xref: /linux/net/core/dev.c (revision 470ac62dfa5732c149adce2cbce84ac678de701f)
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 	N(XDP_FEAT_CHANGE)
1618 	}
1619 #undef N
1620 	return "UNKNOWN_NETDEV_EVENT";
1621 }
1622 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1623 
1624 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1625 				   struct net_device *dev)
1626 {
1627 	struct netdev_notifier_info info = {
1628 		.dev = dev,
1629 	};
1630 
1631 	return nb->notifier_call(nb, val, &info);
1632 }
1633 
1634 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1635 					     struct net_device *dev)
1636 {
1637 	int err;
1638 
1639 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1640 	err = notifier_to_errno(err);
1641 	if (err)
1642 		return err;
1643 
1644 	if (!(dev->flags & IFF_UP))
1645 		return 0;
1646 
1647 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1648 	return 0;
1649 }
1650 
1651 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1652 						struct net_device *dev)
1653 {
1654 	if (dev->flags & IFF_UP) {
1655 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1656 					dev);
1657 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1658 	}
1659 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1660 }
1661 
1662 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1663 						 struct net *net)
1664 {
1665 	struct net_device *dev;
1666 	int err;
1667 
1668 	for_each_netdev(net, dev) {
1669 		err = call_netdevice_register_notifiers(nb, dev);
1670 		if (err)
1671 			goto rollback;
1672 	}
1673 	return 0;
1674 
1675 rollback:
1676 	for_each_netdev_continue_reverse(net, dev)
1677 		call_netdevice_unregister_notifiers(nb, dev);
1678 	return err;
1679 }
1680 
1681 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1682 						    struct net *net)
1683 {
1684 	struct net_device *dev;
1685 
1686 	for_each_netdev(net, dev)
1687 		call_netdevice_unregister_notifiers(nb, dev);
1688 }
1689 
1690 static int dev_boot_phase = 1;
1691 
1692 /**
1693  * register_netdevice_notifier - register a network notifier block
1694  * @nb: notifier
1695  *
1696  * Register a notifier to be called when network device events occur.
1697  * The notifier passed is linked into the kernel structures and must
1698  * not be reused until it has been unregistered. A negative errno code
1699  * is returned on a failure.
1700  *
1701  * When registered all registration and up events are replayed
1702  * to the new notifier to allow device to have a race free
1703  * view of the network device list.
1704  */
1705 
1706 int register_netdevice_notifier(struct notifier_block *nb)
1707 {
1708 	struct net *net;
1709 	int err;
1710 
1711 	/* Close race with setup_net() and cleanup_net() */
1712 	down_write(&pernet_ops_rwsem);
1713 	rtnl_lock();
1714 	err = raw_notifier_chain_register(&netdev_chain, nb);
1715 	if (err)
1716 		goto unlock;
1717 	if (dev_boot_phase)
1718 		goto unlock;
1719 	for_each_net(net) {
1720 		err = call_netdevice_register_net_notifiers(nb, net);
1721 		if (err)
1722 			goto rollback;
1723 	}
1724 
1725 unlock:
1726 	rtnl_unlock();
1727 	up_write(&pernet_ops_rwsem);
1728 	return err;
1729 
1730 rollback:
1731 	for_each_net_continue_reverse(net)
1732 		call_netdevice_unregister_net_notifiers(nb, net);
1733 
1734 	raw_notifier_chain_unregister(&netdev_chain, nb);
1735 	goto unlock;
1736 }
1737 EXPORT_SYMBOL(register_netdevice_notifier);
1738 
1739 /**
1740  * unregister_netdevice_notifier - unregister a network notifier block
1741  * @nb: notifier
1742  *
1743  * Unregister a notifier previously registered by
1744  * register_netdevice_notifier(). The notifier is unlinked into the
1745  * kernel structures and may then be reused. A negative errno code
1746  * is returned on a failure.
1747  *
1748  * After unregistering unregister and down device events are synthesized
1749  * for all devices on the device list to the removed notifier to remove
1750  * the need for special case cleanup code.
1751  */
1752 
1753 int unregister_netdevice_notifier(struct notifier_block *nb)
1754 {
1755 	struct net *net;
1756 	int err;
1757 
1758 	/* Close race with setup_net() and cleanup_net() */
1759 	down_write(&pernet_ops_rwsem);
1760 	rtnl_lock();
1761 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1762 	if (err)
1763 		goto unlock;
1764 
1765 	for_each_net(net)
1766 		call_netdevice_unregister_net_notifiers(nb, net);
1767 
1768 unlock:
1769 	rtnl_unlock();
1770 	up_write(&pernet_ops_rwsem);
1771 	return err;
1772 }
1773 EXPORT_SYMBOL(unregister_netdevice_notifier);
1774 
1775 static int __register_netdevice_notifier_net(struct net *net,
1776 					     struct notifier_block *nb,
1777 					     bool ignore_call_fail)
1778 {
1779 	int err;
1780 
1781 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1782 	if (err)
1783 		return err;
1784 	if (dev_boot_phase)
1785 		return 0;
1786 
1787 	err = call_netdevice_register_net_notifiers(nb, net);
1788 	if (err && !ignore_call_fail)
1789 		goto chain_unregister;
1790 
1791 	return 0;
1792 
1793 chain_unregister:
1794 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1795 	return err;
1796 }
1797 
1798 static int __unregister_netdevice_notifier_net(struct net *net,
1799 					       struct notifier_block *nb)
1800 {
1801 	int err;
1802 
1803 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1804 	if (err)
1805 		return err;
1806 
1807 	call_netdevice_unregister_net_notifiers(nb, net);
1808 	return 0;
1809 }
1810 
1811 /**
1812  * register_netdevice_notifier_net - register a per-netns network notifier block
1813  * @net: network namespace
1814  * @nb: notifier
1815  *
1816  * Register a notifier to be called when network device events occur.
1817  * The notifier passed is linked into the kernel structures and must
1818  * not be reused until it has been unregistered. A negative errno code
1819  * is returned on a failure.
1820  *
1821  * When registered all registration and up events are replayed
1822  * to the new notifier to allow device to have a race free
1823  * view of the network device list.
1824  */
1825 
1826 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1827 {
1828 	int err;
1829 
1830 	rtnl_lock();
1831 	err = __register_netdevice_notifier_net(net, nb, false);
1832 	rtnl_unlock();
1833 	return err;
1834 }
1835 EXPORT_SYMBOL(register_netdevice_notifier_net);
1836 
1837 /**
1838  * unregister_netdevice_notifier_net - unregister a per-netns
1839  *                                     network notifier block
1840  * @net: network namespace
1841  * @nb: notifier
1842  *
1843  * Unregister a notifier previously registered by
1844  * register_netdevice_notifier_net(). The notifier is unlinked from the
1845  * kernel structures and may then be reused. A negative errno code
1846  * is returned on a failure.
1847  *
1848  * After unregistering unregister and down device events are synthesized
1849  * for all devices on the device list to the removed notifier to remove
1850  * the need for special case cleanup code.
1851  */
1852 
1853 int unregister_netdevice_notifier_net(struct net *net,
1854 				      struct notifier_block *nb)
1855 {
1856 	int err;
1857 
1858 	rtnl_lock();
1859 	err = __unregister_netdevice_notifier_net(net, nb);
1860 	rtnl_unlock();
1861 	return err;
1862 }
1863 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1864 
1865 static void __move_netdevice_notifier_net(struct net *src_net,
1866 					  struct net *dst_net,
1867 					  struct notifier_block *nb)
1868 {
1869 	__unregister_netdevice_notifier_net(src_net, nb);
1870 	__register_netdevice_notifier_net(dst_net, nb, true);
1871 }
1872 
1873 void move_netdevice_notifier_net(struct net *src_net, struct net *dst_net,
1874 				 struct notifier_block *nb)
1875 {
1876 	rtnl_lock();
1877 	__move_netdevice_notifier_net(src_net, dst_net, nb);
1878 	rtnl_unlock();
1879 }
1880 
1881 int register_netdevice_notifier_dev_net(struct net_device *dev,
1882 					struct notifier_block *nb,
1883 					struct netdev_net_notifier *nn)
1884 {
1885 	int err;
1886 
1887 	rtnl_lock();
1888 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1889 	if (!err) {
1890 		nn->nb = nb;
1891 		list_add(&nn->list, &dev->net_notifier_list);
1892 	}
1893 	rtnl_unlock();
1894 	return err;
1895 }
1896 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1897 
1898 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1899 					  struct notifier_block *nb,
1900 					  struct netdev_net_notifier *nn)
1901 {
1902 	int err;
1903 
1904 	rtnl_lock();
1905 	list_del(&nn->list);
1906 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1907 	rtnl_unlock();
1908 	return err;
1909 }
1910 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1911 
1912 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1913 					     struct net *net)
1914 {
1915 	struct netdev_net_notifier *nn;
1916 
1917 	list_for_each_entry(nn, &dev->net_notifier_list, list)
1918 		__move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1919 }
1920 
1921 /**
1922  *	call_netdevice_notifiers_info - call all network notifier blocks
1923  *	@val: value passed unmodified to notifier function
1924  *	@info: notifier information data
1925  *
1926  *	Call all network notifier blocks.  Parameters and return value
1927  *	are as for raw_notifier_call_chain().
1928  */
1929 
1930 static int call_netdevice_notifiers_info(unsigned long val,
1931 					 struct netdev_notifier_info *info)
1932 {
1933 	struct net *net = dev_net(info->dev);
1934 	int ret;
1935 
1936 	ASSERT_RTNL();
1937 
1938 	/* Run per-netns notifier block chain first, then run the global one.
1939 	 * Hopefully, one day, the global one is going to be removed after
1940 	 * all notifier block registrators get converted to be per-netns.
1941 	 */
1942 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1943 	if (ret & NOTIFY_STOP_MASK)
1944 		return ret;
1945 	return raw_notifier_call_chain(&netdev_chain, val, info);
1946 }
1947 
1948 /**
1949  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1950  *	                                       for and rollback on error
1951  *	@val_up: value passed unmodified to notifier function
1952  *	@val_down: value passed unmodified to the notifier function when
1953  *	           recovering from an error on @val_up
1954  *	@info: notifier information data
1955  *
1956  *	Call all per-netns network notifier blocks, but not notifier blocks on
1957  *	the global notifier chain. Parameters and return value are as for
1958  *	raw_notifier_call_chain_robust().
1959  */
1960 
1961 static int
1962 call_netdevice_notifiers_info_robust(unsigned long val_up,
1963 				     unsigned long val_down,
1964 				     struct netdev_notifier_info *info)
1965 {
1966 	struct net *net = dev_net(info->dev);
1967 
1968 	ASSERT_RTNL();
1969 
1970 	return raw_notifier_call_chain_robust(&net->netdev_chain,
1971 					      val_up, val_down, info);
1972 }
1973 
1974 static int call_netdevice_notifiers_extack(unsigned long val,
1975 					   struct net_device *dev,
1976 					   struct netlink_ext_ack *extack)
1977 {
1978 	struct netdev_notifier_info info = {
1979 		.dev = dev,
1980 		.extack = extack,
1981 	};
1982 
1983 	return call_netdevice_notifiers_info(val, &info);
1984 }
1985 
1986 /**
1987  *	call_netdevice_notifiers - call all network notifier blocks
1988  *      @val: value passed unmodified to notifier function
1989  *      @dev: net_device pointer passed unmodified to notifier function
1990  *
1991  *	Call all network notifier blocks.  Parameters and return value
1992  *	are as for raw_notifier_call_chain().
1993  */
1994 
1995 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1996 {
1997 	return call_netdevice_notifiers_extack(val, dev, NULL);
1998 }
1999 EXPORT_SYMBOL(call_netdevice_notifiers);
2000 
2001 /**
2002  *	call_netdevice_notifiers_mtu - call all network notifier blocks
2003  *	@val: value passed unmodified to notifier function
2004  *	@dev: net_device pointer passed unmodified to notifier function
2005  *	@arg: additional u32 argument passed to the notifier function
2006  *
2007  *	Call all network notifier blocks.  Parameters and return value
2008  *	are as for raw_notifier_call_chain().
2009  */
2010 static int call_netdevice_notifiers_mtu(unsigned long val,
2011 					struct net_device *dev, u32 arg)
2012 {
2013 	struct netdev_notifier_info_ext info = {
2014 		.info.dev = dev,
2015 		.ext.mtu = arg,
2016 	};
2017 
2018 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2019 
2020 	return call_netdevice_notifiers_info(val, &info.info);
2021 }
2022 
2023 #ifdef CONFIG_NET_INGRESS
2024 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2025 
2026 void net_inc_ingress_queue(void)
2027 {
2028 	static_branch_inc(&ingress_needed_key);
2029 }
2030 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2031 
2032 void net_dec_ingress_queue(void)
2033 {
2034 	static_branch_dec(&ingress_needed_key);
2035 }
2036 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2037 #endif
2038 
2039 #ifdef CONFIG_NET_EGRESS
2040 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2041 
2042 void net_inc_egress_queue(void)
2043 {
2044 	static_branch_inc(&egress_needed_key);
2045 }
2046 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2047 
2048 void net_dec_egress_queue(void)
2049 {
2050 	static_branch_dec(&egress_needed_key);
2051 }
2052 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2053 #endif
2054 
2055 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2056 EXPORT_SYMBOL(netstamp_needed_key);
2057 #ifdef CONFIG_JUMP_LABEL
2058 static atomic_t netstamp_needed_deferred;
2059 static atomic_t netstamp_wanted;
2060 static void netstamp_clear(struct work_struct *work)
2061 {
2062 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2063 	int wanted;
2064 
2065 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2066 	if (wanted > 0)
2067 		static_branch_enable(&netstamp_needed_key);
2068 	else
2069 		static_branch_disable(&netstamp_needed_key);
2070 }
2071 static DECLARE_WORK(netstamp_work, netstamp_clear);
2072 #endif
2073 
2074 void net_enable_timestamp(void)
2075 {
2076 #ifdef CONFIG_JUMP_LABEL
2077 	int wanted = atomic_read(&netstamp_wanted);
2078 
2079 	while (wanted > 0) {
2080 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2081 			return;
2082 	}
2083 	atomic_inc(&netstamp_needed_deferred);
2084 	schedule_work(&netstamp_work);
2085 #else
2086 	static_branch_inc(&netstamp_needed_key);
2087 #endif
2088 }
2089 EXPORT_SYMBOL(net_enable_timestamp);
2090 
2091 void net_disable_timestamp(void)
2092 {
2093 #ifdef CONFIG_JUMP_LABEL
2094 	int wanted = atomic_read(&netstamp_wanted);
2095 
2096 	while (wanted > 1) {
2097 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2098 			return;
2099 	}
2100 	atomic_dec(&netstamp_needed_deferred);
2101 	schedule_work(&netstamp_work);
2102 #else
2103 	static_branch_dec(&netstamp_needed_key);
2104 #endif
2105 }
2106 EXPORT_SYMBOL(net_disable_timestamp);
2107 
2108 static inline void net_timestamp_set(struct sk_buff *skb)
2109 {
2110 	skb->tstamp = 0;
2111 	skb->mono_delivery_time = 0;
2112 	if (static_branch_unlikely(&netstamp_needed_key))
2113 		skb->tstamp = ktime_get_real();
2114 }
2115 
2116 #define net_timestamp_check(COND, SKB)				\
2117 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2118 		if ((COND) && !(SKB)->tstamp)			\
2119 			(SKB)->tstamp = ktime_get_real();	\
2120 	}							\
2121 
2122 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2123 {
2124 	return __is_skb_forwardable(dev, skb, true);
2125 }
2126 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2127 
2128 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2129 			      bool check_mtu)
2130 {
2131 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2132 
2133 	if (likely(!ret)) {
2134 		skb->protocol = eth_type_trans(skb, dev);
2135 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2136 	}
2137 
2138 	return ret;
2139 }
2140 
2141 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2142 {
2143 	return __dev_forward_skb2(dev, skb, true);
2144 }
2145 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2146 
2147 /**
2148  * dev_forward_skb - loopback an skb to another netif
2149  *
2150  * @dev: destination network device
2151  * @skb: buffer to forward
2152  *
2153  * return values:
2154  *	NET_RX_SUCCESS	(no congestion)
2155  *	NET_RX_DROP     (packet was dropped, but freed)
2156  *
2157  * dev_forward_skb can be used for injecting an skb from the
2158  * start_xmit function of one device into the receive queue
2159  * of another device.
2160  *
2161  * The receiving device may be in another namespace, so
2162  * we have to clear all information in the skb that could
2163  * impact namespace isolation.
2164  */
2165 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2166 {
2167 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2168 }
2169 EXPORT_SYMBOL_GPL(dev_forward_skb);
2170 
2171 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2172 {
2173 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2174 }
2175 
2176 static inline int deliver_skb(struct sk_buff *skb,
2177 			      struct packet_type *pt_prev,
2178 			      struct net_device *orig_dev)
2179 {
2180 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2181 		return -ENOMEM;
2182 	refcount_inc(&skb->users);
2183 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2184 }
2185 
2186 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2187 					  struct packet_type **pt,
2188 					  struct net_device *orig_dev,
2189 					  __be16 type,
2190 					  struct list_head *ptype_list)
2191 {
2192 	struct packet_type *ptype, *pt_prev = *pt;
2193 
2194 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2195 		if (ptype->type != type)
2196 			continue;
2197 		if (pt_prev)
2198 			deliver_skb(skb, pt_prev, orig_dev);
2199 		pt_prev = ptype;
2200 	}
2201 	*pt = pt_prev;
2202 }
2203 
2204 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2205 {
2206 	if (!ptype->af_packet_priv || !skb->sk)
2207 		return false;
2208 
2209 	if (ptype->id_match)
2210 		return ptype->id_match(ptype, skb->sk);
2211 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2212 		return true;
2213 
2214 	return false;
2215 }
2216 
2217 /**
2218  * dev_nit_active - return true if any network interface taps are in use
2219  *
2220  * @dev: network device to check for the presence of taps
2221  */
2222 bool dev_nit_active(struct net_device *dev)
2223 {
2224 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2225 }
2226 EXPORT_SYMBOL_GPL(dev_nit_active);
2227 
2228 /*
2229  *	Support routine. Sends outgoing frames to any network
2230  *	taps currently in use.
2231  */
2232 
2233 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2234 {
2235 	struct packet_type *ptype;
2236 	struct sk_buff *skb2 = NULL;
2237 	struct packet_type *pt_prev = NULL;
2238 	struct list_head *ptype_list = &ptype_all;
2239 
2240 	rcu_read_lock();
2241 again:
2242 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2243 		if (ptype->ignore_outgoing)
2244 			continue;
2245 
2246 		/* Never send packets back to the socket
2247 		 * they originated from - MvS (miquels@drinkel.ow.org)
2248 		 */
2249 		if (skb_loop_sk(ptype, skb))
2250 			continue;
2251 
2252 		if (pt_prev) {
2253 			deliver_skb(skb2, pt_prev, skb->dev);
2254 			pt_prev = ptype;
2255 			continue;
2256 		}
2257 
2258 		/* need to clone skb, done only once */
2259 		skb2 = skb_clone(skb, GFP_ATOMIC);
2260 		if (!skb2)
2261 			goto out_unlock;
2262 
2263 		net_timestamp_set(skb2);
2264 
2265 		/* skb->nh should be correctly
2266 		 * set by sender, so that the second statement is
2267 		 * just protection against buggy protocols.
2268 		 */
2269 		skb_reset_mac_header(skb2);
2270 
2271 		if (skb_network_header(skb2) < skb2->data ||
2272 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2273 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2274 					     ntohs(skb2->protocol),
2275 					     dev->name);
2276 			skb_reset_network_header(skb2);
2277 		}
2278 
2279 		skb2->transport_header = skb2->network_header;
2280 		skb2->pkt_type = PACKET_OUTGOING;
2281 		pt_prev = ptype;
2282 	}
2283 
2284 	if (ptype_list == &ptype_all) {
2285 		ptype_list = &dev->ptype_all;
2286 		goto again;
2287 	}
2288 out_unlock:
2289 	if (pt_prev) {
2290 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2291 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2292 		else
2293 			kfree_skb(skb2);
2294 	}
2295 	rcu_read_unlock();
2296 }
2297 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2298 
2299 /**
2300  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2301  * @dev: Network device
2302  * @txq: number of queues available
2303  *
2304  * If real_num_tx_queues is changed the tc mappings may no longer be
2305  * valid. To resolve this verify the tc mapping remains valid and if
2306  * not NULL the mapping. With no priorities mapping to this
2307  * offset/count pair it will no longer be used. In the worst case TC0
2308  * is invalid nothing can be done so disable priority mappings. If is
2309  * expected that drivers will fix this mapping if they can before
2310  * calling netif_set_real_num_tx_queues.
2311  */
2312 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2313 {
2314 	int i;
2315 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2316 
2317 	/* If TC0 is invalidated disable TC mapping */
2318 	if (tc->offset + tc->count > txq) {
2319 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2320 		dev->num_tc = 0;
2321 		return;
2322 	}
2323 
2324 	/* Invalidated prio to tc mappings set to TC0 */
2325 	for (i = 1; i < TC_BITMASK + 1; i++) {
2326 		int q = netdev_get_prio_tc_map(dev, i);
2327 
2328 		tc = &dev->tc_to_txq[q];
2329 		if (tc->offset + tc->count > txq) {
2330 			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",
2331 				    i, q);
2332 			netdev_set_prio_tc_map(dev, i, 0);
2333 		}
2334 	}
2335 }
2336 
2337 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2338 {
2339 	if (dev->num_tc) {
2340 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2341 		int i;
2342 
2343 		/* walk through the TCs and see if it falls into any of them */
2344 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2345 			if ((txq - tc->offset) < tc->count)
2346 				return i;
2347 		}
2348 
2349 		/* didn't find it, just return -1 to indicate no match */
2350 		return -1;
2351 	}
2352 
2353 	return 0;
2354 }
2355 EXPORT_SYMBOL(netdev_txq_to_tc);
2356 
2357 #ifdef CONFIG_XPS
2358 static struct static_key xps_needed __read_mostly;
2359 static struct static_key xps_rxqs_needed __read_mostly;
2360 static DEFINE_MUTEX(xps_map_mutex);
2361 #define xmap_dereference(P)		\
2362 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2363 
2364 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2365 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2366 {
2367 	struct xps_map *map = NULL;
2368 	int pos;
2369 
2370 	if (dev_maps)
2371 		map = xmap_dereference(dev_maps->attr_map[tci]);
2372 	if (!map)
2373 		return false;
2374 
2375 	for (pos = map->len; pos--;) {
2376 		if (map->queues[pos] != index)
2377 			continue;
2378 
2379 		if (map->len > 1) {
2380 			map->queues[pos] = map->queues[--map->len];
2381 			break;
2382 		}
2383 
2384 		if (old_maps)
2385 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2386 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2387 		kfree_rcu(map, rcu);
2388 		return false;
2389 	}
2390 
2391 	return true;
2392 }
2393 
2394 static bool remove_xps_queue_cpu(struct net_device *dev,
2395 				 struct xps_dev_maps *dev_maps,
2396 				 int cpu, u16 offset, u16 count)
2397 {
2398 	int num_tc = dev_maps->num_tc;
2399 	bool active = false;
2400 	int tci;
2401 
2402 	for (tci = cpu * num_tc; num_tc--; tci++) {
2403 		int i, j;
2404 
2405 		for (i = count, j = offset; i--; j++) {
2406 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2407 				break;
2408 		}
2409 
2410 		active |= i < 0;
2411 	}
2412 
2413 	return active;
2414 }
2415 
2416 static void reset_xps_maps(struct net_device *dev,
2417 			   struct xps_dev_maps *dev_maps,
2418 			   enum xps_map_type type)
2419 {
2420 	static_key_slow_dec_cpuslocked(&xps_needed);
2421 	if (type == XPS_RXQS)
2422 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2423 
2424 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2425 
2426 	kfree_rcu(dev_maps, rcu);
2427 }
2428 
2429 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2430 			   u16 offset, u16 count)
2431 {
2432 	struct xps_dev_maps *dev_maps;
2433 	bool active = false;
2434 	int i, j;
2435 
2436 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2437 	if (!dev_maps)
2438 		return;
2439 
2440 	for (j = 0; j < dev_maps->nr_ids; j++)
2441 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2442 	if (!active)
2443 		reset_xps_maps(dev, dev_maps, type);
2444 
2445 	if (type == XPS_CPUS) {
2446 		for (i = offset + (count - 1); count--; i--)
2447 			netdev_queue_numa_node_write(
2448 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2449 	}
2450 }
2451 
2452 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2453 				   u16 count)
2454 {
2455 	if (!static_key_false(&xps_needed))
2456 		return;
2457 
2458 	cpus_read_lock();
2459 	mutex_lock(&xps_map_mutex);
2460 
2461 	if (static_key_false(&xps_rxqs_needed))
2462 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2463 
2464 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2465 
2466 	mutex_unlock(&xps_map_mutex);
2467 	cpus_read_unlock();
2468 }
2469 
2470 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2471 {
2472 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2473 }
2474 
2475 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2476 				      u16 index, bool is_rxqs_map)
2477 {
2478 	struct xps_map *new_map;
2479 	int alloc_len = XPS_MIN_MAP_ALLOC;
2480 	int i, pos;
2481 
2482 	for (pos = 0; map && pos < map->len; pos++) {
2483 		if (map->queues[pos] != index)
2484 			continue;
2485 		return map;
2486 	}
2487 
2488 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2489 	if (map) {
2490 		if (pos < map->alloc_len)
2491 			return map;
2492 
2493 		alloc_len = map->alloc_len * 2;
2494 	}
2495 
2496 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2497 	 *  map
2498 	 */
2499 	if (is_rxqs_map)
2500 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2501 	else
2502 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2503 				       cpu_to_node(attr_index));
2504 	if (!new_map)
2505 		return NULL;
2506 
2507 	for (i = 0; i < pos; i++)
2508 		new_map->queues[i] = map->queues[i];
2509 	new_map->alloc_len = alloc_len;
2510 	new_map->len = pos;
2511 
2512 	return new_map;
2513 }
2514 
2515 /* Copy xps maps at a given index */
2516 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2517 			      struct xps_dev_maps *new_dev_maps, int index,
2518 			      int tc, bool skip_tc)
2519 {
2520 	int i, tci = index * dev_maps->num_tc;
2521 	struct xps_map *map;
2522 
2523 	/* copy maps belonging to foreign traffic classes */
2524 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2525 		if (i == tc && skip_tc)
2526 			continue;
2527 
2528 		/* fill in the new device map from the old device map */
2529 		map = xmap_dereference(dev_maps->attr_map[tci]);
2530 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2531 	}
2532 }
2533 
2534 /* Must be called under cpus_read_lock */
2535 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2536 			  u16 index, enum xps_map_type type)
2537 {
2538 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2539 	const unsigned long *online_mask = NULL;
2540 	bool active = false, copy = false;
2541 	int i, j, tci, numa_node_id = -2;
2542 	int maps_sz, num_tc = 1, tc = 0;
2543 	struct xps_map *map, *new_map;
2544 	unsigned int nr_ids;
2545 
2546 	if (dev->num_tc) {
2547 		/* Do not allow XPS on subordinate device directly */
2548 		num_tc = dev->num_tc;
2549 		if (num_tc < 0)
2550 			return -EINVAL;
2551 
2552 		/* If queue belongs to subordinate dev use its map */
2553 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2554 
2555 		tc = netdev_txq_to_tc(dev, index);
2556 		if (tc < 0)
2557 			return -EINVAL;
2558 	}
2559 
2560 	mutex_lock(&xps_map_mutex);
2561 
2562 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2563 	if (type == XPS_RXQS) {
2564 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2565 		nr_ids = dev->num_rx_queues;
2566 	} else {
2567 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2568 		if (num_possible_cpus() > 1)
2569 			online_mask = cpumask_bits(cpu_online_mask);
2570 		nr_ids = nr_cpu_ids;
2571 	}
2572 
2573 	if (maps_sz < L1_CACHE_BYTES)
2574 		maps_sz = L1_CACHE_BYTES;
2575 
2576 	/* The old dev_maps could be larger or smaller than the one we're
2577 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2578 	 * between. We could try to be smart, but let's be safe instead and only
2579 	 * copy foreign traffic classes if the two map sizes match.
2580 	 */
2581 	if (dev_maps &&
2582 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2583 		copy = true;
2584 
2585 	/* allocate memory for queue storage */
2586 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2587 	     j < nr_ids;) {
2588 		if (!new_dev_maps) {
2589 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2590 			if (!new_dev_maps) {
2591 				mutex_unlock(&xps_map_mutex);
2592 				return -ENOMEM;
2593 			}
2594 
2595 			new_dev_maps->nr_ids = nr_ids;
2596 			new_dev_maps->num_tc = num_tc;
2597 		}
2598 
2599 		tci = j * num_tc + tc;
2600 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2601 
2602 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2603 		if (!map)
2604 			goto error;
2605 
2606 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2607 	}
2608 
2609 	if (!new_dev_maps)
2610 		goto out_no_new_maps;
2611 
2612 	if (!dev_maps) {
2613 		/* Increment static keys at most once per type */
2614 		static_key_slow_inc_cpuslocked(&xps_needed);
2615 		if (type == XPS_RXQS)
2616 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2617 	}
2618 
2619 	for (j = 0; j < nr_ids; j++) {
2620 		bool skip_tc = false;
2621 
2622 		tci = j * num_tc + tc;
2623 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2624 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2625 			/* add tx-queue to CPU/rx-queue maps */
2626 			int pos = 0;
2627 
2628 			skip_tc = true;
2629 
2630 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2631 			while ((pos < map->len) && (map->queues[pos] != index))
2632 				pos++;
2633 
2634 			if (pos == map->len)
2635 				map->queues[map->len++] = index;
2636 #ifdef CONFIG_NUMA
2637 			if (type == XPS_CPUS) {
2638 				if (numa_node_id == -2)
2639 					numa_node_id = cpu_to_node(j);
2640 				else if (numa_node_id != cpu_to_node(j))
2641 					numa_node_id = -1;
2642 			}
2643 #endif
2644 		}
2645 
2646 		if (copy)
2647 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2648 					  skip_tc);
2649 	}
2650 
2651 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2652 
2653 	/* Cleanup old maps */
2654 	if (!dev_maps)
2655 		goto out_no_old_maps;
2656 
2657 	for (j = 0; j < dev_maps->nr_ids; j++) {
2658 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2659 			map = xmap_dereference(dev_maps->attr_map[tci]);
2660 			if (!map)
2661 				continue;
2662 
2663 			if (copy) {
2664 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2665 				if (map == new_map)
2666 					continue;
2667 			}
2668 
2669 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2670 			kfree_rcu(map, rcu);
2671 		}
2672 	}
2673 
2674 	old_dev_maps = dev_maps;
2675 
2676 out_no_old_maps:
2677 	dev_maps = new_dev_maps;
2678 	active = true;
2679 
2680 out_no_new_maps:
2681 	if (type == XPS_CPUS)
2682 		/* update Tx queue numa node */
2683 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2684 					     (numa_node_id >= 0) ?
2685 					     numa_node_id : NUMA_NO_NODE);
2686 
2687 	if (!dev_maps)
2688 		goto out_no_maps;
2689 
2690 	/* removes tx-queue from unused CPUs/rx-queues */
2691 	for (j = 0; j < dev_maps->nr_ids; j++) {
2692 		tci = j * dev_maps->num_tc;
2693 
2694 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2695 			if (i == tc &&
2696 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2697 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2698 				continue;
2699 
2700 			active |= remove_xps_queue(dev_maps,
2701 						   copy ? old_dev_maps : NULL,
2702 						   tci, index);
2703 		}
2704 	}
2705 
2706 	if (old_dev_maps)
2707 		kfree_rcu(old_dev_maps, rcu);
2708 
2709 	/* free map if not active */
2710 	if (!active)
2711 		reset_xps_maps(dev, dev_maps, type);
2712 
2713 out_no_maps:
2714 	mutex_unlock(&xps_map_mutex);
2715 
2716 	return 0;
2717 error:
2718 	/* remove any maps that we added */
2719 	for (j = 0; j < nr_ids; j++) {
2720 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2721 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2722 			map = copy ?
2723 			      xmap_dereference(dev_maps->attr_map[tci]) :
2724 			      NULL;
2725 			if (new_map && new_map != map)
2726 				kfree(new_map);
2727 		}
2728 	}
2729 
2730 	mutex_unlock(&xps_map_mutex);
2731 
2732 	kfree(new_dev_maps);
2733 	return -ENOMEM;
2734 }
2735 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2736 
2737 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2738 			u16 index)
2739 {
2740 	int ret;
2741 
2742 	cpus_read_lock();
2743 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2744 	cpus_read_unlock();
2745 
2746 	return ret;
2747 }
2748 EXPORT_SYMBOL(netif_set_xps_queue);
2749 
2750 #endif
2751 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2752 {
2753 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2754 
2755 	/* Unbind any subordinate channels */
2756 	while (txq-- != &dev->_tx[0]) {
2757 		if (txq->sb_dev)
2758 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2759 	}
2760 }
2761 
2762 void netdev_reset_tc(struct net_device *dev)
2763 {
2764 #ifdef CONFIG_XPS
2765 	netif_reset_xps_queues_gt(dev, 0);
2766 #endif
2767 	netdev_unbind_all_sb_channels(dev);
2768 
2769 	/* Reset TC configuration of device */
2770 	dev->num_tc = 0;
2771 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2772 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2773 }
2774 EXPORT_SYMBOL(netdev_reset_tc);
2775 
2776 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2777 {
2778 	if (tc >= dev->num_tc)
2779 		return -EINVAL;
2780 
2781 #ifdef CONFIG_XPS
2782 	netif_reset_xps_queues(dev, offset, count);
2783 #endif
2784 	dev->tc_to_txq[tc].count = count;
2785 	dev->tc_to_txq[tc].offset = offset;
2786 	return 0;
2787 }
2788 EXPORT_SYMBOL(netdev_set_tc_queue);
2789 
2790 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2791 {
2792 	if (num_tc > TC_MAX_QUEUE)
2793 		return -EINVAL;
2794 
2795 #ifdef CONFIG_XPS
2796 	netif_reset_xps_queues_gt(dev, 0);
2797 #endif
2798 	netdev_unbind_all_sb_channels(dev);
2799 
2800 	dev->num_tc = num_tc;
2801 	return 0;
2802 }
2803 EXPORT_SYMBOL(netdev_set_num_tc);
2804 
2805 void netdev_unbind_sb_channel(struct net_device *dev,
2806 			      struct net_device *sb_dev)
2807 {
2808 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2809 
2810 #ifdef CONFIG_XPS
2811 	netif_reset_xps_queues_gt(sb_dev, 0);
2812 #endif
2813 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2814 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2815 
2816 	while (txq-- != &dev->_tx[0]) {
2817 		if (txq->sb_dev == sb_dev)
2818 			txq->sb_dev = NULL;
2819 	}
2820 }
2821 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2822 
2823 int netdev_bind_sb_channel_queue(struct net_device *dev,
2824 				 struct net_device *sb_dev,
2825 				 u8 tc, u16 count, u16 offset)
2826 {
2827 	/* Make certain the sb_dev and dev are already configured */
2828 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2829 		return -EINVAL;
2830 
2831 	/* We cannot hand out queues we don't have */
2832 	if ((offset + count) > dev->real_num_tx_queues)
2833 		return -EINVAL;
2834 
2835 	/* Record the mapping */
2836 	sb_dev->tc_to_txq[tc].count = count;
2837 	sb_dev->tc_to_txq[tc].offset = offset;
2838 
2839 	/* Provide a way for Tx queue to find the tc_to_txq map or
2840 	 * XPS map for itself.
2841 	 */
2842 	while (count--)
2843 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2844 
2845 	return 0;
2846 }
2847 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2848 
2849 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2850 {
2851 	/* Do not use a multiqueue device to represent a subordinate channel */
2852 	if (netif_is_multiqueue(dev))
2853 		return -ENODEV;
2854 
2855 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2856 	 * Channel 0 is meant to be "native" mode and used only to represent
2857 	 * the main root device. We allow writing 0 to reset the device back
2858 	 * to normal mode after being used as a subordinate channel.
