xref: /openbmc/linux/net/core/dev.c (revision 2d7be507)
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_extack(unsigned long val,
164 					   struct net_device *dev,
165 					   struct netlink_ext_ack *extack);
166 static struct napi_struct *napi_by_id(unsigned int napi_id);
167 
168 /*
169  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
170  * semaphore.
171  *
172  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
173  *
174  * Writers must hold the rtnl semaphore while they loop through the
175  * dev_base_head list, and hold dev_base_lock for writing when they do the
176  * actual updates.  This allows pure readers to access the list even
177  * while a writer is preparing to update it.
178  *
179  * To put it another way, dev_base_lock is held for writing only to
180  * protect against pure readers; the rtnl semaphore provides the
181  * protection against other writers.
182  *
183  * See, for example usages, register_netdevice() and
184  * unregister_netdevice(), which must be called with the rtnl
185  * semaphore held.
186  */
187 DEFINE_RWLOCK(dev_base_lock);
188 EXPORT_SYMBOL(dev_base_lock);
189 
190 static DEFINE_MUTEX(ifalias_mutex);
191 
192 /* protects napi_hash addition/deletion and napi_gen_id */
193 static DEFINE_SPINLOCK(napi_hash_lock);
194 
195 static unsigned int napi_gen_id = NR_CPUS;
196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
197 
198 static DECLARE_RWSEM(devnet_rename_sem);
199 
200 static inline void dev_base_seq_inc(struct net *net)
201 {
202 	while (++net->dev_base_seq == 0)
203 		;
204 }
205 
206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 {
208 	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
209 
210 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
211 }
212 
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 {
215 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
216 }
217 
218 static inline void rps_lock_irqsave(struct softnet_data *sd,
219 				    unsigned long *flags)
220 {
221 	if (IS_ENABLED(CONFIG_RPS))
222 		spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
223 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
224 		local_irq_save(*flags);
225 }
226 
227 static inline void rps_lock_irq_disable(struct softnet_data *sd)
228 {
229 	if (IS_ENABLED(CONFIG_RPS))
230 		spin_lock_irq(&sd->input_pkt_queue.lock);
231 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
232 		local_irq_disable();
233 }
234 
235 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
236 					  unsigned long *flags)
237 {
238 	if (IS_ENABLED(CONFIG_RPS))
239 		spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
240 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
241 		local_irq_restore(*flags);
242 }
243 
244 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
245 {
246 	if (IS_ENABLED(CONFIG_RPS))
247 		spin_unlock_irq(&sd->input_pkt_queue.lock);
248 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
249 		local_irq_enable();
250 }
251 
252 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
253 						       const char *name)
254 {
255 	struct netdev_name_node *name_node;
256 
257 	name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
258 	if (!name_node)
259 		return NULL;
260 	INIT_HLIST_NODE(&name_node->hlist);
261 	name_node->dev = dev;
262 	name_node->name = name;
263 	return name_node;
264 }
265 
266 static struct netdev_name_node *
267 netdev_name_node_head_alloc(struct net_device *dev)
268 {
269 	struct netdev_name_node *name_node;
270 
271 	name_node = netdev_name_node_alloc(dev, dev->name);
272 	if (!name_node)
273 		return NULL;
274 	INIT_LIST_HEAD(&name_node->list);
275 	return name_node;
276 }
277 
278 static void netdev_name_node_free(struct netdev_name_node *name_node)
279 {
280 	kfree(name_node);
281 }
282 
283 static void netdev_name_node_add(struct net *net,
284 				 struct netdev_name_node *name_node)
285 {
286 	hlist_add_head_rcu(&name_node->hlist,
287 			   dev_name_hash(net, name_node->name));
288 }
289 
290 static void netdev_name_node_del(struct netdev_name_node *name_node)
291 {
292 	hlist_del_rcu(&name_node->hlist);
293 }
294 
295 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
296 							const char *name)
297 {
298 	struct hlist_head *head = dev_name_hash(net, name);
299 	struct netdev_name_node *name_node;
300 
301 	hlist_for_each_entry(name_node, head, hlist)
302 		if (!strcmp(name_node->name, name))
303 			return name_node;
304 	return NULL;
305 }
306 
307 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
308 							    const char *name)
309 {
310 	struct hlist_head *head = dev_name_hash(net, name);
311 	struct netdev_name_node *name_node;
312 
313 	hlist_for_each_entry_rcu(name_node, head, hlist)
314 		if (!strcmp(name_node->name, name))
315 			return name_node;
316 	return NULL;
317 }
318 
319 bool netdev_name_in_use(struct net *net, const char *name)
320 {
321 	return netdev_name_node_lookup(net, name);
322 }
323 EXPORT_SYMBOL(netdev_name_in_use);
324 
325 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
326 {
327 	struct netdev_name_node *name_node;
328 	struct net *net = dev_net(dev);
329 
330 	name_node = netdev_name_node_lookup(net, name);
331 	if (name_node)
332 		return -EEXIST;
333 	name_node = netdev_name_node_alloc(dev, name);
334 	if (!name_node)
335 		return -ENOMEM;
336 	netdev_name_node_add(net, name_node);
337 	/* The node that holds dev->name acts as a head of per-device list. */
338 	list_add_tail(&name_node->list, &dev->name_node->list);
339 
340 	return 0;
341 }
342 
343 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
344 {
345 	list_del(&name_node->list);
346 	netdev_name_node_del(name_node);
347 	kfree(name_node->name);
348 	netdev_name_node_free(name_node);
349 }
350 
351 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
352 {
353 	struct netdev_name_node *name_node;
354 	struct net *net = dev_net(dev);
355 
356 	name_node = netdev_name_node_lookup(net, name);
357 	if (!name_node)
358 		return -ENOENT;
359 	/* lookup might have found our primary name or a name belonging
360 	 * to another device.
361 	 */
362 	if (name_node == dev->name_node || name_node->dev != dev)
363 		return -EINVAL;
364 
365 	__netdev_name_node_alt_destroy(name_node);
366 
367 	return 0;
368 }
369 
370 static void netdev_name_node_alt_flush(struct net_device *dev)
371 {
372 	struct netdev_name_node *name_node, *tmp;
373 
374 	list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
375 		__netdev_name_node_alt_destroy(name_node);
376 }
377 
378 /* Device list insertion */
379 static void list_netdevice(struct net_device *dev)
380 {
381 	struct net *net = dev_net(dev);
382 
383 	ASSERT_RTNL();
384 
385 	write_lock(&dev_base_lock);
386 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
387 	netdev_name_node_add(net, dev->name_node);
388 	hlist_add_head_rcu(&dev->index_hlist,
389 			   dev_index_hash(net, dev->ifindex));
390 	write_unlock(&dev_base_lock);
391 
392 	dev_base_seq_inc(net);
393 }
394 
395 /* Device list removal
396  * caller must respect a RCU grace period before freeing/reusing dev
397  */
398 static void unlist_netdevice(struct net_device *dev, bool lock)
399 {
400 	ASSERT_RTNL();
401 
402 	/* Unlink dev from the device chain */
403 	if (lock)
404 		write_lock(&dev_base_lock);
405 	list_del_rcu(&dev->dev_list);
406 	netdev_name_node_del(dev->name_node);
407 	hlist_del_rcu(&dev->index_hlist);
408 	if (lock)
409 		write_unlock(&dev_base_lock);
410 
411 	dev_base_seq_inc(dev_net(dev));
412 }
413 
414 /*
415  *	Our notifier list
416  */
417 
418 static RAW_NOTIFIER_HEAD(netdev_chain);
419 
420 /*
421  *	Device drivers call our routines to queue packets here. We empty the
422  *	queue in the local softnet handler.
423  */
424 
425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
426 EXPORT_PER_CPU_SYMBOL(softnet_data);
427 
428 #ifdef CONFIG_LOCKDEP
429 /*
430  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
431  * according to dev->type
432  */
433 static const unsigned short netdev_lock_type[] = {
434 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
435 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
436 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
437 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
438 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
439 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
440 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
441 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
442 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
443 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
444 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
445 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
446 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
447 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
448 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
449 
450 static const char *const netdev_lock_name[] = {
451 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
452 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
453 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
454 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
455 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
456 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
457 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
458 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
459 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
460 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
461 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
462 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
463 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
464 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
465 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
466 
467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
469 
470 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
471 {
472 	int i;
473 
474 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
475 		if (netdev_lock_type[i] == dev_type)
476 			return i;
477 	/* the last key is used by default */
478 	return ARRAY_SIZE(netdev_lock_type) - 1;
479 }
480 
481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
482 						 unsigned short dev_type)
483 {
484 	int i;
485 
486 	i = netdev_lock_pos(dev_type);
487 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
488 				   netdev_lock_name[i]);
489 }
490 
491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
492 {
493 	int i;
494 
495 	i = netdev_lock_pos(dev->type);
496 	lockdep_set_class_and_name(&dev->addr_list_lock,
497 				   &netdev_addr_lock_key[i],
498 				   netdev_lock_name[i]);
499 }
500 #else
501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
502 						 unsigned short dev_type)
503 {
504 }
505 
506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
507 {
508 }
509 #endif
510 
511 /*******************************************************************************
512  *
513  *		Protocol management and registration routines
514  *
515  *******************************************************************************/
516 
517 
518 /*
519  *	Add a protocol ID to the list. Now that the input handler is
520  *	smarter we can dispense with all the messy stuff that used to be
521  *	here.
522  *
523  *	BEWARE!!! Protocol handlers, mangling input packets,
524  *	MUST BE last in hash buckets and checking protocol handlers
525  *	MUST start from promiscuous ptype_all chain in net_bh.
526  *	It is true now, do not change it.
527  *	Explanation follows: if protocol handler, mangling packet, will
528  *	be the first on list, it is not able to sense, that packet
529  *	is cloned and should be copied-on-write, so that it will
530  *	change it and subsequent readers will get broken packet.
531  *							--ANK (980803)
532  */
533 
534 static inline struct list_head *ptype_head(const struct packet_type *pt)
535 {
536 	if (pt->type == htons(ETH_P_ALL))
537 		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
538 	else
539 		return pt->dev ? &pt->dev->ptype_specific :
540 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
541 }
542 
543 /**
544  *	dev_add_pack - add packet handler
545  *	@pt: packet type declaration
546  *
547  *	Add a protocol handler to the networking stack. The passed &packet_type
548  *	is linked into kernel lists and may not be freed until it has been
549  *	removed from the kernel lists.
550  *
551  *	This call does not sleep therefore it can not
552  *	guarantee all CPU's that are in middle of receiving packets
553  *	will see the new packet type (until the next received packet).
554  */
555 
556 void dev_add_pack(struct packet_type *pt)
557 {
558 	struct list_head *head = ptype_head(pt);
559 
560 	spin_lock(&ptype_lock);
561 	list_add_rcu(&pt->list, head);
562 	spin_unlock(&ptype_lock);
563 }
564 EXPORT_SYMBOL(dev_add_pack);
565 
566 /**
567  *	__dev_remove_pack	 - remove packet handler
568  *	@pt: packet type declaration
569  *
570  *	Remove a protocol handler that was previously added to the kernel
571  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
572  *	from the kernel lists and can be freed or reused once this function
573  *	returns.
574  *
575  *      The packet type might still be in use by receivers
576  *	and must not be freed until after all the CPU's have gone
577  *	through a quiescent state.
578  */
579 void __dev_remove_pack(struct packet_type *pt)
580 {
581 	struct list_head *head = ptype_head(pt);
582 	struct packet_type *pt1;
583 
584 	spin_lock(&ptype_lock);
585 
586 	list_for_each_entry(pt1, head, list) {
587 		if (pt == pt1) {
588 			list_del_rcu(&pt->list);
589 			goto out;
590 		}
591 	}
592 
593 	pr_warn("dev_remove_pack: %p not found\n", pt);
594 out:
595 	spin_unlock(&ptype_lock);
596 }
597 EXPORT_SYMBOL(__dev_remove_pack);
598 
599 /**
600  *	dev_remove_pack	 - remove packet handler
601  *	@pt: packet type declaration
602  *
603  *	Remove a protocol handler that was previously added to the kernel
604  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
605  *	from the kernel lists and can be freed or reused once this function
606  *	returns.
607  *
608  *	This call sleeps to guarantee that no CPU is looking at the packet
609  *	type after return.
610  */
611 void dev_remove_pack(struct packet_type *pt)
612 {
613 	__dev_remove_pack(pt);
614 
615 	synchronize_net();
616 }
617 EXPORT_SYMBOL(dev_remove_pack);
618 
619 
620 /*******************************************************************************
621  *
622  *			    Device Interface Subroutines
623  *
624  *******************************************************************************/
625 
626 /**
627  *	dev_get_iflink	- get 'iflink' value of a interface
628  *	@dev: targeted interface
629  *
630  *	Indicates the ifindex the interface is linked to.
631  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
632  */
633 
634 int dev_get_iflink(const struct net_device *dev)
635 {
636 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
637 		return dev->netdev_ops->ndo_get_iflink(dev);
638 
639 	return dev->ifindex;
640 }
641 EXPORT_SYMBOL(dev_get_iflink);
642 
643 /**
644  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
645  *	@dev: targeted interface
646  *	@skb: The packet.
647  *
648  *	For better visibility of tunnel traffic OVS needs to retrieve
649  *	egress tunnel information for a packet. Following API allows
650  *	user to get this info.
651  */
652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
653 {
654 	struct ip_tunnel_info *info;
655 
656 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
657 		return -EINVAL;
658 
659 	info = skb_tunnel_info_unclone(skb);
660 	if (!info)
661 		return -ENOMEM;
662 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
663 		return -EINVAL;
664 
665 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
666 }
667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
668 
669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
670 {
671 	int k = stack->num_paths++;
672 
673 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
674 		return NULL;
675 
676 	return &stack->path[k];
677 }
678 
679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
680 			  struct net_device_path_stack *stack)
681 {
682 	const struct net_device *last_dev;
683 	struct net_device_path_ctx ctx = {
684 		.dev	= dev,
685 	};
686 	struct net_device_path *path;
687 	int ret = 0;
688 
689 	memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
690 	stack->num_paths = 0;
691 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
692 		last_dev = ctx.dev;
693 		path = dev_fwd_path(stack);
694 		if (!path)
695 			return -1;
696 
697 		memset(path, 0, sizeof(struct net_device_path));
698 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
699 		if (ret < 0)
700 			return -1;
701 
702 		if (WARN_ON_ONCE(last_dev == ctx.dev))
703 			return -1;
704 	}
705 
706 	if (!ctx.dev)
707 		return ret;
708 
709 	path = dev_fwd_path(stack);
710 	if (!path)
711 		return -1;
712 	path->type = DEV_PATH_ETHERNET;
713 	path->dev = ctx.dev;
714 
715 	return ret;
716 }
717 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
718 
719 /**
720  *	__dev_get_by_name	- find a device by its name
721  *	@net: the applicable net namespace
722  *	@name: name to find
723  *
724  *	Find an interface by name. Must be called under RTNL semaphore
725  *	or @dev_base_lock. If the name is found a pointer to the device
726  *	is returned. If the name is not found then %NULL is returned. The
727  *	reference counters are not incremented so the caller must be
728  *	careful with locks.
729  */
730 
731 struct net_device *__dev_get_by_name(struct net *net, const char *name)
732 {
733 	struct netdev_name_node *node_name;
734 
735 	node_name = netdev_name_node_lookup(net, name);
736 	return node_name ? node_name->dev : NULL;
737 }
738 EXPORT_SYMBOL(__dev_get_by_name);
739 
740 /**
741  * dev_get_by_name_rcu	- find a device by its name
742  * @net: the applicable net namespace
743  * @name: name to find
744  *
745  * Find an interface by name.
746  * If the name is found a pointer to the device is returned.
747  * If the name is not found then %NULL is returned.
748  * The reference counters are not incremented so the caller must be
749  * careful with locks. The caller must hold RCU lock.
750  */
751 
752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 {
754 	struct netdev_name_node *node_name;
755 
756 	node_name = netdev_name_node_lookup_rcu(net, name);
757 	return node_name ? node_name->dev : NULL;
758 }
759 EXPORT_SYMBOL(dev_get_by_name_rcu);
760 
761 /**
762  *	dev_get_by_name		- find a device by its name
763  *	@net: the applicable net namespace
764  *	@name: name to find
765  *
766  *	Find an interface by name. This can be called from any
767  *	context and does its own locking. The returned handle has
768  *	the usage count incremented and the caller must use dev_put() to
769  *	release it when it is no longer needed. %NULL is returned if no
770  *	matching device is found.
771  */
772 
773 struct net_device *dev_get_by_name(struct net *net, const char *name)
774 {
775 	struct net_device *dev;
776 
777 	rcu_read_lock();
778 	dev = dev_get_by_name_rcu(net, name);
779 	dev_hold(dev);
780 	rcu_read_unlock();
781 	return dev;
782 }
783 EXPORT_SYMBOL(dev_get_by_name);
784 
785 /**
786  *	__dev_get_by_index - find a device by its ifindex
787  *	@net: the applicable net namespace
788  *	@ifindex: index of device
789  *
790  *	Search for an interface by index. Returns %NULL if the device
791  *	is not found or a pointer to the device. The device has not
792  *	had its reference counter increased so the caller must be careful
793  *	about locking. The caller must hold either the RTNL semaphore
794  *	or @dev_base_lock.
795  */
796 
797 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
798 {
799 	struct net_device *dev;
800 	struct hlist_head *head = dev_index_hash(net, ifindex);
801 
802 	hlist_for_each_entry(dev, head, index_hlist)
803 		if (dev->ifindex == ifindex)
804 			return dev;
805 
806 	return NULL;
807 }
808 EXPORT_SYMBOL(__dev_get_by_index);
809 
810 /**
811  *	dev_get_by_index_rcu - find a device by its ifindex
812  *	@net: the applicable net namespace
813  *	@ifindex: index of device
814  *
815  *	Search for an interface by index. Returns %NULL if the device
816  *	is not found or a pointer to the device. The device has not
817  *	had its reference counter increased so the caller must be careful
818  *	about locking. The caller must hold RCU lock.
819  */
820 
821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
822 {
823 	struct net_device *dev;
824 	struct hlist_head *head = dev_index_hash(net, ifindex);
825 
826 	hlist_for_each_entry_rcu(dev, head, index_hlist)
827 		if (dev->ifindex == ifindex)
828 			return dev;
829 
830 	return NULL;
831 }
832 EXPORT_SYMBOL(dev_get_by_index_rcu);
833 
834 
835 /**
836  *	dev_get_by_index - find a device by its ifindex
837  *	@net: the applicable net namespace
838  *	@ifindex: index of device
839  *
840  *	Search for an interface by index. Returns NULL if the device
841  *	is not found or a pointer to the device. The device returned has
842  *	had a reference added and the pointer is safe until the user calls
843  *	dev_put to indicate they have finished with it.
844  */
845 
846 struct net_device *dev_get_by_index(struct net *net, int ifindex)
847 {
848 	struct net_device *dev;
849 
850 	rcu_read_lock();
851 	dev = dev_get_by_index_rcu(net, ifindex);
852 	dev_hold(dev);
853 	rcu_read_unlock();
854 	return dev;
855 }
856 EXPORT_SYMBOL(dev_get_by_index);
857 
858 /**
859  *	dev_get_by_napi_id - find a device by napi_id
860  *	@napi_id: ID of the NAPI struct
861  *
862  *	Search for an interface by NAPI ID. Returns %NULL if the device
863  *	is not found or a pointer to the device. The device has not had
864  *	its reference counter increased so the caller must be careful
865  *	about locking. The caller must hold RCU lock.
866  */
867 
868 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
869 {
870 	struct napi_struct *napi;
871 
872 	WARN_ON_ONCE(!rcu_read_lock_held());
873 
874 	if (napi_id < MIN_NAPI_ID)
875 		return NULL;
876 
877 	napi = napi_by_id(napi_id);
878 
879 	return napi ? napi->dev : NULL;
880 }
881 EXPORT_SYMBOL(dev_get_by_napi_id);
882 
883 /**
884  *	netdev_get_name - get a netdevice name, knowing its ifindex.
885  *	@net: network namespace
886  *	@name: a pointer to the buffer where the name will be stored.
887  *	@ifindex: the ifindex of the interface to get the name from.
888  */
889 int netdev_get_name(struct net *net, char *name, int ifindex)
890 {
891 	struct net_device *dev;
892 	int ret;
893 
894 	down_read(&devnet_rename_sem);
895 	rcu_read_lock();
896 
897 	dev = dev_get_by_index_rcu(net, ifindex);
898 	if (!dev) {
899 		ret = -ENODEV;
900 		goto out;
901 	}
902 
903 	strcpy(name, dev->name);
904 
905 	ret = 0;
906 out:
907 	rcu_read_unlock();
908 	up_read(&devnet_rename_sem);
909 	return ret;
910 }
911 
912 /**
913  *	dev_getbyhwaddr_rcu - find a device by its hardware address
914  *	@net: the applicable net namespace
915  *	@type: media type of device
916  *	@ha: hardware address
917  *
918  *	Search for an interface by MAC address. Returns NULL if the device
919  *	is not found or a pointer to the device.
920  *	The caller must hold RCU or RTNL.
921  *	The returned device has not had its ref count increased
922  *	and the caller must therefore be careful about locking
923  *
924  */
925 
926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
927 				       const char *ha)
928 {
929 	struct net_device *dev;
930 
931 	for_each_netdev_rcu(net, dev)
932 		if (dev->type == type &&
933 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
934 			return dev;
935 
936 	return NULL;
937 }
938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
939 
940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
941 {
942 	struct net_device *dev, *ret = NULL;
943 
944 	rcu_read_lock();
945 	for_each_netdev_rcu(net, dev)
946 		if (dev->type == type) {
947 			dev_hold(dev);
948 			ret = dev;
949 			break;
950 		}
951 	rcu_read_unlock();
952 	return ret;
953 }
954 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 
956 /**
957  *	__dev_get_by_flags - find any device with given flags
958  *	@net: the applicable net namespace
959  *	@if_flags: IFF_* values
960  *	@mask: bitmask of bits in if_flags to check
961  *
962  *	Search for any interface with the given flags. Returns NULL if a device
963  *	is not found or a pointer to the device. Must be called inside
964  *	rtnl_lock(), and result refcount is unchanged.
965  */
966 
967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 				      unsigned short mask)
969 {
970 	struct net_device *dev, *ret;
971 
972 	ASSERT_RTNL();
973 
974 	ret = NULL;
975 	for_each_netdev(net, dev) {
976 		if (((dev->flags ^ if_flags) & mask) == 0) {
977 			ret = dev;
978 			break;
979 		}
980 	}
981 	return ret;
982 }
983 EXPORT_SYMBOL(__dev_get_by_flags);
984 
985 /**
986  *	dev_valid_name - check if name is okay for network device
987  *	@name: name string
988  *
989  *	Network device names need to be valid file names to
990  *	allow sysfs to work.  We also disallow any kind of
991  *	whitespace.
992  */
993 bool dev_valid_name(const char *name)
994 {
995 	if (*name == '\0')
996 		return false;
997 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
998 		return false;
999 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1000 		return false;
1001 
1002 	while (*name) {
1003 		if (*name == '/' || *name == ':' || isspace(*name))
1004 			return false;
1005 		name++;
1006 	}
1007 	return true;
1008 }
1009 EXPORT_SYMBOL(dev_valid_name);
1010 
1011 /**
1012  *	__dev_alloc_name - allocate a name for a device
1013  *	@net: network namespace to allocate the device name in
1014  *	@name: name format string
1015  *	@buf:  scratch buffer and result name string
1016  *
1017  *	Passed a format string - eg "lt%d" it will try and find a suitable
1018  *	id. It scans list of devices to build up a free map, then chooses
1019  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1020  *	while allocating the name and adding the device in order to avoid
1021  *	duplicates.
1022  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023  *	Returns the number of the unit assigned or a negative errno code.
1024  */
1025 
1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1027 {
1028 	int i = 0;
1029 	const char *p;
1030 	const int max_netdevices = 8*PAGE_SIZE;
1031 	unsigned long *inuse;
1032 	struct net_device *d;
1033 
1034 	if (!dev_valid_name(name))
1035 		return -EINVAL;
1036 
1037 	p = strchr(name, '%');
1038 	if (p) {
1039 		/*
1040 		 * Verify the string as this thing may have come from
1041 		 * the user.  There must be either one "%d" and no other "%"
1042 		 * characters.
1043 		 */
1044 		if (p[1] != 'd' || strchr(p + 2, '%'))
1045 			return -EINVAL;
1046 
1047 		/* Use one page as a bit array of possible slots */
1048 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1049 		if (!inuse)
1050 			return -ENOMEM;
1051 
1052 		for_each_netdev(net, d) {
1053 			struct netdev_name_node *name_node;
1054 			list_for_each_entry(name_node, &d->name_node->list, list) {
1055 				if (!sscanf(name_node->name, name, &i))
1056 					continue;
1057 				if (i < 0 || i >= max_netdevices)
1058 					continue;
1059 
1060 				/*  avoid cases where sscanf is not exact inverse of printf */
1061 				snprintf(buf, IFNAMSIZ, name, i);
1062 				if (!strncmp(buf, name_node->name, IFNAMSIZ))
1063 					__set_bit(i, inuse);
1064 			}
1065 			if (!sscanf(d->name, name, &i))
1066 				continue;
1067 			if (i < 0 || i >= max_netdevices)
1068 				continue;
1069 
1070 			/*  avoid cases where sscanf is not exact inverse of printf */
1071 			snprintf(buf, IFNAMSIZ, name, i);
1072 			if (!strncmp(buf, d->name, IFNAMSIZ))
1073 				__set_bit(i, inuse);
1074 		}
1075 
1076 		i = find_first_zero_bit(inuse, max_netdevices);
1077 		free_page((unsigned long) inuse);
1078 	}
1079 
1080 	snprintf(buf, IFNAMSIZ, name, i);
1081 	if (!netdev_name_in_use(net, buf))
1082 		return i;
1083 
1084 	/* It is possible to run out of possible slots
1085 	 * when the name is long and there isn't enough space left
1086 	 * for the digits, or if all bits are used.
1087 	 */
1088 	return -ENFILE;
1089 }
1090 
1091 static int dev_alloc_name_ns(struct net *net,
1092 			     struct net_device *dev,
1093 			     const char *name)
1094 {
1095 	char buf[IFNAMSIZ];
1096 	int ret;
1097 
1098 	BUG_ON(!net);
1099 	ret = __dev_alloc_name(net, name, buf);
1100 	if (ret >= 0)
1101 		strscpy(dev->name, buf, IFNAMSIZ);
1102 	return ret;
1103 }
1104 
1105 /**
1106  *	dev_alloc_name - allocate a name for a device
1107  *	@dev: device
1108  *	@name: name format string
1109  *
1110  *	Passed a format string - eg "lt%d" it will try and find a suitable
1111  *	id. It scans list of devices to build up a free map, then chooses
1112  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1113  *	while allocating the name and adding the device in order to avoid
1114  *	duplicates.
1115  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1116  *	Returns the number of the unit assigned or a negative errno code.
1117  */
1118 
1119 int dev_alloc_name(struct net_device *dev, const char *name)
1120 {
1121 	return dev_alloc_name_ns(dev_net(dev), dev, name);
1122 }
1123 EXPORT_SYMBOL(dev_alloc_name);
1124 
1125 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1126 			      const char *name)
1127 {
1128 	BUG_ON(!net);
1129 
1130 	if (!dev_valid_name(name))
1131 		return -EINVAL;
1132 
1133 	if (strchr(name, '%'))
1134 		return dev_alloc_name_ns(net, dev, name);
1135 	else if (netdev_name_in_use(net, name))
1136 		return -EEXIST;
1137 	else if (dev->name != name)
1138 		strscpy(dev->name, name, IFNAMSIZ);
1139 
1140 	return 0;
1141 }
1142 
1143 /**
1144  *	dev_change_name - change name of a device
1145  *	@dev: device
1146  *	@newname: name (or format string) must be at least IFNAMSIZ
1147  *
1148  *	Change name of a device, can pass format strings "eth%d".
1149  *	for wildcarding.
1150  */
1151 int dev_change_name(struct net_device *dev, const char *newname)
1152 {
1153 	unsigned char old_assign_type;
1154 	char oldname[IFNAMSIZ];
1155 	int err = 0;
1156 	int ret;
1157 	struct net *net;
1158 
1159 	ASSERT_RTNL();
1160 	BUG_ON(!dev_net(dev));
1161 
1162 	net = dev_net(dev);
1163 
1164 	down_write(&devnet_rename_sem);
1165 
1166 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1167 		up_write(&devnet_rename_sem);
1168 		return 0;
1169 	}
1170 
1171 	memcpy(oldname, dev->name, IFNAMSIZ);
1172 
1173 	err = dev_get_valid_name(net, dev, newname);
1174 	if (err < 0) {
1175 		up_write(&devnet_rename_sem);
1176 		return err;
1177 	}
1178 
1179 	if (oldname[0] && !strchr(oldname, '%'))
1180 		netdev_info(dev, "renamed from %s%s\n", oldname,
1181 			    dev->flags & IFF_UP ? " (while UP)" : "");
1182 
1183 	old_assign_type = dev->name_assign_type;
1184 	dev->name_assign_type = NET_NAME_RENAMED;
1185 
1186 rollback:
1187 	ret = device_rename(&dev->dev, dev->name);
1188 	if (ret) {
1189 		memcpy(dev->name, oldname, IFNAMSIZ);
1190 		dev->name_assign_type = old_assign_type;
1191 		up_write(&devnet_rename_sem);
1192 		return ret;
1193 	}
1194 
1195 	up_write(&devnet_rename_sem);
1196 
1197 	netdev_adjacent_rename_links(dev, oldname);
1198 
1199 	write_lock(&dev_base_lock);
1200 	netdev_name_node_del(dev->name_node);
1201 	write_unlock(&dev_base_lock);
1202 
1203 	synchronize_rcu();
1204 
1205 	write_lock(&dev_base_lock);
1206 	netdev_name_node_add(net, dev->name_node);
1207 	write_unlock(&dev_base_lock);
1208 
1209 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1210 	ret = notifier_to_errno(ret);
1211 
1212 	if (ret) {
1213 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1214 		if (err >= 0) {
1215 			err = ret;
1216 			down_write(&devnet_rename_sem);
1217 			memcpy(dev->name, oldname, IFNAMSIZ);
1218 			memcpy(oldname, newname, IFNAMSIZ);
1219 			dev->name_assign_type = old_assign_type;
1220 			old_assign_type = NET_NAME_RENAMED;
1221 			goto rollback;
1222 		} else {
1223 			netdev_err(dev, "name change rollback failed: %d\n",
1224 				   ret);
1225 		}
1226 	}
1227 
1228 	return err;
1229 }
1230 
1231 /**
1232  *	dev_set_alias - change ifalias of a device
1233  *	@dev: device
1234  *	@alias: name up to IFALIASZ
1235  *	@len: limit of bytes to copy from info
1236  *
1237  *	Set ifalias for a device,
1238  */
1239 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1240 {
1241 	struct dev_ifalias *new_alias = NULL;
1242 
1243 	if (len >= IFALIASZ)
1244 		return -EINVAL;
1245 
1246 	if (len) {
1247 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1248 		if (!new_alias)
1249 			return -ENOMEM;
1250 
1251 		memcpy(new_alias->ifalias, alias, len);
1252 		new_alias->ifalias[len] = 0;
1253 	}
1254 
1255 	mutex_lock(&ifalias_mutex);
1256 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1257 					mutex_is_locked(&ifalias_mutex));
1258 	mutex_unlock(&ifalias_mutex);
1259 
1260 	if (new_alias)
1261 		kfree_rcu(new_alias, rcuhead);
1262 
1263 	return len;
1264 }
1265 EXPORT_SYMBOL(dev_set_alias);
1266 
1267 /**
1268  *	dev_get_alias - get ifalias of a device
1269  *	@dev: device
1270  *	@name: buffer to store name of ifalias
1271  *	@len: size of buffer
1272  *
1273  *	get ifalias for a device.  Caller must make sure dev cannot go
1274  *	away,  e.g. rcu read lock or own a reference count to device.
1275  */
1276 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1277 {
1278 	const struct dev_ifalias *alias;
1279 	int ret = 0;
1280 
1281 	rcu_read_lock();
1282 	alias = rcu_dereference(dev->ifalias);
1283 	if (alias)
1284 		ret = snprintf(name, len, "%s", alias->ifalias);
1285 	rcu_read_unlock();
1286 
1287 	return ret;
1288 }
1289 
1290 /**
1291  *	netdev_features_change - device changes features
1292  *	@dev: device to cause notification
1293  *
1294  *	Called to indicate a device has changed features.
1295  */
1296 void netdev_features_change(struct net_device *dev)
1297 {
1298 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1299 }
1300 EXPORT_SYMBOL(netdev_features_change);
1301 
1302 /**
1303  *	netdev_state_change - device changes state
1304  *	@dev: device to cause notification
1305  *
1306  *	Called to indicate a device has changed state. This function calls
1307  *	the notifier chains for netdev_chain and sends a NEWLINK message
1308  *	to the routing socket.
1309  */
1310 void netdev_state_change(struct net_device *dev)
1311 {
1312 	if (dev->flags & IFF_UP) {
1313 		struct netdev_notifier_change_info change_info = {
1314 			.info.dev = dev,
1315 		};
1316 
1317 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1318 					      &change_info.info);
1319 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1320 	}
1321 }
1322 EXPORT_SYMBOL(netdev_state_change);
1323 
1324 /**
1325  * __netdev_notify_peers - notify network peers about existence of @dev,
1326  * to be called when rtnl lock is already held.
1327  * @dev: network device
1328  *
1329  * Generate traffic such that interested network peers are aware of
1330  * @dev, such as by generating a gratuitous ARP. This may be used when
1331  * a device wants to inform the rest of the network about some sort of
1332  * reconfiguration such as a failover event or virtual machine
1333  * migration.
1334  */
1335 void __netdev_notify_peers(struct net_device *dev)
1336 {
1337 	ASSERT_RTNL();
1338 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1339 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1340 }
1341 EXPORT_SYMBOL(__netdev_notify_peers);
1342 
1343 /**
1344  * netdev_notify_peers - notify network peers about existence of @dev
1345  * @dev: network device
1346  *
1347  * Generate traffic such that interested network peers are aware of
1348  * @dev, such as by generating a gratuitous ARP. This may be used when
1349  * a device wants to inform the rest of the network about some sort of
1350  * reconfiguration such as a failover event or virtual machine
1351  * migration.
1352  */
1353 void netdev_notify_peers(struct net_device *dev)
1354 {
1355 	rtnl_lock();
1356 	__netdev_notify_peers(dev);
1357 	rtnl_unlock();
1358 }
1359 EXPORT_SYMBOL(netdev_notify_peers);
1360 
1361 static int napi_threaded_poll(void *data);
1362 
1363 static int napi_kthread_create(struct napi_struct *n)
1364 {
1365 	int err = 0;
1366 
1367 	/* Create and wake up the kthread once to put it in
1368 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1369 	 * warning and work with loadavg.
1370 	 */
1371 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1372 				n->dev->name, n->napi_id);
1373 	if (IS_ERR(n->thread)) {
1374 		err = PTR_ERR(n->thread);
1375 		pr_err("kthread_run failed with err %d\n", err);
1376 		n->thread = NULL;
1377 	}
1378 
1379 	return err;
1380 }
1381 
1382 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1383 {
1384 	const struct net_device_ops *ops = dev->netdev_ops;
1385 	int ret;
1386 
1387 	ASSERT_RTNL();
1388 	dev_addr_check(dev);
1389 
1390 	if (!netif_device_present(dev)) {
1391 		/* may be detached because parent is runtime-suspended */
1392 		if (dev->dev.parent)
1393 			pm_runtime_resume(dev->dev.parent);
1394 		if (!netif_device_present(dev))
1395 			return -ENODEV;
1396 	}
1397 
1398 	/* Block netpoll from trying to do any rx path servicing.
1399 	 * If we don't do this there is a chance ndo_poll_controller
1400 	 * or ndo_poll may be running while we open the device
1401 	 */
1402 	netpoll_poll_disable(dev);
1403 
1404 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1405 	ret = notifier_to_errno(ret);
1406 	if (ret)
1407 		return ret;
1408 
1409 	set_bit(__LINK_STATE_START, &dev->state);
1410 
1411 	if (ops->ndo_validate_addr)
1412 		ret = ops->ndo_validate_addr(dev);
1413 
1414 	if (!ret && ops->ndo_open)
1415 		ret = ops->ndo_open(dev);
1416 
1417 	netpoll_poll_enable(dev);
1418 
1419 	if (ret)
1420 		clear_bit(__LINK_STATE_START, &dev->state);
1421 	else {
1422 		dev->flags |= IFF_UP;
1423 		dev_set_rx_mode(dev);
1424 		dev_activate(dev);
1425 		add_device_randomness(dev->dev_addr, dev->addr_len);
1426 	}
1427 
1428 	return ret;
1429 }
1430 
1431 /**
1432  *	dev_open	- prepare an interface for use.
1433  *	@dev: device to open
1434  *	@extack: netlink extended ack
1435  *
1436  *	Takes a device from down to up state. The device's private open
1437  *	function is invoked and then the multicast lists are loaded. Finally
1438  *	the device is moved into the up state and a %NETDEV_UP message is
1439  *	sent to the netdev notifier chain.
1440  *
1441  *	Calling this function on an active interface is a nop. On a failure
1442  *	a negative errno code is returned.
1443  */
1444 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1445 {
1446 	int ret;
1447 
1448 	if (dev->flags & IFF_UP)
1449 		return 0;
1450 
1451 	ret = __dev_open(dev, extack);
1452 	if (ret < 0)
1453 		return ret;
1454 
1455 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1456 	call_netdevice_notifiers(NETDEV_UP, dev);
1457 
1458 	return ret;
1459 }
1460 EXPORT_SYMBOL(dev_open);
1461 
1462 static void __dev_close_many(struct list_head *head)
1463 {
1464 	struct net_device *dev;
1465 
1466 	ASSERT_RTNL();
1467 	might_sleep();
1468 
1469 	list_for_each_entry(dev, head, close_list) {
1470 		/* Temporarily disable netpoll until the interface is down */
1471 		netpoll_poll_disable(dev);
1472 
1473 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1474 
1475 		clear_bit(__LINK_STATE_START, &dev->state);
1476 
1477 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1478 		 * can be even on different cpu. So just clear netif_running().
1479 		 *
1480 		 * dev->stop() will invoke napi_disable() on all of it's
1481 		 * napi_struct instances on this device.
1482 		 */
1483 		smp_mb__after_atomic(); /* Commit netif_running(). */
1484 	}
1485 
1486 	dev_deactivate_many(head);
1487 
1488 	list_for_each_entry(dev, head, close_list) {
1489 		const struct net_device_ops *ops = dev->netdev_ops;
1490 
1491 		/*
1492 		 *	Call the device specific close. This cannot fail.
1493 		 *	Only if device is UP
1494 		 *
1495 		 *	We allow it to be called even after a DETACH hot-plug
1496 		 *	event.
1497 		 */
1498 		if (ops->ndo_stop)
1499 			ops->ndo_stop(dev);
1500 
1501 		dev->flags &= ~IFF_UP;
1502 		netpoll_poll_enable(dev);
1503 	}
1504 }
1505 
1506 static void __dev_close(struct net_device *dev)
1507 {
1508 	LIST_HEAD(single);
1509 
1510 	list_add(&dev->close_list, &single);
1511 	__dev_close_many(&single);
1512 	list_del(&single);
1513 }
1514 
1515 void dev_close_many(struct list_head *head, bool unlink)
1516 {
1517 	struct net_device *dev, *tmp;
1518 
1519 	/* Remove the devices that don't need to be closed */
1520 	list_for_each_entry_safe(dev, tmp, head, close_list)
1521 		if (!(dev->flags & IFF_UP))
1522 			list_del_init(&dev->close_list);
1523 
1524 	__dev_close_many(head);
1525 
1526 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1527 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1528 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1529 		if (unlink)
1530 			list_del_init(&dev->close_list);
1531 	}
1532 }
1533 EXPORT_SYMBOL(dev_close_many);
1534 
1535 /**
1536  *	dev_close - shutdown an interface.
1537  *	@dev: device to shutdown
1538  *
1539  *	This function moves an active device into down state. A
1540  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1541  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1542  *	chain.
1543  */
1544 void dev_close(struct net_device *dev)
1545 {
1546 	if (dev->flags & IFF_UP) {
1547 		LIST_HEAD(single);
1548 
1549 		list_add(&dev->close_list, &single);
1550 		dev_close_many(&single, true);
1551 		list_del(&single);
1552 	}
1553 }
1554 EXPORT_SYMBOL(dev_close);
1555 
1556 
1557 /**
1558  *	dev_disable_lro - disable Large Receive Offload on a device
1559  *	@dev: device
1560  *
1561  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1562  *	called under RTNL.  This is needed if received packets may be
1563  *	forwarded to another interface.
1564  */
1565 void dev_disable_lro(struct net_device *dev)
1566 {
1567 	struct net_device *lower_dev;
1568 	struct list_head *iter;
1569 
1570 	dev->wanted_features &= ~NETIF_F_LRO;
1571 	netdev_update_features(dev);
1572 
1573 	if (unlikely(dev->features & NETIF_F_LRO))
1574 		netdev_WARN(dev, "failed to disable LRO!\n");
1575 
1576 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1577 		dev_disable_lro(lower_dev);
1578 }
1579 EXPORT_SYMBOL(dev_disable_lro);
1580 
1581 /**
1582  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1583  *	@dev: device
1584  *
1585  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1586  *	called under RTNL.  This is needed if Generic XDP is installed on
1587  *	the device.
