xref: /linux/net/core/dev.c (revision a751449f8b477e0e1d97f778ed97ae9f6576b690)
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_ingress.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 "net-sysfs.h"
155 
156 #define MAX_GRO_SKBS 8
157 
158 /* This should be increased if a protocol with a bigger head is added. */
159 #define GRO_MAX_HEAD (MAX_HEADER + 128)
160 
161 static DEFINE_SPINLOCK(ptype_lock);
162 static DEFINE_SPINLOCK(offload_lock);
163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
164 struct list_head ptype_all __read_mostly;	/* Taps */
165 static struct list_head offload_base __read_mostly;
166 
167 static int netif_rx_internal(struct sk_buff *skb);
168 static int call_netdevice_notifiers_info(unsigned long val,
169 					 struct netdev_notifier_info *info);
170 static int call_netdevice_notifiers_extack(unsigned long val,
171 					   struct net_device *dev,
172 					   struct netlink_ext_ack *extack);
173 static struct napi_struct *napi_by_id(unsigned int napi_id);
174 
175 /*
176  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177  * semaphore.
178  *
179  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
180  *
181  * Writers must hold the rtnl semaphore while they loop through the
182  * dev_base_head list, and hold dev_base_lock for writing when they do the
183  * actual updates.  This allows pure readers to access the list even
184  * while a writer is preparing to update it.
185  *
186  * To put it another way, dev_base_lock is held for writing only to
187  * protect against pure readers; the rtnl semaphore provides the
188  * protection against other writers.
189  *
190  * See, for example usages, register_netdevice() and
191  * unregister_netdevice(), which must be called with the rtnl
192  * semaphore held.
193  */
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
196 
197 static DEFINE_MUTEX(ifalias_mutex);
198 
199 /* protects napi_hash addition/deletion and napi_gen_id */
200 static DEFINE_SPINLOCK(napi_hash_lock);
201 
202 static unsigned int napi_gen_id = NR_CPUS;
203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
204 
205 static DECLARE_RWSEM(devnet_rename_sem);
206 
207 static inline void dev_base_seq_inc(struct net *net)
208 {
209 	while (++net->dev_base_seq == 0)
210 		;
211 }
212 
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 {
215 	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
216 
217 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 }
219 
220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
221 {
222 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 }
224 
225 static inline void rps_lock(struct softnet_data *sd)
226 {
227 #ifdef CONFIG_RPS
228 	spin_lock(&sd->input_pkt_queue.lock);
229 #endif
230 }
231 
232 static inline void rps_unlock(struct softnet_data *sd)
233 {
234 #ifdef CONFIG_RPS
235 	spin_unlock(&sd->input_pkt_queue.lock);
236 #endif
237 }
238 
239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
240 						       const char *name)
241 {
242 	struct netdev_name_node *name_node;
243 
244 	name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
245 	if (!name_node)
246 		return NULL;
247 	INIT_HLIST_NODE(&name_node->hlist);
248 	name_node->dev = dev;
249 	name_node->name = name;
250 	return name_node;
251 }
252 
253 static struct netdev_name_node *
254 netdev_name_node_head_alloc(struct net_device *dev)
255 {
256 	struct netdev_name_node *name_node;
257 
258 	name_node = netdev_name_node_alloc(dev, dev->name);
259 	if (!name_node)
260 		return NULL;
261 	INIT_LIST_HEAD(&name_node->list);
262 	return name_node;
263 }
264 
265 static void netdev_name_node_free(struct netdev_name_node *name_node)
266 {
267 	kfree(name_node);
268 }
269 
270 static void netdev_name_node_add(struct net *net,
271 				 struct netdev_name_node *name_node)
272 {
273 	hlist_add_head_rcu(&name_node->hlist,
274 			   dev_name_hash(net, name_node->name));
275 }
276 
277 static void netdev_name_node_del(struct netdev_name_node *name_node)
278 {
279 	hlist_del_rcu(&name_node->hlist);
280 }
281 
282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
283 							const char *name)
284 {
285 	struct hlist_head *head = dev_name_hash(net, name);
286 	struct netdev_name_node *name_node;
287 
288 	hlist_for_each_entry(name_node, head, hlist)
289 		if (!strcmp(name_node->name, name))
290 			return name_node;
291 	return NULL;
292 }
293 
294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
295 							    const char *name)
296 {
297 	struct hlist_head *head = dev_name_hash(net, name);
298 	struct netdev_name_node *name_node;
299 
300 	hlist_for_each_entry_rcu(name_node, head, hlist)
301 		if (!strcmp(name_node->name, name))
302 			return name_node;
303 	return NULL;
304 }
305 
306 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
307 {
308 	struct netdev_name_node *name_node;
309 	struct net *net = dev_net(dev);
310 
311 	name_node = netdev_name_node_lookup(net, name);
312 	if (name_node)
313 		return -EEXIST;
314 	name_node = netdev_name_node_alloc(dev, name);
315 	if (!name_node)
316 		return -ENOMEM;
317 	netdev_name_node_add(net, name_node);
318 	/* The node that holds dev->name acts as a head of per-device list. */
319 	list_add_tail(&name_node->list, &dev->name_node->list);
320 
321 	return 0;
322 }
323 EXPORT_SYMBOL(netdev_name_node_alt_create);
324 
325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
326 {
327 	list_del(&name_node->list);
328 	netdev_name_node_del(name_node);
329 	kfree(name_node->name);
330 	netdev_name_node_free(name_node);
331 }
332 
333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
334 {
335 	struct netdev_name_node *name_node;
336 	struct net *net = dev_net(dev);
337 
338 	name_node = netdev_name_node_lookup(net, name);
339 	if (!name_node)
340 		return -ENOENT;
341 	/* lookup might have found our primary name or a name belonging
342 	 * to another device.
343 	 */
344 	if (name_node == dev->name_node || name_node->dev != dev)
345 		return -EINVAL;
346 
347 	__netdev_name_node_alt_destroy(name_node);
348 
349 	return 0;
350 }
351 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
352 
353 static void netdev_name_node_alt_flush(struct net_device *dev)
354 {
355 	struct netdev_name_node *name_node, *tmp;
356 
357 	list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
358 		__netdev_name_node_alt_destroy(name_node);
359 }
360 
361 /* Device list insertion */
362 static void list_netdevice(struct net_device *dev)
363 {
364 	struct net *net = dev_net(dev);
365 
366 	ASSERT_RTNL();
367 
368 	write_lock_bh(&dev_base_lock);
369 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
370 	netdev_name_node_add(net, dev->name_node);
371 	hlist_add_head_rcu(&dev->index_hlist,
372 			   dev_index_hash(net, dev->ifindex));
373 	write_unlock_bh(&dev_base_lock);
374 
375 	dev_base_seq_inc(net);
376 }
377 
378 /* Device list removal
379  * caller must respect a RCU grace period before freeing/reusing dev
380  */
381 static void unlist_netdevice(struct net_device *dev)
382 {
383 	ASSERT_RTNL();
384 
385 	/* Unlink dev from the device chain */
386 	write_lock_bh(&dev_base_lock);
387 	list_del_rcu(&dev->dev_list);
388 	netdev_name_node_del(dev->name_node);
389 	hlist_del_rcu(&dev->index_hlist);
390 	write_unlock_bh(&dev_base_lock);
391 
392 	dev_base_seq_inc(dev_net(dev));
393 }
394 
395 /*
396  *	Our notifier list
397  */
398 
399 static RAW_NOTIFIER_HEAD(netdev_chain);
400 
401 /*
402  *	Device drivers call our routines to queue packets here. We empty the
403  *	queue in the local softnet handler.
404  */
405 
406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
407 EXPORT_PER_CPU_SYMBOL(softnet_data);
408 
409 #ifdef CONFIG_LOCKDEP
410 /*
411  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
412  * according to dev->type
413  */
414 static const unsigned short netdev_lock_type[] = {
415 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
416 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
417 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
418 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
419 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
420 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
421 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
422 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
423 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
424 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
425 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
426 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
427 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
428 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
429 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
430 
431 static const char *const netdev_lock_name[] = {
432 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
433 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
434 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
435 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
436 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
437 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
438 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
439 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
440 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
441 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
442 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
443 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
444 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
445 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
446 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
447 
448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
450 
451 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
452 {
453 	int i;
454 
455 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
456 		if (netdev_lock_type[i] == dev_type)
457 			return i;
458 	/* the last key is used by default */
459 	return ARRAY_SIZE(netdev_lock_type) - 1;
460 }
461 
462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
463 						 unsigned short dev_type)
464 {
465 	int i;
466 
467 	i = netdev_lock_pos(dev_type);
468 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
469 				   netdev_lock_name[i]);
470 }
471 
472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
473 {
474 	int i;
475 
476 	i = netdev_lock_pos(dev->type);
477 	lockdep_set_class_and_name(&dev->addr_list_lock,
478 				   &netdev_addr_lock_key[i],
479 				   netdev_lock_name[i]);
480 }
481 #else
482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
483 						 unsigned short dev_type)
484 {
485 }
486 
487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
488 {
489 }
490 #endif
491 
492 /*******************************************************************************
493  *
494  *		Protocol management and registration routines
495  *
496  *******************************************************************************/
497 
498 
499 /*
500  *	Add a protocol ID to the list. Now that the input handler is
501  *	smarter we can dispense with all the messy stuff that used to be
502  *	here.
503  *
504  *	BEWARE!!! Protocol handlers, mangling input packets,
505  *	MUST BE last in hash buckets and checking protocol handlers
506  *	MUST start from promiscuous ptype_all chain in net_bh.
507  *	It is true now, do not change it.
508  *	Explanation follows: if protocol handler, mangling packet, will
509  *	be the first on list, it is not able to sense, that packet
510  *	is cloned and should be copied-on-write, so that it will
511  *	change it and subsequent readers will get broken packet.
512  *							--ANK (980803)
513  */
514 
515 static inline struct list_head *ptype_head(const struct packet_type *pt)
516 {
517 	if (pt->type == htons(ETH_P_ALL))
518 		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
519 	else
520 		return pt->dev ? &pt->dev->ptype_specific :
521 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
522 }
523 
524 /**
525  *	dev_add_pack - add packet handler
526  *	@pt: packet type declaration
527  *
528  *	Add a protocol handler to the networking stack. The passed &packet_type
529  *	is linked into kernel lists and may not be freed until it has been
530  *	removed from the kernel lists.
531  *
532  *	This call does not sleep therefore it can not
533  *	guarantee all CPU's that are in middle of receiving packets
534  *	will see the new packet type (until the next received packet).
535  */
536 
537 void dev_add_pack(struct packet_type *pt)
538 {
539 	struct list_head *head = ptype_head(pt);
540 
541 	spin_lock(&ptype_lock);
542 	list_add_rcu(&pt->list, head);
543 	spin_unlock(&ptype_lock);
544 }
545 EXPORT_SYMBOL(dev_add_pack);
546 
547 /**
548  *	__dev_remove_pack	 - remove packet handler
549  *	@pt: packet type declaration
550  *
551  *	Remove a protocol handler that was previously added to the kernel
552  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
553  *	from the kernel lists and can be freed or reused once this function
554  *	returns.
555  *
556  *      The packet type might still be in use by receivers
557  *	and must not be freed until after all the CPU's have gone
558  *	through a quiescent state.
559  */
560 void __dev_remove_pack(struct packet_type *pt)
561 {
562 	struct list_head *head = ptype_head(pt);
563 	struct packet_type *pt1;
564 
565 	spin_lock(&ptype_lock);
566 
567 	list_for_each_entry(pt1, head, list) {
568 		if (pt == pt1) {
569 			list_del_rcu(&pt->list);
570 			goto out;
571 		}
572 	}
573 
574 	pr_warn("dev_remove_pack: %p not found\n", pt);
575 out:
576 	spin_unlock(&ptype_lock);
577 }
578 EXPORT_SYMBOL(__dev_remove_pack);
579 
580 /**
581  *	dev_remove_pack	 - remove packet handler
582  *	@pt: packet type declaration
583  *
584  *	Remove a protocol handler that was previously added to the kernel
585  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
586  *	from the kernel lists and can be freed or reused once this function
587  *	returns.
588  *
589  *	This call sleeps to guarantee that no CPU is looking at the packet
590  *	type after return.
591  */
592 void dev_remove_pack(struct packet_type *pt)
593 {
594 	__dev_remove_pack(pt);
595 
596 	synchronize_net();
597 }
598 EXPORT_SYMBOL(dev_remove_pack);
599 
600 
601 /**
602  *	dev_add_offload - register offload handlers
603  *	@po: protocol offload declaration
604  *
605  *	Add protocol offload handlers to the networking stack. The passed
606  *	&proto_offload is linked into kernel lists and may not be freed until
607  *	it has been removed from the kernel lists.
608  *
609  *	This call does not sleep therefore it can not
610  *	guarantee all CPU's that are in middle of receiving packets
611  *	will see the new offload handlers (until the next received packet).
612  */
613 void dev_add_offload(struct packet_offload *po)
614 {
615 	struct packet_offload *elem;
616 
617 	spin_lock(&offload_lock);
618 	list_for_each_entry(elem, &offload_base, list) {
619 		if (po->priority < elem->priority)
620 			break;
621 	}
622 	list_add_rcu(&po->list, elem->list.prev);
623 	spin_unlock(&offload_lock);
624 }
625 EXPORT_SYMBOL(dev_add_offload);
626 
627 /**
628  *	__dev_remove_offload	 - remove offload handler
629  *	@po: packet offload declaration
630  *
631  *	Remove a protocol offload handler that was previously added to the
632  *	kernel offload handlers by dev_add_offload(). The passed &offload_type
633  *	is removed from the kernel lists and can be freed or reused once this
634  *	function returns.
635  *
636  *      The packet type might still be in use by receivers
637  *	and must not be freed until after all the CPU's have gone
638  *	through a quiescent state.