2859 	 */
2860 	if (channel > S16_MAX)
2861 		return -EINVAL;
2862 
2863 	dev->num_tc = -channel;
2864 
2865 	return 0;
2866 }
2867 EXPORT_SYMBOL(netdev_set_sb_channel);
2868 
2869 /*
2870  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2871  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2872  */
2873 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2874 {
2875 	bool disabling;
2876 	int rc;
2877 
2878 	disabling = txq < dev->real_num_tx_queues;
2879 
2880 	if (txq < 1 || txq > dev->num_tx_queues)
2881 		return -EINVAL;
2882 
2883 	if (dev->reg_state == NETREG_REGISTERED ||
2884 	    dev->reg_state == NETREG_UNREGISTERING) {
2885 		ASSERT_RTNL();
2886 
2887 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2888 						  txq);
2889 		if (rc)
2890 			return rc;
2891 
2892 		if (dev->num_tc)
2893 			netif_setup_tc(dev, txq);
2894 
2895 		dev_qdisc_change_real_num_tx(dev, txq);
2896 
2897 		dev->real_num_tx_queues = txq;
2898 
2899 		if (disabling) {
2900 			synchronize_net();
2901 			qdisc_reset_all_tx_gt(dev, txq);
2902 #ifdef CONFIG_XPS
2903 			netif_reset_xps_queues_gt(dev, txq);
2904 #endif
2905 		}
2906 	} else {
2907 		dev->real_num_tx_queues = txq;
2908 	}
2909 
2910 	return 0;
2911 }
2912 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2913 
2914 #ifdef CONFIG_SYSFS
2915 /**
2916  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2917  *	@dev: Network device
2918  *	@rxq: Actual number of RX queues
2919  *
2920  *	This must be called either with the rtnl_lock held or before
2921  *	registration of the net device.  Returns 0 on success, or a
2922  *	negative error code.  If called before registration, it always
2923  *	succeeds.
2924  */
2925 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2926 {
2927 	int rc;
2928 
2929 	if (rxq < 1 || rxq > dev->num_rx_queues)
2930 		return -EINVAL;
2931 
2932 	if (dev->reg_state == NETREG_REGISTERED) {
2933 		ASSERT_RTNL();
2934 
2935 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2936 						  rxq);
2937 		if (rc)
2938 			return rc;
2939 	}
2940 
2941 	dev->real_num_rx_queues = rxq;
2942 	return 0;
2943 }
2944 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2945 #endif
2946 
2947 /**
2948  *	netif_set_real_num_queues - set actual number of RX and TX queues used
2949  *	@dev: Network device
2950  *	@txq: Actual number of TX queues
2951  *	@rxq: Actual number of RX queues
2952  *
2953  *	Set the real number of both TX and RX queues.
2954  *	Does nothing if the number of queues is already correct.
2955  */
2956 int netif_set_real_num_queues(struct net_device *dev,
2957 			      unsigned int txq, unsigned int rxq)
2958 {
2959 	unsigned int old_rxq = dev->real_num_rx_queues;
2960 	int err;
2961 
2962 	if (txq < 1 || txq > dev->num_tx_queues ||
2963 	    rxq < 1 || rxq > dev->num_rx_queues)
2964 		return -EINVAL;
2965 
2966 	/* Start from increases, so the error path only does decreases -
2967 	 * decreases can't fail.
2968 	 */
2969 	if (rxq > dev->real_num_rx_queues) {
2970 		err = netif_set_real_num_rx_queues(dev, rxq);
2971 		if (err)
2972 			return err;
2973 	}
2974 	if (txq > dev->real_num_tx_queues) {
2975 		err = netif_set_real_num_tx_queues(dev, txq);
2976 		if (err)
2977 			goto undo_rx;
2978 	}
2979 	if (rxq < dev->real_num_rx_queues)
2980 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2981 	if (txq < dev->real_num_tx_queues)
2982 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2983 
2984 	return 0;
2985 undo_rx:
2986 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2987 	return err;
2988 }
2989 EXPORT_SYMBOL(netif_set_real_num_queues);
2990 
2991 /**
2992  * netif_set_tso_max_size() - set the max size of TSO frames supported
2993  * @dev:	netdev to update
2994  * @size:	max skb->len of a TSO frame
2995  *
2996  * Set the limit on the size of TSO super-frames the device can handle.
2997  * Unless explicitly set the stack will assume the value of
2998  * %GSO_LEGACY_MAX_SIZE.
2999  */
3000 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3001 {
3002 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
3003 	if (size < READ_ONCE(dev->gso_max_size))
3004 		netif_set_gso_max_size(dev, size);
3005 	if (size < READ_ONCE(dev->gso_ipv4_max_size))
3006 		netif_set_gso_ipv4_max_size(dev, size);
3007 }
3008 EXPORT_SYMBOL(netif_set_tso_max_size);
3009 
3010 /**
3011  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3012  * @dev:	netdev to update
3013  * @segs:	max number of TCP segments
3014  *
3015  * Set the limit on the number of TCP segments the device can generate from
3016  * a single TSO super-frame.
3017  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3018  */
3019 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3020 {
3021 	dev->tso_max_segs = segs;
3022 	if (segs < READ_ONCE(dev->gso_max_segs))
3023 		netif_set_gso_max_segs(dev, segs);
3024 }
3025 EXPORT_SYMBOL(netif_set_tso_max_segs);
3026 
3027 /**
3028  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3029  * @to:		netdev to update
3030  * @from:	netdev from which to copy the limits
3031  */
3032 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3033 {
3034 	netif_set_tso_max_size(to, from->tso_max_size);
3035 	netif_set_tso_max_segs(to, from->tso_max_segs);
3036 }
3037 EXPORT_SYMBOL(netif_inherit_tso_max);
3038 
3039 /**
3040  * netif_get_num_default_rss_queues - default number of RSS queues
3041  *
3042  * Default value is the number of physical cores if there are only 1 or 2, or
3043  * divided by 2 if there are more.
3044  */
3045 int netif_get_num_default_rss_queues(void)
3046 {
3047 	cpumask_var_t cpus;
3048 	int cpu, count = 0;
3049 
3050 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3051 		return 1;
3052 
3053 	cpumask_copy(cpus, cpu_online_mask);
3054 	for_each_cpu(cpu, cpus) {
3055 		++count;
3056 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3057 	}
3058 	free_cpumask_var(cpus);
3059 
3060 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3061 }
3062 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3063 
3064 static void __netif_reschedule(struct Qdisc *q)
3065 {
3066 	struct softnet_data *sd;
3067 	unsigned long flags;
3068 
3069 	local_irq_save(flags);
3070 	sd = this_cpu_ptr(&softnet_data);
3071 	q->next_sched = NULL;
3072 	*sd->output_queue_tailp = q;
3073 	sd->output_queue_tailp = &q->next_sched;
3074 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3075 	local_irq_restore(flags);
3076 }
3077 
3078 void __netif_schedule(struct Qdisc *q)
3079 {
3080 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3081 		__netif_reschedule(q);
3082 }
3083 EXPORT_SYMBOL(__netif_schedule);
3084 
3085 struct dev_kfree_skb_cb {
3086 	enum skb_free_reason reason;
3087 };
3088 
3089 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3090 {
3091 	return (struct dev_kfree_skb_cb *)skb->cb;
3092 }
3093 
3094 void netif_schedule_queue(struct netdev_queue *txq)
3095 {
3096 	rcu_read_lock();
3097 	if (!netif_xmit_stopped(txq)) {
3098 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3099 
3100 		__netif_schedule(q);
3101 	}
3102 	rcu_read_unlock();
3103 }
3104 EXPORT_SYMBOL(netif_schedule_queue);
3105 
3106 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3107 {
3108 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3109 		struct Qdisc *q;
3110 
3111 		rcu_read_lock();
3112 		q = rcu_dereference(dev_queue->qdisc);
3113 		__netif_schedule(q);
3114 		rcu_read_unlock();
3115 	}
3116 }
3117 EXPORT_SYMBOL(netif_tx_wake_queue);
3118 
3119 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3120 {
3121 	unsigned long flags;
3122 
3123 	if (unlikely(!skb))
3124 		return;
3125 
3126 	if (likely(refcount_read(&skb->users) == 1)) {
3127 		smp_rmb();
3128 		refcount_set(&skb->users, 0);
3129 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3130 		return;
3131 	}
3132 	get_kfree_skb_cb(skb)->reason = reason;
3133 	local_irq_save(flags);
3134 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3135 	__this_cpu_write(softnet_data.completion_queue, skb);
3136 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3137 	local_irq_restore(flags);
3138 }
3139 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3140 
3141 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3142 {
3143 	if (in_hardirq() || irqs_disabled())
3144 		__dev_kfree_skb_irq(skb, reason);
3145 	else
3146 		dev_kfree_skb(skb);
3147 }
3148 EXPORT_SYMBOL(__dev_kfree_skb_any);
3149 
3150 
3151 /**
3152  * netif_device_detach - mark device as removed
3153  * @dev: network device
3154  *
3155  * Mark device as removed from system and therefore no longer available.
3156  */
3157 void netif_device_detach(struct net_device *dev)
3158 {
3159 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3160 	    netif_running(dev)) {
3161 		netif_tx_stop_all_queues(dev);
3162 	}
3163 }
3164 EXPORT_SYMBOL(netif_device_detach);
3165 
3166 /**
3167  * netif_device_attach - mark device as attached
3168  * @dev: network device
3169  *
3170  * Mark device as attached from system and restart if needed.
3171  */
3172 void netif_device_attach(struct net_device *dev)
3173 {
3174 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3175 	    netif_running(dev)) {
3176 		netif_tx_wake_all_queues(dev);
3177 		__netdev_watchdog_up(dev);
3178 	}
3179 }
3180 EXPORT_SYMBOL(netif_device_attach);
3181 
3182 /*
3183  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3184  * to be used as a distribution range.
3185  */
3186 static u16 skb_tx_hash(const struct net_device *dev,
3187 		       const struct net_device *sb_dev,
3188 		       struct sk_buff *skb)
3189 {
3190 	u32 hash;
3191 	u16 qoffset = 0;
3192 	u16 qcount = dev->real_num_tx_queues;
3193 
3194 	if (dev->num_tc) {
3195 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3196 
3197 		qoffset = sb_dev->tc_to_txq[tc].offset;
3198 		qcount = sb_dev->tc_to_txq[tc].count;
3199 		if (unlikely(!qcount)) {
3200 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3201 					     sb_dev->name, qoffset, tc);
3202 			qoffset = 0;
3203 			qcount = dev->real_num_tx_queues;
3204 		}
3205 	}
3206 
3207 	if (skb_rx_queue_recorded(skb)) {
3208 		hash = skb_get_rx_queue(skb);
3209 		if (hash >= qoffset)
3210 			hash -= qoffset;
3211 		while (unlikely(hash >= qcount))
3212 			hash -= qcount;
3213 		return hash + qoffset;
3214 	}
3215 
3216 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3217 }
3218 
3219 static void skb_warn_bad_offload(const struct sk_buff *skb)
3220 {
3221 	static const netdev_features_t null_features;
3222 	struct net_device *dev = skb->dev;
3223 	const char *name = "";
3224 
3225 	if (!net_ratelimit())
3226 		return;
3227 
3228 	if (dev) {
3229 		if (dev->dev.parent)
3230 			name = dev_driver_string(dev->dev.parent);
3231 		else
3232 			name = netdev_name(dev);
3233 	}
3234 	skb_dump(KERN_WARNING, skb, false);
3235 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3236 	     name, dev ? &dev->features : &null_features,
3237 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3238 }
3239 
3240 /*
3241  * Invalidate hardware checksum when packet is to be mangled, and
3242  * complete checksum manually on outgoing path.
3243  */
3244 int skb_checksum_help(struct sk_buff *skb)
3245 {
3246 	__wsum csum;
3247 	int ret = 0, offset;
3248 
3249 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3250 		goto out_set_summed;
3251 
3252 	if (unlikely(skb_is_gso(skb))) {
3253 		skb_warn_bad_offload(skb);
3254 		return -EINVAL;
3255 	}
3256 
3257 	/* Before computing a checksum, we should make sure no frag could
3258 	 * be modified by an external entity : checksum could be wrong.
3259 	 */
3260 	if (skb_has_shared_frag(skb)) {
3261 		ret = __skb_linearize(skb);
3262 		if (ret)
3263 			goto out;
3264 	}
3265 
3266 	offset = skb_checksum_start_offset(skb);
3267 	ret = -EINVAL;
3268 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3269 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3270 		goto out;
3271 	}
3272 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3273 
3274 	offset += skb->csum_offset;
3275 	if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3276 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3277 		goto out;
3278 	}
3279 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3280 	if (ret)
3281 		goto out;
3282 
3283 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3284 out_set_summed:
3285 	skb->ip_summed = CHECKSUM_NONE;
3286 out:
3287 	return ret;
3288 }
3289 EXPORT_SYMBOL(skb_checksum_help);
3290 
3291 int skb_crc32c_csum_help(struct sk_buff *skb)
3292 {
3293 	__le32 crc32c_csum;
3294 	int ret = 0, offset, start;
3295 
3296 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3297 		goto out;
3298 
3299 	if (unlikely(skb_is_gso(skb)))
3300 		goto out;
3301 
3302 	/* Before computing a checksum, we should make sure no frag could
3303 	 * be modified by an external entity : checksum could be wrong.
3304 	 */
3305 	if (unlikely(skb_has_shared_frag(skb))) {
3306 		ret = __skb_linearize(skb);
3307 		if (ret)
3308 			goto out;
3309 	}
3310 	start = skb_checksum_start_offset(skb);
3311 	offset = start + offsetof(struct sctphdr, checksum);
3312 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3313 		ret = -EINVAL;
3314 		goto out;
3315 	}
3316 
3317 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3318 	if (ret)
3319 		goto out;
3320 
3321 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3322 						  skb->len - start, ~(__u32)0,
3323 						  crc32c_csum_stub));
3324 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3325 	skb->ip_summed = CHECKSUM_NONE;
3326 	skb->csum_not_inet = 0;
3327 out:
3328 	return ret;
3329 }
3330 
3331 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3332 {
3333 	__be16 type = skb->protocol;
3334 
3335 	/* Tunnel gso handlers can set protocol to ethernet. */
3336 	if (type == htons(ETH_P_TEB)) {
3337 		struct ethhdr *eth;
3338 
3339 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3340 			return 0;
3341 
3342 		eth = (struct ethhdr *)skb->data;
3343 		type = eth->h_proto;
3344 	}
3345 
3346 	return __vlan_get_protocol(skb, type, depth);
3347 }
3348 
3349 /* openvswitch calls this on rx path, so we need a different check.
3350  */
3351 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3352 {
3353 	if (tx_path)
3354 		return skb->ip_summed != CHECKSUM_PARTIAL &&
3355 		       skb->ip_summed != CHECKSUM_UNNECESSARY;
3356 
3357 	return skb->ip_summed == CHECKSUM_NONE;
3358 }
3359 
3360 /**
3361  *	__skb_gso_segment - Perform segmentation on skb.
3362  *	@skb: buffer to segment
3363  *	@features: features for the output path (see dev->features)
3364  *	@tx_path: whether it is called in TX path
3365  *
3366  *	This function segments the given skb and returns a list of segments.
3367  *
3368  *	It may return NULL if the skb requires no segmentation.  This is
3369  *	only possible when GSO is used for verifying header integrity.
3370  *
3371  *	Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3372  */
3373 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3374 				  netdev_features_t features, bool tx_path)
3375 {
3376 	struct sk_buff *segs;
3377 
3378 	if (unlikely(skb_needs_check(skb, tx_path))) {
3379 		int err;
3380 
3381 		/* We're going to init ->check field in TCP or UDP header */
3382 		err = skb_cow_head(skb, 0);
3383 		if (err < 0)
3384 			return ERR_PTR(err);
3385 	}
3386 
3387 	/* Only report GSO partial support if it will enable us to
3388 	 * support segmentation on this frame without needing additional
3389 	 * work.
3390 	 */
3391 	if (features & NETIF_F_GSO_PARTIAL) {
3392 		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3393 		struct net_device *dev = skb->dev;
3394 
3395 		partial_features |= dev->features & dev->gso_partial_features;
3396 		if (!skb_gso_ok(skb, features | partial_features))
3397 			features &= ~NETIF_F_GSO_PARTIAL;
3398 	}
3399 
3400 	BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3401 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3402 
3403 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3404 	SKB_GSO_CB(skb)->encap_level = 0;
3405 
3406 	skb_reset_mac_header(skb);
3407 	skb_reset_mac_len(skb);
3408 
3409 	segs = skb_mac_gso_segment(skb, features);
3410 
3411 	if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3412 		skb_warn_bad_offload(skb);
3413 
3414 	return segs;
3415 }
3416 EXPORT_SYMBOL(__skb_gso_segment);
3417 
3418 /* Take action when hardware reception checksum errors are detected. */
3419 #ifdef CONFIG_BUG
3420 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3421 {
3422 	netdev_err(dev, "hw csum failure\n");
3423 	skb_dump(KERN_ERR, skb, true);
3424 	dump_stack();
3425 }
3426 
3427 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3428 {
3429 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3430 }
3431 EXPORT_SYMBOL(netdev_rx_csum_fault);
3432 #endif
3433 
3434 /* XXX: check that highmem exists at all on the given machine. */
3435 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3436 {
3437 #ifdef CONFIG_HIGHMEM
3438 	int i;
3439 
3440 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3441 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3442 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3443 
3444 			if (PageHighMem(skb_frag_page(frag)))
3445 				return 1;
3446 		}
3447 	}
3448 #endif
3449 	return 0;
3450 }
3451 
3452 /* If MPLS offload request, verify we are testing hardware MPLS features
3453  * instead of standard features for the netdev.
3454  */
3455 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3456 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3457 					   netdev_features_t features,
3458 					   __be16 type)
3459 {
3460 	if (eth_p_mpls(type))
3461 		features &= skb->dev->mpls_features;
3462 
3463 	return features;
3464 }
3465 #else
3466 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3467 					   netdev_features_t features,
3468 					   __be16 type)
3469 {
3470 	return features;
3471 }
3472 #endif
3473 
3474 static netdev_features_t harmonize_features(struct sk_buff *skb,
3475 	netdev_features_t features)
3476 {
3477 	__be16 type;
3478 
3479 	type = skb_network_protocol(skb, NULL);
3480 	features = net_mpls_features(skb, features, type);
3481 
3482 	if (skb->ip_summed != CHECKSUM_NONE &&
3483 	    !can_checksum_protocol(features, type)) {
3484 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3485 	}
3486 	if (illegal_highdma(skb->dev, skb))
3487 		features &= ~NETIF_F_SG;
3488 
3489 	return features;
3490 }
3491 
3492 netdev_features_t passthru_features_check(struct sk_buff *skb,
3493 					  struct net_device *dev,
3494 					  netdev_features_t features)
3495 {
3496 	return features;
3497 }
3498 EXPORT_SYMBOL(passthru_features_check);
3499 
3500 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3501 					     struct net_device *dev,
3502 					     netdev_features_t features)
3503 {
3504 	return vlan_features_check(skb, features);
3505 }
3506 
3507 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3508 					    struct net_device *dev,
3509 					    netdev_features_t features)
3510 {
3511 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3512 
3513 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3514 		return features & ~NETIF_F_GSO_MASK;
3515 
3516 	if (!skb_shinfo(skb)->gso_type) {
3517 		skb_warn_bad_offload(skb);
3518 		return features & ~NETIF_F_GSO_MASK;
3519 	}
3520 
3521 	/* Support for GSO partial features requires software
3522 	 * intervention before we can actually process the packets
3523 	 * so we need to strip support for any partial features now
3524 	 * and we can pull them back in after we have partially
3525 	 * segmented the frame.
3526 	 */
3527 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3528 		features &= ~dev->gso_partial_features;
3529 
3530 	/* Make sure to clear the IPv4 ID mangling feature if the
3531 	 * IPv4 header has the potential to be fragmented.
3532 	 */
3533 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3534 		struct iphdr *iph = skb->encapsulation ?
3535 				    inner_ip_hdr(skb) : ip_hdr(skb);
3536 
3537 		if (!(iph->frag_off & htons(IP_DF)))
3538 			features &= ~NETIF_F_TSO_MANGLEID;
3539 	}
3540 
3541 	return features;
3542 }
3543 
3544 netdev_features_t netif_skb_features(struct sk_buff *skb)
3545 {
3546 	struct net_device *dev = skb->dev;
3547 	netdev_features_t features = dev->features;
3548 
3549 	if (skb_is_gso(skb))
3550 		features = gso_features_check(skb, dev, features);
3551 
3552 	/* If encapsulation offload request, verify we are testing
3553 	 * hardware encapsulation features instead of standard
3554 	 * features for the netdev
3555 	 */
3556 	if (skb->encapsulation)
3557 		features &= dev->hw_enc_features;
3558 
3559 	if (skb_vlan_tagged(skb))
3560 		features = netdev_intersect_features(features,
3561 						     dev->vlan_features |
3562 						     NETIF_F_HW_VLAN_CTAG_TX |
3563 						     NETIF_F_HW_VLAN_STAG_TX);
3564 
3565 	if (dev->netdev_ops->ndo_features_check)
3566 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3567 								features);
3568 	else
3569 		features &= dflt_features_check(skb, dev, features);
3570 
3571 	return harmonize_features(skb, features);
3572 }
3573 EXPORT_SYMBOL(netif_skb_features);
3574 
3575 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3576 		    struct netdev_queue *txq, bool more)
3577 {
3578 	unsigned int len;
3579 	int rc;
3580 
3581 	if (dev_nit_active(dev))
3582 		dev_queue_xmit_nit(skb, dev);
3583 
3584 	len = skb->len;
3585 	trace_net_dev_start_xmit(skb, dev);
3586 	rc = netdev_start_xmit(skb, dev, txq, more);
3587 	trace_net_dev_xmit(skb, rc, dev, len);
3588 
3589 	return rc;
3590 }
3591 
3592 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3593 				    struct netdev_queue *txq, int *ret)
3594 {
3595 	struct sk_buff *skb = first;
3596 	int rc = NETDEV_TX_OK;
3597 
3598 	while (skb) {
3599 		struct sk_buff *next = skb->next;
3600 
3601 		skb_mark_not_on_list(skb);
3602 		rc = xmit_one(skb, dev, txq, next != NULL);
3603 		if (unlikely(!dev_xmit_complete(rc))) {
3604 			skb->next = next;
3605 			goto out;
3606 		}
3607 
3608 		skb = next;
3609 		if (netif_tx_queue_stopped(txq) && skb) {
3610 			rc = NETDEV_TX_BUSY;
3611 			break;
3612 		}
3613 	}
3614 
3615 out:
3616 	*ret = rc;
3617 	return skb;
3618 }
3619 
3620 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3621 					  netdev_features_t features)
3622 {
3623 	if (skb_vlan_tag_present(skb) &&
3624 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3625 		skb = __vlan_hwaccel_push_inside(skb);
3626 	return skb;
3627 }
3628 
3629 int skb_csum_hwoffload_help(struct sk_buff *skb,
3630 			    const netdev_features_t features)
3631 {
3632 	if (unlikely(skb_csum_is_sctp(skb)))
3633 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3634 			skb_crc32c_csum_help(skb);
3635 
3636 	if (features & NETIF_F_HW_CSUM)
3637 		return 0;
3638 
3639 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3640 		switch (skb->csum_offset) {
3641 		case offsetof(struct tcphdr, check):
3642 		case offsetof(struct udphdr, check):
3643 			return 0;
3644 		}
3645 	}
3646 
3647 	return skb_checksum_help(skb);
3648 }
3649 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3650 
3651 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3652 {
3653 	netdev_features_t features;
3654 
3655 	features = netif_skb_features(skb);
3656 	skb = validate_xmit_vlan(skb, features);
3657 	if (unlikely(!skb))
3658 		goto out_null;
3659 
3660 	skb = sk_validate_xmit_skb(skb, dev);
3661 	if (unlikely(!skb))
3662 		goto out_null;
3663 
3664 	if (netif_needs_gso(skb, features)) {
3665 		struct sk_buff *segs;
3666 
3667 		segs = skb_gso_segment(skb, features);
3668 		if (IS_ERR(segs)) {
3669 			goto out_kfree_skb;
3670 		} else if (segs) {
3671 			consume_skb(skb);
3672 			skb = segs;
3673 		}
3674 	} else {
3675 		if (skb_needs_linearize(skb, features) &&
3676 		    __skb_linearize(skb))
3677 			goto out_kfree_skb;
3678 
3679 		/* If packet is not checksummed and device does not
3680 		 * support checksumming for this protocol, complete
3681 		 * checksumming here.
3682 		 */
3683 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3684 			if (skb->encapsulation)
3685 				skb_set_inner_transport_header(skb,
3686 							       skb_checksum_start_offset(skb));
3687 			else
3688 				skb_set_transport_header(skb,
3689 							 skb_checksum_start_offset(skb));
3690 			if (skb_csum_hwoffload_help(skb, features))
3691 				goto out_kfree_skb;
3692 		}
3693 	}
3694 
3695 	skb = validate_xmit_xfrm(skb, features, again);
3696 
3697 	return skb;
3698 
3699 out_kfree_skb:
3700 	kfree_skb(skb);
3701 out_null:
3702 	dev_core_stats_tx_dropped_inc(dev);
3703 	return NULL;
3704 }
3705 
3706 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3707 {
3708 	struct sk_buff *next, *head = NULL, *tail;
3709 
3710 	for (; skb != NULL; skb = next) {
3711 		next = skb->next;
3712 		skb_mark_not_on_list(skb);
3713 
3714 		/* in case skb wont be segmented, point to itself */
3715 		skb->prev = skb;
3716 
3717 		skb = validate_xmit_skb(skb, dev, again);
3718 		if (!skb)
3719 			continue;
3720 
3721 		if (!head)
3722 			head = skb;
3723 		else
3724 			tail->next = skb;
3725 		/* If skb was segmented, skb->prev points to
3726 		 * the last segment. If not, it still contains skb.
3727 		 */
3728 		tail = skb->prev;
3729 	}
3730 	return head;
3731 }
3732 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3733 
3734 static void qdisc_pkt_len_init(struct sk_buff *skb)
3735 {
3736 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3737 
3738 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3739 
3740 	/* To get more precise estimation of bytes sent on wire,
3741 	 * we add to pkt_len the headers size of all segments
3742 	 */
3743 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3744 		unsigned int hdr_len;
3745 		u16 gso_segs = shinfo->gso_segs;
3746 
3747 		/* mac layer + network layer */
3748 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3749 
3750 		/* + transport layer */
3751 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3752 			const struct tcphdr *th;
3753 			struct tcphdr _tcphdr;
3754 
3755 			th = skb_header_pointer(skb, skb_transport_offset(skb),
3756 						sizeof(_tcphdr), &_tcphdr);
3757 			if (likely(th))
3758 				hdr_len += __tcp_hdrlen(th);
3759 		} else {
3760 			struct udphdr _udphdr;
3761 
3762 			if (skb_header_pointer(skb, skb_transport_offset(skb),
3763 					       sizeof(_udphdr), &_udphdr))
3764 				hdr_len += sizeof(struct udphdr);
3765 		}
3766 
3767 		if (shinfo->gso_type & SKB_GSO_DODGY)
3768 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3769 						shinfo->gso_size);
3770 
3771 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3772 	}
3773 }
3774 
3775 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3776 			     struct sk_buff **to_free,
3777 			     struct netdev_queue *txq)
3778 {
3779 	int rc;
3780 
3781 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3782 	if (rc == NET_XMIT_SUCCESS)
3783 		trace_qdisc_enqueue(q, txq, skb);
3784 	return rc;
3785 }
3786 
3787 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3788 				 struct net_device *dev,
3789 				 struct netdev_queue *txq)
3790 {
3791 	spinlock_t *root_lock = qdisc_lock(q);
3792 	struct sk_buff *to_free = NULL;
3793 	bool contended;
3794 	int rc;
3795 
3796 	qdisc_calculate_pkt_len(skb, q);
3797 
3798 	if (q->flags & TCQ_F_NOLOCK) {
3799 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3800 		    qdisc_run_begin(q)) {
3801 			/* Retest nolock_qdisc_is_empty() within the protection
3802 			 * of q->seqlock to protect from racing with requeuing.
3803 			 */
3804 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3805 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3806 				__qdisc_run(q);
3807 				qdisc_run_end(q);
3808 
3809 				goto no_lock_out;
3810 			}
3811 
3812 			qdisc_bstats_cpu_update(q, skb);
3813 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3814 			    !nolock_qdisc_is_empty(q))
3815 				__qdisc_run(q);
3816 
3817 			qdisc_run_end(q);
3818 			return NET_XMIT_SUCCESS;
3819 		}
3820 
3821 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3822 		qdisc_run(q);
3823 
3824 no_lock_out:
3825 		if (unlikely(to_free))
3826 			kfree_skb_list_reason(to_free,
3827 					      SKB_DROP_REASON_QDISC_DROP);
3828 		return rc;
3829 	}
3830 
3831 	/*
3832 	 * Heuristic to force contended enqueues to serialize on a
3833 	 * separate lock before trying to get qdisc main lock.
3834 	 * This permits qdisc->running owner to get the lock more
3835 	 * often and dequeue packets faster.
3836 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3837 	 * and then other tasks will only enqueue packets. The packets will be
3838 	 * sent after the qdisc owner is scheduled again. To prevent this
3839 	 * scenario the task always serialize on the lock.
3840 	 */
3841 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3842 	if (unlikely(contended))
3843 		spin_lock(&q->busylock);
3844 
3845 	spin_lock(root_lock);
3846 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3847 		__qdisc_drop(skb, &to_free);
3848 		rc = NET_XMIT_DROP;
3849 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3850 		   qdisc_run_begin(q)) {
3851 		/*
3852 		 * This is a work-conserving queue; there are no old skbs
3853 		 * waiting to be sent out; and the qdisc is not running -
3854 		 * xmit the skb directly.
3855 		 */
3856 
3857 		qdisc_bstats_update(q, skb);
3858 
3859 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3860 			if (unlikely(contended)) {
3861 				spin_unlock(&q->busylock);
3862 				contended = false;
3863 			}
3864 			__qdisc_run(q);
3865 		}
3866 
3867 		qdisc_run_end(q);
3868 		rc = NET_XMIT_SUCCESS;
3869 	} else {
3870 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3871 		if (qdisc_run_begin(q)) {
3872 			if (unlikely(contended)) {
3873 				spin_unlock(&q->busylock);
3874 				contended = false;
3875 			}
3876 			__qdisc_run(q);
3877 			qdisc_run_end(q);
3878 		}
3879 	}
3880 	spin_unlock(root_lock);
3881 	if (unlikely(to_free))
3882 		kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3883 	if (unlikely(contended))
3884 		spin_unlock(&q->busylock);
3885 	return rc;
3886 }
3887 
3888 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3889 static void skb_update_prio(struct sk_buff *skb)
3890 {
3891 	const struct netprio_map *map;
3892 	const struct sock *sk;
3893 	unsigned int prioidx;
3894 
3895 	if (skb->priority)
3896 		return;
3897 	map = rcu_dereference_bh(skb->dev->priomap);
3898 	if (!map)
3899 		return;
3900 	sk = skb_to_full_sk(skb);
3901 	if (!sk)
3902 		return;
3903 
3904 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3905 
3906 	if (prioidx < map->priomap_len)
3907 		skb->priority = map->priomap[prioidx];
3908 }
3909 #else
3910 #define skb_update_prio(skb)
3911 #endif
3912 
3913 /**
3914  *	dev_loopback_xmit - loop back @skb
3915  *	@net: network namespace this loopback is happening in
3916  *	@sk:  sk needed to be a netfilter okfn
3917  *	@skb: buffer to transmit
3918  */
3919 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3920 {
3921 	skb_reset_mac_header(skb);
3922 	__skb_pull(skb, skb_network_offset(skb));
3923 	skb->pkt_type = PACKET_LOOPBACK;
3924 	if (skb->ip_summed == CHECKSUM_NONE)
3925 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3926 	DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3927 	skb_dst_force(skb);
3928 	netif_rx(skb);
3929 	return 0;
3930 }
3931 EXPORT_SYMBOL(dev_loopback_xmit);
3932 
3933 #ifdef CONFIG_NET_EGRESS
3934 static struct sk_buff *
3935 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3936 {
3937 #ifdef CONFIG_NET_CLS_ACT
3938 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3939 	struct tcf_result cl_res;
3940 
3941 	if (!miniq)
3942 		return skb;
3943 
3944 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3945 	tc_skb_cb(skb)->mru = 0;
3946 	tc_skb_cb(skb)->post_ct = false;
3947 	mini_qdisc_bstats_cpu_update(miniq, skb);
3948 
3949 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3950 	case TC_ACT_OK:
3951 	case TC_ACT_RECLASSIFY:
3952 		skb->tc_index = TC_H_MIN(cl_res.classid);
3953 		break;
3954 	case TC_ACT_SHOT:
3955 		mini_qdisc_qstats_cpu_drop(miniq);
3956 		*ret = NET_XMIT_DROP;
3957 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3958 		return NULL;
3959 	case TC_ACT_STOLEN:
3960 	case TC_ACT_QUEUED:
3961 	case TC_ACT_TRAP:
3962 		*ret = NET_XMIT_SUCCESS;
3963 		consume_skb(skb);
3964 		return NULL;
3965 	case TC_ACT_REDIRECT:
3966 		/* No need to push/pop skb's mac_header here on egress! */
3967 		skb_do_redirect(skb);
3968 		*ret = NET_XMIT_SUCCESS;
3969 		return NULL;
3970 	default:
3971 		break;
3972 	}
3973 #endif /* CONFIG_NET_CLS_ACT */
3974 
3975 	return skb;
3976 }
3977 
3978 static struct netdev_queue *
3979 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3980 {
3981 	int qm = skb_get_queue_mapping(skb);
3982 
3983 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3984 }
3985 
3986 static bool netdev_xmit_txqueue_skipped(void)
3987 {
3988 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3989 }
3990 
3991 void netdev_xmit_skip_txqueue(bool skip)
3992 {
3993 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3994 }
3995 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3996 #endif /* CONFIG_NET_EGRESS */
3997 
3998 #ifdef CONFIG_XPS
3999 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4000 			       struct xps_dev_maps *dev_maps, unsigned int tci)
4001 {
4002 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
4003 	struct xps_map *map;
4004 	int queue_index = -1;
4005 
4006 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4007 		return queue_index;
4008 
4009 	tci *= dev_maps->num_tc;
4010 	tci += tc;
4011 
4012 	map = rcu_dereference(dev_maps->attr_map[tci]);
4013 	if (map) {
4014 		if (map->len == 1)
4015 			queue_index = map->queues[0];
4016 		else
4017 			queue_index = map->queues[reciprocal_scale(
4018 						skb_get_hash(skb), map->len)];
4019 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4020 			queue_index = -1;
4021 	}
4022 	return queue_index;
4023 }
4024 #endif
4025 
4026 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4027 			 struct sk_buff *skb)
4028 {
4029 #ifdef CONFIG_XPS
4030 	struct xps_dev_maps *dev_maps;
4031 	struct sock *sk = skb->sk;
4032 	int queue_index = -1;
4033 
4034 	if (!static_key_false(&xps_needed))
4035 		return -1;
4036 
4037 	rcu_read_lock();
4038 	if (!static_key_false(&xps_rxqs_needed))
4039 		goto get_cpus_map;
4040 
4041 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4042 	if (dev_maps) {
4043 		int tci = sk_rx_queue_get(sk);
4044 
4045 		if (tci >= 0)
4046 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4047 							  tci);
4048 	}
4049 
4050 get_cpus_map:
4051 	if (queue_index < 0) {
4052 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4053 		if (dev_maps) {
4054 			unsigned int tci = skb->sender_cpu - 1;
4055 
4056 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4057 							  tci);
4058 		}
4059 	}
4060 	rcu_read_unlock();
4061 
4062 	return queue_index;
4063 #else
4064 	return -1;
4065 #endif
4066 }
4067 
4068 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4069 		     struct net_device *sb_dev)
4070 {
4071 	return 0;
4072 }
4073 EXPORT_SYMBOL(dev_pick_tx_zero);
4074 
4075 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4076 		       struct net_device *sb_dev)
4077 {
4078 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4079 }
4080 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4081 
4082 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4083 		     struct net_device *sb_dev)
4084 {
4085 	struct sock *sk = skb->sk;
4086 	int queue_index = sk_tx_queue_get(sk);
4087 
4088 	sb_dev = sb_dev ? : dev;
4089 
4090 	if (queue_index < 0 || skb->ooo_okay ||
4091 	    queue_index >= dev->real_num_tx_queues) {
4092 		int new_index = get_xps_queue(dev, sb_dev, skb);
4093 
4094 		if (new_index < 0)
4095 			new_index = skb_tx_hash(dev, sb_dev, skb);
4096 
4097 		if (queue_index != new_index && sk &&
4098 		    sk_fullsock(sk) &&
4099 		    rcu_access_pointer(sk->sk_dst_cache))
4100 			sk_tx_queue_set(sk, new_index);
4101 
4102 		queue_index = new_index;
4103 	}
4104 
4105 	return queue_index;
4106 }
4107 EXPORT_SYMBOL(netdev_pick_tx);
4108 
4109 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4110 					 struct sk_buff *skb,
4111 					 struct net_device *sb_dev)
4112 {
4113 	int queue_index = 0;
4114 
4115 #ifdef CONFIG_XPS
4116 	u32 sender_cpu = skb->sender_cpu - 1;
4117 
4118 	if (sender_cpu >= (u32)NR_CPUS)
4119 		skb->sender_cpu = raw_smp_processor_id() + 1;
4120 #endif
4121 
4122 	if (dev->real_num_tx_queues != 1) {
4123 		const struct net_device_ops *ops = dev->netdev_ops;
4124 
4125 		if (ops->ndo_select_queue)
4126 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4127 		else
4128 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4129 
4130 		queue_index = netdev_cap_txqueue(dev, queue_index);
4131 	}
4132 
4133 	skb_set_queue_mapping(skb, queue_index);
4134 	return netdev_get_tx_queue(dev, queue_index);
4135 }
4136 
4137 /**
4138  * __dev_queue_xmit() - transmit a buffer
4139  * @skb:	buffer to transmit
4140  * @sb_dev:	suboordinate device used for L2 forwarding offload
4141  *
4142  * Queue a buffer for transmission to a network device. The caller must
4143  * have set the device and priority and built the buffer before calling
4144  * this function. The function can be called from an interrupt.