1588  */
1589 static void dev_disable_gro_hw(struct net_device *dev)
1590 {
1591 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1592 	netdev_update_features(dev);
1593 
1594 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1595 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1596 }
1597 
1598 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1599 {
1600 #define N(val) 						\
1601 	case NETDEV_##val:				\
1602 		return "NETDEV_" __stringify(val);
1603 	switch (cmd) {
1604 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1605 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1606 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1607 	N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1608 	N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1609 	N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1610 	N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1611 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1612 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1613 	N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1614 	N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1615 	N(XDP_FEAT_CHANGE)
1616 	}
1617 #undef N
1618 	return "UNKNOWN_NETDEV_EVENT";
1619 }
1620 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1621 
1622 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1623 				   struct net_device *dev)
1624 {
1625 	struct netdev_notifier_info info = {
1626 		.dev = dev,
1627 	};
1628 
1629 	return nb->notifier_call(nb, val, &info);
1630 }
1631 
1632 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1633 					     struct net_device *dev)
1634 {
1635 	int err;
1636 
1637 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1638 	err = notifier_to_errno(err);
1639 	if (err)
1640 		return err;
1641 
1642 	if (!(dev->flags & IFF_UP))
1643 		return 0;
1644 
1645 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1646 	return 0;
1647 }
1648 
1649 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1650 						struct net_device *dev)
1651 {
1652 	if (dev->flags & IFF_UP) {
1653 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1654 					dev);
1655 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1656 	}
1657 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1658 }
1659 
1660 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1661 						 struct net *net)
1662 {
1663 	struct net_device *dev;
1664 	int err;
1665 
1666 	for_each_netdev(net, dev) {
1667 		err = call_netdevice_register_notifiers(nb, dev);
1668 		if (err)
1669 			goto rollback;
1670 	}
1671 	return 0;
1672 
1673 rollback:
1674 	for_each_netdev_continue_reverse(net, dev)
1675 		call_netdevice_unregister_notifiers(nb, dev);
1676 	return err;
1677 }
1678 
1679 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1680 						    struct net *net)
1681 {
1682 	struct net_device *dev;
1683 
1684 	for_each_netdev(net, dev)
1685 		call_netdevice_unregister_notifiers(nb, dev);
1686 }
1687 
1688 static int dev_boot_phase = 1;
1689 
1690 /**
1691  * register_netdevice_notifier - register a network notifier block
1692  * @nb: notifier
1693  *
1694  * Register a notifier to be called when network device events occur.
1695  * The notifier passed is linked into the kernel structures and must
1696  * not be reused until it has been unregistered. A negative errno code
1697  * is returned on a failure.
1698  *
1699  * When registered all registration and up events are replayed
1700  * to the new notifier to allow device to have a race free
1701  * view of the network device list.
1702  */
1703 
1704 int register_netdevice_notifier(struct notifier_block *nb)
1705 {
1706 	struct net *net;
1707 	int err;
1708 
1709 	/* Close race with setup_net() and cleanup_net() */
1710 	down_write(&pernet_ops_rwsem);
1711 	rtnl_lock();
1712 	err = raw_notifier_chain_register(&netdev_chain, nb);
1713 	if (err)
1714 		goto unlock;
1715 	if (dev_boot_phase)
1716 		goto unlock;
1717 	for_each_net(net) {
1718 		err = call_netdevice_register_net_notifiers(nb, net);
1719 		if (err)
1720 			goto rollback;
1721 	}
1722 
1723 unlock:
1724 	rtnl_unlock();
1725 	up_write(&pernet_ops_rwsem);
1726 	return err;
1727 
1728 rollback:
1729 	for_each_net_continue_reverse(net)
1730 		call_netdevice_unregister_net_notifiers(nb, net);
1731 
1732 	raw_notifier_chain_unregister(&netdev_chain, nb);
1733 	goto unlock;
1734 }
1735 EXPORT_SYMBOL(register_netdevice_notifier);
1736 
1737 /**
1738  * unregister_netdevice_notifier - unregister a network notifier block
1739  * @nb: notifier
1740  *
1741  * Unregister a notifier previously registered by
1742  * register_netdevice_notifier(). The notifier is unlinked into the
1743  * kernel structures and may then be reused. A negative errno code
1744  * is returned on a failure.
1745  *
1746  * After unregistering unregister and down device events are synthesized
1747  * for all devices on the device list to the removed notifier to remove
1748  * the need for special case cleanup code.
1749  */
1750 
1751 int unregister_netdevice_notifier(struct notifier_block *nb)
1752 {
1753 	struct net *net;
1754 	int err;
1755 
1756 	/* Close race with setup_net() and cleanup_net() */
1757 	down_write(&pernet_ops_rwsem);
1758 	rtnl_lock();
1759 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1760 	if (err)
1761 		goto unlock;
1762 
1763 	for_each_net(net)
1764 		call_netdevice_unregister_net_notifiers(nb, net);
1765 
1766 unlock:
1767 	rtnl_unlock();
1768 	up_write(&pernet_ops_rwsem);
1769 	return err;
1770 }
1771 EXPORT_SYMBOL(unregister_netdevice_notifier);
1772 
1773 static int __register_netdevice_notifier_net(struct net *net,
1774 					     struct notifier_block *nb,
1775 					     bool ignore_call_fail)
1776 {
1777 	int err;
1778 
1779 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1780 	if (err)
1781 		return err;
1782 	if (dev_boot_phase)
1783 		return 0;
1784 
1785 	err = call_netdevice_register_net_notifiers(nb, net);
1786 	if (err && !ignore_call_fail)
1787 		goto chain_unregister;
1788 
1789 	return 0;
1790 
1791 chain_unregister:
1792 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1793 	return err;
1794 }
1795 
1796 static int __unregister_netdevice_notifier_net(struct net *net,
1797 					       struct notifier_block *nb)
1798 {
1799 	int err;
1800 
1801 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1802 	if (err)
1803 		return err;
1804 
1805 	call_netdevice_unregister_net_notifiers(nb, net);
1806 	return 0;
1807 }
1808 
1809 /**
1810  * register_netdevice_notifier_net - register a per-netns network notifier block
1811  * @net: network namespace
1812  * @nb: notifier
1813  *
1814  * Register a notifier to be called when network device events occur.
1815  * The notifier passed is linked into the kernel structures and must
1816  * not be reused until it has been unregistered. A negative errno code
1817  * is returned on a failure.
1818  *
1819  * When registered all registration and up events are replayed
1820  * to the new notifier to allow device to have a race free
1821  * view of the network device list.
1822  */
1823 
1824 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1825 {
1826 	int err;
1827 
1828 	rtnl_lock();
1829 	err = __register_netdevice_notifier_net(net, nb, false);
1830 	rtnl_unlock();
1831 	return err;
1832 }
1833 EXPORT_SYMBOL(register_netdevice_notifier_net);
1834 
1835 /**
1836  * unregister_netdevice_notifier_net - unregister a per-netns
1837  *                                     network notifier block
1838  * @net: network namespace
1839  * @nb: notifier
1840  *
1841  * Unregister a notifier previously registered by
1842  * register_netdevice_notifier_net(). The notifier is unlinked from the
1843  * kernel structures and may then be reused. A negative errno code
1844  * is returned on a failure.
1845  *
1846  * After unregistering unregister and down device events are synthesized
1847  * for all devices on the device list to the removed notifier to remove
1848  * the need for special case cleanup code.
1849  */
1850 
1851 int unregister_netdevice_notifier_net(struct net *net,
1852 				      struct notifier_block *nb)
1853 {
1854 	int err;
1855 
1856 	rtnl_lock();
1857 	err = __unregister_netdevice_notifier_net(net, nb);
1858 	rtnl_unlock();
1859 	return err;
1860 }
1861 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1862 
1863 static void __move_netdevice_notifier_net(struct net *src_net,
1864 					  struct net *dst_net,
1865 					  struct notifier_block *nb)
1866 {
1867 	__unregister_netdevice_notifier_net(src_net, nb);
1868 	__register_netdevice_notifier_net(dst_net, nb, true);
1869 }
1870 
1871 int register_netdevice_notifier_dev_net(struct net_device *dev,
1872 					struct notifier_block *nb,
1873 					struct netdev_net_notifier *nn)
1874 {
1875 	int err;
1876 
1877 	rtnl_lock();
1878 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1879 	if (!err) {
1880 		nn->nb = nb;
1881 		list_add(&nn->list, &dev->net_notifier_list);
1882 	}
1883 	rtnl_unlock();
1884 	return err;
1885 }
1886 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1887 
1888 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1889 					  struct notifier_block *nb,
1890 					  struct netdev_net_notifier *nn)
1891 {
1892 	int err;
1893 
1894 	rtnl_lock();
1895 	list_del(&nn->list);
1896 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1897 	rtnl_unlock();
1898 	return err;
1899 }
1900 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1901 
1902 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1903 					     struct net *net)
1904 {
1905 	struct netdev_net_notifier *nn;
1906 
1907 	list_for_each_entry(nn, &dev->net_notifier_list, list)
1908 		__move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1909 }
1910 
1911 /**
1912  *	call_netdevice_notifiers_info - call all network notifier blocks
1913  *	@val: value passed unmodified to notifier function
1914  *	@info: notifier information data
1915  *
1916  *	Call all network notifier blocks.  Parameters and return value
1917  *	are as for raw_notifier_call_chain().
1918  */
1919 
1920 int call_netdevice_notifiers_info(unsigned long val,
1921 				  struct netdev_notifier_info *info)
1922 {
1923 	struct net *net = dev_net(info->dev);
1924 	int ret;
1925 
1926 	ASSERT_RTNL();
1927 
1928 	/* Run per-netns notifier block chain first, then run the global one.
1929 	 * Hopefully, one day, the global one is going to be removed after
1930 	 * all notifier block registrators get converted to be per-netns.
1931 	 */
1932 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1933 	if (ret & NOTIFY_STOP_MASK)
1934 		return ret;
1935 	return raw_notifier_call_chain(&netdev_chain, val, info);
1936 }
1937 
1938 /**
1939  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1940  *	                                       for and rollback on error
1941  *	@val_up: value passed unmodified to notifier function
1942  *	@val_down: value passed unmodified to the notifier function when
1943  *	           recovering from an error on @val_up
1944  *	@info: notifier information data
1945  *
1946  *	Call all per-netns network notifier blocks, but not notifier blocks on
1947  *	the global notifier chain. Parameters and return value are as for
1948  *	raw_notifier_call_chain_robust().
1949  */
1950 
1951 static int
1952 call_netdevice_notifiers_info_robust(unsigned long val_up,
1953 				     unsigned long val_down,
1954 				     struct netdev_notifier_info *info)
1955 {
1956 	struct net *net = dev_net(info->dev);
1957 
1958 	ASSERT_RTNL();
1959 
1960 	return raw_notifier_call_chain_robust(&net->netdev_chain,
1961 					      val_up, val_down, info);
1962 }
1963 
1964 static int call_netdevice_notifiers_extack(unsigned long val,
1965 					   struct net_device *dev,
1966 					   struct netlink_ext_ack *extack)
1967 {
1968 	struct netdev_notifier_info info = {
1969 		.dev = dev,
1970 		.extack = extack,
1971 	};
1972 
1973 	return call_netdevice_notifiers_info(val, &info);
1974 }
1975 
1976 /**
1977  *	call_netdevice_notifiers - call all network notifier blocks
1978  *      @val: value passed unmodified to notifier function
1979  *      @dev: net_device pointer passed unmodified to notifier function
1980  *
1981  *	Call all network notifier blocks.  Parameters and return value
1982  *	are as for raw_notifier_call_chain().
1983  */
1984 
1985 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1986 {
1987 	return call_netdevice_notifiers_extack(val, dev, NULL);
1988 }
1989 EXPORT_SYMBOL(call_netdevice_notifiers);
1990 
1991 /**
1992  *	call_netdevice_notifiers_mtu - call all network notifier blocks
1993  *	@val: value passed unmodified to notifier function
1994  *	@dev: net_device pointer passed unmodified to notifier function
1995  *	@arg: additional u32 argument passed to the notifier function
1996  *
1997  *	Call all network notifier blocks.  Parameters and return value
1998  *	are as for raw_notifier_call_chain().
1999  */
2000 static int call_netdevice_notifiers_mtu(unsigned long val,
2001 					struct net_device *dev, u32 arg)
2002 {
2003 	struct netdev_notifier_info_ext info = {
2004 		.info.dev = dev,
2005 		.ext.mtu = arg,
2006 	};
2007 
2008 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2009 
2010 	return call_netdevice_notifiers_info(val, &info.info);
2011 }
2012 
2013 #ifdef CONFIG_NET_INGRESS
2014 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2015 
2016 void net_inc_ingress_queue(void)
2017 {
2018 	static_branch_inc(&ingress_needed_key);
2019 }
2020 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2021 
2022 void net_dec_ingress_queue(void)
2023 {
2024 	static_branch_dec(&ingress_needed_key);
2025 }
2026 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2027 #endif
2028 
2029 #ifdef CONFIG_NET_EGRESS
2030 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2031 
2032 void net_inc_egress_queue(void)
2033 {
2034 	static_branch_inc(&egress_needed_key);
2035 }
2036 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2037 
2038 void net_dec_egress_queue(void)
2039 {
2040 	static_branch_dec(&egress_needed_key);
2041 }
2042 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2043 #endif
2044 
2045 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2046 EXPORT_SYMBOL(netstamp_needed_key);
2047 #ifdef CONFIG_JUMP_LABEL
2048 static atomic_t netstamp_needed_deferred;
2049 static atomic_t netstamp_wanted;
2050 static void netstamp_clear(struct work_struct *work)
2051 {
2052 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2053 	int wanted;
2054 
2055 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2056 	if (wanted > 0)
2057 		static_branch_enable(&netstamp_needed_key);
2058 	else
2059 		static_branch_disable(&netstamp_needed_key);
2060 }
2061 static DECLARE_WORK(netstamp_work, netstamp_clear);
2062 #endif
2063 
2064 void net_enable_timestamp(void)
2065 {
2066 #ifdef CONFIG_JUMP_LABEL
2067 	int wanted = atomic_read(&netstamp_wanted);
2068 
2069 	while (wanted > 0) {
2070 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2071 			return;
2072 	}
2073 	atomic_inc(&netstamp_needed_deferred);
2074 	schedule_work(&netstamp_work);
2075 #else
2076 	static_branch_inc(&netstamp_needed_key);
2077 #endif
2078 }
2079 EXPORT_SYMBOL(net_enable_timestamp);
2080 
2081 void net_disable_timestamp(void)
2082 {
2083 #ifdef CONFIG_JUMP_LABEL
2084 	int wanted = atomic_read(&netstamp_wanted);
2085 
2086 	while (wanted > 1) {
2087 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2088 			return;
2089 	}
2090 	atomic_dec(&netstamp_needed_deferred);
2091 	schedule_work(&netstamp_work);
2092 #else
2093 	static_branch_dec(&netstamp_needed_key);
2094 #endif
2095 }
2096 EXPORT_SYMBOL(net_disable_timestamp);
2097 
2098 static inline void net_timestamp_set(struct sk_buff *skb)
2099 {
2100 	skb->tstamp = 0;
2101 	skb->mono_delivery_time = 0;
2102 	if (static_branch_unlikely(&netstamp_needed_key))
2103 		skb->tstamp = ktime_get_real();
2104 }
2105 
2106 #define net_timestamp_check(COND, SKB)				\
2107 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2108 		if ((COND) && !(SKB)->tstamp)			\
2109 			(SKB)->tstamp = ktime_get_real();	\
2110 	}							\
2111 
2112 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2113 {
2114 	return __is_skb_forwardable(dev, skb, true);
2115 }
2116 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2117 
2118 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2119 			      bool check_mtu)
2120 {
2121 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2122 
2123 	if (likely(!ret)) {
2124 		skb->protocol = eth_type_trans(skb, dev);
2125 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2126 	}
2127 
2128 	return ret;
2129 }
2130 
2131 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2132 {
2133 	return __dev_forward_skb2(dev, skb, true);
2134 }
2135 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2136 
2137 /**
2138  * dev_forward_skb - loopback an skb to another netif
2139  *
2140  * @dev: destination network device
2141  * @skb: buffer to forward
2142  *
2143  * return values:
2144  *	NET_RX_SUCCESS	(no congestion)
2145  *	NET_RX_DROP     (packet was dropped, but freed)
2146  *
2147  * dev_forward_skb can be used for injecting an skb from the
2148  * start_xmit function of one device into the receive queue
2149  * of another device.
2150  *
2151  * The receiving device may be in another namespace, so
2152  * we have to clear all information in the skb that could
2153  * impact namespace isolation.
2154  */
2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2156 {
2157 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2158 }
2159 EXPORT_SYMBOL_GPL(dev_forward_skb);
2160 
2161 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2162 {
2163 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2164 }
2165 
2166 static inline int deliver_skb(struct sk_buff *skb,
2167 			      struct packet_type *pt_prev,
2168 			      struct net_device *orig_dev)
2169 {
2170 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2171 		return -ENOMEM;
2172 	refcount_inc(&skb->users);
2173 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2174 }
2175 
2176 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2177 					  struct packet_type **pt,
2178 					  struct net_device *orig_dev,
2179 					  __be16 type,
2180 					  struct list_head *ptype_list)
2181 {
2182 	struct packet_type *ptype, *pt_prev = *pt;
2183 
2184 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2185 		if (ptype->type != type)
2186 			continue;
2187 		if (pt_prev)
2188 			deliver_skb(skb, pt_prev, orig_dev);
2189 		pt_prev = ptype;
2190 	}
2191 	*pt = pt_prev;
2192 }
2193 
2194 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2195 {
2196 	if (!ptype->af_packet_priv || !skb->sk)
2197 		return false;
2198 
2199 	if (ptype->id_match)
2200 		return ptype->id_match(ptype, skb->sk);
2201 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2202 		return true;
2203 
2204 	return false;
2205 }
2206 
2207 /**
2208  * dev_nit_active - return true if any network interface taps are in use
2209  *
2210  * @dev: network device to check for the presence of taps
2211  */
2212 bool dev_nit_active(struct net_device *dev)
2213 {
2214 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2215 }
2216 EXPORT_SYMBOL_GPL(dev_nit_active);
2217 
2218 /*
2219  *	Support routine. Sends outgoing frames to any network
2220  *	taps currently in use.
2221  */
2222 
2223 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2224 {
2225 	struct packet_type *ptype;
2226 	struct sk_buff *skb2 = NULL;
2227 	struct packet_type *pt_prev = NULL;
2228 	struct list_head *ptype_list = &ptype_all;
2229 
2230 	rcu_read_lock();
2231 again:
2232 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2233 		if (ptype->ignore_outgoing)
2234 			continue;
2235 
2236 		/* Never send packets back to the socket
2237 		 * they originated from - MvS (miquels@drinkel.ow.org)
2238 		 */
2239 		if (skb_loop_sk(ptype, skb))
2240 			continue;
2241 
2242 		if (pt_prev) {
2243 			deliver_skb(skb2, pt_prev, skb->dev);
2244 			pt_prev = ptype;
2245 			continue;
2246 		}
2247 
2248 		/* need to clone skb, done only once */
2249 		skb2 = skb_clone(skb, GFP_ATOMIC);
2250 		if (!skb2)
2251 			goto out_unlock;
2252 
2253 		net_timestamp_set(skb2);
2254 
2255 		/* skb->nh should be correctly
2256 		 * set by sender, so that the second statement is
2257 		 * just protection against buggy protocols.
2258 		 */
2259 		skb_reset_mac_header(skb2);
2260 
2261 		if (skb_network_header(skb2) < skb2->data ||
2262 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2263 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2264 					     ntohs(skb2->protocol),
2265 					     dev->name);
2266 			skb_reset_network_header(skb2);
2267 		}
2268 
2269 		skb2->transport_header = skb2->network_header;
2270 		skb2->pkt_type = PACKET_OUTGOING;
2271 		pt_prev = ptype;
2272 	}
2273 
2274 	if (ptype_list == &ptype_all) {
2275 		ptype_list = &dev->ptype_all;
2276 		goto again;
2277 	}
2278 out_unlock:
2279 	if (pt_prev) {
2280 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2281 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2282 		else
2283 			kfree_skb(skb2);
2284 	}
2285 	rcu_read_unlock();
2286 }
2287 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2288 
2289 /**
2290  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2291  * @dev: Network device
2292  * @txq: number of queues available
2293  *
2294  * If real_num_tx_queues is changed the tc mappings may no longer be
2295  * valid. To resolve this verify the tc mapping remains valid and if
2296  * not NULL the mapping. With no priorities mapping to this
2297  * offset/count pair it will no longer be used. In the worst case TC0
2298  * is invalid nothing can be done so disable priority mappings. If is
2299  * expected that drivers will fix this mapping if they can before
2300  * calling netif_set_real_num_tx_queues.
2301  */
2302 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2303 {
2304 	int i;
2305 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2306 
2307 	/* If TC0 is invalidated disable TC mapping */
2308 	if (tc->offset + tc->count > txq) {
2309 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2310 		dev->num_tc = 0;
2311 		return;
2312 	}
2313 
2314 	/* Invalidated prio to tc mappings set to TC0 */
2315 	for (i = 1; i < TC_BITMASK + 1; i++) {
2316 		int q = netdev_get_prio_tc_map(dev, i);
2317 
2318 		tc = &dev->tc_to_txq[q];
2319 		if (tc->offset + tc->count > txq) {
2320 			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",
2321 				    i, q);
2322 			netdev_set_prio_tc_map(dev, i, 0);
2323 		}
2324 	}
2325 }
2326 
2327 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2328 {
2329 	if (dev->num_tc) {
2330 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2331 		int i;
2332 
2333 		/* walk through the TCs and see if it falls into any of them */
2334 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2335 			if ((txq - tc->offset) < tc->count)
2336 				return i;
2337 		}
2338 
2339 		/* didn't find it, just return -1 to indicate no match */
2340 		return -1;
2341 	}
2342 
2343 	return 0;
2344 }
2345 EXPORT_SYMBOL(netdev_txq_to_tc);
2346 
2347 #ifdef CONFIG_XPS
2348 static struct static_key xps_needed __read_mostly;
2349 static struct static_key xps_rxqs_needed __read_mostly;
2350 static DEFINE_MUTEX(xps_map_mutex);
2351 #define xmap_dereference(P)		\
2352 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2353 
2354 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2355 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2356 {
2357 	struct xps_map *map = NULL;
2358 	int pos;
2359 
2360 	if (dev_maps)
2361 		map = xmap_dereference(dev_maps->attr_map[tci]);
2362 	if (!map)
2363 		return false;
2364 
2365 	for (pos = map->len; pos--;) {
2366 		if (map->queues[pos] != index)
2367 			continue;
2368 
2369 		if (map->len > 1) {
2370 			map->queues[pos] = map->queues[--map->len];
2371 			break;
2372 		}
2373 
2374 		if (old_maps)
2375 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2376 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2377 		kfree_rcu(map, rcu);
2378 		return false;
2379 	}
2380 
2381 	return true;
2382 }
2383 
2384 static bool remove_xps_queue_cpu(struct net_device *dev,
2385 				 struct xps_dev_maps *dev_maps,
2386 				 int cpu, u16 offset, u16 count)
2387 {
2388 	int num_tc = dev_maps->num_tc;
2389 	bool active = false;
2390 	int tci;
2391 
2392 	for (tci = cpu * num_tc; num_tc--; tci++) {
2393 		int i, j;
2394 
2395 		for (i = count, j = offset; i--; j++) {
2396 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2397 				break;
2398 		}
2399 
2400 		active |= i < 0;
2401 	}
2402 
2403 	return active;
2404 }
2405 
2406 static void reset_xps_maps(struct net_device *dev,
2407 			   struct xps_dev_maps *dev_maps,
2408 			   enum xps_map_type type)
2409 {
2410 	static_key_slow_dec_cpuslocked(&xps_needed);
2411 	if (type == XPS_RXQS)
2412 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2413 
2414 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2415 
2416 	kfree_rcu(dev_maps, rcu);
2417 }
2418 
2419 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2420 			   u16 offset, u16 count)
2421 {
2422 	struct xps_dev_maps *dev_maps;
2423 	bool active = false;
2424 	int i, j;
2425 
2426 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2427 	if (!dev_maps)
2428 		return;
2429 
2430 	for (j = 0; j < dev_maps->nr_ids; j++)
2431 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2432 	if (!active)
2433 		reset_xps_maps(dev, dev_maps, type);
2434 
2435 	if (type == XPS_CPUS) {
2436 		for (i = offset + (count - 1); count--; i--)
2437 			netdev_queue_numa_node_write(
2438 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2439 	}
2440 }
2441 
2442 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2443 				   u16 count)
2444 {
2445 	if (!static_key_false(&xps_needed))
2446 		return;
2447 
2448 	cpus_read_lock();
2449 	mutex_lock(&xps_map_mutex);
2450 
2451 	if (static_key_false(&xps_rxqs_needed))
2452 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2453 
2454 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2455 
2456 	mutex_unlock(&xps_map_mutex);
2457 	cpus_read_unlock();
2458 }
2459 
2460 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2461 {
2462 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2463 }
2464 
2465 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2466 				      u16 index, bool is_rxqs_map)
2467 {
2468 	struct xps_map *new_map;
2469 	int alloc_len = XPS_MIN_MAP_ALLOC;
2470 	int i, pos;
2471 
2472 	for (pos = 0; map && pos < map->len; pos++) {
2473 		if (map->queues[pos] != index)
2474 			continue;
2475 		return map;
2476 	}
2477 
2478 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2479 	if (map) {
2480 		if (pos < map->alloc_len)
2481 			return map;
2482 
2483 		alloc_len = map->alloc_len * 2;
2484 	}
2485 
2486 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2487 	 *  map
2488 	 */
2489 	if (is_rxqs_map)
2490 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2491 	else
2492 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2493 				       cpu_to_node(attr_index));
2494 	if (!new_map)
2495 		return NULL;
2496 
2497 	for (i = 0; i < pos; i++)
2498 		new_map->queues[i] = map->queues[i];
2499 	new_map->alloc_len = alloc_len;
2500 	new_map->len = pos;
2501 
2502 	return new_map;
2503 }
2504 
2505 /* Copy xps maps at a given index */
2506 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2507 			      struct xps_dev_maps *new_dev_maps, int index,
2508 			      int tc, bool skip_tc)
2509 {
2510 	int i, tci = index * dev_maps->num_tc;
2511 	struct xps_map *map;
2512 
2513 	/* copy maps belonging to foreign traffic classes */
2514 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2515 		if (i == tc && skip_tc)
2516 			continue;
2517 
2518 		/* fill in the new device map from the old device map */
2519 		map = xmap_dereference(dev_maps->attr_map[tci]);
2520 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2521 	}
2522 }
2523 
2524 /* Must be called under cpus_read_lock */
2525 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2526 			  u16 index, enum xps_map_type type)
2527 {
2528 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2529 	const unsigned long *online_mask = NULL;
2530 	bool active = false, copy = false;
2531 	int i, j, tci, numa_node_id = -2;
2532 	int maps_sz, num_tc = 1, tc = 0;
2533 	struct xps_map *map, *new_map;
2534 	unsigned int nr_ids;
2535 
2536 	WARN_ON_ONCE(index >= dev->num_tx_queues);
2537 
2538 	if (dev->num_tc) {
2539 		/* Do not allow XPS on subordinate device directly */
2540 		num_tc = dev->num_tc;
2541 		if (num_tc < 0)
2542 			return -EINVAL;
2543 
2544 		/* If queue belongs to subordinate dev use its map */
2545 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2546 
2547 		tc = netdev_txq_to_tc(dev, index);
2548 		if (tc < 0)
2549 			return -EINVAL;
2550 	}
2551 
2552 	mutex_lock(&xps_map_mutex);
2553 
2554 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2555 	if (type == XPS_RXQS) {
2556 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2557 		nr_ids = dev->num_rx_queues;
2558 	} else {
2559 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2560 		if (num_possible_cpus() > 1)
2561 			online_mask = cpumask_bits(cpu_online_mask);
2562 		nr_ids = nr_cpu_ids;
2563 	}
2564 
2565 	if (maps_sz < L1_CACHE_BYTES)
2566 		maps_sz = L1_CACHE_BYTES;
2567 
2568 	/* The old dev_maps could be larger or smaller than the one we're
2569 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2570 	 * between. We could try to be smart, but let's be safe instead and only
2571 	 * copy foreign traffic classes if the two map sizes match.
2572 	 */
2573 	if (dev_maps &&
2574 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2575 		copy = true;
2576 
2577 	/* allocate memory for queue storage */
2578 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2579 	     j < nr_ids;) {
2580 		if (!new_dev_maps) {
2581 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2582 			if (!new_dev_maps) {
2583 				mutex_unlock(&xps_map_mutex);
2584 				return -ENOMEM;
2585 			}
2586 
2587 			new_dev_maps->nr_ids = nr_ids;
2588 			new_dev_maps->num_tc = num_tc;
2589 		}
2590 
2591 		tci = j * num_tc + tc;
2592 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2593 
2594 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2595 		if (!map)
2596 			goto error;
2597 
2598 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2599 	}
2600 
2601 	if (!new_dev_maps)
2602 		goto out_no_new_maps;
2603 
2604 	if (!dev_maps) {
2605 		/* Increment static keys at most once per type */
2606 		static_key_slow_inc_cpuslocked(&xps_needed);
2607 		if (type == XPS_RXQS)
2608 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2609 	}
2610 
2611 	for (j = 0; j < nr_ids; j++) {
2612 		bool skip_tc = false;
2613 
2614 		tci = j * num_tc + tc;
2615 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2616 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2617 			/* add tx-queue to CPU/rx-queue maps */
2618 			int pos = 0;
2619 
2620 			skip_tc = true;
2621 
2622 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2623 			while ((pos < map->len) && (map->queues[pos] != index))
2624 				pos++;
2625 
2626 			if (pos == map->len)
2627 				map->queues[map->len++] = index;
2628 #ifdef CONFIG_NUMA
2629 			if (type == XPS_CPUS) {
2630 				if (numa_node_id == -2)
2631 					numa_node_id = cpu_to_node(j);
2632 				else if (numa_node_id != cpu_to_node(j))
2633 					numa_node_id = -1;
2634 			}
2635 #endif
2636 		}
2637 
2638 		if (copy)
2639 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2640 					  skip_tc);
2641 	}
2642 
2643 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2644 
2645 	/* Cleanup old maps */
2646 	if (!dev_maps)
2647 		goto out_no_old_maps;
2648 
2649 	for (j = 0; j < dev_maps->nr_ids; j++) {
2650 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2651 			map = xmap_dereference(dev_maps->attr_map[tci]);
2652 			if (!map)
2653 				continue;
2654 
2655 			if (copy) {
2656 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2657 				if (map == new_map)
2658 					continue;
2659 			}
2660 
2661 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2662 			kfree_rcu(map, rcu);
2663 		}
2664 	}
2665 
2666 	old_dev_maps = dev_maps;
2667 
2668 out_no_old_maps:
2669 	dev_maps = new_dev_maps;
2670 	active = true;
2671 
2672 out_no_new_maps:
2673 	if (type == XPS_CPUS)
2674 		/* update Tx queue numa node */
2675 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2676 					     (numa_node_id >= 0) ?
2677 					     numa_node_id : NUMA_NO_NODE);
2678 
2679 	if (!dev_maps)
2680 		goto out_no_maps;
2681 
2682 	/* removes tx-queue from unused CPUs/rx-queues */
2683 	for (j = 0; j < dev_maps->nr_ids; j++) {
2684 		tci = j * dev_maps->num_tc;
2685 
2686 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2687 			if (i == tc &&
2688 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2689 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2690 				continue;
2691 
2692 			active |= remove_xps_queue(dev_maps,
2693 						   copy ? old_dev_maps : NULL,
2694 						   tci, index);
2695 		}
2696 	}
2697 
2698 	if (old_dev_maps)
2699 		kfree_rcu(old_dev_maps, rcu);
2700 
2701 	/* free map if not active */
2702 	if (!active)
2703 		reset_xps_maps(dev, dev_maps, type);
2704 
2705 out_no_maps:
2706 	mutex_unlock(&xps_map_mutex);
2707 
2708 	return 0;
2709 error:
2710 	/* remove any maps that we added */
2711 	for (j = 0; j < nr_ids; j++) {
2712 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2713 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2714 			map = copy ?
2715 			      xmap_dereference(dev_maps->attr_map[tci]) :
2716 			      NULL;
2717 			if (new_map && new_map != map)
2718 				kfree(new_map);
2719 		}
2720 	}
2721 
2722 	mutex_unlock(&xps_map_mutex);
2723 
2724 	kfree(new_dev_maps);
2725 	return -ENOMEM;
2726 }
2727 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2728 
2729 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2730 			u16 index)
2731 {
2732 	int ret;
2733 
2734 	cpus_read_lock();
2735 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2736 	cpus_read_unlock();
2737 
2738 	return ret;
2739 }
2740 EXPORT_SYMBOL(netif_set_xps_queue);
2741 
2742 #endif
2743 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2744 {
2745 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2746 
2747 	/* Unbind any subordinate channels */
2748 	while (txq-- != &dev->_tx[0]) {
2749 		if (txq->sb_dev)
2750 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2751 	}
2752 }
2753 
2754 void netdev_reset_tc(struct net_device *dev)
2755 {
2756 #ifdef CONFIG_XPS
2757 	netif_reset_xps_queues_gt(dev, 0);
2758 #endif
2759 	netdev_unbind_all_sb_channels(dev);
2760 
2761 	/* Reset TC configuration of device */
2762 	dev->num_tc = 0;
2763 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2764 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2765 }
2766 EXPORT_SYMBOL(netdev_reset_tc);
2767 
2768 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2769 {
2770 	if (tc >= dev->num_tc)
2771 		return -EINVAL;
2772 
2773 #ifdef CONFIG_XPS
2774 	netif_reset_xps_queues(dev, offset, count);
2775 #endif
2776 	dev->tc_to_txq[tc].count = count;
2777 	dev->tc_to_txq[tc].offset = offset;
2778 	return 0;
2779 }
2780 EXPORT_SYMBOL(netdev_set_tc_queue);
2781 
2782 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2783 {
2784 	if (num_tc > TC_MAX_QUEUE)
2785 		return -EINVAL;
2786 
2787 #ifdef CONFIG_XPS
2788 	netif_reset_xps_queues_gt(dev, 0);
2789 #endif
2790 	netdev_unbind_all_sb_channels(dev);
2791 
2792 	dev->num_tc = num_tc;
2793 	return 0;
2794 }
2795 EXPORT_SYMBOL(netdev_set_num_tc);
2796 
2797 void netdev_unbind_sb_channel(struct net_device *dev,
2798 			      struct net_device *sb_dev)
2799 {
2800 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2801 
2802 #ifdef CONFIG_XPS
2803 	netif_reset_xps_queues_gt(sb_dev, 0);
2804 #endif
2805 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2806 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2807 
2808 	while (txq-- != &dev->_tx[0]) {
2809 		if (txq->sb_dev == sb_dev)
2810 			txq->sb_dev = NULL;
2811 	}
2812 }
2813 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2814 
2815 int netdev_bind_sb_channel_queue(struct net_device *dev,
2816 				 struct net_device *sb_dev,
2817 				 u8 tc, u16 count, u16 offset)
2818 {
2819 	/* Make certain the sb_dev and dev are already configured */
2820 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2821 		return -EINVAL;
2822 
2823 	/* We cannot hand out queues we don't have */
2824 	if ((offset + count) > dev->real_num_tx_queues)
2825 		return -EINVAL;
2826 
2827 	/* Record the mapping */
2828 	sb_dev->tc_to_txq[tc].count = count;
2829 	sb_dev->tc_to_txq[tc].offset = offset;
2830 
2831 	/* Provide a way for Tx queue to find the tc_to_txq map or
2832 	 * XPS map for itself.
2833 	 */
2834 	while (count--)
2835 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2836 
2837 	return 0;
2838 }
2839 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2840 
2841 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2842 {
2843 	/* Do not use a multiqueue device to represent a subordinate channel */
2844 	if (netif_is_multiqueue(dev))
2845 		return -ENODEV;
2846 
2847 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2848 	 * Channel 0 is meant to be "native" mode and used only to represent
2849 	 * the main root device. We allow writing 0 to reset the device back
2850 	 * to normal mode after being used as a subordinate channel.
2851 	 */
2852 	if (channel > S16_MAX)
2853 		return -EINVAL;
2854 
2855 	dev->num_tc = -channel;
2856 
2857 	return 0;
2858 }
2859 EXPORT_SYMBOL(netdev_set_sb_channel);
2860 
2861 /*
2862  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2863  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2864  */
2865 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2866 {
2867 	bool disabling;
2868 	int rc;
2869 
2870 	disabling = txq < dev->real_num_tx_queues;
2871 
2872 	if (txq < 1 || txq > dev->num_tx_queues)
2873 		return -EINVAL;
2874 
2875 	if (dev->reg_state == NETREG_REGISTERED ||
2876 	    dev->reg_state == NETREG_UNREGISTERING) {
2877 		ASSERT_RTNL();
2878 
2879 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2880 						  txq);
2881 		if (rc)
2882 			return rc;
2883 
2884 		if (dev->num_tc)
2885 			netif_setup_tc(dev, txq);
2886 
2887 		dev_qdisc_change_real_num_tx(dev, txq);
2888 
2889 		dev->real_num_tx_queues = txq;
2890 
2891 		if (disabling) {
2892 			synchronize_net();
2893 			qdisc_reset_all_tx_gt(dev, txq);
2894 #ifdef CONFIG_XPS
2895 			netif_reset_xps_queues_gt(dev, txq);
2896 #endif
2897 		}
2898 	} else {
2899 		dev->real_num_tx_queues = txq;
2900 	}
2901 
2902 	return 0;
2903 }
2904 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2905 
2906 #ifdef CONFIG_SYSFS
2907 /**
2908  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2909  *	@dev: Network device
2910  *	@rxq: Actual number of RX queues
2911  *
2912  *	This must be called either with the rtnl_lock held or before
2913  *	registration of the net device.  Returns 0 on success, or a
2914  *	negative error code.  If called before registration, it always
2915  *	succeeds.
2916  */
2917 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2918 {
2919 	int rc;
2920 
2921 	if (rxq < 1 || rxq > dev->num_rx_queues)
2922 		return -EINVAL;
2923 
2924 	if (dev->reg_state == NETREG_REGISTERED) {
2925 		ASSERT_RTNL();
2926 
2927 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2928 						  rxq);
2929 		if (rc)
2930 			return rc;
2931 	}
2932 
2933 	dev->real_num_rx_queues = rxq;
2934 	return 0;
2935 }
2936 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2937 #endif
2938 
2939 /**
2940  *	netif_set_real_num_queues - set actual number of RX and TX queues used
2941  *	@dev: Network device
2942  *	@txq: Actual number of TX queues
2943  *	@rxq: Actual number of RX queues
2944  *
2945  *	Set the real number of both TX and RX queues.
2946  *	Does nothing if the number of queues is already correct.
2947  */
2948 int netif_set_real_num_queues(struct net_device *dev,
2949 			      unsigned int txq, unsigned int rxq)
2950 {
2951 	unsigned int old_rxq = dev->real_num_rx_queues;
2952 	int err;
2953 
2954 	if (txq < 1 || txq > dev->num_tx_queues ||
2955 	    rxq < 1 || rxq > dev->num_rx_queues)
2956 		return -EINVAL;
2957 
2958 	/* Start from increases, so the error path only does decreases -
2959 	 * decreases can't fail.
2960 	 */
2961 	if (rxq > dev->real_num_rx_queues) {
2962 		err = netif_set_real_num_rx_queues(dev, rxq);
2963 		if (err)
2964 			return err;
2965 	}
2966 	if (txq > dev->real_num_tx_queues) {
2967 		err = netif_set_real_num_tx_queues(dev, txq);
2968 		if (err)
2969 			goto undo_rx;
2970 	}
2971 	if (rxq < dev->real_num_rx_queues)
2972 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2973 	if (txq < dev->real_num_tx_queues)
2974 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2975 
2976 	return 0;
2977 undo_rx:
2978 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2979 	return err;
2980 }
2981 EXPORT_SYMBOL(netif_set_real_num_queues);
2982 
2983 /**
2984  * netif_set_tso_max_size() - set the max size of TSO frames supported
2985  * @dev:	netdev to update
2986  * @size:	max skb->len of a TSO frame
2987  *
2988  * Set the limit on the size of TSO super-frames the device can handle.
2989  * Unless explicitly set the stack will assume the value of
2990  * %GSO_LEGACY_MAX_SIZE.
2991  */
2992 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
2993 {
2994 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
2995 	if (size < READ_ONCE(dev->gso_max_size))
2996 		netif_set_gso_max_size(dev, size);
2997 	if (size < READ_ONCE(dev->gso_ipv4_max_size))
2998 		netif_set_gso_ipv4_max_size(dev, size);
2999 }
3000 EXPORT_SYMBOL(netif_set_tso_max_size);
3001 
3002 /**
3003  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3004  * @dev:	netdev to update
3005  * @segs:	max number of TCP segments
3006  *
3007  * Set the limit on the number of TCP segments the device can generate from
3008  * a single TSO super-frame.
3009  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3010  */
3011 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3012 {
3013 	dev->tso_max_segs = segs;
3014 	if (segs < READ_ONCE(dev->gso_max_segs))
3015 		netif_set_gso_max_segs(dev, segs);
3016 }
3017 EXPORT_SYMBOL(netif_set_tso_max_segs);
3018 
3019 /**
3020  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3021  * @to:		netdev to update
3022  * @from:	netdev from which to copy the limits
3023  */
3024 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3025 {
3026 	netif_set_tso_max_size(to, from->tso_max_size);
3027 	netif_set_tso_max_segs(to, from->tso_max_segs);
3028 }
3029 EXPORT_SYMBOL(netif_inherit_tso_max);
3030 
3031 /**
3032  * netif_get_num_default_rss_queues - default number of RSS queues
3033  *
3034  * Default value is the number of physical cores if there are only 1 or 2, or
3035  * divided by 2 if there are more.