639  */
640 static void __dev_remove_offload(struct packet_offload *po)
641 {
642 	struct list_head *head = &offload_base;
643 	struct packet_offload *po1;
644 
645 	spin_lock(&offload_lock);
646 
647 	list_for_each_entry(po1, head, list) {
648 		if (po == po1) {
649 			list_del_rcu(&po->list);
650 			goto out;
651 		}
652 	}
653 
654 	pr_warn("dev_remove_offload: %p not found\n", po);
655 out:
656 	spin_unlock(&offload_lock);
657 }
658 
659 /**
660  *	dev_remove_offload	 - remove packet offload handler
661  *	@po: packet offload declaration
662  *
663  *	Remove a packet offload handler that was previously added to the kernel
664  *	offload handlers by dev_add_offload(). The passed &offload_type is
665  *	removed from the kernel lists and can be freed or reused once this
666  *	function returns.
667  *
668  *	This call sleeps to guarantee that no CPU is looking at the packet
669  *	type after return.
670  */
671 void dev_remove_offload(struct packet_offload *po)
672 {
673 	__dev_remove_offload(po);
674 
675 	synchronize_net();
676 }
677 EXPORT_SYMBOL(dev_remove_offload);
678 
679 /******************************************************************************
680  *
681  *		      Device Boot-time Settings Routines
682  *
683  ******************************************************************************/
684 
685 /* Boot time configuration table */
686 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
687 
688 /**
689  *	netdev_boot_setup_add	- add new setup entry
690  *	@name: name of the device
691  *	@map: configured settings for the device
692  *
693  *	Adds new setup entry to the dev_boot_setup list.  The function
694  *	returns 0 on error and 1 on success.  This is a generic routine to
695  *	all netdevices.
696  */
697 static int netdev_boot_setup_add(char *name, struct ifmap *map)
698 {
699 	struct netdev_boot_setup *s;
700 	int i;
701 
702 	s = dev_boot_setup;
703 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
704 		if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
705 			memset(s[i].name, 0, sizeof(s[i].name));
706 			strlcpy(s[i].name, name, IFNAMSIZ);
707 			memcpy(&s[i].map, map, sizeof(s[i].map));
708 			break;
709 		}
710 	}
711 
712 	return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
713 }
714 
715 /**
716  * netdev_boot_setup_check	- check boot time settings
717  * @dev: the netdevice
718  *
719  * Check boot time settings for the device.
720  * The found settings are set for the device to be used
721  * later in the device probing.
722  * Returns 0 if no settings found, 1 if they are.
723  */
724 int netdev_boot_setup_check(struct net_device *dev)
725 {
726 	struct netdev_boot_setup *s = dev_boot_setup;
727 	int i;
728 
729 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
730 		if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
731 		    !strcmp(dev->name, s[i].name)) {
732 			dev->irq = s[i].map.irq;
733 			dev->base_addr = s[i].map.base_addr;
734 			dev->mem_start = s[i].map.mem_start;
735 			dev->mem_end = s[i].map.mem_end;
736 			return 1;
737 		}
738 	}
739 	return 0;
740 }
741 EXPORT_SYMBOL(netdev_boot_setup_check);
742 
743 
744 /**
745  * netdev_boot_base	- get address from boot time settings
746  * @prefix: prefix for network device
747  * @unit: id for network device
748  *
749  * Check boot time settings for the base address of device.
750  * The found settings are set for the device to be used
751  * later in the device probing.
752  * Returns 0 if no settings found.
753  */
754 unsigned long netdev_boot_base(const char *prefix, int unit)
755 {
756 	const struct netdev_boot_setup *s = dev_boot_setup;
757 	char name[IFNAMSIZ];
758 	int i;
759 
760 	sprintf(name, "%s%d", prefix, unit);
761 
762 	/*
763 	 * If device already registered then return base of 1
764 	 * to indicate not to probe for this interface
765 	 */
766 	if (__dev_get_by_name(&init_net, name))
767 		return 1;
768 
769 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
770 		if (!strcmp(name, s[i].name))
771 			return s[i].map.base_addr;
772 	return 0;
773 }
774 
775 /*
776  * Saves at boot time configured settings for any netdevice.
777  */
778 int __init netdev_boot_setup(char *str)
779 {
780 	int ints[5];
781 	struct ifmap map;
782 
783 	str = get_options(str, ARRAY_SIZE(ints), ints);
784 	if (!str || !*str)
785 		return 0;
786 
787 	/* Save settings */
788 	memset(&map, 0, sizeof(map));
789 	if (ints[0] > 0)
790 		map.irq = ints[1];
791 	if (ints[0] > 1)
792 		map.base_addr = ints[2];
793 	if (ints[0] > 2)
794 		map.mem_start = ints[3];
795 	if (ints[0] > 3)
796 		map.mem_end = ints[4];
797 
798 	/* Add new entry to the list */
799 	return netdev_boot_setup_add(str, &map);
800 }
801 
802 __setup("netdev=", netdev_boot_setup);
803 
804 /*******************************************************************************
805  *
806  *			    Device Interface Subroutines
807  *
808  *******************************************************************************/
809 
810 /**
811  *	dev_get_iflink	- get 'iflink' value of a interface
812  *	@dev: targeted interface
813  *
814  *	Indicates the ifindex the interface is linked to.
815  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
816  */
817 
818 int dev_get_iflink(const struct net_device *dev)
819 {
820 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
821 		return dev->netdev_ops->ndo_get_iflink(dev);
822 
823 	return dev->ifindex;
824 }
825 EXPORT_SYMBOL(dev_get_iflink);
826 
827 /**
828  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
829  *	@dev: targeted interface
830  *	@skb: The packet.
831  *
832  *	For better visibility of tunnel traffic OVS needs to retrieve
833  *	egress tunnel information for a packet. Following API allows
834  *	user to get this info.
835  */
836 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
837 {
838 	struct ip_tunnel_info *info;
839 
840 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
841 		return -EINVAL;
842 
843 	info = skb_tunnel_info_unclone(skb);
844 	if (!info)
845 		return -ENOMEM;
846 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
847 		return -EINVAL;
848 
849 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
850 }
851 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
852 
853 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
854 {
855 	int k = stack->num_paths++;
856 
857 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
858 		return NULL;
859 
860 	return &stack->path[k];
861 }
862 
863 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
864 			  struct net_device_path_stack *stack)
865 {
866 	const struct net_device *last_dev;
867 	struct net_device_path_ctx ctx = {
868 		.dev	= dev,
869 		.daddr	= daddr,
870 	};
871 	struct net_device_path *path;
872 	int ret = 0;
873 
874 	stack->num_paths = 0;
875 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
876 		last_dev = ctx.dev;
877 		path = dev_fwd_path(stack);
878 		if (!path)
879 			return -1;
880 
881 		memset(path, 0, sizeof(struct net_device_path));
882 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
883 		if (ret < 0)
884 			return -1;
885 
886 		if (WARN_ON_ONCE(last_dev == ctx.dev))
887 			return -1;
888 	}
889 	path = dev_fwd_path(stack);
890 	if (!path)
891 		return -1;
892 	path->type = DEV_PATH_ETHERNET;
893 	path->dev = ctx.dev;
894 
895 	return ret;
896 }
897 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
898 
899 /**
900  *	__dev_get_by_name	- find a device by its name
901  *	@net: the applicable net namespace
902  *	@name: name to find
903  *
904  *	Find an interface by name. Must be called under RTNL semaphore
905  *	or @dev_base_lock. If the name is found a pointer to the device
906  *	is returned. If the name is not found then %NULL is returned. The
907  *	reference counters are not incremented so the caller must be
908  *	careful with locks.
909  */
910 
911 struct net_device *__dev_get_by_name(struct net *net, const char *name)
912 {
913 	struct netdev_name_node *node_name;
914 
915 	node_name = netdev_name_node_lookup(net, name);
916 	return node_name ? node_name->dev : NULL;
917 }
918 EXPORT_SYMBOL(__dev_get_by_name);
919 
920 /**
921  * dev_get_by_name_rcu	- find a device by its name
922  * @net: the applicable net namespace
923  * @name: name to find
924  *
925  * Find an interface by name.
926  * If the name is found a pointer to the device is returned.
927  * If the name is not found then %NULL is returned.
928  * The reference counters are not incremented so the caller must be
929  * careful with locks. The caller must hold RCU lock.
930  */
931 
932 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
933 {
934 	struct netdev_name_node *node_name;
935 
936 	node_name = netdev_name_node_lookup_rcu(net, name);
937 	return node_name ? node_name->dev : NULL;
938 }
939 EXPORT_SYMBOL(dev_get_by_name_rcu);
940 
941 /**
942  *	dev_get_by_name		- find a device by its name
943  *	@net: the applicable net namespace
944  *	@name: name to find
945  *
946  *	Find an interface by name. This can be called from any
947  *	context and does its own locking. The returned handle has
948  *	the usage count incremented and the caller must use dev_put() to
949  *	release it when it is no longer needed. %NULL is returned if no
950  *	matching device is found.
951  */
952 
953 struct net_device *dev_get_by_name(struct net *net, const char *name)
954 {
955 	struct net_device *dev;
956 
957 	rcu_read_lock();
958 	dev = dev_get_by_name_rcu(net, name);
959 	if (dev)
960 		dev_hold(dev);
961 	rcu_read_unlock();
962 	return dev;
963 }
964 EXPORT_SYMBOL(dev_get_by_name);
965 
966 /**
967  *	__dev_get_by_index - find a device by its ifindex
968  *	@net: the applicable net namespace
969  *	@ifindex: index of device
970  *
971  *	Search for an interface by index. Returns %NULL if the device
972  *	is not found or a pointer to the device. The device has not
973  *	had its reference counter increased so the caller must be careful
974  *	about locking. The caller must hold either the RTNL semaphore
975  *	or @dev_base_lock.
976  */
977 
978 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
979 {
980 	struct net_device *dev;
981 	struct hlist_head *head = dev_index_hash(net, ifindex);
982 
983 	hlist_for_each_entry(dev, head, index_hlist)
984 		if (dev->ifindex == ifindex)
985 			return dev;
986 
987 	return NULL;
988 }
989 EXPORT_SYMBOL(__dev_get_by_index);
990 
991 /**
992  *	dev_get_by_index_rcu - find a device by its ifindex
993  *	@net: the applicable net namespace
994  *	@ifindex: index of device
995  *
996  *	Search for an interface by index. Returns %NULL if the device
997  *	is not found or a pointer to the device. The device has not
998  *	had its reference counter increased so the caller must be careful
999  *	about locking. The caller must hold RCU lock.
1000  */
1001 
1002 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
1003 {
1004 	struct net_device *dev;
1005 	struct hlist_head *head = dev_index_hash(net, ifindex);
1006 
1007 	hlist_for_each_entry_rcu(dev, head, index_hlist)
1008 		if (dev->ifindex == ifindex)
1009 			return dev;
1010 
1011 	return NULL;
1012 }
1013 EXPORT_SYMBOL(dev_get_by_index_rcu);
1014 
1015 
1016 /**
1017  *	dev_get_by_index - find a device by its ifindex
1018  *	@net: the applicable net namespace
1019  *	@ifindex: index of device
1020  *
1021  *	Search for an interface by index. Returns NULL if the device
1022  *	is not found or a pointer to the device. The device returned has
1023  *	had a reference added and the pointer is safe until the user calls
1024  *	dev_put to indicate they have finished with it.
1025  */
1026 
1027 struct net_device *dev_get_by_index(struct net *net, int ifindex)
1028 {
1029 	struct net_device *dev;
1030 
1031 	rcu_read_lock();
1032 	dev = dev_get_by_index_rcu(net, ifindex);
1033 	if (dev)
1034 		dev_hold(dev);
1035 	rcu_read_unlock();
1036 	return dev;
1037 }
1038 EXPORT_SYMBOL(dev_get_by_index);
1039 
1040 /**
1041  *	dev_get_by_napi_id - find a device by napi_id
1042  *	@napi_id: ID of the NAPI struct
1043  *
1044  *	Search for an interface by NAPI ID. Returns %NULL if the device
1045  *	is not found or a pointer to the device. The device has not had
1046  *	its reference counter increased so the caller must be careful
1047  *	about locking. The caller must hold RCU lock.
1048  */
1049 
1050 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
1051 {
1052 	struct napi_struct *napi;
1053 
1054 	WARN_ON_ONCE(!rcu_read_lock_held());
1055 
1056 	if (napi_id < MIN_NAPI_ID)
1057 		return NULL;
1058 
1059 	napi = napi_by_id(napi_id);
1060 
1061 	return napi ? napi->dev : NULL;
1062 }
1063 EXPORT_SYMBOL(dev_get_by_napi_id);
1064 
1065 /**
1066  *	netdev_get_name - get a netdevice name, knowing its ifindex.
1067  *	@net: network namespace
1068  *	@name: a pointer to the buffer where the name will be stored.
1069  *	@ifindex: the ifindex of the interface to get the name from.
1070  */
1071 int netdev_get_name(struct net *net, char *name, int ifindex)
1072 {
1073 	struct net_device *dev;
1074 	int ret;
1075 
1076 	down_read(&devnet_rename_sem);
1077 	rcu_read_lock();
1078 
1079 	dev = dev_get_by_index_rcu(net, ifindex);
1080 	if (!dev) {
1081 		ret = -ENODEV;
1082 		goto out;
1083 	}
1084 
1085 	strcpy(name, dev->name);
1086 
1087 	ret = 0;
1088 out:
1089 	rcu_read_unlock();
1090 	up_read(&devnet_rename_sem);
1091 	return ret;
1092 }
1093 
1094 /**
1095  *	dev_getbyhwaddr_rcu - find a device by its hardware address
1096  *	@net: the applicable net namespace
1097  *	@type: media type of device
1098  *	@ha: hardware address
1099  *
1100  *	Search for an interface by MAC address. Returns NULL if the device
1101  *	is not found or a pointer to the device.
1102  *	The caller must hold RCU or RTNL.