4145  *
4146  * When calling this method, interrupts MUST be enabled. This is because
4147  * the BH enable code must have IRQs enabled so that it will not deadlock.
4148  *
4149  * Regardless of the return value, the skb is consumed, so it is currently
4150  * difficult to retry a send to this method. (You can bump the ref count
4151  * before sending to hold a reference for retry if you are careful.)
4152  *
4153  * Return:
4154  * * 0				- buffer successfully transmitted
4155  * * positive qdisc return code	- NET_XMIT_DROP etc.
4156  * * negative errno		- other errors
4157  */
4158 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4159 {
4160 	struct net_device *dev = skb->dev;
4161 	struct netdev_queue *txq = NULL;
4162 	struct Qdisc *q;
4163 	int rc = -ENOMEM;
4164 	bool again = false;
4165 
4166 	skb_reset_mac_header(skb);
4167 	skb_assert_len(skb);
4168 
4169 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4170 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4171 
4172 	/* Disable soft irqs for various locks below. Also
4173 	 * stops preemption for RCU.
4174 	 */
4175 	rcu_read_lock_bh();
4176 
4177 	skb_update_prio(skb);
4178 
4179 	qdisc_pkt_len_init(skb);
4180 #ifdef CONFIG_NET_CLS_ACT
4181 	skb->tc_at_ingress = 0;
4182 #endif
4183 #ifdef CONFIG_NET_EGRESS
4184 	if (static_branch_unlikely(&egress_needed_key)) {
4185 		if (nf_hook_egress_active()) {
4186 			skb = nf_hook_egress(skb, &rc, dev);
4187 			if (!skb)
4188 				goto out;
4189 		}
4190 
4191 		netdev_xmit_skip_txqueue(false);
4192 
4193 		nf_skip_egress(skb, true);
4194 		skb = sch_handle_egress(skb, &rc, dev);
4195 		if (!skb)
4196 			goto out;
4197 		nf_skip_egress(skb, false);
4198 
4199 		if (netdev_xmit_txqueue_skipped())
4200 			txq = netdev_tx_queue_mapping(dev, skb);
4201 	}
4202 #endif
4203 	/* If device/qdisc don't need skb->dst, release it right now while
4204 	 * its hot in this cpu cache.
4205 	 */
4206 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4207 		skb_dst_drop(skb);
4208 	else
4209 		skb_dst_force(skb);
4210 
4211 	if (!txq)
4212 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4213 
4214 	q = rcu_dereference_bh(txq->qdisc);
4215 
4216 	trace_net_dev_queue(skb);
4217 	if (q->enqueue) {
4218 		rc = __dev_xmit_skb(skb, q, dev, txq);
4219 		goto out;
4220 	}
4221 
4222 	/* The device has no queue. Common case for software devices:
4223 	 * loopback, all the sorts of tunnels...
4224 
4225 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4226 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4227 	 * counters.)
4228 	 * However, it is possible, that they rely on protection
4229 	 * made by us here.
4230 
4231 	 * Check this and shot the lock. It is not prone from deadlocks.
4232 	 *Either shot noqueue qdisc, it is even simpler 8)
4233 	 */
4234 	if (dev->flags & IFF_UP) {
4235 		int cpu = smp_processor_id(); /* ok because BHs are off */
4236 
4237 		/* Other cpus might concurrently change txq->xmit_lock_owner
4238 		 * to -1 or to their cpu id, but not to our id.
4239 		 */
4240 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4241 			if (dev_xmit_recursion())
4242 				goto recursion_alert;
4243 
4244 			skb = validate_xmit_skb(skb, dev, &again);
4245 			if (!skb)
4246 				goto out;
4247 
4248 			HARD_TX_LOCK(dev, txq, cpu);
4249 
4250 			if (!netif_xmit_stopped(txq)) {
4251 				dev_xmit_recursion_inc();
4252 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4253 				dev_xmit_recursion_dec();
4254 				if (dev_xmit_complete(rc)) {
4255 					HARD_TX_UNLOCK(dev, txq);
4256 					goto out;
4257 				}
4258 			}
4259 			HARD_TX_UNLOCK(dev, txq);
4260 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4261 					     dev->name);
4262 		} else {
4263 			/* Recursion is detected! It is possible,
4264 			 * unfortunately
4265 			 */
4266 recursion_alert:
4267 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4268 					     dev->name);
4269 		}
4270 	}
4271 
4272 	rc = -ENETDOWN;
4273 	rcu_read_unlock_bh();
4274 
4275 	dev_core_stats_tx_dropped_inc(dev);
4276 	kfree_skb_list(skb);
4277 	return rc;
4278 out:
4279 	rcu_read_unlock_bh();
4280 	return rc;
4281 }
4282 EXPORT_SYMBOL(__dev_queue_xmit);
4283 
4284 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4285 {
4286 	struct net_device *dev = skb->dev;
4287 	struct sk_buff *orig_skb = skb;
4288 	struct netdev_queue *txq;
4289 	int ret = NETDEV_TX_BUSY;
4290 	bool again = false;
4291 
4292 	if (unlikely(!netif_running(dev) ||
4293 		     !netif_carrier_ok(dev)))
4294 		goto drop;
4295 
4296 	skb = validate_xmit_skb_list(skb, dev, &again);
4297 	if (skb != orig_skb)
4298 		goto drop;
4299 
4300 	skb_set_queue_mapping(skb, queue_id);
4301 	txq = skb_get_tx_queue(dev, skb);
4302 
4303 	local_bh_disable();
4304 
4305 	dev_xmit_recursion_inc();
4306 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4307 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4308 		ret = netdev_start_xmit(skb, dev, txq, false);
4309 	HARD_TX_UNLOCK(dev, txq);
4310 	dev_xmit_recursion_dec();
4311 
4312 	local_bh_enable();
4313 	return ret;
4314 drop:
4315 	dev_core_stats_tx_dropped_inc(dev);
4316 	kfree_skb_list(skb);
4317 	return NET_XMIT_DROP;
4318 }
4319 EXPORT_SYMBOL(__dev_direct_xmit);
4320 
4321 /*************************************************************************
4322  *			Receiver routines
4323  *************************************************************************/
4324 
4325 int netdev_max_backlog __read_mostly = 1000;
4326 EXPORT_SYMBOL(netdev_max_backlog);
4327 
4328 int netdev_tstamp_prequeue __read_mostly = 1;
4329 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4330 int netdev_budget __read_mostly = 300;
4331 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4332 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4333 int weight_p __read_mostly = 64;           /* old backlog weight */
4334 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4335 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4336 int dev_rx_weight __read_mostly = 64;
4337 int dev_tx_weight __read_mostly = 64;
4338 
4339 /* Called with irq disabled */
4340 static inline void ____napi_schedule(struct softnet_data *sd,
4341 				     struct napi_struct *napi)
4342 {
4343 	struct task_struct *thread;
4344 
4345 	lockdep_assert_irqs_disabled();
4346 
4347 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4348 		/* Paired with smp_mb__before_atomic() in
4349 		 * napi_enable()/dev_set_threaded().
4350 		 * Use READ_ONCE() to guarantee a complete
4351 		 * read on napi->thread. Only call
4352 		 * wake_up_process() when it's not NULL.
4353 		 */
4354 		thread = READ_ONCE(napi->thread);
4355 		if (thread) {
4356 			/* Avoid doing set_bit() if the thread is in
4357 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4358 			 * makes sure to proceed with napi polling
4359 			 * if the thread is explicitly woken from here.
4360 			 */
4361 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4362 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4363 			wake_up_process(thread);
4364 			return;
4365 		}
4366 	}
4367 
4368 	list_add_tail(&napi->poll_list, &sd->poll_list);
4369 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4370 }
4371 
4372 #ifdef CONFIG_RPS
4373 
4374 /* One global table that all flow-based protocols share. */
4375 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4376 EXPORT_SYMBOL(rps_sock_flow_table);
4377 u32 rps_cpu_mask __read_mostly;
4378 EXPORT_SYMBOL(rps_cpu_mask);
4379 
4380 struct static_key_false rps_needed __read_mostly;
4381 EXPORT_SYMBOL(rps_needed);
4382 struct static_key_false rfs_needed __read_mostly;
4383 EXPORT_SYMBOL(rfs_needed);
4384 
4385 static struct rps_dev_flow *
4386 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4387 	    struct rps_dev_flow *rflow, u16 next_cpu)
4388 {
4389 	if (next_cpu < nr_cpu_ids) {
4390 #ifdef CONFIG_RFS_ACCEL
4391 		struct netdev_rx_queue *rxqueue;
4392 		struct rps_dev_flow_table *flow_table;
4393 		struct rps_dev_flow *old_rflow;
4394 		u32 flow_id;
4395 		u16 rxq_index;
4396 		int rc;
4397 
4398 		/* Should we steer this flow to a different hardware queue? */
4399 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4400 		    !(dev->features & NETIF_F_NTUPLE))
4401 			goto out;
4402 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4403 		if (rxq_index == skb_get_rx_queue(skb))
4404 			goto out;
4405 
4406 		rxqueue = dev->_rx + rxq_index;
4407 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4408 		if (!flow_table)
4409 			goto out;
4410 		flow_id = skb_get_hash(skb) & flow_table->mask;
4411 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4412 							rxq_index, flow_id);
4413 		if (rc < 0)
4414 			goto out;
4415 		old_rflow = rflow;
4416 		rflow = &flow_table->flows[flow_id];
4417 		rflow->filter = rc;
4418 		if (old_rflow->filter == rflow->filter)
4419 			old_rflow->filter = RPS_NO_FILTER;
4420 	out:
4421 #endif
4422 		rflow->last_qtail =
4423 			per_cpu(softnet_data, next_cpu).input_queue_head;
4424 	}
4425 
4426 	rflow->cpu = next_cpu;
4427 	return rflow;
4428 }
4429 
4430 /*
4431  * get_rps_cpu is called from netif_receive_skb and returns the target
4432  * CPU from the RPS map of the receiving queue for a given skb.
4433  * rcu_read_lock must be held on entry.
4434  */
4435 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4436 		       struct rps_dev_flow **rflowp)
4437 {
4438 	const struct rps_sock_flow_table *sock_flow_table;
4439 	struct netdev_rx_queue *rxqueue = dev->_rx;
4440 	struct rps_dev_flow_table *flow_table;
4441 	struct rps_map *map;
4442 	int cpu = -1;
4443 	u32 tcpu;
4444 	u32 hash;
4445 
4446 	if (skb_rx_queue_recorded(skb)) {
4447 		u16 index = skb_get_rx_queue(skb);
4448 
4449 		if (unlikely(index >= dev->real_num_rx_queues)) {
4450 			WARN_ONCE(dev->real_num_rx_queues > 1,
4451 				  "%s received packet on queue %u, but number "
4452 				  "of RX queues is %u\n",
4453 				  dev->name, index, dev->real_num_rx_queues);
4454 			goto done;
4455 		}
4456 		rxqueue += index;
4457 	}
4458 
4459 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4460 
4461 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4462 	map = rcu_dereference(rxqueue->rps_map);
4463 	if (!flow_table && !map)
4464 		goto done;
4465 
4466 	skb_reset_network_header(skb);
4467 	hash = skb_get_hash(skb);
4468 	if (!hash)
4469 		goto done;
4470 
4471 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4472 	if (flow_table && sock_flow_table) {
4473 		struct rps_dev_flow *rflow;
4474 		u32 next_cpu;
4475 		u32 ident;
4476 
4477 		/* First check into global flow table if there is a match */
4478 		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4479 		if ((ident ^ hash) & ~rps_cpu_mask)
4480 			goto try_rps;
4481 
4482 		next_cpu = ident & rps_cpu_mask;
4483 
4484 		/* OK, now we know there is a match,
4485 		 * we can look at the local (per receive queue) flow table
4486 		 */
4487 		rflow = &flow_table->flows[hash & flow_table->mask];
4488 		tcpu = rflow->cpu;
4489 
4490 		/*
4491 		 * If the desired CPU (where last recvmsg was done) is
4492 		 * different from current CPU (one in the rx-queue flow
4493 		 * table entry), switch if one of the following holds:
4494 		 *   - Current CPU is unset (>= nr_cpu_ids).
4495 		 *   - Current CPU is offline.
4496 		 *   - The current CPU's queue tail has advanced beyond the
4497 		 *     last packet that was enqueued using this table entry.
4498 		 *     This guarantees that all previous packets for the flow
4499 		 *     have been dequeued, thus preserving in order delivery.
4500 		 */
4501 		if (unlikely(tcpu != next_cpu) &&
4502 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4503 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4504 		      rflow->last_qtail)) >= 0)) {
4505 			tcpu = next_cpu;
4506 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4507 		}
4508 
4509 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4510 			*rflowp = rflow;
4511 			cpu = tcpu;
4512 			goto done;
4513 		}
4514 	}
4515 
4516 try_rps:
4517 
4518 	if (map) {
4519 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4520 		if (cpu_online(tcpu)) {
4521 			cpu = tcpu;
4522 			goto done;
4523 		}
4524 	}
4525 
4526 done:
4527 	return cpu;
4528 }
4529 
4530 #ifdef CONFIG_RFS_ACCEL
4531 
4532 /**
4533  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4534  * @dev: Device on which the filter was set
4535  * @rxq_index: RX queue index
4536  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4537  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4538  *
4539  * Drivers that implement ndo_rx_flow_steer() should periodically call
4540  * this function for each installed filter and remove the filters for
4541  * which it returns %true.
4542  */
4543 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4544 			 u32 flow_id, u16 filter_id)
4545 {
4546 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4547 	struct rps_dev_flow_table *flow_table;
4548 	struct rps_dev_flow *rflow;
4549 	bool expire = true;
4550 	unsigned int cpu;
4551 
4552 	rcu_read_lock();
4553 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4554 	if (flow_table && flow_id <= flow_table->mask) {
4555 		rflow = &flow_table->flows[flow_id];
4556 		cpu = READ_ONCE(rflow->cpu);
4557 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4558 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4559 			   rflow->last_qtail) <
4560 		     (int)(10 * flow_table->mask)))
4561 			expire = false;
4562 	}
4563 	rcu_read_unlock();
4564 	return expire;
4565 }
4566 EXPORT_SYMBOL(rps_may_expire_flow);
4567 
4568 #endif /* CONFIG_RFS_ACCEL */
4569 
4570 /* Called from hardirq (IPI) context */
4571 static void rps_trigger_softirq(void *data)
4572 {
4573 	struct softnet_data *sd = data;
4574 
4575 	____napi_schedule(sd, &sd->backlog);
4576 	sd->received_rps++;
4577 }
4578 
4579 #endif /* CONFIG_RPS */
4580 
4581 /* Called from hardirq (IPI) context */
4582 static void trigger_rx_softirq(void *data)
4583 {
4584 	struct softnet_data *sd = data;
4585 
4586 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4587 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4588 }
4589 
4590 /*
4591  * Check if this softnet_data structure is another cpu one
4592  * If yes, queue it to our IPI list and return 1
4593  * If no, return 0
4594  */
4595 static int napi_schedule_rps(struct softnet_data *sd)
4596 {
4597 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4598 
4599 #ifdef CONFIG_RPS
4600 	if (sd != mysd) {
4601 		sd->rps_ipi_next = mysd->rps_ipi_list;
4602 		mysd->rps_ipi_list = sd;
4603 
4604 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4605 		return 1;
4606 	}
4607 #endif /* CONFIG_RPS */
4608 	__napi_schedule_irqoff(&mysd->backlog);
4609 	return 0;
4610 }
4611 
4612 #ifdef CONFIG_NET_FLOW_LIMIT
4613 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4614 #endif
4615 
4616 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4617 {
4618 #ifdef CONFIG_NET_FLOW_LIMIT
4619 	struct sd_flow_limit *fl;
4620 	struct softnet_data *sd;
4621 	unsigned int old_flow, new_flow;
4622 
4623 	if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4624 		return false;
4625 
4626 	sd = this_cpu_ptr(&softnet_data);
4627 
4628 	rcu_read_lock();
4629 	fl = rcu_dereference(sd->flow_limit);
4630 	if (fl) {
4631 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4632 		old_flow = fl->history[fl->history_head];
4633 		fl->history[fl->history_head] = new_flow;
4634 
4635 		fl->history_head++;
4636 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4637 
4638 		if (likely(fl->buckets[old_flow]))
4639 			fl->buckets[old_flow]--;
4640 
4641 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4642 			fl->count++;
4643 			rcu_read_unlock();
4644 			return true;
4645 		}
4646 	}
4647 	rcu_read_unlock();
4648 #endif
4649 	return false;
4650 }
4651 
4652 /*
4653  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4654  * queue (may be a remote CPU queue).
4655  */
4656 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4657 			      unsigned int *qtail)
4658 {
4659 	enum skb_drop_reason reason;
4660 	struct softnet_data *sd;
4661 	unsigned long flags;
4662 	unsigned int qlen;
4663 
4664 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
4665 	sd = &per_cpu(softnet_data, cpu);
4666 
4667 	rps_lock_irqsave(sd, &flags);
4668 	if (!netif_running(skb->dev))
4669 		goto drop;
4670 	qlen = skb_queue_len(&sd->input_pkt_queue);
4671 	if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4672 		if (qlen) {
4673 enqueue:
4674 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4675 			input_queue_tail_incr_save(sd, qtail);
4676 			rps_unlock_irq_restore(sd, &flags);
4677 			return NET_RX_SUCCESS;
4678 		}
4679 
4680 		/* Schedule NAPI for backlog device
4681 		 * We can use non atomic operation since we own the queue lock
4682 		 */
4683 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4684 			napi_schedule_rps(sd);
4685 		goto enqueue;
4686 	}
4687 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4688 
4689 drop:
4690 	sd->dropped++;
4691 	rps_unlock_irq_restore(sd, &flags);
4692 
4693 	dev_core_stats_rx_dropped_inc(skb->dev);
4694 	kfree_skb_reason(skb, reason);
4695 	return NET_RX_DROP;
4696 }
4697 
4698 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4699 {
4700 	struct net_device *dev = skb->dev;
4701 	struct netdev_rx_queue *rxqueue;
4702 
4703 	rxqueue = dev->_rx;
4704 
4705 	if (skb_rx_queue_recorded(skb)) {
4706 		u16 index = skb_get_rx_queue(skb);
4707 
4708 		if (unlikely(index >= dev->real_num_rx_queues)) {
4709 			WARN_ONCE(dev->real_num_rx_queues > 1,
4710 				  "%s received packet on queue %u, but number "
4711 				  "of RX queues is %u\n",
4712 				  dev->name, index, dev->real_num_rx_queues);
4713 
4714 			return rxqueue; /* Return first rxqueue */
4715 		}
4716 		rxqueue += index;
4717 	}
4718 	return rxqueue;
4719 }
4720 
4721 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4722 			     struct bpf_prog *xdp_prog)
4723 {
4724 	void *orig_data, *orig_data_end, *hard_start;
4725 	struct netdev_rx_queue *rxqueue;
4726 	bool orig_bcast, orig_host;
4727 	u32 mac_len, frame_sz;
4728 	__be16 orig_eth_type;
4729 	struct ethhdr *eth;
4730 	u32 metalen, act;
4731 	int off;
4732 
4733 	/* The XDP program wants to see the packet starting at the MAC
4734 	 * header.
4735 	 */
4736 	mac_len = skb->data - skb_mac_header(skb);
4737 	hard_start = skb->data - skb_headroom(skb);
4738 
4739 	/* SKB "head" area always have tailroom for skb_shared_info */
4740 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4741 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4742 
4743 	rxqueue = netif_get_rxqueue(skb);
4744 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4745 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4746 			 skb_headlen(skb) + mac_len, true);
4747 
4748 	orig_data_end = xdp->data_end;
4749 	orig_data = xdp->data;
4750 	eth = (struct ethhdr *)xdp->data;
4751 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4752 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4753 	orig_eth_type = eth->h_proto;
4754 
4755 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4756 
4757 	/* check if bpf_xdp_adjust_head was used */
4758 	off = xdp->data - orig_data;
4759 	if (off) {
4760 		if (off > 0)
4761 			__skb_pull(skb, off);
4762 		else if (off < 0)
4763 			__skb_push(skb, -off);
4764 
4765 		skb->mac_header += off;
4766 		skb_reset_network_header(skb);
4767 	}
4768 
4769 	/* check if bpf_xdp_adjust_tail was used */
4770 	off = xdp->data_end - orig_data_end;
4771 	if (off != 0) {
4772 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4773 		skb->len += off; /* positive on grow, negative on shrink */
4774 	}
4775 
4776 	/* check if XDP changed eth hdr such SKB needs update */
4777 	eth = (struct ethhdr *)xdp->data;
4778 	if ((orig_eth_type != eth->h_proto) ||
4779 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4780 						  skb->dev->dev_addr)) ||
4781 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4782 		__skb_push(skb, ETH_HLEN);
4783 		skb->pkt_type = PACKET_HOST;
4784 		skb->protocol = eth_type_trans(skb, skb->dev);
4785 	}
4786 
4787 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4788 	 * before calling us again on redirect path. We do not call do_redirect
4789 	 * as we leave that up to the caller.
4790 	 *
4791 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4792 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4793 	 */
4794 	switch (act) {
4795 	case XDP_REDIRECT:
4796 	case XDP_TX:
4797 		__skb_push(skb, mac_len);
4798 		break;
4799 	case XDP_PASS:
4800 		metalen = xdp->data - xdp->data_meta;
4801 		if (metalen)
4802 			skb_metadata_set(skb, metalen);
4803 		break;
4804 	}
4805 
4806 	return act;
4807 }
4808 
4809 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4810 				     struct xdp_buff *xdp,
4811 				     struct bpf_prog *xdp_prog)
4812 {
4813 	u32 act = XDP_DROP;
4814 
4815 	/* Reinjected packets coming from act_mirred or similar should
4816 	 * not get XDP generic processing.
4817 	 */
4818 	if (skb_is_redirected(skb))
4819 		return XDP_PASS;
4820 
4821 	/* XDP packets must be linear and must have sufficient headroom
4822 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4823 	 * native XDP provides, thus we need to do it here as well.
4824 	 */
4825 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4826 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4827 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4828 		int troom = skb->tail + skb->data_len - skb->end;
4829 
4830 		/* In case we have to go down the path and also linearize,
4831 		 * then lets do the pskb_expand_head() work just once here.
4832 		 */
4833 		if (pskb_expand_head(skb,
4834 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4835 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4836 			goto do_drop;
4837 		if (skb_linearize(skb))
4838 			goto do_drop;
4839 	}
4840 
4841 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4842 	switch (act) {
4843 	case XDP_REDIRECT:
4844 	case XDP_TX:
4845 	case XDP_PASS:
4846 		break;
4847 	default:
4848 		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4849 		fallthrough;
4850 	case XDP_ABORTED:
4851 		trace_xdp_exception(skb->dev, xdp_prog, act);
4852 		fallthrough;
4853 	case XDP_DROP:
4854 	do_drop:
4855 		kfree_skb(skb);
4856 		break;
4857 	}
4858 
4859 	return act;
4860 }
4861 
4862 /* When doing generic XDP we have to bypass the qdisc layer and the
4863  * network taps in order to match in-driver-XDP behavior. This also means
4864  * that XDP packets are able to starve other packets going through a qdisc,
4865  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4866  * queues, so they do not have this starvation issue.
4867  */
4868 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4869 {
4870 	struct net_device *dev = skb->dev;
4871 	struct netdev_queue *txq;
4872 	bool free_skb = true;
4873 	int cpu, rc;
4874 
4875 	txq = netdev_core_pick_tx(dev, skb, NULL);
4876 	cpu = smp_processor_id();
4877 	HARD_TX_LOCK(dev, txq, cpu);
4878 	if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4879 		rc = netdev_start_xmit(skb, dev, txq, 0);
4880 		if (dev_xmit_complete(rc))
4881 			free_skb = false;
4882 	}
4883 	HARD_TX_UNLOCK(dev, txq);
4884 	if (free_skb) {
4885 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4886 		dev_core_stats_tx_dropped_inc(dev);
4887 		kfree_skb(skb);
4888 	}
4889 }
4890 
4891 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4892 
4893 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4894 {
4895 	if (xdp_prog) {
4896 		struct xdp_buff xdp;
4897 		u32 act;
4898 		int err;
4899 
4900 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4901 		if (act != XDP_PASS) {
4902 			switch (act) {
4903 			case XDP_REDIRECT:
4904 				err = xdp_do_generic_redirect(skb->dev, skb,
4905 							      &xdp, xdp_prog);
4906 				if (err)
4907 					goto out_redir;
4908 				break;
4909 			case XDP_TX:
4910 				generic_xdp_tx(skb, xdp_prog);
4911 				break;
4912 			}
4913 			return XDP_DROP;
4914 		}
4915 	}
4916 	return XDP_PASS;
4917 out_redir:
4918 	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4919 	return XDP_DROP;
4920 }
4921 EXPORT_SYMBOL_GPL(do_xdp_generic);
4922 
4923 static int netif_rx_internal(struct sk_buff *skb)
4924 {
4925 	int ret;
4926 
4927 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4928 
4929 	trace_netif_rx(skb);
4930 
4931 #ifdef CONFIG_RPS
4932 	if (static_branch_unlikely(&rps_needed)) {
4933 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4934 		int cpu;
4935 
4936 		rcu_read_lock();
4937 
4938 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4939 		if (cpu < 0)
4940 			cpu = smp_processor_id();
4941 
4942 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4943 
4944 		rcu_read_unlock();
4945 	} else
4946 #endif
4947 	{
4948 		unsigned int qtail;
4949 
4950 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4951 	}
4952 	return ret;
4953 }
4954 
4955 /**
4956  *	__netif_rx	-	Slightly optimized version of netif_rx
4957  *	@skb: buffer to post
4958  *
4959  *	This behaves as netif_rx except that it does not disable bottom halves.
4960  *	As a result this function may only be invoked from the interrupt context
4961  *	(either hard or soft interrupt).
4962  */
4963 int __netif_rx(struct sk_buff *skb)
4964 {
4965 	int ret;
4966 
4967 	lockdep_assert_once(hardirq_count() | softirq_count());
4968 
4969 	trace_netif_rx_entry(skb);
4970 	ret = netif_rx_internal(skb);
4971 	trace_netif_rx_exit(ret);
4972 	return ret;
4973 }
4974 EXPORT_SYMBOL(__netif_rx);
4975 
4976 /**
4977  *	netif_rx	-	post buffer to the network code
4978  *	@skb: buffer to post
4979  *
4980  *	This function receives a packet from a device driver and queues it for
4981  *	the upper (protocol) levels to process via the backlog NAPI device. It
4982  *	always succeeds. The buffer may be dropped during processing for
4983  *	congestion control or by the protocol layers.
4984  *	The network buffer is passed via the backlog NAPI device. Modern NIC
4985  *	driver should use NAPI and GRO.
4986  *	This function can used from interrupt and from process context. The
4987  *	caller from process context must not disable interrupts before invoking
4988  *	this function.
4989  *
4990  *	return values:
4991  *	NET_RX_SUCCESS	(no congestion)
4992  *	NET_RX_DROP     (packet was dropped)
4993  *
4994  */
4995 int netif_rx(struct sk_buff *skb)
4996 {
4997 	bool need_bh_off = !(hardirq_count() | softirq_count());
4998 	int ret;
4999 
5000 	if (need_bh_off)
5001 		local_bh_disable();
5002 	trace_netif_rx_entry(skb);
5003 	ret = netif_rx_internal(skb);
5004 	trace_netif_rx_exit(ret);
5005 	if (need_bh_off)
5006 		local_bh_enable();
5007 	return ret;
5008 }
5009 EXPORT_SYMBOL(netif_rx);
5010 
5011 static __latent_entropy void net_tx_action(struct softirq_action *h)
5012 {
5013 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5014 
5015 	if (sd->completion_queue) {
5016 		struct sk_buff *clist;
5017 
5018 		local_irq_disable();
5019 		clist = sd->completion_queue;
5020 		sd->completion_queue = NULL;
5021 		local_irq_enable();
5022 
5023 		while (clist) {
5024 			struct sk_buff *skb = clist;
5025 
5026 			clist = clist->next;
5027 
5028 			WARN_ON(refcount_read(&skb->users));
5029 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5030 				trace_consume_skb(skb);
5031 			else
5032 				trace_kfree_skb(skb, net_tx_action,
5033 						SKB_DROP_REASON_NOT_SPECIFIED);
5034 
5035 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5036 				__kfree_skb(skb);
5037 			else
5038 				__kfree_skb_defer(skb);
5039 		}
5040 	}
5041 
5042 	if (sd->output_queue) {
5043 		struct Qdisc *head;
5044 
5045 		local_irq_disable();
5046 		head = sd->output_queue;
5047 		sd->output_queue = NULL;
5048 		sd->output_queue_tailp = &sd->output_queue;
5049 		local_irq_enable();
5050 
5051 		rcu_read_lock();
5052 
5053 		while (head) {
5054 			struct Qdisc *q = head;
5055 			spinlock_t *root_lock = NULL;
5056 
5057 			head = head->next_sched;
5058 
5059 			/* We need to make sure head->next_sched is read
5060 			 * before clearing __QDISC_STATE_SCHED
5061 			 */
5062 			smp_mb__before_atomic();
5063 
5064 			if (!(q->flags & TCQ_F_NOLOCK)) {
5065 				root_lock = qdisc_lock(q);
5066 				spin_lock(root_lock);
5067 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5068 						     &q->state))) {
5069 				/* There is a synchronize_net() between
5070 				 * STATE_DEACTIVATED flag being set and
5071 				 * qdisc_reset()/some_qdisc_is_busy() in
5072 				 * dev_deactivate(), so we can safely bail out
5073 				 * early here to avoid data race between
5074 				 * qdisc_deactivate() and some_qdisc_is_busy()
5075 				 * for lockless qdisc.
5076 				 */
5077 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5078 				continue;
5079 			}
5080 
5081 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5082 			qdisc_run(q);
5083 			if (root_lock)
5084 				spin_unlock(root_lock);
5085 		}
5086 
5087 		rcu_read_unlock();
5088 	}
5089 
5090 	xfrm_dev_backlog(sd);
5091 }
5092 
5093 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5094 /* This hook is defined here for ATM LANE */
5095 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5096 			     unsigned char *addr) __read_mostly;
5097 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5098 #endif
5099 
5100 static inline struct sk_buff *
5101 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5102 		   struct net_device *orig_dev, bool *another)
5103 {
5104 #ifdef CONFIG_NET_CLS_ACT
5105 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5106 	struct tcf_result cl_res;
5107 
5108 	/* If there's at least one ingress present somewhere (so
5109 	 * we get here via enabled static key), remaining devices
5110 	 * that are not configured with an ingress qdisc will bail
5111 	 * out here.
5112 	 */
5113 	if (!miniq)
5114 		return skb;
5115 
5116 	if (*pt_prev) {
5117 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5118 		*pt_prev = NULL;
5119 	}
5120 
5121 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5122 	tc_skb_cb(skb)->mru = 0;
5123 	tc_skb_cb(skb)->post_ct = false;
5124 	skb->tc_at_ingress = 1;
5125 	mini_qdisc_bstats_cpu_update(miniq, skb);
5126 
5127 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5128 	case TC_ACT_OK:
5129 	case TC_ACT_RECLASSIFY:
5130 		skb->tc_index = TC_H_MIN(cl_res.classid);
5131 		break;
5132 	case TC_ACT_SHOT:
5133 		mini_qdisc_qstats_cpu_drop(miniq);
5134 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5135 		*ret = NET_RX_DROP;
5136 		return NULL;
5137 	case TC_ACT_STOLEN:
5138 	case TC_ACT_QUEUED:
5139 	case TC_ACT_TRAP:
5140 		consume_skb(skb);
5141 		*ret = NET_RX_SUCCESS;
5142 		return NULL;
5143 	case TC_ACT_REDIRECT:
5144 		/* skb_mac_header check was done by cls/act_bpf, so
5145 		 * we can safely push the L2 header back before
5146 		 * redirecting to another netdev
5147 		 */
5148 		__skb_push(skb, skb->mac_len);
5149 		if (skb_do_redirect(skb) == -EAGAIN) {
5150 			__skb_pull(skb, skb->mac_len);
5151 			*another = true;
5152 			break;
5153 		}
5154 		*ret = NET_RX_SUCCESS;
5155 		return NULL;
5156 	case TC_ACT_CONSUMED:
5157 		*ret = NET_RX_SUCCESS;
5158 		return NULL;
5159 	default:
5160 		break;
5161 	}
5162 #endif /* CONFIG_NET_CLS_ACT */
5163 	return skb;
5164 }
5165 
5166 /**
5167  *	netdev_is_rx_handler_busy - check if receive handler is registered
5168  *	@dev: device to check
5169  *
5170  *	Check if a receive handler is already registered for a given device.
5171  *	Return true if there one.
5172  *
5173  *	The caller must hold the rtnl_mutex.
5174  */
5175 bool netdev_is_rx_handler_busy(struct net_device *dev)
5176 {
5177 	ASSERT_RTNL();
5178 	return dev && rtnl_dereference(dev->rx_handler);
5179 }
5180 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5181 
5182 /**
5183  *	netdev_rx_handler_register - register receive handler
5184  *	@dev: device to register a handler for
5185  *	@rx_handler: receive handler to register
5186  *	@rx_handler_data: data pointer that is used by rx handler
5187  *
5188  *	Register a receive handler for a device. This handler will then be
5189  *	called from __netif_receive_skb. A negative errno code is returned
5190  *	on a failure.
5191  *
5192  *	The caller must hold the rtnl_mutex.
5193  *
5194  *	For a general description of rx_handler, see enum rx_handler_result.
5195  */
5196 int netdev_rx_handler_register(struct net_device *dev,
5197 			       rx_handler_func_t *rx_handler,
5198 			       void *rx_handler_data)
5199 {
5200 	if (netdev_is_rx_handler_busy(dev))
5201 		return -EBUSY;
5202 
5203 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5204 		return -EINVAL;
5205 
5206 	/* Note: rx_handler_data must be set before rx_handler */
5207 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5208 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5209 
5210 	return 0;
5211 }
5212 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5213 
5214 /**
5215  *	netdev_rx_handler_unregister - unregister receive handler
5216  *	@dev: device to unregister a handler from
5217  *
5218  *	Unregister a receive handler from a device.
5219  *
5220  *	The caller must hold the rtnl_mutex.
5221  */
5222 void netdev_rx_handler_unregister(struct net_device *dev)
5223 {
5224 
5225 	ASSERT_RTNL();
5226 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5227 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5228 	 * section has a guarantee to see a non NULL rx_handler_data
5229 	 * as well.
5230 	 */
5231 	synchronize_net();
5232 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5233 }
5234 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5235 
5236 /*
5237  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5238  * the special handling of PFMEMALLOC skbs.