3036  */
3037 int netif_get_num_default_rss_queues(void)
3038 {
3039 	cpumask_var_t cpus;
3040 	int cpu, count = 0;
3041 
3042 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3043 		return 1;
3044 
3045 	cpumask_copy(cpus, cpu_online_mask);
3046 	for_each_cpu(cpu, cpus) {
3047 		++count;
3048 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3049 	}
3050 	free_cpumask_var(cpus);
3051 
3052 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3053 }
3054 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3055 
3056 static void __netif_reschedule(struct Qdisc *q)
3057 {
3058 	struct softnet_data *sd;
3059 	unsigned long flags;
3060 
3061 	local_irq_save(flags);
3062 	sd = this_cpu_ptr(&softnet_data);
3063 	q->next_sched = NULL;
3064 	*sd->output_queue_tailp = q;
3065 	sd->output_queue_tailp = &q->next_sched;
3066 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3067 	local_irq_restore(flags);
3068 }
3069 
3070 void __netif_schedule(struct Qdisc *q)
3071 {
3072 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3073 		__netif_reschedule(q);
3074 }
3075 EXPORT_SYMBOL(__netif_schedule);
3076 
3077 struct dev_kfree_skb_cb {
3078 	enum skb_drop_reason reason;
3079 };
3080 
3081 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3082 {
3083 	return (struct dev_kfree_skb_cb *)skb->cb;
3084 }
3085 
3086 void netif_schedule_queue(struct netdev_queue *txq)
3087 {
3088 	rcu_read_lock();
3089 	if (!netif_xmit_stopped(txq)) {
3090 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3091 
3092 		__netif_schedule(q);
3093 	}
3094 	rcu_read_unlock();
3095 }
3096 EXPORT_SYMBOL(netif_schedule_queue);
3097 
3098 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3099 {
3100 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3101 		struct Qdisc *q;
3102 
3103 		rcu_read_lock();
3104 		q = rcu_dereference(dev_queue->qdisc);
3105 		__netif_schedule(q);
3106 		rcu_read_unlock();
3107 	}
3108 }
3109 EXPORT_SYMBOL(netif_tx_wake_queue);
3110 
3111 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3112 {
3113 	unsigned long flags;
3114 
3115 	if (unlikely(!skb))
3116 		return;
3117 
3118 	if (likely(refcount_read(&skb->users) == 1)) {
3119 		smp_rmb();
3120 		refcount_set(&skb->users, 0);
3121 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3122 		return;
3123 	}
3124 	get_kfree_skb_cb(skb)->reason = reason;
3125 	local_irq_save(flags);
3126 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3127 	__this_cpu_write(softnet_data.completion_queue, skb);
3128 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3129 	local_irq_restore(flags);
3130 }
3131 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3132 
3133 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3134 {
3135 	if (in_hardirq() || irqs_disabled())
3136 		dev_kfree_skb_irq_reason(skb, reason);
3137 	else
3138 		kfree_skb_reason(skb, reason);
3139 }
3140 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3141 
3142 
3143 /**
3144  * netif_device_detach - mark device as removed
3145  * @dev: network device
3146  *
3147  * Mark device as removed from system and therefore no longer available.
3148  */
3149 void netif_device_detach(struct net_device *dev)
3150 {
3151 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3152 	    netif_running(dev)) {
3153 		netif_tx_stop_all_queues(dev);
3154 	}
3155 }
3156 EXPORT_SYMBOL(netif_device_detach);
3157 
3158 /**
3159  * netif_device_attach - mark device as attached
3160  * @dev: network device
3161  *
3162  * Mark device as attached from system and restart if needed.
3163  */
3164 void netif_device_attach(struct net_device *dev)
3165 {
3166 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3167 	    netif_running(dev)) {
3168 		netif_tx_wake_all_queues(dev);
3169 		__netdev_watchdog_up(dev);
3170 	}
3171 }
3172 EXPORT_SYMBOL(netif_device_attach);
3173 
3174 /*
3175  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3176  * to be used as a distribution range.
3177  */
3178 static u16 skb_tx_hash(const struct net_device *dev,
3179 		       const struct net_device *sb_dev,
3180 		       struct sk_buff *skb)
3181 {
3182 	u32 hash;
3183 	u16 qoffset = 0;
3184 	u16 qcount = dev->real_num_tx_queues;
3185 
3186 	if (dev->num_tc) {
3187 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3188 
3189 		qoffset = sb_dev->tc_to_txq[tc].offset;
3190 		qcount = sb_dev->tc_to_txq[tc].count;
3191 		if (unlikely(!qcount)) {
3192 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3193 					     sb_dev->name, qoffset, tc);
3194 			qoffset = 0;
3195 			qcount = dev->real_num_tx_queues;
3196 		}
3197 	}
3198 
3199 	if (skb_rx_queue_recorded(skb)) {
3200 		DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3201 		hash = skb_get_rx_queue(skb);
3202 		if (hash >= qoffset)
3203 			hash -= qoffset;
3204 		while (unlikely(hash >= qcount))
3205 			hash -= qcount;
3206 		return hash + qoffset;
3207 	}
3208 
3209 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3210 }
3211 
3212 void skb_warn_bad_offload(const struct sk_buff *skb)
3213 {
3214 	static const netdev_features_t null_features;
3215 	struct net_device *dev = skb->dev;
3216 	const char *name = "";
3217 
3218 	if (!net_ratelimit())
3219 		return;
3220 
3221 	if (dev) {
3222 		if (dev->dev.parent)
3223 			name = dev_driver_string(dev->dev.parent);
3224 		else
3225 			name = netdev_name(dev);
3226 	}
3227 	skb_dump(KERN_WARNING, skb, false);
3228 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3229 	     name, dev ? &dev->features : &null_features,
3230 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3231 }
3232 
3233 /*
3234  * Invalidate hardware checksum when packet is to be mangled, and
3235  * complete checksum manually on outgoing path.
3236  */
3237 int skb_checksum_help(struct sk_buff *skb)
3238 {
3239 	__wsum csum;
3240 	int ret = 0, offset;
3241 
3242 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3243 		goto out_set_summed;
3244 
3245 	if (unlikely(skb_is_gso(skb))) {
3246 		skb_warn_bad_offload(skb);
3247 		return -EINVAL;
3248 	}
3249 
3250 	/* Before computing a checksum, we should make sure no frag could
3251 	 * be modified by an external entity : checksum could be wrong.
3252 	 */
3253 	if (skb_has_shared_frag(skb)) {
3254 		ret = __skb_linearize(skb);
3255 		if (ret)
3256 			goto out;
3257 	}
3258 
3259 	offset = skb_checksum_start_offset(skb);
3260 	ret = -EINVAL;
3261 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3262 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3263 		goto out;
3264 	}
3265 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3266 
3267 	offset += skb->csum_offset;
3268 	if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3269 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3270 		goto out;
3271 	}
3272 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3273 	if (ret)
3274 		goto out;
3275 
3276 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3277 out_set_summed:
3278 	skb->ip_summed = CHECKSUM_NONE;
3279 out:
3280 	return ret;
3281 }
3282 EXPORT_SYMBOL(skb_checksum_help);
3283 
3284 int skb_crc32c_csum_help(struct sk_buff *skb)
3285 {
3286 	__le32 crc32c_csum;
3287 	int ret = 0, offset, start;
3288 
3289 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3290 		goto out;
3291 
3292 	if (unlikely(skb_is_gso(skb)))
3293 		goto out;
3294 
3295 	/* Before computing a checksum, we should make sure no frag could
3296 	 * be modified by an external entity : checksum could be wrong.
3297 	 */
3298 	if (unlikely(skb_has_shared_frag(skb))) {
3299 		ret = __skb_linearize(skb);
3300 		if (ret)
3301 			goto out;
3302 	}
3303 	start = skb_checksum_start_offset(skb);
3304 	offset = start + offsetof(struct sctphdr, checksum);
3305 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3306 		ret = -EINVAL;
3307 		goto out;
3308 	}
3309 
3310 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3311 	if (ret)
3312 		goto out;
3313 
3314 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3315 						  skb->len - start, ~(__u32)0,
3316 						  crc32c_csum_stub));
3317 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3318 	skb_reset_csum_not_inet(skb);
3319 out:
3320 	return ret;
3321 }
3322 
3323 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3324 {
3325 	__be16 type = skb->protocol;
3326 
3327 	/* Tunnel gso handlers can set protocol to ethernet. */
3328 	if (type == htons(ETH_P_TEB)) {
3329 		struct ethhdr *eth;
3330 
3331 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3332 			return 0;
3333 
3334 		eth = (struct ethhdr *)skb->data;
3335 		type = eth->h_proto;
3336 	}
3337 
3338 	return vlan_get_protocol_and_depth(skb, type, depth);
3339 }
3340 
3341 
3342 /* Take action when hardware reception checksum errors are detected. */
3343 #ifdef CONFIG_BUG
3344 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3345 {
3346 	netdev_err(dev, "hw csum failure\n");
3347 	skb_dump(KERN_ERR, skb, true);
3348 	dump_stack();
3349 }
3350 
3351 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3352 {
3353 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3354 }
3355 EXPORT_SYMBOL(netdev_rx_csum_fault);
3356 #endif
3357 
3358 /* XXX: check that highmem exists at all on the given machine. */
3359 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3360 {
3361 #ifdef CONFIG_HIGHMEM
3362 	int i;
3363 
3364 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3365 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3366 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3367 
3368 			if (PageHighMem(skb_frag_page(frag)))
3369 				return 1;
3370 		}
3371 	}
3372 #endif
3373 	return 0;
3374 }
3375 
3376 /* If MPLS offload request, verify we are testing hardware MPLS features
3377  * instead of standard features for the netdev.
3378  */
3379 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3380 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3381 					   netdev_features_t features,
3382 					   __be16 type)
3383 {
3384 	if (eth_p_mpls(type))
3385 		features &= skb->dev->mpls_features;
3386 
3387 	return features;
3388 }
3389 #else
3390 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3391 					   netdev_features_t features,
3392 					   __be16 type)
3393 {
3394 	return features;
3395 }
3396 #endif
3397 
3398 static netdev_features_t harmonize_features(struct sk_buff *skb,
3399 	netdev_features_t features)
3400 {
3401 	__be16 type;
3402 
3403 	type = skb_network_protocol(skb, NULL);
3404 	features = net_mpls_features(skb, features, type);
3405 
3406 	if (skb->ip_summed != CHECKSUM_NONE &&
3407 	    !can_checksum_protocol(features, type)) {
3408 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3409 	}
3410 	if (illegal_highdma(skb->dev, skb))
3411 		features &= ~NETIF_F_SG;
3412 
3413 	return features;
3414 }
3415 
3416 netdev_features_t passthru_features_check(struct sk_buff *skb,
3417 					  struct net_device *dev,
3418 					  netdev_features_t features)
3419 {
3420 	return features;
3421 }
3422 EXPORT_SYMBOL(passthru_features_check);
3423 
3424 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3425 					     struct net_device *dev,
3426 					     netdev_features_t features)
3427 {
3428 	return vlan_features_check(skb, features);
3429 }
3430 
3431 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3432 					    struct net_device *dev,
3433 					    netdev_features_t features)
3434 {
3435 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3436 
3437 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3438 		return features & ~NETIF_F_GSO_MASK;
3439 
3440 	if (!skb_shinfo(skb)->gso_type) {
3441 		skb_warn_bad_offload(skb);
3442 		return features & ~NETIF_F_GSO_MASK;
3443 	}
3444 
3445 	/* Support for GSO partial features requires software
3446 	 * intervention before we can actually process the packets
3447 	 * so we need to strip support for any partial features now
3448 	 * and we can pull them back in after we have partially
3449 	 * segmented the frame.
3450 	 */
3451 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3452 		features &= ~dev->gso_partial_features;
3453 
3454 	/* Make sure to clear the IPv4 ID mangling feature if the
3455 	 * IPv4 header has the potential to be fragmented.
3456 	 */
3457 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3458 		struct iphdr *iph = skb->encapsulation ?
3459 				    inner_ip_hdr(skb) : ip_hdr(skb);
3460 
3461 		if (!(iph->frag_off & htons(IP_DF)))
3462 			features &= ~NETIF_F_TSO_MANGLEID;
3463 	}
3464 
3465 	return features;
3466 }
3467 
3468 netdev_features_t netif_skb_features(struct sk_buff *skb)
3469 {
3470 	struct net_device *dev = skb->dev;
3471 	netdev_features_t features = dev->features;
3472 
3473 	if (skb_is_gso(skb))
3474 		features = gso_features_check(skb, dev, features);
3475 
3476 	/* If encapsulation offload request, verify we are testing
3477 	 * hardware encapsulation features instead of standard
3478 	 * features for the netdev
3479 	 */
3480 	if (skb->encapsulation)
3481 		features &= dev->hw_enc_features;
3482 
3483 	if (skb_vlan_tagged(skb))
3484 		features = netdev_intersect_features(features,
3485 						     dev->vlan_features |
3486 						     NETIF_F_HW_VLAN_CTAG_TX |
3487 						     NETIF_F_HW_VLAN_STAG_TX);
3488 
3489 	if (dev->netdev_ops->ndo_features_check)
3490 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3491 								features);
3492 	else
3493 		features &= dflt_features_check(skb, dev, features);
3494 
3495 	return harmonize_features(skb, features);
3496 }
3497 EXPORT_SYMBOL(netif_skb_features);
3498 
3499 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3500 		    struct netdev_queue *txq, bool more)
3501 {
3502 	unsigned int len;
3503 	int rc;
3504 
3505 	if (dev_nit_active(dev))
3506 		dev_queue_xmit_nit(skb, dev);
3507 
3508 	len = skb->len;
3509 	trace_net_dev_start_xmit(skb, dev);
3510 	rc = netdev_start_xmit(skb, dev, txq, more);
3511 	trace_net_dev_xmit(skb, rc, dev, len);
3512 
3513 	return rc;
3514 }
3515 
3516 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3517 				    struct netdev_queue *txq, int *ret)
3518 {
3519 	struct sk_buff *skb = first;
3520 	int rc = NETDEV_TX_OK;
3521 
3522 	while (skb) {
3523 		struct sk_buff *next = skb->next;
3524 
3525 		skb_mark_not_on_list(skb);
3526 		rc = xmit_one(skb, dev, txq, next != NULL);
3527 		if (unlikely(!dev_xmit_complete(rc))) {
3528 			skb->next = next;
3529 			goto out;
3530 		}
3531 
3532 		skb = next;
3533 		if (netif_tx_queue_stopped(txq) && skb) {
3534 			rc = NETDEV_TX_BUSY;
3535 			break;
3536 		}
3537 	}
3538 
3539 out:
3540 	*ret = rc;
3541 	return skb;
3542 }
3543 
3544 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3545 					  netdev_features_t features)
3546 {
3547 	if (skb_vlan_tag_present(skb) &&
3548 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3549 		skb = __vlan_hwaccel_push_inside(skb);
3550 	return skb;
3551 }
3552 
3553 int skb_csum_hwoffload_help(struct sk_buff *skb,
3554 			    const netdev_features_t features)
3555 {
3556 	if (unlikely(skb_csum_is_sctp(skb)))
3557 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3558 			skb_crc32c_csum_help(skb);
3559 
3560 	if (features & NETIF_F_HW_CSUM)
3561 		return 0;
3562 
3563 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3564 		switch (skb->csum_offset) {
3565 		case offsetof(struct tcphdr, check):
3566 		case offsetof(struct udphdr, check):
3567 			return 0;
3568 		}
3569 	}
3570 
3571 	return skb_checksum_help(skb);
3572 }
3573 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3574 
3575 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3576 {
3577 	netdev_features_t features;
3578 
3579 	features = netif_skb_features(skb);
3580 	skb = validate_xmit_vlan(skb, features);
3581 	if (unlikely(!skb))
3582 		goto out_null;
3583 
3584 	skb = sk_validate_xmit_skb(skb, dev);
3585 	if (unlikely(!skb))
3586 		goto out_null;
3587 
3588 	if (netif_needs_gso(skb, features)) {
3589 		struct sk_buff *segs;
3590 
3591 		segs = skb_gso_segment(skb, features);
3592 		if (IS_ERR(segs)) {
3593 			goto out_kfree_skb;
3594 		} else if (segs) {
3595 			consume_skb(skb);
3596 			skb = segs;
3597 		}
3598 	} else {
3599 		if (skb_needs_linearize(skb, features) &&
3600 		    __skb_linearize(skb))
3601 			goto out_kfree_skb;
3602 
3603 		/* If packet is not checksummed and device does not
3604 		 * support checksumming for this protocol, complete
3605 		 * checksumming here.
3606 		 */
3607 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3608 			if (skb->encapsulation)
3609 				skb_set_inner_transport_header(skb,
3610 							       skb_checksum_start_offset(skb));
3611 			else
3612 				skb_set_transport_header(skb,
3613 							 skb_checksum_start_offset(skb));
3614 			if (skb_csum_hwoffload_help(skb, features))
3615 				goto out_kfree_skb;
3616 		}
3617 	}
3618 
3619 	skb = validate_xmit_xfrm(skb, features, again);
3620 
3621 	return skb;
3622 
3623 out_kfree_skb:
3624 	kfree_skb(skb);
3625 out_null:
3626 	dev_core_stats_tx_dropped_inc(dev);
3627 	return NULL;
3628 }
3629 
3630 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3631 {
3632 	struct sk_buff *next, *head = NULL, *tail;
3633 
3634 	for (; skb != NULL; skb = next) {
3635 		next = skb->next;
3636 		skb_mark_not_on_list(skb);
3637 
3638 		/* in case skb wont be segmented, point to itself */
3639 		skb->prev = skb;
3640 
3641 		skb = validate_xmit_skb(skb, dev, again);
3642 		if (!skb)
3643 			continue;
3644 
3645 		if (!head)
3646 			head = skb;
3647 		else
3648 			tail->next = skb;
3649 		/* If skb was segmented, skb->prev points to
3650 		 * the last segment. If not, it still contains skb.
3651 		 */
3652 		tail = skb->prev;
3653 	}
3654 	return head;
3655 }
3656 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3657 
3658 static void qdisc_pkt_len_init(struct sk_buff *skb)
3659 {
3660 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3661 
3662 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3663 
3664 	/* To get more precise estimation of bytes sent on wire,
3665 	 * we add to pkt_len the headers size of all segments
3666 	 */
3667 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3668 		u16 gso_segs = shinfo->gso_segs;
3669 		unsigned int hdr_len;
3670 
3671 		/* mac layer + network layer */
3672 		hdr_len = skb_transport_offset(skb);
3673 
3674 		/* + transport layer */
3675 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3676 			const struct tcphdr *th;
3677 			struct tcphdr _tcphdr;
3678 
3679 			th = skb_header_pointer(skb, hdr_len,
3680 						sizeof(_tcphdr), &_tcphdr);
3681 			if (likely(th))
3682 				hdr_len += __tcp_hdrlen(th);
3683 		} else {
3684 			struct udphdr _udphdr;
3685 
3686 			if (skb_header_pointer(skb, hdr_len,
3687 					       sizeof(_udphdr), &_udphdr))
3688 				hdr_len += sizeof(struct udphdr);
3689 		}
3690 
3691 		if (shinfo->gso_type & SKB_GSO_DODGY)
3692 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3693 						shinfo->gso_size);
3694 
3695 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3696 	}
3697 }
3698 
3699 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3700 			     struct sk_buff **to_free,
3701 			     struct netdev_queue *txq)
3702 {
3703 	int rc;
3704 
3705 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3706 	if (rc == NET_XMIT_SUCCESS)
3707 		trace_qdisc_enqueue(q, txq, skb);
3708 	return rc;
3709 }
3710 
3711 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3712 				 struct net_device *dev,
3713 				 struct netdev_queue *txq)
3714 {
3715 	spinlock_t *root_lock = qdisc_lock(q);
3716 	struct sk_buff *to_free = NULL;
3717 	bool contended;
3718 	int rc;
3719 
3720 	qdisc_calculate_pkt_len(skb, q);
3721 
3722 	if (q->flags & TCQ_F_NOLOCK) {
3723 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3724 		    qdisc_run_begin(q)) {
3725 			/* Retest nolock_qdisc_is_empty() within the protection
3726 			 * of q->seqlock to protect from racing with requeuing.
3727 			 */
3728 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3729 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3730 				__qdisc_run(q);
3731 				qdisc_run_end(q);
3732 
3733 				goto no_lock_out;
3734 			}
3735 
3736 			qdisc_bstats_cpu_update(q, skb);
3737 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3738 			    !nolock_qdisc_is_empty(q))
3739 				__qdisc_run(q);
3740 
3741 			qdisc_run_end(q);
3742 			return NET_XMIT_SUCCESS;
3743 		}
3744 
3745 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3746 		qdisc_run(q);
3747 
3748 no_lock_out:
3749 		if (unlikely(to_free))
3750 			kfree_skb_list_reason(to_free,
3751 					      SKB_DROP_REASON_QDISC_DROP);
3752 		return rc;
3753 	}
3754 
3755 	/*
3756 	 * Heuristic to force contended enqueues to serialize on a
3757 	 * separate lock before trying to get qdisc main lock.
3758 	 * This permits qdisc->running owner to get the lock more
3759 	 * often and dequeue packets faster.
3760 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3761 	 * and then other tasks will only enqueue packets. The packets will be
3762 	 * sent after the qdisc owner is scheduled again. To prevent this
3763 	 * scenario the task always serialize on the lock.
3764 	 */
3765 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3766 	if (unlikely(contended))
3767 		spin_lock(&q->busylock);
3768 
3769 	spin_lock(root_lock);
3770 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3771 		__qdisc_drop(skb, &to_free);
3772 		rc = NET_XMIT_DROP;
3773 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3774 		   qdisc_run_begin(q)) {
3775 		/*
3776 		 * This is a work-conserving queue; there are no old skbs
3777 		 * waiting to be sent out; and the qdisc is not running -
3778 		 * xmit the skb directly.
3779 		 */
3780 
3781 		qdisc_bstats_update(q, skb);
3782 
3783 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3784 			if (unlikely(contended)) {
3785 				spin_unlock(&q->busylock);
3786 				contended = false;
3787 			}
3788 			__qdisc_run(q);
3789 		}
3790 
3791 		qdisc_run_end(q);
3792 		rc = NET_XMIT_SUCCESS;
3793 	} else {
3794 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3795 		if (qdisc_run_begin(q)) {
3796 			if (unlikely(contended)) {
3797 				spin_unlock(&q->busylock);
3798 				contended = false;
3799 			}
3800 			__qdisc_run(q);
3801 			qdisc_run_end(q);
3802 		}
3803 	}
3804 	spin_unlock(root_lock);
3805 	if (unlikely(to_free))
3806 		kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3807 	if (unlikely(contended))
3808 		spin_unlock(&q->busylock);
3809 	return rc;
3810 }
3811 
3812 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3813 static void skb_update_prio(struct sk_buff *skb)
3814 {
3815 	const struct netprio_map *map;
3816 	const struct sock *sk;
3817 	unsigned int prioidx;
3818 
3819 	if (skb->priority)
3820 		return;
3821 	map = rcu_dereference_bh(skb->dev->priomap);
3822 	if (!map)
3823 		return;
3824 	sk = skb_to_full_sk(skb);
3825 	if (!sk)
3826 		return;
3827 
3828 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3829 
3830 	if (prioidx < map->priomap_len)
3831 		skb->priority = map->priomap[prioidx];
3832 }
3833 #else
3834 #define skb_update_prio(skb)
3835 #endif
3836 
3837 /**
3838  *	dev_loopback_xmit - loop back @skb
3839  *	@net: network namespace this loopback is happening in
3840  *	@sk:  sk needed to be a netfilter okfn
3841  *	@skb: buffer to transmit
3842  */
3843 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3844 {
3845 	skb_reset_mac_header(skb);
3846 	__skb_pull(skb, skb_network_offset(skb));
3847 	skb->pkt_type = PACKET_LOOPBACK;
3848 	if (skb->ip_summed == CHECKSUM_NONE)
3849 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3850 	DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3851 	skb_dst_force(skb);
3852 	netif_rx(skb);
3853 	return 0;
3854 }
3855 EXPORT_SYMBOL(dev_loopback_xmit);
3856 
3857 #ifdef CONFIG_NET_EGRESS
3858 static struct sk_buff *
3859 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3860 {
3861 #ifdef CONFIG_NET_CLS_ACT
3862 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3863 	struct tcf_result cl_res;
3864 
3865 	if (!miniq)
3866 		return skb;
3867 
3868 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3869 	tc_skb_cb(skb)->mru = 0;
3870 	tc_skb_cb(skb)->post_ct = false;
3871 	mini_qdisc_bstats_cpu_update(miniq, skb);
3872 
3873 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3874 	case TC_ACT_OK:
3875 	case TC_ACT_RECLASSIFY:
3876 		skb->tc_index = TC_H_MIN(cl_res.classid);
3877 		break;
3878 	case TC_ACT_SHOT:
3879 		mini_qdisc_qstats_cpu_drop(miniq);
3880 		*ret = NET_XMIT_DROP;
3881 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3882 		return NULL;
3883 	case TC_ACT_STOLEN:
3884 	case TC_ACT_QUEUED:
3885 	case TC_ACT_TRAP:
3886 		*ret = NET_XMIT_SUCCESS;
3887 		consume_skb(skb);
3888 		return NULL;
3889 	case TC_ACT_REDIRECT:
3890 		/* No need to push/pop skb's mac_header here on egress! */
3891 		skb_do_redirect(skb);
3892 		*ret = NET_XMIT_SUCCESS;
3893 		return NULL;
3894 	default:
3895 		break;
3896 	}
3897 #endif /* CONFIG_NET_CLS_ACT */
3898 
3899 	return skb;
3900 }
3901 
3902 static struct netdev_queue *
3903 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3904 {
3905 	int qm = skb_get_queue_mapping(skb);
3906 
3907 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3908 }
3909 
3910 static bool netdev_xmit_txqueue_skipped(void)
3911 {
3912 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3913 }
3914 
3915 void netdev_xmit_skip_txqueue(bool skip)
3916 {
3917 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3918 }
3919 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3920 #endif /* CONFIG_NET_EGRESS */
3921 
3922 #ifdef CONFIG_XPS
3923 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3924 			       struct xps_dev_maps *dev_maps, unsigned int tci)
3925 {
3926 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
3927 	struct xps_map *map;
3928 	int queue_index = -1;
3929 
3930 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3931 		return queue_index;
3932 
3933 	tci *= dev_maps->num_tc;
3934 	tci += tc;
3935 
3936 	map = rcu_dereference(dev_maps->attr_map[tci]);
3937 	if (map) {
3938 		if (map->len == 1)
3939 			queue_index = map->queues[0];
3940 		else
3941 			queue_index = map->queues[reciprocal_scale(
3942 						skb_get_hash(skb), map->len)];
3943 		if (unlikely(queue_index >= dev->real_num_tx_queues))
3944 			queue_index = -1;
3945 	}
3946 	return queue_index;
3947 }
3948 #endif
3949 
3950 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3951 			 struct sk_buff *skb)
3952 {
3953 #ifdef CONFIG_XPS
3954 	struct xps_dev_maps *dev_maps;
3955 	struct sock *sk = skb->sk;
3956 	int queue_index = -1;
3957 
3958 	if (!static_key_false(&xps_needed))
3959 		return -1;
3960 
3961 	rcu_read_lock();
3962 	if (!static_key_false(&xps_rxqs_needed))
3963 		goto get_cpus_map;
3964 
3965 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
3966 	if (dev_maps) {
3967 		int tci = sk_rx_queue_get(sk);
3968 
3969 		if (tci >= 0)
3970 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3971 							  tci);
3972 	}
3973 
3974 get_cpus_map:
3975 	if (queue_index < 0) {
3976 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
3977 		if (dev_maps) {
3978 			unsigned int tci = skb->sender_cpu - 1;
3979 
3980 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3981 							  tci);
3982 		}
3983 	}
3984 	rcu_read_unlock();
3985 
3986 	return queue_index;
3987 #else
3988 	return -1;
3989 #endif
3990 }
3991 
3992 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3993 		     struct net_device *sb_dev)
3994 {
3995 	return 0;
3996 }
3997 EXPORT_SYMBOL(dev_pick_tx_zero);
3998 
3999 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4000 		       struct net_device *sb_dev)
4001 {
4002 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4003 }
4004 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4005 
4006 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4007 		     struct net_device *sb_dev)
4008 {
4009 	struct sock *sk = skb->sk;
4010 	int queue_index = sk_tx_queue_get(sk);
4011 
4012 	sb_dev = sb_dev ? : dev;
4013 
4014 	if (queue_index < 0 || skb->ooo_okay ||
4015 	    queue_index >= dev->real_num_tx_queues) {
4016 		int new_index = get_xps_queue(dev, sb_dev, skb);
4017 
4018 		if (new_index < 0)
4019 			new_index = skb_tx_hash(dev, sb_dev, skb);
4020 
4021 		if (queue_index != new_index && sk &&
4022 		    sk_fullsock(sk) &&
4023 		    rcu_access_pointer(sk->sk_dst_cache))
4024 			sk_tx_queue_set(sk, new_index);
4025 
4026 		queue_index = new_index;
4027 	}
4028 
4029 	return queue_index;
4030 }
4031 EXPORT_SYMBOL(netdev_pick_tx);
4032 
4033 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4034 					 struct sk_buff *skb,
4035 					 struct net_device *sb_dev)
4036 {
4037 	int queue_index = 0;
4038 
4039 #ifdef CONFIG_XPS
4040 	u32 sender_cpu = skb->sender_cpu - 1;
4041 
4042 	if (sender_cpu >= (u32)NR_CPUS)
4043 		skb->sender_cpu = raw_smp_processor_id() + 1;
4044 #endif
4045 
4046 	if (dev->real_num_tx_queues != 1) {
4047 		const struct net_device_ops *ops = dev->netdev_ops;
4048 
4049 		if (ops->ndo_select_queue)
4050 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4051 		else
4052 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4053 
4054 		queue_index = netdev_cap_txqueue(dev, queue_index);
4055 	}
4056 
4057 	skb_set_queue_mapping(skb, queue_index);
4058 	return netdev_get_tx_queue(dev, queue_index);
4059 }
4060 
4061 /**
4062  * __dev_queue_xmit() - transmit a buffer
4063  * @skb:	buffer to transmit
4064  * @sb_dev:	suboordinate device used for L2 forwarding offload
4065  *
4066  * Queue a buffer for transmission to a network device. The caller must
4067  * have set the device and priority and built the buffer before calling
4068  * this function. The function can be called from an interrupt.
4069  *
4070  * When calling this method, interrupts MUST be enabled. This is because
4071  * the BH enable code must have IRQs enabled so that it will not deadlock.
4072  *
4073  * Regardless of the return value, the skb is consumed, so it is currently
4074  * difficult to retry a send to this method. (You can bump the ref count
4075  * before sending to hold a reference for retry if you are careful.)
4076  *
4077  * Return:
4078  * * 0				- buffer successfully transmitted
4079  * * positive qdisc return code	- NET_XMIT_DROP etc.
4080  * * negative errno		- other errors
4081  */
4082 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4083 {
4084 	struct net_device *dev = skb->dev;
4085 	struct netdev_queue *txq = NULL;
4086 	struct Qdisc *q;
4087 	int rc = -ENOMEM;
4088 	bool again = false;
4089 
4090 	skb_reset_mac_header(skb);
4091 	skb_assert_len(skb);
4092 
4093 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4094 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4095 
4096 	/* Disable soft irqs for various locks below. Also
4097 	 * stops preemption for RCU.
4098 	 */
4099 	rcu_read_lock_bh();
4100 
4101 	skb_update_prio(skb);
4102 
4103 	qdisc_pkt_len_init(skb);
4104 #ifdef CONFIG_NET_CLS_ACT
4105 	skb->tc_at_ingress = 0;
4106 #endif
4107 #ifdef CONFIG_NET_EGRESS
4108 	if (static_branch_unlikely(&egress_needed_key)) {
4109 		if (nf_hook_egress_active()) {
4110 			skb = nf_hook_egress(skb, &rc, dev);
4111 			if (!skb)
4112 				goto out;
4113 		}
4114 
4115 		netdev_xmit_skip_txqueue(false);
4116 
4117 		nf_skip_egress(skb, true);
4118 		skb = sch_handle_egress(skb, &rc, dev);
4119 		if (!skb)
4120 			goto out;
4121 		nf_skip_egress(skb, false);
4122 
4123 		if (netdev_xmit_txqueue_skipped())
4124 			txq = netdev_tx_queue_mapping(dev, skb);
4125 	}
4126 #endif
4127 	/* If device/qdisc don't need skb->dst, release it right now while
4128 	 * its hot in this cpu cache.
4129 	 */
4130 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4131 		skb_dst_drop(skb);
4132 	else
4133 		skb_dst_force(skb);
4134 
4135 	if (!txq)
4136 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4137 
4138 	q = rcu_dereference_bh(txq->qdisc);
4139 
4140 	trace_net_dev_queue(skb);
4141 	if (q->enqueue) {
4142 		rc = __dev_xmit_skb(skb, q, dev, txq);
4143 		goto out;
4144 	}
4145 
4146 	/* The device has no queue. Common case for software devices:
4147 	 * loopback, all the sorts of tunnels...
4148 
4149 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4150 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4151 	 * counters.)
4152 	 * However, it is possible, that they rely on protection
4153 	 * made by us here.
4154 
4155 	 * Check this and shot the lock. It is not prone from deadlocks.
4156 	 *Either shot noqueue qdisc, it is even simpler 8)
4157 	 */
4158 	if (dev->flags & IFF_UP) {
4159 		int cpu = smp_processor_id(); /* ok because BHs are off */
4160 
4161 		/* Other cpus might concurrently change txq->xmit_lock_owner
4162 		 * to -1 or to their cpu id, but not to our id.
4163 		 */
4164 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4165 			if (dev_xmit_recursion())
4166 				goto recursion_alert;
4167 
4168 			skb = validate_xmit_skb(skb, dev, &again);
4169 			if (!skb)
4170 				goto out;
4171 
4172 			HARD_TX_LOCK(dev, txq, cpu);
4173 
4174 			if (!netif_xmit_stopped(txq)) {
4175 				dev_xmit_recursion_inc();
4176 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4177 				dev_xmit_recursion_dec();
4178 				if (dev_xmit_complete(rc)) {
4179 					HARD_TX_UNLOCK(dev, txq);
4180 					goto out;
4181 				}
4182 			}
4183 			HARD_TX_UNLOCK(dev, txq);
4184 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4185 					     dev->name);
4186 		} else {
4187 			/* Recursion is detected! It is possible,
4188 			 * unfortunately
4189 			 */
4190 recursion_alert:
4191 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4192 					     dev->name);
4193 		}
4194 	}
4195 
4196 	rc = -ENETDOWN;
4197 	rcu_read_unlock_bh();
4198 
4199 	dev_core_stats_tx_dropped_inc(dev);
4200 	kfree_skb_list(skb);
4201 	return rc;
4202 out:
4203 	rcu_read_unlock_bh();
4204 	return rc;
4205 }
4206 EXPORT_SYMBOL(__dev_queue_xmit);
4207 
4208 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4209 {
4210 	struct net_device *dev = skb->dev;
4211 	struct sk_buff *orig_skb = skb;
4212 	struct netdev_queue *txq;
4213 	int ret = NETDEV_TX_BUSY;
4214 	bool again = false;
4215 
4216 	if (unlikely(!netif_running(dev) ||
4217 		     !netif_carrier_ok(dev)))
4218 		goto drop;
4219 
4220 	skb = validate_xmit_skb_list(skb, dev, &again);
4221 	if (skb != orig_skb)
4222 		goto drop;
4223 
4224 	skb_set_queue_mapping(skb, queue_id);
4225 	txq = skb_get_tx_queue(dev, skb);
4226 
4227 	local_bh_disable();
4228 
4229 	dev_xmit_recursion_inc();
4230 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4231 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4232 		ret = netdev_start_xmit(skb, dev, txq, false);
4233 	HARD_TX_UNLOCK(dev, txq);
4234 	dev_xmit_recursion_dec();
4235 
4236 	local_bh_enable();
4237 	return ret;
4238 drop:
4239 	dev_core_stats_tx_dropped_inc(dev);
4240 	kfree_skb_list(skb);
4241 	return NET_XMIT_DROP;
4242 }
4243 EXPORT_SYMBOL(__dev_direct_xmit);
4244 
4245 /*************************************************************************
4246  *			Receiver routines
4247  *************************************************************************/
4248 
4249 int netdev_max_backlog __read_mostly = 1000;
4250 EXPORT_SYMBOL(netdev_max_backlog);
4251 
4252 int netdev_tstamp_prequeue __read_mostly = 1;
4253 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4254 int netdev_budget __read_mostly = 300;
4255 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4256 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4257 int weight_p __read_mostly = 64;           /* old backlog weight */
4258 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4259 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4260 int dev_rx_weight __read_mostly = 64;
4261 int dev_tx_weight __read_mostly = 64;
4262 
4263 /* Called with irq disabled */
4264 static inline void ____napi_schedule(struct softnet_data *sd,
4265 				     struct napi_struct *napi)
4266 {
4267 	struct task_struct *thread;
4268 
4269 	lockdep_assert_irqs_disabled();
4270 
4271 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4272 		/* Paired with smp_mb__before_atomic() in
4273 		 * napi_enable()/dev_set_threaded().
4274 		 * Use READ_ONCE() to guarantee a complete
4275 		 * read on napi->thread. Only call
4276 		 * wake_up_process() when it's not NULL.
4277 		 */
4278 		thread = READ_ONCE(napi->thread);
4279 		if (thread) {
4280 			/* Avoid doing set_bit() if the thread is in
4281 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4282 			 * makes sure to proceed with napi polling
4283 			 * if the thread is explicitly woken from here.
4284 			 */
4285 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4286 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4287 			wake_up_process(thread);
4288 			return;
4289 		}
4290 	}
4291 
4292 	list_add_tail(&napi->poll_list, &sd->poll_list);
4293 	WRITE_ONCE(napi->list_owner, smp_processor_id());
4294 	/* If not called from net_rx_action()
4295 	 * we have to raise NET_RX_SOFTIRQ.
4296 	 */
4297 	if (!sd->in_net_rx_action)
4298 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4299 }
4300 
4301 #ifdef CONFIG_RPS
4302 
4303 /* One global table that all flow-based protocols share. */
4304 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4305 EXPORT_SYMBOL(rps_sock_flow_table);
4306 u32 rps_cpu_mask __read_mostly;
4307 EXPORT_SYMBOL(rps_cpu_mask);
4308 
4309 struct static_key_false rps_needed __read_mostly;
4310 EXPORT_SYMBOL(rps_needed);
4311 struct static_key_false rfs_needed __read_mostly;
4312 EXPORT_SYMBOL(rfs_needed);
4313 
4314 static struct rps_dev_flow *
4315 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4316 	    struct rps_dev_flow *rflow, u16 next_cpu)
4317 {
4318 	if (next_cpu < nr_cpu_ids) {
4319 #ifdef CONFIG_RFS_ACCEL
4320 		struct netdev_rx_queue *rxqueue;
4321 		struct rps_dev_flow_table *flow_table;
4322 		struct rps_dev_flow *old_rflow;
4323 		u32 flow_id;
4324 		u16 rxq_index;
4325 		int rc;
4326 
4327 		/* Should we steer this flow to a different hardware queue? */
4328 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4329 		    !(dev->features & NETIF_F_NTUPLE))
4330 			goto out;
4331 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4332 		if (rxq_index == skb_get_rx_queue(skb))
4333 			goto out;
4334 
4335 		rxqueue = dev->_rx + rxq_index;
4336 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4337 		if (!flow_table)
4338 			goto out;
4339 		flow_id = skb_get_hash(skb) & flow_table->mask;
4340 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4341 							rxq_index, flow_id);
4342 		if (rc < 0)
4343 			goto out;
4344 		old_rflow = rflow;
4345 		rflow = &flow_table->flows[flow_id];
4346 		rflow->filter = rc;
4347 		if (old_rflow->filter == rflow->filter)
4348 			old_rflow->filter = RPS_NO_FILTER;
4349 	out:
4350 #endif
4351 		rflow->last_qtail =
4352 			per_cpu(softnet_data, next_cpu).input_queue_head;
4353 	}
4354 
4355 	rflow->cpu = next_cpu;
4356 	return rflow;
4357 }
4358 
4359 /*
4360  * get_rps_cpu is called from netif_receive_skb and returns the target
4361  * CPU from the RPS map of the receiving queue for a given skb.
4362  * rcu_read_lock must be held on entry.
4363  */
4364 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4365 		       struct rps_dev_flow **rflowp)
4366 {
4367 	const struct rps_sock_flow_table *sock_flow_table;
4368 	struct netdev_rx_queue *rxqueue = dev->_rx;
4369 	struct rps_dev_flow_table *flow_table;
4370 	struct rps_map *map;
4371 	int cpu = -1;
4372 	u32 tcpu;
4373 	u32 hash;
4374 
4375 	if (skb_rx_queue_recorded(skb)) {
4376 		u16 index = skb_get_rx_queue(skb);
4377 
4378 		if (unlikely(index >= dev->real_num_rx_queues)) {
4379 			WARN_ONCE(dev->real_num_rx_queues > 1,
4380 				  "%s received packet on queue %u, but number "
4381 				  "of RX queues is %u\n",
4382 				  dev->name, index, dev->real_num_rx_queues);
4383 			goto done;
4384 		}
4385 		rxqueue += index;
4386 	}
4387 
4388 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4389 
4390 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4391 	map = rcu_dereference(rxqueue->rps_map);
4392 	if (!flow_table && !map)
4393 		goto done;
4394 
4395 	skb_reset_network_header(skb);
4396 	hash = skb_get_hash(skb);
4397 	if (!hash)
4398 		goto done;
4399 
4400 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4401 	if (flow_table && sock_flow_table) {
4402 		struct rps_dev_flow *rflow;
4403 		u32 next_cpu;
4404 		u32 ident;
4405 
4406 		/* First check into global flow table if there is a match.
4407 		 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4408 		 */
4409 		ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4410 		if ((ident ^ hash) & ~rps_cpu_mask)
4411 			goto try_rps;
4412 
4413 		next_cpu = ident & rps_cpu_mask;
4414 
4415 		/* OK, now we know there is a match,
4416 		 * we can look at the local (per receive queue) flow table
4417 		 */
4418 		rflow = &flow_table->flows[hash & flow_table->mask];
4419 		tcpu = rflow->cpu;
4420 
4421 		/*
4422 		 * If the desired CPU (where last recvmsg was done) is
4423 		 * different from current CPU (one in the rx-queue flow
4424 		 * table entry), switch if one of the following holds:
4425 		 *   - Current CPU is unset (>= nr_cpu_ids).