1103  *	The returned device has not had its ref count increased
1104  *	and the caller must therefore be careful about locking
1105  *
1106  */
1107 
1108 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1109 				       const char *ha)
1110 {
1111 	struct net_device *dev;
1112 
1113 	for_each_netdev_rcu(net, dev)
1114 		if (dev->type == type &&
1115 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
1116 			return dev;
1117 
1118 	return NULL;
1119 }
1120 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1121 
1122 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1123 {
1124 	struct net_device *dev, *ret = NULL;
1125 
1126 	rcu_read_lock();
1127 	for_each_netdev_rcu(net, dev)
1128 		if (dev->type == type) {
1129 			dev_hold(dev);
1130 			ret = dev;
1131 			break;
1132 		}
1133 	rcu_read_unlock();
1134 	return ret;
1135 }
1136 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1137 
1138 /**
1139  *	__dev_get_by_flags - find any device with given flags
1140  *	@net: the applicable net namespace
1141  *	@if_flags: IFF_* values
1142  *	@mask: bitmask of bits in if_flags to check
1143  *
1144  *	Search for any interface with the given flags. Returns NULL if a device
1145  *	is not found or a pointer to the device. Must be called inside
1146  *	rtnl_lock(), and result refcount is unchanged.
1147  */
1148 
1149 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1150 				      unsigned short mask)
1151 {
1152 	struct net_device *dev, *ret;
1153 
1154 	ASSERT_RTNL();
1155 
1156 	ret = NULL;
1157 	for_each_netdev(net, dev) {
1158 		if (((dev->flags ^ if_flags) & mask) == 0) {
1159 			ret = dev;
1160 			break;
1161 		}
1162 	}
1163 	return ret;
1164 }
1165 EXPORT_SYMBOL(__dev_get_by_flags);
1166 
1167 /**
1168  *	dev_valid_name - check if name is okay for network device
1169  *	@name: name string
1170  *
1171  *	Network device names need to be valid file names to
1172  *	allow sysfs to work.  We also disallow any kind of
1173  *	whitespace.
1174  */
1175 bool dev_valid_name(const char *name)
1176 {
1177 	if (*name == '\0')
1178 		return false;
1179 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1180 		return false;
1181 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1182 		return false;
1183 
1184 	while (*name) {
1185 		if (*name == '/' || *name == ':' || isspace(*name))
1186 			return false;
1187 		name++;
1188 	}
1189 	return true;
1190 }
1191 EXPORT_SYMBOL(dev_valid_name);
1192 
1193 /**
1194  *	__dev_alloc_name - allocate a name for a device
1195  *	@net: network namespace to allocate the device name in
1196  *	@name: name format string
1197  *	@buf:  scratch buffer and result name string
1198  *
1199  *	Passed a format string - eg "lt%d" it will try and find a suitable
1200  *	id. It scans list of devices to build up a free map, then chooses
1201  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1202  *	while allocating the name and adding the device in order to avoid
1203  *	duplicates.
1204  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1205  *	Returns the number of the unit assigned or a negative errno code.
1206  */
1207 
1208 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1209 {
1210 	int i = 0;
1211 	const char *p;
1212 	const int max_netdevices = 8*PAGE_SIZE;
1213 	unsigned long *inuse;
1214 	struct net_device *d;
1215 
1216 	if (!dev_valid_name(name))
1217 		return -EINVAL;
1218 
1219 	p = strchr(name, '%');
1220 	if (p) {
1221 		/*
1222 		 * Verify the string as this thing may have come from
1223 		 * the user.  There must be either one "%d" and no other "%"
1224 		 * characters.
1225 		 */
1226 		if (p[1] != 'd' || strchr(p + 2, '%'))
1227 			return -EINVAL;
1228 
1229 		/* Use one page as a bit array of possible slots */
1230 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1231 		if (!inuse)
1232 			return -ENOMEM;
1233 
1234 		for_each_netdev(net, d) {
1235 			struct netdev_name_node *name_node;
1236 			list_for_each_entry(name_node, &d->name_node->list, list) {
1237 				if (!sscanf(name_node->name, name, &i))
1238 					continue;
1239 				if (i < 0 || i >= max_netdevices)
1240 					continue;
1241 
1242 				/*  avoid cases where sscanf is not exact inverse of printf */
1243 				snprintf(buf, IFNAMSIZ, name, i);
1244 				if (!strncmp(buf, name_node->name, IFNAMSIZ))
1245 					set_bit(i, inuse);
1246 			}
1247 			if (!sscanf(d->name, name, &i))
1248 				continue;
1249 			if (i < 0 || i >= max_netdevices)
1250 				continue;
1251 
1252 			/*  avoid cases where sscanf is not exact inverse of printf */
1253 			snprintf(buf, IFNAMSIZ, name, i);
1254 			if (!strncmp(buf, d->name, IFNAMSIZ))
1255 				set_bit(i, inuse);
1256 		}
1257 
1258 		i = find_first_zero_bit(inuse, max_netdevices);
1259 		free_page((unsigned long) inuse);
1260 	}
1261 
1262 	snprintf(buf, IFNAMSIZ, name, i);
1263 	if (!__dev_get_by_name(net, buf))
1264 		return i;
1265 
1266 	/* It is possible to run out of possible slots
1267 	 * when the name is long and there isn't enough space left
1268 	 * for the digits, or if all bits are used.
1269 	 */
1270 	return -ENFILE;
1271 }
1272 
1273 static int dev_alloc_name_ns(struct net *net,
1274 			     struct net_device *dev,
1275 			     const char *name)
1276 {
1277 	char buf[IFNAMSIZ];
1278 	int ret;
1279 
1280 	BUG_ON(!net);
1281 	ret = __dev_alloc_name(net, name, buf);
1282 	if (ret >= 0)
1283 		strlcpy(dev->name, buf, IFNAMSIZ);
1284 	return ret;
1285 }
1286 
1287 /**
1288  *	dev_alloc_name - allocate a name for a device
1289  *	@dev: device
1290  *	@name: name format string
1291  *
1292  *	Passed a format string - eg "lt%d" it will try and find a suitable
1293  *	id. It scans list of devices to build up a free map, then chooses
1294  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1295  *	while allocating the name and adding the device in order to avoid
1296  *	duplicates.
1297  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1298  *	Returns the number of the unit assigned or a negative errno code.
1299  */
1300 
1301 int dev_alloc_name(struct net_device *dev, const char *name)
1302 {
1303 	return dev_alloc_name_ns(dev_net(dev), dev, name);
1304 }
1305 EXPORT_SYMBOL(dev_alloc_name);
1306 
1307 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1308 			      const char *name)
1309 {
1310 	BUG_ON(!net);
1311 
1312 	if (!dev_valid_name(name))
1313 		return -EINVAL;
1314 
1315 	if (strchr(name, '%'))
1316 		return dev_alloc_name_ns(net, dev, name);
1317 	else if (__dev_get_by_name(net, name))
1318 		return -EEXIST;
1319 	else if (dev->name != name)
1320 		strlcpy(dev->name, name, IFNAMSIZ);
1321 
1322 	return 0;
1323 }
1324 
1325 /**
1326  *	dev_change_name - change name of a device
1327  *	@dev: device
1328  *	@newname: name (or format string) must be at least IFNAMSIZ
1329  *
1330  *	Change name of a device, can pass format strings "eth%d".
1331  *	for wildcarding.
1332  */
1333 int dev_change_name(struct net_device *dev, const char *newname)
1334 {
1335 	unsigned char old_assign_type;
1336 	char oldname[IFNAMSIZ];
1337 	int err = 0;
1338 	int ret;
1339 	struct net *net;
1340 
1341 	ASSERT_RTNL();
1342 	BUG_ON(!dev_net(dev));
1343 
1344 	net = dev_net(dev);
1345 
1346 	/* Some auto-enslaved devices e.g. failover slaves are
1347 	 * special, as userspace might rename the device after
1348 	 * the interface had been brought up and running since
1349 	 * the point kernel initiated auto-enslavement. Allow
1350 	 * live name change even when these slave devices are
1351 	 * up and running.
1352 	 *
1353 	 * Typically, users of these auto-enslaving devices
1354 	 * don't actually care about slave name change, as
1355 	 * they are supposed to operate on master interface
1356 	 * directly.
1357 	 */
1358 	if (dev->flags & IFF_UP &&
1359 	    likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1360 		return -EBUSY;
1361 
1362 	down_write(&devnet_rename_sem);
1363 
1364 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1365 		up_write(&devnet_rename_sem);
1366 		return 0;
1367 	}
1368 
1369 	memcpy(oldname, dev->name, IFNAMSIZ);
1370 
1371 	err = dev_get_valid_name(net, dev, newname);
1372 	if (err < 0) {
1373 		up_write(&devnet_rename_sem);
1374 		return err;
1375 	}
1376 
1377 	if (oldname[0] && !strchr(oldname, '%'))
1378 		netdev_info(dev, "renamed from %s\n", oldname);
1379 
1380 	old_assign_type = dev->name_assign_type;
1381 	dev->name_assign_type = NET_NAME_RENAMED;
1382 
1383 rollback:
1384 	ret = device_rename(&dev->dev, dev->name);
1385 	if (ret) {
1386 		memcpy(dev->name, oldname, IFNAMSIZ);
1387 		dev->name_assign_type = old_assign_type;
1388 		up_write(&devnet_rename_sem);
1389 		return ret;
1390 	}
1391 
1392 	up_write(&devnet_rename_sem);
1393 
1394 	netdev_adjacent_rename_links(dev, oldname);
1395 
1396 	write_lock_bh(&dev_base_lock);
1397 	netdev_name_node_del(dev->name_node);
1398 	write_unlock_bh(&dev_base_lock);
1399 
1400 	synchronize_rcu();
1401 
1402 	write_lock_bh(&dev_base_lock);
1403 	netdev_name_node_add(net, dev->name_node);
1404 	write_unlock_bh(&dev_base_lock);
1405 
1406 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1407 	ret = notifier_to_errno(ret);
1408 
1409 	if (ret) {
1410 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1411 		if (err >= 0) {
1412 			err = ret;
1413 			down_write(&devnet_rename_sem);
1414 			memcpy(dev->name, oldname, IFNAMSIZ);
1415 			memcpy(oldname, newname, IFNAMSIZ);
1416 			dev->name_assign_type = old_assign_type;
1417 			old_assign_type = NET_NAME_RENAMED;
1418 			goto rollback;
1419 		} else {
1420 			pr_err("%s: name change rollback failed: %d\n",
1421 			       dev->name, ret);
1422 		}
1423 	}
1424 
1425 	return err;
1426 }
1427 
1428 /**
1429  *	dev_set_alias - change ifalias of a device
1430  *	@dev: device
1431  *	@alias: name up to IFALIASZ
1432  *	@len: limit of bytes to copy from info
1433  *
1434  *	Set ifalias for a device,
1435  */
1436 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1437 {
1438 	struct dev_ifalias *new_alias = NULL;
1439 
1440 	if (len >= IFALIASZ)
1441 		return -EINVAL;
1442 
1443 	if (len) {
1444 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1445 		if (!new_alias)
1446 			return -ENOMEM;
1447 
1448 		memcpy(new_alias->ifalias, alias, len);
1449 		new_alias->ifalias[len] = 0;
1450 	}
1451 
1452 	mutex_lock(&ifalias_mutex);
1453 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1454 					mutex_is_locked(&ifalias_mutex));
1455 	mutex_unlock(&ifalias_mutex);
1456 
1457 	if (new_alias)
1458 		kfree_rcu(new_alias, rcuhead);
1459 
1460 	return len;
1461 }
1462 EXPORT_SYMBOL(dev_set_alias);
1463 
1464 /**
1465  *	dev_get_alias - get ifalias of a device
1466  *	@dev: device
1467  *	@name: buffer to store name of ifalias
1468  *	@len: size of buffer
1469  *
1470  *	get ifalias for a device.  Caller must make sure dev cannot go
1471  *	away,  e.g. rcu read lock or own a reference count to device.
1472  */
1473 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1474 {
1475 	const struct dev_ifalias *alias;
1476 	int ret = 0;
1477 
1478 	rcu_read_lock();
1479 	alias = rcu_dereference(dev->ifalias);
1480 	if (alias)
1481 		ret = snprintf(name, len, "%s", alias->ifalias);
1482 	rcu_read_unlock();
1483 
1484 	return ret;
1485 }
1486 
1487 /**
1488  *	netdev_features_change - device changes features
1489  *	@dev: device to cause notification
1490  *
1491  *	Called to indicate a device has changed features.
1492  */
1493 void netdev_features_change(struct net_device *dev)
1494 {
1495 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1496 }
1497 EXPORT_SYMBOL(netdev_features_change);
1498 
1499 /**
1500  *	netdev_state_change - device changes state
1501  *	@dev: device to cause notification
1502  *
1503  *	Called to indicate a device has changed state. This function calls
1504  *	the notifier chains for netdev_chain and sends a NEWLINK message
1505  *	to the routing socket.
1506  */
1507 void netdev_state_change(struct net_device *dev)
1508 {
1509 	if (dev->flags & IFF_UP) {
1510 		struct netdev_notifier_change_info change_info = {
1511 			.info.dev = dev,
1512 		};
1513 
1514 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1515 					      &change_info.info);
1516 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1517 	}
1518 }
1519 EXPORT_SYMBOL(netdev_state_change);
1520 
1521 /**
1522  * __netdev_notify_peers - notify network peers about existence of @dev,
1523  * to be called when rtnl lock is already held.
1524  * @dev: network device
1525  *
1526  * Generate traffic such that interested network peers are aware of
1527  * @dev, such as by generating a gratuitous ARP. This may be used when
1528  * a device wants to inform the rest of the network about some sort of
1529  * reconfiguration such as a failover event or virtual machine
1530  * migration.