5239  */
5240 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5241 {
5242 	switch (skb->protocol) {
5243 	case htons(ETH_P_ARP):
5244 	case htons(ETH_P_IP):
5245 	case htons(ETH_P_IPV6):
5246 	case htons(ETH_P_8021Q):
5247 	case htons(ETH_P_8021AD):
5248 		return true;
5249 	default:
5250 		return false;
5251 	}
5252 }
5253 
5254 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5255 			     int *ret, struct net_device *orig_dev)
5256 {
5257 	if (nf_hook_ingress_active(skb)) {
5258 		int ingress_retval;
5259 
5260 		if (*pt_prev) {
5261 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5262 			*pt_prev = NULL;
5263 		}
5264 
5265 		rcu_read_lock();
5266 		ingress_retval = nf_hook_ingress(skb);
5267 		rcu_read_unlock();
5268 		return ingress_retval;
5269 	}
5270 	return 0;
5271 }
5272 
5273 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5274 				    struct packet_type **ppt_prev)
5275 {
5276 	struct packet_type *ptype, *pt_prev;
5277 	rx_handler_func_t *rx_handler;
5278 	struct sk_buff *skb = *pskb;
5279 	struct net_device *orig_dev;
5280 	bool deliver_exact = false;
5281 	int ret = NET_RX_DROP;
5282 	__be16 type;
5283 
5284 	net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5285 
5286 	trace_netif_receive_skb(skb);
5287 
5288 	orig_dev = skb->dev;
5289 
5290 	skb_reset_network_header(skb);
5291 	if (!skb_transport_header_was_set(skb))
5292 		skb_reset_transport_header(skb);
5293 	skb_reset_mac_len(skb);
5294 
5295 	pt_prev = NULL;
5296 
5297 another_round:
5298 	skb->skb_iif = skb->dev->ifindex;
5299 
5300 	__this_cpu_inc(softnet_data.processed);
5301 
5302 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5303 		int ret2;
5304 
5305 		migrate_disable();
5306 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5307 		migrate_enable();
5308 
5309 		if (ret2 != XDP_PASS) {
5310 			ret = NET_RX_DROP;
5311 			goto out;
5312 		}
5313 	}
5314 
5315 	if (eth_type_vlan(skb->protocol)) {
5316 		skb = skb_vlan_untag(skb);
5317 		if (unlikely(!skb))
5318 			goto out;
5319 	}
5320 
5321 	if (skb_skip_tc_classify(skb))
5322 		goto skip_classify;
5323 
5324 	if (pfmemalloc)
5325 		goto skip_taps;
5326 
5327 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5328 		if (pt_prev)
5329 			ret = deliver_skb(skb, pt_prev, orig_dev);
5330 		pt_prev = ptype;
5331 	}
5332 
5333 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5334 		if (pt_prev)
5335 			ret = deliver_skb(skb, pt_prev, orig_dev);
5336 		pt_prev = ptype;
5337 	}
5338 
5339 skip_taps:
5340 #ifdef CONFIG_NET_INGRESS
5341 	if (static_branch_unlikely(&ingress_needed_key)) {
5342 		bool another = false;
5343 
5344 		nf_skip_egress(skb, true);
5345 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5346 					 &another);
5347 		if (another)
5348 			goto another_round;
5349 		if (!skb)
5350 			goto out;
5351 
5352 		nf_skip_egress(skb, false);
5353 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5354 			goto out;
5355 	}
5356 #endif
5357 	skb_reset_redirect(skb);
5358 skip_classify:
5359 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5360 		goto drop;
5361 
5362 	if (skb_vlan_tag_present(skb)) {
5363 		if (pt_prev) {
5364 			ret = deliver_skb(skb, pt_prev, orig_dev);
5365 			pt_prev = NULL;
5366 		}
5367 		if (vlan_do_receive(&skb))
5368 			goto another_round;
5369 		else if (unlikely(!skb))
5370 			goto out;
5371 	}
5372 
5373 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5374 	if (rx_handler) {
5375 		if (pt_prev) {
5376 			ret = deliver_skb(skb, pt_prev, orig_dev);
5377 			pt_prev = NULL;
5378 		}
5379 		switch (rx_handler(&skb)) {
5380 		case RX_HANDLER_CONSUMED:
5381 			ret = NET_RX_SUCCESS;
5382 			goto out;
5383 		case RX_HANDLER_ANOTHER:
5384 			goto another_round;
5385 		case RX_HANDLER_EXACT:
5386 			deliver_exact = true;
5387 			break;
5388 		case RX_HANDLER_PASS:
5389 			break;
5390 		default:
5391 			BUG();
5392 		}
5393 	}
5394 
5395 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5396 check_vlan_id:
5397 		if (skb_vlan_tag_get_id(skb)) {
5398 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5399 			 * find vlan device.
5400 			 */
5401 			skb->pkt_type = PACKET_OTHERHOST;
5402 		} else if (eth_type_vlan(skb->protocol)) {
5403 			/* Outer header is 802.1P with vlan 0, inner header is
5404 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5405 			 * not find vlan dev for vlan id 0.
5406 			 */
5407 			__vlan_hwaccel_clear_tag(skb);
5408 			skb = skb_vlan_untag(skb);
5409 			if (unlikely(!skb))
5410 				goto out;
5411 			if (vlan_do_receive(&skb))
5412 				/* After stripping off 802.1P header with vlan 0
5413 				 * vlan dev is found for inner header.
5414 				 */
5415 				goto another_round;
5416 			else if (unlikely(!skb))
5417 				goto out;
5418 			else
5419 				/* We have stripped outer 802.1P vlan 0 header.
5420 				 * But could not find vlan dev.
5421 				 * check again for vlan id to set OTHERHOST.
5422 				 */
5423 				goto check_vlan_id;
5424 		}
5425 		/* Note: we might in the future use prio bits
5426 		 * and set skb->priority like in vlan_do_receive()
5427 		 * For the time being, just ignore Priority Code Point
5428 		 */
5429 		__vlan_hwaccel_clear_tag(skb);
5430 	}
5431 
5432 	type = skb->protocol;
5433 
5434 	/* deliver only exact match when indicated */
5435 	if (likely(!deliver_exact)) {
5436 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5437 				       &ptype_base[ntohs(type) &
5438 						   PTYPE_HASH_MASK]);
5439 	}
5440 
5441 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5442 			       &orig_dev->ptype_specific);
5443 
5444 	if (unlikely(skb->dev != orig_dev)) {
5445 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5446 				       &skb->dev->ptype_specific);
5447 	}
5448 
5449 	if (pt_prev) {
5450 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5451 			goto drop;
5452 		*ppt_prev = pt_prev;
5453 	} else {
5454 drop:
5455 		if (!deliver_exact)
5456 			dev_core_stats_rx_dropped_inc(skb->dev);
5457 		else
5458 			dev_core_stats_rx_nohandler_inc(skb->dev);
5459 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5460 		/* Jamal, now you will not able to escape explaining
5461 		 * me how you were going to use this. :-)
5462 		 */
5463 		ret = NET_RX_DROP;
5464 	}
5465 
5466 out:
5467 	/* The invariant here is that if *ppt_prev is not NULL
5468 	 * then skb should also be non-NULL.
5469 	 *
5470 	 * Apparently *ppt_prev assignment above holds this invariant due to
5471 	 * skb dereferencing near it.
5472 	 */
5473 	*pskb = skb;
5474 	return ret;
5475 }
5476 
5477 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5478 {
5479 	struct net_device *orig_dev = skb->dev;
5480 	struct packet_type *pt_prev = NULL;
5481 	int ret;
5482 
5483 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5484 	if (pt_prev)
5485 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5486 					 skb->dev, pt_prev, orig_dev);
5487 	return ret;
5488 }
5489 
5490 /**
5491  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5492  *	@skb: buffer to process
5493  *
5494  *	More direct receive version of netif_receive_skb().  It should
5495  *	only be used by callers that have a need to skip RPS and Generic XDP.
5496  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5497  *
5498  *	This function may only be called from softirq context and interrupts
5499  *	should be enabled.
5500  *
5501  *	Return values (usually ignored):
5502  *	NET_RX_SUCCESS: no congestion
5503  *	NET_RX_DROP: packet was dropped
5504  */
5505 int netif_receive_skb_core(struct sk_buff *skb)
5506 {
5507 	int ret;
5508 
5509 	rcu_read_lock();
5510 	ret = __netif_receive_skb_one_core(skb, false);
5511 	rcu_read_unlock();
5512 
5513 	return ret;
5514 }
5515 EXPORT_SYMBOL(netif_receive_skb_core);
5516 
5517 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5518 						  struct packet_type *pt_prev,
5519 						  struct net_device *orig_dev)
5520 {
5521 	struct sk_buff *skb, *next;
5522 
5523 	if (!pt_prev)
5524 		return;
5525 	if (list_empty(head))
5526 		return;
5527 	if (pt_prev->list_func != NULL)
5528 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5529 				   ip_list_rcv, head, pt_prev, orig_dev);
5530 	else
5531 		list_for_each_entry_safe(skb, next, head, list) {
5532 			skb_list_del_init(skb);
5533 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5534 		}
5535 }
5536 
5537 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5538 {
5539 	/* Fast-path assumptions:
5540 	 * - There is no RX handler.
5541 	 * - Only one packet_type matches.
5542 	 * If either of these fails, we will end up doing some per-packet
5543 	 * processing in-line, then handling the 'last ptype' for the whole
5544 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5545 	 * because the 'last ptype' must be constant across the sublist, and all
5546 	 * other ptypes are handled per-packet.
5547 	 */
5548 	/* Current (common) ptype of sublist */
5549 	struct packet_type *pt_curr = NULL;
5550 	/* Current (common) orig_dev of sublist */
5551 	struct net_device *od_curr = NULL;
5552 	struct list_head sublist;
5553 	struct sk_buff *skb, *next;
5554 
5555 	INIT_LIST_HEAD(&sublist);
5556 	list_for_each_entry_safe(skb, next, head, list) {
5557 		struct net_device *orig_dev = skb->dev;
5558 		struct packet_type *pt_prev = NULL;
5559 
5560 		skb_list_del_init(skb);
5561 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5562 		if (!pt_prev)
5563 			continue;
5564 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5565 			/* dispatch old sublist */
5566 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5567 			/* start new sublist */
5568 			INIT_LIST_HEAD(&sublist);
5569 			pt_curr = pt_prev;
5570 			od_curr = orig_dev;
5571 		}
5572 		list_add_tail(&skb->list, &sublist);
5573 	}
5574 
5575 	/* dispatch final sublist */
5576 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5577 }
5578 
5579 static int __netif_receive_skb(struct sk_buff *skb)
5580 {
5581 	int ret;
5582 
5583 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5584 		unsigned int noreclaim_flag;
5585 
5586 		/*
5587 		 * PFMEMALLOC skbs are special, they should
5588 		 * - be delivered to SOCK_MEMALLOC sockets only
5589 		 * - stay away from userspace
5590 		 * - have bounded memory usage
5591 		 *
5592 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5593 		 * context down to all allocation sites.
5594 		 */
5595 		noreclaim_flag = memalloc_noreclaim_save();
5596 		ret = __netif_receive_skb_one_core(skb, true);
5597 		memalloc_noreclaim_restore(noreclaim_flag);
5598 	} else
5599 		ret = __netif_receive_skb_one_core(skb, false);
5600 
5601 	return ret;
5602 }
5603 
5604 static void __netif_receive_skb_list(struct list_head *head)
5605 {
5606 	unsigned long noreclaim_flag = 0;
5607 	struct sk_buff *skb, *next;
5608 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5609 
5610 	list_for_each_entry_safe(skb, next, head, list) {
5611 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5612 			struct list_head sublist;
5613 
5614 			/* Handle the previous sublist */
5615 			list_cut_before(&sublist, head, &skb->list);
5616 			if (!list_empty(&sublist))
5617 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5618 			pfmemalloc = !pfmemalloc;
5619 			/* See comments in __netif_receive_skb */
5620 			if (pfmemalloc)
5621 				noreclaim_flag = memalloc_noreclaim_save();
5622 			else
5623 				memalloc_noreclaim_restore(noreclaim_flag);
5624 		}
5625 	}
5626 	/* Handle the remaining sublist */
5627 	if (!list_empty(head))
5628 		__netif_receive_skb_list_core(head, pfmemalloc);
5629 	/* Restore pflags */
5630 	if (pfmemalloc)
5631 		memalloc_noreclaim_restore(noreclaim_flag);
5632 }
5633 
5634 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5635 {
5636 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5637 	struct bpf_prog *new = xdp->prog;
5638 	int ret = 0;
5639 
5640 	switch (xdp->command) {
5641 	case XDP_SETUP_PROG:
5642 		rcu_assign_pointer(dev->xdp_prog, new);
5643 		if (old)
5644 			bpf_prog_put(old);
5645 
5646 		if (old && !new) {
5647 			static_branch_dec(&generic_xdp_needed_key);
5648 		} else if (new && !old) {
5649 			static_branch_inc(&generic_xdp_needed_key);
5650 			dev_disable_lro(dev);
5651 			dev_disable_gro_hw(dev);
5652 		}
5653 		break;
5654 
5655 	default:
5656 		ret = -EINVAL;
5657 		break;
5658 	}
5659 
5660 	return ret;
5661 }
5662 
5663 static int netif_receive_skb_internal(struct sk_buff *skb)
5664 {
5665 	int ret;
5666 
5667 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5668 
5669 	if (skb_defer_rx_timestamp(skb))
5670 		return NET_RX_SUCCESS;
5671 
5672 	rcu_read_lock();
5673 #ifdef CONFIG_RPS
5674 	if (static_branch_unlikely(&rps_needed)) {
5675 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5676 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5677 
5678 		if (cpu >= 0) {
5679 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5680 			rcu_read_unlock();
5681 			return ret;
5682 		}
5683 	}
5684 #endif
5685 	ret = __netif_receive_skb(skb);
5686 	rcu_read_unlock();
5687 	return ret;
5688 }
5689 
5690 void netif_receive_skb_list_internal(struct list_head *head)
5691 {
5692 	struct sk_buff *skb, *next;
5693 	struct list_head sublist;
5694 
5695 	INIT_LIST_HEAD(&sublist);
5696 	list_for_each_entry_safe(skb, next, head, list) {
5697 		net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5698 		skb_list_del_init(skb);
5699 		if (!skb_defer_rx_timestamp(skb))
5700 			list_add_tail(&skb->list, &sublist);
5701 	}
5702 	list_splice_init(&sublist, head);
5703 
5704 	rcu_read_lock();
5705 #ifdef CONFIG_RPS
5706 	if (static_branch_unlikely(&rps_needed)) {
5707 		list_for_each_entry_safe(skb, next, head, list) {
5708 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5709 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5710 
5711 			if (cpu >= 0) {
5712 				/* Will be handled, remove from list */
5713 				skb_list_del_init(skb);
5714 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5715 			}
5716 		}
5717 	}
5718 #endif
5719 	__netif_receive_skb_list(head);
5720 	rcu_read_unlock();
5721 }
5722 
5723 /**
5724  *	netif_receive_skb - process receive buffer from network
5725  *	@skb: buffer to process
5726  *
5727  *	netif_receive_skb() is the main receive data processing function.
5728  *	It always succeeds. The buffer may be dropped during processing
5729  *	for congestion control or by the protocol layers.
5730  *
5731  *	This function may only be called from softirq context and interrupts
5732  *	should be enabled.
5733  *
5734  *	Return values (usually ignored):
5735  *	NET_RX_SUCCESS: no congestion
5736  *	NET_RX_DROP: packet was dropped
5737  */
5738 int netif_receive_skb(struct sk_buff *skb)
5739 {
5740 	int ret;
5741 
5742 	trace_netif_receive_skb_entry(skb);
5743 
5744 	ret = netif_receive_skb_internal(skb);
5745 	trace_netif_receive_skb_exit(ret);
5746 
5747 	return ret;
5748 }
5749 EXPORT_SYMBOL(netif_receive_skb);
5750 
5751 /**
5752  *	netif_receive_skb_list - process many receive buffers from network
5753  *	@head: list of skbs to process.
5754  *
5755  *	Since return value of netif_receive_skb() is normally ignored, and
5756  *	wouldn't be meaningful for a list, this function returns void.
5757  *
5758  *	This function may only be called from softirq context and interrupts
5759  *	should be enabled.
5760  */
5761 void netif_receive_skb_list(struct list_head *head)
5762 {
5763 	struct sk_buff *skb;
5764 
5765 	if (list_empty(head))
5766 		return;
5767 	if (trace_netif_receive_skb_list_entry_enabled()) {
5768 		list_for_each_entry(skb, head, list)
5769 			trace_netif_receive_skb_list_entry(skb);
5770 	}
5771 	netif_receive_skb_list_internal(head);
5772 	trace_netif_receive_skb_list_exit(0);
5773 }
5774 EXPORT_SYMBOL(netif_receive_skb_list);
5775 
5776 static DEFINE_PER_CPU(struct work_struct, flush_works);
5777 
5778 /* Network device is going away, flush any packets still pending */
5779 static void flush_backlog(struct work_struct *work)
5780 {
5781 	struct sk_buff *skb, *tmp;
5782 	struct softnet_data *sd;
5783 
5784 	local_bh_disable();
5785 	sd = this_cpu_ptr(&softnet_data);
5786 
5787 	rps_lock_irq_disable(sd);
5788 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5789 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5790 			__skb_unlink(skb, &sd->input_pkt_queue);
5791 			dev_kfree_skb_irq(skb);
5792 			input_queue_head_incr(sd);
5793 		}
5794 	}
5795 	rps_unlock_irq_enable(sd);
5796 
5797 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5798 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5799 			__skb_unlink(skb, &sd->process_queue);
5800 			kfree_skb(skb);
5801 			input_queue_head_incr(sd);
5802 		}
5803 	}
5804 	local_bh_enable();
5805 }
5806 
5807 static bool flush_required(int cpu)
5808 {
5809 #if IS_ENABLED(CONFIG_RPS)
5810 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5811 	bool do_flush;
5812 
5813 	rps_lock_irq_disable(sd);
5814 
5815 	/* as insertion into process_queue happens with the rps lock held,
5816 	 * process_queue access may race only with dequeue
5817 	 */
5818 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5819 		   !skb_queue_empty_lockless(&sd->process_queue);
5820 	rps_unlock_irq_enable(sd);
5821 
5822 	return do_flush;
5823 #endif
5824 	/* without RPS we can't safely check input_pkt_queue: during a
5825 	 * concurrent remote skb_queue_splice() we can detect as empty both
5826 	 * input_pkt_queue and process_queue even if the latter could end-up
5827 	 * containing a lot of packets.
5828 	 */
5829 	return true;
5830 }
5831 
5832 static void flush_all_backlogs(void)
5833 {
5834 	static cpumask_t flush_cpus;
5835 	unsigned int cpu;
5836 
5837 	/* since we are under rtnl lock protection we can use static data
5838 	 * for the cpumask and avoid allocating on stack the possibly
5839 	 * large mask
5840 	 */
5841 	ASSERT_RTNL();
5842 
5843 	cpus_read_lock();
5844 
5845 	cpumask_clear(&flush_cpus);
5846 	for_each_online_cpu(cpu) {
5847 		if (flush_required(cpu)) {
5848 			queue_work_on(cpu, system_highpri_wq,
5849 				      per_cpu_ptr(&flush_works, cpu));
5850 			cpumask_set_cpu(cpu, &flush_cpus);
5851 		}
5852 	}
5853 
5854 	/* we can have in flight packet[s] on the cpus we are not flushing,
5855 	 * synchronize_net() in unregister_netdevice_many() will take care of
5856 	 * them
5857 	 */
5858 	for_each_cpu(cpu, &flush_cpus)
5859 		flush_work(per_cpu_ptr(&flush_works, cpu));
5860 
5861 	cpus_read_unlock();
5862 }
5863 
5864 static void net_rps_send_ipi(struct softnet_data *remsd)
5865 {
5866 #ifdef CONFIG_RPS
5867 	while (remsd) {
5868 		struct softnet_data *next = remsd->rps_ipi_next;
5869 
5870 		if (cpu_online(remsd->cpu))
5871 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
5872 		remsd = next;
5873 	}
5874 #endif
5875 }
5876 
5877 /*
5878  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5879  * Note: called with local irq disabled, but exits with local irq enabled.
5880  */
5881 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5882 {
5883 #ifdef CONFIG_RPS
5884 	struct softnet_data *remsd = sd->rps_ipi_list;
5885 
5886 	if (remsd) {
5887 		sd->rps_ipi_list = NULL;
5888 
5889 		local_irq_enable();
5890 
5891 		/* Send pending IPI's to kick RPS processing on remote cpus. */
5892 		net_rps_send_ipi(remsd);
5893 	} else
5894 #endif
5895 		local_irq_enable();
5896 }
5897 
5898 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5899 {
5900 #ifdef CONFIG_RPS
5901 	return sd->rps_ipi_list != NULL;
5902 #else
5903 	return false;
5904 #endif
5905 }
5906 
5907 static int process_backlog(struct napi_struct *napi, int quota)
5908 {
5909 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5910 	bool again = true;
5911 	int work = 0;
5912 
5913 	/* Check if we have pending ipi, its better to send them now,
5914 	 * not waiting net_rx_action() end.
5915 	 */
5916 	if (sd_has_rps_ipi_waiting(sd)) {
5917 		local_irq_disable();
5918 		net_rps_action_and_irq_enable(sd);
5919 	}
5920 
5921 	napi->weight = READ_ONCE(dev_rx_weight);
5922 	while (again) {
5923 		struct sk_buff *skb;
5924 
5925 		while ((skb = __skb_dequeue(&sd->process_queue))) {
5926 			rcu_read_lock();
5927 			__netif_receive_skb(skb);
5928 			rcu_read_unlock();
5929 			input_queue_head_incr(sd);
5930 			if (++work >= quota)
5931 				return work;
5932 
5933 		}
5934 
5935 		rps_lock_irq_disable(sd);
5936 		if (skb_queue_empty(&sd->input_pkt_queue)) {
5937 			/*
5938 			 * Inline a custom version of __napi_complete().
5939 			 * only current cpu owns and manipulates this napi,
5940 			 * and NAPI_STATE_SCHED is the only possible flag set
5941 			 * on backlog.
5942 			 * We can use a plain write instead of clear_bit(),
5943 			 * and we dont need an smp_mb() memory barrier.
5944 			 */
5945 			napi->state = 0;
5946 			again = false;
5947 		} else {
5948 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
5949 						   &sd->process_queue);
5950 		}
5951 		rps_unlock_irq_enable(sd);
5952 	}
5953 
5954 	return work;
5955 }
5956 
5957 /**
5958  * __napi_schedule - schedule for receive
5959  * @n: entry to schedule
5960  *
5961  * The entry's receive function will be scheduled to run.
5962  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5963  */
5964 void __napi_schedule(struct napi_struct *n)
5965 {
5966 	unsigned long flags;
5967 
5968 	local_irq_save(flags);
5969 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5970 	local_irq_restore(flags);
5971 }
5972 EXPORT_SYMBOL(__napi_schedule);
5973 
5974 /**
5975  *	napi_schedule_prep - check if napi can be scheduled
5976  *	@n: napi context
5977  *
5978  * Test if NAPI routine is already running, and if not mark
5979  * it as running.  This is used as a condition variable to
5980  * insure only one NAPI poll instance runs.  We also make
5981  * sure there is no pending NAPI disable.
5982  */
5983 bool napi_schedule_prep(struct napi_struct *n)
5984 {
5985 	unsigned long new, val = READ_ONCE(n->state);
5986 
5987 	do {
5988 		if (unlikely(val & NAPIF_STATE_DISABLE))
5989 			return false;
5990 		new = val | NAPIF_STATE_SCHED;
5991 
5992 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
5993 		 * This was suggested by Alexander Duyck, as compiler
5994 		 * emits better code than :
5995 		 * if (val & NAPIF_STATE_SCHED)
5996 		 *     new |= NAPIF_STATE_MISSED;
5997 		 */
5998 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5999 						   NAPIF_STATE_MISSED;
6000 	} while (!try_cmpxchg(&n->state, &val, new));
6001 
6002 	return !(val & NAPIF_STATE_SCHED);
6003 }
6004 EXPORT_SYMBOL(napi_schedule_prep);
6005 
6006 /**
6007  * __napi_schedule_irqoff - schedule for receive
6008  * @n: entry to schedule
6009  *
6010  * Variant of __napi_schedule() assuming hard irqs are masked.
6011  *
6012  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6013  * because the interrupt disabled assumption might not be true
6014  * due to force-threaded interrupts and spinlock substitution.
6015  */
6016 void __napi_schedule_irqoff(struct napi_struct *n)
6017 {
6018 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6019 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6020 	else
6021 		__napi_schedule(n);
6022 }
6023 EXPORT_SYMBOL(__napi_schedule_irqoff);
6024 
6025 bool napi_complete_done(struct napi_struct *n, int work_done)
6026 {
6027 	unsigned long flags, val, new, timeout = 0;
6028 	bool ret = true;
6029 
6030 	/*
6031 	 * 1) Don't let napi dequeue from the cpu poll list
6032 	 *    just in case its running on a different cpu.
6033 	 * 2) If we are busy polling, do nothing here, we have
6034 	 *    the guarantee we will be called later.
6035 	 */
6036 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6037 				 NAPIF_STATE_IN_BUSY_POLL)))
6038 		return false;
6039 
6040 	if (work_done) {
6041 		if (n->gro_bitmask)
6042 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6043 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6044 	}
6045 	if (n->defer_hard_irqs_count > 0) {
6046 		n->defer_hard_irqs_count--;
6047 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6048 		if (timeout)
6049 			ret = false;
6050 	}
6051 	if (n->gro_bitmask) {
6052 		/* When the NAPI instance uses a timeout and keeps postponing
6053 		 * it, we need to bound somehow the time packets are kept in
6054 		 * the GRO layer
6055 		 */
6056 		napi_gro_flush(n, !!timeout);
6057 	}
6058 
6059 	gro_normal_list(n);
6060 
6061 	if (unlikely(!list_empty(&n->poll_list))) {
6062 		/* If n->poll_list is not empty, we need to mask irqs */
6063 		local_irq_save(flags);
6064 		list_del_init(&n->poll_list);
6065 		local_irq_restore(flags);
6066 	}
6067 
6068 	val = READ_ONCE(n->state);
6069 	do {
6070 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6071 
6072 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6073 			      NAPIF_STATE_SCHED_THREADED |
6074 			      NAPIF_STATE_PREFER_BUSY_POLL);
6075 
6076 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6077 		 * because we will call napi->poll() one more time.
6078 		 * This C code was suggested by Alexander Duyck to help gcc.
6079 		 */
6080 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6081 						    NAPIF_STATE_SCHED;
6082 	} while (!try_cmpxchg(&n->state, &val, new));
6083 
6084 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6085 		__napi_schedule(n);
6086 		return false;
6087 	}
6088 
6089 	if (timeout)
6090 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6091 			      HRTIMER_MODE_REL_PINNED);
6092 	return ret;
6093 }
6094 EXPORT_SYMBOL(napi_complete_done);
6095 
6096 /* must be called under rcu_read_lock(), as we dont take a reference */
6097 static struct napi_struct *napi_by_id(unsigned int napi_id)
6098 {
6099 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6100 	struct napi_struct *napi;
6101 
6102 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6103 		if (napi->napi_id == napi_id)
6104 			return napi;
6105 
6106 	return NULL;
6107 }
6108 
6109 #if defined(CONFIG_NET_RX_BUSY_POLL)
6110 
6111 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6112 {
6113 	if (!skip_schedule) {
6114 		gro_normal_list(napi);
6115 		__napi_schedule(napi);
6116 		return;
6117 	}
6118 
6119 	if (napi->gro_bitmask) {
6120 		/* flush too old packets
6121 		 * If HZ < 1000, flush all packets.
6122 		 */
6123 		napi_gro_flush(napi, HZ >= 1000);
6124 	}
6125 
6126 	gro_normal_list(napi);
6127 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6128 }
6129 
6130 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6131 			   u16 budget)
6132 {
6133 	bool skip_schedule = false;
6134 	unsigned long timeout;
6135 	int rc;
6136 
6137 	/* Busy polling means there is a high chance device driver hard irq
6138 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6139 	 * set in napi_schedule_prep().
6140 	 * Since we are about to call napi->poll() once more, we can safely
6141 	 * clear NAPI_STATE_MISSED.
6142 	 *
6143 	 * Note: x86 could use a single "lock and ..." instruction
6144 	 * to perform these two clear_bit()
6145 	 */
6146 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6147 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6148 
6149 	local_bh_disable();
6150 
6151 	if (prefer_busy_poll) {
6152 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6153 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6154 		if (napi->defer_hard_irqs_count && timeout) {
6155 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6156 			skip_schedule = true;
6157 		}
6158 	}
6159 
6160 	/* All we really want here is to re-enable device interrupts.
6161 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6162 	 */
6163 	rc = napi->poll(napi, budget);
6164 	/* We can't gro_normal_list() here, because napi->poll() might have
6165 	 * rearmed the napi (napi_complete_done()) in which case it could
6166 	 * already be running on another CPU.
6167 	 */
6168 	trace_napi_poll(napi, rc, budget);
6169 	netpoll_poll_unlock(have_poll_lock);
6170 	if (rc == budget)
6171 		__busy_poll_stop(napi, skip_schedule);
6172 	local_bh_enable();
6173 }
6174 
6175 void napi_busy_loop(unsigned int napi_id,
6176 		    bool (*loop_end)(void *, unsigned long),
6177 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6178 {
6179 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6180 	int (*napi_poll)(struct napi_struct *napi, int budget);
6181 	void *have_poll_lock = NULL;
6182 	struct napi_struct *napi;
6183 
6184 restart:
6185 	napi_poll = NULL;
6186 
6187 	rcu_read_lock();
6188 
6189 	napi = napi_by_id(napi_id);
6190 	if (!napi)
6191 		goto out;
6192 
6193 	preempt_disable();
6194 	for (;;) {
6195 		int work = 0;
6196 
6197 		local_bh_disable();
6198 		if (!napi_poll) {
6199 			unsigned long val = READ_ONCE(napi->state);
6200 
6201 			/* If multiple threads are competing for this napi,
6202 			 * we avoid dirtying napi->state as much as we can.
6203 			 */
6204 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6205 				   NAPIF_STATE_IN_BUSY_POLL)) {
6206 				if (prefer_busy_poll)
6207 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6208 				goto count;
6209 			}
6210 			if (cmpxchg(&napi->state, val,
6211 				    val | NAPIF_STATE_IN_BUSY_POLL |
6212 					  NAPIF_STATE_SCHED) != val) {
6213 				if (prefer_busy_poll)
6214 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6215 				goto count;
6216 			}
6217 			have_poll_lock = netpoll_poll_lock(napi);
6218 			napi_poll = napi->poll;
6219 		}
6220 		work = napi_poll(napi, budget);
6221 		trace_napi_poll(napi, work, budget);
6222 		gro_normal_list(napi);
6223 count:
6224 		if (work > 0)
6225 			__NET_ADD_STATS(dev_net(napi->dev),
6226 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6227 		local_bh_enable();
6228 
6229 		if (!loop_end || loop_end(loop_end_arg, start_time))
6230 			break;
6231 
6232 		if (unlikely(need_resched())) {
6233 			if (napi_poll)
6234 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6235 			preempt_enable();
6236 			rcu_read_unlock();
6237 			cond_resched();
6238 			if (loop_end(loop_end_arg, start_time))
6239 				return;
6240 			goto restart;
6241 		}
6242 		cpu_relax();
6243 	}
6244 	if (napi_poll)
6245 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6246 	preempt_enable();
6247 out:
6248 	rcu_read_unlock();
6249 }
6250 EXPORT_SYMBOL(napi_busy_loop);
6251 
6252 #endif /* CONFIG_NET_RX_BUSY_POLL */
6253 
6254 static void napi_hash_add(struct napi_struct *napi)
6255 {
6256 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6257 		return;
6258 
6259 	spin_lock(&napi_hash_lock);
6260 
6261 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6262 	do {
6263 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6264 			napi_gen_id = MIN_NAPI_ID;
6265 	} while (napi_by_id(napi_gen_id));
6266 	napi->napi_id = napi_gen_id;
6267 
6268 	hlist_add_head_rcu(&napi->napi_hash_node,
6269 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6270 
6271 	spin_unlock(&napi_hash_lock);
6272 }
6273 
6274 /* Warning : caller is responsible to make sure rcu grace period
6275  * is respected before freeing memory containing @napi
6276  */
6277 static void napi_hash_del(struct napi_struct *napi)
6278 {
6279 	spin_lock(&napi_hash_lock);
6280 
6281 	hlist_del_init_rcu(&napi->napi_hash_node);
6282 
6283 	spin_unlock(&napi_hash_lock);
6284 }
6285 
6286 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6287 {
6288 	struct napi_struct *napi;
6289 
6290 	napi = container_of(timer, struct napi_struct, timer);
6291 
6292 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6293 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6294 	 */
6295 	if (!napi_disable_pending(napi) &&
6296 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6297 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6298 		__napi_schedule_irqoff(napi);
6299 	}
6300 
6301 	return HRTIMER_NORESTART;
6302 }
6303 
6304 static void init_gro_hash(struct napi_struct *napi)
6305 {
6306 	int i;
6307 
6308 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6309 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6310 		napi->gro_hash[i].count = 0;
6311 	}
6312 	napi->gro_bitmask = 0;
6313 }
6314 
6315 int dev_set_threaded(struct net_device *dev, bool threaded)
6316 {
6317 	struct napi_struct *napi;
6318 	int err = 0;
6319 
6320 	if (dev->threaded == threaded)
6321 		return 0;
6322 
6323 	if (threaded) {
6324 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6325 			if (!napi->thread) {
6326 				err = napi_kthread_create(napi);
6327 				if (err) {
6328 					threaded = false;
6329 					break;
6330 				}
6331 			}
6332 		}
6333 	}
6334 
6335 	dev->threaded = threaded;
6336 
6337 	/* Make sure kthread is created before THREADED bit
6338 	 * is set.
6339 	 */
6340 	smp_mb__before_atomic();
6341 
6342 	/* Setting/unsetting threaded mode on a napi might not immediately
6343 	 * take effect, if the current napi instance is actively being
6344 	 * polled. In this case, the switch between threaded mode and
6345 	 * softirq mode will happen in the next round of napi_schedule().
6346 	 * This should not cause hiccups/stalls to the live traffic.
6347 	 */
6348 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6349 		if (threaded)
6350 			set_bit(NAPI_STATE_THREADED, &napi->state);
6351 		else
6352 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6353 	}
6354 
6355 	return err;
6356 }
6357 EXPORT_SYMBOL(dev_set_threaded);
6358 
6359 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6360 			   int (*poll)(struct napi_struct *, int), int weight)
6361 {
6362 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6363 		return;
6364 
6365 	INIT_LIST_HEAD(&napi->poll_list);
6366 	INIT_HLIST_NODE(&napi->napi_hash_node);
6367 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6368 	napi->timer.function = napi_watchdog;
6369 	init_gro_hash(napi);
6370 	napi->skb = NULL;
6371 	INIT_LIST_HEAD(&napi->rx_list);
6372 	napi->rx_count = 0;
6373 	napi->poll = poll;
6374 	if (weight > NAPI_POLL_WEIGHT)
6375 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6376 				weight);
6377 	napi->weight = weight;
6378 	napi->dev = dev;
6379 #ifdef CONFIG_NETPOLL
6380 	napi->poll_owner = -1;
6381 #endif
6382 	set_bit(NAPI_STATE_SCHED, &napi->state);
6383 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6384 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6385 	napi_hash_add(napi);
6386 	napi_get_frags_check(napi);
6387 	/* Create kthread for this napi if dev->threaded is set.
6388 	 * Clear dev->threaded if kthread creation failed so that
6389 	 * threaded mode will not be enabled in napi_enable().
6390 	 */
6391 	if (dev->threaded && napi_kthread_create(napi))
6392 		dev->threaded = 0;
6393 }
6394 EXPORT_SYMBOL(netif_napi_add_weight);
6395 
6396 void napi_disable(struct napi_struct *n)
6397 {
6398 	unsigned long val, new;
6399 
6400 	might_sleep();
6401 	set_bit(NAPI_STATE_DISABLE, &n->state);
6402 
6403 	val = READ_ONCE(n->state);
6404 	do {
6405 		while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6406 			usleep_range(20, 200);
6407 			val = READ_ONCE(n->state);
6408 		}
6409 
6410 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6411 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6412 	} while (!try_cmpxchg(&n->state, &val, new));
6413 
6414 	hrtimer_cancel(&n->timer);
6415 
6416 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6417 }
6418 EXPORT_SYMBOL(napi_disable);
6419 
6420 /**
6421  *	napi_enable - enable NAPI scheduling
6422  *	@n: NAPI context
6423  *
6424  * Resume NAPI from being scheduled on this context.
6425  * Must be paired with napi_disable.
6426  */
6427 void napi_enable(struct napi_struct *n)
6428 {
6429 	unsigned long new, val = READ_ONCE(n->state);
6430 
6431 	do {
6432 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6433 
6434 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6435 		if (n->dev->threaded && n->thread)
6436 			new |= NAPIF_STATE_THREADED;
6437 	} while (!try_cmpxchg(&n->state, &val, new));
6438 }
6439 EXPORT_SYMBOL(napi_enable);
6440 
6441 static void flush_gro_hash(struct napi_struct *napi)
6442 {
6443 	int i;
6444 
6445 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6446 		struct sk_buff *skb, *n;
6447 
6448 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6449 			kfree_skb(skb);
6450 		napi->gro_hash[i].count = 0;
6451 	}
6452 }
6453 
6454 /* Must be called in process context */
6455 void __netif_napi_del(struct napi_struct *napi)
6456 {
6457 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6458 		return;
6459 
6460 	napi_hash_del(napi);
6461 	list_del_rcu(&napi->dev_list);
6462 	napi_free_frags(napi);
6463 
6464 	flush_gro_hash(napi);
6465 	napi->gro_bitmask = 0;
6466 
6467 	if (napi->thread) {
6468 		kthread_stop(napi->thread);
6469 		napi->thread = NULL;
6470 	}
6471 }
6472 EXPORT_SYMBOL(__netif_napi_del);
6473 
6474 static int __napi_poll(struct napi_struct *n, bool *repoll)
6475 {
6476 	int work, weight;
6477 
6478 	weight = n->weight;
6479 
6480 	/* This NAPI_STATE_SCHED test is for avoiding a race
6481 	 * with netpoll's poll_napi().  Only the entity which
6482 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6483 	 * actually make the ->poll() call.  Therefore we avoid
6484 	 * accidentally calling ->poll() when NAPI is not scheduled.