4426 		 *   - Current CPU is offline.
4427 		 *   - The current CPU's queue tail has advanced beyond the
4428 		 *     last packet that was enqueued using this table entry.
4429 		 *     This guarantees that all previous packets for the flow
4430 		 *     have been dequeued, thus preserving in order delivery.
4431 		 */
4432 		if (unlikely(tcpu != next_cpu) &&
4433 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4434 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4435 		      rflow->last_qtail)) >= 0)) {
4436 			tcpu = next_cpu;
4437 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4438 		}
4439 
4440 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4441 			*rflowp = rflow;
4442 			cpu = tcpu;
4443 			goto done;
4444 		}
4445 	}
4446 
4447 try_rps:
4448 
4449 	if (map) {
4450 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4451 		if (cpu_online(tcpu)) {
4452 			cpu = tcpu;
4453 			goto done;
4454 		}
4455 	}
4456 
4457 done:
4458 	return cpu;
4459 }
4460 
4461 #ifdef CONFIG_RFS_ACCEL
4462 
4463 /**
4464  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4465  * @dev: Device on which the filter was set
4466  * @rxq_index: RX queue index
4467  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4468  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4469  *
4470  * Drivers that implement ndo_rx_flow_steer() should periodically call
4471  * this function for each installed filter and remove the filters for
4472  * which it returns %true.
4473  */
4474 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4475 			 u32 flow_id, u16 filter_id)
4476 {
4477 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4478 	struct rps_dev_flow_table *flow_table;
4479 	struct rps_dev_flow *rflow;
4480 	bool expire = true;
4481 	unsigned int cpu;
4482 
4483 	rcu_read_lock();
4484 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4485 	if (flow_table && flow_id <= flow_table->mask) {
4486 		rflow = &flow_table->flows[flow_id];
4487 		cpu = READ_ONCE(rflow->cpu);
4488 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4489 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4490 			   rflow->last_qtail) <
4491 		     (int)(10 * flow_table->mask)))
4492 			expire = false;
4493 	}
4494 	rcu_read_unlock();
4495 	return expire;
4496 }
4497 EXPORT_SYMBOL(rps_may_expire_flow);
4498 
4499 #endif /* CONFIG_RFS_ACCEL */
4500 
4501 /* Called from hardirq (IPI) context */
4502 static void rps_trigger_softirq(void *data)
4503 {
4504 	struct softnet_data *sd = data;
4505 
4506 	____napi_schedule(sd, &sd->backlog);
4507 	sd->received_rps++;
4508 }
4509 
4510 #endif /* CONFIG_RPS */
4511 
4512 /* Called from hardirq (IPI) context */
4513 static void trigger_rx_softirq(void *data)
4514 {
4515 	struct softnet_data *sd = data;
4516 
4517 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4518 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4519 }
4520 
4521 /*
4522  * After we queued a packet into sd->input_pkt_queue,
4523  * we need to make sure this queue is serviced soon.
4524  *
4525  * - If this is another cpu queue, link it to our rps_ipi_list,
4526  *   and make sure we will process rps_ipi_list from net_rx_action().
4527  *
4528  * - If this is our own queue, NAPI schedule our backlog.
4529  *   Note that this also raises NET_RX_SOFTIRQ.
4530  */
4531 static void napi_schedule_rps(struct softnet_data *sd)
4532 {
4533 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4534 
4535 #ifdef CONFIG_RPS
4536 	if (sd != mysd) {
4537 		sd->rps_ipi_next = mysd->rps_ipi_list;
4538 		mysd->rps_ipi_list = sd;
4539 
4540 		/* If not called from net_rx_action() or napi_threaded_poll()
4541 		 * we have to raise NET_RX_SOFTIRQ.
4542 		 */
4543 		if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4544 			__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4545 		return;
4546 	}
4547 #endif /* CONFIG_RPS */
4548 	__napi_schedule_irqoff(&mysd->backlog);
4549 }
4550 
4551 #ifdef CONFIG_NET_FLOW_LIMIT
4552 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4553 #endif
4554 
4555 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4556 {
4557 #ifdef CONFIG_NET_FLOW_LIMIT
4558 	struct sd_flow_limit *fl;
4559 	struct softnet_data *sd;
4560 	unsigned int old_flow, new_flow;
4561 
4562 	if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4563 		return false;
4564 
4565 	sd = this_cpu_ptr(&softnet_data);
4566 
4567 	rcu_read_lock();
4568 	fl = rcu_dereference(sd->flow_limit);
4569 	if (fl) {
4570 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4571 		old_flow = fl->history[fl->history_head];
4572 		fl->history[fl->history_head] = new_flow;
4573 
4574 		fl->history_head++;
4575 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4576 
4577 		if (likely(fl->buckets[old_flow]))
4578 			fl->buckets[old_flow]--;
4579 
4580 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4581 			fl->count++;
4582 			rcu_read_unlock();
4583 			return true;
4584 		}
4585 	}
4586 	rcu_read_unlock();
4587 #endif
4588 	return false;
4589 }
4590 
4591 /*
4592  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4593  * queue (may be a remote CPU queue).
4594  */
4595 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4596 			      unsigned int *qtail)
4597 {
4598 	enum skb_drop_reason reason;
4599 	struct softnet_data *sd;
4600 	unsigned long flags;
4601 	unsigned int qlen;
4602 
4603 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
4604 	sd = &per_cpu(softnet_data, cpu);
4605 
4606 	rps_lock_irqsave(sd, &flags);
4607 	if (!netif_running(skb->dev))
4608 		goto drop;
4609 	qlen = skb_queue_len(&sd->input_pkt_queue);
4610 	if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4611 		if (qlen) {
4612 enqueue:
4613 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4614 			input_queue_tail_incr_save(sd, qtail);
4615 			rps_unlock_irq_restore(sd, &flags);
4616 			return NET_RX_SUCCESS;
4617 		}
4618 
4619 		/* Schedule NAPI for backlog device
4620 		 * We can use non atomic operation since we own the queue lock
4621 		 */
4622 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4623 			napi_schedule_rps(sd);
4624 		goto enqueue;
4625 	}
4626 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4627 
4628 drop:
4629 	sd->dropped++;
4630 	rps_unlock_irq_restore(sd, &flags);
4631 
4632 	dev_core_stats_rx_dropped_inc(skb->dev);
4633 	kfree_skb_reason(skb, reason);
4634 	return NET_RX_DROP;
4635 }
4636 
4637 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4638 {
4639 	struct net_device *dev = skb->dev;
4640 	struct netdev_rx_queue *rxqueue;
4641 
4642 	rxqueue = dev->_rx;
4643 
4644 	if (skb_rx_queue_recorded(skb)) {
4645 		u16 index = skb_get_rx_queue(skb);
4646 
4647 		if (unlikely(index >= dev->real_num_rx_queues)) {
4648 			WARN_ONCE(dev->real_num_rx_queues > 1,
4649 				  "%s received packet on queue %u, but number "
4650 				  "of RX queues is %u\n",
4651 				  dev->name, index, dev->real_num_rx_queues);
4652 
4653 			return rxqueue; /* Return first rxqueue */
4654 		}
4655 		rxqueue += index;
4656 	}
4657 	return rxqueue;
4658 }
4659 
4660 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4661 			     struct bpf_prog *xdp_prog)
4662 {
4663 	void *orig_data, *orig_data_end, *hard_start;
4664 	struct netdev_rx_queue *rxqueue;
4665 	bool orig_bcast, orig_host;
4666 	u32 mac_len, frame_sz;
4667 	__be16 orig_eth_type;
4668 	struct ethhdr *eth;
4669 	u32 metalen, act;
4670 	int off;
4671 
4672 	/* The XDP program wants to see the packet starting at the MAC
4673 	 * header.
4674 	 */
4675 	mac_len = skb->data - skb_mac_header(skb);
4676 	hard_start = skb->data - skb_headroom(skb);
4677 
4678 	/* SKB "head" area always have tailroom for skb_shared_info */
4679 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4680 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4681 
4682 	rxqueue = netif_get_rxqueue(skb);
4683 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4684 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4685 			 skb_headlen(skb) + mac_len, true);
4686 
4687 	orig_data_end = xdp->data_end;
4688 	orig_data = xdp->data;
4689 	eth = (struct ethhdr *)xdp->data;
4690 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4691 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4692 	orig_eth_type = eth->h_proto;
4693 
4694 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4695 
4696 	/* check if bpf_xdp_adjust_head was used */
4697 	off = xdp->data - orig_data;
4698 	if (off) {
4699 		if (off > 0)
4700 			__skb_pull(skb, off);
4701 		else if (off < 0)
4702 			__skb_push(skb, -off);
4703 
4704 		skb->mac_header += off;
4705 		skb_reset_network_header(skb);
4706 	}
4707 
4708 	/* check if bpf_xdp_adjust_tail was used */
4709 	off = xdp->data_end - orig_data_end;
4710 	if (off != 0) {
4711 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4712 		skb->len += off; /* positive on grow, negative on shrink */
4713 	}
4714 
4715 	/* check if XDP changed eth hdr such SKB needs update */
4716 	eth = (struct ethhdr *)xdp->data;
4717 	if ((orig_eth_type != eth->h_proto) ||
4718 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4719 						  skb->dev->dev_addr)) ||
4720 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4721 		__skb_push(skb, ETH_HLEN);
4722 		skb->pkt_type = PACKET_HOST;
4723 		skb->protocol = eth_type_trans(skb, skb->dev);
4724 	}
4725 
4726 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4727 	 * before calling us again on redirect path. We do not call do_redirect
4728 	 * as we leave that up to the caller.
4729 	 *
4730 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4731 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4732 	 */
4733 	switch (act) {
4734 	case XDP_REDIRECT:
4735 	case XDP_TX:
4736 		__skb_push(skb, mac_len);
4737 		break;
4738 	case XDP_PASS:
4739 		metalen = xdp->data - xdp->data_meta;
4740 		if (metalen)
4741 			skb_metadata_set(skb, metalen);
4742 		break;
4743 	}
4744 
4745 	return act;
4746 }
4747 
4748 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4749 				     struct xdp_buff *xdp,
4750 				     struct bpf_prog *xdp_prog)
4751 {
4752 	u32 act = XDP_DROP;
4753 
4754 	/* Reinjected packets coming from act_mirred or similar should
4755 	 * not get XDP generic processing.
4756 	 */
4757 	if (skb_is_redirected(skb))
4758 		return XDP_PASS;
4759 
4760 	/* XDP packets must be linear and must have sufficient headroom
4761 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4762 	 * native XDP provides, thus we need to do it here as well.
4763 	 */
4764 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4765 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4766 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4767 		int troom = skb->tail + skb->data_len - skb->end;
4768 
4769 		/* In case we have to go down the path and also linearize,
4770 		 * then lets do the pskb_expand_head() work just once here.
4771 		 */
4772 		if (pskb_expand_head(skb,
4773 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4774 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4775 			goto do_drop;
4776 		if (skb_linearize(skb))
4777 			goto do_drop;
4778 	}
4779 
4780 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4781 	switch (act) {
4782 	case XDP_REDIRECT:
4783 	case XDP_TX:
4784 	case XDP_PASS:
4785 		break;
4786 	default:
4787 		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4788 		fallthrough;
4789 	case XDP_ABORTED:
4790 		trace_xdp_exception(skb->dev, xdp_prog, act);
4791 		fallthrough;
4792 	case XDP_DROP:
4793 	do_drop:
4794 		kfree_skb(skb);
4795 		break;
4796 	}
4797 
4798 	return act;
4799 }
4800 
4801 /* When doing generic XDP we have to bypass the qdisc layer and the
4802  * network taps in order to match in-driver-XDP behavior. This also means
4803  * that XDP packets are able to starve other packets going through a qdisc,
4804  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4805  * queues, so they do not have this starvation issue.
4806  */
4807 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4808 {
4809 	struct net_device *dev = skb->dev;
4810 	struct netdev_queue *txq;
4811 	bool free_skb = true;
4812 	int cpu, rc;
4813 
4814 	txq = netdev_core_pick_tx(dev, skb, NULL);
4815 	cpu = smp_processor_id();
4816 	HARD_TX_LOCK(dev, txq, cpu);
4817 	if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4818 		rc = netdev_start_xmit(skb, dev, txq, 0);
4819 		if (dev_xmit_complete(rc))
4820 			free_skb = false;
4821 	}
4822 	HARD_TX_UNLOCK(dev, txq);
4823 	if (free_skb) {
4824 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4825 		dev_core_stats_tx_dropped_inc(dev);
4826 		kfree_skb(skb);
4827 	}
4828 }
4829 
4830 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4831 
4832 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4833 {
4834 	if (xdp_prog) {
4835 		struct xdp_buff xdp;
4836 		u32 act;
4837 		int err;
4838 
4839 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4840 		if (act != XDP_PASS) {
4841 			switch (act) {
4842 			case XDP_REDIRECT:
4843 				err = xdp_do_generic_redirect(skb->dev, skb,
4844 							      &xdp, xdp_prog);
4845 				if (err)
4846 					goto out_redir;
4847 				break;
4848 			case XDP_TX:
4849 				generic_xdp_tx(skb, xdp_prog);
4850 				break;
4851 			}
4852 			return XDP_DROP;
4853 		}
4854 	}
4855 	return XDP_PASS;
4856 out_redir:
4857 	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4858 	return XDP_DROP;
4859 }
4860 EXPORT_SYMBOL_GPL(do_xdp_generic);
4861 
4862 static int netif_rx_internal(struct sk_buff *skb)
4863 {
4864 	int ret;
4865 
4866 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4867 
4868 	trace_netif_rx(skb);
4869 
4870 #ifdef CONFIG_RPS
4871 	if (static_branch_unlikely(&rps_needed)) {
4872 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4873 		int cpu;
4874 
4875 		rcu_read_lock();
4876 
4877 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4878 		if (cpu < 0)
4879 			cpu = smp_processor_id();
4880 
4881 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4882 
4883 		rcu_read_unlock();
4884 	} else
4885 #endif
4886 	{
4887 		unsigned int qtail;
4888 
4889 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4890 	}
4891 	return ret;
4892 }
4893 
4894 /**
4895  *	__netif_rx	-	Slightly optimized version of netif_rx
4896  *	@skb: buffer to post
4897  *
4898  *	This behaves as netif_rx except that it does not disable bottom halves.
4899  *	As a result this function may only be invoked from the interrupt context
4900  *	(either hard or soft interrupt).
4901  */
4902 int __netif_rx(struct sk_buff *skb)
4903 {
4904 	int ret;
4905 
4906 	lockdep_assert_once(hardirq_count() | softirq_count());
4907 
4908 	trace_netif_rx_entry(skb);
4909 	ret = netif_rx_internal(skb);
4910 	trace_netif_rx_exit(ret);
4911 	return ret;
4912 }
4913 EXPORT_SYMBOL(__netif_rx);
4914 
4915 /**
4916  *	netif_rx	-	post buffer to the network code
4917  *	@skb: buffer to post
4918  *
4919  *	This function receives a packet from a device driver and queues it for
4920  *	the upper (protocol) levels to process via the backlog NAPI device. It
4921  *	always succeeds. The buffer may be dropped during processing for
4922  *	congestion control or by the protocol layers.
4923  *	The network buffer is passed via the backlog NAPI device. Modern NIC
4924  *	driver should use NAPI and GRO.
4925  *	This function can used from interrupt and from process context. The
4926  *	caller from process context must not disable interrupts before invoking
4927  *	this function.
4928  *
4929  *	return values:
4930  *	NET_RX_SUCCESS	(no congestion)
4931  *	NET_RX_DROP     (packet was dropped)
4932  *
4933  */
4934 int netif_rx(struct sk_buff *skb)
4935 {
4936 	bool need_bh_off = !(hardirq_count() | softirq_count());
4937 	int ret;
4938 
4939 	if (need_bh_off)
4940 		local_bh_disable();
4941 	trace_netif_rx_entry(skb);
4942 	ret = netif_rx_internal(skb);
4943 	trace_netif_rx_exit(ret);
4944 	if (need_bh_off)
4945 		local_bh_enable();
4946 	return ret;
4947 }
4948 EXPORT_SYMBOL(netif_rx);
4949 
4950 static __latent_entropy void net_tx_action(struct softirq_action *h)
4951 {
4952 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4953 
4954 	if (sd->completion_queue) {
4955 		struct sk_buff *clist;
4956 
4957 		local_irq_disable();
4958 		clist = sd->completion_queue;
4959 		sd->completion_queue = NULL;
4960 		local_irq_enable();
4961 
4962 		while (clist) {
4963 			struct sk_buff *skb = clist;
4964 
4965 			clist = clist->next;
4966 
4967 			WARN_ON(refcount_read(&skb->users));
4968 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
4969 				trace_consume_skb(skb, net_tx_action);
4970 			else
4971 				trace_kfree_skb(skb, net_tx_action,
4972 						get_kfree_skb_cb(skb)->reason);
4973 
4974 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4975 				__kfree_skb(skb);
4976 			else
4977 				__napi_kfree_skb(skb,
4978 						 get_kfree_skb_cb(skb)->reason);
4979 		}
4980 	}
4981 
4982 	if (sd->output_queue) {
4983 		struct Qdisc *head;
4984 
4985 		local_irq_disable();
4986 		head = sd->output_queue;
4987 		sd->output_queue = NULL;
4988 		sd->output_queue_tailp = &sd->output_queue;
4989 		local_irq_enable();
4990 
4991 		rcu_read_lock();
4992 
4993 		while (head) {
4994 			struct Qdisc *q = head;
4995 			spinlock_t *root_lock = NULL;
4996 
4997 			head = head->next_sched;
4998 
4999 			/* We need to make sure head->next_sched is read
5000 			 * before clearing __QDISC_STATE_SCHED
5001 			 */
5002 			smp_mb__before_atomic();
5003 
5004 			if (!(q->flags & TCQ_F_NOLOCK)) {
5005 				root_lock = qdisc_lock(q);
5006 				spin_lock(root_lock);
5007 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5008 						     &q->state))) {
5009 				/* There is a synchronize_net() between
5010 				 * STATE_DEACTIVATED flag being set and
5011 				 * qdisc_reset()/some_qdisc_is_busy() in
5012 				 * dev_deactivate(), so we can safely bail out
5013 				 * early here to avoid data race between
5014 				 * qdisc_deactivate() and some_qdisc_is_busy()
5015 				 * for lockless qdisc.
5016 				 */
5017 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5018 				continue;
5019 			}
5020 
5021 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5022 			qdisc_run(q);
5023 			if (root_lock)
5024 				spin_unlock(root_lock);
5025 		}
5026 
5027 		rcu_read_unlock();
5028 	}
5029 
5030 	xfrm_dev_backlog(sd);
5031 }
5032 
5033 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5034 /* This hook is defined here for ATM LANE */
5035 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5036 			     unsigned char *addr) __read_mostly;
5037 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5038 #endif
5039 
5040 static inline struct sk_buff *
5041 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5042 		   struct net_device *orig_dev, bool *another)
5043 {
5044 #ifdef CONFIG_NET_CLS_ACT
5045 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5046 	struct tcf_result cl_res;
5047 
5048 	/* If there's at least one ingress present somewhere (so
5049 	 * we get here via enabled static key), remaining devices
5050 	 * that are not configured with an ingress qdisc will bail
5051 	 * out here.
5052 	 */
5053 	if (!miniq)
5054 		return skb;
5055 
5056 	if (*pt_prev) {
5057 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5058 		*pt_prev = NULL;
5059 	}
5060 
5061 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5062 	tc_skb_cb(skb)->mru = 0;
5063 	tc_skb_cb(skb)->post_ct = false;
5064 	skb->tc_at_ingress = 1;
5065 	mini_qdisc_bstats_cpu_update(miniq, skb);
5066 
5067 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5068 	case TC_ACT_OK:
5069 	case TC_ACT_RECLASSIFY:
5070 		skb->tc_index = TC_H_MIN(cl_res.classid);
5071 		break;
5072 	case TC_ACT_SHOT:
5073 		mini_qdisc_qstats_cpu_drop(miniq);
5074 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5075 		*ret = NET_RX_DROP;
5076 		return NULL;
5077 	case TC_ACT_STOLEN:
5078 	case TC_ACT_QUEUED:
5079 	case TC_ACT_TRAP:
5080 		consume_skb(skb);
5081 		*ret = NET_RX_SUCCESS;
5082 		return NULL;
5083 	case TC_ACT_REDIRECT:
5084 		/* skb_mac_header check was done by cls/act_bpf, so
5085 		 * we can safely push the L2 header back before
5086 		 * redirecting to another netdev
5087 		 */
5088 		__skb_push(skb, skb->mac_len);
5089 		if (skb_do_redirect(skb) == -EAGAIN) {
5090 			__skb_pull(skb, skb->mac_len);
5091 			*another = true;
5092 			break;
5093 		}
5094 		*ret = NET_RX_SUCCESS;
5095 		return NULL;
5096 	case TC_ACT_CONSUMED:
5097 		*ret = NET_RX_SUCCESS;
5098 		return NULL;
5099 	default:
5100 		break;
5101 	}
5102 #endif /* CONFIG_NET_CLS_ACT */
5103 	return skb;
5104 }
5105 
5106 /**
5107  *	netdev_is_rx_handler_busy - check if receive handler is registered
5108  *	@dev: device to check
5109  *
5110  *	Check if a receive handler is already registered for a given device.
5111  *	Return true if there one.
5112  *
5113  *	The caller must hold the rtnl_mutex.
5114  */
5115 bool netdev_is_rx_handler_busy(struct net_device *dev)
5116 {
5117 	ASSERT_RTNL();
5118 	return dev && rtnl_dereference(dev->rx_handler);
5119 }
5120 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5121 
5122 /**
5123  *	netdev_rx_handler_register - register receive handler
5124  *	@dev: device to register a handler for
5125  *	@rx_handler: receive handler to register
5126  *	@rx_handler_data: data pointer that is used by rx handler
5127  *
5128  *	Register a receive handler for a device. This handler will then be
5129  *	called from __netif_receive_skb. A negative errno code is returned
5130  *	on a failure.
5131  *
5132  *	The caller must hold the rtnl_mutex.
5133  *
5134  *	For a general description of rx_handler, see enum rx_handler_result.
5135  */
5136 int netdev_rx_handler_register(struct net_device *dev,
5137 			       rx_handler_func_t *rx_handler,
5138 			       void *rx_handler_data)
5139 {
5140 	if (netdev_is_rx_handler_busy(dev))
5141 		return -EBUSY;
5142 
5143 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5144 		return -EINVAL;
5145 
5146 	/* Note: rx_handler_data must be set before rx_handler */
5147 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5148 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5149 
5150 	return 0;
5151 }
5152 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5153 
5154 /**
5155  *	netdev_rx_handler_unregister - unregister receive handler
5156  *	@dev: device to unregister a handler from
5157  *
5158  *	Unregister a receive handler from a device.
5159  *
5160  *	The caller must hold the rtnl_mutex.
5161  */
5162 void netdev_rx_handler_unregister(struct net_device *dev)
5163 {
5164 
5165 	ASSERT_RTNL();
5166 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5167 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5168 	 * section has a guarantee to see a non NULL rx_handler_data
5169 	 * as well.
5170 	 */
5171 	synchronize_net();
5172 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5173 }
5174 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5175 
5176 /*
5177  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5178  * the special handling of PFMEMALLOC skbs.
5179  */
5180 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5181 {
5182 	switch (skb->protocol) {
5183 	case htons(ETH_P_ARP):
5184 	case htons(ETH_P_IP):
5185 	case htons(ETH_P_IPV6):
5186 	case htons(ETH_P_8021Q):
5187 	case htons(ETH_P_8021AD):
5188 		return true;
5189 	default:
5190 		return false;
5191 	}
5192 }
5193 
5194 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5195 			     int *ret, struct net_device *orig_dev)
5196 {
5197 	if (nf_hook_ingress_active(skb)) {
5198 		int ingress_retval;
5199 
5200 		if (*pt_prev) {
5201 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5202 			*pt_prev = NULL;
5203 		}
5204 
5205 		rcu_read_lock();
5206 		ingress_retval = nf_hook_ingress(skb);
5207 		rcu_read_unlock();
5208 		return ingress_retval;
5209 	}
5210 	return 0;
5211 }
5212 
5213 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5214 				    struct packet_type **ppt_prev)
5215 {
5216 	struct packet_type *ptype, *pt_prev;
5217 	rx_handler_func_t *rx_handler;
5218 	struct sk_buff *skb = *pskb;
5219 	struct net_device *orig_dev;
5220 	bool deliver_exact = false;
5221 	int ret = NET_RX_DROP;
5222 	__be16 type;
5223 
5224 	net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5225 
5226 	trace_netif_receive_skb(skb);
5227 
5228 	orig_dev = skb->dev;
5229 
5230 	skb_reset_network_header(skb);
5231 	if (!skb_transport_header_was_set(skb))
5232 		skb_reset_transport_header(skb);
5233 	skb_reset_mac_len(skb);
5234 
5235 	pt_prev = NULL;
5236 
5237 another_round:
5238 	skb->skb_iif = skb->dev->ifindex;
5239 
5240 	__this_cpu_inc(softnet_data.processed);
5241 
5242 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5243 		int ret2;
5244 
5245 		migrate_disable();
5246 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5247 		migrate_enable();
5248 
5249 		if (ret2 != XDP_PASS) {
5250 			ret = NET_RX_DROP;
5251 			goto out;
5252 		}
5253 	}
5254 
5255 	if (eth_type_vlan(skb->protocol)) {
5256 		skb = skb_vlan_untag(skb);
5257 		if (unlikely(!skb))
5258 			goto out;
5259 	}
5260 
5261 	if (skb_skip_tc_classify(skb))
5262 		goto skip_classify;
5263 
5264 	if (pfmemalloc)
5265 		goto skip_taps;
5266 
5267 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5268 		if (pt_prev)
5269 			ret = deliver_skb(skb, pt_prev, orig_dev);
5270 		pt_prev = ptype;
5271 	}
5272 
5273 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5274 		if (pt_prev)
5275 			ret = deliver_skb(skb, pt_prev, orig_dev);
5276 		pt_prev = ptype;
5277 	}
5278 
5279 skip_taps:
5280 #ifdef CONFIG_NET_INGRESS
5281 	if (static_branch_unlikely(&ingress_needed_key)) {
5282 		bool another = false;
5283 
5284 		nf_skip_egress(skb, true);
5285 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5286 					 &another);
5287 		if (another)
5288 			goto another_round;
5289 		if (!skb)
5290 			goto out;
5291 
5292 		nf_skip_egress(skb, false);
5293 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5294 			goto out;
5295 	}
5296 #endif
5297 	skb_reset_redirect(skb);
5298 skip_classify:
5299 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5300 		goto drop;
5301 
5302 	if (skb_vlan_tag_present(skb)) {
5303 		if (pt_prev) {
5304 			ret = deliver_skb(skb, pt_prev, orig_dev);
5305 			pt_prev = NULL;
5306 		}
5307 		if (vlan_do_receive(&skb))
5308 			goto another_round;
5309 		else if (unlikely(!skb))
5310 			goto out;
5311 	}
5312 
5313 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5314 	if (rx_handler) {
5315 		if (pt_prev) {
5316 			ret = deliver_skb(skb, pt_prev, orig_dev);
5317 			pt_prev = NULL;
5318 		}
5319 		switch (rx_handler(&skb)) {
5320 		case RX_HANDLER_CONSUMED:
5321 			ret = NET_RX_SUCCESS;
5322 			goto out;
5323 		case RX_HANDLER_ANOTHER:
5324 			goto another_round;
5325 		case RX_HANDLER_EXACT:
5326 			deliver_exact = true;
5327 			break;
5328 		case RX_HANDLER_PASS:
5329 			break;
5330 		default:
5331 			BUG();
5332 		}
5333 	}
5334 
5335 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5336 check_vlan_id:
5337 		if (skb_vlan_tag_get_id(skb)) {
5338 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5339 			 * find vlan device.
5340 			 */
5341 			skb->pkt_type = PACKET_OTHERHOST;
5342 		} else if (eth_type_vlan(skb->protocol)) {
5343 			/* Outer header is 802.1P with vlan 0, inner header is
5344 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5345 			 * not find vlan dev for vlan id 0.
5346 			 */
5347 			__vlan_hwaccel_clear_tag(skb);
5348 			skb = skb_vlan_untag(skb);
5349 			if (unlikely(!skb))
5350 				goto out;
5351 			if (vlan_do_receive(&skb))
5352 				/* After stripping off 802.1P header with vlan 0
5353 				 * vlan dev is found for inner header.
5354 				 */
5355 				goto another_round;
5356 			else if (unlikely(!skb))
5357 				goto out;
5358 			else
5359 				/* We have stripped outer 802.1P vlan 0 header.
5360 				 * But could not find vlan dev.
5361 				 * check again for vlan id to set OTHERHOST.
5362 				 */
5363 				goto check_vlan_id;
5364 		}
5365 		/* Note: we might in the future use prio bits
5366 		 * and set skb->priority like in vlan_do_receive()
5367 		 * For the time being, just ignore Priority Code Point
5368 		 */
5369 		__vlan_hwaccel_clear_tag(skb);
5370 	}
5371 
5372 	type = skb->protocol;
5373 
5374 	/* deliver only exact match when indicated */
5375 	if (likely(!deliver_exact)) {
5376 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5377 				       &ptype_base[ntohs(type) &
5378 						   PTYPE_HASH_MASK]);
5379 	}
5380 
5381 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5382 			       &orig_dev->ptype_specific);
5383 
5384 	if (unlikely(skb->dev != orig_dev)) {
5385 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5386 				       &skb->dev->ptype_specific);
5387 	}
5388 
5389 	if (pt_prev) {
5390 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5391 			goto drop;
5392 		*ppt_prev = pt_prev;
5393 	} else {
5394 drop:
5395 		if (!deliver_exact)
5396 			dev_core_stats_rx_dropped_inc(skb->dev);
5397 		else
5398 			dev_core_stats_rx_nohandler_inc(skb->dev);
5399 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5400 		/* Jamal, now you will not able to escape explaining
5401 		 * me how you were going to use this. :-)
5402 		 */
5403 		ret = NET_RX_DROP;
5404 	}
5405 
5406 out:
5407 	/* The invariant here is that if *ppt_prev is not NULL
5408 	 * then skb should also be non-NULL.
5409 	 *
5410 	 * Apparently *ppt_prev assignment above holds this invariant due to
5411 	 * skb dereferencing near it.
5412 	 */
5413 	*pskb = skb;
5414 	return ret;
5415 }
5416 
5417 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5418 {
5419 	struct net_device *orig_dev = skb->dev;
5420 	struct packet_type *pt_prev = NULL;
5421 	int ret;
5422 
5423 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5424 	if (pt_prev)
5425 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5426 					 skb->dev, pt_prev, orig_dev);
5427 	return ret;
5428 }
5429 
5430 /**
5431  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5432  *	@skb: buffer to process
5433  *
5434  *	More direct receive version of netif_receive_skb().  It should
5435  *	only be used by callers that have a need to skip RPS and Generic XDP.
5436  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5437  *
5438  *	This function may only be called from softirq context and interrupts
5439  *	should be enabled.
5440  *
5441  *	Return values (usually ignored):
5442  *	NET_RX_SUCCESS: no congestion
5443  *	NET_RX_DROP: packet was dropped
5444  */
5445 int netif_receive_skb_core(struct sk_buff *skb)
5446 {
5447 	int ret;
5448 
5449 	rcu_read_lock();
5450 	ret = __netif_receive_skb_one_core(skb, false);
5451 	rcu_read_unlock();
5452 
5453 	return ret;
5454 }
5455 EXPORT_SYMBOL(netif_receive_skb_core);
5456 
5457 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5458 						  struct packet_type *pt_prev,
5459 						  struct net_device *orig_dev)
5460 {
5461 	struct sk_buff *skb, *next;
5462 
5463 	if (!pt_prev)
5464 		return;
5465 	if (list_empty(head))
5466 		return;
5467 	if (pt_prev->list_func != NULL)
5468 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5469 				   ip_list_rcv, head, pt_prev, orig_dev);
5470 	else
5471 		list_for_each_entry_safe(skb, next, head, list) {
5472 			skb_list_del_init(skb);
5473 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5474 		}
5475 }
5476 
5477 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5478 {
5479 	/* Fast-path assumptions:
5480 	 * - There is no RX handler.
5481 	 * - Only one packet_type matches.
5482 	 * If either of these fails, we will end up doing some per-packet
5483 	 * processing in-line, then handling the 'last ptype' for the whole
5484 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5485 	 * because the 'last ptype' must be constant across the sublist, and all
5486 	 * other ptypes are handled per-packet.
5487 	 */
5488 	/* Current (common) ptype of sublist */
5489 	struct packet_type *pt_curr = NULL;
5490 	/* Current (common) orig_dev of sublist */
5491 	struct net_device *od_curr = NULL;
5492 	struct list_head sublist;
5493 	struct sk_buff *skb, *next;
5494 
5495 	INIT_LIST_HEAD(&sublist);
5496 	list_for_each_entry_safe(skb, next, head, list) {
5497 		struct net_device *orig_dev = skb->dev;
5498 		struct packet_type *pt_prev = NULL;
5499 
5500 		skb_list_del_init(skb);
5501 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5502 		if (!pt_prev)
5503 			continue;
5504 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5505 			/* dispatch old sublist */
5506 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5507 			/* start new sublist */
5508 			INIT_LIST_HEAD(&sublist);
5509 			pt_curr = pt_prev;
5510 			od_curr = orig_dev;
5511 		}
5512 		list_add_tail(&skb->list, &sublist);
5513 	}
5514 
5515 	/* dispatch final sublist */
5516 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5517 }
5518 
5519 static int __netif_receive_skb(struct sk_buff *skb)
5520 {
5521 	int ret;
5522 
5523 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5524 		unsigned int noreclaim_flag;
5525 
5526 		/*
5527 		 * PFMEMALLOC skbs are special, they should
5528 		 * - be delivered to SOCK_MEMALLOC sockets only
5529 		 * - stay away from userspace
5530 		 * - have bounded memory usage
5531 		 *
5532 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5533 		 * context down to all allocation sites.
5534 		 */
5535 		noreclaim_flag = memalloc_noreclaim_save();
5536 		ret = __netif_receive_skb_one_core(skb, true);
5537 		memalloc_noreclaim_restore(noreclaim_flag);
5538 	} else
5539 		ret = __netif_receive_skb_one_core(skb, false);
5540 
5541 	return ret;
5542 }
5543 
5544 static void __netif_receive_skb_list(struct list_head *head)
5545 {
5546 	unsigned long noreclaim_flag = 0;
5547 	struct sk_buff *skb, *next;
5548 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5549 
5550 	list_for_each_entry_safe(skb, next, head, list) {
5551 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5552 			struct list_head sublist;
5553 
5554 			/* Handle the previous sublist */
5555 			list_cut_before(&sublist, head, &skb->list);
5556 			if (!list_empty(&sublist))
5557 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5558 			pfmemalloc = !pfmemalloc;
5559 			/* See comments in __netif_receive_skb */
5560 			if (pfmemalloc)
5561 				noreclaim_flag = memalloc_noreclaim_save();
5562 			else
5563 				memalloc_noreclaim_restore(noreclaim_flag);
5564 		}
5565 	}
5566 	/* Handle the remaining sublist */
5567 	if (!list_empty(head))
5568 		__netif_receive_skb_list_core(head, pfmemalloc);
5569 	/* Restore pflags */
5570 	if (pfmemalloc)
5571 		memalloc_noreclaim_restore(noreclaim_flag);
5572 }
5573 
5574 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5575 {
5576 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5577 	struct bpf_prog *new = xdp->prog;
5578 	int ret = 0;
5579 
5580 	switch (xdp->command) {
5581 	case XDP_SETUP_PROG:
5582 		rcu_assign_pointer(dev->xdp_prog, new);
5583 		if (old)
5584 			bpf_prog_put(old);
5585 
5586 		if (old && !new) {
5587 			static_branch_dec(&generic_xdp_needed_key);
5588 		} else if (new && !old) {
5589 			static_branch_inc(&generic_xdp_needed_key);
5590 			dev_disable_lro(dev);
5591 			dev_disable_gro_hw(dev);
5592 		}
5593 		break;
5594 
5595 	default:
5596 		ret = -EINVAL;
5597 		break;
5598 	}
5599 
5600 	return ret;
5601 }
5602 
5603 static int netif_receive_skb_internal(struct sk_buff *skb)
5604 {
5605 	int ret;
5606 
5607 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5608 
5609 	if (skb_defer_rx_timestamp(skb))
5610 		return NET_RX_SUCCESS;
5611 
5612 	rcu_read_lock();
5613 #ifdef CONFIG_RPS
5614 	if (static_branch_unlikely(&rps_needed)) {
5615 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5616 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5617 
5618 		if (cpu >= 0) {
5619 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5620 			rcu_read_unlock();
5621 			return ret;
5622 		}
5623 	}
5624 #endif
5625 	ret = __netif_receive_skb(skb);
5626 	rcu_read_unlock();
5627 	return ret;
5628 }
5629 
5630 void netif_receive_skb_list_internal(struct list_head *head)
5631 {
5632 	struct sk_buff *skb, *next;
5633 	struct list_head sublist;
5634 
5635 	INIT_LIST_HEAD(&sublist);
5636 	list_for_each_entry_safe(skb, next, head, list) {
5637 		net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5638 		skb_list_del_init(skb);
5639 		if (!skb_defer_rx_timestamp(skb))
5640 			list_add_tail(&skb->list, &sublist);
5641 	}
5642 	list_splice_init(&sublist, head);
5643 
5644 	rcu_read_lock();
5645 #ifdef CONFIG_RPS
5646 	if (static_branch_unlikely(&rps_needed)) {
5647 		list_for_each_entry_safe(skb, next, head, list) {
5648 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5649 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5650 
5651 			if (cpu >= 0) {
5652 				/* Will be handled, remove from list */
5653 				skb_list_del_init(skb);
5654 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5655 			}
5656 		}
5657 	}
5658 #endif
5659 	__netif_receive_skb_list(head);
5660 	rcu_read_unlock();
5661 }
5662 
5663 /**
5664  *	netif_receive_skb - process receive buffer from network
5665  *	@skb: buffer to process
5666  *
5667  *	netif_receive_skb() is the main receive data processing function.
5668  *	It always succeeds. The buffer may be dropped during processing
5669  *	for congestion control or by the protocol layers.
5670  *
5671  *	This function may only be called from softirq context and interrupts
5672  *	should be enabled.
5673  *
5674  *	Return values (usually ignored):
5675  *	NET_RX_SUCCESS: no congestion
5676  *	NET_RX_DROP: packet was dropped
5677  */
5678 int netif_receive_skb(struct sk_buff *skb)
5679 {
5680 	int ret;
5681 
5682 	trace_netif_receive_skb_entry(skb);
5683 
5684 	ret = netif_receive_skb_internal(skb);
5685 	trace_netif_receive_skb_exit(ret);
5686 
5687 	return ret;
5688 }
5689 EXPORT_SYMBOL(netif_receive_skb);
5690 
5691 /**
5692  *	netif_receive_skb_list - process many receive buffers from network
5693  *	@head: list of skbs to process.
5694  *
5695  *	Since return value of netif_receive_skb() is normally ignored, and
5696  *	wouldn't be meaningful for a list, this function returns void.
5697  *
5698  *	This function may only be called from softirq context and interrupts
5699  *	should be enabled.
5700  */
5701 void netif_receive_skb_list(struct list_head *head)
5702 {
5703 	struct sk_buff *skb;
5704 
5705 	if (list_empty(head))
5706 		return;
5707 	if (trace_netif_receive_skb_list_entry_enabled()) {
5708 		list_for_each_entry(skb, head, list)
5709 			trace_netif_receive_skb_list_entry(skb);
5710 	}
5711 	netif_receive_skb_list_internal(head);
5712 	trace_netif_receive_skb_list_exit(0);
5713 }
5714 EXPORT_SYMBOL(netif_receive_skb_list);
5715 
5716 static DEFINE_PER_CPU(struct work_struct, flush_works);
5717 
5718 /* Network device is going away, flush any packets still pending */
5719 static void flush_backlog(struct work_struct *work)
5720 {
5721 	struct sk_buff *skb, *tmp;
5722 	struct softnet_data *sd;
5723 
5724 	local_bh_disable();
5725 	sd = this_cpu_ptr(&softnet_data);
5726 
5727 	rps_lock_irq_disable(sd);
5728 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5729 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5730 			__skb_unlink(skb, &sd->input_pkt_queue);
5731 			dev_kfree_skb_irq(skb);
5732 			input_queue_head_incr(sd);
5733 		}
5734 	}
5735 	rps_unlock_irq_enable(sd);
5736 
5737 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5738 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5739 			__skb_unlink(skb, &sd->process_queue);
5740 			kfree_skb(skb);
5741 			input_queue_head_incr(sd);
5742 		}
5743 	}
5744 	local_bh_enable();
5745 }
5746 
5747 static bool flush_required(int cpu)
5748 {
5749 #if IS_ENABLED(CONFIG_RPS)
5750 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5751 	bool do_flush;
5752 
5753 	rps_lock_irq_disable(sd);
5754 
5755 	/* as insertion into process_queue happens with the rps lock held,
5756 	 * process_queue access may race only with dequeue
5757 	 */
5758 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5759 		   !skb_queue_empty_lockless(&sd->process_queue);
5760 	rps_unlock_irq_enable(sd);
5761 
5762 	return do_flush;
5763 #endif
5764 	/* without RPS we can't safely check input_pkt_queue: during a
5765 	 * concurrent remote skb_queue_splice() we can detect as empty both
5766 	 * input_pkt_queue and process_queue even if the latter could end-up
5767 	 * containing a lot of packets.