1531  */
1532 void __netdev_notify_peers(struct net_device *dev)
1533 {
1534 	ASSERT_RTNL();
1535 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1536 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1537 }
1538 EXPORT_SYMBOL(__netdev_notify_peers);
1539 
1540 /**
1541  * netdev_notify_peers - notify network peers about existence of @dev
1542  * @dev: network device
1543  *
1544  * Generate traffic such that interested network peers are aware of
1545  * @dev, such as by generating a gratuitous ARP. This may be used when
1546  * a device wants to inform the rest of the network about some sort of
1547  * reconfiguration such as a failover event or virtual machine
1548  * migration.
1549  */
1550 void netdev_notify_peers(struct net_device *dev)
1551 {
1552 	rtnl_lock();
1553 	__netdev_notify_peers(dev);
1554 	rtnl_unlock();
1555 }
1556 EXPORT_SYMBOL(netdev_notify_peers);
1557 
1558 static int napi_threaded_poll(void *data);
1559 
1560 static int napi_kthread_create(struct napi_struct *n)
1561 {
1562 	int err = 0;
1563 
1564 	/* Create and wake up the kthread once to put it in
1565 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1566 	 * warning and work with loadavg.
1567 	 */
1568 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1569 				n->dev->name, n->napi_id);
1570 	if (IS_ERR(n->thread)) {
1571 		err = PTR_ERR(n->thread);
1572 		pr_err("kthread_run failed with err %d\n", err);
1573 		n->thread = NULL;
1574 	}
1575 
1576 	return err;
1577 }
1578 
1579 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1580 {
1581 	const struct net_device_ops *ops = dev->netdev_ops;
1582 	int ret;
1583 
1584 	ASSERT_RTNL();
1585 
1586 	if (!netif_device_present(dev)) {
1587 		/* may be detached because parent is runtime-suspended */
1588 		if (dev->dev.parent)
1589 			pm_runtime_resume(dev->dev.parent);
1590 		if (!netif_device_present(dev))
1591 			return -ENODEV;
1592 	}
1593 
1594 	/* Block netpoll from trying to do any rx path servicing.
1595 	 * If we don't do this there is a chance ndo_poll_controller
1596 	 * or ndo_poll may be running while we open the device
1597 	 */
1598 	netpoll_poll_disable(dev);
1599 
1600 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1601 	ret = notifier_to_errno(ret);
1602 	if (ret)
1603 		return ret;
1604 
1605 	set_bit(__LINK_STATE_START, &dev->state);
1606 
1607 	if (ops->ndo_validate_addr)
1608 		ret = ops->ndo_validate_addr(dev);
1609 
1610 	if (!ret && ops->ndo_open)
1611 		ret = ops->ndo_open(dev);
1612 
1613 	netpoll_poll_enable(dev);
1614 
1615 	if (ret)
1616 		clear_bit(__LINK_STATE_START, &dev->state);
1617 	else {
1618 		dev->flags |= IFF_UP;
1619 		dev_set_rx_mode(dev);
1620 		dev_activate(dev);
1621 		add_device_randomness(dev->dev_addr, dev->addr_len);
1622 	}
1623 
1624 	return ret;
1625 }
1626 
1627 /**
1628  *	dev_open	- prepare an interface for use.
1629  *	@dev: device to open
1630  *	@extack: netlink extended ack
1631  *
1632  *	Takes a device from down to up state. The device's private open
1633  *	function is invoked and then the multicast lists are loaded. Finally
1634  *	the device is moved into the up state and a %NETDEV_UP message is
1635  *	sent to the netdev notifier chain.
1636  *
1637  *	Calling this function on an active interface is a nop. On a failure
1638  *	a negative errno code is returned.
1639  */
1640 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1641 {
1642 	int ret;
1643 
1644 	if (dev->flags & IFF_UP)
1645 		return 0;
1646 
1647 	ret = __dev_open(dev, extack);
1648 	if (ret < 0)
1649 		return ret;
1650 
1651 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1652 	call_netdevice_notifiers(NETDEV_UP, dev);
1653 
1654 	return ret;
1655 }
1656 EXPORT_SYMBOL(dev_open);
1657 
1658 static void __dev_close_many(struct list_head *head)
1659 {
1660 	struct net_device *dev;
1661 
1662 	ASSERT_RTNL();
1663 	might_sleep();
1664 
1665 	list_for_each_entry(dev, head, close_list) {
1666 		/* Temporarily disable netpoll until the interface is down */
1667 		netpoll_poll_disable(dev);
1668 
1669 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1670 
1671 		clear_bit(__LINK_STATE_START, &dev->state);
1672 
1673 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1674 		 * can be even on different cpu. So just clear netif_running().
1675 		 *
1676 		 * dev->stop() will invoke napi_disable() on all of it's
1677 		 * napi_struct instances on this device.
1678 		 */
1679 		smp_mb__after_atomic(); /* Commit netif_running(). */
1680 	}
1681 
1682 	dev_deactivate_many(head);
1683 
1684 	list_for_each_entry(dev, head, close_list) {
1685 		const struct net_device_ops *ops = dev->netdev_ops;
1686 
1687 		/*
1688 		 *	Call the device specific close. This cannot fail.
1689 		 *	Only if device is UP
1690 		 *
1691 		 *	We allow it to be called even after a DETACH hot-plug
1692 		 *	event.
1693 		 */
1694 		if (ops->ndo_stop)
1695 			ops->ndo_stop(dev);
1696 
1697 		dev->flags &= ~IFF_UP;
1698 		netpoll_poll_enable(dev);
1699 	}
1700 }
1701 
1702 static void __dev_close(struct net_device *dev)
1703 {
1704 	LIST_HEAD(single);
1705 
1706 	list_add(&dev->close_list, &single);
1707 	__dev_close_many(&single);
1708 	list_del(&single);
1709 }
1710 
1711 void dev_close_many(struct list_head *head, bool unlink)
1712 {
1713 	struct net_device *dev, *tmp;
1714 
1715 	/* Remove the devices that don't need to be closed */
1716 	list_for_each_entry_safe(dev, tmp, head, close_list)
1717 		if (!(dev->flags & IFF_UP))
1718 			list_del_init(&dev->close_list);
1719 
1720 	__dev_close_many(head);
1721 
1722 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1723 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1724 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1725 		if (unlink)
1726 			list_del_init(&dev->close_list);
1727 	}
1728 }
1729 EXPORT_SYMBOL(dev_close_many);
1730 
1731 /**
1732  *	dev_close - shutdown an interface.
1733  *	@dev: device to shutdown
1734  *
1735  *	This function moves an active device into down state. A
1736  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1737  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1738  *	chain.
1739  */
1740 void dev_close(struct net_device *dev)
1741 {
1742 	if (dev->flags & IFF_UP) {
1743 		LIST_HEAD(single);
1744 
1745 		list_add(&dev->close_list, &single);
1746 		dev_close_many(&single, true);
1747 		list_del(&single);
1748 	}
1749 }
1750 EXPORT_SYMBOL(dev_close);
1751 
1752 
1753 /**
1754  *	dev_disable_lro - disable Large Receive Offload on a device
1755  *	@dev: device
1756  *
1757  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1758  *	called under RTNL.  This is needed if received packets may be
1759  *	forwarded to another interface.
1760  */
1761 void dev_disable_lro(struct net_device *dev)
1762 {
1763 	struct net_device *lower_dev;
1764 	struct list_head *iter;
1765 
1766 	dev->wanted_features &= ~NETIF_F_LRO;
1767 	netdev_update_features(dev);
1768 
1769 	if (unlikely(dev->features & NETIF_F_LRO))
1770 		netdev_WARN(dev, "failed to disable LRO!\n");
1771 
1772 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1773 		dev_disable_lro(lower_dev);
1774 }
1775 EXPORT_SYMBOL(dev_disable_lro);
1776 
1777 /**
1778  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1779  *	@dev: device
1780  *
1781  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1782  *	called under RTNL.  This is needed if Generic XDP is installed on
1783  *	the device.
1784  */
1785 static void dev_disable_gro_hw(struct net_device *dev)
1786 {
1787 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1788 	netdev_update_features(dev);
1789 
1790 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1791 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1792 }
1793 
1794 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1795 {
1796 #define N(val) 						\
1797 	case NETDEV_##val:				\
1798 		return "NETDEV_" __stringify(val);
1799 	switch (cmd) {
1800 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1801 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1802 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1803 	N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1804 	N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1805 	N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1806 	N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1807 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1808 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1809 	N(PRE_CHANGEADDR)
1810 	}
1811 #undef N
1812 	return "UNKNOWN_NETDEV_EVENT";
1813 }
1814 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1815 
1816 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1817 				   struct net_device *dev)
1818 {
1819 	struct netdev_notifier_info info = {
1820 		.dev = dev,
1821 	};
1822 
1823 	return nb->notifier_call(nb, val, &info);
1824 }
1825 
1826 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1827 					     struct net_device *dev)
1828 {
1829 	int err;
1830 
1831 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1832 	err = notifier_to_errno(err);
1833 	if (err)
1834 		return err;
1835 
1836 	if (!(dev->flags & IFF_UP))
1837 		return 0;
1838 
1839 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1840 	return 0;
1841 }
1842 
1843 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1844 						struct net_device *dev)
1845 {
1846 	if (dev->flags & IFF_UP) {
1847 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1848 					dev);
1849 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1850 	}
1851 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1852 }
1853 
1854 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1855 						 struct net *net)
1856 {
1857 	struct net_device *dev;
1858 	int err;
1859 
1860 	for_each_netdev(net, dev) {
1861 		err = call_netdevice_register_notifiers(nb, dev);
1862 		if (err)
1863 			goto rollback;
1864 	}
1865 	return 0;
1866 
1867 rollback:
1868 	for_each_netdev_continue_reverse(net, dev)
1869 		call_netdevice_unregister_notifiers(nb, dev);
1870 	return err;
1871 }
1872 
1873 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1874 						    struct net *net)
1875 {
1876 	struct net_device *dev;
1877 
1878 	for_each_netdev(net, dev)
1879 		call_netdevice_unregister_notifiers(nb, dev);
1880 }
1881 
1882 static int dev_boot_phase = 1;
1883 
1884 /**
1885  * register_netdevice_notifier - register a network notifier block
1886  * @nb: notifier
1887  *
1888  * Register a notifier to be called when network device events occur.
1889  * The notifier passed is linked into the kernel structures and must
1890  * not be reused until it has been unregistered. A negative errno code
1891  * is returned on a failure.
1892  *
1893  * When registered all registration and up events are replayed
1894  * to the new notifier to allow device to have a race free
1895  * view of the network device list.
1896  */
1897 
1898 int register_netdevice_notifier(struct notifier_block *nb)
1899 {
1900 	struct net *net;
1901 	int err;
1902 
1903 	/* Close race with setup_net() and cleanup_net() */
1904 	down_write(&pernet_ops_rwsem);
1905 	rtnl_lock();
1906 	err = raw_notifier_chain_register(&netdev_chain, nb);
1907 	if (err)
1908 		goto unlock;
1909 	if (dev_boot_phase)
1910 		goto unlock;
1911 	for_each_net(net) {
1912 		err = call_netdevice_register_net_notifiers(nb, net);
1913 		if (err)
1914 			goto rollback;
1915 	}
1916 
1917 unlock:
1918 	rtnl_unlock();
1919 	up_write(&pernet_ops_rwsem);
1920 	return err;
1921 
1922 rollback:
1923 	for_each_net_continue_reverse(net)
1924 		call_netdevice_unregister_net_notifiers(nb, net);
1925 
1926 	raw_notifier_chain_unregister(&netdev_chain, nb);
1927 	goto unlock;
1928 }
1929 EXPORT_SYMBOL(register_netdevice_notifier);
1930 
1931 /**
1932  * unregister_netdevice_notifier - unregister a network notifier block
1933  * @nb: notifier
1934  *
1935  * Unregister a notifier previously registered by
1936  * register_netdevice_notifier(). The notifier is unlinked into the
1937  * kernel structures and may then be reused. A negative errno code
1938  * is returned on a failure.
1939  *
1940  * After unregistering unregister and down device events are synthesized
1941  * for all devices on the device list to the removed notifier to remove
1942  * the need for special case cleanup code.
1943  */
1944 
1945 int unregister_netdevice_notifier(struct notifier_block *nb)
1946 {
1947 	struct net *net;
1948 	int err;
1949 
1950 	/* Close race with setup_net() and cleanup_net() */
1951 	down_write(&pernet_ops_rwsem);
1952 	rtnl_lock();
1953 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1954 	if (err)
1955 		goto unlock;
1956 
1957 	for_each_net(net)
1958 		call_netdevice_unregister_net_notifiers(nb, net);
1959 
1960 unlock:
1961 	rtnl_unlock();
1962 	up_write(&pernet_ops_rwsem);
1963 	return err;
1964 }
1965 EXPORT_SYMBOL(unregister_netdevice_notifier);
1966 
1967 static int __register_netdevice_notifier_net(struct net *net,
1968 					     struct notifier_block *nb,
1969 					     bool ignore_call_fail)
1970 {
1971 	int err;
1972 
1973 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1974 	if (err)
1975 		return err;
1976 	if (dev_boot_phase)
1977 		return 0;
1978 
1979 	err = call_netdevice_register_net_notifiers(nb, net);
1980 	if (err && !ignore_call_fail)
1981 		goto chain_unregister;
1982 
1983 	return 0;
1984 
1985 chain_unregister:
1986 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1987 	return err;
1988 }
1989 
1990 static int __unregister_netdevice_notifier_net(struct net *net,
1991 					       struct notifier_block *nb)
1992 {
1993 	int err;
1994 
1995 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1996 	if (err)
1997 		return err;
1998 
1999 	call_netdevice_unregister_net_notifiers(nb, net);
2000 	return 0;
2001 }
2002 
2003 /**
2004  * register_netdevice_notifier_net - register a per-netns network notifier block
2005  * @net: network namespace
2006  * @nb: notifier
2007  *
2008  * Register a notifier to be called when network device events occur.
2009  * The notifier passed is linked into the kernel structures and must
2010  * not be reused until it has been unregistered. A negative errno code
2011  * is returned on a failure.