6485 	 */
6486 	work = 0;
6487 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6488 		work = n->poll(n, weight);
6489 		trace_napi_poll(n, work, weight);
6490 	}
6491 
6492 	if (unlikely(work > weight))
6493 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6494 				n->poll, work, weight);
6495 
6496 	if (likely(work < weight))
6497 		return work;
6498 
6499 	/* Drivers must not modify the NAPI state if they
6500 	 * consume the entire weight.  In such cases this code
6501 	 * still "owns" the NAPI instance and therefore can
6502 	 * move the instance around on the list at-will.
6503 	 */
6504 	if (unlikely(napi_disable_pending(n))) {
6505 		napi_complete(n);
6506 		return work;
6507 	}
6508 
6509 	/* The NAPI context has more processing work, but busy-polling
6510 	 * is preferred. Exit early.
6511 	 */
6512 	if (napi_prefer_busy_poll(n)) {
6513 		if (napi_complete_done(n, work)) {
6514 			/* If timeout is not set, we need to make sure
6515 			 * that the NAPI is re-scheduled.
6516 			 */
6517 			napi_schedule(n);
6518 		}
6519 		return work;
6520 	}
6521 
6522 	if (n->gro_bitmask) {
6523 		/* flush too old packets
6524 		 * If HZ < 1000, flush all packets.
6525 		 */
6526 		napi_gro_flush(n, HZ >= 1000);
6527 	}
6528 
6529 	gro_normal_list(n);
6530 
6531 	/* Some drivers may have called napi_schedule
6532 	 * prior to exhausting their budget.
6533 	 */
6534 	if (unlikely(!list_empty(&n->poll_list))) {
6535 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6536 			     n->dev ? n->dev->name : "backlog");
6537 		return work;
6538 	}
6539 
6540 	*repoll = true;
6541 
6542 	return work;
6543 }
6544 
6545 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6546 {
6547 	bool do_repoll = false;
6548 	void *have;
6549 	int work;
6550 
6551 	list_del_init(&n->poll_list);
6552 
6553 	have = netpoll_poll_lock(n);
6554 
6555 	work = __napi_poll(n, &do_repoll);
6556 
6557 	if (do_repoll)
6558 		list_add_tail(&n->poll_list, repoll);
6559 
6560 	netpoll_poll_unlock(have);
6561 
6562 	return work;
6563 }
6564 
6565 static int napi_thread_wait(struct napi_struct *napi)
6566 {
6567 	bool woken = false;
6568 
6569 	set_current_state(TASK_INTERRUPTIBLE);
6570 
6571 	while (!kthread_should_stop()) {
6572 		/* Testing SCHED_THREADED bit here to make sure the current
6573 		 * kthread owns this napi and could poll on this napi.
6574 		 * Testing SCHED bit is not enough because SCHED bit might be
6575 		 * set by some other busy poll thread or by napi_disable().
6576 		 */
6577 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6578 			WARN_ON(!list_empty(&napi->poll_list));
6579 			__set_current_state(TASK_RUNNING);
6580 			return 0;
6581 		}
6582 
6583 		schedule();
6584 		/* woken being true indicates this thread owns this napi. */
6585 		woken = true;
6586 		set_current_state(TASK_INTERRUPTIBLE);
6587 	}
6588 	__set_current_state(TASK_RUNNING);
6589 
6590 	return -1;
6591 }
6592 
6593 static int napi_threaded_poll(void *data)
6594 {
6595 	struct napi_struct *napi = data;
6596 	void *have;
6597 
6598 	while (!napi_thread_wait(napi)) {
6599 		for (;;) {
6600 			bool repoll = false;
6601 
6602 			local_bh_disable();
6603 
6604 			have = netpoll_poll_lock(napi);
6605 			__napi_poll(napi, &repoll);
6606 			netpoll_poll_unlock(have);
6607 
6608 			local_bh_enable();
6609 
6610 			if (!repoll)
6611 				break;
6612 
6613 			cond_resched();
6614 		}
6615 	}
6616 	return 0;
6617 }
6618 
6619 static void skb_defer_free_flush(struct softnet_data *sd)
6620 {
6621 	struct sk_buff *skb, *next;
6622 
6623 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6624 	if (!READ_ONCE(sd->defer_list))
6625 		return;
6626 
6627 	spin_lock_irq(&sd->defer_lock);
6628 	skb = sd->defer_list;
6629 	sd->defer_list = NULL;
6630 	sd->defer_count = 0;
6631 	spin_unlock_irq(&sd->defer_lock);
6632 
6633 	while (skb != NULL) {
6634 		next = skb->next;
6635 		napi_consume_skb(skb, 1);
6636 		skb = next;
6637 	}
6638 }
6639 
6640 static __latent_entropy void net_rx_action(struct softirq_action *h)
6641 {
6642 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6643 	unsigned long time_limit = jiffies +
6644 		usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6645 	int budget = READ_ONCE(netdev_budget);
6646 	LIST_HEAD(list);
6647 	LIST_HEAD(repoll);
6648 
6649 	local_irq_disable();
6650 	list_splice_init(&sd->poll_list, &list);
6651 	local_irq_enable();
6652 
6653 	for (;;) {
6654 		struct napi_struct *n;
6655 
6656 		skb_defer_free_flush(sd);
6657 
6658 		if (list_empty(&list)) {
6659 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6660 				goto end;
6661 			break;
6662 		}
6663 
6664 		n = list_first_entry(&list, struct napi_struct, poll_list);
6665 		budget -= napi_poll(n, &repoll);
6666 
6667 		/* If softirq window is exhausted then punt.
6668 		 * Allow this to run for 2 jiffies since which will allow
6669 		 * an average latency of 1.5/HZ.
6670 		 */
6671 		if (unlikely(budget <= 0 ||
6672 			     time_after_eq(jiffies, time_limit))) {
6673 			sd->time_squeeze++;
6674 			break;
6675 		}
6676 	}
6677 
6678 	local_irq_disable();
6679 
6680 	list_splice_tail_init(&sd->poll_list, &list);
6681 	list_splice_tail(&repoll, &list);
6682 	list_splice(&list, &sd->poll_list);
6683 	if (!list_empty(&sd->poll_list))
6684 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6685 
6686 	net_rps_action_and_irq_enable(sd);
6687 end:;
6688 }
6689 
6690 struct netdev_adjacent {
6691 	struct net_device *dev;
6692 	netdevice_tracker dev_tracker;
6693 
6694 	/* upper master flag, there can only be one master device per list */
6695 	bool master;
6696 
6697 	/* lookup ignore flag */
6698 	bool ignore;
6699 
6700 	/* counter for the number of times this device was added to us */
6701 	u16 ref_nr;
6702 
6703 	/* private field for the users */
6704 	void *private;
6705 
6706 	struct list_head list;
6707 	struct rcu_head rcu;
6708 };
6709 
6710 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6711 						 struct list_head *adj_list)
6712 {
6713 	struct netdev_adjacent *adj;
6714 
6715 	list_for_each_entry(adj, adj_list, list) {
6716 		if (adj->dev == adj_dev)
6717 			return adj;
6718 	}
6719 	return NULL;
6720 }
6721 
6722 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6723 				    struct netdev_nested_priv *priv)
6724 {
6725 	struct net_device *dev = (struct net_device *)priv->data;
6726 
6727 	return upper_dev == dev;
6728 }
6729 
6730 /**
6731  * netdev_has_upper_dev - Check if device is linked to an upper device
6732  * @dev: device
6733  * @upper_dev: upper device to check
6734  *
6735  * Find out if a device is linked to specified upper device and return true
6736  * in case it is. Note that this checks only immediate upper device,
6737  * not through a complete stack of devices. The caller must hold the RTNL lock.
6738  */
6739 bool netdev_has_upper_dev(struct net_device *dev,
6740 			  struct net_device *upper_dev)
6741 {
6742 	struct netdev_nested_priv priv = {
6743 		.data = (void *)upper_dev,
6744 	};
6745 
6746 	ASSERT_RTNL();
6747 
6748 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6749 					     &priv);
6750 }
6751 EXPORT_SYMBOL(netdev_has_upper_dev);
6752 
6753 /**
6754  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6755  * @dev: device
6756  * @upper_dev: upper device to check
6757  *
6758  * Find out if a device is linked to specified upper device and return true
6759  * in case it is. Note that this checks the entire upper device chain.
6760  * The caller must hold rcu lock.
6761  */
6762 
6763 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6764 				  struct net_device *upper_dev)
6765 {
6766 	struct netdev_nested_priv priv = {
6767 		.data = (void *)upper_dev,
6768 	};
6769 
6770 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6771 					       &priv);
6772 }
6773 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6774 
6775 /**
6776  * netdev_has_any_upper_dev - Check if device is linked to some device
6777  * @dev: device
6778  *
6779  * Find out if a device is linked to an upper device and return true in case
6780  * it is. The caller must hold the RTNL lock.
6781  */
6782 bool netdev_has_any_upper_dev(struct net_device *dev)
6783 {
6784 	ASSERT_RTNL();
6785 
6786 	return !list_empty(&dev->adj_list.upper);
6787 }
6788 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6789 
6790 /**
6791  * netdev_master_upper_dev_get - Get master upper device
6792  * @dev: device
6793  *
6794  * Find a master upper device and return pointer to it or NULL in case
6795  * it's not there. The caller must hold the RTNL lock.
6796  */
6797 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6798 {
6799 	struct netdev_adjacent *upper;
6800 
6801 	ASSERT_RTNL();
6802 
6803 	if (list_empty(&dev->adj_list.upper))
6804 		return NULL;
6805 
6806 	upper = list_first_entry(&dev->adj_list.upper,
6807 				 struct netdev_adjacent, list);
6808 	if (likely(upper->master))
6809 		return upper->dev;
6810 	return NULL;
6811 }
6812 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6813 
6814 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6815 {
6816 	struct netdev_adjacent *upper;
6817 
6818 	ASSERT_RTNL();
6819 
6820 	if (list_empty(&dev->adj_list.upper))
6821 		return NULL;
6822 
6823 	upper = list_first_entry(&dev->adj_list.upper,
6824 				 struct netdev_adjacent, list);
6825 	if (likely(upper->master) && !upper->ignore)
6826 		return upper->dev;
6827 	return NULL;
6828 }
6829 
6830 /**
6831  * netdev_has_any_lower_dev - Check if device is linked to some device
6832  * @dev: device
6833  *
6834  * Find out if a device is linked to a lower device and return true in case
6835  * it is. The caller must hold the RTNL lock.
6836  */
6837 static bool netdev_has_any_lower_dev(struct net_device *dev)
6838 {
6839 	ASSERT_RTNL();
6840 
6841 	return !list_empty(&dev->adj_list.lower);
6842 }
6843 
6844 void *netdev_adjacent_get_private(struct list_head *adj_list)
6845 {
6846 	struct netdev_adjacent *adj;
6847 
6848 	adj = list_entry(adj_list, struct netdev_adjacent, list);
6849 
6850 	return adj->private;
6851 }
6852 EXPORT_SYMBOL(netdev_adjacent_get_private);
6853 
6854 /**
6855  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6856  * @dev: device
6857  * @iter: list_head ** of the current position
6858  *
6859  * Gets the next device from the dev's upper list, starting from iter
6860  * position. The caller must hold RCU read lock.
6861  */
6862 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6863 						 struct list_head **iter)
6864 {
6865 	struct netdev_adjacent *upper;
6866 
6867 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6868 
6869 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6870 
6871 	if (&upper->list == &dev->adj_list.upper)
6872 		return NULL;
6873 
6874 	*iter = &upper->list;
6875 
6876 	return upper->dev;
6877 }
6878 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6879 
6880 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6881 						  struct list_head **iter,
6882 						  bool *ignore)
6883 {
6884 	struct netdev_adjacent *upper;
6885 
6886 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6887 
6888 	if (&upper->list == &dev->adj_list.upper)
6889 		return NULL;
6890 
6891 	*iter = &upper->list;
6892 	*ignore = upper->ignore;
6893 
6894 	return upper->dev;
6895 }
6896 
6897 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6898 						    struct list_head **iter)
6899 {
6900 	struct netdev_adjacent *upper;
6901 
6902 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6903 
6904 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6905 
6906 	if (&upper->list == &dev->adj_list.upper)
6907 		return NULL;
6908 
6909 	*iter = &upper->list;
6910 
6911 	return upper->dev;
6912 }
6913 
6914 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6915 				       int (*fn)(struct net_device *dev,
6916 					 struct netdev_nested_priv *priv),
6917 				       struct netdev_nested_priv *priv)
6918 {
6919 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6920 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6921 	int ret, cur = 0;
6922 	bool ignore;
6923 
6924 	now = dev;
6925 	iter = &dev->adj_list.upper;
6926 
6927 	while (1) {
6928 		if (now != dev) {
6929 			ret = fn(now, priv);
6930 			if (ret)
6931 				return ret;
6932 		}
6933 
6934 		next = NULL;
6935 		while (1) {
6936 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
6937 			if (!udev)
6938 				break;
6939 			if (ignore)
6940 				continue;
6941 
6942 			next = udev;
6943 			niter = &udev->adj_list.upper;
6944 			dev_stack[cur] = now;
6945 			iter_stack[cur++] = iter;
6946 			break;
6947 		}
6948 
6949 		if (!next) {
6950 			if (!cur)
6951 				return 0;
6952 			next = dev_stack[--cur];
6953 			niter = iter_stack[cur];
6954 		}
6955 
6956 		now = next;
6957 		iter = niter;
6958 	}
6959 
6960 	return 0;
6961 }
6962 
6963 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6964 				  int (*fn)(struct net_device *dev,
6965 					    struct netdev_nested_priv *priv),
6966 				  struct netdev_nested_priv *priv)
6967 {
6968 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6969 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6970 	int ret, cur = 0;
6971 
6972 	now = dev;
6973 	iter = &dev->adj_list.upper;
6974 
6975 	while (1) {
6976 		if (now != dev) {
6977 			ret = fn(now, priv);
6978 			if (ret)
6979 				return ret;
6980 		}
6981 
6982 		next = NULL;
6983 		while (1) {
6984 			udev = netdev_next_upper_dev_rcu(now, &iter);
6985 			if (!udev)
6986 				break;
6987 
6988 			next = udev;
6989 			niter = &udev->adj_list.upper;
6990 			dev_stack[cur] = now;
6991 			iter_stack[cur++] = iter;
6992 			break;
6993 		}
6994 
6995 		if (!next) {
6996 			if (!cur)
6997 				return 0;
6998 			next = dev_stack[--cur];
6999 			niter = iter_stack[cur];
7000 		}
7001 
7002 		now = next;
7003 		iter = niter;
7004 	}
7005 
7006 	return 0;
7007 }
7008 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7009 
7010 static bool __netdev_has_upper_dev(struct net_device *dev,
7011 				   struct net_device *upper_dev)
7012 {
7013 	struct netdev_nested_priv priv = {
7014 		.flags = 0,
7015 		.data = (void *)upper_dev,
7016 	};
7017 
7018 	ASSERT_RTNL();
7019 
7020 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7021 					   &priv);
7022 }
7023 
7024 /**
7025  * netdev_lower_get_next_private - Get the next ->private from the
7026  *				   lower neighbour list
7027  * @dev: device
7028  * @iter: list_head ** of the current position
7029  *
7030  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7031  * list, starting from iter position. The caller must hold either hold the
7032  * RTNL lock or its own locking that guarantees that the neighbour lower
7033  * list will remain unchanged.
7034  */
7035 void *netdev_lower_get_next_private(struct net_device *dev,
7036 				    struct list_head **iter)
7037 {
7038 	struct netdev_adjacent *lower;
7039 
7040 	lower = list_entry(*iter, struct netdev_adjacent, list);
7041 
7042 	if (&lower->list == &dev->adj_list.lower)
7043 		return NULL;
7044 
7045 	*iter = lower->list.next;
7046 
7047 	return lower->private;
7048 }
7049 EXPORT_SYMBOL(netdev_lower_get_next_private);
7050 
7051 /**
7052  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7053  *				       lower neighbour list, RCU
7054  *				       variant
7055  * @dev: device
7056  * @iter: list_head ** of the current position
7057  *
7058  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7059  * list, starting from iter position. The caller must hold RCU read lock.
7060  */
7061 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7062 					struct list_head **iter)
7063 {
7064 	struct netdev_adjacent *lower;
7065 
7066 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7067 
7068 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7069 
7070 	if (&lower->list == &dev->adj_list.lower)
7071 		return NULL;
7072 
7073 	*iter = &lower->list;
7074 
7075 	return lower->private;
7076 }
7077 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7078 
7079 /**
7080  * netdev_lower_get_next - Get the next device from the lower neighbour
7081  *                         list
7082  * @dev: device
7083  * @iter: list_head ** of the current position
7084  *
7085  * Gets the next netdev_adjacent from the dev's lower neighbour
7086  * list, starting from iter position. The caller must hold RTNL lock or
7087  * its own locking that guarantees that the neighbour lower
7088  * list will remain unchanged.
7089  */
7090 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7091 {
7092 	struct netdev_adjacent *lower;
7093 
7094 	lower = list_entry(*iter, struct netdev_adjacent, list);
7095 
7096 	if (&lower->list == &dev->adj_list.lower)
7097 		return NULL;
7098 
7099 	*iter = lower->list.next;
7100 
7101 	return lower->dev;
7102 }
7103 EXPORT_SYMBOL(netdev_lower_get_next);
7104 
7105 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7106 						struct list_head **iter)
7107 {
7108 	struct netdev_adjacent *lower;
7109 
7110 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7111 
7112 	if (&lower->list == &dev->adj_list.lower)
7113 		return NULL;
7114 
7115 	*iter = &lower->list;
7116 
7117 	return lower->dev;
7118 }
7119 
7120 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7121 						  struct list_head **iter,
7122 						  bool *ignore)
7123 {
7124 	struct netdev_adjacent *lower;
7125 
7126 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7127 
7128 	if (&lower->list == &dev->adj_list.lower)
7129 		return NULL;
7130 
7131 	*iter = &lower->list;
7132 	*ignore = lower->ignore;
7133 
7134 	return lower->dev;
7135 }
7136 
7137 int netdev_walk_all_lower_dev(struct net_device *dev,
7138 			      int (*fn)(struct net_device *dev,
7139 					struct netdev_nested_priv *priv),
7140 			      struct netdev_nested_priv *priv)
7141 {
7142 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7143 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7144 	int ret, cur = 0;
7145 
7146 	now = dev;
7147 	iter = &dev->adj_list.lower;
7148 
7149 	while (1) {
7150 		if (now != dev) {
7151 			ret = fn(now, priv);
7152 			if (ret)
7153 				return ret;
7154 		}
7155 
7156 		next = NULL;
7157 		while (1) {
7158 			ldev = netdev_next_lower_dev(now, &iter);
7159 			if (!ldev)
7160 				break;
7161 
7162 			next = ldev;
7163 			niter = &ldev->adj_list.lower;
7164 			dev_stack[cur] = now;
7165 			iter_stack[cur++] = iter;
7166 			break;
7167 		}
7168 
7169 		if (!next) {
7170 			if (!cur)
7171 				return 0;
7172 			next = dev_stack[--cur];
7173 			niter = iter_stack[cur];
7174 		}
7175 
7176 		now = next;
7177 		iter = niter;
7178 	}
7179 
7180 	return 0;
7181 }
7182 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7183 
7184 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7185 				       int (*fn)(struct net_device *dev,
7186 					 struct netdev_nested_priv *priv),
7187 				       struct netdev_nested_priv *priv)
7188 {
7189 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7190 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7191 	int ret, cur = 0;
7192 	bool ignore;
7193 
7194 	now = dev;
7195 	iter = &dev->adj_list.lower;
7196 
7197 	while (1) {
7198 		if (now != dev) {
7199 			ret = fn(now, priv);
7200 			if (ret)
7201 				return ret;
7202 		}
7203 
7204 		next = NULL;
7205 		while (1) {
7206 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7207 			if (!ldev)
7208 				break;
7209 			if (ignore)
7210 				continue;
7211 
7212 			next = ldev;
7213 			niter = &ldev->adj_list.lower;
7214 			dev_stack[cur] = now;
7215 			iter_stack[cur++] = iter;
7216 			break;
7217 		}
7218 
7219 		if (!next) {
7220 			if (!cur)
7221 				return 0;
7222 			next = dev_stack[--cur];
7223 			niter = iter_stack[cur];
7224 		}
7225 
7226 		now = next;
7227 		iter = niter;
7228 	}
7229 
7230 	return 0;
7231 }
7232 
7233 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7234 					     struct list_head **iter)
7235 {
7236 	struct netdev_adjacent *lower;
7237 
7238 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7239 	if (&lower->list == &dev->adj_list.lower)
7240 		return NULL;
7241 
7242 	*iter = &lower->list;
7243 
7244 	return lower->dev;
7245 }
7246 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7247 
7248 static u8 __netdev_upper_depth(struct net_device *dev)
7249 {
7250 	struct net_device *udev;
7251 	struct list_head *iter;
7252 	u8 max_depth = 0;
7253 	bool ignore;
7254 
7255 	for (iter = &dev->adj_list.upper,
7256 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7257 	     udev;
7258 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7259 		if (ignore)
7260 			continue;
7261 		if (max_depth < udev->upper_level)
7262 			max_depth = udev->upper_level;
7263 	}
7264 
7265 	return max_depth;
7266 }
7267 
7268 static u8 __netdev_lower_depth(struct net_device *dev)
7269 {
7270 	struct net_device *ldev;
7271 	struct list_head *iter;
7272 	u8 max_depth = 0;
7273 	bool ignore;
7274 
7275 	for (iter = &dev->adj_list.lower,
7276 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7277 	     ldev;
7278 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7279 		if (ignore)
7280 			continue;
7281 		if (max_depth < ldev->lower_level)
7282 			max_depth = ldev->lower_level;
7283 	}
7284 
7285 	return max_depth;
7286 }
7287 
7288 static int __netdev_update_upper_level(struct net_device *dev,
7289 				       struct netdev_nested_priv *__unused)
7290 {
7291 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7292 	return 0;
7293 }
7294 
7295 #ifdef CONFIG_LOCKDEP
7296 static LIST_HEAD(net_unlink_list);
7297 
7298 static void net_unlink_todo(struct net_device *dev)
7299 {
7300 	if (list_empty(&dev->unlink_list))
7301 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7302 }
7303 #endif
7304 
7305 static int __netdev_update_lower_level(struct net_device *dev,
7306 				       struct netdev_nested_priv *priv)
7307 {
7308 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7309 
7310 #ifdef CONFIG_LOCKDEP
7311 	if (!priv)
7312 		return 0;
7313 
7314 	if (priv->flags & NESTED_SYNC_IMM)
7315 		dev->nested_level = dev->lower_level - 1;
7316 	if (priv->flags & NESTED_SYNC_TODO)
7317 		net_unlink_todo(dev);
7318 #endif
7319 	return 0;
7320 }
7321 
7322 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7323 				  int (*fn)(struct net_device *dev,
7324 					    struct netdev_nested_priv *priv),
7325 				  struct netdev_nested_priv *priv)
7326 {
7327 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7328 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7329 	int ret, cur = 0;
7330 
7331 	now = dev;
7332 	iter = &dev->adj_list.lower;
7333 
7334 	while (1) {
7335 		if (now != dev) {
7336 			ret = fn(now, priv);
7337 			if (ret)
7338 				return ret;
7339 		}
7340 
7341 		next = NULL;
7342 		while (1) {
7343 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7344 			if (!ldev)
7345 				break;
7346 
7347 			next = ldev;
7348 			niter = &ldev->adj_list.lower;
7349 			dev_stack[cur] = now;
7350 			iter_stack[cur++] = iter;
7351 			break;
7352 		}
7353 
7354 		if (!next) {
7355 			if (!cur)
7356 				return 0;
7357 			next = dev_stack[--cur];
7358 			niter = iter_stack[cur];
7359 		}
7360 
7361 		now = next;
7362 		iter = niter;
7363 	}
7364 
7365 	return 0;
7366 }
7367 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7368 
7369 /**
7370  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7371  *				       lower neighbour list, RCU
7372  *				       variant
7373  * @dev: device
7374  *
7375  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7376  * list. The caller must hold RCU read lock.
7377  */
7378 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7379 {
7380 	struct netdev_adjacent *lower;
7381 
7382 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7383 			struct netdev_adjacent, list);
7384 	if (lower)
7385 		return lower->private;
7386 	return NULL;
7387 }
7388 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7389 
7390 /**
7391  * netdev_master_upper_dev_get_rcu - Get master upper device
7392  * @dev: device
7393  *
7394  * Find a master upper device and return pointer to it or NULL in case
7395  * it's not there. The caller must hold the RCU read lock.
7396  */
7397 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7398 {
7399 	struct netdev_adjacent *upper;
7400 
7401 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7402 				       struct netdev_adjacent, list);
7403 	if (upper && likely(upper->master))
7404 		return upper->dev;
7405 	return NULL;
7406 }
7407 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7408 
7409 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7410 			      struct net_device *adj_dev,
7411 			      struct list_head *dev_list)
7412 {
7413 	char linkname[IFNAMSIZ+7];
7414 
7415 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7416 		"upper_%s" : "lower_%s", adj_dev->name);
7417 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7418 				 linkname);
7419 }
7420 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7421 			       char *name,
7422 			       struct list_head *dev_list)
7423 {
7424 	char linkname[IFNAMSIZ+7];
7425 
7426 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7427 		"upper_%s" : "lower_%s", name);
7428 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7429 }
7430 
7431 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7432 						 struct net_device *adj_dev,
7433 						 struct list_head *dev_list)
7434 {
7435 	return (dev_list == &dev->adj_list.upper ||
7436 		dev_list == &dev->adj_list.lower) &&
7437 		net_eq(dev_net(dev), dev_net(adj_dev));
7438 }
7439 
7440 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7441 					struct net_device *adj_dev,
7442 					struct list_head *dev_list,
7443 					void *private, bool master)
7444 {
7445 	struct netdev_adjacent *adj;
7446 	int ret;
7447 
7448 	adj = __netdev_find_adj(adj_dev, dev_list);
7449 
7450 	if (adj) {
7451 		adj->ref_nr += 1;
7452 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7453 			 dev->name, adj_dev->name, adj->ref_nr);
7454 
7455 		return 0;
7456 	}
7457 
7458 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7459 	if (!adj)
7460 		return -ENOMEM;
7461 
7462 	adj->dev = adj_dev;
7463 	adj->master = master;
7464 	adj->ref_nr = 1;
7465 	adj->private = private;
7466 	adj->ignore = false;
7467 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7468 
7469 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7470 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7471 
7472 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7473 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7474 		if (ret)
7475 			goto free_adj;
7476 	}
7477 
7478 	/* Ensure that master link is always the first item in list. */
7479 	if (master) {
7480 		ret = sysfs_create_link(&(dev->dev.kobj),
7481 					&(adj_dev->dev.kobj), "master");
7482 		if (ret)
7483 			goto remove_symlinks;
7484 
7485 		list_add_rcu(&adj->list, dev_list);
7486 	} else {
7487 		list_add_tail_rcu(&adj->list, dev_list);
7488 	}
7489 
7490 	return 0;
7491 
7492 remove_symlinks:
7493 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7494 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7495 free_adj:
7496 	netdev_put(adj_dev, &adj->dev_tracker);
7497 	kfree(adj);
7498 
7499 	return ret;
7500 }
7501 
7502 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7503 					 struct net_device *adj_dev,
7504 					 u16 ref_nr,
7505 					 struct list_head *dev_list)
7506 {
7507 	struct netdev_adjacent *adj;
7508 
7509 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7510 		 dev->name, adj_dev->name, ref_nr);
7511 
7512 	adj = __netdev_find_adj(adj_dev, dev_list);
7513 
7514 	if (!adj) {
7515 		pr_err("Adjacency does not exist for device %s from %s\n",
7516 		       dev->name, adj_dev->name);
7517 		WARN_ON(1);
7518 		return;
7519 	}
7520 
7521 	if (adj->ref_nr > ref_nr) {
7522 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7523 			 dev->name, adj_dev->name, ref_nr,
7524 			 adj->ref_nr - ref_nr);
7525 		adj->ref_nr -= ref_nr;
7526 		return;
7527 	}
7528 
7529 	if (adj->master)
7530 		sysfs_remove_link(&(dev->dev.kobj), "master");
7531 
7532 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7533 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7534 
7535 	list_del_rcu(&adj->list);
7536 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7537 		 adj_dev->name, dev->name, adj_dev->name);
7538 	netdev_put(adj_dev, &adj->dev_tracker);
7539 	kfree_rcu(adj, rcu);
7540 }
7541 
7542 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7543 					    struct net_device *upper_dev,
7544 					    struct list_head *up_list,
7545 					    struct list_head *down_list,
7546 					    void *private, bool master)
7547 {
7548 	int ret;
7549 
7550 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7551 					   private, master);
7552 	if (ret)
7553 		return ret;
7554 
7555 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7556 					   private, false);
7557 	if (ret) {
7558 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7559 		return ret;
7560 	}
7561 
7562 	return 0;
7563 }
7564 
7565 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7566 					       struct net_device *upper_dev,
7567 					       u16 ref_nr,
7568 					       struct list_head *up_list,
7569 					       struct list_head *down_list)
7570 {
7571 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7572 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7573 }
7574 
7575 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7576 						struct net_device *upper_dev,
7577 						void *private, bool master)
7578 {
7579 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7580 						&dev->adj_list.upper,
7581 						&upper_dev->adj_list.lower,
7582 						private, master);
7583 }
7584 
7585 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7586 						   struct net_device *upper_dev)
7587 {
7588 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7589 					   &dev->adj_list.upper,
7590 					   &upper_dev->adj_list.lower);
7591 }
7592 
7593 static int __netdev_upper_dev_link(struct net_device *dev,
7594 				   struct net_device *upper_dev, bool master,
7595 				   void *upper_priv, void *upper_info,
7596 				   struct netdev_nested_priv *priv,
7597 				   struct netlink_ext_ack *extack)
7598 {
7599 	struct netdev_notifier_changeupper_info changeupper_info = {
7600 		.info = {
7601 			.dev = dev,
7602 			.extack = extack,
7603 		},
7604 		.upper_dev = upper_dev,
7605 		.master = master,
7606 		.linking = true,
7607 		.upper_info = upper_info,
7608 	};
7609 	struct net_device *master_dev;
7610 	int ret = 0;
7611 
7612 	ASSERT_RTNL();
7613 
7614 	if (dev == upper_dev)
7615 		return -EBUSY;
7616 
7617 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7618 	if (__netdev_has_upper_dev(upper_dev, dev))
7619 		return -EBUSY;
7620 
7621 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7622 		return -EMLINK;
7623 
7624 	if (!master) {
7625 		if (__netdev_has_upper_dev(dev, upper_dev))
7626 			return -EEXIST;
7627 	} else {
7628 		master_dev = __netdev_master_upper_dev_get(dev);
7629 		if (master_dev)
7630 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7631 	}
7632 
7633 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7634 					    &changeupper_info.info);
7635 	ret = notifier_to_errno(ret);
7636 	if (ret)
7637 		return ret;
7638 
7639 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7640 						   master);
7641 	if (ret)
7642 		return ret;
7643 
7644 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7645 					    &changeupper_info.info);
7646 	ret = notifier_to_errno(ret);
7647 	if (ret)
7648 		goto rollback;
7649 
7650 	__netdev_update_upper_level(dev, NULL);
7651 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7652 
7653 	__netdev_update_lower_level(upper_dev, priv);
7654 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7655 				    priv);
7656 
7657 	return 0;
7658 
7659 rollback:
7660 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7661 
7662 	return ret;
7663 }
7664 
7665 /**
7666  * netdev_upper_dev_link - Add a link to the upper device
7667  * @dev: device
7668  * @upper_dev: new upper device
7669  * @extack: netlink extended ack
7670  *
7671  * Adds a link to device which is upper to this one. The caller must hold
7672  * the RTNL lock. On a failure a negative errno code is returned.
7673  * On success the reference counts are adjusted and the function
7674  * returns zero.
7675  */
7676 int netdev_upper_dev_link(struct net_device *dev,
7677 			  struct net_device *upper_dev,
7678 			  struct netlink_ext_ack *extack)
7679 {
7680 	struct netdev_nested_priv priv = {
7681 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7682 		.data = NULL,
7683 	};
7684 
7685 	return __netdev_upper_dev_link(dev, upper_dev, false,
7686 				       NULL, NULL, &priv, extack);
7687 }
7688 EXPORT_SYMBOL(netdev_upper_dev_link);
7689 
7690 /**
7691  * netdev_master_upper_dev_link - Add a master link to the upper device
7692  * @dev: device
7693  * @upper_dev: new upper device
7694  * @upper_priv: upper device private
7695  * @upper_info: upper info to be passed down via notifier
7696  * @extack: netlink extended ack
7697  *
7698  * Adds a link to device which is upper to this one. In this case, only
7699  * one master upper device can be linked, although other non-master devices
7700  * might be linked as well. The caller must hold the RTNL lock.
7701  * On a failure a negative errno code is returned. On success the reference
7702  * counts are adjusted and the function returns zero.
7703  */
7704 int netdev_master_upper_dev_link(struct net_device *dev,
7705 				 struct net_device *upper_dev,
7706 				 void *upper_priv, void *upper_info,
7707 				 struct netlink_ext_ack *extack)
7708 {
7709 	struct netdev_nested_priv priv = {
7710 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7711 		.data = NULL,
7712 	};
7713 
7714 	return __netdev_upper_dev_link(dev, upper_dev, true,
7715 				       upper_priv, upper_info, &priv, extack);
7716 }
7717 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7718 
7719 static void __netdev_upper_dev_unlink(struct net_device *dev,
7720 				      struct net_device *upper_dev,
7721 				      struct netdev_nested_priv *priv)
7722 {
7723 	struct netdev_notifier_changeupper_info changeupper_info = {
7724 		.info = {
7725 			.dev = dev,
7726 		},
7727 		.upper_dev = upper_dev,
7728 		.linking = false,
7729 	};
7730 
7731 	ASSERT_RTNL();
7732 
7733 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7734 
7735 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7736 				      &changeupper_info.info);
7737 
7738 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7739 
7740 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7741 				      &changeupper_info.info);
7742 
7743 	__netdev_update_upper_level(dev, NULL);
7744 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7745 
7746 	__netdev_update_lower_level(upper_dev, priv);
7747 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7748 				    priv);
7749 }
7750 
7751 /**
7752  * netdev_upper_dev_unlink - Removes a link to upper device
7753  * @dev: device
7754  * @upper_dev: new upper device
7755  *
7756  * Removes a link to device which is upper to this one. The caller must hold
7757  * the RTNL lock.
7758  */
7759 void netdev_upper_dev_unlink(struct net_device *dev,
7760 			     struct net_device *upper_dev)
7761 {
7762 	struct netdev_nested_priv priv = {
7763 		.flags = NESTED_SYNC_TODO,
7764 		.data = NULL,
7765 	};
7766 
7767 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
7768 }
7769 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7770 
7771 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7772 				      struct net_device *lower_dev,
7773 				      bool val)
7774 {
7775 	struct netdev_adjacent *adj;
7776 
7777 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7778 	if (adj)
7779 		adj->ignore = val;
7780 
7781 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7782 	if (adj)
7783 		adj->ignore = val;
7784 }
7785 
7786 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7787 					struct net_device *lower_dev)
7788 {
7789 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7790 }
7791 
7792 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7793 				       struct net_device *lower_dev)
7794 {
7795 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7796 }
7797 
7798 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7799 				   struct net_device *new_dev,
7800 				   struct net_device *dev,
7801 				   struct netlink_ext_ack *extack)
7802 {
7803 	struct netdev_nested_priv priv = {
7804 		.flags = 0,
7805 		.data = NULL,
7806 	};
7807 	int err;
7808 
7809 	if (!new_dev)
7810 		return 0;
7811 
7812 	if (old_dev && new_dev != old_dev)
7813 		netdev_adjacent_dev_disable(dev, old_dev);
7814 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7815 				      extack);
7816 	if (err) {
7817 		if (old_dev && new_dev != old_dev)
7818 			netdev_adjacent_dev_enable(dev, old_dev);
7819 		return err;
7820 	}
7821 
7822 	return 0;
7823 }
7824 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7825 
7826 void netdev_adjacent_change_commit(struct net_device *old_dev,
7827 				   struct net_device *new_dev,
7828 				   struct net_device *dev)
7829 {
7830 	struct netdev_nested_priv priv = {
7831 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7832 		.data = NULL,
7833 	};
7834 
7835 	if (!new_dev || !old_dev)
7836 		return;
7837 
7838 	if (new_dev == old_dev)
7839 		return;
7840 
7841 	netdev_adjacent_dev_enable(dev, old_dev);
7842 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
7843 }
7844 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7845 
7846 void netdev_adjacent_change_abort(struct net_device *old_dev,
7847 				  struct net_device *new_dev,
7848 				  struct net_device *dev)
7849 {
7850 	struct netdev_nested_priv priv = {
7851 		.flags = 0,
7852 		.data = NULL,
7853 	};
7854 
7855 	if (!new_dev)
7856 		return;
7857 
7858 	if (old_dev && new_dev != old_dev)
7859 		netdev_adjacent_dev_enable(dev, old_dev);
7860 
7861 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
7862 }
7863 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7864 
7865 /**
7866  * netdev_bonding_info_change - Dispatch event about slave change
7867  * @dev: device
7868  * @bonding_info: info to dispatch
7869  *
7870  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7871  * The caller must hold the RTNL lock.