5768 	 */
5769 	return true;
5770 }
5771 
5772 static void flush_all_backlogs(void)
5773 {
5774 	static cpumask_t flush_cpus;
5775 	unsigned int cpu;
5776 
5777 	/* since we are under rtnl lock protection we can use static data
5778 	 * for the cpumask and avoid allocating on stack the possibly
5779 	 * large mask
5780 	 */
5781 	ASSERT_RTNL();
5782 
5783 	cpus_read_lock();
5784 
5785 	cpumask_clear(&flush_cpus);
5786 	for_each_online_cpu(cpu) {
5787 		if (flush_required(cpu)) {
5788 			queue_work_on(cpu, system_highpri_wq,
5789 				      per_cpu_ptr(&flush_works, cpu));
5790 			cpumask_set_cpu(cpu, &flush_cpus);
5791 		}
5792 	}
5793 
5794 	/* we can have in flight packet[s] on the cpus we are not flushing,
5795 	 * synchronize_net() in unregister_netdevice_many() will take care of
5796 	 * them
5797 	 */
5798 	for_each_cpu(cpu, &flush_cpus)
5799 		flush_work(per_cpu_ptr(&flush_works, cpu));
5800 
5801 	cpus_read_unlock();
5802 }
5803 
5804 static void net_rps_send_ipi(struct softnet_data *remsd)
5805 {
5806 #ifdef CONFIG_RPS
5807 	while (remsd) {
5808 		struct softnet_data *next = remsd->rps_ipi_next;
5809 
5810 		if (cpu_online(remsd->cpu))
5811 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
5812 		remsd = next;
5813 	}
5814 #endif
5815 }
5816 
5817 /*
5818  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5819  * Note: called with local irq disabled, but exits with local irq enabled.
5820  */
5821 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5822 {
5823 #ifdef CONFIG_RPS
5824 	struct softnet_data *remsd = sd->rps_ipi_list;
5825 
5826 	if (remsd) {
5827 		sd->rps_ipi_list = NULL;
5828 
5829 		local_irq_enable();
5830 
5831 		/* Send pending IPI's to kick RPS processing on remote cpus. */
5832 		net_rps_send_ipi(remsd);
5833 	} else
5834 #endif
5835 		local_irq_enable();
5836 }
5837 
5838 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5839 {
5840 #ifdef CONFIG_RPS
5841 	return sd->rps_ipi_list != NULL;
5842 #else
5843 	return false;
5844 #endif
5845 }
5846 
5847 static int process_backlog(struct napi_struct *napi, int quota)
5848 {
5849 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5850 	bool again = true;
5851 	int work = 0;
5852 
5853 	/* Check if we have pending ipi, its better to send them now,
5854 	 * not waiting net_rx_action() end.
5855 	 */
5856 	if (sd_has_rps_ipi_waiting(sd)) {
5857 		local_irq_disable();
5858 		net_rps_action_and_irq_enable(sd);
5859 	}
5860 
5861 	napi->weight = READ_ONCE(dev_rx_weight);
5862 	while (again) {
5863 		struct sk_buff *skb;
5864 
5865 		while ((skb = __skb_dequeue(&sd->process_queue))) {
5866 			rcu_read_lock();
5867 			__netif_receive_skb(skb);
5868 			rcu_read_unlock();
5869 			input_queue_head_incr(sd);
5870 			if (++work >= quota)
5871 				return work;
5872 
5873 		}
5874 
5875 		rps_lock_irq_disable(sd);
5876 		if (skb_queue_empty(&sd->input_pkt_queue)) {
5877 			/*
5878 			 * Inline a custom version of __napi_complete().
5879 			 * only current cpu owns and manipulates this napi,
5880 			 * and NAPI_STATE_SCHED is the only possible flag set
5881 			 * on backlog.
5882 			 * We can use a plain write instead of clear_bit(),
5883 			 * and we dont need an smp_mb() memory barrier.
5884 			 */
5885 			napi->state = 0;
5886 			again = false;
5887 		} else {
5888 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
5889 						   &sd->process_queue);
5890 		}
5891 		rps_unlock_irq_enable(sd);
5892 	}
5893 
5894 	return work;
5895 }
5896 
5897 /**
5898  * __napi_schedule - schedule for receive
5899  * @n: entry to schedule
5900  *
5901  * The entry's receive function will be scheduled to run.
5902  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5903  */
5904 void __napi_schedule(struct napi_struct *n)
5905 {
5906 	unsigned long flags;
5907 
5908 	local_irq_save(flags);
5909 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5910 	local_irq_restore(flags);
5911 }
5912 EXPORT_SYMBOL(__napi_schedule);
5913 
5914 /**
5915  *	napi_schedule_prep - check if napi can be scheduled
5916  *	@n: napi context
5917  *
5918  * Test if NAPI routine is already running, and if not mark
5919  * it as running.  This is used as a condition variable to
5920  * insure only one NAPI poll instance runs.  We also make
5921  * sure there is no pending NAPI disable.
5922  */
5923 bool napi_schedule_prep(struct napi_struct *n)
5924 {
5925 	unsigned long new, val = READ_ONCE(n->state);
5926 
5927 	do {
5928 		if (unlikely(val & NAPIF_STATE_DISABLE))
5929 			return false;
5930 		new = val | NAPIF_STATE_SCHED;
5931 
5932 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
5933 		 * This was suggested by Alexander Duyck, as compiler
5934 		 * emits better code than :
5935 		 * if (val & NAPIF_STATE_SCHED)
5936 		 *     new |= NAPIF_STATE_MISSED;
5937 		 */
5938 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5939 						   NAPIF_STATE_MISSED;
5940 	} while (!try_cmpxchg(&n->state, &val, new));
5941 
5942 	return !(val & NAPIF_STATE_SCHED);
5943 }
5944 EXPORT_SYMBOL(napi_schedule_prep);
5945 
5946 /**
5947  * __napi_schedule_irqoff - schedule for receive
5948  * @n: entry to schedule
5949  *
5950  * Variant of __napi_schedule() assuming hard irqs are masked.
5951  *
5952  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5953  * because the interrupt disabled assumption might not be true
5954  * due to force-threaded interrupts and spinlock substitution.
5955  */
5956 void __napi_schedule_irqoff(struct napi_struct *n)
5957 {
5958 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5959 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
5960 	else
5961 		__napi_schedule(n);
5962 }
5963 EXPORT_SYMBOL(__napi_schedule_irqoff);
5964 
5965 bool napi_complete_done(struct napi_struct *n, int work_done)
5966 {
5967 	unsigned long flags, val, new, timeout = 0;
5968 	bool ret = true;
5969 
5970 	/*
5971 	 * 1) Don't let napi dequeue from the cpu poll list
5972 	 *    just in case its running on a different cpu.
5973 	 * 2) If we are busy polling, do nothing here, we have
5974 	 *    the guarantee we will be called later.
5975 	 */
5976 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5977 				 NAPIF_STATE_IN_BUSY_POLL)))
5978 		return false;
5979 
5980 	if (work_done) {
5981 		if (n->gro_bitmask)
5982 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
5983 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
5984 	}
5985 	if (n->defer_hard_irqs_count > 0) {
5986 		n->defer_hard_irqs_count--;
5987 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
5988 		if (timeout)
5989 			ret = false;
5990 	}
5991 	if (n->gro_bitmask) {
5992 		/* When the NAPI instance uses a timeout and keeps postponing
5993 		 * it, we need to bound somehow the time packets are kept in
5994 		 * the GRO layer
5995 		 */
5996 		napi_gro_flush(n, !!timeout);
5997 	}
5998 
5999 	gro_normal_list(n);
6000 
6001 	if (unlikely(!list_empty(&n->poll_list))) {
6002 		/* If n->poll_list is not empty, we need to mask irqs */
6003 		local_irq_save(flags);
6004 		list_del_init(&n->poll_list);
6005 		local_irq_restore(flags);
6006 	}
6007 	WRITE_ONCE(n->list_owner, -1);
6008 
6009 	val = READ_ONCE(n->state);
6010 	do {
6011 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6012 
6013 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6014 			      NAPIF_STATE_SCHED_THREADED |
6015 			      NAPIF_STATE_PREFER_BUSY_POLL);
6016 
6017 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6018 		 * because we will call napi->poll() one more time.
6019 		 * This C code was suggested by Alexander Duyck to help gcc.
6020 		 */
6021 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6022 						    NAPIF_STATE_SCHED;
6023 	} while (!try_cmpxchg(&n->state, &val, new));
6024 
6025 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6026 		__napi_schedule(n);
6027 		return false;
6028 	}
6029 
6030 	if (timeout)
6031 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6032 			      HRTIMER_MODE_REL_PINNED);
6033 	return ret;
6034 }
6035 EXPORT_SYMBOL(napi_complete_done);
6036 
6037 /* must be called under rcu_read_lock(), as we dont take a reference */
6038 static struct napi_struct *napi_by_id(unsigned int napi_id)
6039 {
6040 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6041 	struct napi_struct *napi;
6042 
6043 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6044 		if (napi->napi_id == napi_id)
6045 			return napi;
6046 
6047 	return NULL;
6048 }
6049 
6050 #if defined(CONFIG_NET_RX_BUSY_POLL)
6051 
6052 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6053 {
6054 	if (!skip_schedule) {
6055 		gro_normal_list(napi);
6056 		__napi_schedule(napi);
6057 		return;
6058 	}
6059 
6060 	if (napi->gro_bitmask) {
6061 		/* flush too old packets
6062 		 * If HZ < 1000, flush all packets.
6063 		 */
6064 		napi_gro_flush(napi, HZ >= 1000);
6065 	}
6066 
6067 	gro_normal_list(napi);
6068 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6069 }
6070 
6071 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6072 			   u16 budget)
6073 {
6074 	bool skip_schedule = false;
6075 	unsigned long timeout;
6076 	int rc;
6077 
6078 	/* Busy polling means there is a high chance device driver hard irq
6079 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6080 	 * set in napi_schedule_prep().
6081 	 * Since we are about to call napi->poll() once more, we can safely
6082 	 * clear NAPI_STATE_MISSED.
6083 	 *
6084 	 * Note: x86 could use a single "lock and ..." instruction
6085 	 * to perform these two clear_bit()
6086 	 */
6087 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6088 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6089 
6090 	local_bh_disable();
6091 
6092 	if (prefer_busy_poll) {
6093 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6094 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6095 		if (napi->defer_hard_irqs_count && timeout) {
6096 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6097 			skip_schedule = true;
6098 		}
6099 	}
6100 
6101 	/* All we really want here is to re-enable device interrupts.
6102 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6103 	 */
6104 	rc = napi->poll(napi, budget);
6105 	/* We can't gro_normal_list() here, because napi->poll() might have
6106 	 * rearmed the napi (napi_complete_done()) in which case it could
6107 	 * already be running on another CPU.
6108 	 */
6109 	trace_napi_poll(napi, rc, budget);
6110 	netpoll_poll_unlock(have_poll_lock);
6111 	if (rc == budget)
6112 		__busy_poll_stop(napi, skip_schedule);
6113 	local_bh_enable();
6114 }
6115 
6116 void napi_busy_loop(unsigned int napi_id,
6117 		    bool (*loop_end)(void *, unsigned long),
6118 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6119 {
6120 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6121 	int (*napi_poll)(struct napi_struct *napi, int budget);
6122 	void *have_poll_lock = NULL;
6123 	struct napi_struct *napi;
6124 
6125 restart:
6126 	napi_poll = NULL;
6127 
6128 	rcu_read_lock();
6129 
6130 	napi = napi_by_id(napi_id);
6131 	if (!napi)
6132 		goto out;
6133 
6134 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6135 		preempt_disable();
6136 	for (;;) {
6137 		int work = 0;
6138 
6139 		local_bh_disable();
6140 		if (!napi_poll) {
6141 			unsigned long val = READ_ONCE(napi->state);
6142 
6143 			/* If multiple threads are competing for this napi,
6144 			 * we avoid dirtying napi->state as much as we can.
6145 			 */
6146 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6147 				   NAPIF_STATE_IN_BUSY_POLL)) {
6148 				if (prefer_busy_poll)
6149 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6150 				goto count;
6151 			}
6152 			if (cmpxchg(&napi->state, val,
6153 				    val | NAPIF_STATE_IN_BUSY_POLL |
6154 					  NAPIF_STATE_SCHED) != val) {
6155 				if (prefer_busy_poll)
6156 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6157 				goto count;
6158 			}
6159 			have_poll_lock = netpoll_poll_lock(napi);
6160 			napi_poll = napi->poll;
6161 		}
6162 		work = napi_poll(napi, budget);
6163 		trace_napi_poll(napi, work, budget);
6164 		gro_normal_list(napi);
6165 count:
6166 		if (work > 0)
6167 			__NET_ADD_STATS(dev_net(napi->dev),
6168 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6169 		local_bh_enable();
6170 
6171 		if (!loop_end || loop_end(loop_end_arg, start_time))
6172 			break;
6173 
6174 		if (unlikely(need_resched())) {
6175 			if (napi_poll)
6176 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6177 			if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6178 				preempt_enable();
6179 			rcu_read_unlock();
6180 			cond_resched();
6181 			if (loop_end(loop_end_arg, start_time))
6182 				return;
6183 			goto restart;
6184 		}
6185 		cpu_relax();
6186 	}
6187 	if (napi_poll)
6188 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6189 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6190 		preempt_enable();
6191 out:
6192 	rcu_read_unlock();
6193 }
6194 EXPORT_SYMBOL(napi_busy_loop);
6195 
6196 #endif /* CONFIG_NET_RX_BUSY_POLL */
6197 
6198 static void napi_hash_add(struct napi_struct *napi)
6199 {
6200 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6201 		return;
6202 
6203 	spin_lock(&napi_hash_lock);
6204 
6205 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6206 	do {
6207 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6208 			napi_gen_id = MIN_NAPI_ID;
6209 	} while (napi_by_id(napi_gen_id));
6210 	napi->napi_id = napi_gen_id;
6211 
6212 	hlist_add_head_rcu(&napi->napi_hash_node,
6213 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6214 
6215 	spin_unlock(&napi_hash_lock);
6216 }
6217 
6218 /* Warning : caller is responsible to make sure rcu grace period
6219  * is respected before freeing memory containing @napi
6220  */
6221 static void napi_hash_del(struct napi_struct *napi)
6222 {
6223 	spin_lock(&napi_hash_lock);
6224 
6225 	hlist_del_init_rcu(&napi->napi_hash_node);
6226 
6227 	spin_unlock(&napi_hash_lock);
6228 }
6229 
6230 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6231 {
6232 	struct napi_struct *napi;
6233 
6234 	napi = container_of(timer, struct napi_struct, timer);
6235 
6236 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6237 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6238 	 */
6239 	if (!napi_disable_pending(napi) &&
6240 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6241 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6242 		__napi_schedule_irqoff(napi);
6243 	}
6244 
6245 	return HRTIMER_NORESTART;
6246 }
6247 
6248 static void init_gro_hash(struct napi_struct *napi)
6249 {
6250 	int i;
6251 
6252 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6253 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6254 		napi->gro_hash[i].count = 0;
6255 	}
6256 	napi->gro_bitmask = 0;
6257 }
6258 
6259 int dev_set_threaded(struct net_device *dev, bool threaded)
6260 {
6261 	struct napi_struct *napi;
6262 	int err = 0;
6263 
6264 	if (dev->threaded == threaded)
6265 		return 0;
6266 
6267 	if (threaded) {
6268 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6269 			if (!napi->thread) {
6270 				err = napi_kthread_create(napi);
6271 				if (err) {
6272 					threaded = false;
6273 					break;
6274 				}
6275 			}
6276 		}
6277 	}
6278 
6279 	dev->threaded = threaded;
6280 
6281 	/* Make sure kthread is created before THREADED bit
6282 	 * is set.
6283 	 */
6284 	smp_mb__before_atomic();
6285 
6286 	/* Setting/unsetting threaded mode on a napi might not immediately
6287 	 * take effect, if the current napi instance is actively being
6288 	 * polled. In this case, the switch between threaded mode and
6289 	 * softirq mode will happen in the next round of napi_schedule().
6290 	 * This should not cause hiccups/stalls to the live traffic.
6291 	 */
6292 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6293 		if (threaded)
6294 			set_bit(NAPI_STATE_THREADED, &napi->state);
6295 		else
6296 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6297 	}
6298 
6299 	return err;
6300 }
6301 EXPORT_SYMBOL(dev_set_threaded);
6302 
6303 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6304 			   int (*poll)(struct napi_struct *, int), int weight)
6305 {
6306 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6307 		return;
6308 
6309 	INIT_LIST_HEAD(&napi->poll_list);
6310 	INIT_HLIST_NODE(&napi->napi_hash_node);
6311 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6312 	napi->timer.function = napi_watchdog;
6313 	init_gro_hash(napi);
6314 	napi->skb = NULL;
6315 	INIT_LIST_HEAD(&napi->rx_list);
6316 	napi->rx_count = 0;
6317 	napi->poll = poll;
6318 	if (weight > NAPI_POLL_WEIGHT)
6319 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6320 				weight);
6321 	napi->weight = weight;
6322 	napi->dev = dev;
6323 #ifdef CONFIG_NETPOLL
6324 	napi->poll_owner = -1;
6325 #endif
6326 	napi->list_owner = -1;
6327 	set_bit(NAPI_STATE_SCHED, &napi->state);
6328 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6329 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6330 	napi_hash_add(napi);
6331 	napi_get_frags_check(napi);
6332 	/* Create kthread for this napi if dev->threaded is set.
6333 	 * Clear dev->threaded if kthread creation failed so that
6334 	 * threaded mode will not be enabled in napi_enable().
6335 	 */
6336 	if (dev->threaded && napi_kthread_create(napi))
6337 		dev->threaded = 0;
6338 }
6339 EXPORT_SYMBOL(netif_napi_add_weight);
6340 
6341 void napi_disable(struct napi_struct *n)
6342 {
6343 	unsigned long val, new;
6344 
6345 	might_sleep();
6346 	set_bit(NAPI_STATE_DISABLE, &n->state);
6347 
6348 	val = READ_ONCE(n->state);
6349 	do {
6350 		while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6351 			usleep_range(20, 200);
6352 			val = READ_ONCE(n->state);
6353 		}
6354 
6355 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6356 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6357 	} while (!try_cmpxchg(&n->state, &val, new));
6358 
6359 	hrtimer_cancel(&n->timer);
6360 
6361 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6362 }
6363 EXPORT_SYMBOL(napi_disable);
6364 
6365 /**
6366  *	napi_enable - enable NAPI scheduling
6367  *	@n: NAPI context
6368  *
6369  * Resume NAPI from being scheduled on this context.
6370  * Must be paired with napi_disable.
6371  */
6372 void napi_enable(struct napi_struct *n)
6373 {
6374 	unsigned long new, val = READ_ONCE(n->state);
6375 
6376 	do {
6377 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6378 
6379 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6380 		if (n->dev->threaded && n->thread)
6381 			new |= NAPIF_STATE_THREADED;
6382 	} while (!try_cmpxchg(&n->state, &val, new));
6383 }
6384 EXPORT_SYMBOL(napi_enable);
6385 
6386 static void flush_gro_hash(struct napi_struct *napi)
6387 {
6388 	int i;
6389 
6390 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6391 		struct sk_buff *skb, *n;
6392 
6393 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6394 			kfree_skb(skb);
6395 		napi->gro_hash[i].count = 0;
6396 	}
6397 }
6398 
6399 /* Must be called in process context */
6400 void __netif_napi_del(struct napi_struct *napi)
6401 {
6402 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6403 		return;
6404 
6405 	napi_hash_del(napi);
6406 	list_del_rcu(&napi->dev_list);
6407 	napi_free_frags(napi);
6408 
6409 	flush_gro_hash(napi);
6410 	napi->gro_bitmask = 0;
6411 
6412 	if (napi->thread) {
6413 		kthread_stop(napi->thread);
6414 		napi->thread = NULL;
6415 	}
6416 }
6417 EXPORT_SYMBOL(__netif_napi_del);
6418 
6419 static int __napi_poll(struct napi_struct *n, bool *repoll)
6420 {
6421 	int work, weight;
6422 
6423 	weight = n->weight;
6424 
6425 	/* This NAPI_STATE_SCHED test is for avoiding a race
6426 	 * with netpoll's poll_napi().  Only the entity which
6427 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6428 	 * actually make the ->poll() call.  Therefore we avoid
6429 	 * accidentally calling ->poll() when NAPI is not scheduled.
6430 	 */
6431 	work = 0;
6432 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6433 		work = n->poll(n, weight);
6434 		trace_napi_poll(n, work, weight);
6435 	}
6436 
6437 	if (unlikely(work > weight))
6438 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6439 				n->poll, work, weight);
6440 
6441 	if (likely(work < weight))
6442 		return work;
6443 
6444 	/* Drivers must not modify the NAPI state if they
6445 	 * consume the entire weight.  In such cases this code
6446 	 * still "owns" the NAPI instance and therefore can
6447 	 * move the instance around on the list at-will.
6448 	 */
6449 	if (unlikely(napi_disable_pending(n))) {
6450 		napi_complete(n);
6451 		return work;
6452 	}
6453 
6454 	/* The NAPI context has more processing work, but busy-polling
6455 	 * is preferred. Exit early.
6456 	 */
6457 	if (napi_prefer_busy_poll(n)) {
6458 		if (napi_complete_done(n, work)) {
6459 			/* If timeout is not set, we need to make sure
6460 			 * that the NAPI is re-scheduled.
6461 			 */
6462 			napi_schedule(n);
6463 		}
6464 		return work;
6465 	}
6466 
6467 	if (n->gro_bitmask) {
6468 		/* flush too old packets
6469 		 * If HZ < 1000, flush all packets.
6470 		 */
6471 		napi_gro_flush(n, HZ >= 1000);
6472 	}
6473 
6474 	gro_normal_list(n);
6475 
6476 	/* Some drivers may have called napi_schedule
6477 	 * prior to exhausting their budget.
6478 	 */
6479 	if (unlikely(!list_empty(&n->poll_list))) {
6480 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6481 			     n->dev ? n->dev->name : "backlog");
6482 		return work;
6483 	}
6484 
6485 	*repoll = true;
6486 
6487 	return work;
6488 }
6489 
6490 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6491 {
6492 	bool do_repoll = false;
6493 	void *have;
6494 	int work;
6495 
6496 	list_del_init(&n->poll_list);
6497 
6498 	have = netpoll_poll_lock(n);
6499 
6500 	work = __napi_poll(n, &do_repoll);
6501 
6502 	if (do_repoll)
6503 		list_add_tail(&n->poll_list, repoll);
6504 
6505 	netpoll_poll_unlock(have);
6506 
6507 	return work;
6508 }
6509 
6510 static int napi_thread_wait(struct napi_struct *napi)
6511 {
6512 	bool woken = false;
6513 
6514 	set_current_state(TASK_INTERRUPTIBLE);
6515 
6516 	while (!kthread_should_stop()) {
6517 		/* Testing SCHED_THREADED bit here to make sure the current
6518 		 * kthread owns this napi and could poll on this napi.
6519 		 * Testing SCHED bit is not enough because SCHED bit might be
6520 		 * set by some other busy poll thread or by napi_disable().
6521 		 */
6522 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6523 			WARN_ON(!list_empty(&napi->poll_list));
6524 			__set_current_state(TASK_RUNNING);
6525 			return 0;
6526 		}
6527 
6528 		schedule();
6529 		/* woken being true indicates this thread owns this napi. */
6530 		woken = true;
6531 		set_current_state(TASK_INTERRUPTIBLE);
6532 	}
6533 	__set_current_state(TASK_RUNNING);
6534 
6535 	return -1;
6536 }
6537 
6538 static void skb_defer_free_flush(struct softnet_data *sd)
6539 {
6540 	struct sk_buff *skb, *next;
6541 
6542 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6543 	if (!READ_ONCE(sd->defer_list))
6544 		return;
6545 
6546 	spin_lock(&sd->defer_lock);
6547 	skb = sd->defer_list;
6548 	sd->defer_list = NULL;
6549 	sd->defer_count = 0;
6550 	spin_unlock(&sd->defer_lock);
6551 
6552 	while (skb != NULL) {
6553 		next = skb->next;
6554 		napi_consume_skb(skb, 1);
6555 		skb = next;
6556 	}
6557 }
6558 
6559 static int napi_threaded_poll(void *data)
6560 {
6561 	struct napi_struct *napi = data;
6562 	struct softnet_data *sd;
6563 	void *have;
6564 
6565 	while (!napi_thread_wait(napi)) {
6566 		for (;;) {
6567 			bool repoll = false;
6568 
6569 			local_bh_disable();
6570 			sd = this_cpu_ptr(&softnet_data);
6571 			sd->in_napi_threaded_poll = true;
6572 
6573 			have = netpoll_poll_lock(napi);
6574 			__napi_poll(napi, &repoll);
6575 			netpoll_poll_unlock(have);
6576 
6577 			sd->in_napi_threaded_poll = false;
6578 			barrier();
6579 
6580 			if (sd_has_rps_ipi_waiting(sd)) {
6581 				local_irq_disable();
6582 				net_rps_action_and_irq_enable(sd);
6583 			}
6584 			skb_defer_free_flush(sd);
6585 			local_bh_enable();
6586 
6587 			if (!repoll)
6588 				break;
6589 
6590 			cond_resched();
6591 		}
6592 	}
6593 	return 0;
6594 }
6595 
6596 static __latent_entropy void net_rx_action(struct softirq_action *h)
6597 {
6598 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6599 	unsigned long time_limit = jiffies +
6600 		usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6601 	int budget = READ_ONCE(netdev_budget);
6602 	LIST_HEAD(list);
6603 	LIST_HEAD(repoll);
6604 
6605 start:
6606 	sd->in_net_rx_action = true;
6607 	local_irq_disable();
6608 	list_splice_init(&sd->poll_list, &list);
6609 	local_irq_enable();
6610 
6611 	for (;;) {
6612 		struct napi_struct *n;
6613 
6614 		skb_defer_free_flush(sd);
6615 
6616 		if (list_empty(&list)) {
6617 			if (list_empty(&repoll)) {
6618 				sd->in_net_rx_action = false;
6619 				barrier();
6620 				/* We need to check if ____napi_schedule()
6621 				 * had refilled poll_list while
6622 				 * sd->in_net_rx_action was true.
6623 				 */
6624 				if (!list_empty(&sd->poll_list))
6625 					goto start;
6626 				if (!sd_has_rps_ipi_waiting(sd))
6627 					goto end;
6628 			}
6629 			break;
6630 		}
6631 
6632 		n = list_first_entry(&list, struct napi_struct, poll_list);
6633 		budget -= napi_poll(n, &repoll);
6634 
6635 		/* If softirq window is exhausted then punt.
6636 		 * Allow this to run for 2 jiffies since which will allow
6637 		 * an average latency of 1.5/HZ.
6638 		 */
6639 		if (unlikely(budget <= 0 ||
6640 			     time_after_eq(jiffies, time_limit))) {
6641 			sd->time_squeeze++;
6642 			break;
6643 		}
6644 	}
6645 
6646 	local_irq_disable();
6647 
6648 	list_splice_tail_init(&sd->poll_list, &list);
6649 	list_splice_tail(&repoll, &list);
6650 	list_splice(&list, &sd->poll_list);
6651 	if (!list_empty(&sd->poll_list))
6652 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6653 	else
6654 		sd->in_net_rx_action = false;
6655 
6656 	net_rps_action_and_irq_enable(sd);
6657 end:;
6658 }
6659 
6660 struct netdev_adjacent {
6661 	struct net_device *dev;
6662 	netdevice_tracker dev_tracker;
6663 
6664 	/* upper master flag, there can only be one master device per list */
6665 	bool master;
6666 
6667 	/* lookup ignore flag */
6668 	bool ignore;
6669 
6670 	/* counter for the number of times this device was added to us */
6671 	u16 ref_nr;
6672 
6673 	/* private field for the users */
6674 	void *private;
6675 
6676 	struct list_head list;
6677 	struct rcu_head rcu;
6678 };
6679 
6680 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6681 						 struct list_head *adj_list)
6682 {
6683 	struct netdev_adjacent *adj;
6684 
6685 	list_for_each_entry(adj, adj_list, list) {
6686 		if (adj->dev == adj_dev)
6687 			return adj;
6688 	}
6689 	return NULL;
6690 }
6691 
6692 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6693 				    struct netdev_nested_priv *priv)
6694 {
6695 	struct net_device *dev = (struct net_device *)priv->data;
6696 
6697 	return upper_dev == dev;
6698 }
6699 
6700 /**
6701  * netdev_has_upper_dev - Check if device is linked to an upper device
6702  * @dev: device
6703  * @upper_dev: upper device to check
6704  *
6705  * Find out if a device is linked to specified upper device and return true
6706  * in case it is. Note that this checks only immediate upper device,
6707  * not through a complete stack of devices. The caller must hold the RTNL lock.
6708  */
6709 bool netdev_has_upper_dev(struct net_device *dev,
6710 			  struct net_device *upper_dev)
6711 {
6712 	struct netdev_nested_priv priv = {
6713 		.data = (void *)upper_dev,
6714 	};
6715 
6716 	ASSERT_RTNL();
6717 
6718 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6719 					     &priv);
6720 }
6721 EXPORT_SYMBOL(netdev_has_upper_dev);
6722 
6723 /**
6724  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6725  * @dev: device
6726  * @upper_dev: upper device to check
6727  *
6728  * Find out if a device is linked to specified upper device and return true
6729  * in case it is. Note that this checks the entire upper device chain.
6730  * The caller must hold rcu lock.
6731  */
6732 
6733 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6734 				  struct net_device *upper_dev)
6735 {
6736 	struct netdev_nested_priv priv = {
6737 		.data = (void *)upper_dev,
6738 	};
6739 
6740 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6741 					       &priv);
6742 }
6743 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6744 
6745 /**
6746  * netdev_has_any_upper_dev - Check if device is linked to some device
6747  * @dev: device
6748  *
6749  * Find out if a device is linked to an upper device and return true in case
6750  * it is. The caller must hold the RTNL lock.
6751  */
6752 bool netdev_has_any_upper_dev(struct net_device *dev)
6753 {
6754 	ASSERT_RTNL();
6755 
6756 	return !list_empty(&dev->adj_list.upper);
6757 }
6758 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6759 
6760 /**
6761  * netdev_master_upper_dev_get - Get master upper device
6762  * @dev: device
6763  *
6764  * Find a master upper device and return pointer to it or NULL in case
6765  * it's not there. The caller must hold the RTNL lock.
6766  */
6767 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6768 {
6769 	struct netdev_adjacent *upper;
6770 
6771 	ASSERT_RTNL();
6772 
6773 	if (list_empty(&dev->adj_list.upper))
6774 		return NULL;
6775 
6776 	upper = list_first_entry(&dev->adj_list.upper,
6777 				 struct netdev_adjacent, list);
6778 	if (likely(upper->master))
6779 		return upper->dev;
6780 	return NULL;
6781 }
6782 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6783 
6784 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6785 {
6786 	struct netdev_adjacent *upper;
6787 
6788 	ASSERT_RTNL();
6789 
6790 	if (list_empty(&dev->adj_list.upper))
6791 		return NULL;
6792 
6793 	upper = list_first_entry(&dev->adj_list.upper,
6794 				 struct netdev_adjacent, list);
6795 	if (likely(upper->master) && !upper->ignore)
6796 		return upper->dev;
6797 	return NULL;
6798 }
6799 
6800 /**
6801  * netdev_has_any_lower_dev - Check if device is linked to some device
6802  * @dev: device
6803  *
6804  * Find out if a device is linked to a lower device and return true in case
6805  * it is. The caller must hold the RTNL lock.
6806  */
6807 static bool netdev_has_any_lower_dev(struct net_device *dev)
6808 {
6809 	ASSERT_RTNL();
6810 
6811 	return !list_empty(&dev->adj_list.lower);
6812 }
6813 
6814 void *netdev_adjacent_get_private(struct list_head *adj_list)
6815 {
6816 	struct netdev_adjacent *adj;
6817 
6818 	adj = list_entry(adj_list, struct netdev_adjacent, list);
6819 
6820 	return adj->private;
6821 }
6822 EXPORT_SYMBOL(netdev_adjacent_get_private);
6823 
6824 /**
6825  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6826  * @dev: device
6827  * @iter: list_head ** of the current position
6828  *
6829  * Gets the next device from the dev's upper list, starting from iter
6830  * position. The caller must hold RCU read lock.
6831  */
6832 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6833 						 struct list_head **iter)
6834 {
6835 	struct netdev_adjacent *upper;
6836 
6837 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6838 
6839 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6840 
6841 	if (&upper->list == &dev->adj_list.upper)
6842 		return NULL;
6843 
6844 	*iter = &upper->list;
6845 
6846 	return upper->dev;
6847 }
6848 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6849 
6850 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6851 						  struct list_head **iter,
6852 						  bool *ignore)
6853 {
6854 	struct netdev_adjacent *upper;
6855 
6856 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6857 
6858 	if (&upper->list == &dev->adj_list.upper)
6859 		return NULL;
6860 
6861 	*iter = &upper->list;
6862 	*ignore = upper->ignore;
6863 
6864 	return upper->dev;
6865 }
6866 
6867 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6868 						    struct list_head **iter)
6869 {
6870 	struct netdev_adjacent *upper;
6871 
6872 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6873 
6874 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6875 
6876 	if (&upper->list == &dev->adj_list.upper)
6877 		return NULL;
6878 
6879 	*iter = &upper->list;
6880 
6881 	return upper->dev;
6882 }
6883 
6884 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6885 				       int (*fn)(struct net_device *dev,
6886 					 struct netdev_nested_priv *priv),
6887 				       struct netdev_nested_priv *priv)
6888 {
6889 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6890 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6891 	int ret, cur = 0;
6892 	bool ignore;
6893 
6894 	now = dev;
6895 	iter = &dev->adj_list.upper;
6896 
6897 	while (1) {
6898 		if (now != dev) {
6899 			ret = fn(now, priv);
6900 			if (ret)
6901 				return ret;
6902 		}
6903 
6904 		next = NULL;
6905 		while (1) {
6906 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
6907 			if (!udev)
6908 				break;
6909 			if (ignore)
6910 				continue;
6911 
6912 			next = udev;
6913 			niter = &udev->adj_list.upper;
6914 			dev_stack[cur] = now;
6915 			iter_stack[cur++] = iter;
6916 			break;
6917 		}
6918 
6919 		if (!next) {
6920 			if (!cur)
6921 				return 0;
6922 			next = dev_stack[--cur];
6923 			niter = iter_stack[cur];
6924 		}
6925 
6926 		now = next;
6927 		iter = niter;
6928 	}
6929 
6930 	return 0;
6931 }
6932 
6933 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6934 				  int (*fn)(struct net_device *dev,
6935 					    struct netdev_nested_priv *priv),
6936 				  struct netdev_nested_priv *priv)
6937 {
6938 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6939 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6940 	int ret, cur = 0;
6941 
6942 	now = dev;
6943 	iter = &dev->adj_list.upper;
6944 
6945 	while (1) {
6946 		if (now != dev) {
6947 			ret = fn(now, priv);
6948 			if (ret)
6949 				return ret;
6950 		}
6951 
6952 		next = NULL;
6953 		while (1) {
6954 			udev = netdev_next_upper_dev_rcu(now, &iter);
6955 			if (!udev)
6956 				break;
6957 
6958 			next = udev;
6959 			niter = &udev->adj_list.upper;
6960 			dev_stack[cur] = now;
6961 			iter_stack[cur++] = iter;
6962 			break;
6963 		}
6964 
6965 		if (!next) {
6966 			if (!cur)
6967 				return 0;
6968 			next = dev_stack[--cur];
6969 			niter = iter_stack[cur];
6970 		}
6971 
6972 		now = next;
6973 		iter = niter;
6974 	}
6975 
6976 	return 0;
6977 }
6978 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6979 
6980 static bool __netdev_has_upper_dev(struct net_device *dev,
6981 				   struct net_device *upper_dev)
6982 {
6983 	struct netdev_nested_priv priv = {
6984 		.flags = 0,
6985 		.data = (void *)upper_dev,
6986 	};
6987 
6988 	ASSERT_RTNL();
6989 
6990 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6991 					   &priv);
6992 }
6993 
6994 /**
6995  * netdev_lower_get_next_private - Get the next ->private from the
6996  *				   lower neighbour list
6997  * @dev: device
6998  * @iter: list_head ** of the current position
6999  *
7000  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7001  * list, starting from iter position. The caller must hold either hold the
7002  * RTNL lock or its own locking that guarantees that the neighbour lower
7003  * list will remain unchanged.
7004  */
7005 void *netdev_lower_get_next_private(struct net_device *dev,
7006 				    struct list_head **iter)
7007 {
7008 	struct netdev_adjacent *lower;
7009 
7010 	lower = list_entry(*iter, struct netdev_adjacent, list);
7011 
7012 	if (&lower->list == &dev->adj_list.lower)
7013 		return NULL;
7014 
7015 	*iter = lower->list.next;
7016 
7017 	return lower->private;
7018 }
7019 EXPORT_SYMBOL(netdev_lower_get_next_private);
7020 
7021 /**
7022  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7023  *				       lower neighbour list, RCU
7024  *				       variant
7025  * @dev: device
7026  * @iter: list_head ** of the current position
7027  *
7028  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7029  * list, starting from iter position. The caller must hold RCU read lock.
7030  */
7031 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7032 					struct list_head **iter)
7033 {
7034 	struct netdev_adjacent *lower;
7035 
7036 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7037 
7038 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7039 
7040 	if (&lower->list == &dev->adj_list.lower)
7041 		return NULL;
7042 
7043 	*iter = &lower->list;
7044 
7045 	return lower->private;
7046 }
7047 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7048 
7049 /**
7050  * netdev_lower_get_next - Get the next device from the lower neighbour
7051  *                         list
7052  * @dev: device
7053  * @iter: list_head ** of the current position
7054  *
7055  * Gets the next netdev_adjacent from the dev's lower neighbour
7056  * list, starting from iter position. The caller must hold RTNL lock or
7057  * its own locking that guarantees that the neighbour lower
7058  * list will remain unchanged.
7059  */
7060 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7061 {
7062 	struct netdev_adjacent *lower;
7063 
7064 	lower = list_entry(*iter, struct netdev_adjacent, list);
7065 
7066 	if (&lower->list == &dev->adj_list.lower)
7067 		return NULL;
7068 
7069 	*iter = lower->list.next;
7070 
7071 	return lower->dev;
7072 }
7073 EXPORT_SYMBOL(netdev_lower_get_next);
7074 
7075 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7076 						struct list_head **iter)
7077 {
7078 	struct netdev_adjacent *lower;
7079 
7080 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7081 
7082 	if (&lower->list == &dev->adj_list.lower)
7083 		return NULL;
7084 
7085 	*iter = &lower->list;
7086 
7087 	return lower->dev;
7088 }
7089 
7090 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7091 						  struct list_head **iter,
7092 						  bool *ignore)
7093 {
7094 	struct netdev_adjacent *lower;
7095 
7096 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7097 
7098 	if (&lower->list == &dev->adj_list.lower)
7099 		return NULL;
7100 
7101 	*iter = &lower->list;
7102 	*ignore = lower->ignore;
7103 
7104 	return lower->dev;
7105 }
7106 
7107 int netdev_walk_all_lower_dev(struct net_device *dev,
7108 			      int (*fn)(struct net_device *dev,
7109 					struct netdev_nested_priv *priv),
7110 			      struct netdev_nested_priv *priv)
7111 {
7112 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7113 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7114 	int ret, cur = 0;
7115 
7116 	now = dev;
7117 	iter = &dev->adj_list.lower;
7118 
7119 	while (1) {
7120 		if (now != dev) {
7121 			ret = fn(now, priv);
7122 			if (ret)
7123 				return ret;
7124 		}
7125 
7126 		next = NULL;
7127 		while (1) {
7128 			ldev = netdev_next_lower_dev(now, &iter);
7129 			if (!ldev)
7130 				break;
7131 
7132 			next = ldev;
7133 			niter = &ldev->adj_list.lower;
7134 			dev_stack[cur] = now;
7135 			iter_stack[cur++] = iter;
7136 			break;
7137 		}
7138 
7139 		if (!next) {
7140 			if (!cur)
7141 				return 0;
7142 			next = dev_stack[--cur];
7143 			niter = iter_stack[cur];
7144 		}
7145 
7146 		now = next;
7147 		iter = niter;
7148 	}
7149 
7150 	return 0;
7151 }
7152 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7153 
7154 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7155 				       int (*fn)(struct net_device *dev,
7156 					 struct netdev_nested_priv *priv),
7157 				       struct netdev_nested_priv *priv)
7158 {
7159 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7160 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7161 	int ret, cur = 0;
7162 	bool ignore;
7163 
7164 	now = dev;
7165 	iter = &dev->adj_list.lower;
7166 
7167 	while (1) {
7168 		if (now != dev) {
7169 			ret = fn(now, priv);
7170 			if (ret)
7171 				return ret;
7172 		}
7173 
7174 		next = NULL;
7175 		while (1) {
7176 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7177 			if (!ldev)
7178 				break;
7179 			if (ignore)
7180 				continue;
7181 
7182 			next = ldev;
7183 			niter = &ldev->adj_list.lower;
7184 			dev_stack[cur] = now;
7185 			iter_stack[cur++] = iter;
7186 			break;
7187 		}
7188 
7189 		if (!next) {
7190 			if (!cur)
7191 				return 0;
7192 			next = dev_stack[--cur];
7193 			niter = iter_stack[cur];
7194 		}
7195 
7196 		now = next;
7197 		iter = niter;
7198 	}
7199 
7200 	return 0;
7201 }
7202 
7203 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7204 					     struct list_head **iter)
7205 {
7206 	struct netdev_adjacent *lower;
7207 
7208 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7209 	if (&lower->list == &dev->adj_list.lower)
7210 		return NULL;
7211 
7212 	*iter = &lower->list;
7213 
7214 	return lower->dev;
7215 }
7216 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7217 
7218 static u8 __netdev_upper_depth(struct net_device *dev)
7219 {
7220 	struct net_device *udev;
7221 	struct list_head *iter;
7222 	u8 max_depth = 0;
7223 	bool ignore;
7224 
7225 	for (iter = &dev->adj_list.upper,
7226 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7227 	     udev;
7228 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7229 		if (ignore)
7230 			continue;
7231 		if (max_depth < udev->upper_level)
7232 			max_depth = udev->upper_level;
7233 	}
7234 
7235 	return max_depth;
7236 }
7237 
7238 static u8 __netdev_lower_depth(struct net_device *dev)
7239 {
7240 	struct net_device *ldev;
7241 	struct list_head *iter;
7242 	u8 max_depth = 0;
7243 	bool ignore;
7244 
7245 	for (iter = &dev->adj_list.lower,
7246 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7247 	     ldev;
7248 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7249 		if (ignore)
7250 			continue;
7251 		if (max_depth < ldev->lower_level)
7252 			max_depth = ldev->lower_level;
7253 	}
7254 
7255 	return max_depth;
7256 }
7257 
7258 static int __netdev_update_upper_level(struct net_device *dev,
7259 				       struct netdev_nested_priv *__unused)
7260 {
7261 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7262 	return 0;
7263 }
7264 
7265 #ifdef CONFIG_LOCKDEP
7266 static LIST_HEAD(net_unlink_list);
7267 
7268 static void net_unlink_todo(struct net_device *dev)
7269 {
7270 	if (list_empty(&dev->unlink_list))
7271 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7272 }
7273 #endif
7274 
7275 static int __netdev_update_lower_level(struct net_device *dev,
7276 				       struct netdev_nested_priv *priv)
7277 {
7278 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7279 
7280 #ifdef CONFIG_LOCKDEP
7281 	if (!priv)
7282 		return 0;
7283 
7284 	if (priv->flags & NESTED_SYNC_IMM)
7285 		dev->nested_level = dev->lower_level - 1;
7286 	if (priv->flags & NESTED_SYNC_TODO)
7287 		net_unlink_todo(dev);
7288 #endif
7289 	return 0;
7290 }
7291 
7292 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7293 				  int (*fn)(struct net_device *dev,
7294 					    struct netdev_nested_priv *priv),
7295 				  struct netdev_nested_priv *priv)
7296 {
7297 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7298 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7299 	int ret, cur = 0;
7300 
7301 	now = dev;
7302 	iter = &dev->adj_list.lower;
7303 
7304 	while (1) {
7305 		if (now != dev) {
7306 			ret = fn(now, priv);
7307 			if (ret)
7308 				return ret;
7309 		}
7310 
7311 		next = NULL;
7312 		while (1) {
7313 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7314 			if (!ldev)
7315 				break;
7316 
7317 			next = ldev;
7318 			niter = &ldev->adj_list.lower;
7319 			dev_stack[cur] = now;
7320 			iter_stack[cur++] = iter;
7321 			break;
7322 		}
7323 
7324 		if (!next) {
7325 			if (!cur)
7326 				return 0;
7327 			next = dev_stack[--cur];
7328 			niter = iter_stack[cur];
7329 		}
7330 
7331 		now = next;
7332 		iter = niter;
7333 	}
7334 
7335 	return 0;
7336 }
7337 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7338 
7339 /**
7340  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7341  *				       lower neighbour list, RCU
7342  *				       variant
7343  * @dev: device
7344  *
7345  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7346  * list. The caller must hold RCU read lock.