2012  *
2013  * When registered all registration and up events are replayed
2014  * to the new notifier to allow device to have a race free
2015  * view of the network device list.
2016  */
2017 
2018 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
2019 {
2020 	int err;
2021 
2022 	rtnl_lock();
2023 	err = __register_netdevice_notifier_net(net, nb, false);
2024 	rtnl_unlock();
2025 	return err;
2026 }
2027 EXPORT_SYMBOL(register_netdevice_notifier_net);
2028 
2029 /**
2030  * unregister_netdevice_notifier_net - unregister a per-netns
2031  *                                     network notifier block
2032  * @net: network namespace
2033  * @nb: notifier
2034  *
2035  * Unregister a notifier previously registered by
2036  * register_netdevice_notifier(). The notifier is unlinked into the
2037  * kernel structures and may then be reused. A negative errno code
2038  * is returned on a failure.
2039  *
2040  * After unregistering unregister and down device events are synthesized
2041  * for all devices on the device list to the removed notifier to remove
2042  * the need for special case cleanup code.
2043  */
2044 
2045 int unregister_netdevice_notifier_net(struct net *net,
2046 				      struct notifier_block *nb)
2047 {
2048 	int err;
2049 
2050 	rtnl_lock();
2051 	err = __unregister_netdevice_notifier_net(net, nb);
2052 	rtnl_unlock();
2053 	return err;
2054 }
2055 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
2056 
2057 int register_netdevice_notifier_dev_net(struct net_device *dev,
2058 					struct notifier_block *nb,
2059 					struct netdev_net_notifier *nn)
2060 {
2061 	int err;
2062 
2063 	rtnl_lock();
2064 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
2065 	if (!err) {
2066 		nn->nb = nb;
2067 		list_add(&nn->list, &dev->net_notifier_list);
2068 	}
2069 	rtnl_unlock();
2070 	return err;
2071 }
2072 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
2073 
2074 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
2075 					  struct notifier_block *nb,
2076 					  struct netdev_net_notifier *nn)
2077 {
2078 	int err;
2079 
2080 	rtnl_lock();
2081 	list_del(&nn->list);
2082 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
2083 	rtnl_unlock();
2084 	return err;
2085 }
2086 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
2087 
2088 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
2089 					     struct net *net)
2090 {
2091 	struct netdev_net_notifier *nn;
2092 
2093 	list_for_each_entry(nn, &dev->net_notifier_list, list) {
2094 		__unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
2095 		__register_netdevice_notifier_net(net, nn->nb, true);
2096 	}
2097 }
2098 
2099 /**
2100  *	call_netdevice_notifiers_info - call all network notifier blocks
2101  *	@val: value passed unmodified to notifier function
2102  *	@info: notifier information data
2103  *
2104  *	Call all network notifier blocks.  Parameters and return value
2105  *	are as for raw_notifier_call_chain().
2106  */
2107 
2108 static int call_netdevice_notifiers_info(unsigned long val,
2109 					 struct netdev_notifier_info *info)
2110 {
2111 	struct net *net = dev_net(info->dev);
2112 	int ret;
2113 
2114 	ASSERT_RTNL();
2115 
2116 	/* Run per-netns notifier block chain first, then run the global one.
2117 	 * Hopefully, one day, the global one is going to be removed after
2118 	 * all notifier block registrators get converted to be per-netns.
2119 	 */
2120 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2121 	if (ret & NOTIFY_STOP_MASK)
2122 		return ret;
2123 	return raw_notifier_call_chain(&netdev_chain, val, info);
2124 }
2125 
2126 static int call_netdevice_notifiers_extack(unsigned long val,
2127 					   struct net_device *dev,
2128 					   struct netlink_ext_ack *extack)
2129 {
2130 	struct netdev_notifier_info info = {
2131 		.dev = dev,
2132 		.extack = extack,
2133 	};
2134 
2135 	return call_netdevice_notifiers_info(val, &info);
2136 }
2137 
2138 /**
2139  *	call_netdevice_notifiers - call all network notifier blocks
2140  *      @val: value passed unmodified to notifier function
2141  *      @dev: net_device pointer passed unmodified to notifier function
2142  *
2143  *	Call all network notifier blocks.  Parameters and return value
2144  *	are as for raw_notifier_call_chain().
2145  */
2146 
2147 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2148 {
2149 	return call_netdevice_notifiers_extack(val, dev, NULL);
2150 }
2151 EXPORT_SYMBOL(call_netdevice_notifiers);
2152 
2153 /**
2154  *	call_netdevice_notifiers_mtu - call all network notifier blocks
2155  *	@val: value passed unmodified to notifier function
2156  *	@dev: net_device pointer passed unmodified to notifier function
2157  *	@arg: additional u32 argument passed to the notifier function
2158  *
2159  *	Call all network notifier blocks.  Parameters and return value
2160  *	are as for raw_notifier_call_chain().
2161  */
2162 static int call_netdevice_notifiers_mtu(unsigned long val,
2163 					struct net_device *dev, u32 arg)
2164 {
2165 	struct netdev_notifier_info_ext info = {
2166 		.info.dev = dev,
2167 		.ext.mtu = arg,
2168 	};
2169 
2170 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2171 
2172 	return call_netdevice_notifiers_info(val, &info.info);
2173 }
2174 
2175 #ifdef CONFIG_NET_INGRESS
2176 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2177 
2178 void net_inc_ingress_queue(void)
2179 {
2180 	static_branch_inc(&ingress_needed_key);
2181 }
2182 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2183 
2184 void net_dec_ingress_queue(void)
2185 {
2186 	static_branch_dec(&ingress_needed_key);
2187 }
2188 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2189 #endif
2190 
2191 #ifdef CONFIG_NET_EGRESS
2192 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2193 
2194 void net_inc_egress_queue(void)
2195 {
2196 	static_branch_inc(&egress_needed_key);
2197 }
2198 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2199 
2200 void net_dec_egress_queue(void)
2201 {
2202 	static_branch_dec(&egress_needed_key);
2203 }
2204 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2205 #endif
2206 
2207 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2208 #ifdef CONFIG_JUMP_LABEL
2209 static atomic_t netstamp_needed_deferred;
2210 static atomic_t netstamp_wanted;
2211 static void netstamp_clear(struct work_struct *work)
2212 {
2213 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2214 	int wanted;
2215 
2216 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2217 	if (wanted > 0)
2218 		static_branch_enable(&netstamp_needed_key);
2219 	else
2220 		static_branch_disable(&netstamp_needed_key);
2221 }
2222 static DECLARE_WORK(netstamp_work, netstamp_clear);
2223 #endif
2224 
2225 void net_enable_timestamp(void)
2226 {
2227 #ifdef CONFIG_JUMP_LABEL
2228 	int wanted;
2229 
2230 	while (1) {
2231 		wanted = atomic_read(&netstamp_wanted);
2232 		if (wanted <= 0)
2233 			break;
2234 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2235 			return;
2236 	}
2237 	atomic_inc(&netstamp_needed_deferred);
2238 	schedule_work(&netstamp_work);
2239 #else
2240 	static_branch_inc(&netstamp_needed_key);
2241 #endif
2242 }
2243 EXPORT_SYMBOL(net_enable_timestamp);
2244 
2245 void net_disable_timestamp(void)
2246 {
2247 #ifdef CONFIG_JUMP_LABEL
2248 	int wanted;
2249 
2250 	while (1) {
2251 		wanted = atomic_read(&netstamp_wanted);
2252 		if (wanted <= 1)
2253 			break;
2254 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2255 			return;
2256 	}
2257 	atomic_dec(&netstamp_needed_deferred);
2258 	schedule_work(&netstamp_work);
2259 #else
2260 	static_branch_dec(&netstamp_needed_key);
2261 #endif
2262 }
2263 EXPORT_SYMBOL(net_disable_timestamp);
2264 
2265 static inline void net_timestamp_set(struct sk_buff *skb)
2266 {
2267 	skb->tstamp = 0;
2268 	if (static_branch_unlikely(&netstamp_needed_key))
2269 		__net_timestamp(skb);
2270 }
2271 
2272 #define net_timestamp_check(COND, SKB)				\
2273 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2274 		if ((COND) && !(SKB)->tstamp)			\
2275 			__net_timestamp(SKB);			\
2276 	}							\
2277 
2278 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2279 {
2280 	return __is_skb_forwardable(dev, skb, true);
2281 }
2282 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2283 
2284 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2285 			      bool check_mtu)
2286 {
2287 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2288 
2289 	if (likely(!ret)) {
2290 		skb->protocol = eth_type_trans(skb, dev);
2291 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2292 	}
2293 
2294 	return ret;
2295 }
2296 
2297 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2298 {
2299 	return __dev_forward_skb2(dev, skb, true);
2300 }
2301 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2302 
2303 /**
2304  * dev_forward_skb - loopback an skb to another netif
2305  *
2306  * @dev: destination network device
2307  * @skb: buffer to forward
2308  *
2309  * return values:
2310  *	NET_RX_SUCCESS	(no congestion)
2311  *	NET_RX_DROP     (packet was dropped, but freed)
2312  *
2313  * dev_forward_skb can be used for injecting an skb from the
2314  * start_xmit function of one device into the receive queue
2315  * of another device.
2316  *
2317  * The receiving device may be in another namespace, so
2318  * we have to clear all information in the skb that could
2319  * impact namespace isolation.
2320  */
2321 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2322 {
2323 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2324 }
2325 EXPORT_SYMBOL_GPL(dev_forward_skb);
2326 
2327 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2328 {
2329 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2330 }
2331 
2332 static inline int deliver_skb(struct sk_buff *skb,
2333 			      struct packet_type *pt_prev,
2334 			      struct net_device *orig_dev)
2335 {
2336 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2337 		return -ENOMEM;
2338 	refcount_inc(&skb->users);
2339 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2340 }
2341 
2342 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2343 					  struct packet_type **pt,
2344 					  struct net_device *orig_dev,
2345 					  __be16 type,
2346 					  struct list_head *ptype_list)
2347 {
2348 	struct packet_type *ptype, *pt_prev = *pt;
2349 
2350 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2351 		if (ptype->type != type)
2352 			continue;
2353 		if (pt_prev)
2354 			deliver_skb(skb, pt_prev, orig_dev);
2355 		pt_prev = ptype;
2356 	}
2357 	*pt = pt_prev;
2358 }
2359 
2360 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2361 {
2362 	if (!ptype->af_packet_priv || !skb->sk)
2363 		return false;
2364 
2365 	if (ptype->id_match)
2366 		return ptype->id_match(ptype, skb->sk);
2367 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2368 		return true;
2369 
2370 	return false;
2371 }
2372 
2373 /**
2374  * dev_nit_active - return true if any network interface taps are in use
2375  *
2376  * @dev: network device to check for the presence of taps
2377  */
2378 bool dev_nit_active(struct net_device *dev)
2379 {
2380 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2381 }
2382 EXPORT_SYMBOL_GPL(dev_nit_active);
2383 
2384 /*
2385  *	Support routine. Sends outgoing frames to any network
2386  *	taps currently in use.
2387  */
2388 
2389 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2390 {
2391 	struct packet_type *ptype;
2392 	struct sk_buff *skb2 = NULL;
2393 	struct packet_type *pt_prev = NULL;
2394 	struct list_head *ptype_list = &ptype_all;
2395 
2396 	rcu_read_lock();
2397 again:
2398 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2399 		if (ptype->ignore_outgoing)
2400 			continue;
2401 
2402 		/* Never send packets back to the socket
2403 		 * they originated from - MvS (miquels@drinkel.ow.org)
2404 		 */
2405 		if (skb_loop_sk(ptype, skb))
2406 			continue;
2407 
2408 		if (pt_prev) {
2409 			deliver_skb(skb2, pt_prev, skb->dev);
2410 			pt_prev = ptype;
2411 			continue;
2412 		}
2413 
2414 		/* need to clone skb, done only once */
2415 		skb2 = skb_clone(skb, GFP_ATOMIC);
2416 		if (!skb2)
2417 			goto out_unlock;
2418 
2419 		net_timestamp_set(skb2);
2420 
2421 		/* skb->nh should be correctly
2422 		 * set by sender, so that the second statement is
2423 		 * just protection against buggy protocols.
2424 		 */
2425 		skb_reset_mac_header(skb2);
2426 
2427 		if (skb_network_header(skb2) < skb2->data ||
2428 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2429 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2430 					     ntohs(skb2->protocol),
2431 					     dev->name);
2432 			skb_reset_network_header(skb2);
2433 		}
2434 
2435 		skb2->transport_header = skb2->network_header;
2436 		skb2->pkt_type = PACKET_OUTGOING;
2437 		pt_prev = ptype;
2438 	}
2439 
2440 	if (ptype_list == &ptype_all) {
2441 		ptype_list = &dev->ptype_all;
2442 		goto again;
2443 	}
2444 out_unlock:
2445 	if (pt_prev) {
2446 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2447 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2448 		else
2449 			kfree_skb(skb2);
2450 	}
2451 	rcu_read_unlock();
2452 }
2453 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2454 
2455 /**
2456  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2457  * @dev: Network device
2458  * @txq: number of queues available
2459  *
2460  * If real_num_tx_queues is changed the tc mappings may no longer be
2461  * valid. To resolve this verify the tc mapping remains valid and if
2462  * not NULL the mapping. With no priorities mapping to this
2463  * offset/count pair it will no longer be used. In the worst case TC0
2464  * is invalid nothing can be done so disable priority mappings. If is
2465  * expected that drivers will fix this mapping if they can before
2466  * calling netif_set_real_num_tx_queues.