7872  */
7873 void netdev_bonding_info_change(struct net_device *dev,
7874 				struct netdev_bonding_info *bonding_info)
7875 {
7876 	struct netdev_notifier_bonding_info info = {
7877 		.info.dev = dev,
7878 	};
7879 
7880 	memcpy(&info.bonding_info, bonding_info,
7881 	       sizeof(struct netdev_bonding_info));
7882 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7883 				      &info.info);
7884 }
7885 EXPORT_SYMBOL(netdev_bonding_info_change);
7886 
7887 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7888 					   struct netlink_ext_ack *extack)
7889 {
7890 	struct netdev_notifier_offload_xstats_info info = {
7891 		.info.dev = dev,
7892 		.info.extack = extack,
7893 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7894 	};
7895 	int err;
7896 	int rc;
7897 
7898 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7899 					 GFP_KERNEL);
7900 	if (!dev->offload_xstats_l3)
7901 		return -ENOMEM;
7902 
7903 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7904 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
7905 						  &info.info);
7906 	err = notifier_to_errno(rc);
7907 	if (err)
7908 		goto free_stats;
7909 
7910 	return 0;
7911 
7912 free_stats:
7913 	kfree(dev->offload_xstats_l3);
7914 	dev->offload_xstats_l3 = NULL;
7915 	return err;
7916 }
7917 
7918 int netdev_offload_xstats_enable(struct net_device *dev,
7919 				 enum netdev_offload_xstats_type type,
7920 				 struct netlink_ext_ack *extack)
7921 {
7922 	ASSERT_RTNL();
7923 
7924 	if (netdev_offload_xstats_enabled(dev, type))
7925 		return -EALREADY;
7926 
7927 	switch (type) {
7928 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7929 		return netdev_offload_xstats_enable_l3(dev, extack);
7930 	}
7931 
7932 	WARN_ON(1);
7933 	return -EINVAL;
7934 }
7935 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7936 
7937 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7938 {
7939 	struct netdev_notifier_offload_xstats_info info = {
7940 		.info.dev = dev,
7941 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7942 	};
7943 
7944 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7945 				      &info.info);
7946 	kfree(dev->offload_xstats_l3);
7947 	dev->offload_xstats_l3 = NULL;
7948 }
7949 
7950 int netdev_offload_xstats_disable(struct net_device *dev,
7951 				  enum netdev_offload_xstats_type type)
7952 {
7953 	ASSERT_RTNL();
7954 
7955 	if (!netdev_offload_xstats_enabled(dev, type))
7956 		return -EALREADY;
7957 
7958 	switch (type) {
7959 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7960 		netdev_offload_xstats_disable_l3(dev);
7961 		return 0;
7962 	}
7963 
7964 	WARN_ON(1);
7965 	return -EINVAL;
7966 }
7967 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7968 
7969 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7970 {
7971 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7972 }
7973 
7974 static struct rtnl_hw_stats64 *
7975 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7976 			      enum netdev_offload_xstats_type type)
7977 {
7978 	switch (type) {
7979 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7980 		return dev->offload_xstats_l3;
7981 	}
7982 
7983 	WARN_ON(1);
7984 	return NULL;
7985 }
7986 
7987 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7988 				   enum netdev_offload_xstats_type type)
7989 {
7990 	ASSERT_RTNL();
7991 
7992 	return netdev_offload_xstats_get_ptr(dev, type);
7993 }
7994 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
7995 
7996 struct netdev_notifier_offload_xstats_ru {
7997 	bool used;
7998 };
7999 
8000 struct netdev_notifier_offload_xstats_rd {
8001 	struct rtnl_hw_stats64 stats;
8002 	bool used;
8003 };
8004 
8005 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8006 				  const struct rtnl_hw_stats64 *src)
8007 {
8008 	dest->rx_packets	  += src->rx_packets;
8009 	dest->tx_packets	  += src->tx_packets;
8010 	dest->rx_bytes		  += src->rx_bytes;
8011 	dest->tx_bytes		  += src->tx_bytes;
8012 	dest->rx_errors		  += src->rx_errors;
8013 	dest->tx_errors		  += src->tx_errors;
8014 	dest->rx_dropped	  += src->rx_dropped;
8015 	dest->tx_dropped	  += src->tx_dropped;
8016 	dest->multicast		  += src->multicast;
8017 }
8018 
8019 static int netdev_offload_xstats_get_used(struct net_device *dev,
8020 					  enum netdev_offload_xstats_type type,
8021 					  bool *p_used,
8022 					  struct netlink_ext_ack *extack)
8023 {
8024 	struct netdev_notifier_offload_xstats_ru report_used = {};
8025 	struct netdev_notifier_offload_xstats_info info = {
8026 		.info.dev = dev,
8027 		.info.extack = extack,
8028 		.type = type,
8029 		.report_used = &report_used,
8030 	};
8031 	int rc;
8032 
8033 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8034 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8035 					   &info.info);
8036 	*p_used = report_used.used;
8037 	return notifier_to_errno(rc);
8038 }
8039 
8040 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8041 					   enum netdev_offload_xstats_type type,
8042 					   struct rtnl_hw_stats64 *p_stats,
8043 					   bool *p_used,
8044 					   struct netlink_ext_ack *extack)
8045 {
8046 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8047 	struct netdev_notifier_offload_xstats_info info = {
8048 		.info.dev = dev,
8049 		.info.extack = extack,
8050 		.type = type,
8051 		.report_delta = &report_delta,
8052 	};
8053 	struct rtnl_hw_stats64 *stats;
8054 	int rc;
8055 
8056 	stats = netdev_offload_xstats_get_ptr(dev, type);
8057 	if (WARN_ON(!stats))
8058 		return -EINVAL;
8059 
8060 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8061 					   &info.info);
8062 
8063 	/* Cache whatever we got, even if there was an error, otherwise the
8064 	 * successful stats retrievals would get lost.
8065 	 */
8066 	netdev_hw_stats64_add(stats, &report_delta.stats);
8067 
8068 	if (p_stats)
8069 		*p_stats = *stats;
8070 	*p_used = report_delta.used;
8071 
8072 	return notifier_to_errno(rc);
8073 }
8074 
8075 int netdev_offload_xstats_get(struct net_device *dev,
8076 			      enum netdev_offload_xstats_type type,
8077 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8078 			      struct netlink_ext_ack *extack)
8079 {
8080 	ASSERT_RTNL();
8081 
8082 	if (p_stats)
8083 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8084 						       p_used, extack);
8085 	else
8086 		return netdev_offload_xstats_get_used(dev, type, p_used,
8087 						      extack);
8088 }
8089 EXPORT_SYMBOL(netdev_offload_xstats_get);
8090 
8091 void
8092 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8093 				   const struct rtnl_hw_stats64 *stats)
8094 {
8095 	report_delta->used = true;
8096 	netdev_hw_stats64_add(&report_delta->stats, stats);
8097 }
8098 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8099 
8100 void
8101 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8102 {
8103 	report_used->used = true;
8104 }
8105 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8106 
8107 void netdev_offload_xstats_push_delta(struct net_device *dev,
8108 				      enum netdev_offload_xstats_type type,
8109 				      const struct rtnl_hw_stats64 *p_stats)
8110 {
8111 	struct rtnl_hw_stats64 *stats;
8112 
8113 	ASSERT_RTNL();
8114 
8115 	stats = netdev_offload_xstats_get_ptr(dev, type);
8116 	if (WARN_ON(!stats))
8117 		return;
8118 
8119 	netdev_hw_stats64_add(stats, p_stats);
8120 }
8121 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8122 
8123 /**
8124  * netdev_get_xmit_slave - Get the xmit slave of master device
8125  * @dev: device
8126  * @skb: The packet
8127  * @all_slaves: assume all the slaves are active
8128  *
8129  * The reference counters are not incremented so the caller must be
8130  * careful with locks. The caller must hold RCU lock.
8131  * %NULL is returned if no slave is found.
8132  */
8133 
8134 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8135 					 struct sk_buff *skb,
8136 					 bool all_slaves)
8137 {
8138 	const struct net_device_ops *ops = dev->netdev_ops;
8139 
8140 	if (!ops->ndo_get_xmit_slave)
8141 		return NULL;
8142 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8143 }
8144 EXPORT_SYMBOL(netdev_get_xmit_slave);
8145 
8146 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8147 						  struct sock *sk)
8148 {
8149 	const struct net_device_ops *ops = dev->netdev_ops;
8150 
8151 	if (!ops->ndo_sk_get_lower_dev)
8152 		return NULL;
8153 	return ops->ndo_sk_get_lower_dev(dev, sk);
8154 }
8155 
8156 /**
8157  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8158  * @dev: device
8159  * @sk: the socket
8160  *
8161  * %NULL is returned if no lower device is found.
8162  */
8163 
8164 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8165 					    struct sock *sk)
8166 {
8167 	struct net_device *lower;
8168 
8169 	lower = netdev_sk_get_lower_dev(dev, sk);
8170 	while (lower) {
8171 		dev = lower;
8172 		lower = netdev_sk_get_lower_dev(dev, sk);
8173 	}
8174 
8175 	return dev;
8176 }
8177 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8178 
8179 static void netdev_adjacent_add_links(struct net_device *dev)
8180 {
8181 	struct netdev_adjacent *iter;
8182 
8183 	struct net *net = dev_net(dev);
8184 
8185 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8186 		if (!net_eq(net, dev_net(iter->dev)))
8187 			continue;
8188 		netdev_adjacent_sysfs_add(iter->dev, dev,
8189 					  &iter->dev->adj_list.lower);
8190 		netdev_adjacent_sysfs_add(dev, iter->dev,
8191 					  &dev->adj_list.upper);
8192 	}
8193 
8194 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8195 		if (!net_eq(net, dev_net(iter->dev)))
8196 			continue;
8197 		netdev_adjacent_sysfs_add(iter->dev, dev,
8198 					  &iter->dev->adj_list.upper);
8199 		netdev_adjacent_sysfs_add(dev, iter->dev,
8200 					  &dev->adj_list.lower);
8201 	}
8202 }
8203 
8204 static void netdev_adjacent_del_links(struct net_device *dev)
8205 {
8206 	struct netdev_adjacent *iter;
8207 
8208 	struct net *net = dev_net(dev);
8209 
8210 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8211 		if (!net_eq(net, dev_net(iter->dev)))
8212 			continue;
8213 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8214 					  &iter->dev->adj_list.lower);
8215 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8216 					  &dev->adj_list.upper);
8217 	}
8218 
8219 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8220 		if (!net_eq(net, dev_net(iter->dev)))
8221 			continue;
8222 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8223 					  &iter->dev->adj_list.upper);
8224 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8225 					  &dev->adj_list.lower);
8226 	}
8227 }
8228 
8229 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8230 {
8231 	struct netdev_adjacent *iter;
8232 
8233 	struct net *net = dev_net(dev);
8234 
8235 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8236 		if (!net_eq(net, dev_net(iter->dev)))
8237 			continue;
8238 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8239 					  &iter->dev->adj_list.lower);
8240 		netdev_adjacent_sysfs_add(iter->dev, dev,
8241 					  &iter->dev->adj_list.lower);
8242 	}
8243 
8244 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8245 		if (!net_eq(net, dev_net(iter->dev)))
8246 			continue;
8247 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8248 					  &iter->dev->adj_list.upper);
8249 		netdev_adjacent_sysfs_add(iter->dev, dev,
8250 					  &iter->dev->adj_list.upper);
8251 	}
8252 }
8253 
8254 void *netdev_lower_dev_get_private(struct net_device *dev,
8255 				   struct net_device *lower_dev)
8256 {
8257 	struct netdev_adjacent *lower;
8258 
8259 	if (!lower_dev)
8260 		return NULL;
8261 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8262 	if (!lower)
8263 		return NULL;
8264 
8265 	return lower->private;
8266 }
8267 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8268 
8269 
8270 /**
8271  * netdev_lower_state_changed - Dispatch event about lower device state change
8272  * @lower_dev: device
8273  * @lower_state_info: state to dispatch
8274  *
8275  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8276  * The caller must hold the RTNL lock.
8277  */
8278 void netdev_lower_state_changed(struct net_device *lower_dev,
8279 				void *lower_state_info)
8280 {
8281 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8282 		.info.dev = lower_dev,
8283 	};
8284 
8285 	ASSERT_RTNL();
8286 	changelowerstate_info.lower_state_info = lower_state_info;
8287 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8288 				      &changelowerstate_info.info);
8289 }
8290 EXPORT_SYMBOL(netdev_lower_state_changed);
8291 
8292 static void dev_change_rx_flags(struct net_device *dev, int flags)
8293 {
8294 	const struct net_device_ops *ops = dev->netdev_ops;
8295 
8296 	if (ops->ndo_change_rx_flags)
8297 		ops->ndo_change_rx_flags(dev, flags);
8298 }
8299 
8300 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8301 {
8302 	unsigned int old_flags = dev->flags;
8303 	kuid_t uid;
8304 	kgid_t gid;
8305 
8306 	ASSERT_RTNL();
8307 
8308 	dev->flags |= IFF_PROMISC;
8309 	dev->promiscuity += inc;
8310 	if (dev->promiscuity == 0) {
8311 		/*
8312 		 * Avoid overflow.
8313 		 * If inc causes overflow, untouch promisc and return error.
8314 		 */
8315 		if (inc < 0)
8316 			dev->flags &= ~IFF_PROMISC;
8317 		else {
8318 			dev->promiscuity -= inc;
8319 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8320 			return -EOVERFLOW;
8321 		}
8322 	}
8323 	if (dev->flags != old_flags) {
8324 		netdev_info(dev, "%s promiscuous mode\n",
8325 			    dev->flags & IFF_PROMISC ? "entered" : "left");
8326 		if (audit_enabled) {
8327 			current_uid_gid(&uid, &gid);
8328 			audit_log(audit_context(), GFP_ATOMIC,
8329 				  AUDIT_ANOM_PROMISCUOUS,
8330 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8331 				  dev->name, (dev->flags & IFF_PROMISC),
8332 				  (old_flags & IFF_PROMISC),
8333 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8334 				  from_kuid(&init_user_ns, uid),
8335 				  from_kgid(&init_user_ns, gid),
8336 				  audit_get_sessionid(current));
8337 		}
8338 
8339 		dev_change_rx_flags(dev, IFF_PROMISC);
8340 	}
8341 	if (notify)
8342 		__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8343 	return 0;
8344 }
8345 
8346 /**
8347  *	dev_set_promiscuity	- update promiscuity count on a device
8348  *	@dev: device
8349  *	@inc: modifier
8350  *
8351  *	Add or remove promiscuity from a device. While the count in the device
8352  *	remains above zero the interface remains promiscuous. Once it hits zero
8353  *	the device reverts back to normal filtering operation. A negative inc
8354  *	value is used to drop promiscuity on the device.
8355  *	Return 0 if successful or a negative errno code on error.
8356  */
8357 int dev_set_promiscuity(struct net_device *dev, int inc)
8358 {
8359 	unsigned int old_flags = dev->flags;
8360 	int err;
8361 
8362 	err = __dev_set_promiscuity(dev, inc, true);
8363 	if (err < 0)
8364 		return err;
8365 	if (dev->flags != old_flags)
8366 		dev_set_rx_mode(dev);
8367 	return err;
8368 }
8369 EXPORT_SYMBOL(dev_set_promiscuity);
8370 
8371 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8372 {
8373 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8374 
8375 	ASSERT_RTNL();
8376 
8377 	dev->flags |= IFF_ALLMULTI;
8378 	dev->allmulti += inc;
8379 	if (dev->allmulti == 0) {
8380 		/*
8381 		 * Avoid overflow.
8382 		 * If inc causes overflow, untouch allmulti and return error.
8383 		 */
8384 		if (inc < 0)
8385 			dev->flags &= ~IFF_ALLMULTI;
8386 		else {
8387 			dev->allmulti -= inc;
8388 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8389 			return -EOVERFLOW;
8390 		}
8391 	}
8392 	if (dev->flags ^ old_flags) {
8393 		netdev_info(dev, "%s allmulticast mode\n",
8394 			    dev->flags & IFF_ALLMULTI ? "entered" : "left");
8395 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8396 		dev_set_rx_mode(dev);
8397 		if (notify)
8398 			__dev_notify_flags(dev, old_flags,
8399 					   dev->gflags ^ old_gflags, 0, NULL);
8400 	}
8401 	return 0;
8402 }
8403 
8404 /**
8405  *	dev_set_allmulti	- update allmulti count on a device
8406  *	@dev: device
8407  *	@inc: modifier
8408  *
8409  *	Add or remove reception of all multicast frames to a device. While the
8410  *	count in the device remains above zero the interface remains listening
8411  *	to all interfaces. Once it hits zero the device reverts back to normal
8412  *	filtering operation. A negative @inc value is used to drop the counter
8413  *	when releasing a resource needing all multicasts.
8414  *	Return 0 if successful or a negative errno code on error.
8415  */
8416 
8417 int dev_set_allmulti(struct net_device *dev, int inc)
8418 {
8419 	return __dev_set_allmulti(dev, inc, true);
8420 }
8421 EXPORT_SYMBOL(dev_set_allmulti);
8422 
8423 /*
8424  *	Upload unicast and multicast address lists to device and
8425  *	configure RX filtering. When the device doesn't support unicast
8426  *	filtering it is put in promiscuous mode while unicast addresses
8427  *	are present.
8428  */
8429 void __dev_set_rx_mode(struct net_device *dev)
8430 {
8431 	const struct net_device_ops *ops = dev->netdev_ops;
8432 
8433 	/* dev_open will call this function so the list will stay sane. */
8434 	if (!(dev->flags&IFF_UP))
8435 		return;
8436 
8437 	if (!netif_device_present(dev))
8438 		return;
8439 
8440 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8441 		/* Unicast addresses changes may only happen under the rtnl,
8442 		 * therefore calling __dev_set_promiscuity here is safe.
8443 		 */
8444 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8445 			__dev_set_promiscuity(dev, 1, false);
8446 			dev->uc_promisc = true;
8447 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8448 			__dev_set_promiscuity(dev, -1, false);
8449 			dev->uc_promisc = false;
8450 		}
8451 	}
8452 
8453 	if (ops->ndo_set_rx_mode)
8454 		ops->ndo_set_rx_mode(dev);
8455 }
8456 
8457 void dev_set_rx_mode(struct net_device *dev)
8458 {
8459 	netif_addr_lock_bh(dev);
8460 	__dev_set_rx_mode(dev);
8461 	netif_addr_unlock_bh(dev);
8462 }
8463 
8464 /**
8465  *	dev_get_flags - get flags reported to userspace
8466  *	@dev: device
8467  *
8468  *	Get the combination of flag bits exported through APIs to userspace.
8469  */
8470 unsigned int dev_get_flags(const struct net_device *dev)
8471 {
8472 	unsigned int flags;
8473 
8474 	flags = (dev->flags & ~(IFF_PROMISC |
8475 				IFF_ALLMULTI |
8476 				IFF_RUNNING |
8477 				IFF_LOWER_UP |
8478 				IFF_DORMANT)) |
8479 		(dev->gflags & (IFF_PROMISC |
8480 				IFF_ALLMULTI));
8481 
8482 	if (netif_running(dev)) {
8483 		if (netif_oper_up(dev))
8484 			flags |= IFF_RUNNING;
8485 		if (netif_carrier_ok(dev))
8486 			flags |= IFF_LOWER_UP;
8487 		if (netif_dormant(dev))
8488 			flags |= IFF_DORMANT;
8489 	}
8490 
8491 	return flags;
8492 }
8493 EXPORT_SYMBOL(dev_get_flags);
8494 
8495 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8496 		       struct netlink_ext_ack *extack)
8497 {
8498 	unsigned int old_flags = dev->flags;
8499 	int ret;
8500 
8501 	ASSERT_RTNL();
8502 
8503 	/*
8504 	 *	Set the flags on our device.
8505 	 */
8506 
8507 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8508 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8509 			       IFF_AUTOMEDIA)) |
8510 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8511 				    IFF_ALLMULTI));
8512 
8513 	/*
8514 	 *	Load in the correct multicast list now the flags have changed.
8515 	 */
8516 
8517 	if ((old_flags ^ flags) & IFF_MULTICAST)
8518 		dev_change_rx_flags(dev, IFF_MULTICAST);
8519 
8520 	dev_set_rx_mode(dev);
8521 
8522 	/*
8523 	 *	Have we downed the interface. We handle IFF_UP ourselves
8524 	 *	according to user attempts to set it, rather than blindly
8525 	 *	setting it.
8526 	 */
8527 
8528 	ret = 0;
8529 	if ((old_flags ^ flags) & IFF_UP) {
8530 		if (old_flags & IFF_UP)
8531 			__dev_close(dev);
8532 		else
8533 			ret = __dev_open(dev, extack);
8534 	}
8535 
8536 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8537 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8538 		unsigned int old_flags = dev->flags;
8539 
8540 		dev->gflags ^= IFF_PROMISC;
8541 
8542 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8543 			if (dev->flags != old_flags)
8544 				dev_set_rx_mode(dev);
8545 	}
8546 
8547 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8548 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8549 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8550 	 */
8551 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8552 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8553 
8554 		dev->gflags ^= IFF_ALLMULTI;
8555 		__dev_set_allmulti(dev, inc, false);
8556 	}
8557 
8558 	return ret;
8559 }
8560 
8561 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8562 			unsigned int gchanges, u32 portid,
8563 			const struct nlmsghdr *nlh)
8564 {
8565 	unsigned int changes = dev->flags ^ old_flags;
8566 
8567 	if (gchanges)
8568 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8569 
8570 	if (changes & IFF_UP) {
8571 		if (dev->flags & IFF_UP)
8572 			call_netdevice_notifiers(NETDEV_UP, dev);
8573 		else
8574 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8575 	}
8576 
8577 	if (dev->flags & IFF_UP &&
8578 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8579 		struct netdev_notifier_change_info change_info = {
8580 			.info = {
8581 				.dev = dev,
8582 			},
8583 			.flags_changed = changes,
8584 		};
8585 
8586 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8587 	}
8588 }
8589 
8590 /**
8591  *	dev_change_flags - change device settings
8592  *	@dev: device
8593  *	@flags: device state flags
8594  *	@extack: netlink extended ack
8595  *
8596  *	Change settings on device based state flags. The flags are
8597  *	in the userspace exported format.
8598  */
8599 int dev_change_flags(struct net_device *dev, unsigned int flags,
8600 		     struct netlink_ext_ack *extack)
8601 {
8602 	int ret;
8603 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8604 
8605 	ret = __dev_change_flags(dev, flags, extack);
8606 	if (ret < 0)
8607 		return ret;
8608 
8609 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8610 	__dev_notify_flags(dev, old_flags, changes, 0, NULL);
8611 	return ret;
8612 }
8613 EXPORT_SYMBOL(dev_change_flags);
8614 
8615 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8616 {
8617 	const struct net_device_ops *ops = dev->netdev_ops;
8618 
8619 	if (ops->ndo_change_mtu)
8620 		return ops->ndo_change_mtu(dev, new_mtu);
8621 
8622 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8623 	WRITE_ONCE(dev->mtu, new_mtu);
8624 	return 0;
8625 }
8626 EXPORT_SYMBOL(__dev_set_mtu);
8627 
8628 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8629 		     struct netlink_ext_ack *extack)
8630 {
8631 	/* MTU must be positive, and in range */
8632 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8633 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8634 		return -EINVAL;
8635 	}
8636 
8637 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8638 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8639 		return -EINVAL;
8640 	}
8641 	return 0;
8642 }
8643 
8644 /**
8645  *	dev_set_mtu_ext - Change maximum transfer unit
8646  *	@dev: device
8647  *	@new_mtu: new transfer unit
8648  *	@extack: netlink extended ack
8649  *
8650  *	Change the maximum transfer size of the network device.
8651  */
8652 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8653 		    struct netlink_ext_ack *extack)
8654 {
8655 	int err, orig_mtu;
8656 
8657 	if (new_mtu == dev->mtu)
8658 		return 0;
8659 
8660 	err = dev_validate_mtu(dev, new_mtu, extack);
8661 	if (err)
8662 		return err;
8663 
8664 	if (!netif_device_present(dev))
8665 		return -ENODEV;
8666 
8667 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8668 	err = notifier_to_errno(err);
8669 	if (err)
8670 		return err;
8671 
8672 	orig_mtu = dev->mtu;
8673 	err = __dev_set_mtu(dev, new_mtu);
8674 
8675 	if (!err) {
8676 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8677 						   orig_mtu);
8678 		err = notifier_to_errno(err);
8679 		if (err) {
8680 			/* setting mtu back and notifying everyone again,
8681 			 * so that they have a chance to revert changes.
8682 			 */
8683 			__dev_set_mtu(dev, orig_mtu);
8684 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8685 						     new_mtu);
8686 		}
8687 	}
8688 	return err;
8689 }
8690 
8691 int dev_set_mtu(struct net_device *dev, int new_mtu)
8692 {
8693 	struct netlink_ext_ack extack;
8694 	int err;
8695 
8696 	memset(&extack, 0, sizeof(extack));
8697 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8698 	if (err && extack._msg)
8699 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8700 	return err;
8701 }
8702 EXPORT_SYMBOL(dev_set_mtu);
8703 
8704 /**
8705  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8706  *	@dev: device
8707  *	@new_len: new tx queue length
8708  */
8709 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8710 {
8711 	unsigned int orig_len = dev->tx_queue_len;
8712 	int res;
8713 
8714 	if (new_len != (unsigned int)new_len)
8715 		return -ERANGE;
8716 
8717 	if (new_len != orig_len) {
8718 		dev->tx_queue_len = new_len;
8719 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8720 		res = notifier_to_errno(res);
8721 		if (res)
8722 			goto err_rollback;
8723 		res = dev_qdisc_change_tx_queue_len(dev);
8724 		if (res)
8725 			goto err_rollback;
8726 	}
8727 
8728 	return 0;
8729 
8730 err_rollback:
8731 	netdev_err(dev, "refused to change device tx_queue_len\n");
8732 	dev->tx_queue_len = orig_len;
8733 	return res;
8734 }
8735 
8736 /**
8737  *	dev_set_group - Change group this device belongs to
8738  *	@dev: device
8739  *	@new_group: group this device should belong to
8740  */
8741 void dev_set_group(struct net_device *dev, int new_group)
8742 {
8743 	dev->group = new_group;
8744 }
8745 
8746 /**
8747  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8748  *	@dev: device
8749  *	@addr: new address
8750  *	@extack: netlink extended ack
8751  */
8752 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8753 			      struct netlink_ext_ack *extack)
8754 {
8755 	struct netdev_notifier_pre_changeaddr_info info = {
8756 		.info.dev = dev,
8757 		.info.extack = extack,
8758 		.dev_addr = addr,
8759 	};
8760 	int rc;
8761 
8762 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8763 	return notifier_to_errno(rc);
8764 }
8765 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8766 
8767 /**
8768  *	dev_set_mac_address - Change Media Access Control Address
8769  *	@dev: device
8770  *	@sa: new address
8771  *	@extack: netlink extended ack
8772  *
8773  *	Change the hardware (MAC) address of the device
8774  */
8775 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8776 			struct netlink_ext_ack *extack)
8777 {
8778 	const struct net_device_ops *ops = dev->netdev_ops;
8779 	int err;
8780 
8781 	if (!ops->ndo_set_mac_address)
8782 		return -EOPNOTSUPP;
8783 	if (sa->sa_family != dev->type)
8784 		return -EINVAL;
8785 	if (!netif_device_present(dev))
8786 		return -ENODEV;
8787 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8788 	if (err)
8789 		return err;
8790 	err = ops->ndo_set_mac_address(dev, sa);
8791 	if (err)
8792 		return err;
8793 	dev->addr_assign_type = NET_ADDR_SET;
8794 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8795 	add_device_randomness(dev->dev_addr, dev->addr_len);
8796 	return 0;
8797 }
8798 EXPORT_SYMBOL(dev_set_mac_address);
8799 
8800 static DECLARE_RWSEM(dev_addr_sem);
8801 
8802 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8803 			     struct netlink_ext_ack *extack)
8804 {
8805 	int ret;
8806 
8807 	down_write(&dev_addr_sem);
8808 	ret = dev_set_mac_address(dev, sa, extack);
8809 	up_write(&dev_addr_sem);
8810 	return ret;
8811 }
8812 EXPORT_SYMBOL(dev_set_mac_address_user);
8813 
8814 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8815 {
8816 	size_t size = sizeof(sa->sa_data_min);
8817 	struct net_device *dev;
8818 	int ret = 0;
8819 
8820 	down_read(&dev_addr_sem);
8821 	rcu_read_lock();
8822 
8823 	dev = dev_get_by_name_rcu(net, dev_name);
8824 	if (!dev) {
8825 		ret = -ENODEV;
8826 		goto unlock;
8827 	}
8828 	if (!dev->addr_len)
8829 		memset(sa->sa_data, 0, size);
8830 	else
8831 		memcpy(sa->sa_data, dev->dev_addr,
8832 		       min_t(size_t, size, dev->addr_len));
8833 	sa->sa_family = dev->type;
8834 
8835 unlock:
8836 	rcu_read_unlock();
8837 	up_read(&dev_addr_sem);
8838 	return ret;
8839 }
8840 EXPORT_SYMBOL(dev_get_mac_address);
8841 
8842 /**
8843  *	dev_change_carrier - Change device carrier
8844  *	@dev: device
8845  *	@new_carrier: new value
8846  *
8847  *	Change device carrier
8848  */
8849 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8850 {
8851 	const struct net_device_ops *ops = dev->netdev_ops;
8852 
8853 	if (!ops->ndo_change_carrier)
8854 		return -EOPNOTSUPP;
8855 	if (!netif_device_present(dev))
8856 		return -ENODEV;
8857 	return ops->ndo_change_carrier(dev, new_carrier);
8858 }
8859 
8860 /**
8861  *	dev_get_phys_port_id - Get device physical port ID
8862  *	@dev: device
8863  *	@ppid: port ID
8864  *
8865  *	Get device physical port ID
8866  */
8867 int dev_get_phys_port_id(struct net_device *dev,
8868 			 struct netdev_phys_item_id *ppid)
8869 {
8870 	const struct net_device_ops *ops = dev->netdev_ops;
8871 
8872 	if (!ops->ndo_get_phys_port_id)
8873 		return -EOPNOTSUPP;
8874 	return ops->ndo_get_phys_port_id(dev, ppid);
8875 }
8876 
8877 /**
8878  *	dev_get_phys_port_name - Get device physical port name
8879  *	@dev: device
8880  *	@name: port name
8881  *	@len: limit of bytes to copy to name
8882  *
8883  *	Get device physical port name
8884  */
8885 int dev_get_phys_port_name(struct net_device *dev,
8886 			   char *name, size_t len)
8887 {
8888 	const struct net_device_ops *ops = dev->netdev_ops;
8889 	int err;
8890 
8891 	if (ops->ndo_get_phys_port_name) {
8892 		err = ops->ndo_get_phys_port_name(dev, name, len);
8893 		if (err != -EOPNOTSUPP)
8894 			return err;
8895 	}
8896 	return devlink_compat_phys_port_name_get(dev, name, len);
8897 }
8898 
8899 /**
8900  *	dev_get_port_parent_id - Get the device's port parent identifier
8901  *	@dev: network device
8902  *	@ppid: pointer to a storage for the port's parent identifier
8903  *	@recurse: allow/disallow recursion to lower devices
8904  *
8905  *	Get the devices's port parent identifier
8906  */
8907 int dev_get_port_parent_id(struct net_device *dev,
8908 			   struct netdev_phys_item_id *ppid,
8909 			   bool recurse)
8910 {
8911 	const struct net_device_ops *ops = dev->netdev_ops;
8912 	struct netdev_phys_item_id first = { };
8913 	struct net_device *lower_dev;
8914 	struct list_head *iter;
8915 	int err;
8916 
8917 	if (ops->ndo_get_port_parent_id) {
8918 		err = ops->ndo_get_port_parent_id(dev, ppid);
8919 		if (err != -EOPNOTSUPP)
8920 			return err;
8921 	}
8922 
8923 	err = devlink_compat_switch_id_get(dev, ppid);
8924 	if (!recurse || err != -EOPNOTSUPP)
8925 		return err;
8926 
8927 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
8928 		err = dev_get_port_parent_id(lower_dev, ppid, true);
8929 		if (err)
8930 			break;
8931 		if (!first.id_len)
8932 			first = *ppid;
8933 		else if (memcmp(&first, ppid, sizeof(*ppid)))
8934 			return -EOPNOTSUPP;
8935 	}
8936 
8937 	return err;
8938 }
8939 EXPORT_SYMBOL(dev_get_port_parent_id);
8940 
8941 /**
8942  *	netdev_port_same_parent_id - Indicate if two network devices have
8943  *	the same port parent identifier
8944  *	@a: first network device
8945  *	@b: second network device
8946  */
8947 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8948 {
8949 	struct netdev_phys_item_id a_id = { };
8950 	struct netdev_phys_item_id b_id = { };
8951 
8952 	if (dev_get_port_parent_id(a, &a_id, true) ||
8953 	    dev_get_port_parent_id(b, &b_id, true))
8954 		return false;
8955 
8956 	return netdev_phys_item_id_same(&a_id, &b_id);
8957 }
8958 EXPORT_SYMBOL(netdev_port_same_parent_id);
8959 
8960 /**
8961  *	dev_change_proto_down - set carrier according to proto_down.
8962  *
8963  *	@dev: device
8964  *	@proto_down: new value
8965  */
8966 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8967 {
8968 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8969 		return -EOPNOTSUPP;
8970 	if (!netif_device_present(dev))
8971 		return -ENODEV;
8972 	if (proto_down)
8973 		netif_carrier_off(dev);
8974 	else
8975 		netif_carrier_on(dev);
8976 	dev->proto_down = proto_down;
8977 	return 0;
8978 }
8979 
8980 /**
8981  *	dev_change_proto_down_reason - proto down reason
8982  *
8983  *	@dev: device
8984  *	@mask: proto down mask
8985  *	@value: proto down value
8986  */
8987 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8988 				  u32 value)
8989 {
8990 	int b;
8991 
8992 	if (!mask) {
8993 		dev->proto_down_reason = value;
8994 	} else {
8995 		for_each_set_bit(b, &mask, 32) {
8996 			if (value & (1 << b))
8997 				dev->proto_down_reason |= BIT(b);
8998 			else
8999 				dev->proto_down_reason &= ~BIT(b);
9000 		}
9001 	}
9002 }
9003 
9004 struct bpf_xdp_link {
9005 	struct bpf_link link;
9006 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9007 	int flags;
9008 };
9009 
9010 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9011 {
9012 	if (flags & XDP_FLAGS_HW_MODE)
9013 		return XDP_MODE_HW;
9014 	if (flags & XDP_FLAGS_DRV_MODE)
9015 		return XDP_MODE_DRV;
9016 	if (flags & XDP_FLAGS_SKB_MODE)
9017 		return XDP_MODE_SKB;
9018 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9019 }
9020 
9021 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9022 {
9023 	switch (mode) {
9024 	case XDP_MODE_SKB:
9025 		return generic_xdp_install;
9026 	case XDP_MODE_DRV:
9027 	case XDP_MODE_HW:
9028 		return dev->netdev_ops->ndo_bpf;
9029 	default:
9030 		return NULL;
9031 	}
9032 }
9033 
9034 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9035 					 enum bpf_xdp_mode mode)
9036 {
9037 	return dev->xdp_state[mode].link;
9038 }
9039 
9040 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9041 				     enum bpf_xdp_mode mode)
9042 {
9043 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9044 
9045 	if (link)
9046 		return link->link.prog;
9047 	return dev->xdp_state[mode].prog;
9048 }
9049 
9050 u8 dev_xdp_prog_count(struct net_device *dev)
9051 {
9052 	u8 count = 0;
9053 	int i;
9054 
9055 	for (i = 0; i < __MAX_XDP_MODE; i++)
9056 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9057 			count++;
9058 	return count;
9059 }
9060 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9061 
9062 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9063 {
9064 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9065 
9066 	return prog ? prog->aux->id : 0;
9067 }
9068 
9069 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9070 			     struct bpf_xdp_link *link)
9071 {
9072 	dev->xdp_state[mode].link = link;
9073 	dev->xdp_state[mode].prog = NULL;
9074 }
9075 
9076 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9077 			     struct bpf_prog *prog)
9078 {
9079 	dev->xdp_state[mode].link = NULL;
9080 	dev->xdp_state[mode].prog = prog;
9081 }
9082 
9083 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9084 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9085 			   u32 flags, struct bpf_prog *prog)
9086 {
9087 	struct netdev_bpf xdp;
9088 	int err;
9089 
9090 	memset(&xdp, 0, sizeof(xdp));
9091 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9092 	xdp.extack = extack;
9093 	xdp.flags = flags;
9094 	xdp.prog = prog;
9095 
9096 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9097 	 * "moved" into driver), so they don't increment it on their own, but
9098 	 * they do decrement refcnt when program is detached or replaced.