7347  */
7348 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7349 {
7350 	struct netdev_adjacent *lower;
7351 
7352 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7353 			struct netdev_adjacent, list);
7354 	if (lower)
7355 		return lower->private;
7356 	return NULL;
7357 }
7358 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7359 
7360 /**
7361  * netdev_master_upper_dev_get_rcu - Get master upper device
7362  * @dev: device
7363  *
7364  * Find a master upper device and return pointer to it or NULL in case
7365  * it's not there. The caller must hold the RCU read lock.
7366  */
7367 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7368 {
7369 	struct netdev_adjacent *upper;
7370 
7371 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7372 				       struct netdev_adjacent, list);
7373 	if (upper && likely(upper->master))
7374 		return upper->dev;
7375 	return NULL;
7376 }
7377 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7378 
7379 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7380 			      struct net_device *adj_dev,
7381 			      struct list_head *dev_list)
7382 {
7383 	char linkname[IFNAMSIZ+7];
7384 
7385 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7386 		"upper_%s" : "lower_%s", adj_dev->name);
7387 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7388 				 linkname);
7389 }
7390 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7391 			       char *name,
7392 			       struct list_head *dev_list)
7393 {
7394 	char linkname[IFNAMSIZ+7];
7395 
7396 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7397 		"upper_%s" : "lower_%s", name);
7398 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7399 }
7400 
7401 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7402 						 struct net_device *adj_dev,
7403 						 struct list_head *dev_list)
7404 {
7405 	return (dev_list == &dev->adj_list.upper ||
7406 		dev_list == &dev->adj_list.lower) &&
7407 		net_eq(dev_net(dev), dev_net(adj_dev));
7408 }
7409 
7410 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7411 					struct net_device *adj_dev,
7412 					struct list_head *dev_list,
7413 					void *private, bool master)
7414 {
7415 	struct netdev_adjacent *adj;
7416 	int ret;
7417 
7418 	adj = __netdev_find_adj(adj_dev, dev_list);
7419 
7420 	if (adj) {
7421 		adj->ref_nr += 1;
7422 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7423 			 dev->name, adj_dev->name, adj->ref_nr);
7424 
7425 		return 0;
7426 	}
7427 
7428 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7429 	if (!adj)
7430 		return -ENOMEM;
7431 
7432 	adj->dev = adj_dev;
7433 	adj->master = master;
7434 	adj->ref_nr = 1;
7435 	adj->private = private;
7436 	adj->ignore = false;
7437 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7438 
7439 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7440 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7441 
7442 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7443 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7444 		if (ret)
7445 			goto free_adj;
7446 	}
7447 
7448 	/* Ensure that master link is always the first item in list. */
7449 	if (master) {
7450 		ret = sysfs_create_link(&(dev->dev.kobj),
7451 					&(adj_dev->dev.kobj), "master");
7452 		if (ret)
7453 			goto remove_symlinks;
7454 
7455 		list_add_rcu(&adj->list, dev_list);
7456 	} else {
7457 		list_add_tail_rcu(&adj->list, dev_list);
7458 	}
7459 
7460 	return 0;
7461 
7462 remove_symlinks:
7463 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7464 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7465 free_adj:
7466 	netdev_put(adj_dev, &adj->dev_tracker);
7467 	kfree(adj);
7468 
7469 	return ret;
7470 }
7471 
7472 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7473 					 struct net_device *adj_dev,
7474 					 u16 ref_nr,
7475 					 struct list_head *dev_list)
7476 {
7477 	struct netdev_adjacent *adj;
7478 
7479 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7480 		 dev->name, adj_dev->name, ref_nr);
7481 
7482 	adj = __netdev_find_adj(adj_dev, dev_list);
7483 
7484 	if (!adj) {
7485 		pr_err("Adjacency does not exist for device %s from %s\n",
7486 		       dev->name, adj_dev->name);
7487 		WARN_ON(1);
7488 		return;
7489 	}
7490 
7491 	if (adj->ref_nr > ref_nr) {
7492 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7493 			 dev->name, adj_dev->name, ref_nr,
7494 			 adj->ref_nr - ref_nr);
7495 		adj->ref_nr -= ref_nr;
7496 		return;
7497 	}
7498 
7499 	if (adj->master)
7500 		sysfs_remove_link(&(dev->dev.kobj), "master");
7501 
7502 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7503 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7504 
7505 	list_del_rcu(&adj->list);
7506 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7507 		 adj_dev->name, dev->name, adj_dev->name);
7508 	netdev_put(adj_dev, &adj->dev_tracker);
7509 	kfree_rcu(adj, rcu);
7510 }
7511 
7512 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7513 					    struct net_device *upper_dev,
7514 					    struct list_head *up_list,
7515 					    struct list_head *down_list,
7516 					    void *private, bool master)
7517 {
7518 	int ret;
7519 
7520 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7521 					   private, master);
7522 	if (ret)
7523 		return ret;
7524 
7525 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7526 					   private, false);
7527 	if (ret) {
7528 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7529 		return ret;
7530 	}
7531 
7532 	return 0;
7533 }
7534 
7535 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7536 					       struct net_device *upper_dev,
7537 					       u16 ref_nr,
7538 					       struct list_head *up_list,
7539 					       struct list_head *down_list)
7540 {
7541 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7542 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7543 }
7544 
7545 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7546 						struct net_device *upper_dev,
7547 						void *private, bool master)
7548 {
7549 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7550 						&dev->adj_list.upper,
7551 						&upper_dev->adj_list.lower,
7552 						private, master);
7553 }
7554 
7555 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7556 						   struct net_device *upper_dev)
7557 {
7558 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7559 					   &dev->adj_list.upper,
7560 					   &upper_dev->adj_list.lower);
7561 }
7562 
7563 static int __netdev_upper_dev_link(struct net_device *dev,
7564 				   struct net_device *upper_dev, bool master,
7565 				   void *upper_priv, void *upper_info,
7566 				   struct netdev_nested_priv *priv,
7567 				   struct netlink_ext_ack *extack)
7568 {
7569 	struct netdev_notifier_changeupper_info changeupper_info = {
7570 		.info = {
7571 			.dev = dev,
7572 			.extack = extack,
7573 		},
7574 		.upper_dev = upper_dev,
7575 		.master = master,
7576 		.linking = true,
7577 		.upper_info = upper_info,
7578 	};
7579 	struct net_device *master_dev;
7580 	int ret = 0;
7581 
7582 	ASSERT_RTNL();
7583 
7584 	if (dev == upper_dev)
7585 		return -EBUSY;
7586 
7587 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7588 	if (__netdev_has_upper_dev(upper_dev, dev))
7589 		return -EBUSY;
7590 
7591 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7592 		return -EMLINK;
7593 
7594 	if (!master) {
7595 		if (__netdev_has_upper_dev(dev, upper_dev))
7596 			return -EEXIST;
7597 	} else {
7598 		master_dev = __netdev_master_upper_dev_get(dev);
7599 		if (master_dev)
7600 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7601 	}
7602 
7603 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7604 					    &changeupper_info.info);
7605 	ret = notifier_to_errno(ret);
7606 	if (ret)
7607 		return ret;
7608 
7609 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7610 						   master);
7611 	if (ret)
7612 		return ret;
7613 
7614 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7615 					    &changeupper_info.info);
7616 	ret = notifier_to_errno(ret);
7617 	if (ret)
7618 		goto rollback;
7619 
7620 	__netdev_update_upper_level(dev, NULL);
7621 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7622 
7623 	__netdev_update_lower_level(upper_dev, priv);
7624 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7625 				    priv);
7626 
7627 	return 0;
7628 
7629 rollback:
7630 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7631 
7632 	return ret;
7633 }
7634 
7635 /**
7636  * netdev_upper_dev_link - Add a link to the upper device
7637  * @dev: device
7638  * @upper_dev: new upper device
7639  * @extack: netlink extended ack
7640  *
7641  * Adds a link to device which is upper to this one. The caller must hold
7642  * the RTNL lock. On a failure a negative errno code is returned.
7643  * On success the reference counts are adjusted and the function
7644  * returns zero.
7645  */
7646 int netdev_upper_dev_link(struct net_device *dev,
7647 			  struct net_device *upper_dev,
7648 			  struct netlink_ext_ack *extack)
7649 {
7650 	struct netdev_nested_priv priv = {
7651 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7652 		.data = NULL,
7653 	};
7654 
7655 	return __netdev_upper_dev_link(dev, upper_dev, false,
7656 				       NULL, NULL, &priv, extack);
7657 }
7658 EXPORT_SYMBOL(netdev_upper_dev_link);
7659 
7660 /**
7661  * netdev_master_upper_dev_link - Add a master link to the upper device
7662  * @dev: device
7663  * @upper_dev: new upper device
7664  * @upper_priv: upper device private
7665  * @upper_info: upper info to be passed down via notifier
7666  * @extack: netlink extended ack
7667  *
7668  * Adds a link to device which is upper to this one. In this case, only
7669  * one master upper device can be linked, although other non-master devices
7670  * might be linked as well. The caller must hold the RTNL lock.
7671  * On a failure a negative errno code is returned. On success the reference
7672  * counts are adjusted and the function returns zero.
7673  */
7674 int netdev_master_upper_dev_link(struct net_device *dev,
7675 				 struct net_device *upper_dev,
7676 				 void *upper_priv, void *upper_info,
7677 				 struct netlink_ext_ack *extack)
7678 {
7679 	struct netdev_nested_priv priv = {
7680 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7681 		.data = NULL,
7682 	};
7683 
7684 	return __netdev_upper_dev_link(dev, upper_dev, true,
7685 				       upper_priv, upper_info, &priv, extack);
7686 }
7687 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7688 
7689 static void __netdev_upper_dev_unlink(struct net_device *dev,
7690 				      struct net_device *upper_dev,
7691 				      struct netdev_nested_priv *priv)
7692 {
7693 	struct netdev_notifier_changeupper_info changeupper_info = {
7694 		.info = {
7695 			.dev = dev,
7696 		},
7697 		.upper_dev = upper_dev,
7698 		.linking = false,
7699 	};
7700 
7701 	ASSERT_RTNL();
7702 
7703 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7704 
7705 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7706 				      &changeupper_info.info);
7707 
7708 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7709 
7710 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7711 				      &changeupper_info.info);
7712 
7713 	__netdev_update_upper_level(dev, NULL);
7714 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7715 
7716 	__netdev_update_lower_level(upper_dev, priv);
7717 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7718 				    priv);
7719 }
7720 
7721 /**
7722  * netdev_upper_dev_unlink - Removes a link to upper device
7723  * @dev: device
7724  * @upper_dev: new upper device
7725  *
7726  * Removes a link to device which is upper to this one. The caller must hold
7727  * the RTNL lock.
7728  */
7729 void netdev_upper_dev_unlink(struct net_device *dev,
7730 			     struct net_device *upper_dev)
7731 {
7732 	struct netdev_nested_priv priv = {
7733 		.flags = NESTED_SYNC_TODO,
7734 		.data = NULL,
7735 	};
7736 
7737 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
7738 }
7739 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7740 
7741 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7742 				      struct net_device *lower_dev,
7743 				      bool val)
7744 {
7745 	struct netdev_adjacent *adj;
7746 
7747 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7748 	if (adj)
7749 		adj->ignore = val;
7750 
7751 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7752 	if (adj)
7753 		adj->ignore = val;
7754 }
7755 
7756 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7757 					struct net_device *lower_dev)
7758 {
7759 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7760 }
7761 
7762 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7763 				       struct net_device *lower_dev)
7764 {
7765 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7766 }
7767 
7768 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7769 				   struct net_device *new_dev,
7770 				   struct net_device *dev,
7771 				   struct netlink_ext_ack *extack)
7772 {
7773 	struct netdev_nested_priv priv = {
7774 		.flags = 0,
7775 		.data = NULL,
7776 	};
7777 	int err;
7778 
7779 	if (!new_dev)
7780 		return 0;
7781 
7782 	if (old_dev && new_dev != old_dev)
7783 		netdev_adjacent_dev_disable(dev, old_dev);
7784 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7785 				      extack);
7786 	if (err) {
7787 		if (old_dev && new_dev != old_dev)
7788 			netdev_adjacent_dev_enable(dev, old_dev);
7789 		return err;
7790 	}
7791 
7792 	return 0;
7793 }
7794 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7795 
7796 void netdev_adjacent_change_commit(struct net_device *old_dev,
7797 				   struct net_device *new_dev,
7798 				   struct net_device *dev)
7799 {
7800 	struct netdev_nested_priv priv = {
7801 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7802 		.data = NULL,
7803 	};
7804 
7805 	if (!new_dev || !old_dev)
7806 		return;
7807 
7808 	if (new_dev == old_dev)
7809 		return;
7810 
7811 	netdev_adjacent_dev_enable(dev, old_dev);
7812 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
7813 }
7814 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7815 
7816 void netdev_adjacent_change_abort(struct net_device *old_dev,
7817 				  struct net_device *new_dev,
7818 				  struct net_device *dev)
7819 {
7820 	struct netdev_nested_priv priv = {
7821 		.flags = 0,
7822 		.data = NULL,
7823 	};
7824 
7825 	if (!new_dev)
7826 		return;
7827 
7828 	if (old_dev && new_dev != old_dev)
7829 		netdev_adjacent_dev_enable(dev, old_dev);
7830 
7831 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
7832 }
7833 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7834 
7835 /**
7836  * netdev_bonding_info_change - Dispatch event about slave change
7837  * @dev: device
7838  * @bonding_info: info to dispatch
7839  *
7840  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7841  * The caller must hold the RTNL lock.
7842  */
7843 void netdev_bonding_info_change(struct net_device *dev,
7844 				struct netdev_bonding_info *bonding_info)
7845 {
7846 	struct netdev_notifier_bonding_info info = {
7847 		.info.dev = dev,
7848 	};
7849 
7850 	memcpy(&info.bonding_info, bonding_info,
7851 	       sizeof(struct netdev_bonding_info));
7852 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7853 				      &info.info);
7854 }
7855 EXPORT_SYMBOL(netdev_bonding_info_change);
7856 
7857 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7858 					   struct netlink_ext_ack *extack)
7859 {
7860 	struct netdev_notifier_offload_xstats_info info = {
7861 		.info.dev = dev,
7862 		.info.extack = extack,
7863 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7864 	};
7865 	int err;
7866 	int rc;
7867 
7868 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7869 					 GFP_KERNEL);
7870 	if (!dev->offload_xstats_l3)
7871 		return -ENOMEM;
7872 
7873 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7874 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
7875 						  &info.info);
7876 	err = notifier_to_errno(rc);
7877 	if (err)
7878 		goto free_stats;
7879 
7880 	return 0;
7881 
7882 free_stats:
7883 	kfree(dev->offload_xstats_l3);
7884 	dev->offload_xstats_l3 = NULL;
7885 	return err;
7886 }
7887 
7888 int netdev_offload_xstats_enable(struct net_device *dev,
7889 				 enum netdev_offload_xstats_type type,
7890 				 struct netlink_ext_ack *extack)
7891 {
7892 	ASSERT_RTNL();
7893 
7894 	if (netdev_offload_xstats_enabled(dev, type))
7895 		return -EALREADY;
7896 
7897 	switch (type) {
7898 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7899 		return netdev_offload_xstats_enable_l3(dev, extack);
7900 	}
7901 
7902 	WARN_ON(1);
7903 	return -EINVAL;
7904 }
7905 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7906 
7907 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7908 {
7909 	struct netdev_notifier_offload_xstats_info info = {
7910 		.info.dev = dev,
7911 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7912 	};
7913 
7914 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7915 				      &info.info);
7916 	kfree(dev->offload_xstats_l3);
7917 	dev->offload_xstats_l3 = NULL;
7918 }
7919 
7920 int netdev_offload_xstats_disable(struct net_device *dev,
7921 				  enum netdev_offload_xstats_type type)
7922 {
7923 	ASSERT_RTNL();
7924 
7925 	if (!netdev_offload_xstats_enabled(dev, type))
7926 		return -EALREADY;
7927 
7928 	switch (type) {
7929 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7930 		netdev_offload_xstats_disable_l3(dev);
7931 		return 0;
7932 	}
7933 
7934 	WARN_ON(1);
7935 	return -EINVAL;
7936 }
7937 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7938 
7939 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7940 {
7941 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7942 }
7943 
7944 static struct rtnl_hw_stats64 *
7945 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7946 			      enum netdev_offload_xstats_type type)
7947 {
7948 	switch (type) {
7949 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7950 		return dev->offload_xstats_l3;
7951 	}
7952 
7953 	WARN_ON(1);
7954 	return NULL;
7955 }
7956 
7957 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7958 				   enum netdev_offload_xstats_type type)
7959 {
7960 	ASSERT_RTNL();
7961 
7962 	return netdev_offload_xstats_get_ptr(dev, type);
7963 }
7964 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
7965 
7966 struct netdev_notifier_offload_xstats_ru {
7967 	bool used;
7968 };
7969 
7970 struct netdev_notifier_offload_xstats_rd {
7971 	struct rtnl_hw_stats64 stats;
7972 	bool used;
7973 };
7974 
7975 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
7976 				  const struct rtnl_hw_stats64 *src)
7977 {
7978 	dest->rx_packets	  += src->rx_packets;
7979 	dest->tx_packets	  += src->tx_packets;
7980 	dest->rx_bytes		  += src->rx_bytes;
7981 	dest->tx_bytes		  += src->tx_bytes;
7982 	dest->rx_errors		  += src->rx_errors;
7983 	dest->tx_errors		  += src->tx_errors;
7984 	dest->rx_dropped	  += src->rx_dropped;
7985 	dest->tx_dropped	  += src->tx_dropped;
7986 	dest->multicast		  += src->multicast;
7987 }
7988 
7989 static int netdev_offload_xstats_get_used(struct net_device *dev,
7990 					  enum netdev_offload_xstats_type type,
7991 					  bool *p_used,
7992 					  struct netlink_ext_ack *extack)
7993 {
7994 	struct netdev_notifier_offload_xstats_ru report_used = {};
7995 	struct netdev_notifier_offload_xstats_info info = {
7996 		.info.dev = dev,
7997 		.info.extack = extack,
7998 		.type = type,
7999 		.report_used = &report_used,
8000 	};
8001 	int rc;
8002 
8003 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8004 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8005 					   &info.info);
8006 	*p_used = report_used.used;
8007 	return notifier_to_errno(rc);
8008 }
8009 
8010 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8011 					   enum netdev_offload_xstats_type type,
8012 					   struct rtnl_hw_stats64 *p_stats,
8013 					   bool *p_used,
8014 					   struct netlink_ext_ack *extack)
8015 {
8016 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8017 	struct netdev_notifier_offload_xstats_info info = {
8018 		.info.dev = dev,
8019 		.info.extack = extack,
8020 		.type = type,
8021 		.report_delta = &report_delta,
8022 	};
8023 	struct rtnl_hw_stats64 *stats;
8024 	int rc;
8025 
8026 	stats = netdev_offload_xstats_get_ptr(dev, type);
8027 	if (WARN_ON(!stats))
8028 		return -EINVAL;
8029 
8030 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8031 					   &info.info);
8032 
8033 	/* Cache whatever we got, even if there was an error, otherwise the
8034 	 * successful stats retrievals would get lost.
8035 	 */
8036 	netdev_hw_stats64_add(stats, &report_delta.stats);
8037 
8038 	if (p_stats)
8039 		*p_stats = *stats;
8040 	*p_used = report_delta.used;
8041 
8042 	return notifier_to_errno(rc);
8043 }
8044 
8045 int netdev_offload_xstats_get(struct net_device *dev,
8046 			      enum netdev_offload_xstats_type type,
8047 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8048 			      struct netlink_ext_ack *extack)
8049 {
8050 	ASSERT_RTNL();
8051 
8052 	if (p_stats)
8053 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8054 						       p_used, extack);
8055 	else
8056 		return netdev_offload_xstats_get_used(dev, type, p_used,
8057 						      extack);
8058 }
8059 EXPORT_SYMBOL(netdev_offload_xstats_get);
8060 
8061 void
8062 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8063 				   const struct rtnl_hw_stats64 *stats)
8064 {
8065 	report_delta->used = true;
8066 	netdev_hw_stats64_add(&report_delta->stats, stats);
8067 }
8068 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8069 
8070 void
8071 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8072 {
8073 	report_used->used = true;
8074 }
8075 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8076 
8077 void netdev_offload_xstats_push_delta(struct net_device *dev,
8078 				      enum netdev_offload_xstats_type type,
8079 				      const struct rtnl_hw_stats64 *p_stats)
8080 {
8081 	struct rtnl_hw_stats64 *stats;
8082 
8083 	ASSERT_RTNL();
8084 
8085 	stats = netdev_offload_xstats_get_ptr(dev, type);
8086 	if (WARN_ON(!stats))
8087 		return;
8088 
8089 	netdev_hw_stats64_add(stats, p_stats);
8090 }
8091 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8092 
8093 /**
8094  * netdev_get_xmit_slave - Get the xmit slave of master device
8095  * @dev: device
8096  * @skb: The packet
8097  * @all_slaves: assume all the slaves are active
8098  *
8099  * The reference counters are not incremented so the caller must be
8100  * careful with locks. The caller must hold RCU lock.
8101  * %NULL is returned if no slave is found.
8102  */
8103 
8104 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8105 					 struct sk_buff *skb,
8106 					 bool all_slaves)
8107 {
8108 	const struct net_device_ops *ops = dev->netdev_ops;
8109 
8110 	if (!ops->ndo_get_xmit_slave)
8111 		return NULL;
8112 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8113 }
8114 EXPORT_SYMBOL(netdev_get_xmit_slave);
8115 
8116 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8117 						  struct sock *sk)
8118 {
8119 	const struct net_device_ops *ops = dev->netdev_ops;
8120 
8121 	if (!ops->ndo_sk_get_lower_dev)
8122 		return NULL;
8123 	return ops->ndo_sk_get_lower_dev(dev, sk);
8124 }
8125 
8126 /**
8127  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8128  * @dev: device
8129  * @sk: the socket
8130  *
8131  * %NULL is returned if no lower device is found.
8132  */
8133 
8134 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8135 					    struct sock *sk)
8136 {
8137 	struct net_device *lower;
8138 
8139 	lower = netdev_sk_get_lower_dev(dev, sk);
8140 	while (lower) {
8141 		dev = lower;
8142 		lower = netdev_sk_get_lower_dev(dev, sk);
8143 	}
8144 
8145 	return dev;
8146 }
8147 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8148 
8149 static void netdev_adjacent_add_links(struct net_device *dev)
8150 {
8151 	struct netdev_adjacent *iter;
8152 
8153 	struct net *net = dev_net(dev);
8154 
8155 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8156 		if (!net_eq(net, dev_net(iter->dev)))
8157 			continue;
8158 		netdev_adjacent_sysfs_add(iter->dev, dev,
8159 					  &iter->dev->adj_list.lower);
8160 		netdev_adjacent_sysfs_add(dev, iter->dev,
8161 					  &dev->adj_list.upper);
8162 	}
8163 
8164 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8165 		if (!net_eq(net, dev_net(iter->dev)))
8166 			continue;
8167 		netdev_adjacent_sysfs_add(iter->dev, dev,
8168 					  &iter->dev->adj_list.upper);
8169 		netdev_adjacent_sysfs_add(dev, iter->dev,
8170 					  &dev->adj_list.lower);
8171 	}
8172 }
8173 
8174 static void netdev_adjacent_del_links(struct net_device *dev)
8175 {
8176 	struct netdev_adjacent *iter;
8177 
8178 	struct net *net = dev_net(dev);
8179 
8180 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8181 		if (!net_eq(net, dev_net(iter->dev)))
8182 			continue;
8183 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8184 					  &iter->dev->adj_list.lower);
8185 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8186 					  &dev->adj_list.upper);
8187 	}
8188 
8189 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8190 		if (!net_eq(net, dev_net(iter->dev)))
8191 			continue;
8192 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8193 					  &iter->dev->adj_list.upper);
8194 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8195 					  &dev->adj_list.lower);
8196 	}
8197 }
8198 
8199 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8200 {
8201 	struct netdev_adjacent *iter;
8202 
8203 	struct net *net = dev_net(dev);
8204 
8205 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8206 		if (!net_eq(net, dev_net(iter->dev)))
8207 			continue;
8208 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8209 					  &iter->dev->adj_list.lower);
8210 		netdev_adjacent_sysfs_add(iter->dev, dev,
8211 					  &iter->dev->adj_list.lower);
8212 	}
8213 
8214 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8215 		if (!net_eq(net, dev_net(iter->dev)))
8216 			continue;
8217 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8218 					  &iter->dev->adj_list.upper);
8219 		netdev_adjacent_sysfs_add(iter->dev, dev,
8220 					  &iter->dev->adj_list.upper);
8221 	}
8222 }
8223 
8224 void *netdev_lower_dev_get_private(struct net_device *dev,
8225 				   struct net_device *lower_dev)
8226 {
8227 	struct netdev_adjacent *lower;
8228 
8229 	if (!lower_dev)
8230 		return NULL;
8231 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8232 	if (!lower)
8233 		return NULL;
8234 
8235 	return lower->private;
8236 }
8237 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8238 
8239 
8240 /**
8241  * netdev_lower_state_changed - Dispatch event about lower device state change
8242  * @lower_dev: device
8243  * @lower_state_info: state to dispatch
8244  *
8245  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8246  * The caller must hold the RTNL lock.
8247  */
8248 void netdev_lower_state_changed(struct net_device *lower_dev,
8249 				void *lower_state_info)
8250 {
8251 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8252 		.info.dev = lower_dev,
8253 	};
8254 
8255 	ASSERT_RTNL();
8256 	changelowerstate_info.lower_state_info = lower_state_info;
8257 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8258 				      &changelowerstate_info.info);
8259 }
8260 EXPORT_SYMBOL(netdev_lower_state_changed);
8261 
8262 static void dev_change_rx_flags(struct net_device *dev, int flags)
8263 {
8264 	const struct net_device_ops *ops = dev->netdev_ops;
8265 
8266 	if (ops->ndo_change_rx_flags)
8267 		ops->ndo_change_rx_flags(dev, flags);
8268 }
8269 
8270 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8271 {
8272 	unsigned int old_flags = dev->flags;
8273 	kuid_t uid;
8274 	kgid_t gid;
8275 
8276 	ASSERT_RTNL();
8277 
8278 	dev->flags |= IFF_PROMISC;
8279 	dev->promiscuity += inc;
8280 	if (dev->promiscuity == 0) {
8281 		/*
8282 		 * Avoid overflow.
8283 		 * If inc causes overflow, untouch promisc and return error.
8284 		 */
8285 		if (inc < 0)
8286 			dev->flags &= ~IFF_PROMISC;
8287 		else {
8288 			dev->promiscuity -= inc;
8289 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8290 			return -EOVERFLOW;
8291 		}
8292 	}
8293 	if (dev->flags != old_flags) {
8294 		netdev_info(dev, "%s promiscuous mode\n",
8295 			    dev->flags & IFF_PROMISC ? "entered" : "left");
8296 		if (audit_enabled) {
8297 			current_uid_gid(&uid, &gid);
8298 			audit_log(audit_context(), GFP_ATOMIC,
8299 				  AUDIT_ANOM_PROMISCUOUS,
8300 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8301 				  dev->name, (dev->flags & IFF_PROMISC),
8302 				  (old_flags & IFF_PROMISC),
8303 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8304 				  from_kuid(&init_user_ns, uid),
8305 				  from_kgid(&init_user_ns, gid),
8306 				  audit_get_sessionid(current));
8307 		}
8308 
8309 		dev_change_rx_flags(dev, IFF_PROMISC);
8310 	}
8311 	if (notify)
8312 		__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8313 	return 0;
8314 }
8315 
8316 /**
8317  *	dev_set_promiscuity	- update promiscuity count on a device
8318  *	@dev: device
8319  *	@inc: modifier
8320  *
8321  *	Add or remove promiscuity from a device. While the count in the device
8322  *	remains above zero the interface remains promiscuous. Once it hits zero
8323  *	the device reverts back to normal filtering operation. A negative inc
8324  *	value is used to drop promiscuity on the device.
8325  *	Return 0 if successful or a negative errno code on error.
8326  */
8327 int dev_set_promiscuity(struct net_device *dev, int inc)
8328 {
8329 	unsigned int old_flags = dev->flags;
8330 	int err;
8331 
8332 	err = __dev_set_promiscuity(dev, inc, true);
8333 	if (err < 0)
8334 		return err;
8335 	if (dev->flags != old_flags)
8336 		dev_set_rx_mode(dev);
8337 	return err;
8338 }
8339 EXPORT_SYMBOL(dev_set_promiscuity);
8340 
8341 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8342 {
8343 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8344 
8345 	ASSERT_RTNL();
8346 
8347 	dev->flags |= IFF_ALLMULTI;
8348 	dev->allmulti += inc;
8349 	if (dev->allmulti == 0) {
8350 		/*
8351 		 * Avoid overflow.
8352 		 * If inc causes overflow, untouch allmulti and return error.
8353 		 */
8354 		if (inc < 0)
8355 			dev->flags &= ~IFF_ALLMULTI;
8356 		else {
8357 			dev->allmulti -= inc;
8358 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8359 			return -EOVERFLOW;
8360 		}
8361 	}
8362 	if (dev->flags ^ old_flags) {
8363 		netdev_info(dev, "%s allmulticast mode\n",
8364 			    dev->flags & IFF_ALLMULTI ? "entered" : "left");
8365 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8366 		dev_set_rx_mode(dev);
8367 		if (notify)
8368 			__dev_notify_flags(dev, old_flags,
8369 					   dev->gflags ^ old_gflags, 0, NULL);
8370 	}
8371 	return 0;
8372 }
8373 
8374 /**
8375  *	dev_set_allmulti	- update allmulti count on a device
8376  *	@dev: device
8377  *	@inc: modifier
8378  *
8379  *	Add or remove reception of all multicast frames to a device. While the
8380  *	count in the device remains above zero the interface remains listening
8381  *	to all interfaces. Once it hits zero the device reverts back to normal
8382  *	filtering operation. A negative @inc value is used to drop the counter
8383  *	when releasing a resource needing all multicasts.
8384  *	Return 0 if successful or a negative errno code on error.
8385  */
8386 
8387 int dev_set_allmulti(struct net_device *dev, int inc)
8388 {
8389 	return __dev_set_allmulti(dev, inc, true);
8390 }
8391 EXPORT_SYMBOL(dev_set_allmulti);
8392 
8393 /*
8394  *	Upload unicast and multicast address lists to device and
8395  *	configure RX filtering. When the device doesn't support unicast
8396  *	filtering it is put in promiscuous mode while unicast addresses
8397  *	are present.
8398  */
8399 void __dev_set_rx_mode(struct net_device *dev)
8400 {
8401 	const struct net_device_ops *ops = dev->netdev_ops;
8402 
8403 	/* dev_open will call this function so the list will stay sane. */
8404 	if (!(dev->flags&IFF_UP))
8405 		return;
8406 
8407 	if (!netif_device_present(dev))
8408 		return;
8409 
8410 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8411 		/* Unicast addresses changes may only happen under the rtnl,
8412 		 * therefore calling __dev_set_promiscuity here is safe.
8413 		 */
8414 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8415 			__dev_set_promiscuity(dev, 1, false);
8416 			dev->uc_promisc = true;
8417 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8418 			__dev_set_promiscuity(dev, -1, false);
8419 			dev->uc_promisc = false;
8420 		}
8421 	}
8422 
8423 	if (ops->ndo_set_rx_mode)
8424 		ops->ndo_set_rx_mode(dev);
8425 }
8426 
8427 void dev_set_rx_mode(struct net_device *dev)
8428 {
8429 	netif_addr_lock_bh(dev);
8430 	__dev_set_rx_mode(dev);
8431 	netif_addr_unlock_bh(dev);
8432 }
8433 
8434 /**
8435  *	dev_get_flags - get flags reported to userspace
8436  *	@dev: device
8437  *
8438  *	Get the combination of flag bits exported through APIs to userspace.
8439  */
8440 unsigned int dev_get_flags(const struct net_device *dev)
8441 {
8442 	unsigned int flags;
8443 
8444 	flags = (dev->flags & ~(IFF_PROMISC |
8445 				IFF_ALLMULTI |
8446 				IFF_RUNNING |
8447 				IFF_LOWER_UP |
8448 				IFF_DORMANT)) |
8449 		(dev->gflags & (IFF_PROMISC |
8450 				IFF_ALLMULTI));
8451 
8452 	if (netif_running(dev)) {
8453 		if (netif_oper_up(dev))
8454 			flags |= IFF_RUNNING;
8455 		if (netif_carrier_ok(dev))
8456 			flags |= IFF_LOWER_UP;
8457 		if (netif_dormant(dev))
8458 			flags |= IFF_DORMANT;
8459 	}
8460 
8461 	return flags;
8462 }
8463 EXPORT_SYMBOL(dev_get_flags);
8464 
8465 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8466 		       struct netlink_ext_ack *extack)
8467 {
8468 	unsigned int old_flags = dev->flags;
8469 	int ret;
8470 
8471 	ASSERT_RTNL();
8472 
8473 	/*
8474 	 *	Set the flags on our device.
8475 	 */
8476 
8477 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8478 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8479 			       IFF_AUTOMEDIA)) |
8480 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8481 				    IFF_ALLMULTI));
8482 
8483 	/*
8484 	 *	Load in the correct multicast list now the flags have changed.
8485 	 */
8486 
8487 	if ((old_flags ^ flags) & IFF_MULTICAST)
8488 		dev_change_rx_flags(dev, IFF_MULTICAST);
8489 
8490 	dev_set_rx_mode(dev);
8491 
8492 	/*
8493 	 *	Have we downed the interface. We handle IFF_UP ourselves
8494 	 *	according to user attempts to set it, rather than blindly
8495 	 *	setting it.
8496 	 */
8497 
8498 	ret = 0;
8499 	if ((old_flags ^ flags) & IFF_UP) {
8500 		if (old_flags & IFF_UP)
8501 			__dev_close(dev);
8502 		else
8503 			ret = __dev_open(dev, extack);
8504 	}
8505 
8506 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8507 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8508 		unsigned int old_flags = dev->flags;
8509 
8510 		dev->gflags ^= IFF_PROMISC;
8511 
8512 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8513 			if (dev->flags != old_flags)
8514 				dev_set_rx_mode(dev);
8515 	}
8516 
8517 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8518 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8519 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8520 	 */
8521 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8522 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8523 
8524 		dev->gflags ^= IFF_ALLMULTI;
8525 		__dev_set_allmulti(dev, inc, false);
8526 	}
8527 
8528 	return ret;
8529 }
8530 
8531 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8532 			unsigned int gchanges, u32 portid,
8533 			const struct nlmsghdr *nlh)
8534 {
8535 	unsigned int changes = dev->flags ^ old_flags;
8536 
8537 	if (gchanges)
8538 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8539 
8540 	if (changes & IFF_UP) {
8541 		if (dev->flags & IFF_UP)
8542 			call_netdevice_notifiers(NETDEV_UP, dev);
8543 		else
8544 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8545 	}
8546 
8547 	if (dev->flags & IFF_UP &&
8548 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8549 		struct netdev_notifier_change_info change_info = {
8550 			.info = {
8551 				.dev = dev,
8552 			},
8553 			.flags_changed = changes,
8554 		};
8555 
8556 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8557 	}
8558 }
8559 
8560 /**
8561  *	dev_change_flags - change device settings
8562  *	@dev: device
8563  *	@flags: device state flags
8564  *	@extack: netlink extended ack
8565  *
8566  *	Change settings on device based state flags. The flags are
8567  *	in the userspace exported format.
8568  */
8569 int dev_change_flags(struct net_device *dev, unsigned int flags,
8570 		     struct netlink_ext_ack *extack)
8571 {
8572 	int ret;
8573 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8574 
8575 	ret = __dev_change_flags(dev, flags, extack);
8576 	if (ret < 0)
8577 		return ret;
8578 
8579 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8580 	__dev_notify_flags(dev, old_flags, changes, 0, NULL);
8581 	return ret;
8582 }
8583 EXPORT_SYMBOL(dev_change_flags);
8584 
8585 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8586 {
8587 	const struct net_device_ops *ops = dev->netdev_ops;
8588 
8589 	if (ops->ndo_change_mtu)
8590 		return ops->ndo_change_mtu(dev, new_mtu);
8591 
8592 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8593 	WRITE_ONCE(dev->mtu, new_mtu);
8594 	return 0;
8595 }
8596 EXPORT_SYMBOL(__dev_set_mtu);
8597 
8598 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8599 		     struct netlink_ext_ack *extack)
8600 {
8601 	/* MTU must be positive, and in range */
8602 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8603 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8604 		return -EINVAL;
8605 	}
8606 
8607 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8608 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8609 		return -EINVAL;
8610 	}
8611 	return 0;
8612 }
8613 
8614 /**
8615  *	dev_set_mtu_ext - Change maximum transfer unit
8616  *	@dev: device
8617  *	@new_mtu: new transfer unit
8618  *	@extack: netlink extended ack
8619  *
8620  *	Change the maximum transfer size of the network device.
8621  */
8622 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8623 		    struct netlink_ext_ack *extack)
8624 {
8625 	int err, orig_mtu;
8626 
8627 	if (new_mtu == dev->mtu)
8628 		return 0;
8629 
8630 	err = dev_validate_mtu(dev, new_mtu, extack);
8631 	if (err)
8632 		return err;
8633 
8634 	if (!netif_device_present(dev))
8635 		return -ENODEV;
8636 
8637 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8638 	err = notifier_to_errno(err);
8639 	if (err)
8640 		return err;
8641 
8642 	orig_mtu = dev->mtu;
8643 	err = __dev_set_mtu(dev, new_mtu);
8644 
8645 	if (!err) {
8646 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8647 						   orig_mtu);
8648 		err = notifier_to_errno(err);
8649 		if (err) {
8650 			/* setting mtu back and notifying everyone again,
8651 			 * so that they have a chance to revert changes.