2467  */
2468 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2469 {
2470 	int i;
2471 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2472 
2473 	/* If TC0 is invalidated disable TC mapping */
2474 	if (tc->offset + tc->count > txq) {
2475 		pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2476 		dev->num_tc = 0;
2477 		return;
2478 	}
2479 
2480 	/* Invalidated prio to tc mappings set to TC0 */
2481 	for (i = 1; i < TC_BITMASK + 1; i++) {
2482 		int q = netdev_get_prio_tc_map(dev, i);
2483 
2484 		tc = &dev->tc_to_txq[q];
2485 		if (tc->offset + tc->count > txq) {
2486 			pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2487 				i, q);
2488 			netdev_set_prio_tc_map(dev, i, 0);
2489 		}
2490 	}
2491 }
2492 
2493 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2494 {
2495 	if (dev->num_tc) {
2496 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2497 		int i;
2498 
2499 		/* walk through the TCs and see if it falls into any of them */
2500 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2501 			if ((txq - tc->offset) < tc->count)
2502 				return i;
2503 		}
2504 
2505 		/* didn't find it, just return -1 to indicate no match */
2506 		return -1;
2507 	}
2508 
2509 	return 0;
2510 }
2511 EXPORT_SYMBOL(netdev_txq_to_tc);
2512 
2513 #ifdef CONFIG_XPS
2514 static struct static_key xps_needed __read_mostly;
2515 static struct static_key xps_rxqs_needed __read_mostly;
2516 static DEFINE_MUTEX(xps_map_mutex);
2517 #define xmap_dereference(P)		\
2518 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2519 
2520 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2521 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2522 {
2523 	struct xps_map *map = NULL;
2524 	int pos;
2525 
2526 	if (dev_maps)
2527 		map = xmap_dereference(dev_maps->attr_map[tci]);
2528 	if (!map)
2529 		return false;
2530 
2531 	for (pos = map->len; pos--;) {
2532 		if (map->queues[pos] != index)
2533 			continue;
2534 
2535 		if (map->len > 1) {
2536 			map->queues[pos] = map->queues[--map->len];
2537 			break;
2538 		}
2539 
2540 		if (old_maps)
2541 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2542 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2543 		kfree_rcu(map, rcu);
2544 		return false;
2545 	}
2546 
2547 	return true;
2548 }
2549 
2550 static bool remove_xps_queue_cpu(struct net_device *dev,
2551 				 struct xps_dev_maps *dev_maps,
2552 				 int cpu, u16 offset, u16 count)
2553 {
2554 	int num_tc = dev_maps->num_tc;
2555 	bool active = false;
2556 	int tci;
2557 
2558 	for (tci = cpu * num_tc; num_tc--; tci++) {
2559 		int i, j;
2560 
2561 		for (i = count, j = offset; i--; j++) {
2562 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2563 				break;
2564 		}
2565 
2566 		active |= i < 0;
2567 	}
2568 
2569 	return active;
2570 }
2571 
2572 static void reset_xps_maps(struct net_device *dev,
2573 			   struct xps_dev_maps *dev_maps,
2574 			   enum xps_map_type type)
2575 {
2576 	static_key_slow_dec_cpuslocked(&xps_needed);
2577 	if (type == XPS_RXQS)
2578 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2579 
2580 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2581 
2582 	kfree_rcu(dev_maps, rcu);
2583 }
2584 
2585 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2586 			   u16 offset, u16 count)
2587 {
2588 	struct xps_dev_maps *dev_maps;
2589 	bool active = false;
2590 	int i, j;
2591 
2592 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2593 	if (!dev_maps)
2594 		return;
2595 
2596 	for (j = 0; j < dev_maps->nr_ids; j++)
2597 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2598 	if (!active)
2599 		reset_xps_maps(dev, dev_maps, type);
2600 
2601 	if (type == XPS_CPUS) {
2602 		for (i = offset + (count - 1); count--; i--)
2603 			netdev_queue_numa_node_write(
2604 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2605 	}
2606 }
2607 
2608 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2609 				   u16 count)
2610 {
2611 	if (!static_key_false(&xps_needed))
2612 		return;
2613 
2614 	cpus_read_lock();
2615 	mutex_lock(&xps_map_mutex);
2616 
2617 	if (static_key_false(&xps_rxqs_needed))
2618 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2619 
2620 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2621 
2622 	mutex_unlock(&xps_map_mutex);
2623 	cpus_read_unlock();
2624 }
2625 
2626 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2627 {
2628 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2629 }
2630 
2631 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2632 				      u16 index, bool is_rxqs_map)
2633 {
2634 	struct xps_map *new_map;
2635 	int alloc_len = XPS_MIN_MAP_ALLOC;
2636 	int i, pos;
2637 
2638 	for (pos = 0; map && pos < map->len; pos++) {
2639 		if (map->queues[pos] != index)
2640 			continue;
2641 		return map;
2642 	}
2643 
2644 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2645 	if (map) {
2646 		if (pos < map->alloc_len)
2647 			return map;
2648 
2649 		alloc_len = map->alloc_len * 2;
2650 	}
2651 
2652 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2653 	 *  map
2654 	 */
2655 	if (is_rxqs_map)
2656 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2657 	else
2658 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2659 				       cpu_to_node(attr_index));
2660 	if (!new_map)
2661 		return NULL;
2662 
2663 	for (i = 0; i < pos; i++)
2664 		new_map->queues[i] = map->queues[i];
2665 	new_map->alloc_len = alloc_len;
2666 	new_map->len = pos;
2667 
2668 	return new_map;
2669 }
2670 
2671 /* Copy xps maps at a given index */
2672 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2673 			      struct xps_dev_maps *new_dev_maps, int index,
2674 			      int tc, bool skip_tc)
2675 {
2676 	int i, tci = index * dev_maps->num_tc;
2677 	struct xps_map *map;
2678 
2679 	/* copy maps belonging to foreign traffic classes */
2680 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2681 		if (i == tc && skip_tc)
2682 			continue;
2683 
2684 		/* fill in the new device map from the old device map */
2685 		map = xmap_dereference(dev_maps->attr_map[tci]);
2686 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2687 	}
2688 }
2689 
2690 /* Must be called under cpus_read_lock */
2691 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2692 			  u16 index, enum xps_map_type type)
2693 {
2694 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2695 	const unsigned long *online_mask = NULL;
2696 	bool active = false, copy = false;
2697 	int i, j, tci, numa_node_id = -2;
2698 	int maps_sz, num_tc = 1, tc = 0;
2699 	struct xps_map *map, *new_map;
2700 	unsigned int nr_ids;
2701 
2702 	if (dev->num_tc) {
2703 		/* Do not allow XPS on subordinate device directly */
2704 		num_tc = dev->num_tc;
2705 		if (num_tc < 0)
2706 			return -EINVAL;
2707 
2708 		/* If queue belongs to subordinate dev use its map */
2709 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2710 
2711 		tc = netdev_txq_to_tc(dev, index);
2712 		if (tc < 0)
2713 			return -EINVAL;
2714 	}
2715 
2716 	mutex_lock(&xps_map_mutex);
2717 
2718 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2719 	if (type == XPS_RXQS) {
2720 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2721 		nr_ids = dev->num_rx_queues;
2722 	} else {
2723 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2724 		if (num_possible_cpus() > 1)
2725 			online_mask = cpumask_bits(cpu_online_mask);
2726 		nr_ids = nr_cpu_ids;
2727 	}
2728 
2729 	if (maps_sz < L1_CACHE_BYTES)
2730 		maps_sz = L1_CACHE_BYTES;
2731 
2732 	/* The old dev_maps could be larger or smaller than the one we're
2733 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2734 	 * between. We could try to be smart, but let's be safe instead and only
2735 	 * copy foreign traffic classes if the two map sizes match.
2736 	 */
2737 	if (dev_maps &&
2738 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2739 		copy = true;
2740 
2741 	/* allocate memory for queue storage */
2742 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2743 	     j < nr_ids;) {
2744 		if (!new_dev_maps) {
2745 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2746 			if (!new_dev_maps) {
2747 				mutex_unlock(&xps_map_mutex);
2748 				return -ENOMEM;
2749 			}
2750 
2751 			new_dev_maps->nr_ids = nr_ids;
2752 			new_dev_maps->num_tc = num_tc;
2753 		}
2754 
2755 		tci = j * num_tc + tc;
2756 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2757 
2758 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2759 		if (!map)
2760 			goto error;
2761 
2762 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2763 	}
2764 
2765 	if (!new_dev_maps)
2766 		goto out_no_new_maps;
2767 
2768 	if (!dev_maps) {
2769 		/* Increment static keys at most once per type */
2770 		static_key_slow_inc_cpuslocked(&xps_needed);
2771 		if (type == XPS_RXQS)
2772 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2773 	}
2774 
2775 	for (j = 0; j < nr_ids; j++) {
2776 		bool skip_tc = false;
2777 
2778 		tci = j * num_tc + tc;
2779 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2780 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2781 			/* add tx-queue to CPU/rx-queue maps */
2782 			int pos = 0;
2783 
2784 			skip_tc = true;
2785 
2786 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2787 			while ((pos < map->len) && (map->queues[pos] != index))
2788 				pos++;
2789 
2790 			if (pos == map->len)
2791 				map->queues[map->len++] = index;
2792 #ifdef CONFIG_NUMA
2793 			if (type == XPS_CPUS) {
2794 				if (numa_node_id == -2)
2795 					numa_node_id = cpu_to_node(j);
2796 				else if (numa_node_id != cpu_to_node(j))
2797 					numa_node_id = -1;
2798 			}
2799 #endif
2800 		}
2801 
2802 		if (copy)
2803 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2804 					  skip_tc);
2805 	}
2806 
2807 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2808 
2809 	/* Cleanup old maps */
2810 	if (!dev_maps)
2811 		goto out_no_old_maps;
2812 
2813 	for (j = 0; j < dev_maps->nr_ids; j++) {
2814 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2815 			map = xmap_dereference(dev_maps->attr_map[tci]);
2816 			if (!map)
2817 				continue;
2818 
2819 			if (copy) {
2820 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2821 				if (map == new_map)
2822 					continue;
2823 			}
2824 
2825 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2826 			kfree_rcu(map, rcu);
2827 		}
2828 	}
2829 
2830 	old_dev_maps = dev_maps;
2831 
2832 out_no_old_maps:
2833 	dev_maps = new_dev_maps;
2834 	active = true;
2835 
2836 out_no_new_maps:
2837 	if (type == XPS_CPUS)
2838 		/* update Tx queue numa node */
2839 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2840 					     (numa_node_id >= 0) ?
2841 					     numa_node_id : NUMA_NO_NODE);
2842 
2843 	if (!dev_maps)
2844 		goto out_no_maps;
2845 
2846 	/* removes tx-queue from unused CPUs/rx-queues */
2847 	for (j = 0; j < dev_maps->nr_ids; j++) {
2848 		tci = j * dev_maps->num_tc;
2849 
2850 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2851 			if (i == tc &&
2852 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2853 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2854 				continue;
2855 
2856 			active |= remove_xps_queue(dev_maps,
2857 						   copy ? old_dev_maps : NULL,
2858 						   tci, index);
2859 		}
2860 	}
2861 
2862 	if (old_dev_maps)
2863 		kfree_rcu(old_dev_maps, rcu);
2864 
2865 	/* free map if not active */
2866 	if (!active)
2867 		reset_xps_maps(dev, dev_maps, type);
2868 
2869 out_no_maps:
2870 	mutex_unlock(&xps_map_mutex);
2871 
2872 	return 0;
2873 error:
2874 	/* remove any maps that we added */
2875 	for (j = 0; j < nr_ids; j++) {
2876 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2877 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2878 			map = copy ?
2879 			      xmap_dereference(dev_maps->attr_map[tci]) :
2880 			      NULL;
2881 			if (new_map && new_map != map)
2882 				kfree(new_map);
2883 		}
2884 	}
2885 
2886 	mutex_unlock(&xps_map_mutex);
2887 
2888 	kfree(new_dev_maps);
2889 	return -ENOMEM;
2890 }
2891 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2892 
2893 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2894 			u16 index)
2895 {
2896 	int ret;
2897 
2898 	cpus_read_lock();
2899 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2900 	cpus_read_unlock();
2901 
2902 	return ret;
2903 }
2904 EXPORT_SYMBOL(netif_set_xps_queue);
2905 
2906 #endif
2907 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2908 {
2909 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2910 
2911 	/* Unbind any subordinate channels */
2912 	while (txq-- != &dev->_tx[0]) {
2913 		if (txq->sb_dev)
2914 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2915 	}
2916 }
2917 
2918 void netdev_reset_tc(struct net_device *dev)
2919 {
2920 #ifdef CONFIG_XPS
2921 	netif_reset_xps_queues_gt(dev, 0);
2922 #endif
2923 	netdev_unbind_all_sb_channels(dev);
2924 
2925 	/* Reset TC configuration of device */
2926 	dev->num_tc = 0;
2927 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2928 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2929 }
2930 EXPORT_SYMBOL(netdev_reset_tc);
2931 
2932 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2933 {
2934 	if (tc >= dev->num_tc)
2935 		return -EINVAL;
2936 
2937 #ifdef CONFIG_XPS
2938 	netif_reset_xps_queues(dev, offset, count);
2939 #endif
2940 	dev->tc_to_txq[tc].count = count;
2941 	dev->tc_to_txq[tc].offset = offset;
2942 	return 0;
2943 }
2944 EXPORT_SYMBOL(netdev_set_tc_queue);
2945 
2946 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2947 {
2948 	if (num_tc > TC_MAX_QUEUE)
2949 		return -EINVAL;
2950 
2951 #ifdef CONFIG_XPS
2952 	netif_reset_xps_queues_gt(dev, 0);
2953 #endif
2954 	netdev_unbind_all_sb_channels(dev);
2955 
2956 	dev->num_tc = num_tc;
2957 	return 0;
2958 }
2959 EXPORT_SYMBOL(netdev_set_num_tc);
2960 
2961 void netdev_unbind_sb_channel(struct net_device *dev,
2962 			      struct net_device *sb_dev)
2963 {
2964 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2965 
2966 #ifdef CONFIG_XPS
2967 	netif_reset_xps_queues_gt(sb_dev, 0);
2968 #endif
2969 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2970 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2971 
2972 	while (txq-- != &dev->_tx[0]) {
2973 		if (txq->sb_dev == sb_dev)
2974 			txq->sb_dev = NULL;
2975 	}
2976 }
2977 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2978 
2979 int netdev_bind_sb_channel_queue(struct net_device *dev,
2980 				 struct net_device *sb_dev,
2981 				 u8 tc, u16 count, u16 offset)
2982 {
2983 	/* Make certain the sb_dev and dev are already configured */
2984 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2985 		return -EINVAL;
2986 
2987 	/* We cannot hand out queues we don't have */
2988 	if ((offset + count) > dev->real_num_tx_queues)
2989 		return -EINVAL;
2990 
2991 	/* Record the mapping */
2992 	sb_dev->tc_to_txq[tc].count = count;
2993 	sb_dev->tc_to_txq[tc].offset = offset;
2994 
2995 	/* Provide a way for Tx queue to find the tc_to_txq map or
2996 	 * XPS map for itself.