9099 	 * Given net_device also owns link/prog, we need to bump refcnt here
9100 	 * to prevent drivers from underflowing it.
9101 	 */
9102 	if (prog)
9103 		bpf_prog_inc(prog);
9104 	err = bpf_op(dev, &xdp);
9105 	if (err) {
9106 		if (prog)
9107 			bpf_prog_put(prog);
9108 		return err;
9109 	}
9110 
9111 	if (mode != XDP_MODE_HW)
9112 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9113 
9114 	return 0;
9115 }
9116 
9117 static void dev_xdp_uninstall(struct net_device *dev)
9118 {
9119 	struct bpf_xdp_link *link;
9120 	struct bpf_prog *prog;
9121 	enum bpf_xdp_mode mode;
9122 	bpf_op_t bpf_op;
9123 
9124 	ASSERT_RTNL();
9125 
9126 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9127 		prog = dev_xdp_prog(dev, mode);
9128 		if (!prog)
9129 			continue;
9130 
9131 		bpf_op = dev_xdp_bpf_op(dev, mode);
9132 		if (!bpf_op)
9133 			continue;
9134 
9135 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9136 
9137 		/* auto-detach link from net device */
9138 		link = dev_xdp_link(dev, mode);
9139 		if (link)
9140 			link->dev = NULL;
9141 		else
9142 			bpf_prog_put(prog);
9143 
9144 		dev_xdp_set_link(dev, mode, NULL);
9145 	}
9146 }
9147 
9148 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9149 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9150 			  struct bpf_prog *old_prog, u32 flags)
9151 {
9152 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9153 	struct bpf_prog *cur_prog;
9154 	struct net_device *upper;
9155 	struct list_head *iter;
9156 	enum bpf_xdp_mode mode;
9157 	bpf_op_t bpf_op;
9158 	int err;
9159 
9160 	ASSERT_RTNL();
9161 
9162 	/* either link or prog attachment, never both */
9163 	if (link && (new_prog || old_prog))
9164 		return -EINVAL;
9165 	/* link supports only XDP mode flags */
9166 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9167 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9168 		return -EINVAL;
9169 	}
9170 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9171 	if (num_modes > 1) {
9172 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9173 		return -EINVAL;
9174 	}
9175 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9176 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9177 		NL_SET_ERR_MSG(extack,
9178 			       "More than one program loaded, unset mode is ambiguous");
9179 		return -EINVAL;
9180 	}
9181 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9182 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9183 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9184 		return -EINVAL;
9185 	}
9186 
9187 	mode = dev_xdp_mode(dev, flags);
9188 	/* can't replace attached link */
9189 	if (dev_xdp_link(dev, mode)) {
9190 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9191 		return -EBUSY;
9192 	}
9193 
9194 	/* don't allow if an upper device already has a program */
9195 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9196 		if (dev_xdp_prog_count(upper) > 0) {
9197 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9198 			return -EEXIST;
9199 		}
9200 	}
9201 
9202 	cur_prog = dev_xdp_prog(dev, mode);
9203 	/* can't replace attached prog with link */
9204 	if (link && cur_prog) {
9205 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9206 		return -EBUSY;
9207 	}
9208 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9209 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9210 		return -EEXIST;
9211 	}
9212 
9213 	/* put effective new program into new_prog */
9214 	if (link)
9215 		new_prog = link->link.prog;
9216 
9217 	if (new_prog) {
9218 		bool offload = mode == XDP_MODE_HW;
9219 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9220 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9221 
9222 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9223 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9224 			return -EBUSY;
9225 		}
9226 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9227 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9228 			return -EEXIST;
9229 		}
9230 		if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9231 			NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9232 			return -EINVAL;
9233 		}
9234 		if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9235 			NL_SET_ERR_MSG(extack, "Program bound to different device");
9236 			return -EINVAL;
9237 		}
9238 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9239 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9240 			return -EINVAL;
9241 		}
9242 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9243 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9244 			return -EINVAL;
9245 		}
9246 	}
9247 
9248 	/* don't call drivers if the effective program didn't change */
9249 	if (new_prog != cur_prog) {
9250 		bpf_op = dev_xdp_bpf_op(dev, mode);
9251 		if (!bpf_op) {
9252 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9253 			return -EOPNOTSUPP;
9254 		}
9255 
9256 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9257 		if (err)
9258 			return err;
9259 	}
9260 
9261 	if (link)
9262 		dev_xdp_set_link(dev, mode, link);
9263 	else
9264 		dev_xdp_set_prog(dev, mode, new_prog);
9265 	if (cur_prog)
9266 		bpf_prog_put(cur_prog);
9267 
9268 	return 0;
9269 }
9270 
9271 static int dev_xdp_attach_link(struct net_device *dev,
9272 			       struct netlink_ext_ack *extack,
9273 			       struct bpf_xdp_link *link)
9274 {
9275 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9276 }
9277 
9278 static int dev_xdp_detach_link(struct net_device *dev,
9279 			       struct netlink_ext_ack *extack,
9280 			       struct bpf_xdp_link *link)
9281 {
9282 	enum bpf_xdp_mode mode;
9283 	bpf_op_t bpf_op;
9284 
9285 	ASSERT_RTNL();
9286 
9287 	mode = dev_xdp_mode(dev, link->flags);
9288 	if (dev_xdp_link(dev, mode) != link)
9289 		return -EINVAL;
9290 
9291 	bpf_op = dev_xdp_bpf_op(dev, mode);
9292 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9293 	dev_xdp_set_link(dev, mode, NULL);
9294 	return 0;
9295 }
9296 
9297 static void bpf_xdp_link_release(struct bpf_link *link)
9298 {
9299 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9300 
9301 	rtnl_lock();
9302 
9303 	/* if racing with net_device's tear down, xdp_link->dev might be
9304 	 * already NULL, in which case link was already auto-detached
9305 	 */
9306 	if (xdp_link->dev) {
9307 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9308 		xdp_link->dev = NULL;
9309 	}
9310 
9311 	rtnl_unlock();
9312 }
9313 
9314 static int bpf_xdp_link_detach(struct bpf_link *link)
9315 {
9316 	bpf_xdp_link_release(link);
9317 	return 0;
9318 }
9319 
9320 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9321 {
9322 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9323 
9324 	kfree(xdp_link);
9325 }
9326 
9327 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9328 				     struct seq_file *seq)
9329 {
9330 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9331 	u32 ifindex = 0;
9332 
9333 	rtnl_lock();
9334 	if (xdp_link->dev)
9335 		ifindex = xdp_link->dev->ifindex;
9336 	rtnl_unlock();
9337 
9338 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9339 }
9340 
9341 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9342 				       struct bpf_link_info *info)
9343 {
9344 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9345 	u32 ifindex = 0;
9346 
9347 	rtnl_lock();
9348 	if (xdp_link->dev)
9349 		ifindex = xdp_link->dev->ifindex;
9350 	rtnl_unlock();
9351 
9352 	info->xdp.ifindex = ifindex;
9353 	return 0;
9354 }
9355 
9356 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9357 			       struct bpf_prog *old_prog)
9358 {
9359 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9360 	enum bpf_xdp_mode mode;
9361 	bpf_op_t bpf_op;
9362 	int err = 0;
9363 
9364 	rtnl_lock();
9365 
9366 	/* link might have been auto-released already, so fail */
9367 	if (!xdp_link->dev) {
9368 		err = -ENOLINK;
9369 		goto out_unlock;
9370 	}
9371 
9372 	if (old_prog && link->prog != old_prog) {
9373 		err = -EPERM;
9374 		goto out_unlock;
9375 	}
9376 	old_prog = link->prog;
9377 	if (old_prog->type != new_prog->type ||
9378 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9379 		err = -EINVAL;
9380 		goto out_unlock;
9381 	}
9382 
9383 	if (old_prog == new_prog) {
9384 		/* no-op, don't disturb drivers */
9385 		bpf_prog_put(new_prog);
9386 		goto out_unlock;
9387 	}
9388 
9389 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9390 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9391 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9392 			      xdp_link->flags, new_prog);
9393 	if (err)
9394 		goto out_unlock;
9395 
9396 	old_prog = xchg(&link->prog, new_prog);
9397 	bpf_prog_put(old_prog);
9398 
9399 out_unlock:
9400 	rtnl_unlock();
9401 	return err;
9402 }
9403 
9404 static const struct bpf_link_ops bpf_xdp_link_lops = {
9405 	.release = bpf_xdp_link_release,
9406 	.dealloc = bpf_xdp_link_dealloc,
9407 	.detach = bpf_xdp_link_detach,
9408 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9409 	.fill_link_info = bpf_xdp_link_fill_link_info,
9410 	.update_prog = bpf_xdp_link_update,
9411 };
9412 
9413 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9414 {
9415 	struct net *net = current->nsproxy->net_ns;
9416 	struct bpf_link_primer link_primer;
9417 	struct bpf_xdp_link *link;
9418 	struct net_device *dev;
9419 	int err, fd;
9420 
9421 	rtnl_lock();
9422 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9423 	if (!dev) {
9424 		rtnl_unlock();
9425 		return -EINVAL;
9426 	}
9427 
9428 	link = kzalloc(sizeof(*link), GFP_USER);
9429 	if (!link) {
9430 		err = -ENOMEM;
9431 		goto unlock;
9432 	}
9433 
9434 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9435 	link->dev = dev;
9436 	link->flags = attr->link_create.flags;
9437 
9438 	err = bpf_link_prime(&link->link, &link_primer);
9439 	if (err) {
9440 		kfree(link);
9441 		goto unlock;
9442 	}
9443 
9444 	err = dev_xdp_attach_link(dev, NULL, link);
9445 	rtnl_unlock();
9446 
9447 	if (err) {
9448 		link->dev = NULL;
9449 		bpf_link_cleanup(&link_primer);
9450 		goto out_put_dev;
9451 	}
9452 
9453 	fd = bpf_link_settle(&link_primer);
9454 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9455 	dev_put(dev);
9456 	return fd;
9457 
9458 unlock:
9459 	rtnl_unlock();
9460 
9461 out_put_dev:
9462 	dev_put(dev);
9463 	return err;
9464 }
9465 
9466 /**
9467  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9468  *	@dev: device
9469  *	@extack: netlink extended ack
9470  *	@fd: new program fd or negative value to clear
9471  *	@expected_fd: old program fd that userspace expects to replace or clear
9472  *	@flags: xdp-related flags
9473  *
9474  *	Set or clear a bpf program for a device
9475  */
9476 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9477 		      int fd, int expected_fd, u32 flags)
9478 {
9479 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9480 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9481 	int err;
9482 
9483 	ASSERT_RTNL();
9484 
9485 	if (fd >= 0) {
9486 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9487 						 mode != XDP_MODE_SKB);
9488 		if (IS_ERR(new_prog))
9489 			return PTR_ERR(new_prog);
9490 	}
9491 
9492 	if (expected_fd >= 0) {
9493 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9494 						 mode != XDP_MODE_SKB);
9495 		if (IS_ERR(old_prog)) {
9496 			err = PTR_ERR(old_prog);
9497 			old_prog = NULL;
9498 			goto err_out;
9499 		}
9500 	}
9501 
9502 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9503 
9504 err_out:
9505 	if (err && new_prog)
9506 		bpf_prog_put(new_prog);
9507 	if (old_prog)
9508 		bpf_prog_put(old_prog);
9509 	return err;
9510 }
9511 
9512 /**
9513  *	dev_new_index	-	allocate an ifindex
9514  *	@net: the applicable net namespace
9515  *
9516  *	Returns a suitable unique value for a new device interface
9517  *	number.  The caller must hold the rtnl semaphore or the
9518  *	dev_base_lock to be sure it remains unique.
9519  */
9520 static int dev_new_index(struct net *net)
9521 {
9522 	int ifindex = net->ifindex;
9523 
9524 	for (;;) {
9525 		if (++ifindex <= 0)
9526 			ifindex = 1;
9527 		if (!__dev_get_by_index(net, ifindex))
9528 			return net->ifindex = ifindex;
9529 	}
9530 }
9531 
9532 /* Delayed registration/unregisteration */
9533 LIST_HEAD(net_todo_list);
9534 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9535 
9536 static void net_set_todo(struct net_device *dev)
9537 {
9538 	list_add_tail(&dev->todo_list, &net_todo_list);
9539 	atomic_inc(&dev_net(dev)->dev_unreg_count);
9540 }
9541 
9542 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9543 	struct net_device *upper, netdev_features_t features)
9544 {
9545 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9546 	netdev_features_t feature;
9547 	int feature_bit;
9548 
9549 	for_each_netdev_feature(upper_disables, feature_bit) {
9550 		feature = __NETIF_F_BIT(feature_bit);
9551 		if (!(upper->wanted_features & feature)
9552 		    && (features & feature)) {
9553 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9554 				   &feature, upper->name);
9555 			features &= ~feature;
9556 		}
9557 	}
9558 
9559 	return features;
9560 }
9561 
9562 static void netdev_sync_lower_features(struct net_device *upper,
9563 	struct net_device *lower, netdev_features_t features)
9564 {
9565 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9566 	netdev_features_t feature;
9567 	int feature_bit;
9568 
9569 	for_each_netdev_feature(upper_disables, feature_bit) {
9570 		feature = __NETIF_F_BIT(feature_bit);
9571 		if (!(features & feature) && (lower->features & feature)) {
9572 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9573 				   &feature, lower->name);
9574 			lower->wanted_features &= ~feature;
9575 			__netdev_update_features(lower);
9576 
9577 			if (unlikely(lower->features & feature))
9578 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9579 					    &feature, lower->name);
9580 			else
9581 				netdev_features_change(lower);
9582 		}
9583 	}
9584 }
9585 
9586 static netdev_features_t netdev_fix_features(struct net_device *dev,
9587 	netdev_features_t features)
9588 {
9589 	/* Fix illegal checksum combinations */
9590 	if ((features & NETIF_F_HW_CSUM) &&
9591 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9592 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9593 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9594 	}
9595 
9596 	/* TSO requires that SG is present as well. */
9597 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9598 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9599 		features &= ~NETIF_F_ALL_TSO;
9600 	}
9601 
9602 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9603 					!(features & NETIF_F_IP_CSUM)) {
9604 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9605 		features &= ~NETIF_F_TSO;
9606 		features &= ~NETIF_F_TSO_ECN;
9607 	}
9608 
9609 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9610 					 !(features & NETIF_F_IPV6_CSUM)) {
9611 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9612 		features &= ~NETIF_F_TSO6;
9613 	}
9614 
9615 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9616 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9617 		features &= ~NETIF_F_TSO_MANGLEID;
9618 
9619 	/* TSO ECN requires that TSO is present as well. */
9620 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9621 		features &= ~NETIF_F_TSO_ECN;
9622 
9623 	/* Software GSO depends on SG. */
9624 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9625 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9626 		features &= ~NETIF_F_GSO;
9627 	}
9628 
9629 	/* GSO partial features require GSO partial be set */
9630 	if ((features & dev->gso_partial_features) &&
9631 	    !(features & NETIF_F_GSO_PARTIAL)) {
9632 		netdev_dbg(dev,
9633 			   "Dropping partially supported GSO features since no GSO partial.\n");
9634 		features &= ~dev->gso_partial_features;
9635 	}
9636 
9637 	if (!(features & NETIF_F_RXCSUM)) {
9638 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9639 		 * successfully merged by hardware must also have the
9640 		 * checksum verified by hardware.  If the user does not
9641 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9642 		 */
9643 		if (features & NETIF_F_GRO_HW) {
9644 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9645 			features &= ~NETIF_F_GRO_HW;
9646 		}
9647 	}
9648 
9649 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9650 	if (features & NETIF_F_RXFCS) {
9651 		if (features & NETIF_F_LRO) {
9652 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9653 			features &= ~NETIF_F_LRO;
9654 		}
9655 
9656 		if (features & NETIF_F_GRO_HW) {
9657 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9658 			features &= ~NETIF_F_GRO_HW;
9659 		}
9660 	}
9661 
9662 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9663 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9664 		features &= ~NETIF_F_LRO;
9665 	}
9666 
9667 	if (features & NETIF_F_HW_TLS_TX) {
9668 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9669 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9670 		bool hw_csum = features & NETIF_F_HW_CSUM;
9671 
9672 		if (!ip_csum && !hw_csum) {
9673 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9674 			features &= ~NETIF_F_HW_TLS_TX;
9675 		}
9676 	}
9677 
9678 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9679 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9680 		features &= ~NETIF_F_HW_TLS_RX;
9681 	}
9682 
9683 	return features;
9684 }
9685 
9686 int __netdev_update_features(struct net_device *dev)
9687 {
9688 	struct net_device *upper, *lower;
9689 	netdev_features_t features;
9690 	struct list_head *iter;
9691 	int err = -1;
9692 
9693 	ASSERT_RTNL();
9694 
9695 	features = netdev_get_wanted_features(dev);
9696 
9697 	if (dev->netdev_ops->ndo_fix_features)
9698 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9699 
9700 	/* driver might be less strict about feature dependencies */
9701 	features = netdev_fix_features(dev, features);
9702 
9703 	/* some features can't be enabled if they're off on an upper device */
9704 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9705 		features = netdev_sync_upper_features(dev, upper, features);
9706 
9707 	if (dev->features == features)
9708 		goto sync_lower;
9709 
9710 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9711 		&dev->features, &features);
9712 
9713 	if (dev->netdev_ops->ndo_set_features)
9714 		err = dev->netdev_ops->ndo_set_features(dev, features);
9715 	else
9716 		err = 0;
9717 
9718 	if (unlikely(err < 0)) {
9719 		netdev_err(dev,
9720 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
9721 			err, &features, &dev->features);
9722 		/* return non-0 since some features might have changed and
9723 		 * it's better to fire a spurious notification than miss it
9724 		 */
9725 		return -1;
9726 	}
9727 
9728 sync_lower:
9729 	/* some features must be disabled on lower devices when disabled
9730 	 * on an upper device (think: bonding master or bridge)
9731 	 */
9732 	netdev_for_each_lower_dev(dev, lower, iter)
9733 		netdev_sync_lower_features(dev, lower, features);
9734 
9735 	if (!err) {
9736 		netdev_features_t diff = features ^ dev->features;
9737 
9738 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9739 			/* udp_tunnel_{get,drop}_rx_info both need
9740 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9741 			 * device, or they won't do anything.
9742 			 * Thus we need to update dev->features
9743 			 * *before* calling udp_tunnel_get_rx_info,
9744 			 * but *after* calling udp_tunnel_drop_rx_info.
9745 			 */
9746 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9747 				dev->features = features;
9748 				udp_tunnel_get_rx_info(dev);
9749 			} else {
9750 				udp_tunnel_drop_rx_info(dev);
9751 			}
9752 		}
9753 
9754 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9755 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9756 				dev->features = features;
9757 				err |= vlan_get_rx_ctag_filter_info(dev);
9758 			} else {
9759 				vlan_drop_rx_ctag_filter_info(dev);
9760 			}
9761 		}
9762 
9763 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9764 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9765 				dev->features = features;
9766 				err |= vlan_get_rx_stag_filter_info(dev);
9767 			} else {
9768 				vlan_drop_rx_stag_filter_info(dev);
9769 			}
9770 		}
9771 
9772 		dev->features = features;
9773 	}
9774 
9775 	return err < 0 ? 0 : 1;
9776 }
9777 
9778 /**
9779  *	netdev_update_features - recalculate device features
9780  *	@dev: the device to check
9781  *
9782  *	Recalculate dev->features set and send notifications if it
9783  *	has changed. Should be called after driver or hardware dependent
9784  *	conditions might have changed that influence the features.
9785  */
9786 void netdev_update_features(struct net_device *dev)
9787 {
9788 	if (__netdev_update_features(dev))
9789 		netdev_features_change(dev);
9790 }
9791 EXPORT_SYMBOL(netdev_update_features);
9792 
9793 /**
9794  *	netdev_change_features - recalculate device features
9795  *	@dev: the device to check
9796  *
9797  *	Recalculate dev->features set and send notifications even
9798  *	if they have not changed. Should be called instead of
9799  *	netdev_update_features() if also dev->vlan_features might
9800  *	have changed to allow the changes to be propagated to stacked
9801  *	VLAN devices.
9802  */
9803 void netdev_change_features(struct net_device *dev)
9804 {
9805 	__netdev_update_features(dev);
9806 	netdev_features_change(dev);
9807 }
9808 EXPORT_SYMBOL(netdev_change_features);
9809 
9810 /**
9811  *	netif_stacked_transfer_operstate -	transfer operstate
9812  *	@rootdev: the root or lower level device to transfer state from
9813  *	@dev: the device to transfer operstate to
9814  *
9815  *	Transfer operational state from root to device. This is normally
9816  *	called when a stacking relationship exists between the root
9817  *	device and the device(a leaf device).
9818  */
9819 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9820 					struct net_device *dev)
9821 {
9822 	if (rootdev->operstate == IF_OPER_DORMANT)
9823 		netif_dormant_on(dev);
9824 	else
9825 		netif_dormant_off(dev);
9826 
9827 	if (rootdev->operstate == IF_OPER_TESTING)
9828 		netif_testing_on(dev);
9829 	else
9830 		netif_testing_off(dev);
9831 
9832 	if (netif_carrier_ok(rootdev))
9833 		netif_carrier_on(dev);
9834 	else
9835 		netif_carrier_off(dev);
9836 }
9837 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9838 
9839 static int netif_alloc_rx_queues(struct net_device *dev)
9840 {
9841 	unsigned int i, count = dev->num_rx_queues;
9842 	struct netdev_rx_queue *rx;
9843 	size_t sz = count * sizeof(*rx);
9844 	int err = 0;
9845 
9846 	BUG_ON(count < 1);
9847 
9848 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9849 	if (!rx)
9850 		return -ENOMEM;
9851 
9852 	dev->_rx = rx;
9853 
9854 	for (i = 0; i < count; i++) {
9855 		rx[i].dev = dev;
9856 
9857 		/* XDP RX-queue setup */
9858 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9859 		if (err < 0)
9860 			goto err_rxq_info;
9861 	}
9862 	return 0;
9863 
9864 err_rxq_info:
9865 	/* Rollback successful reg's and free other resources */
9866 	while (i--)
9867 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9868 	kvfree(dev->_rx);
9869 	dev->_rx = NULL;
9870 	return err;
9871 }
9872 
9873 static void netif_free_rx_queues(struct net_device *dev)
9874 {
9875 	unsigned int i, count = dev->num_rx_queues;
9876 
9877 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9878 	if (!dev->_rx)
9879 		return;
9880 
9881 	for (i = 0; i < count; i++)
9882 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9883 
9884 	kvfree(dev->_rx);
9885 }
9886 
9887 static void netdev_init_one_queue(struct net_device *dev,
9888 				  struct netdev_queue *queue, void *_unused)
9889 {
9890 	/* Initialize queue lock */
9891 	spin_lock_init(&queue->_xmit_lock);
9892 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9893 	queue->xmit_lock_owner = -1;
9894 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9895 	queue->dev = dev;
9896 #ifdef CONFIG_BQL
9897 	dql_init(&queue->dql, HZ);
9898 #endif
9899 }
9900 
9901 static void netif_free_tx_queues(struct net_device *dev)
9902 {
9903 	kvfree(dev->_tx);
9904 }
9905 
9906 static int netif_alloc_netdev_queues(struct net_device *dev)
9907 {
9908 	unsigned int count = dev->num_tx_queues;
9909 	struct netdev_queue *tx;
9910 	size_t sz = count * sizeof(*tx);
9911 
9912 	if (count < 1 || count > 0xffff)
9913 		return -EINVAL;
9914 
9915 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9916 	if (!tx)
9917 		return -ENOMEM;
9918 
9919 	dev->_tx = tx;
9920 
9921 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9922 	spin_lock_init(&dev->tx_global_lock);
9923 
9924 	return 0;
9925 }
9926 
9927 void netif_tx_stop_all_queues(struct net_device *dev)
9928 {
9929 	unsigned int i;
9930 
9931 	for (i = 0; i < dev->num_tx_queues; i++) {
9932 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9933 
9934 		netif_tx_stop_queue(txq);
9935 	}
9936 }
9937 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9938 
9939 /**
9940  * register_netdevice() - register a network device
9941  * @dev: device to register
9942  *
9943  * Take a prepared network device structure and make it externally accessible.
9944  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9945  * Callers must hold the rtnl lock - you may want register_netdev()
9946  * instead of this.
9947  */
9948 int register_netdevice(struct net_device *dev)
9949 {
9950 	int ret;
9951 	struct net *net = dev_net(dev);
9952 
9953 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9954 		     NETDEV_FEATURE_COUNT);
9955 	BUG_ON(dev_boot_phase);
9956 	ASSERT_RTNL();
9957 
9958 	might_sleep();
9959 
9960 	/* When net_device's are persistent, this will be fatal. */
9961 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9962 	BUG_ON(!net);
9963 
9964 	ret = ethtool_check_ops(dev->ethtool_ops);
9965 	if (ret)
9966 		return ret;
9967 
9968 	spin_lock_init(&dev->addr_list_lock);
9969 	netdev_set_addr_lockdep_class(dev);
9970 
9971 	ret = dev_get_valid_name(net, dev, dev->name);
9972 	if (ret < 0)
9973 		goto out;
9974 
9975 	ret = -ENOMEM;
9976 	dev->name_node = netdev_name_node_head_alloc(dev);
9977 	if (!dev->name_node)
9978 		goto out;
9979 
9980 	/* Init, if this function is available */
9981 	if (dev->netdev_ops->ndo_init) {
9982 		ret = dev->netdev_ops->ndo_init(dev);
9983 		if (ret) {
9984 			if (ret > 0)
9985 				ret = -EIO;
9986 			goto err_free_name;
9987 		}
9988 	}
9989 
9990 	if (((dev->hw_features | dev->features) &
9991 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
9992 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9993 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9994 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9995 		ret = -EINVAL;
9996 		goto err_uninit;
9997 	}
9998 
9999 	ret = -EBUSY;
10000 	if (!dev->ifindex)
10001 		dev->ifindex = dev_new_index(net);
10002 	else if (__dev_get_by_index(net, dev->ifindex))
10003 		goto err_uninit;
10004 
10005 	/* Transfer changeable features to wanted_features and enable
10006 	 * software offloads (GSO and GRO).
10007 	 */
10008 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10009 	dev->features |= NETIF_F_SOFT_FEATURES;
10010 
10011 	if (dev->udp_tunnel_nic_info) {
10012 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10013 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10014 	}
10015 
10016 	dev->wanted_features = dev->features & dev->hw_features;
10017 
10018 	if (!(dev->flags & IFF_LOOPBACK))
10019 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10020 
10021 	/* If IPv4 TCP segmentation offload is supported we should also
10022 	 * allow the device to enable segmenting the frame with the option
10023 	 * of ignoring a static IP ID value.  This doesn't enable the
10024 	 * feature itself but allows the user to enable it later.
10025 	 */
10026 	if (dev->hw_features & NETIF_F_TSO)
10027 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10028 	if (dev->vlan_features & NETIF_F_TSO)
10029 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10030 	if (dev->mpls_features & NETIF_F_TSO)
10031 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10032 	if (dev->hw_enc_features & NETIF_F_TSO)
10033 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10034 
10035 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10036 	 */
10037 	dev->vlan_features |= NETIF_F_HIGHDMA;
10038 
10039 	/* Make NETIF_F_SG inheritable to tunnel devices.
10040 	 */
10041 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10042 
10043 	/* Make NETIF_F_SG inheritable to MPLS.
10044 	 */
10045 	dev->mpls_features |= NETIF_F_SG;
10046 
10047 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10048 	ret = notifier_to_errno(ret);
10049 	if (ret)
10050 		goto err_uninit;
10051 
10052 	ret = netdev_register_kobject(dev);
10053 	write_lock(&dev_base_lock);
10054 	dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10055 	write_unlock(&dev_base_lock);
10056 	if (ret)
10057 		goto err_uninit_notify;
10058 
10059 	__netdev_update_features(dev);
10060 
10061 	/*
10062 	 *	Default initial state at registry is that the
10063 	 *	device is present.
10064 	 */
10065 
10066 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10067 
10068 	linkwatch_init_dev(dev);
10069 
10070 	dev_init_scheduler(dev);
10071 
10072 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10073 	list_netdevice(dev);
10074 
10075 	add_device_randomness(dev->dev_addr, dev->addr_len);
10076 
10077 	/* If the device has permanent device address, driver should
10078 	 * set dev_addr and also addr_assign_type should be set to
10079 	 * NET_ADDR_PERM (default value).
10080 	 */
10081 	if (dev->addr_assign_type == NET_ADDR_PERM)
10082 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10083 
10084 	/* Notify protocols, that a new device appeared. */
10085 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10086 	ret = notifier_to_errno(ret);
10087 	if (ret) {
10088 		/* Expect explicit free_netdev() on failure */
10089 		dev->needs_free_netdev = false;
10090 		unregister_netdevice_queue(dev, NULL);
10091 		goto out;
10092 	}
10093 	/*
10094 	 *	Prevent userspace races by waiting until the network
10095 	 *	device is fully setup before sending notifications.
10096 	 */
10097 	if (!dev->rtnl_link_ops ||
10098 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10099 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10100 
10101 out:
10102 	return ret;
10103 
10104 err_uninit_notify:
10105 	call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10106 err_uninit:
10107 	if (dev->netdev_ops->ndo_uninit)
10108 		dev->netdev_ops->ndo_uninit(dev);
10109 	if (dev->priv_destructor)
10110 		dev->priv_destructor(dev);
10111 err_free_name:
10112 	netdev_name_node_free(dev->name_node);
10113 	goto out;
10114 }
10115 EXPORT_SYMBOL(register_netdevice);
10116 
10117 /**
10118  *	init_dummy_netdev	- init a dummy network device for NAPI
10119  *	@dev: device to init
10120  *
10121  *	This takes a network device structure and initialize the minimum
10122  *	amount of fields so it can be used to schedule NAPI polls without
10123  *	registering a full blown interface. This is to be used by drivers
10124  *	that need to tie several hardware interfaces to a single NAPI
10125  *	poll scheduler due to HW limitations.
10126  */
10127 int init_dummy_netdev(struct net_device *dev)
10128 {
10129 	/* Clear everything. Note we don't initialize spinlocks
10130 	 * are they aren't supposed to be taken by any of the
10131 	 * NAPI code and this dummy netdev is supposed to be
10132 	 * only ever used for NAPI polls
10133 	 */
10134 	memset(dev, 0, sizeof(struct net_device));
10135 
10136 	/* make sure we BUG if trying to hit standard
10137 	 * register/unregister code path
10138 	 */
10139 	dev->reg_state = NETREG_DUMMY;
10140 
10141 	/* NAPI wants this */
10142 	INIT_LIST_HEAD(&dev->napi_list);
10143 
10144 	/* a dummy interface is started by default */
10145 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10146 	set_bit(__LINK_STATE_START, &dev->state);
10147 
10148 	/* napi_busy_loop stats accounting wants this */
10149 	dev_net_set(dev, &init_net);
10150 
10151 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10152 	 * because users of this 'device' dont need to change
10153 	 * its refcount.
10154 	 */
10155 
10156 	return 0;
10157 }
10158 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10159 
10160 
10161 /**
10162  *	register_netdev	- register a network device
10163  *	@dev: device to register
10164  *
10165  *	Take a completed network device structure and add it to the kernel
10166  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10167  *	chain. 0 is returned on success. A negative errno code is returned
10168  *	on a failure to set up the device, or if the name is a duplicate.
10169  *
10170  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10171  *	and expands the device name if you passed a format string to
10172  *	alloc_netdev.
10173  */
10174 int register_netdev(struct net_device *dev)
10175 {
10176 	int err;
10177 
10178 	if (rtnl_lock_killable())
10179 		return -EINTR;
10180 	err = register_netdevice(dev);
10181 	rtnl_unlock();
10182 	return err;
10183 }
10184 EXPORT_SYMBOL(register_netdev);
10185 
10186 int netdev_refcnt_read(const struct net_device *dev)
10187 {
10188 #ifdef CONFIG_PCPU_DEV_REFCNT
10189 	int i, refcnt = 0;
10190 
10191 	for_each_possible_cpu(i)
10192 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10193 	return refcnt;
10194 #else
10195 	return refcount_read(&dev->dev_refcnt);
10196 #endif
10197 }
10198 EXPORT_SYMBOL(netdev_refcnt_read);
10199 
10200 int netdev_unregister_timeout_secs __read_mostly = 10;
10201 
10202 #define WAIT_REFS_MIN_MSECS 1
10203 #define WAIT_REFS_MAX_MSECS 250
10204 /**
10205  * netdev_wait_allrefs_any - wait until all references are gone.
10206  * @list: list of net_devices to wait on
10207  *
10208  * This is called when unregistering network devices.
10209  *
10210  * Any protocol or device that holds a reference should register
10211  * for netdevice notification, and cleanup and put back the
10212  * reference if they receive an UNREGISTER event.
10213  * We can get stuck here if buggy protocols don't correctly
10214  * call dev_put.
10215  */
10216 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10217 {
10218 	unsigned long rebroadcast_time, warning_time;
10219 	struct net_device *dev;
10220 	int wait = 0;
10221 
10222 	rebroadcast_time = warning_time = jiffies;
10223 
10224 	list_for_each_entry(dev, list, todo_list)
10225 		if (netdev_refcnt_read(dev) == 1)
10226 			return dev;
10227 
10228 	while (true) {
10229 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10230 			rtnl_lock();
10231 
10232 			/* Rebroadcast unregister notification */
10233 			list_for_each_entry(dev, list, todo_list)
10234 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10235 
10236 			__rtnl_unlock();
10237 			rcu_barrier();
10238 			rtnl_lock();
10239 
10240 			list_for_each_entry(dev, list, todo_list)
10241 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10242 					     &dev->state)) {
10243 					/* We must not have linkwatch events
10244 					 * pending on unregister. If this
10245 					 * happens, we simply run the queue
10246 					 * unscheduled, resulting in a noop
10247 					 * for this device.
10248 					 */
10249 					linkwatch_run_queue();
10250 					break;
10251 				}
10252 
10253 			__rtnl_unlock();
10254 
10255 			rebroadcast_time = jiffies;
10256 		}
10257 
10258 		if (!wait) {
10259 			rcu_barrier();
10260 			wait = WAIT_REFS_MIN_MSECS;
10261 		} else {
10262 			msleep(wait);
10263 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10264 		}
10265 
10266 		list_for_each_entry(dev, list, todo_list)
10267 			if (netdev_refcnt_read(dev) == 1)
10268 				return dev;
10269 
10270 		if (time_after(jiffies, warning_time +
10271 			       READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10272 			list_for_each_entry(dev, list, todo_list) {
10273 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10274 					 dev->name, netdev_refcnt_read(dev));
10275 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10276 			}
10277 
10278 			warning_time = jiffies;
10279 		}
10280 	}
10281 }
10282 
10283 /* The sequence is:
10284  *
10285  *	rtnl_lock();
10286  *	...
10287  *	register_netdevice(x1);
10288  *	register_netdevice(x2);
10289  *	...
10290  *	unregister_netdevice(y1);
10291  *	unregister_netdevice(y2);
10292  *      ...
10293  *	rtnl_unlock();
10294  *	free_netdev(y1);
10295  *	free_netdev(y2);
10296  *
10297  * We are invoked by rtnl_unlock().
10298  * This allows us to deal with problems:
10299  * 1) We can delete sysfs objects which invoke hotplug
10300  *    without deadlocking with linkwatch via keventd.
10301  * 2) Since we run with the RTNL semaphore not held, we can sleep
10302  *    safely in order to wait for the netdev refcnt to drop to zero.
10303  *
10304  * We must not return until all unregister events added during
10305  * the interval the lock was held have been completed.