8652 			 */
8653 			__dev_set_mtu(dev, orig_mtu);
8654 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8655 						     new_mtu);
8656 		}
8657 	}
8658 	return err;
8659 }
8660 
8661 int dev_set_mtu(struct net_device *dev, int new_mtu)
8662 {
8663 	struct netlink_ext_ack extack;
8664 	int err;
8665 
8666 	memset(&extack, 0, sizeof(extack));
8667 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8668 	if (err && extack._msg)
8669 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8670 	return err;
8671 }
8672 EXPORT_SYMBOL(dev_set_mtu);
8673 
8674 /**
8675  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8676  *	@dev: device
8677  *	@new_len: new tx queue length
8678  */
8679 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8680 {
8681 	unsigned int orig_len = dev->tx_queue_len;
8682 	int res;
8683 
8684 	if (new_len != (unsigned int)new_len)
8685 		return -ERANGE;
8686 
8687 	if (new_len != orig_len) {
8688 		dev->tx_queue_len = new_len;
8689 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8690 		res = notifier_to_errno(res);
8691 		if (res)
8692 			goto err_rollback;
8693 		res = dev_qdisc_change_tx_queue_len(dev);
8694 		if (res)
8695 			goto err_rollback;
8696 	}
8697 
8698 	return 0;
8699 
8700 err_rollback:
8701 	netdev_err(dev, "refused to change device tx_queue_len\n");
8702 	dev->tx_queue_len = orig_len;
8703 	return res;
8704 }
8705 
8706 /**
8707  *	dev_set_group - Change group this device belongs to
8708  *	@dev: device
8709  *	@new_group: group this device should belong to
8710  */
8711 void dev_set_group(struct net_device *dev, int new_group)
8712 {
8713 	dev->group = new_group;
8714 }
8715 
8716 /**
8717  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8718  *	@dev: device
8719  *	@addr: new address
8720  *	@extack: netlink extended ack
8721  */
8722 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8723 			      struct netlink_ext_ack *extack)
8724 {
8725 	struct netdev_notifier_pre_changeaddr_info info = {
8726 		.info.dev = dev,
8727 		.info.extack = extack,
8728 		.dev_addr = addr,
8729 	};
8730 	int rc;
8731 
8732 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8733 	return notifier_to_errno(rc);
8734 }
8735 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8736 
8737 /**
8738  *	dev_set_mac_address - Change Media Access Control Address
8739  *	@dev: device
8740  *	@sa: new address
8741  *	@extack: netlink extended ack
8742  *
8743  *	Change the hardware (MAC) address of the device
8744  */
8745 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8746 			struct netlink_ext_ack *extack)
8747 {
8748 	const struct net_device_ops *ops = dev->netdev_ops;
8749 	int err;
8750 
8751 	if (!ops->ndo_set_mac_address)
8752 		return -EOPNOTSUPP;
8753 	if (sa->sa_family != dev->type)
8754 		return -EINVAL;
8755 	if (!netif_device_present(dev))
8756 		return -ENODEV;
8757 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8758 	if (err)
8759 		return err;
8760 	err = ops->ndo_set_mac_address(dev, sa);
8761 	if (err)
8762 		return err;
8763 	dev->addr_assign_type = NET_ADDR_SET;
8764 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8765 	add_device_randomness(dev->dev_addr, dev->addr_len);
8766 	return 0;
8767 }
8768 EXPORT_SYMBOL(dev_set_mac_address);
8769 
8770 static DECLARE_RWSEM(dev_addr_sem);
8771 
8772 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8773 			     struct netlink_ext_ack *extack)
8774 {
8775 	int ret;
8776 
8777 	down_write(&dev_addr_sem);
8778 	ret = dev_set_mac_address(dev, sa, extack);
8779 	up_write(&dev_addr_sem);
8780 	return ret;
8781 }
8782 EXPORT_SYMBOL(dev_set_mac_address_user);
8783 
8784 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8785 {
8786 	size_t size = sizeof(sa->sa_data_min);
8787 	struct net_device *dev;
8788 	int ret = 0;
8789 
8790 	down_read(&dev_addr_sem);
8791 	rcu_read_lock();
8792 
8793 	dev = dev_get_by_name_rcu(net, dev_name);
8794 	if (!dev) {
8795 		ret = -ENODEV;
8796 		goto unlock;
8797 	}
8798 	if (!dev->addr_len)
8799 		memset(sa->sa_data, 0, size);
8800 	else
8801 		memcpy(sa->sa_data, dev->dev_addr,
8802 		       min_t(size_t, size, dev->addr_len));
8803 	sa->sa_family = dev->type;
8804 
8805 unlock:
8806 	rcu_read_unlock();
8807 	up_read(&dev_addr_sem);
8808 	return ret;
8809 }
8810 EXPORT_SYMBOL(dev_get_mac_address);
8811 
8812 /**
8813  *	dev_change_carrier - Change device carrier
8814  *	@dev: device
8815  *	@new_carrier: new value
8816  *
8817  *	Change device carrier
8818  */
8819 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8820 {
8821 	const struct net_device_ops *ops = dev->netdev_ops;
8822 
8823 	if (!ops->ndo_change_carrier)
8824 		return -EOPNOTSUPP;
8825 	if (!netif_device_present(dev))
8826 		return -ENODEV;
8827 	return ops->ndo_change_carrier(dev, new_carrier);
8828 }
8829 
8830 /**
8831  *	dev_get_phys_port_id - Get device physical port ID
8832  *	@dev: device
8833  *	@ppid: port ID
8834  *
8835  *	Get device physical port ID
8836  */
8837 int dev_get_phys_port_id(struct net_device *dev,
8838 			 struct netdev_phys_item_id *ppid)
8839 {
8840 	const struct net_device_ops *ops = dev->netdev_ops;
8841 
8842 	if (!ops->ndo_get_phys_port_id)
8843 		return -EOPNOTSUPP;
8844 	return ops->ndo_get_phys_port_id(dev, ppid);
8845 }
8846 
8847 /**
8848  *	dev_get_phys_port_name - Get device physical port name
8849  *	@dev: device
8850  *	@name: port name
8851  *	@len: limit of bytes to copy to name
8852  *
8853  *	Get device physical port name
8854  */
8855 int dev_get_phys_port_name(struct net_device *dev,
8856 			   char *name, size_t len)
8857 {
8858 	const struct net_device_ops *ops = dev->netdev_ops;
8859 	int err;
8860 
8861 	if (ops->ndo_get_phys_port_name) {
8862 		err = ops->ndo_get_phys_port_name(dev, name, len);
8863 		if (err != -EOPNOTSUPP)
8864 			return err;
8865 	}
8866 	return devlink_compat_phys_port_name_get(dev, name, len);
8867 }
8868 
8869 /**
8870  *	dev_get_port_parent_id - Get the device's port parent identifier
8871  *	@dev: network device
8872  *	@ppid: pointer to a storage for the port's parent identifier
8873  *	@recurse: allow/disallow recursion to lower devices
8874  *
8875  *	Get the devices's port parent identifier
8876  */
8877 int dev_get_port_parent_id(struct net_device *dev,
8878 			   struct netdev_phys_item_id *ppid,
8879 			   bool recurse)
8880 {
8881 	const struct net_device_ops *ops = dev->netdev_ops;
8882 	struct netdev_phys_item_id first = { };
8883 	struct net_device *lower_dev;
8884 	struct list_head *iter;
8885 	int err;
8886 
8887 	if (ops->ndo_get_port_parent_id) {
8888 		err = ops->ndo_get_port_parent_id(dev, ppid);
8889 		if (err != -EOPNOTSUPP)
8890 			return err;
8891 	}
8892 
8893 	err = devlink_compat_switch_id_get(dev, ppid);
8894 	if (!recurse || err != -EOPNOTSUPP)
8895 		return err;
8896 
8897 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
8898 		err = dev_get_port_parent_id(lower_dev, ppid, true);
8899 		if (err)
8900 			break;
8901 		if (!first.id_len)
8902 			first = *ppid;
8903 		else if (memcmp(&first, ppid, sizeof(*ppid)))
8904 			return -EOPNOTSUPP;
8905 	}
8906 
8907 	return err;
8908 }
8909 EXPORT_SYMBOL(dev_get_port_parent_id);
8910 
8911 /**
8912  *	netdev_port_same_parent_id - Indicate if two network devices have
8913  *	the same port parent identifier
8914  *	@a: first network device
8915  *	@b: second network device
8916  */
8917 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8918 {
8919 	struct netdev_phys_item_id a_id = { };
8920 	struct netdev_phys_item_id b_id = { };
8921 
8922 	if (dev_get_port_parent_id(a, &a_id, true) ||
8923 	    dev_get_port_parent_id(b, &b_id, true))
8924 		return false;
8925 
8926 	return netdev_phys_item_id_same(&a_id, &b_id);
8927 }
8928 EXPORT_SYMBOL(netdev_port_same_parent_id);
8929 
8930 /**
8931  *	dev_change_proto_down - set carrier according to proto_down.
8932  *
8933  *	@dev: device
8934  *	@proto_down: new value
8935  */
8936 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8937 {
8938 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8939 		return -EOPNOTSUPP;
8940 	if (!netif_device_present(dev))
8941 		return -ENODEV;
8942 	if (proto_down)
8943 		netif_carrier_off(dev);
8944 	else
8945 		netif_carrier_on(dev);
8946 	dev->proto_down = proto_down;
8947 	return 0;
8948 }
8949 
8950 /**
8951  *	dev_change_proto_down_reason - proto down reason
8952  *
8953  *	@dev: device
8954  *	@mask: proto down mask
8955  *	@value: proto down value
8956  */
8957 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8958 				  u32 value)
8959 {
8960 	int b;
8961 
8962 	if (!mask) {
8963 		dev->proto_down_reason = value;
8964 	} else {
8965 		for_each_set_bit(b, &mask, 32) {
8966 			if (value & (1 << b))
8967 				dev->proto_down_reason |= BIT(b);
8968 			else
8969 				dev->proto_down_reason &= ~BIT(b);
8970 		}
8971 	}
8972 }
8973 
8974 struct bpf_xdp_link {
8975 	struct bpf_link link;
8976 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
8977 	int flags;
8978 };
8979 
8980 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
8981 {
8982 	if (flags & XDP_FLAGS_HW_MODE)
8983 		return XDP_MODE_HW;
8984 	if (flags & XDP_FLAGS_DRV_MODE)
8985 		return XDP_MODE_DRV;
8986 	if (flags & XDP_FLAGS_SKB_MODE)
8987 		return XDP_MODE_SKB;
8988 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
8989 }
8990 
8991 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
8992 {
8993 	switch (mode) {
8994 	case XDP_MODE_SKB:
8995 		return generic_xdp_install;
8996 	case XDP_MODE_DRV:
8997 	case XDP_MODE_HW:
8998 		return dev->netdev_ops->ndo_bpf;
8999 	default:
9000 		return NULL;
9001 	}
9002 }
9003 
9004 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9005 					 enum bpf_xdp_mode mode)
9006 {
9007 	return dev->xdp_state[mode].link;
9008 }
9009 
9010 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9011 				     enum bpf_xdp_mode mode)
9012 {
9013 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9014 
9015 	if (link)
9016 		return link->link.prog;
9017 	return dev->xdp_state[mode].prog;
9018 }
9019 
9020 u8 dev_xdp_prog_count(struct net_device *dev)
9021 {
9022 	u8 count = 0;
9023 	int i;
9024 
9025 	for (i = 0; i < __MAX_XDP_MODE; i++)
9026 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9027 			count++;
9028 	return count;
9029 }
9030 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9031 
9032 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9033 {
9034 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9035 
9036 	return prog ? prog->aux->id : 0;
9037 }
9038 
9039 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9040 			     struct bpf_xdp_link *link)
9041 {
9042 	dev->xdp_state[mode].link = link;
9043 	dev->xdp_state[mode].prog = NULL;
9044 }
9045 
9046 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9047 			     struct bpf_prog *prog)
9048 {
9049 	dev->xdp_state[mode].link = NULL;
9050 	dev->xdp_state[mode].prog = prog;
9051 }
9052 
9053 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9054 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9055 			   u32 flags, struct bpf_prog *prog)
9056 {
9057 	struct netdev_bpf xdp;
9058 	int err;
9059 
9060 	memset(&xdp, 0, sizeof(xdp));
9061 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9062 	xdp.extack = extack;
9063 	xdp.flags = flags;
9064 	xdp.prog = prog;
9065 
9066 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9067 	 * "moved" into driver), so they don't increment it on their own, but
9068 	 * they do decrement refcnt when program is detached or replaced.
9069 	 * Given net_device also owns link/prog, we need to bump refcnt here
9070 	 * to prevent drivers from underflowing it.
9071 	 */
9072 	if (prog)
9073 		bpf_prog_inc(prog);
9074 	err = bpf_op(dev, &xdp);
9075 	if (err) {
9076 		if (prog)
9077 			bpf_prog_put(prog);
9078 		return err;
9079 	}
9080 
9081 	if (mode != XDP_MODE_HW)
9082 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9083 
9084 	return 0;
9085 }
9086 
9087 static void dev_xdp_uninstall(struct net_device *dev)
9088 {
9089 	struct bpf_xdp_link *link;
9090 	struct bpf_prog *prog;
9091 	enum bpf_xdp_mode mode;
9092 	bpf_op_t bpf_op;
9093 
9094 	ASSERT_RTNL();
9095 
9096 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9097 		prog = dev_xdp_prog(dev, mode);
9098 		if (!prog)
9099 			continue;
9100 
9101 		bpf_op = dev_xdp_bpf_op(dev, mode);
9102 		if (!bpf_op)
9103 			continue;
9104 
9105 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9106 
9107 		/* auto-detach link from net device */
9108 		link = dev_xdp_link(dev, mode);
9109 		if (link)
9110 			link->dev = NULL;
9111 		else
9112 			bpf_prog_put(prog);
9113 
9114 		dev_xdp_set_link(dev, mode, NULL);
9115 	}
9116 }
9117 
9118 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9119 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9120 			  struct bpf_prog *old_prog, u32 flags)
9121 {
9122 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9123 	struct bpf_prog *cur_prog;
9124 	struct net_device *upper;
9125 	struct list_head *iter;
9126 	enum bpf_xdp_mode mode;
9127 	bpf_op_t bpf_op;
9128 	int err;
9129 
9130 	ASSERT_RTNL();
9131 
9132 	/* either link or prog attachment, never both */
9133 	if (link && (new_prog || old_prog))
9134 		return -EINVAL;
9135 	/* link supports only XDP mode flags */
9136 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9137 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9138 		return -EINVAL;
9139 	}
9140 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9141 	if (num_modes > 1) {
9142 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9143 		return -EINVAL;
9144 	}
9145 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9146 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9147 		NL_SET_ERR_MSG(extack,
9148 			       "More than one program loaded, unset mode is ambiguous");
9149 		return -EINVAL;
9150 	}
9151 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9152 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9153 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9154 		return -EINVAL;
9155 	}
9156 
9157 	mode = dev_xdp_mode(dev, flags);
9158 	/* can't replace attached link */
9159 	if (dev_xdp_link(dev, mode)) {
9160 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9161 		return -EBUSY;
9162 	}
9163 
9164 	/* don't allow if an upper device already has a program */
9165 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9166 		if (dev_xdp_prog_count(upper) > 0) {
9167 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9168 			return -EEXIST;
9169 		}
9170 	}
9171 
9172 	cur_prog = dev_xdp_prog(dev, mode);
9173 	/* can't replace attached prog with link */
9174 	if (link && cur_prog) {
9175 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9176 		return -EBUSY;
9177 	}
9178 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9179 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9180 		return -EEXIST;
9181 	}
9182 
9183 	/* put effective new program into new_prog */
9184 	if (link)
9185 		new_prog = link->link.prog;
9186 
9187 	if (new_prog) {
9188 		bool offload = mode == XDP_MODE_HW;
9189 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9190 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9191 
9192 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9193 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9194 			return -EBUSY;
9195 		}
9196 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9197 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9198 			return -EEXIST;
9199 		}
9200 		if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9201 			NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9202 			return -EINVAL;
9203 		}
9204 		if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9205 			NL_SET_ERR_MSG(extack, "Program bound to different device");
9206 			return -EINVAL;
9207 		}
9208 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9209 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9210 			return -EINVAL;
9211 		}
9212 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9213 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9214 			return -EINVAL;
9215 		}
9216 	}
9217 
9218 	/* don't call drivers if the effective program didn't change */
9219 	if (new_prog != cur_prog) {
9220 		bpf_op = dev_xdp_bpf_op(dev, mode);
9221 		if (!bpf_op) {
9222 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9223 			return -EOPNOTSUPP;
9224 		}
9225 
9226 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9227 		if (err)
9228 			return err;
9229 	}
9230 
9231 	if (link)
9232 		dev_xdp_set_link(dev, mode, link);
9233 	else
9234 		dev_xdp_set_prog(dev, mode, new_prog);
9235 	if (cur_prog)
9236 		bpf_prog_put(cur_prog);
9237 
9238 	return 0;
9239 }
9240 
9241 static int dev_xdp_attach_link(struct net_device *dev,
9242 			       struct netlink_ext_ack *extack,
9243 			       struct bpf_xdp_link *link)
9244 {
9245 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9246 }
9247 
9248 static int dev_xdp_detach_link(struct net_device *dev,
9249 			       struct netlink_ext_ack *extack,
9250 			       struct bpf_xdp_link *link)
9251 {
9252 	enum bpf_xdp_mode mode;
9253 	bpf_op_t bpf_op;
9254 
9255 	ASSERT_RTNL();
9256 
9257 	mode = dev_xdp_mode(dev, link->flags);
9258 	if (dev_xdp_link(dev, mode) != link)
9259 		return -EINVAL;
9260 
9261 	bpf_op = dev_xdp_bpf_op(dev, mode);
9262 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9263 	dev_xdp_set_link(dev, mode, NULL);
9264 	return 0;
9265 }
9266 
9267 static void bpf_xdp_link_release(struct bpf_link *link)
9268 {
9269 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9270 
9271 	rtnl_lock();
9272 
9273 	/* if racing with net_device's tear down, xdp_link->dev might be
9274 	 * already NULL, in which case link was already auto-detached
9275 	 */
9276 	if (xdp_link->dev) {
9277 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9278 		xdp_link->dev = NULL;
9279 	}
9280 
9281 	rtnl_unlock();
9282 }
9283 
9284 static int bpf_xdp_link_detach(struct bpf_link *link)
9285 {
9286 	bpf_xdp_link_release(link);
9287 	return 0;
9288 }
9289 
9290 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9291 {
9292 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9293 
9294 	kfree(xdp_link);
9295 }
9296 
9297 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9298 				     struct seq_file *seq)
9299 {
9300 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9301 	u32 ifindex = 0;
9302 
9303 	rtnl_lock();
9304 	if (xdp_link->dev)
9305 		ifindex = xdp_link->dev->ifindex;
9306 	rtnl_unlock();
9307 
9308 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9309 }
9310 
9311 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9312 				       struct bpf_link_info *info)
9313 {
9314 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9315 	u32 ifindex = 0;
9316 
9317 	rtnl_lock();
9318 	if (xdp_link->dev)
9319 		ifindex = xdp_link->dev->ifindex;
9320 	rtnl_unlock();
9321 
9322 	info->xdp.ifindex = ifindex;
9323 	return 0;
9324 }
9325 
9326 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9327 			       struct bpf_prog *old_prog)
9328 {
9329 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9330 	enum bpf_xdp_mode mode;
9331 	bpf_op_t bpf_op;
9332 	int err = 0;
9333 
9334 	rtnl_lock();
9335 
9336 	/* link might have been auto-released already, so fail */
9337 	if (!xdp_link->dev) {
9338 		err = -ENOLINK;
9339 		goto out_unlock;
9340 	}
9341 
9342 	if (old_prog && link->prog != old_prog) {
9343 		err = -EPERM;
9344 		goto out_unlock;
9345 	}
9346 	old_prog = link->prog;
9347 	if (old_prog->type != new_prog->type ||
9348 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9349 		err = -EINVAL;
9350 		goto out_unlock;
9351 	}
9352 
9353 	if (old_prog == new_prog) {
9354 		/* no-op, don't disturb drivers */
9355 		bpf_prog_put(new_prog);
9356 		goto out_unlock;
9357 	}
9358 
9359 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9360 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9361 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9362 			      xdp_link->flags, new_prog);
9363 	if (err)
9364 		goto out_unlock;
9365 
9366 	old_prog = xchg(&link->prog, new_prog);
9367 	bpf_prog_put(old_prog);
9368 
9369 out_unlock:
9370 	rtnl_unlock();
9371 	return err;
9372 }
9373 
9374 static const struct bpf_link_ops bpf_xdp_link_lops = {
9375 	.release = bpf_xdp_link_release,
9376 	.dealloc = bpf_xdp_link_dealloc,
9377 	.detach = bpf_xdp_link_detach,
9378 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9379 	.fill_link_info = bpf_xdp_link_fill_link_info,
9380 	.update_prog = bpf_xdp_link_update,
9381 };
9382 
9383 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9384 {
9385 	struct net *net = current->nsproxy->net_ns;
9386 	struct bpf_link_primer link_primer;
9387 	struct bpf_xdp_link *link;
9388 	struct net_device *dev;
9389 	int err, fd;
9390 
9391 	rtnl_lock();
9392 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9393 	if (!dev) {
9394 		rtnl_unlock();
9395 		return -EINVAL;
9396 	}
9397 
9398 	link = kzalloc(sizeof(*link), GFP_USER);
9399 	if (!link) {
9400 		err = -ENOMEM;
9401 		goto unlock;
9402 	}
9403 
9404 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9405 	link->dev = dev;
9406 	link->flags = attr->link_create.flags;
9407 
9408 	err = bpf_link_prime(&link->link, &link_primer);
9409 	if (err) {
9410 		kfree(link);
9411 		goto unlock;
9412 	}
9413 
9414 	err = dev_xdp_attach_link(dev, NULL, link);
9415 	rtnl_unlock();
9416 
9417 	if (err) {
9418 		link->dev = NULL;
9419 		bpf_link_cleanup(&link_primer);
9420 		goto out_put_dev;
9421 	}
9422 
9423 	fd = bpf_link_settle(&link_primer);
9424 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9425 	dev_put(dev);
9426 	return fd;
9427 
9428 unlock:
9429 	rtnl_unlock();
9430 
9431 out_put_dev:
9432 	dev_put(dev);
9433 	return err;
9434 }
9435 
9436 /**
9437  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9438  *	@dev: device
9439  *	@extack: netlink extended ack
9440  *	@fd: new program fd or negative value to clear
9441  *	@expected_fd: old program fd that userspace expects to replace or clear
9442  *	@flags: xdp-related flags
9443  *
9444  *	Set or clear a bpf program for a device
9445  */
9446 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9447 		      int fd, int expected_fd, u32 flags)
9448 {
9449 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9450 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9451 	int err;
9452 
9453 	ASSERT_RTNL();
9454 
9455 	if (fd >= 0) {
9456 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9457 						 mode != XDP_MODE_SKB);
9458 		if (IS_ERR(new_prog))
9459 			return PTR_ERR(new_prog);
9460 	}
9461 
9462 	if (expected_fd >= 0) {
9463 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9464 						 mode != XDP_MODE_SKB);
9465 		if (IS_ERR(old_prog)) {
9466 			err = PTR_ERR(old_prog);
9467 			old_prog = NULL;
9468 			goto err_out;
9469 		}
9470 	}
9471 
9472 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9473 
9474 err_out:
9475 	if (err && new_prog)
9476 		bpf_prog_put(new_prog);
9477 	if (old_prog)
9478 		bpf_prog_put(old_prog);
9479 	return err;
9480 }
9481 
9482 /**
9483  *	dev_new_index	-	allocate an ifindex
9484  *	@net: the applicable net namespace
9485  *
9486  *	Returns a suitable unique value for a new device interface
9487  *	number.  The caller must hold the rtnl semaphore or the
9488  *	dev_base_lock to be sure it remains unique.
9489  */
9490 static int dev_new_index(struct net *net)
9491 {
9492 	int ifindex = net->ifindex;
9493 
9494 	for (;;) {
9495 		if (++ifindex <= 0)
9496 			ifindex = 1;
9497 		if (!__dev_get_by_index(net, ifindex))
9498 			return net->ifindex = ifindex;
9499 	}
9500 }
9501 
9502 /* Delayed registration/unregisteration */
9503 LIST_HEAD(net_todo_list);
9504 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9505 
9506 static void net_set_todo(struct net_device *dev)
9507 {
9508 	list_add_tail(&dev->todo_list, &net_todo_list);
9509 	atomic_inc(&dev_net(dev)->dev_unreg_count);
9510 }
9511 
9512 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9513 	struct net_device *upper, netdev_features_t features)
9514 {
9515 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9516 	netdev_features_t feature;
9517 	int feature_bit;
9518 
9519 	for_each_netdev_feature(upper_disables, feature_bit) {
9520 		feature = __NETIF_F_BIT(feature_bit);
9521 		if (!(upper->wanted_features & feature)
9522 		    && (features & feature)) {
9523 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9524 				   &feature, upper->name);
9525 			features &= ~feature;
9526 		}
9527 	}
9528 
9529 	return features;
9530 }
9531 
9532 static void netdev_sync_lower_features(struct net_device *upper,
9533 	struct net_device *lower, netdev_features_t features)
9534 {
9535 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9536 	netdev_features_t feature;
9537 	int feature_bit;
9538 
9539 	for_each_netdev_feature(upper_disables, feature_bit) {
9540 		feature = __NETIF_F_BIT(feature_bit);
9541 		if (!(features & feature) && (lower->features & feature)) {
9542 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9543 				   &feature, lower->name);
9544 			lower->wanted_features &= ~feature;
9545 			__netdev_update_features(lower);
9546 
9547 			if (unlikely(lower->features & feature))
9548 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9549 					    &feature, lower->name);
9550 			else
9551 				netdev_features_change(lower);
9552 		}
9553 	}
9554 }
9555 
9556 static netdev_features_t netdev_fix_features(struct net_device *dev,
9557 	netdev_features_t features)
9558 {
9559 	/* Fix illegal checksum combinations */
9560 	if ((features & NETIF_F_HW_CSUM) &&
9561 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9562 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9563 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9564 	}
9565 
9566 	/* TSO requires that SG is present as well. */
9567 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9568 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9569 		features &= ~NETIF_F_ALL_TSO;
9570 	}
9571 
9572 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9573 					!(features & NETIF_F_IP_CSUM)) {
9574 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9575 		features &= ~NETIF_F_TSO;
9576 		features &= ~NETIF_F_TSO_ECN;
9577 	}
9578 
9579 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9580 					 !(features & NETIF_F_IPV6_CSUM)) {
9581 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9582 		features &= ~NETIF_F_TSO6;
9583 	}
9584 
9585 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9586 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9587 		features &= ~NETIF_F_TSO_MANGLEID;
9588 
9589 	/* TSO ECN requires that TSO is present as well. */
9590 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9591 		features &= ~NETIF_F_TSO_ECN;
9592 
9593 	/* Software GSO depends on SG. */
9594 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9595 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9596 		features &= ~NETIF_F_GSO;
9597 	}
9598 
9599 	/* GSO partial features require GSO partial be set */
9600 	if ((features & dev->gso_partial_features) &&
9601 	    !(features & NETIF_F_GSO_PARTIAL)) {
9602 		netdev_dbg(dev,
9603 			   "Dropping partially supported GSO features since no GSO partial.\n");
9604 		features &= ~dev->gso_partial_features;
9605 	}
9606 
9607 	if (!(features & NETIF_F_RXCSUM)) {
9608 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9609 		 * successfully merged by hardware must also have the
9610 		 * checksum verified by hardware.  If the user does not
9611 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9612 		 */
9613 		if (features & NETIF_F_GRO_HW) {
9614 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9615 			features &= ~NETIF_F_GRO_HW;
9616 		}
9617 	}
9618 
9619 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9620 	if (features & NETIF_F_RXFCS) {
9621 		if (features & NETIF_F_LRO) {
9622 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9623 			features &= ~NETIF_F_LRO;
9624 		}
9625 
9626 		if (features & NETIF_F_GRO_HW) {
9627 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9628 			features &= ~NETIF_F_GRO_HW;
9629 		}
9630 	}
9631 
9632 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9633 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9634 		features &= ~NETIF_F_LRO;
9635 	}
9636 
9637 	if (features & NETIF_F_HW_TLS_TX) {
9638 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9639 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9640 		bool hw_csum = features & NETIF_F_HW_CSUM;
9641 
9642 		if (!ip_csum && !hw_csum) {
9643 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9644 			features &= ~NETIF_F_HW_TLS_TX;
9645 		}
9646 	}
9647 
9648 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9649 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9650 		features &= ~NETIF_F_HW_TLS_RX;
9651 	}
9652 
9653 	return features;
9654 }
9655 
9656 int __netdev_update_features(struct net_device *dev)
9657 {
9658 	struct net_device *upper, *lower;
9659 	netdev_features_t features;
9660 	struct list_head *iter;
9661 	int err = -1;
9662 
9663 	ASSERT_RTNL();
9664 
9665 	features = netdev_get_wanted_features(dev);
9666 
9667 	if (dev->netdev_ops->ndo_fix_features)
9668 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9669 
9670 	/* driver might be less strict about feature dependencies */
9671 	features = netdev_fix_features(dev, features);
9672 
9673 	/* some features can't be enabled if they're off on an upper device */
9674 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9675 		features = netdev_sync_upper_features(dev, upper, features);
9676 
9677 	if (dev->features == features)
9678 		goto sync_lower;
9679 
9680 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9681 		&dev->features, &features);
9682 
9683 	if (dev->netdev_ops->ndo_set_features)
9684 		err = dev->netdev_ops->ndo_set_features(dev, features);
9685 	else
9686 		err = 0;
9687 
9688 	if (unlikely(err < 0)) {
9689 		netdev_err(dev,
9690 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
9691 			err, &features, &dev->features);
9692 		/* return non-0 since some features might have changed and
9693 		 * it's better to fire a spurious notification than miss it
9694 		 */
9695 		return -1;
9696 	}
9697 
9698 sync_lower:
9699 	/* some features must be disabled on lower devices when disabled
9700 	 * on an upper device (think: bonding master or bridge)
9701 	 */
9702 	netdev_for_each_lower_dev(dev, lower, iter)
9703 		netdev_sync_lower_features(dev, lower, features);
9704 
9705 	if (!err) {
9706 		netdev_features_t diff = features ^ dev->features;
9707 
9708 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9709 			/* udp_tunnel_{get,drop}_rx_info both need
9710 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9711 			 * device, or they won't do anything.
9712 			 * Thus we need to update dev->features
9713 			 * *before* calling udp_tunnel_get_rx_info,
9714 			 * but *after* calling udp_tunnel_drop_rx_info.
9715 			 */
9716 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9717 				dev->features = features;
9718 				udp_tunnel_get_rx_info(dev);
9719 			} else {
9720 				udp_tunnel_drop_rx_info(dev);
9721 			}
9722 		}
9723 
9724 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9725 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9726 				dev->features = features;
9727 				err |= vlan_get_rx_ctag_filter_info(dev);
9728 			} else {
9729 				vlan_drop_rx_ctag_filter_info(dev);
9730 			}
9731 		}
9732 
9733 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9734 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9735 				dev->features = features;
9736 				err |= vlan_get_rx_stag_filter_info(dev);
9737 			} else {
9738 				vlan_drop_rx_stag_filter_info(dev);
9739 			}
9740 		}
9741 
9742 		dev->features = features;
9743 	}
9744 
9745 	return err < 0 ? 0 : 1;
9746 }
9747 
9748 /**
9749  *	netdev_update_features - recalculate device features
9750  *	@dev: the device to check
9751  *
9752  *	Recalculate dev->features set and send notifications if it
9753  *	has changed. Should be called after driver or hardware dependent
9754  *	conditions might have changed that influence the features.
9755  */
9756 void netdev_update_features(struct net_device *dev)
9757 {
9758 	if (__netdev_update_features(dev))
9759 		netdev_features_change(dev);
9760 }
9761 EXPORT_SYMBOL(netdev_update_features);
9762 
9763 /**
9764  *	netdev_change_features - recalculate device features
9765  *	@dev: the device to check
9766  *
9767  *	Recalculate dev->features set and send notifications even
9768  *	if they have not changed. Should be called instead of
9769  *	netdev_update_features() if also dev->vlan_features might
9770  *	have changed to allow the changes to be propagated to stacked
9771  *	VLAN devices.
9772  */
9773 void netdev_change_features(struct net_device *dev)
9774 {
9775 	__netdev_update_features(dev);
9776 	netdev_features_change(dev);
9777 }
9778 EXPORT_SYMBOL(netdev_change_features);
9779 
9780 /**
9781  *	netif_stacked_transfer_operstate -	transfer operstate
9782  *	@rootdev: the root or lower level device to transfer state from
9783  *	@dev: the device to transfer operstate to
9784  *
9785  *	Transfer operational state from root to device. This is normally
9786  *	called when a stacking relationship exists between the root
9787  *	device and the device(a leaf device).
9788  */
9789 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9790 					struct net_device *dev)
9791 {
9792 	if (rootdev->operstate == IF_OPER_DORMANT)
9793 		netif_dormant_on(dev);
9794 	else
9795 		netif_dormant_off(dev);
9796 
9797 	if (rootdev->operstate == IF_OPER_TESTING)
9798 		netif_testing_on(dev);
9799 	else
9800 		netif_testing_off(dev);
9801 
9802 	if (netif_carrier_ok(rootdev))
9803 		netif_carrier_on(dev);
9804 	else
9805 		netif_carrier_off(dev);
9806 }
9807 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9808 
9809 static int netif_alloc_rx_queues(struct net_device *dev)
9810 {
9811 	unsigned int i, count = dev->num_rx_queues;
9812 	struct netdev_rx_queue *rx;
9813 	size_t sz = count * sizeof(*rx);
9814 	int err = 0;
9815 
9816 	BUG_ON(count < 1);
9817 
9818 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9819 	if (!rx)
9820 		return -ENOMEM;
9821 
9822 	dev->_rx = rx;
9823 
9824 	for (i = 0; i < count; i++) {
9825 		rx[i].dev = dev;
9826 
9827 		/* XDP RX-queue setup */
9828 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9829 		if (err < 0)
9830 			goto err_rxq_info;
9831 	}
9832 	return 0;
9833 
9834 err_rxq_info:
9835 	/* Rollback successful reg's and free other resources */
9836 	while (i--)
9837 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9838 	kvfree(dev->_rx);
9839 	dev->_rx = NULL;
9840 	return err;
9841 }
9842 
9843 static void netif_free_rx_queues(struct net_device *dev)
9844 {
9845 	unsigned int i, count = dev->num_rx_queues;
9846 
9847 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9848 	if (!dev->_rx)
9849 		return;
9850 
9851 	for (i = 0; i < count; i++)
9852 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9853 
9854 	kvfree(dev->_rx);
9855 }
9856 
9857 static void netdev_init_one_queue(struct net_device *dev,
9858 				  struct netdev_queue *queue, void *_unused)
9859 {
9860 	/* Initialize queue lock */
9861 	spin_lock_init(&queue->_xmit_lock);
9862 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9863 	queue->xmit_lock_owner = -1;
9864 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9865 	queue->dev = dev;
9866 #ifdef CONFIG_BQL
9867 	dql_init(&queue->dql, HZ);
9868 #endif
9869 }
9870 
9871 static void netif_free_tx_queues(struct net_device *dev)
9872 {
9873 	kvfree(dev->_tx);
9874 }
9875 
9876 static int netif_alloc_netdev_queues(struct net_device *dev)
9877 {
9878 	unsigned int count = dev->num_tx_queues;
9879 	struct netdev_queue *tx;
9880 	size_t sz = count * sizeof(*tx);
9881 
9882 	if (count < 1 || count > 0xffff)
9883 		return -EINVAL;
9884 
9885 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9886 	if (!tx)
9887 		return -ENOMEM;
9888 
9889 	dev->_tx = tx;
9890 
9891 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9892 	spin_lock_init(&dev->tx_global_lock);
9893 
9894 	return 0;
9895 }
9896 
9897 void netif_tx_stop_all_queues(struct net_device *dev)
9898 {
9899 	unsigned int i;
9900 
9901 	for (i = 0; i < dev->num_tx_queues; i++) {
9902 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9903 
9904 		netif_tx_stop_queue(txq);
9905 	}
9906 }
9907 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9908 
9909 /**
9910  * register_netdevice() - register a network device
9911  * @dev: device to register
9912  *
9913  * Take a prepared network device structure and make it externally accessible.
9914  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9915  * Callers must hold the rtnl lock - you may want register_netdev()
9916  * instead of this.
9917  */
9918 int register_netdevice(struct net_device *dev)
9919 {
9920 	int ret;
9921 	struct net *net = dev_net(dev);
9922 
9923 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9924 		     NETDEV_FEATURE_COUNT);
9925 	BUG_ON(dev_boot_phase);
9926 	ASSERT_RTNL();
9927 
9928 	might_sleep();
9929 
9930 	/* When net_device's are persistent, this will be fatal. */
9931 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9932 	BUG_ON(!net);
9933 
9934 	ret = ethtool_check_ops(dev->ethtool_ops);
9935 	if (ret)
9936 		return ret;
9937 
9938 	spin_lock_init(&dev->addr_list_lock);
9939 	netdev_set_addr_lockdep_class(dev);
9940 
9941 	ret = dev_get_valid_name(net, dev, dev->name);
9942 	if (ret < 0)
9943 		goto out;
9944 
9945 	ret = -ENOMEM;
9946 	dev->name_node = netdev_name_node_head_alloc(dev);
9947 	if (!dev->name_node)
9948 		goto out;
9949 
9950 	/* Init, if this function is available */
9951 	if (dev->netdev_ops->ndo_init) {
9952 		ret = dev->netdev_ops->ndo_init(dev);
9953 		if (ret) {
9954 			if (ret > 0)
9955 				ret = -EIO;
9956 			goto err_free_name;
9957 		}
9958 	}
9959 
9960 	if (((dev->hw_features | dev->features) &
9961 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
9962 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9963 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9964 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9965 		ret = -EINVAL;
9966 		goto err_uninit;
9967 	}
9968 
9969 	ret = -EBUSY;
9970 	if (!dev->ifindex)
9971 		dev->ifindex = dev_new_index(net);
9972 	else if (__dev_get_by_index(net, dev->ifindex))
9973 		goto err_uninit;
9974 
9975 	/* Transfer changeable features to wanted_features and enable
9976 	 * software offloads (GSO and GRO).
9977 	 */
9978 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9979 	dev->features |= NETIF_F_SOFT_FEATURES;
9980 
9981 	if (dev->udp_tunnel_nic_info) {
9982 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9983 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9984 	}
9985 
9986 	dev->wanted_features = dev->features & dev->hw_features;
9987 
9988 	if (!(dev->flags & IFF_LOOPBACK))
9989 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
9990 
9991 	/* If IPv4 TCP segmentation offload is supported we should also
9992 	 * allow the device to enable segmenting the frame with the option
9993 	 * of ignoring a static IP ID value.  This doesn't enable the
9994 	 * feature itself but allows the user to enable it later.
9995 	 */
9996 	if (dev->hw_features & NETIF_F_TSO)
9997 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
9998 	if (dev->vlan_features & NETIF_F_TSO)
9999 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10000 	if (dev->mpls_features & NETIF_F_TSO)
10001 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10002 	if (dev->hw_enc_features & NETIF_F_TSO)
10003 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10004 
10005 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10006 	 */
10007 	dev->vlan_features |= NETIF_F_HIGHDMA;
10008 
10009 	/* Make NETIF_F_SG inheritable to tunnel devices.
10010 	 */
10011 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10012 
10013 	/* Make NETIF_F_SG inheritable to MPLS.
10014 	 */
10015 	dev->mpls_features |= NETIF_F_SG;
10016 
10017 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10018 	ret = notifier_to_errno(ret);
10019 	if (ret)
10020 		goto err_uninit;
10021 
10022 	ret = netdev_register_kobject(dev);
10023 	write_lock(&dev_base_lock);
10024 	dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10025 	write_unlock(&dev_base_lock);
10026 	if (ret)
10027 		goto err_uninit_notify;
10028 
10029 	__netdev_update_features(dev);
10030 
10031 	/*
10032 	 *	Default initial state at registry is that the
10033 	 *	device is present.
10034 	 */
10035 
10036 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10037 
10038 	linkwatch_init_dev(dev);
10039 
10040 	dev_init_scheduler(dev);
10041 
10042 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10043 	list_netdevice(dev);
10044 
10045 	add_device_randomness(dev->dev_addr, dev->addr_len);
10046 
10047 	/* If the device has permanent device address, driver should
10048 	 * set dev_addr and also addr_assign_type should be set to
10049 	 * NET_ADDR_PERM (default value).
10050 	 */
10051 	if (dev->addr_assign_type == NET_ADDR_PERM)
10052 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10053 
10054 	/* Notify protocols, that a new device appeared. */
10055 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10056 	ret = notifier_to_errno(ret);
10057 	if (ret) {
10058 		/* Expect explicit free_netdev() on failure */
10059 		dev->needs_free_netdev = false;
10060 		unregister_netdevice_queue(dev, NULL);
10061 		goto out;
10062 	}
10063 	/*
10064 	 *	Prevent userspace races by waiting until the network
10065 	 *	device is fully setup before sending notifications.
10066 	 */
10067 	if (!dev->rtnl_link_ops ||
10068 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10069 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10070 
10071 out:
10072 	return ret;
10073 
10074 err_uninit_notify:
10075 	call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10076 err_uninit:
10077 	if (dev->netdev_ops->ndo_uninit)
10078 		dev->netdev_ops->ndo_uninit(dev);
10079 	if (dev->priv_destructor)
10080 		dev->priv_destructor(dev);
10081 err_free_name:
10082 	netdev_name_node_free(dev->name_node);
10083 	goto out;
10084 }
10085 EXPORT_SYMBOL(register_netdevice);
10086 
10087 /**
10088  *	init_dummy_netdev	- init a dummy network device for NAPI
10089  *	@dev: device to init
10090  *
10091  *	This takes a network device structure and initialize the minimum
10092  *	amount of fields so it can be used to schedule NAPI polls without
10093  *	registering a full blown interface. This is to be used by drivers
10094  *	that need to tie several hardware interfaces to a single NAPI
10095  *	poll scheduler due to HW limitations.