2997 	 */
2998 	while (count--)
2999 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
3000 
3001 	return 0;
3002 }
3003 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
3004 
3005 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
3006 {
3007 	/* Do not use a multiqueue device to represent a subordinate channel */
3008 	if (netif_is_multiqueue(dev))
3009 		return -ENODEV;
3010 
3011 	/* We allow channels 1 - 32767 to be used for subordinate channels.
3012 	 * Channel 0 is meant to be "native" mode and used only to represent
3013 	 * the main root device. We allow writing 0 to reset the device back
3014 	 * to normal mode after being used as a subordinate channel.
3015 	 */
3016 	if (channel > S16_MAX)
3017 		return -EINVAL;
3018 
3019 	dev->num_tc = -channel;
3020 
3021 	return 0;
3022 }
3023 EXPORT_SYMBOL(netdev_set_sb_channel);
3024 
3025 /*
3026  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
3027  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
3028  */
3029 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
3030 {
3031 	bool disabling;
3032 	int rc;
3033 
3034 	disabling = txq < dev->real_num_tx_queues;
3035 
3036 	if (txq < 1 || txq > dev->num_tx_queues)
3037 		return -EINVAL;
3038 
3039 	if (dev->reg_state == NETREG_REGISTERED ||
3040 	    dev->reg_state == NETREG_UNREGISTERING) {
3041 		ASSERT_RTNL();
3042 
3043 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
3044 						  txq);
3045 		if (rc)
3046 			return rc;
3047 
3048 		if (dev->num_tc)
3049 			netif_setup_tc(dev, txq);
3050 
3051 		dev->real_num_tx_queues = txq;
3052 
3053 		if (disabling) {
3054 			synchronize_net();
3055 			qdisc_reset_all_tx_gt(dev, txq);
3056 #ifdef CONFIG_XPS
3057 			netif_reset_xps_queues_gt(dev, txq);
3058 #endif
3059 		}
3060 	} else {
3061 		dev->real_num_tx_queues = txq;
3062 	}
3063 
3064 	return 0;
3065 }
3066 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
3067 
3068 #ifdef CONFIG_SYSFS
3069 /**
3070  *	netif_set_real_num_rx_queues - set actual number of RX queues used
3071  *	@dev: Network device
3072  *	@rxq: Actual number of RX queues
3073  *
3074  *	This must be called either with the rtnl_lock held or before
3075  *	registration of the net device.  Returns 0 on success, or a
3076  *	negative error code.  If called before registration, it always
3077  *	succeeds.
3078  */
3079 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
3080 {
3081 	int rc;
3082 
3083 	if (rxq < 1 || rxq > dev->num_rx_queues)
3084 		return -EINVAL;
3085 
3086 	if (dev->reg_state == NETREG_REGISTERED) {
3087 		ASSERT_RTNL();
3088 
3089 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3090 						  rxq);
3091 		if (rc)
3092 			return rc;
3093 	}
3094 
3095 	dev->real_num_rx_queues = rxq;
3096 	return 0;
3097 }
3098 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3099 #endif
3100 
3101 /**
3102  * netif_get_num_default_rss_queues - default number of RSS queues
3103  *
3104  * This routine should set an upper limit on the number of RSS queues
3105  * used by default by multiqueue devices.
3106  */
3107 int netif_get_num_default_rss_queues(void)
3108 {
3109 	return is_kdump_kernel() ?
3110 		1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3111 }
3112 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3113 
3114 static void __netif_reschedule(struct Qdisc *q)
3115 {
3116 	struct softnet_data *sd;
3117 	unsigned long flags;
3118 
3119 	local_irq_save(flags);
3120 	sd = this_cpu_ptr(&softnet_data);
3121 	q->next_sched = NULL;
3122 	*sd->output_queue_tailp = q;
3123 	sd->output_queue_tailp = &q->next_sched;
3124 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3125 	local_irq_restore(flags);
3126 }
3127 
3128 void __netif_schedule(struct Qdisc *q)
3129 {
3130 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3131 		__netif_reschedule(q);
3132 }
3133 EXPORT_SYMBOL(__netif_schedule);
3134 
3135 struct dev_kfree_skb_cb {
3136 	enum skb_free_reason reason;
3137 };
3138 
3139 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3140 {
3141 	return (struct dev_kfree_skb_cb *)skb->cb;
3142 }
3143 
3144 void netif_schedule_queue(struct netdev_queue *txq)
3145 {
3146 	rcu_read_lock();
3147 	if (!netif_xmit_stopped(txq)) {
3148 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3149 
3150 		__netif_schedule(q);
3151 	}
3152 	rcu_read_unlock();
3153 }
3154 EXPORT_SYMBOL(netif_schedule_queue);
3155 
3156 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3157 {
3158 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3159 		struct Qdisc *q;
3160 
3161 		rcu_read_lock();
3162 		q = rcu_dereference(dev_queue->qdisc);
3163 		__netif_schedule(q);
3164 		rcu_read_unlock();
3165 	}
3166 }
3167 EXPORT_SYMBOL(netif_tx_wake_queue);
3168 
3169 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3170 {
3171 	unsigned long flags;
3172 
3173 	if (unlikely(!skb))
3174 		return;
3175 
3176 	if (likely(refcount_read(&skb->users) == 1)) {
3177 		smp_rmb();
3178 		refcount_set(&skb->users, 0);
3179 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3180 		return;
3181 	}
3182 	get_kfree_skb_cb(skb)->reason = reason;
3183 	local_irq_save(flags);
3184 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3185 	__this_cpu_write(softnet_data.completion_queue, skb);
3186 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3187 	local_irq_restore(flags);
3188 }
3189 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3190 
3191 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3192 {
3193 	if (in_irq() || irqs_disabled())
3194 		__dev_kfree_skb_irq(skb, reason);
3195 	else
3196 		dev_kfree_skb(skb);
3197 }
3198 EXPORT_SYMBOL(__dev_kfree_skb_any);
3199 
3200 
3201 /**
3202  * netif_device_detach - mark device as removed
3203  * @dev: network device
3204  *
3205  * Mark device as removed from system and therefore no longer available.
3206  */
3207 void netif_device_detach(struct net_device *dev)
3208 {
3209 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3210 	    netif_running(dev)) {
3211 		netif_tx_stop_all_queues(dev);
3212 	}
3213 }
3214 EXPORT_SYMBOL(netif_device_detach);
3215 
3216 /**
3217  * netif_device_attach - mark device as attached
3218  * @dev: network device
3219  *
3220  * Mark device as attached from system and restart if needed.
3221  */
3222 void netif_device_attach(struct net_device *dev)
3223 {
3224 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3225 	    netif_running(dev)) {
3226 		netif_tx_wake_all_queues(dev);
3227 		__netdev_watchdog_up(dev);
3228 	}
3229 }
3230 EXPORT_SYMBOL(netif_device_attach);
3231 
3232 /*
3233  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3234  * to be used as a distribution range.
3235  */
3236 static u16 skb_tx_hash(const struct net_device *dev,
3237 		       const struct net_device *sb_dev,
3238 		       struct sk_buff *skb)
3239 {
3240 	u32 hash;
3241 	u16 qoffset = 0;
3242 	u16 qcount = dev->real_num_tx_queues;
3243 
3244 	if (dev->num_tc) {
3245 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3246 
3247 		qoffset = sb_dev->tc_to_txq[tc].offset;
3248 		qcount = sb_dev->tc_to_txq[tc].count;
3249 	}
3250 
3251 	if (skb_rx_queue_recorded(skb)) {
3252 		hash = skb_get_rx_queue(skb);
3253 		if (hash >= qoffset)
3254 			hash -= qoffset;
3255 		while (unlikely(hash >= qcount))
3256 			hash -= qcount;
3257 		return hash + qoffset;
3258 	}
3259 
3260 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3261 }
3262 
3263 static void skb_warn_bad_offload(const struct sk_buff *skb)
3264 {
3265 	static const netdev_features_t null_features;
3266 	struct net_device *dev = skb->dev;
3267 	const char *name = "";
3268 
3269 	if (!net_ratelimit())
3270 		return;
3271 
3272 	if (dev) {
3273 		if (dev->dev.parent)
3274 			name = dev_driver_string(dev->dev.parent);
3275 		else
3276 			name = netdev_name(dev);
3277 	}
3278 	skb_dump(KERN_WARNING, skb, false);
3279 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3280 	     name, dev ? &dev->features : &null_features,
3281 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3282 }
3283 
3284 /*
3285  * Invalidate hardware checksum when packet is to be mangled, and
3286  * complete checksum manually on outgoing path.
3287  */
3288 int skb_checksum_help(struct sk_buff *skb)
3289 {
3290 	__wsum csum;
3291 	int ret = 0, offset;
3292 
3293 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3294 		goto out_set_summed;
3295 
3296 	if (unlikely(skb_is_gso(skb))) {
3297 		skb_warn_bad_offload(skb);
3298 		return -EINVAL;
3299 	}
3300 
3301 	/* Before computing a checksum, we should make sure no frag could
3302 	 * be modified by an external entity : checksum could be wrong.
3303 	 */
3304 	if (skb_has_shared_frag(skb)) {
3305 		ret = __skb_linearize(skb);
3306 		if (ret)
3307 			goto out;
3308 	}
3309 
3310 	offset = skb_checksum_start_offset(skb);
3311 	BUG_ON(offset >= skb_headlen(skb));
3312 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3313 
3314 	offset += skb->csum_offset;
3315 	BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3316 
3317 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3318 	if (ret)
3319 		goto out;
3320 
3321 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3322 out_set_summed:
3323 	skb->ip_summed = CHECKSUM_NONE;
3324 out:
3325 	return ret;
3326 }
3327 EXPORT_SYMBOL(skb_checksum_help);
3328 
3329 int skb_crc32c_csum_help(struct sk_buff *skb)
3330 {
3331 	__le32 crc32c_csum;
3332 	int ret = 0, offset, start;
3333 
3334 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3335 		goto out;
3336 
3337 	if (unlikely(skb_is_gso(skb)))
3338 		goto out;
3339 
3340 	/* Before computing a checksum, we should make sure no frag could
3341 	 * be modified by an external entity : checksum could be wrong.
3342 	 */
3343 	if (unlikely(skb_has_shared_frag(skb))) {
3344 		ret = __skb_linearize(skb);
3345 		if (ret)
3346 			goto out;
3347 	}
3348 	start = skb_checksum_start_offset(skb);
3349 	offset = start + offsetof(struct sctphdr, checksum);
3350 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3351 		ret = -EINVAL;
3352 		goto out;
3353 	}
3354 
3355 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3356 	if (ret)
3357 		goto out;
3358 
3359 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3360 						  skb->len - start, ~(__u32)0,
3361 						  crc32c_csum_stub));
3362 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3363 	skb->ip_summed = CHECKSUM_NONE;
3364 	skb->csum_not_inet = 0;
3365 out:
3366 	return ret;
3367 }
3368 
3369 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3370 {
3371 	__be16 type = skb->protocol;
3372 
3373 	/* Tunnel gso handlers can set protocol to ethernet. */
3374 	if (type == htons(ETH_P_TEB)) {
3375 		struct ethhdr *eth;
3376 
3377 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3378 			return 0;
3379 
3380 		eth = (struct ethhdr *)skb->data;
3381 		type = eth->h_proto;
3382 	}
3383 
3384 	return __vlan_get_protocol(skb, type, depth);
3385 }
3386 
3387 /**
3388  *	skb_mac_gso_segment - mac layer segmentation handler.
3389  *	@skb: buffer to segment
3390  *	@features: features for the output path (see dev->features)
3391  */
3392 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3393 				    netdev_features_t features)
3394 {
3395 	struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3396 	struct packet_offload *ptype;
3397 	int vlan_depth = skb->mac_len;
3398 	__be16 type = skb_network_protocol(skb, &vlan_depth);
3399 
3400 	if (unlikely(!type))
3401 		return ERR_PTR(-EINVAL);
3402 
3403 	__skb_pull(skb, vlan_depth);
3404 
3405 	rcu_read_lock();
3406 	list_for_each_entry_rcu(ptype, &offload_base, list) {
3407 		if (ptype->type == type && ptype->callbacks.gso_segment) {
3408 			segs = ptype->callbacks.gso_segment(skb, features);
3409 			break;
3410 		}
3411 	}
3412 	rcu_read_unlock();
3413 
3414 	__skb_push(skb, skb->data - skb_mac_header(skb));
3415 
3416 	return segs;
3417 }
3418 EXPORT_SYMBOL(skb_mac_gso_segment);
3419 
3420 
3421 /* openvswitch calls this on rx path, so we need a different check.
3422  */
3423 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3424 {
3425 	if (tx_path)
3426 		return skb->ip_summed != CHECKSUM_PARTIAL &&
3427 		       skb->ip_summed != CHECKSUM_UNNECESSARY;
3428 
3429 	return skb->ip_summed == CHECKSUM_NONE;
3430 }
3431 
3432 /**
3433  *	__skb_gso_segment - Perform segmentation on skb.