10306  */
10307 void netdev_run_todo(void)
10308 {
10309 	struct net_device *dev, *tmp;
10310 	struct list_head list;
10311 #ifdef CONFIG_LOCKDEP
10312 	struct list_head unlink_list;
10313 
10314 	list_replace_init(&net_unlink_list, &unlink_list);
10315 
10316 	while (!list_empty(&unlink_list)) {
10317 		struct net_device *dev = list_first_entry(&unlink_list,
10318 							  struct net_device,
10319 							  unlink_list);
10320 		list_del_init(&dev->unlink_list);
10321 		dev->nested_level = dev->lower_level - 1;
10322 	}
10323 #endif
10324 
10325 	/* Snapshot list, allow later requests */
10326 	list_replace_init(&net_todo_list, &list);
10327 
10328 	__rtnl_unlock();
10329 
10330 	/* Wait for rcu callbacks to finish before next phase */
10331 	if (!list_empty(&list))
10332 		rcu_barrier();
10333 
10334 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10335 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10336 			netdev_WARN(dev, "run_todo but not unregistering\n");
10337 			list_del(&dev->todo_list);
10338 			continue;
10339 		}
10340 
10341 		write_lock(&dev_base_lock);
10342 		dev->reg_state = NETREG_UNREGISTERED;
10343 		write_unlock(&dev_base_lock);
10344 		linkwatch_forget_dev(dev);
10345 	}
10346 
10347 	while (!list_empty(&list)) {
10348 		dev = netdev_wait_allrefs_any(&list);
10349 		list_del(&dev->todo_list);
10350 
10351 		/* paranoia */
10352 		BUG_ON(netdev_refcnt_read(dev) != 1);
10353 		BUG_ON(!list_empty(&dev->ptype_all));
10354 		BUG_ON(!list_empty(&dev->ptype_specific));
10355 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10356 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10357 
10358 		if (dev->priv_destructor)
10359 			dev->priv_destructor(dev);
10360 		if (dev->needs_free_netdev)
10361 			free_netdev(dev);
10362 
10363 		if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10364 			wake_up(&netdev_unregistering_wq);
10365 
10366 		/* Free network device */
10367 		kobject_put(&dev->dev.kobj);
10368 	}
10369 }
10370 
10371 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10372  * all the same fields in the same order as net_device_stats, with only
10373  * the type differing, but rtnl_link_stats64 may have additional fields
10374  * at the end for newer counters.
10375  */
10376 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10377 			     const struct net_device_stats *netdev_stats)
10378 {
10379 	size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10380 	const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10381 	u64 *dst = (u64 *)stats64;
10382 
10383 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10384 	for (i = 0; i < n; i++)
10385 		dst[i] = atomic_long_read(&src[i]);
10386 	/* zero out counters that only exist in rtnl_link_stats64 */
10387 	memset((char *)stats64 + n * sizeof(u64), 0,
10388 	       sizeof(*stats64) - n * sizeof(u64));
10389 }
10390 EXPORT_SYMBOL(netdev_stats_to_stats64);
10391 
10392 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10393 {
10394 	struct net_device_core_stats __percpu *p;
10395 
10396 	p = alloc_percpu_gfp(struct net_device_core_stats,
10397 			     GFP_ATOMIC | __GFP_NOWARN);
10398 
10399 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10400 		free_percpu(p);
10401 
10402 	/* This READ_ONCE() pairs with the cmpxchg() above */
10403 	return READ_ONCE(dev->core_stats);
10404 }
10405 EXPORT_SYMBOL(netdev_core_stats_alloc);
10406 
10407 /**
10408  *	dev_get_stats	- get network device statistics
10409  *	@dev: device to get statistics from
10410  *	@storage: place to store stats
10411  *
10412  *	Get network statistics from device. Return @storage.
10413  *	The device driver may provide its own method by setting
10414  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10415  *	otherwise the internal statistics structure is used.
10416  */
10417 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10418 					struct rtnl_link_stats64 *storage)
10419 {
10420 	const struct net_device_ops *ops = dev->netdev_ops;
10421 	const struct net_device_core_stats __percpu *p;
10422 
10423 	if (ops->ndo_get_stats64) {
10424 		memset(storage, 0, sizeof(*storage));
10425 		ops->ndo_get_stats64(dev, storage);
10426 	} else if (ops->ndo_get_stats) {
10427 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10428 	} else {
10429 		netdev_stats_to_stats64(storage, &dev->stats);
10430 	}
10431 
10432 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10433 	p = READ_ONCE(dev->core_stats);
10434 	if (p) {
10435 		const struct net_device_core_stats *core_stats;
10436 		int i;
10437 
10438 		for_each_possible_cpu(i) {
10439 			core_stats = per_cpu_ptr(p, i);
10440 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10441 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10442 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10443 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10444 		}
10445 	}
10446 	return storage;
10447 }
10448 EXPORT_SYMBOL(dev_get_stats);
10449 
10450 /**
10451  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10452  *	@s: place to store stats
10453  *	@netstats: per-cpu network stats to read from
10454  *
10455  *	Read per-cpu network statistics and populate the related fields in @s.
10456  */
10457 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10458 			   const struct pcpu_sw_netstats __percpu *netstats)
10459 {
10460 	int cpu;
10461 
10462 	for_each_possible_cpu(cpu) {
10463 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10464 		const struct pcpu_sw_netstats *stats;
10465 		unsigned int start;
10466 
10467 		stats = per_cpu_ptr(netstats, cpu);
10468 		do {
10469 			start = u64_stats_fetch_begin(&stats->syncp);
10470 			rx_packets = u64_stats_read(&stats->rx_packets);
10471 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10472 			tx_packets = u64_stats_read(&stats->tx_packets);
10473 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10474 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10475 
10476 		s->rx_packets += rx_packets;
10477 		s->rx_bytes   += rx_bytes;
10478 		s->tx_packets += tx_packets;
10479 		s->tx_bytes   += tx_bytes;
10480 	}
10481 }
10482 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10483 
10484 /**
10485  *	dev_get_tstats64 - ndo_get_stats64 implementation
10486  *	@dev: device to get statistics from
10487  *	@s: place to store stats
10488  *
10489  *	Populate @s from dev->stats and dev->tstats. Can be used as
10490  *	ndo_get_stats64() callback.
10491  */
10492 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10493 {
10494 	netdev_stats_to_stats64(s, &dev->stats);
10495 	dev_fetch_sw_netstats(s, dev->tstats);
10496 }
10497 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10498 
10499 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10500 {
10501 	struct netdev_queue *queue = dev_ingress_queue(dev);
10502 
10503 #ifdef CONFIG_NET_CLS_ACT
10504 	if (queue)
10505 		return queue;
10506 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10507 	if (!queue)
10508 		return NULL;
10509 	netdev_init_one_queue(dev, queue, NULL);
10510 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10511 	queue->qdisc_sleeping = &noop_qdisc;
10512 	rcu_assign_pointer(dev->ingress_queue, queue);
10513 #endif
10514 	return queue;
10515 }
10516 
10517 static const struct ethtool_ops default_ethtool_ops;
10518 
10519 void netdev_set_default_ethtool_ops(struct net_device *dev,
10520 				    const struct ethtool_ops *ops)
10521 {
10522 	if (dev->ethtool_ops == &default_ethtool_ops)
10523 		dev->ethtool_ops = ops;
10524 }
10525 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10526 
10527 /**
10528  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10529  * @dev: netdev to enable the IRQ coalescing on
10530  *
10531  * Sets a conservative default for SW IRQ coalescing. Users can use
10532  * sysfs attributes to override the default values.
10533  */
10534 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10535 {
10536 	WARN_ON(dev->reg_state == NETREG_REGISTERED);
10537 
10538 	dev->gro_flush_timeout = 20000;
10539 	dev->napi_defer_hard_irqs = 1;
10540 }
10541 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10542 
10543 void netdev_freemem(struct net_device *dev)
10544 {
10545 	char *addr = (char *)dev - dev->padded;
10546 
10547 	kvfree(addr);
10548 }
10549 
10550 /**
10551  * alloc_netdev_mqs - allocate network device
10552  * @sizeof_priv: size of private data to allocate space for
10553  * @name: device name format string
10554  * @name_assign_type: origin of device name
10555  * @setup: callback to initialize device
10556  * @txqs: the number of TX subqueues to allocate
10557  * @rxqs: the number of RX subqueues to allocate
10558  *
10559  * Allocates a struct net_device with private data area for driver use
10560  * and performs basic initialization.  Also allocates subqueue structs
10561  * for each queue on the device.
10562  */
10563 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10564 		unsigned char name_assign_type,
10565 		void (*setup)(struct net_device *),
10566 		unsigned int txqs, unsigned int rxqs)
10567 {
10568 	struct net_device *dev;
10569 	unsigned int alloc_size;
10570 	struct net_device *p;
10571 
10572 	BUG_ON(strlen(name) >= sizeof(dev->name));
10573 
10574 	if (txqs < 1) {
10575 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10576 		return NULL;
10577 	}
10578 
10579 	if (rxqs < 1) {
10580 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10581 		return NULL;
10582 	}
10583 
10584 	alloc_size = sizeof(struct net_device);
10585 	if (sizeof_priv) {
10586 		/* ensure 32-byte alignment of private area */
10587 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10588 		alloc_size += sizeof_priv;
10589 	}
10590 	/* ensure 32-byte alignment of whole construct */
10591 	alloc_size += NETDEV_ALIGN - 1;
10592 
10593 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10594 	if (!p)
10595 		return NULL;
10596 
10597 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10598 	dev->padded = (char *)dev - (char *)p;
10599 
10600 	ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10601 #ifdef CONFIG_PCPU_DEV_REFCNT
10602 	dev->pcpu_refcnt = alloc_percpu(int);
10603 	if (!dev->pcpu_refcnt)
10604 		goto free_dev;
10605 	__dev_hold(dev);
10606 #else
10607 	refcount_set(&dev->dev_refcnt, 1);
10608 #endif
10609 
10610 	if (dev_addr_init(dev))
10611 		goto free_pcpu;
10612 
10613 	dev_mc_init(dev);
10614 	dev_uc_init(dev);
10615 
10616 	dev_net_set(dev, &init_net);
10617 
10618 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10619 	dev->gso_max_segs = GSO_MAX_SEGS;
10620 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10621 	dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
10622 	dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
10623 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10624 	dev->tso_max_segs = TSO_MAX_SEGS;
10625 	dev->upper_level = 1;
10626 	dev->lower_level = 1;
10627 #ifdef CONFIG_LOCKDEP
10628 	dev->nested_level = 0;
10629 	INIT_LIST_HEAD(&dev->unlink_list);
10630 #endif
10631 
10632 	INIT_LIST_HEAD(&dev->napi_list);
10633 	INIT_LIST_HEAD(&dev->unreg_list);
10634 	INIT_LIST_HEAD(&dev->close_list);
10635 	INIT_LIST_HEAD(&dev->link_watch_list);
10636 	INIT_LIST_HEAD(&dev->adj_list.upper);
10637 	INIT_LIST_HEAD(&dev->adj_list.lower);
10638 	INIT_LIST_HEAD(&dev->ptype_all);
10639 	INIT_LIST_HEAD(&dev->ptype_specific);
10640 	INIT_LIST_HEAD(&dev->net_notifier_list);
10641 #ifdef CONFIG_NET_SCHED
10642 	hash_init(dev->qdisc_hash);
10643 #endif
10644 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10645 	setup(dev);
10646 
10647 	if (!dev->tx_queue_len) {
10648 		dev->priv_flags |= IFF_NO_QUEUE;
10649 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10650 	}
10651 
10652 	dev->num_tx_queues = txqs;
10653 	dev->real_num_tx_queues = txqs;
10654 	if (netif_alloc_netdev_queues(dev))
10655 		goto free_all;
10656 
10657 	dev->num_rx_queues = rxqs;
10658 	dev->real_num_rx_queues = rxqs;
10659 	if (netif_alloc_rx_queues(dev))
10660 		goto free_all;
10661 
10662 	strcpy(dev->name, name);
10663 	dev->name_assign_type = name_assign_type;
10664 	dev->group = INIT_NETDEV_GROUP;
10665 	if (!dev->ethtool_ops)
10666 		dev->ethtool_ops = &default_ethtool_ops;
10667 
10668 	nf_hook_netdev_init(dev);
10669 
10670 	return dev;
10671 
10672 free_all:
10673 	free_netdev(dev);
10674 	return NULL;
10675 
10676 free_pcpu:
10677 #ifdef CONFIG_PCPU_DEV_REFCNT
10678 	free_percpu(dev->pcpu_refcnt);
10679 free_dev:
10680 #endif
10681 	netdev_freemem(dev);
10682 	return NULL;
10683 }
10684 EXPORT_SYMBOL(alloc_netdev_mqs);
10685 
10686 /**
10687  * free_netdev - free network device
10688  * @dev: device
10689  *
10690  * This function does the last stage of destroying an allocated device
10691  * interface. The reference to the device object is released. If this
10692  * is the last reference then it will be freed.Must be called in process
10693  * context.
10694  */
10695 void free_netdev(struct net_device *dev)
10696 {
10697 	struct napi_struct *p, *n;
10698 
10699 	might_sleep();
10700 
10701 	/* When called immediately after register_netdevice() failed the unwind
10702 	 * handling may still be dismantling the device. Handle that case by
10703 	 * deferring the free.
10704 	 */
10705 	if (dev->reg_state == NETREG_UNREGISTERING) {
10706 		ASSERT_RTNL();
10707 		dev->needs_free_netdev = true;
10708 		return;
10709 	}
10710 
10711 	netif_free_tx_queues(dev);
10712 	netif_free_rx_queues(dev);
10713 
10714 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10715 
10716 	/* Flush device addresses */
10717 	dev_addr_flush(dev);
10718 
10719 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10720 		netif_napi_del(p);
10721 
10722 	ref_tracker_dir_exit(&dev->refcnt_tracker);
10723 #ifdef CONFIG_PCPU_DEV_REFCNT
10724 	free_percpu(dev->pcpu_refcnt);
10725 	dev->pcpu_refcnt = NULL;
10726 #endif
10727 	free_percpu(dev->core_stats);
10728 	dev->core_stats = NULL;
10729 	free_percpu(dev->xdp_bulkq);
10730 	dev->xdp_bulkq = NULL;
10731 
10732 	/*  Compatibility with error handling in drivers */
10733 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10734 		netdev_freemem(dev);
10735 		return;
10736 	}
10737 
10738 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10739 	dev->reg_state = NETREG_RELEASED;
10740 
10741 	/* will free via device release */
10742 	put_device(&dev->dev);
10743 }
10744 EXPORT_SYMBOL(free_netdev);
10745 
10746 /**
10747  *	synchronize_net -  Synchronize with packet receive processing
10748  *
10749  *	Wait for packets currently being received to be done.
10750  *	Does not block later packets from starting.
10751  */
10752 void synchronize_net(void)
10753 {
10754 	might_sleep();
10755 	if (rtnl_is_locked())
10756 		synchronize_rcu_expedited();
10757 	else
10758 		synchronize_rcu();
10759 }
10760 EXPORT_SYMBOL(synchronize_net);
10761 
10762 /**
10763  *	unregister_netdevice_queue - remove device from the kernel
10764  *	@dev: device
10765  *	@head: list
10766  *
10767  *	This function shuts down a device interface and removes it
10768  *	from the kernel tables.
10769  *	If head not NULL, device is queued to be unregistered later.
10770  *
10771  *	Callers must hold the rtnl semaphore.  You may want
10772  *	unregister_netdev() instead of this.
10773  */
10774 
10775 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10776 {
10777 	ASSERT_RTNL();
10778 
10779 	if (head) {
10780 		list_move_tail(&dev->unreg_list, head);
10781 	} else {
10782 		LIST_HEAD(single);
10783 
10784 		list_add(&dev->unreg_list, &single);
10785 		unregister_netdevice_many(&single);
10786 	}
10787 }
10788 EXPORT_SYMBOL(unregister_netdevice_queue);
10789 
10790 void unregister_netdevice_many_notify(struct list_head *head,
10791 				      u32 portid, const struct nlmsghdr *nlh)
10792 {
10793 	struct net_device *dev, *tmp;
10794 	LIST_HEAD(close_head);
10795 
10796 	BUG_ON(dev_boot_phase);
10797 	ASSERT_RTNL();
10798 
10799 	if (list_empty(head))
10800 		return;
10801 
10802 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10803 		/* Some devices call without registering
10804 		 * for initialization unwind. Remove those
10805 		 * devices and proceed with the remaining.
10806 		 */
10807 		if (dev->reg_state == NETREG_UNINITIALIZED) {
10808 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10809 				 dev->name, dev);
10810 
10811 			WARN_ON(1);
10812 			list_del(&dev->unreg_list);
10813 			continue;
10814 		}
10815 		dev->dismantle = true;
10816 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
10817 	}
10818 
10819 	/* If device is running, close it first. */
10820 	list_for_each_entry(dev, head, unreg_list)
10821 		list_add_tail(&dev->close_list, &close_head);
10822 	dev_close_many(&close_head, true);
10823 
10824 	list_for_each_entry(dev, head, unreg_list) {
10825 		/* And unlink it from device chain. */
10826 		write_lock(&dev_base_lock);
10827 		unlist_netdevice(dev, false);
10828 		dev->reg_state = NETREG_UNREGISTERING;
10829 		write_unlock(&dev_base_lock);
10830 	}
10831 	flush_all_backlogs();
10832 
10833 	synchronize_net();
10834 
10835 	list_for_each_entry(dev, head, unreg_list) {
10836 		struct sk_buff *skb = NULL;
10837 
10838 		/* Shutdown queueing discipline. */
10839 		dev_shutdown(dev);
10840 
10841 		dev_xdp_uninstall(dev);
10842 		bpf_dev_bound_netdev_unregister(dev);
10843 
10844 		netdev_offload_xstats_disable_all(dev);
10845 
10846 		/* Notify protocols, that we are about to destroy
10847 		 * this device. They should clean all the things.
10848 		 */
10849 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10850 
10851 		if (!dev->rtnl_link_ops ||
10852 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10853 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10854 						     GFP_KERNEL, NULL, 0,
10855 						     portid, nlmsg_seq(nlh));
10856 
10857 		/*
10858 		 *	Flush the unicast and multicast chains
10859 		 */
10860 		dev_uc_flush(dev);
10861 		dev_mc_flush(dev);
10862 
10863 		netdev_name_node_alt_flush(dev);
10864 		netdev_name_node_free(dev->name_node);
10865 
10866 		call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10867 
10868 		if (dev->netdev_ops->ndo_uninit)
10869 			dev->netdev_ops->ndo_uninit(dev);
10870 
10871 		if (skb)
10872 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10873 
10874 		/* Notifier chain MUST detach us all upper devices. */
10875 		WARN_ON(netdev_has_any_upper_dev(dev));
10876 		WARN_ON(netdev_has_any_lower_dev(dev));
10877 
10878 		/* Remove entries from kobject tree */
10879 		netdev_unregister_kobject(dev);
10880 #ifdef CONFIG_XPS
10881 		/* Remove XPS queueing entries */
10882 		netif_reset_xps_queues_gt(dev, 0);
10883 #endif
10884 	}
10885 
10886 	synchronize_net();
10887 
10888 	list_for_each_entry(dev, head, unreg_list) {
10889 		netdev_put(dev, &dev->dev_registered_tracker);
10890 		net_set_todo(dev);
10891 	}
10892 
10893 	list_del(head);
10894 }
10895 
10896 /**
10897  *	unregister_netdevice_many - unregister many devices
10898  *	@head: list of devices
10899  *
10900  *  Note: As most callers use a stack allocated list_head,
10901  *  we force a list_del() to make sure stack wont be corrupted later.
10902  */
10903 void unregister_netdevice_many(struct list_head *head)
10904 {
10905 	unregister_netdevice_many_notify(head, 0, NULL);
10906 }
10907 EXPORT_SYMBOL(unregister_netdevice_many);
10908 
10909 /**
10910  *	unregister_netdev - remove device from the kernel
10911  *	@dev: device
10912  *
10913  *	This function shuts down a device interface and removes it
10914  *	from the kernel tables.
10915  *
10916  *	This is just a wrapper for unregister_netdevice that takes
10917  *	the rtnl semaphore.  In general you want to use this and not
10918  *	unregister_netdevice.
10919  */
10920 void unregister_netdev(struct net_device *dev)
10921 {
10922 	rtnl_lock();
10923 	unregister_netdevice(dev);
10924 	rtnl_unlock();
10925 }
10926 EXPORT_SYMBOL(unregister_netdev);
10927 
10928 /**
10929  *	__dev_change_net_namespace - move device to different nethost namespace
10930  *	@dev: device
10931  *	@net: network namespace
10932  *	@pat: If not NULL name pattern to try if the current device name
10933  *	      is already taken in the destination network namespace.
10934  *	@new_ifindex: If not zero, specifies device index in the target
10935  *	              namespace.
10936  *
10937  *	This function shuts down a device interface and moves it
10938  *	to a new network namespace. On success 0 is returned, on
10939  *	a failure a netagive errno code is returned.
10940  *
10941  *	Callers must hold the rtnl semaphore.
10942  */
10943 
10944 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10945 			       const char *pat, int new_ifindex)
10946 {
10947 	struct net *net_old = dev_net(dev);
10948 	int err, new_nsid;
10949 
10950 	ASSERT_RTNL();
10951 
10952 	/* Don't allow namespace local devices to be moved. */
10953 	err = -EINVAL;
10954 	if (dev->features & NETIF_F_NETNS_LOCAL)
10955 		goto out;
10956 
10957 	/* Ensure the device has been registrered */
10958 	if (dev->reg_state != NETREG_REGISTERED)
10959 		goto out;
10960 
10961 	/* Get out if there is nothing todo */
10962 	err = 0;
10963 	if (net_eq(net_old, net))
10964 		goto out;
10965 
10966 	/* Pick the destination device name, and ensure
10967 	 * we can use it in the destination network namespace.
10968 	 */
10969 	err = -EEXIST;
10970 	if (netdev_name_in_use(net, dev->name)) {
10971 		/* We get here if we can't use the current device name */
10972 		if (!pat)
10973 			goto out;
10974 		err = dev_get_valid_name(net, dev, pat);
10975 		if (err < 0)
10976 			goto out;
10977 	}
10978 
10979 	/* Check that new_ifindex isn't used yet. */
10980 	err = -EBUSY;
10981 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10982 		goto out;
10983 
10984 	/*
10985 	 * And now a mini version of register_netdevice unregister_netdevice.
10986 	 */
10987 
10988 	/* If device is running close it first. */
10989 	dev_close(dev);
10990 
10991 	/* And unlink it from device chain */
10992 	unlist_netdevice(dev, true);
10993 
10994 	synchronize_net();
10995 
10996 	/* Shutdown queueing discipline. */
10997 	dev_shutdown(dev);
10998 
10999 	/* Notify protocols, that we are about to destroy
11000 	 * this device. They should clean all the things.
11001 	 *
11002 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
11003 	 * This is wanted because this way 8021q and macvlan know
11004 	 * the device is just moving and can keep their slaves up.
11005 	 */
11006 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11007 	rcu_barrier();
11008 
11009 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11010 	/* If there is an ifindex conflict assign a new one */
11011 	if (!new_ifindex) {
11012 		if (__dev_get_by_index(net, dev->ifindex))
11013 			new_ifindex = dev_new_index(net);
11014 		else
11015 			new_ifindex = dev->ifindex;
11016 	}
11017 
11018 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11019 			    new_ifindex);
11020 
11021 	/*
11022 	 *	Flush the unicast and multicast chains
11023 	 */
11024 	dev_uc_flush(dev);
11025 	dev_mc_flush(dev);
11026 
11027 	/* Send a netdev-removed uevent to the old namespace */
11028 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11029 	netdev_adjacent_del_links(dev);
11030 
11031 	/* Move per-net netdevice notifiers that are following the netdevice */
11032 	move_netdevice_notifiers_dev_net(dev, net);
11033 
11034 	/* Actually switch the network namespace */
11035 	dev_net_set(dev, net);
11036 	dev->ifindex = new_ifindex;
11037 
11038 	/* Send a netdev-add uevent to the new namespace */
11039 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11040 	netdev_adjacent_add_links(dev);
11041 
11042 	/* Fixup kobjects */
11043 	err = device_rename(&dev->dev, dev->name);
11044 	WARN_ON(err);
11045 
11046 	/* Adapt owner in case owning user namespace of target network
11047 	 * namespace is different from the original one.
11048 	 */
11049 	err = netdev_change_owner(dev, net_old, net);
11050 	WARN_ON(err);
11051 
11052 	/* Add the device back in the hashes */
11053 	list_netdevice(dev);
11054 
11055 	/* Notify protocols, that a new device appeared. */
11056 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11057 
11058 	/*
11059 	 *	Prevent userspace races by waiting until the network
11060 	 *	device is fully setup before sending notifications.
11061 	 */
11062 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11063 
11064 	synchronize_net();
11065 	err = 0;
11066 out:
11067 	return err;
11068 }
11069 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11070 
11071 static int dev_cpu_dead(unsigned int oldcpu)
11072 {
11073 	struct sk_buff **list_skb;
11074 	struct sk_buff *skb;
11075 	unsigned int cpu;
11076 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11077 
11078 	local_irq_disable();
11079 	cpu = smp_processor_id();
11080 	sd = &per_cpu(softnet_data, cpu);
11081 	oldsd = &per_cpu(softnet_data, oldcpu);
11082 
11083 	/* Find end of our completion_queue. */
11084 	list_skb = &sd->completion_queue;
11085 	while (*list_skb)
11086 		list_skb = &(*list_skb)->next;
11087 	/* Append completion queue from offline CPU. */
11088 	*list_skb = oldsd->completion_queue;
11089 	oldsd->completion_queue = NULL;
11090 
11091 	/* Append output queue from offline CPU. */
11092 	if (oldsd->output_queue) {
11093 		*sd->output_queue_tailp = oldsd->output_queue;
11094 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11095 		oldsd->output_queue = NULL;
11096 		oldsd->output_queue_tailp = &oldsd->output_queue;
11097 	}
11098 	/* Append NAPI poll list from offline CPU, with one exception :
11099 	 * process_backlog() must be called by cpu owning percpu backlog.
11100 	 * We properly handle process_queue & input_pkt_queue later.
11101 	 */
11102 	while (!list_empty(&oldsd->poll_list)) {
11103 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11104 							    struct napi_struct,
11105 							    poll_list);
11106 
11107 		list_del_init(&napi->poll_list);
11108 		if (napi->poll == process_backlog)
11109 			napi->state = 0;
11110 		else
11111 			____napi_schedule(sd, napi);
11112 	}
11113 
11114 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11115 	local_irq_enable();
11116 
11117 #ifdef CONFIG_RPS
11118 	remsd = oldsd->rps_ipi_list;
11119 	oldsd->rps_ipi_list = NULL;
11120 #endif
11121 	/* send out pending IPI's on offline CPU */
11122 	net_rps_send_ipi(remsd);
11123 
11124 	/* Process offline CPU's input_pkt_queue */
11125 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11126 		netif_rx(skb);
11127 		input_queue_head_incr(oldsd);
11128 	}
11129 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11130 		netif_rx(skb);
11131 		input_queue_head_incr(oldsd);
11132 	}
11133 
11134 	return 0;
11135 }
11136 
11137 /**
11138  *	netdev_increment_features - increment feature set by one
11139  *	@all: current feature set
11140  *	@one: new feature set
11141  *	@mask: mask feature set
11142  *
11143  *	Computes a new feature set after adding a device with feature set
11144  *	@one to the master device with current feature set @all.  Will not
11145  *	enable anything that is off in @mask. Returns the new feature set.
11146  */
11147 netdev_features_t netdev_increment_features(netdev_features_t all,
11148 	netdev_features_t one, netdev_features_t mask)
11149 {
11150 	if (mask & NETIF_F_HW_CSUM)
11151 		mask |= NETIF_F_CSUM_MASK;
11152 	mask |= NETIF_F_VLAN_CHALLENGED;
11153 
11154 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11155 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11156 
11157 	/* If one device supports hw checksumming, set for all. */
11158 	if (all & NETIF_F_HW_CSUM)
11159 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11160 
11161 	return all;
11162 }
11163 EXPORT_SYMBOL(netdev_increment_features);
11164 
11165 static struct hlist_head * __net_init netdev_create_hash(void)
11166 {
11167 	int i;
11168 	struct hlist_head *hash;
11169 
11170 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11171 	if (hash != NULL)
11172 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11173 			INIT_HLIST_HEAD(&hash[i]);
11174 
11175 	return hash;
11176 }
11177 
11178 /* Initialize per network namespace state */
11179 static int __net_init netdev_init(struct net *net)
11180 {
11181 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11182 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11183 
11184 	INIT_LIST_HEAD(&net->dev_base_head);
11185 
11186 	net->dev_name_head = netdev_create_hash();
11187 	if (net->dev_name_head == NULL)
11188 		goto err_name;
11189 
11190 	net->dev_index_head = netdev_create_hash();
11191 	if (net->dev_index_head == NULL)
11192 		goto err_idx;
11193 
11194 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11195 
11196 	return 0;
11197 
11198 err_idx:
11199 	kfree(net->dev_name_head);
11200 err_name:
11201 	return -ENOMEM;
11202 }
11203 
11204 /**
11205  *	netdev_drivername - network driver for the device
11206  *	@dev: network device
11207  *
11208  *	Determine network driver for device.
11209  */
11210 const char *netdev_drivername(const struct net_device *dev)
11211 {
11212 	const struct device_driver *driver;
11213 	const struct device *parent;
11214 	const char *empty = "";
11215 
11216 	parent = dev->dev.parent;
11217 	if (!parent)
11218 		return empty;
11219 
11220 	driver = parent->driver;
11221 	if (driver && driver->name)
11222 		return driver->name;
11223 	return empty;
11224 }
11225 
11226 static void __netdev_printk(const char *level, const struct net_device *dev,
11227 			    struct va_format *vaf)
11228 {
11229 	if (dev && dev->dev.parent) {
11230 		dev_printk_emit(level[1] - '0',
11231 				dev->dev.parent,
11232 				"%s %s %s%s: %pV",
11233 				dev_driver_string(dev->dev.parent),
11234 				dev_name(dev->dev.parent),
11235 				netdev_name(dev), netdev_reg_state(dev),
11236 				vaf);
11237 	} else if (dev) {
11238 		printk("%s%s%s: %pV",
11239 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11240 	} else {
11241 		printk("%s(NULL net_device): %pV", level, vaf);
11242 	}
11243 }
11244 
11245 void netdev_printk(const char *level, const struct net_device *dev,
11246 		   const char *format, ...)
11247 {
11248 	struct va_format vaf;
11249 	va_list args;
11250 
11251 	va_start(args, format);
11252 
11253 	vaf.fmt = format;
11254 	vaf.va = &args;
11255 
11256 	__netdev_printk(level, dev, &vaf);
11257 
11258 	va_end(args);
11259 }
11260 EXPORT_SYMBOL(netdev_printk);
11261 
11262 #define define_netdev_printk_level(func, level)			\
11263 void func(const struct net_device *dev, const char *fmt, ...)	\
11264 {								\
11265 	struct va_format vaf;					\
11266 	va_list args;						\
11267 								\
11268 	va_start(args, fmt);					\
11269 								\
11270 	vaf.fmt = fmt;						\
11271 	vaf.va = &args;						\
11272 								\
11273 	__netdev_printk(level, dev, &vaf);			\
11274 								\
11275 	va_end(args);						\
11276 }								\
11277 EXPORT_SYMBOL(func);
11278 
11279 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11280 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11281 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11282 define_netdev_printk_level(netdev_err, KERN_ERR);
11283 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11284 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11285 define_netdev_printk_level(netdev_info, KERN_INFO);
11286 
11287 static void __net_exit netdev_exit(struct net *net)
11288 {
11289 	kfree(net->dev_name_head);
11290 	kfree(net->dev_index_head);
11291 	if (net != &init_net)
11292 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11293 }
11294 
11295 static struct pernet_operations __net_initdata netdev_net_ops = {
11296 	.init = netdev_init,
11297 	.exit = netdev_exit,
11298 };
11299 
11300 static void __net_exit default_device_exit_net(struct net *net)
11301 {
11302 	struct net_device *dev, *aux;
11303 	/*
11304 	 * Push all migratable network devices back to the
11305 	 * initial network namespace
11306 	 */
11307 	ASSERT_RTNL();
11308 	for_each_netdev_safe(net, dev, aux) {
11309 		int err;
11310 		char fb_name[IFNAMSIZ];
11311 
11312 		/* Ignore unmoveable devices (i.e. loopback) */
11313 		if (dev->features & NETIF_F_NETNS_LOCAL)
11314 			continue;
11315 
11316 		/* Leave virtual devices for the generic cleanup */
11317 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11318 			continue;
11319 
11320 		/* Push remaining network devices to init_net */
11321 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11322 		if (netdev_name_in_use(&init_net, fb_name))
11323 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11324 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11325 		if (err) {
11326 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11327 				 __func__, dev->name, err);
11328 			BUG();
11329 		}
11330 	}
11331 }
11332 
11333 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11334 {
11335 	/* At exit all network devices most be removed from a network
11336 	 * namespace.  Do this in the reverse order of registration.
11337 	 * Do this across as many network namespaces as possible to
11338 	 * improve batching efficiency.
11339 	 */
11340 	struct net_device *dev;
11341 	struct net *net;
11342 	LIST_HEAD(dev_kill_list);
11343 
11344 	rtnl_lock();
11345 	list_for_each_entry(net, net_list, exit_list) {
11346 		default_device_exit_net(net);
11347 		cond_resched();
11348 	}
11349 
11350 	list_for_each_entry(net, net_list, exit_list) {
11351 		for_each_netdev_reverse(net, dev) {
11352 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11353 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11354 			else
11355 				unregister_netdevice_queue(dev, &dev_kill_list);
11356 		}
11357 	}
11358 	unregister_netdevice_many(&dev_kill_list);
11359 	rtnl_unlock();
11360 }
11361 
11362 static struct pernet_operations __net_initdata default_device_ops = {
11363 	.exit_batch = default_device_exit_batch,
11364 };
11365 
11366 /*
11367  *	Initialize the DEV module. At boot time this walks the device list and
11368  *	unhooks any devices that fail to initialise (normally hardware not
11369  *	present) and leaves us with a valid list of present and active devices.
11370  *
11371  */
11372 
11373 /*
11374  *       This is called single threaded during boot, so no need
11375  *       to take the rtnl semaphore.
11376  */
11377 static int __init net_dev_init(void)
11378 {
11379 	int i, rc = -ENOMEM;
11380 
11381 	BUG_ON(!dev_boot_phase);
11382 
11383 	if (dev_proc_init())
11384 		goto out;
11385 
11386 	if (netdev_kobject_init())
11387 		goto out;
11388 
11389 	INIT_LIST_HEAD(&ptype_all);
11390 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11391 		INIT_LIST_HEAD(&ptype_base[i]);
11392 
11393 	if (register_pernet_subsys(&netdev_net_ops))
11394 		goto out;
11395 
11396 	/*
11397 	 *	Initialise the packet receive queues.
11398 	 */
11399 
11400 	for_each_possible_cpu(i) {
11401 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11402 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11403 
11404 		INIT_WORK(flush, flush_backlog);
11405 
11406 		skb_queue_head_init(&sd->input_pkt_queue);
11407 		skb_queue_head_init(&sd->process_queue);
11408 #ifdef CONFIG_XFRM_OFFLOAD
11409 		skb_queue_head_init(&sd->xfrm_backlog);
11410 #endif
11411 		INIT_LIST_HEAD(&sd->poll_list);
11412 		sd->output_queue_tailp = &sd->output_queue;
11413 #ifdef CONFIG_RPS
11414 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11415 		sd->cpu = i;
11416 #endif
11417 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11418 		spin_lock_init(&sd->defer_lock);
11419 
11420 		init_gro_hash(&sd->backlog);
11421 		sd->backlog.poll = process_backlog;
11422 		sd->backlog.weight = weight_p;
11423 	}
11424 
11425 	dev_boot_phase = 0;
11426 
11427 	/* The loopback device is special if any other network devices
11428 	 * is present in a network namespace the loopback device must
11429 	 * be present. Since we now dynamically allocate and free the
11430 	 * loopback device ensure this invariant is maintained by
11431 	 * keeping the loopback device as the first device on the
11432 	 * list of network devices.  Ensuring the loopback devices
11433 	 * is the first device that appears and the last network device
11434 	 * that disappears.
11435 	 */
11436 	if (register_pernet_device(&loopback_net_ops))
11437 		goto out;
11438 
11439 	if (register_pernet_device(&default_device_ops))
11440 		goto out;
11441 
11442 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11443 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11444 
11445 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11446 				       NULL, dev_cpu_dead);
11447 	WARN_ON(rc < 0);
11448 	rc = 0;
11449 out:
11450 	return rc;
11451 }
11452 
11453 subsys_initcall(net_dev_init);
11454