10096  */
10097 int init_dummy_netdev(struct net_device *dev)
10098 {
10099 	/* Clear everything. Note we don't initialize spinlocks
10100 	 * are they aren't supposed to be taken by any of the
10101 	 * NAPI code and this dummy netdev is supposed to be
10102 	 * only ever used for NAPI polls
10103 	 */
10104 	memset(dev, 0, sizeof(struct net_device));
10105 
10106 	/* make sure we BUG if trying to hit standard
10107 	 * register/unregister code path
10108 	 */
10109 	dev->reg_state = NETREG_DUMMY;
10110 
10111 	/* NAPI wants this */
10112 	INIT_LIST_HEAD(&dev->napi_list);
10113 
10114 	/* a dummy interface is started by default */
10115 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10116 	set_bit(__LINK_STATE_START, &dev->state);
10117 
10118 	/* napi_busy_loop stats accounting wants this */
10119 	dev_net_set(dev, &init_net);
10120 
10121 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10122 	 * because users of this 'device' dont need to change
10123 	 * its refcount.
10124 	 */
10125 
10126 	return 0;
10127 }
10128 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10129 
10130 
10131 /**
10132  *	register_netdev	- register a network device
10133  *	@dev: device to register
10134  *
10135  *	Take a completed network device structure and add it to the kernel
10136  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10137  *	chain. 0 is returned on success. A negative errno code is returned
10138  *	on a failure to set up the device, or if the name is a duplicate.
10139  *
10140  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10141  *	and expands the device name if you passed a format string to
10142  *	alloc_netdev.
10143  */
10144 int register_netdev(struct net_device *dev)
10145 {
10146 	int err;
10147 
10148 	if (rtnl_lock_killable())
10149 		return -EINTR;
10150 	err = register_netdevice(dev);
10151 	rtnl_unlock();
10152 	return err;
10153 }
10154 EXPORT_SYMBOL(register_netdev);
10155 
10156 int netdev_refcnt_read(const struct net_device *dev)
10157 {
10158 #ifdef CONFIG_PCPU_DEV_REFCNT
10159 	int i, refcnt = 0;
10160 
10161 	for_each_possible_cpu(i)
10162 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10163 	return refcnt;
10164 #else
10165 	return refcount_read(&dev->dev_refcnt);
10166 #endif
10167 }
10168 EXPORT_SYMBOL(netdev_refcnt_read);
10169 
10170 int netdev_unregister_timeout_secs __read_mostly = 10;
10171 
10172 #define WAIT_REFS_MIN_MSECS 1
10173 #define WAIT_REFS_MAX_MSECS 250
10174 /**
10175  * netdev_wait_allrefs_any - wait until all references are gone.
10176  * @list: list of net_devices to wait on
10177  *
10178  * This is called when unregistering network devices.
10179  *
10180  * Any protocol or device that holds a reference should register
10181  * for netdevice notification, and cleanup and put back the
10182  * reference if they receive an UNREGISTER event.
10183  * We can get stuck here if buggy protocols don't correctly
10184  * call dev_put.
10185  */
10186 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10187 {
10188 	unsigned long rebroadcast_time, warning_time;
10189 	struct net_device *dev;
10190 	int wait = 0;
10191 
10192 	rebroadcast_time = warning_time = jiffies;
10193 
10194 	list_for_each_entry(dev, list, todo_list)
10195 		if (netdev_refcnt_read(dev) == 1)
10196 			return dev;
10197 
10198 	while (true) {
10199 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10200 			rtnl_lock();
10201 
10202 			/* Rebroadcast unregister notification */
10203 			list_for_each_entry(dev, list, todo_list)
10204 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10205 
10206 			__rtnl_unlock();
10207 			rcu_barrier();
10208 			rtnl_lock();
10209 
10210 			list_for_each_entry(dev, list, todo_list)
10211 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10212 					     &dev->state)) {
10213 					/* We must not have linkwatch events
10214 					 * pending on unregister. If this
10215 					 * happens, we simply run the queue
10216 					 * unscheduled, resulting in a noop
10217 					 * for this device.
10218 					 */
10219 					linkwatch_run_queue();
10220 					break;
10221 				}
10222 
10223 			__rtnl_unlock();
10224 
10225 			rebroadcast_time = jiffies;
10226 		}
10227 
10228 		if (!wait) {
10229 			rcu_barrier();
10230 			wait = WAIT_REFS_MIN_MSECS;
10231 		} else {
10232 			msleep(wait);
10233 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10234 		}
10235 
10236 		list_for_each_entry(dev, list, todo_list)
10237 			if (netdev_refcnt_read(dev) == 1)
10238 				return dev;
10239 
10240 		if (time_after(jiffies, warning_time +
10241 			       READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10242 			list_for_each_entry(dev, list, todo_list) {
10243 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10244 					 dev->name, netdev_refcnt_read(dev));
10245 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10246 			}
10247 
10248 			warning_time = jiffies;
10249 		}
10250 	}
10251 }
10252 
10253 /* The sequence is:
10254  *
10255  *	rtnl_lock();
10256  *	...
10257  *	register_netdevice(x1);
10258  *	register_netdevice(x2);
10259  *	...
10260  *	unregister_netdevice(y1);
10261  *	unregister_netdevice(y2);
10262  *      ...
10263  *	rtnl_unlock();
10264  *	free_netdev(y1);
10265  *	free_netdev(y2);
10266  *
10267  * We are invoked by rtnl_unlock().
10268  * This allows us to deal with problems:
10269  * 1) We can delete sysfs objects which invoke hotplug
10270  *    without deadlocking with linkwatch via keventd.
10271  * 2) Since we run with the RTNL semaphore not held, we can sleep
10272  *    safely in order to wait for the netdev refcnt to drop to zero.
10273  *
10274  * We must not return until all unregister events added during
10275  * the interval the lock was held have been completed.
10276  */
10277 void netdev_run_todo(void)
10278 {
10279 	struct net_device *dev, *tmp;
10280 	struct list_head list;
10281 #ifdef CONFIG_LOCKDEP
10282 	struct list_head unlink_list;
10283 
10284 	list_replace_init(&net_unlink_list, &unlink_list);
10285 
10286 	while (!list_empty(&unlink_list)) {
10287 		struct net_device *dev = list_first_entry(&unlink_list,
10288 							  struct net_device,
10289 							  unlink_list);
10290 		list_del_init(&dev->unlink_list);
10291 		dev->nested_level = dev->lower_level - 1;
10292 	}
10293 #endif
10294 
10295 	/* Snapshot list, allow later requests */
10296 	list_replace_init(&net_todo_list, &list);
10297 
10298 	__rtnl_unlock();
10299 
10300 	/* Wait for rcu callbacks to finish before next phase */
10301 	if (!list_empty(&list))
10302 		rcu_barrier();
10303 
10304 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10305 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10306 			netdev_WARN(dev, "run_todo but not unregistering\n");
10307 			list_del(&dev->todo_list);
10308 			continue;
10309 		}
10310 
10311 		write_lock(&dev_base_lock);
10312 		dev->reg_state = NETREG_UNREGISTERED;
10313 		write_unlock(&dev_base_lock);
10314 		linkwatch_forget_dev(dev);
10315 	}
10316 
10317 	while (!list_empty(&list)) {
10318 		dev = netdev_wait_allrefs_any(&list);
10319 		list_del(&dev->todo_list);
10320 
10321 		/* paranoia */
10322 		BUG_ON(netdev_refcnt_read(dev) != 1);
10323 		BUG_ON(!list_empty(&dev->ptype_all));
10324 		BUG_ON(!list_empty(&dev->ptype_specific));
10325 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10326 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10327 
10328 		if (dev->priv_destructor)
10329 			dev->priv_destructor(dev);
10330 		if (dev->needs_free_netdev)
10331 			free_netdev(dev);
10332 
10333 		if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10334 			wake_up(&netdev_unregistering_wq);
10335 
10336 		/* Free network device */
10337 		kobject_put(&dev->dev.kobj);
10338 	}
10339 }
10340 
10341 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10342  * all the same fields in the same order as net_device_stats, with only
10343  * the type differing, but rtnl_link_stats64 may have additional fields
10344  * at the end for newer counters.
10345  */
10346 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10347 			     const struct net_device_stats *netdev_stats)
10348 {
10349 	size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10350 	const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10351 	u64 *dst = (u64 *)stats64;
10352 
10353 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10354 	for (i = 0; i < n; i++)
10355 		dst[i] = (unsigned long)atomic_long_read(&src[i]);
10356 	/* zero out counters that only exist in rtnl_link_stats64 */
10357 	memset((char *)stats64 + n * sizeof(u64), 0,
10358 	       sizeof(*stats64) - n * sizeof(u64));
10359 }
10360 EXPORT_SYMBOL(netdev_stats_to_stats64);
10361 
10362 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10363 {
10364 	struct net_device_core_stats __percpu *p;
10365 
10366 	p = alloc_percpu_gfp(struct net_device_core_stats,
10367 			     GFP_ATOMIC | __GFP_NOWARN);
10368 
10369 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10370 		free_percpu(p);
10371 
10372 	/* This READ_ONCE() pairs with the cmpxchg() above */
10373 	return READ_ONCE(dev->core_stats);
10374 }
10375 EXPORT_SYMBOL(netdev_core_stats_alloc);
10376 
10377 /**
10378  *	dev_get_stats	- get network device statistics
10379  *	@dev: device to get statistics from
10380  *	@storage: place to store stats
10381  *
10382  *	Get network statistics from device. Return @storage.
10383  *	The device driver may provide its own method by setting
10384  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10385  *	otherwise the internal statistics structure is used.
10386  */
10387 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10388 					struct rtnl_link_stats64 *storage)
10389 {
10390 	const struct net_device_ops *ops = dev->netdev_ops;
10391 	const struct net_device_core_stats __percpu *p;
10392 
10393 	if (ops->ndo_get_stats64) {
10394 		memset(storage, 0, sizeof(*storage));
10395 		ops->ndo_get_stats64(dev, storage);
10396 	} else if (ops->ndo_get_stats) {
10397 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10398 	} else {
10399 		netdev_stats_to_stats64(storage, &dev->stats);
10400 	}
10401 
10402 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10403 	p = READ_ONCE(dev->core_stats);
10404 	if (p) {
10405 		const struct net_device_core_stats *core_stats;
10406 		int i;
10407 
10408 		for_each_possible_cpu(i) {
10409 			core_stats = per_cpu_ptr(p, i);
10410 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10411 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10412 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10413 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10414 		}
10415 	}
10416 	return storage;
10417 }
10418 EXPORT_SYMBOL(dev_get_stats);
10419 
10420 /**
10421  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10422  *	@s: place to store stats
10423  *	@netstats: per-cpu network stats to read from
10424  *
10425  *	Read per-cpu network statistics and populate the related fields in @s.
10426  */
10427 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10428 			   const struct pcpu_sw_netstats __percpu *netstats)
10429 {
10430 	int cpu;
10431 
10432 	for_each_possible_cpu(cpu) {
10433 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10434 		const struct pcpu_sw_netstats *stats;
10435 		unsigned int start;
10436 
10437 		stats = per_cpu_ptr(netstats, cpu);
10438 		do {
10439 			start = u64_stats_fetch_begin(&stats->syncp);
10440 			rx_packets = u64_stats_read(&stats->rx_packets);
10441 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10442 			tx_packets = u64_stats_read(&stats->tx_packets);
10443 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10444 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10445 
10446 		s->rx_packets += rx_packets;
10447 		s->rx_bytes   += rx_bytes;
10448 		s->tx_packets += tx_packets;
10449 		s->tx_bytes   += tx_bytes;
10450 	}
10451 }
10452 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10453 
10454 /**
10455  *	dev_get_tstats64 - ndo_get_stats64 implementation
10456  *	@dev: device to get statistics from
10457  *	@s: place to store stats
10458  *
10459  *	Populate @s from dev->stats and dev->tstats. Can be used as
10460  *	ndo_get_stats64() callback.
10461  */
10462 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10463 {
10464 	netdev_stats_to_stats64(s, &dev->stats);
10465 	dev_fetch_sw_netstats(s, dev->tstats);
10466 }
10467 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10468 
10469 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10470 {
10471 	struct netdev_queue *queue = dev_ingress_queue(dev);
10472 
10473 #ifdef CONFIG_NET_CLS_ACT
10474 	if (queue)
10475 		return queue;
10476 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10477 	if (!queue)
10478 		return NULL;
10479 	netdev_init_one_queue(dev, queue, NULL);
10480 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10481 	RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10482 	rcu_assign_pointer(dev->ingress_queue, queue);
10483 #endif
10484 	return queue;
10485 }
10486 
10487 static const struct ethtool_ops default_ethtool_ops;
10488 
10489 void netdev_set_default_ethtool_ops(struct net_device *dev,
10490 				    const struct ethtool_ops *ops)
10491 {
10492 	if (dev->ethtool_ops == &default_ethtool_ops)
10493 		dev->ethtool_ops = ops;
10494 }
10495 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10496 
10497 /**
10498  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10499  * @dev: netdev to enable the IRQ coalescing on
10500  *
10501  * Sets a conservative default for SW IRQ coalescing. Users can use
10502  * sysfs attributes to override the default values.
10503  */
10504 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10505 {
10506 	WARN_ON(dev->reg_state == NETREG_REGISTERED);
10507 
10508 	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
10509 		dev->gro_flush_timeout = 20000;
10510 		dev->napi_defer_hard_irqs = 1;
10511 	}
10512 }
10513 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10514 
10515 void netdev_freemem(struct net_device *dev)
10516 {
10517 	char *addr = (char *)dev - dev->padded;
10518 
10519 	kvfree(addr);
10520 }
10521 
10522 /**
10523  * alloc_netdev_mqs - allocate network device
10524  * @sizeof_priv: size of private data to allocate space for
10525  * @name: device name format string
10526  * @name_assign_type: origin of device name
10527  * @setup: callback to initialize device
10528  * @txqs: the number of TX subqueues to allocate
10529  * @rxqs: the number of RX subqueues to allocate
10530  *
10531  * Allocates a struct net_device with private data area for driver use
10532  * and performs basic initialization.  Also allocates subqueue structs
10533  * for each queue on the device.
10534  */
10535 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10536 		unsigned char name_assign_type,
10537 		void (*setup)(struct net_device *),
10538 		unsigned int txqs, unsigned int rxqs)
10539 {
10540 	struct net_device *dev;
10541 	unsigned int alloc_size;
10542 	struct net_device *p;
10543 
10544 	BUG_ON(strlen(name) >= sizeof(dev->name));
10545 
10546 	if (txqs < 1) {
10547 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10548 		return NULL;
10549 	}
10550 
10551 	if (rxqs < 1) {
10552 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10553 		return NULL;
10554 	}
10555 
10556 	alloc_size = sizeof(struct net_device);
10557 	if (sizeof_priv) {
10558 		/* ensure 32-byte alignment of private area */
10559 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10560 		alloc_size += sizeof_priv;
10561 	}
10562 	/* ensure 32-byte alignment of whole construct */
10563 	alloc_size += NETDEV_ALIGN - 1;
10564 
10565 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10566 	if (!p)
10567 		return NULL;
10568 
10569 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10570 	dev->padded = (char *)dev - (char *)p;
10571 
10572 	ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
10573 #ifdef CONFIG_PCPU_DEV_REFCNT
10574 	dev->pcpu_refcnt = alloc_percpu(int);
10575 	if (!dev->pcpu_refcnt)
10576 		goto free_dev;
10577 	__dev_hold(dev);
10578 #else
10579 	refcount_set(&dev->dev_refcnt, 1);
10580 #endif
10581 
10582 	if (dev_addr_init(dev))
10583 		goto free_pcpu;
10584 
10585 	dev_mc_init(dev);
10586 	dev_uc_init(dev);
10587 
10588 	dev_net_set(dev, &init_net);
10589 
10590 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10591 	dev->gso_max_segs = GSO_MAX_SEGS;
10592 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10593 	dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
10594 	dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
10595 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10596 	dev->tso_max_segs = TSO_MAX_SEGS;
10597 	dev->upper_level = 1;
10598 	dev->lower_level = 1;
10599 #ifdef CONFIG_LOCKDEP
10600 	dev->nested_level = 0;
10601 	INIT_LIST_HEAD(&dev->unlink_list);
10602 #endif
10603 
10604 	INIT_LIST_HEAD(&dev->napi_list);
10605 	INIT_LIST_HEAD(&dev->unreg_list);
10606 	INIT_LIST_HEAD(&dev->close_list);
10607 	INIT_LIST_HEAD(&dev->link_watch_list);
10608 	INIT_LIST_HEAD(&dev->adj_list.upper);
10609 	INIT_LIST_HEAD(&dev->adj_list.lower);
10610 	INIT_LIST_HEAD(&dev->ptype_all);
10611 	INIT_LIST_HEAD(&dev->ptype_specific);
10612 	INIT_LIST_HEAD(&dev->net_notifier_list);
10613 #ifdef CONFIG_NET_SCHED
10614 	hash_init(dev->qdisc_hash);
10615 #endif
10616 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10617 	setup(dev);
10618 
10619 	if (!dev->tx_queue_len) {
10620 		dev->priv_flags |= IFF_NO_QUEUE;
10621 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10622 	}
10623 
10624 	dev->num_tx_queues = txqs;
10625 	dev->real_num_tx_queues = txqs;
10626 	if (netif_alloc_netdev_queues(dev))
10627 		goto free_all;
10628 
10629 	dev->num_rx_queues = rxqs;
10630 	dev->real_num_rx_queues = rxqs;
10631 	if (netif_alloc_rx_queues(dev))
10632 		goto free_all;
10633 
10634 	strcpy(dev->name, name);
10635 	dev->name_assign_type = name_assign_type;
10636 	dev->group = INIT_NETDEV_GROUP;
10637 	if (!dev->ethtool_ops)
10638 		dev->ethtool_ops = &default_ethtool_ops;
10639 
10640 	nf_hook_netdev_init(dev);
10641 
10642 	return dev;
10643 
10644 free_all:
10645 	free_netdev(dev);
10646 	return NULL;
10647 
10648 free_pcpu:
10649 #ifdef CONFIG_PCPU_DEV_REFCNT
10650 	free_percpu(dev->pcpu_refcnt);
10651 free_dev:
10652 #endif
10653 	netdev_freemem(dev);
10654 	return NULL;
10655 }
10656 EXPORT_SYMBOL(alloc_netdev_mqs);
10657 
10658 /**
10659  * free_netdev - free network device
10660  * @dev: device
10661  *
10662  * This function does the last stage of destroying an allocated device
10663  * interface. The reference to the device object is released. If this
10664  * is the last reference then it will be freed.Must be called in process
10665  * context.
10666  */
10667 void free_netdev(struct net_device *dev)
10668 {
10669 	struct napi_struct *p, *n;
10670 
10671 	might_sleep();
10672 
10673 	/* When called immediately after register_netdevice() failed the unwind
10674 	 * handling may still be dismantling the device. Handle that case by
10675 	 * deferring the free.
10676 	 */
10677 	if (dev->reg_state == NETREG_UNREGISTERING) {
10678 		ASSERT_RTNL();
10679 		dev->needs_free_netdev = true;
10680 		return;
10681 	}
10682 
10683 	netif_free_tx_queues(dev);
10684 	netif_free_rx_queues(dev);
10685 
10686 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10687 
10688 	/* Flush device addresses */
10689 	dev_addr_flush(dev);
10690 
10691 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10692 		netif_napi_del(p);
10693 
10694 	ref_tracker_dir_exit(&dev->refcnt_tracker);
10695 #ifdef CONFIG_PCPU_DEV_REFCNT
10696 	free_percpu(dev->pcpu_refcnt);
10697 	dev->pcpu_refcnt = NULL;
10698 #endif
10699 	free_percpu(dev->core_stats);
10700 	dev->core_stats = NULL;
10701 	free_percpu(dev->xdp_bulkq);
10702 	dev->xdp_bulkq = NULL;
10703 
10704 	/*  Compatibility with error handling in drivers */
10705 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10706 		netdev_freemem(dev);
10707 		return;
10708 	}
10709 
10710 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10711 	dev->reg_state = NETREG_RELEASED;
10712 
10713 	/* will free via device release */
10714 	put_device(&dev->dev);
10715 }
10716 EXPORT_SYMBOL(free_netdev);
10717 
10718 /**
10719  *	synchronize_net -  Synchronize with packet receive processing
10720  *
10721  *	Wait for packets currently being received to be done.
10722  *	Does not block later packets from starting.
10723  */
10724 void synchronize_net(void)
10725 {
10726 	might_sleep();
10727 	if (rtnl_is_locked())
10728 		synchronize_rcu_expedited();
10729 	else
10730 		synchronize_rcu();
10731 }
10732 EXPORT_SYMBOL(synchronize_net);
10733 
10734 /**
10735  *	unregister_netdevice_queue - remove device from the kernel
10736  *	@dev: device
10737  *	@head: list
10738  *
10739  *	This function shuts down a device interface and removes it
10740  *	from the kernel tables.
10741  *	If head not NULL, device is queued to be unregistered later.
10742  *
10743  *	Callers must hold the rtnl semaphore.  You may want
10744  *	unregister_netdev() instead of this.
10745  */
10746 
10747 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10748 {
10749 	ASSERT_RTNL();
10750 
10751 	if (head) {
10752 		list_move_tail(&dev->unreg_list, head);
10753 	} else {
10754 		LIST_HEAD(single);
10755 
10756 		list_add(&dev->unreg_list, &single);
10757 		unregister_netdevice_many(&single);
10758 	}
10759 }
10760 EXPORT_SYMBOL(unregister_netdevice_queue);
10761 
10762 void unregister_netdevice_many_notify(struct list_head *head,
10763 				      u32 portid, const struct nlmsghdr *nlh)
10764 {
10765 	struct net_device *dev, *tmp;
10766 	LIST_HEAD(close_head);
10767 
10768 	BUG_ON(dev_boot_phase);
10769 	ASSERT_RTNL();
10770 
10771 	if (list_empty(head))
10772 		return;
10773 
10774 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10775 		/* Some devices call without registering
10776 		 * for initialization unwind. Remove those
10777 		 * devices and proceed with the remaining.
10778 		 */
10779 		if (dev->reg_state == NETREG_UNINITIALIZED) {
10780 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10781 				 dev->name, dev);
10782 
10783 			WARN_ON(1);
10784 			list_del(&dev->unreg_list);
10785 			continue;
10786 		}
10787 		dev->dismantle = true;
10788 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
10789 	}
10790 
10791 	/* If device is running, close it first. */
10792 	list_for_each_entry(dev, head, unreg_list)
10793 		list_add_tail(&dev->close_list, &close_head);
10794 	dev_close_many(&close_head, true);
10795 
10796 	list_for_each_entry(dev, head, unreg_list) {
10797 		/* And unlink it from device chain. */
10798 		write_lock(&dev_base_lock);
10799 		unlist_netdevice(dev, false);
10800 		dev->reg_state = NETREG_UNREGISTERING;
10801 		write_unlock(&dev_base_lock);
10802 	}
10803 	flush_all_backlogs();
10804 
10805 	synchronize_net();
10806 
10807 	list_for_each_entry(dev, head, unreg_list) {
10808 		struct sk_buff *skb = NULL;
10809 
10810 		/* Shutdown queueing discipline. */
10811 		dev_shutdown(dev);
10812 
10813 		dev_xdp_uninstall(dev);
10814 		bpf_dev_bound_netdev_unregister(dev);
10815 
10816 		netdev_offload_xstats_disable_all(dev);
10817 
10818 		/* Notify protocols, that we are about to destroy
10819 		 * this device. They should clean all the things.
10820 		 */
10821 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10822 
10823 		if (!dev->rtnl_link_ops ||
10824 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10825 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10826 						     GFP_KERNEL, NULL, 0,
10827 						     portid, nlh);
10828 
10829 		/*
10830 		 *	Flush the unicast and multicast chains
10831 		 */
10832 		dev_uc_flush(dev);
10833 		dev_mc_flush(dev);
10834 
10835 		netdev_name_node_alt_flush(dev);
10836 		netdev_name_node_free(dev->name_node);
10837 
10838 		call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10839 
10840 		if (dev->netdev_ops->ndo_uninit)
10841 			dev->netdev_ops->ndo_uninit(dev);
10842 
10843 		if (skb)
10844 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10845 
10846 		/* Notifier chain MUST detach us all upper devices. */
10847 		WARN_ON(netdev_has_any_upper_dev(dev));
10848 		WARN_ON(netdev_has_any_lower_dev(dev));
10849 
10850 		/* Remove entries from kobject tree */
10851 		netdev_unregister_kobject(dev);
10852 #ifdef CONFIG_XPS
10853 		/* Remove XPS queueing entries */
10854 		netif_reset_xps_queues_gt(dev, 0);
10855 #endif
10856 	}
10857 
10858 	synchronize_net();
10859 
10860 	list_for_each_entry(dev, head, unreg_list) {
10861 		netdev_put(dev, &dev->dev_registered_tracker);
10862 		net_set_todo(dev);
10863 	}
10864 
10865 	list_del(head);
10866 }
10867 
10868 /**
10869  *	unregister_netdevice_many - unregister many devices
10870  *	@head: list of devices
10871  *
10872  *  Note: As most callers use a stack allocated list_head,
10873  *  we force a list_del() to make sure stack wont be corrupted later.
10874  */
10875 void unregister_netdevice_many(struct list_head *head)
10876 {
10877 	unregister_netdevice_many_notify(head, 0, NULL);
10878 }
10879 EXPORT_SYMBOL(unregister_netdevice_many);
10880 
10881 /**
10882  *	unregister_netdev - remove device from the kernel
10883  *	@dev: device
10884  *
10885  *	This function shuts down a device interface and removes it
10886  *	from the kernel tables.
10887  *
10888  *	This is just a wrapper for unregister_netdevice that takes
10889  *	the rtnl semaphore.  In general you want to use this and not
10890  *	unregister_netdevice.
10891  */
10892 void unregister_netdev(struct net_device *dev)
10893 {
10894 	rtnl_lock();
10895 	unregister_netdevice(dev);
10896 	rtnl_unlock();
10897 }
10898 EXPORT_SYMBOL(unregister_netdev);
10899 
10900 /**
10901  *	__dev_change_net_namespace - move device to different nethost namespace
10902  *	@dev: device
10903  *	@net: network namespace
10904  *	@pat: If not NULL name pattern to try if the current device name
10905  *	      is already taken in the destination network namespace.
10906  *	@new_ifindex: If not zero, specifies device index in the target
10907  *	              namespace.
10908  *
10909  *	This function shuts down a device interface and moves it
10910  *	to a new network namespace. On success 0 is returned, on
10911  *	a failure a netagive errno code is returned.
10912  *
10913  *	Callers must hold the rtnl semaphore.
10914  */
10915 
10916 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10917 			       const char *pat, int new_ifindex)
10918 {
10919 	struct net *net_old = dev_net(dev);
10920 	int err, new_nsid;
10921 
10922 	ASSERT_RTNL();
10923 
10924 	/* Don't allow namespace local devices to be moved. */
10925 	err = -EINVAL;
10926 	if (dev->features & NETIF_F_NETNS_LOCAL)
10927 		goto out;
10928 
10929 	/* Ensure the device has been registrered */
10930 	if (dev->reg_state != NETREG_REGISTERED)
10931 		goto out;
10932 
10933 	/* Get out if there is nothing todo */
10934 	err = 0;
10935 	if (net_eq(net_old, net))
10936 		goto out;
10937 
10938 	/* Pick the destination device name, and ensure
10939 	 * we can use it in the destination network namespace.
10940 	 */
10941 	err = -EEXIST;
10942 	if (netdev_name_in_use(net, dev->name)) {
10943 		/* We get here if we can't use the current device name */
10944 		if (!pat)
10945 			goto out;
10946 		err = dev_get_valid_name(net, dev, pat);
10947 		if (err < 0)
10948 			goto out;
10949 	}
10950 
10951 	/* Check that new_ifindex isn't used yet. */
10952 	err = -EBUSY;
10953 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10954 		goto out;
10955 
10956 	/*
10957 	 * And now a mini version of register_netdevice unregister_netdevice.
10958 	 */
10959 
10960 	/* If device is running close it first. */
10961 	dev_close(dev);
10962 
10963 	/* And unlink it from device chain */
10964 	unlist_netdevice(dev, true);
10965 
10966 	synchronize_net();
10967 
10968 	/* Shutdown queueing discipline. */
10969 	dev_shutdown(dev);
10970 
10971 	/* Notify protocols, that we are about to destroy
10972 	 * this device. They should clean all the things.
10973 	 *
10974 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
10975 	 * This is wanted because this way 8021q and macvlan know
10976 	 * the device is just moving and can keep their slaves up.
10977 	 */
10978 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10979 	rcu_barrier();
10980 
10981 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10982 	/* If there is an ifindex conflict assign a new one */
10983 	if (!new_ifindex) {
10984 		if (__dev_get_by_index(net, dev->ifindex))
10985 			new_ifindex = dev_new_index(net);
10986 		else
10987 			new_ifindex = dev->ifindex;
10988 	}
10989 
10990 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10991 			    new_ifindex);
10992 
10993 	/*
10994 	 *	Flush the unicast and multicast chains
10995 	 */
10996 	dev_uc_flush(dev);
10997 	dev_mc_flush(dev);
10998 
10999 	/* Send a netdev-removed uevent to the old namespace */
11000 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11001 	netdev_adjacent_del_links(dev);
11002 
11003 	/* Move per-net netdevice notifiers that are following the netdevice */
11004 	move_netdevice_notifiers_dev_net(dev, net);
11005 
11006 	/* Actually switch the network namespace */
11007 	dev_net_set(dev, net);
11008 	dev->ifindex = new_ifindex;
11009 
11010 	/* Send a netdev-add uevent to the new namespace */
11011 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11012 	netdev_adjacent_add_links(dev);
11013 
11014 	/* Fixup kobjects */
11015 	err = device_rename(&dev->dev, dev->name);
11016 	WARN_ON(err);
11017 
11018 	/* Adapt owner in case owning user namespace of target network
11019 	 * namespace is different from the original one.
11020 	 */
11021 	err = netdev_change_owner(dev, net_old, net);
11022 	WARN_ON(err);
11023 
11024 	/* Add the device back in the hashes */
11025 	list_netdevice(dev);
11026 
11027 	/* Notify protocols, that a new device appeared. */
11028 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11029 
11030 	/*
11031 	 *	Prevent userspace races by waiting until the network
11032 	 *	device is fully setup before sending notifications.
11033 	 */
11034 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11035 
11036 	synchronize_net();
11037 	err = 0;
11038 out:
11039 	return err;
11040 }
11041 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11042 
11043 static int dev_cpu_dead(unsigned int oldcpu)
11044 {
11045 	struct sk_buff **list_skb;
11046 	struct sk_buff *skb;
11047 	unsigned int cpu;
11048 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11049 
11050 	local_irq_disable();
11051 	cpu = smp_processor_id();
11052 	sd = &per_cpu(softnet_data, cpu);
11053 	oldsd = &per_cpu(softnet_data, oldcpu);
11054 
11055 	/* Find end of our completion_queue. */
11056 	list_skb = &sd->completion_queue;
11057 	while (*list_skb)
11058 		list_skb = &(*list_skb)->next;
11059 	/* Append completion queue from offline CPU. */
11060 	*list_skb = oldsd->completion_queue;
11061 	oldsd->completion_queue = NULL;
11062 
11063 	/* Append output queue from offline CPU. */
11064 	if (oldsd->output_queue) {
11065 		*sd->output_queue_tailp = oldsd->output_queue;
11066 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11067 		oldsd->output_queue = NULL;
11068 		oldsd->output_queue_tailp = &oldsd->output_queue;
11069 	}
11070 	/* Append NAPI poll list from offline CPU, with one exception :
11071 	 * process_backlog() must be called by cpu owning percpu backlog.
11072 	 * We properly handle process_queue & input_pkt_queue later.
11073 	 */
11074 	while (!list_empty(&oldsd->poll_list)) {
11075 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11076 							    struct napi_struct,
11077 							    poll_list);
11078 
11079 		list_del_init(&napi->poll_list);
11080 		if (napi->poll == process_backlog)
11081 			napi->state = 0;
11082 		else
11083 			____napi_schedule(sd, napi);
11084 	}
11085 
11086 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11087 	local_irq_enable();
11088 
11089 #ifdef CONFIG_RPS
11090 	remsd = oldsd->rps_ipi_list;
11091 	oldsd->rps_ipi_list = NULL;
11092 #endif
11093 	/* send out pending IPI's on offline CPU */
11094 	net_rps_send_ipi(remsd);
11095 
11096 	/* Process offline CPU's input_pkt_queue */
11097 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11098 		netif_rx(skb);
11099 		input_queue_head_incr(oldsd);
11100 	}
11101 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11102 		netif_rx(skb);
11103 		input_queue_head_incr(oldsd);
11104 	}
11105 
11106 	return 0;
11107 }
11108 
11109 /**
11110  *	netdev_increment_features - increment feature set by one
11111  *	@all: current feature set
11112  *	@one: new feature set
11113  *	@mask: mask feature set
11114  *
11115  *	Computes a new feature set after adding a device with feature set
11116  *	@one to the master device with current feature set @all.  Will not
11117  *	enable anything that is off in @mask. Returns the new feature set.
11118  */
11119 netdev_features_t netdev_increment_features(netdev_features_t all,
11120 	netdev_features_t one, netdev_features_t mask)
11121 {
11122 	if (mask & NETIF_F_HW_CSUM)
11123 		mask |= NETIF_F_CSUM_MASK;
11124 	mask |= NETIF_F_VLAN_CHALLENGED;
11125 
11126 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11127 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11128 
11129 	/* If one device supports hw checksumming, set for all. */
11130 	if (all & NETIF_F_HW_CSUM)
11131 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11132 
11133 	return all;
11134 }
11135 EXPORT_SYMBOL(netdev_increment_features);
11136 
11137 static struct hlist_head * __net_init netdev_create_hash(void)
11138 {
11139 	int i;
11140 	struct hlist_head *hash;
11141 
11142 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11143 	if (hash != NULL)
11144 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11145 			INIT_HLIST_HEAD(&hash[i]);
11146 
11147 	return hash;
11148 }
11149 
11150 /* Initialize per network namespace state */
11151 static int __net_init netdev_init(struct net *net)
11152 {
11153 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11154 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11155 
11156 	INIT_LIST_HEAD(&net->dev_base_head);
11157 
11158 	net->dev_name_head = netdev_create_hash();
11159 	if (net->dev_name_head == NULL)
11160 		goto err_name;
11161 
11162 	net->dev_index_head = netdev_create_hash();
11163 	if (net->dev_index_head == NULL)
11164 		goto err_idx;
11165 
11166 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11167 
11168 	return 0;
11169 
11170 err_idx:
11171 	kfree(net->dev_name_head);
11172 err_name:
11173 	return -ENOMEM;
11174 }
11175 
11176 /**
11177  *	netdev_drivername - network driver for the device
11178  *	@dev: network device
11179  *
11180  *	Determine network driver for device.
11181  */
11182 const char *netdev_drivername(const struct net_device *dev)
11183 {
11184 	const struct device_driver *driver;
11185 	const struct device *parent;
11186 	const char *empty = "";
11187 
11188 	parent = dev->dev.parent;
11189 	if (!parent)
11190 		return empty;
11191 
11192 	driver = parent->driver;
11193 	if (driver && driver->name)
11194 		return driver->name;
11195 	return empty;
11196 }
11197 
11198 static void __netdev_printk(const char *level, const struct net_device *dev,
11199 			    struct va_format *vaf)
11200 {
11201 	if (dev && dev->dev.parent) {
11202 		dev_printk_emit(level[1] - '0',
11203 				dev->dev.parent,
11204 				"%s %s %s%s: %pV",
11205 				dev_driver_string(dev->dev.parent),
11206 				dev_name(dev->dev.parent),
11207 				netdev_name(dev), netdev_reg_state(dev),
11208 				vaf);
11209 	} else if (dev) {
11210 		printk("%s%s%s: %pV",
11211 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11212 	} else {
11213 		printk("%s(NULL net_device): %pV", level, vaf);
11214 	}
11215 }
11216 
11217 void netdev_printk(const char *level, const struct net_device *dev,
11218 		   const char *format, ...)
11219 {
11220 	struct va_format vaf;
11221 	va_list args;
11222 
11223 	va_start(args, format);
11224 
11225 	vaf.fmt = format;
11226 	vaf.va = &args;
11227 
11228 	__netdev_printk(level, dev, &vaf);
11229 
11230 	va_end(args);
11231 }
11232 EXPORT_SYMBOL(netdev_printk);
11233 
11234 #define define_netdev_printk_level(func, level)			\
11235 void func(const struct net_device *dev, const char *fmt, ...)	\
11236 {								\
11237 	struct va_format vaf;					\
11238 	va_list args;						\
11239 								\
11240 	va_start(args, fmt);					\
11241 								\
11242 	vaf.fmt = fmt;						\
11243 	vaf.va = &args;						\
11244 								\
11245 	__netdev_printk(level, dev, &vaf);			\
11246 								\
11247 	va_end(args);						\
11248 }								\
11249 EXPORT_SYMBOL(func);
11250 
11251 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11252 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11253 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11254 define_netdev_printk_level(netdev_err, KERN_ERR);
11255 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11256 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11257 define_netdev_printk_level(netdev_info, KERN_INFO);
11258 
11259 static void __net_exit netdev_exit(struct net *net)
11260 {
11261 	kfree(net->dev_name_head);
11262 	kfree(net->dev_index_head);
11263 	if (net != &init_net)
11264 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11265 }
11266 
11267 static struct pernet_operations __net_initdata netdev_net_ops = {
11268 	.init = netdev_init,
11269 	.exit = netdev_exit,
11270 };
11271 
11272 static void __net_exit default_device_exit_net(struct net *net)
11273 {
11274 	struct net_device *dev, *aux;
11275 	/*
11276 	 * Push all migratable network devices back to the
11277 	 * initial network namespace
11278 	 */
11279 	ASSERT_RTNL();
11280 	for_each_netdev_safe(net, dev, aux) {
11281 		int err;
11282 		char fb_name[IFNAMSIZ];
11283 
11284 		/* Ignore unmoveable devices (i.e. loopback) */
11285 		if (dev->features & NETIF_F_NETNS_LOCAL)
11286 			continue;
11287 
11288 		/* Leave virtual devices for the generic cleanup */
11289 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11290 			continue;
11291 
11292 		/* Push remaining network devices to init_net */
11293 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11294 		if (netdev_name_in_use(&init_net, fb_name))
11295 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11296 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11297 		if (err) {
11298 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11299 				 __func__, dev->name, err);
11300 			BUG();
11301 		}
11302 	}
11303 }
11304 
11305 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11306 {
11307 	/* At exit all network devices most be removed from a network
11308 	 * namespace.  Do this in the reverse order of registration.
11309 	 * Do this across as many network namespaces as possible to
11310 	 * improve batching efficiency.
11311 	 */
11312 	struct net_device *dev;
11313 	struct net *net;
11314 	LIST_HEAD(dev_kill_list);
11315 
11316 	rtnl_lock();
11317 	list_for_each_entry(net, net_list, exit_list) {
11318 		default_device_exit_net(net);
11319 		cond_resched();
11320 	}
11321 
11322 	list_for_each_entry(net, net_list, exit_list) {
11323 		for_each_netdev_reverse(net, dev) {
11324 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11325 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11326 			else
11327 				unregister_netdevice_queue(dev, &dev_kill_list);
11328 		}
11329 	}
11330 	unregister_netdevice_many(&dev_kill_list);
11331 	rtnl_unlock();
11332 }
11333 
11334 static struct pernet_operations __net_initdata default_device_ops = {
11335 	.exit_batch = default_device_exit_batch,
11336 };
11337 
11338 /*
11339  *	Initialize the DEV module. At boot time this walks the device list and
11340  *	unhooks any devices that fail to initialise (normally hardware not
11341  *	present) and leaves us with a valid list of present and active devices.
11342  *
11343  */
11344 
11345 /*
11346  *       This is called single threaded during boot, so no need
11347  *       to take the rtnl semaphore.
11348  */
11349 static int __init net_dev_init(void)
11350 {
11351 	int i, rc = -ENOMEM;
11352 
11353 	BUG_ON(!dev_boot_phase);
11354 
11355 	if (dev_proc_init())
11356 		goto out;
11357 
11358 	if (netdev_kobject_init())
11359 		goto out;
11360 
11361 	INIT_LIST_HEAD(&ptype_all);
11362 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11363 		INIT_LIST_HEAD(&ptype_base[i]);
11364 
11365 	if (register_pernet_subsys(&netdev_net_ops))
11366 		goto out;
11367 
11368 	/*
11369 	 *	Initialise the packet receive queues.
11370 	 */
11371 
11372 	for_each_possible_cpu(i) {
11373 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11374 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11375 
11376 		INIT_WORK(flush, flush_backlog);
11377 
11378 		skb_queue_head_init(&sd->input_pkt_queue);
11379 		skb_queue_head_init(&sd->process_queue);
11380 #ifdef CONFIG_XFRM_OFFLOAD
11381 		skb_queue_head_init(&sd->xfrm_backlog);
11382 #endif
11383 		INIT_LIST_HEAD(&sd->poll_list);
11384 		sd->output_queue_tailp = &sd->output_queue;
11385 #ifdef CONFIG_RPS
11386 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11387 		sd->cpu = i;
11388 #endif
11389 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11390 		spin_lock_init(&sd->defer_lock);
11391 
11392 		init_gro_hash(&sd->backlog);
11393 		sd->backlog.poll = process_backlog;
11394 		sd->backlog.weight = weight_p;
11395 	}
11396 
11397 	dev_boot_phase = 0;
11398 
11399 	/* The loopback device is special if any other network devices
11400 	 * is present in a network namespace the loopback device must
11401 	 * be present. Since we now dynamically allocate and free the
11402 	 * loopback device ensure this invariant is maintained by
11403 	 * keeping the loopback device as the first device on the
11404 	 * list of network devices.  Ensuring the loopback devices
11405 	 * is the first device that appears and the last network device
11406 	 * that disappears.
11407 	 */
11408 	if (register_pernet_device(&loopback_net_ops))
11409 		goto out;
11410 
11411 	if (register_pernet_device(&default_device_ops))
11412 		goto out;
11413 
11414 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11415 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11416 
11417 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11418 				       NULL, dev_cpu_dead);
11419 	WARN_ON(rc < 0);
11420 	rc = 0;
11421 out:
11422 	return rc;
11423 }
11424 
11425 subsys_initcall(net_dev_init);
11426