3434  *	@skb: buffer to segment
3435  *	@features: features for the output path (see dev->features)
3436  *	@tx_path: whether it is called in TX path
3437  *
3438  *	This function segments the given skb and returns a list of segments.
3439  *
3440  *	It may return NULL if the skb requires no segmentation.  This is
3441  *	only possible when GSO is used for verifying header integrity.
3442  *
3443  *	Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3444  */
3445 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3446 				  netdev_features_t features, bool tx_path)
3447 {
3448 	struct sk_buff *segs;
3449 
3450 	if (unlikely(skb_needs_check(skb, tx_path))) {
3451 		int err;
3452 
3453 		/* We're going to init ->check field in TCP or UDP header */
3454 		err = skb_cow_head(skb, 0);
3455 		if (err < 0)
3456 			return ERR_PTR(err);
3457 	}
3458 
3459 	/* Only report GSO partial support if it will enable us to
3460 	 * support segmentation on this frame without needing additional
3461 	 * work.
3462 	 */
3463 	if (features & NETIF_F_GSO_PARTIAL) {
3464 		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3465 		struct net_device *dev = skb->dev;
3466 
3467 		partial_features |= dev->features & dev->gso_partial_features;
3468 		if (!skb_gso_ok(skb, features | partial_features))
3469 			features &= ~NETIF_F_GSO_PARTIAL;
3470 	}
3471 
3472 	BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3473 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3474 
3475 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3476 	SKB_GSO_CB(skb)->encap_level = 0;
3477 
3478 	skb_reset_mac_header(skb);
3479 	skb_reset_mac_len(skb);
3480 
3481 	segs = skb_mac_gso_segment(skb, features);
3482 
3483 	if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3484 		skb_warn_bad_offload(skb);
3485 
3486 	return segs;
3487 }
3488 EXPORT_SYMBOL(__skb_gso_segment);
3489 
3490 /* Take action when hardware reception checksum errors are detected. */
3491 #ifdef CONFIG_BUG
3492 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3493 {
3494 	pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3495 	skb_dump(KERN_ERR, skb, true);
3496 	dump_stack();
3497 }
3498 
3499 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3500 {
3501 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3502 }
3503 EXPORT_SYMBOL(netdev_rx_csum_fault);
3504 #endif
3505 
3506 /* XXX: check that highmem exists at all on the given machine. */
3507 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3508 {
3509 #ifdef CONFIG_HIGHMEM
3510 	int i;
3511 
3512 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3513 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3514 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3515 
3516 			if (PageHighMem(skb_frag_page(frag)))
3517 				return 1;
3518 		}
3519 	}
3520 #endif
3521 	return 0;
3522 }
3523 
3524 /* If MPLS offload request, verify we are testing hardware MPLS features
3525  * instead of standard features for the netdev.
3526  */
3527 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3528 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3529 					   netdev_features_t features,
3530 					   __be16 type)
3531 {
3532 	if (eth_p_mpls(type))
3533 		features &= skb->dev->mpls_features;
3534 
3535 	return features;
3536 }
3537 #else
3538 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3539 					   netdev_features_t features,
3540 					   __be16 type)
3541 {
3542 	return features;
3543 }
3544 #endif
3545 
3546 static netdev_features_t harmonize_features(struct sk_buff *skb,
3547 	netdev_features_t features)
3548 {
3549 	__be16 type;
3550 
3551 	type = skb_network_protocol(skb, NULL);
3552 	features = net_mpls_features(skb, features, type);
3553 
3554 	if (skb->ip_summed != CHECKSUM_NONE &&
3555 	    !can_checksum_protocol(features, type)) {
3556 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3557 	}
3558 	if (illegal_highdma(skb->dev, skb))
3559 		features &= ~NETIF_F_SG;
3560 
3561 	return features;
3562 }
3563 
3564 netdev_features_t passthru_features_check(struct sk_buff *skb,
3565 					  struct net_device *dev,
3566 					  netdev_features_t features)
3567 {
3568 	return features;
3569 }
3570 EXPORT_SYMBOL(passthru_features_check);
3571 
3572 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3573 					     struct net_device *dev,
3574 					     netdev_features_t features)
3575 {
3576 	return vlan_features_check(skb, features);
3577 }
3578 
3579 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3580 					    struct net_device *dev,
3581 					    netdev_features_t features)
3582 {
3583 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3584 
3585 	if (gso_segs > dev->gso_max_segs)
3586 		return features & ~NETIF_F_GSO_MASK;
3587 
3588 	if (!skb_shinfo(skb)->gso_type) {
3589 		skb_warn_bad_offload(skb);
3590 		return features & ~NETIF_F_GSO_MASK;
3591 	}
3592 
3593 	/* Support for GSO partial features requires software
3594 	 * intervention before we can actually process the packets
3595 	 * so we need to strip support for any partial features now
3596 	 * and we can pull them back in after we have partially
3597 	 * segmented the frame.
3598 	 */
3599 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3600 		features &= ~dev->gso_partial_features;
3601 
3602 	/* Make sure to clear the IPv4 ID mangling feature if the
3603 	 * IPv4 header has the potential to be fragmented.
3604 	 */
3605 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3606 		struct iphdr *iph = skb->encapsulation ?
3607 				    inner_ip_hdr(skb) : ip_hdr(skb);
3608 
3609 		if (!(iph->frag_off & htons(IP_DF)))
3610 			features &= ~NETIF_F_TSO_MANGLEID;
3611 	}
3612 
3613 	return features;
3614 }
3615 
3616 netdev_features_t netif_skb_features(struct sk_buff *skb)
3617 {
3618 	struct net_device *dev = skb->dev;
3619 	netdev_features_t features = dev->features;
3620 
3621 	if (skb_is_gso(skb))
3622 		features = gso_features_check(skb, dev, features);
3623 
3624 	/* If encapsulation offload request, verify we are testing
3625 	 * hardware encapsulation features instead of standard
3626 	 * features for the netdev
3627 	 */
3628 	if (skb->encapsulation)
3629 		features &= dev->hw_enc_features;
3630 
3631 	if (skb_vlan_tagged(skb))
3632 		features = netdev_intersect_features(features,
3633 						     dev->vlan_features |
3634 						     NETIF_F_HW_VLAN_CTAG_TX |
3635 						     NETIF_F_HW_VLAN_STAG_TX);
3636 
3637 	if (dev->netdev_ops->ndo_features_check)
3638 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3639 								features);
3640 	else
3641 		features &= dflt_features_check(skb, dev, features);
3642 
3643 	return harmonize_features(skb, features);
3644 }
3645 EXPORT_SYMBOL(netif_skb_features);
3646 
3647 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3648 		    struct netdev_queue *txq, bool more)
3649 {
3650 	unsigned int len;
3651 	int rc;
3652 
3653 	if (dev_nit_active(dev))
3654 		dev_queue_xmit_nit(skb, dev);
3655 
3656 	len = skb->len;
3657 	PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3658 	trace_net_dev_start_xmit(skb, dev);
3659 	rc = netdev_start_xmit(skb, dev, txq, more);
3660 	trace_net_dev_xmit(skb, rc, dev, len);
3661 
3662 	return rc;
3663 }
3664 
3665 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3666 				    struct netdev_queue *txq, int *ret)
3667 {
3668 	struct sk_buff *skb = first;
3669 	int rc = NETDEV_TX_OK;
3670 
3671 	while (skb) {
3672 		struct sk_buff *next = skb->next;
3673 
3674 		skb_mark_not_on_list(skb);
3675 		rc = xmit_one(skb, dev, txq, next != NULL);
3676 		if (unlikely(!dev_xmit_complete(rc))) {
3677 			skb->next = next;
3678 			goto out;
3679 		}
3680 
3681 		skb = next;
3682 		if (netif_tx_queue_stopped(txq) && skb) {
3683 			rc = NETDEV_TX_BUSY;
3684 			break;
3685 		}
3686 	}
3687 
3688 out:
3689 	*ret = rc;
3690 	return skb;
3691 }
3692 
3693 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3694 					  netdev_features_t features)
3695 {
3696 	if (skb_vlan_tag_present(skb) &&
3697 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3698 		skb = __vlan_hwaccel_push_inside(skb);
3699 	return skb;
3700 }
3701 
3702 int skb_csum_hwoffload_help(struct sk_buff *skb,
3703 			    const netdev_features_t features)
3704 {
3705 	if (unlikely(skb_csum_is_sctp(skb)))
3706 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3707 			skb_crc32c_csum_help(skb);
3708 
3709 	if (features & NETIF_F_HW_CSUM)
3710 		return 0;
3711 
3712 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3713 		switch (skb->csum_offset) {
3714 		case offsetof(struct tcphdr, check):
3715 		case offsetof(struct udphdr, check):
3716 			return 0;
3717 		}
3718 	}
3719 
3720 	return skb_checksum_help(skb);
3721 }
3722 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3723 
3724 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3725 {
3726 	netdev_features_t features;
3727 
3728 	features = netif_skb_features(skb);
3729 	skb = validate_xmit_vlan(skb, features);
3730 	if (unlikely(!skb))
3731 		goto out_null;
3732 
3733 	skb = sk_validate_xmit_skb(skb, dev);
3734 	if (unlikely(!skb))
3735 		goto out_null;
3736 
3737 	if (netif_needs_gso(skb, features)) {
3738 		struct sk_buff *segs;
3739 
3740 		segs = skb_gso_segment(skb, features);
3741 		if (IS_ERR(segs)) {
3742 			goto out_kfree_skb;
3743 		} else if (segs) {
3744 			consume_skb(skb);
3745 			skb = segs;
3746 		}
3747 	} else {
3748 		if (skb_needs_linearize(skb, features) &&
3749 		    __skb_linearize(skb))
3750 			goto out_kfree_skb;
3751 
3752 		/* If packet is not checksummed and device does not
3753 		 * support checksumming for this protocol, complete
3754 		 * checksumming here.
3755 		 */
3756 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3757 			if (skb->encapsulation)
3758 				skb_set_inner_transport_header(skb,
3759 							       skb_checksum_start_offset(skb));
3760 			else
3761 				skb_set_transport_header(skb,
3762 							 skb_checksum_start_offset(skb));
3763 			if (skb_csum_hwoffload_help(skb, features))
3764 				goto out_kfree_skb;
3765 		}
3766 	}
3767 
3768 	skb = validate_xmit_xfrm(skb, features, again);
3769 
3770 	return skb;
3771 
3772 out_kfree_skb:
3773 	kfree_skb(skb);
3774 out_null:
3775 	atomic_long_inc(&dev->tx_dropped);
3776 	return NULL;
3777 }
3778 
3779 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3780 {
3781 	struct sk_buff *next, *head = NULL, *tail;
3782 
3783 	for (; skb != NULL; skb = next) {
3784 		next = skb->next;
3785 		skb_mark_not_on_list(skb);
3786 
3787 		/* in case skb wont be segmented, point to itself */
3788 		skb->prev = skb;
3789 
3790 		skb = validate_xmit_skb(skb, dev, again);
3791 		if (!skb)
3792 			continue;
3793 
3794 		if (!head)
3795 			head = skb;
3796 		else
3797 			tail->next = skb;
3798 		/* If skb was segmented, skb->prev points to
3799 		 * the last segment. If not, it still contains skb.
3800 		 */
3801 		tail = skb->prev;
3802 	}
3803 	return head;
3804 }
3805 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3806 
3807 static void qdisc_pkt_len_init(struct sk_buff *skb)
3808 {
3809 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3810 
3811 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3812 
3813 	/* To get more precise estimation of bytes sent on wire,
3814 	 * we add to pkt_len the headers size of all segments
3815 	 */
3816 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3817 		unsigned int hdr_len;
3818 		u16 gso_segs = shinfo->gso_segs;
3819 
3820 		/* mac layer + network layer */
3821 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3822 
3823 		/* + transport layer */
3824 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3825 			const struct tcphdr *th;
3826 			struct tcphdr _tcphdr;
3827 
3828 			th = skb_header_pointer(skb, skb_transport_offset(skb),
3829 						sizeof(_tcphdr), &_tcphdr);
3830 			if (likely(th))
3831 				hdr_len += __tcp_hdrlen(th);
3832 		} else {
3833 			struct udphdr _udphdr;
3834 
3835 			if (skb_header_pointer(skb, skb_transport_offset(skb),
3836 					       sizeof(_udphdr), &_udphdr))
3837 				hdr_len += sizeof(struct udphdr);
3838 		}
3839 
3840 		if (shinfo->gso_type & SKB_GSO_DODGY)
3841 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3842 						shinfo->gso_size);
3843 
3844 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3845 	}
3846 }
3847 
3848 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3849 			     struct sk_buff **to_free,
3850 			     struct netdev_queue *txq)
3851 {
3852 	int rc;
3853 
3854 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3855 	if (rc == NET_XMIT_SUCCESS)
3856 		trace_qdisc_enqueue(q, txq, skb);
3857 	return rc;
3858 }
3859 
3860 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3861 				 struct net_device *dev,
3862 				 struct netdev_queue *txq)
3863 {
3864 	spinlock_t *root_lock = qdisc_lock(q);
3865 	struct sk_buff *to_free = NULL;
3866 	bool contended;
3867 	int rc;
3868 
3869 	qdisc_calculate_pkt_len(skb, q);
3870 
3871 	if (q->flags & TCQ_F_NOLOCK) {
3872 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3873 		    qdisc_run_begin(q)) {
3874 			/* Retest nolock_qdisc_is_empty() within the protection
3875 			 * of q->seqlock to protect from racing with requeuing.
3876 			 */
3877 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3878 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3879 				__qdisc_run(q);
3880 				qdisc_run_end(q);
3881 
3882 				goto no_lock_out;
3883 			}
3884 
3885 			qdisc_bstats_cpu_update(q, skb);
3886 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3887 			    !nolock_qdisc_is_empty(q))
3888 				__qdisc_run(q);
3889 
3890 			qdisc_run_end(q);
3891 			return NET_XMIT_SUCCESS;
3892 		}
3893 
3894 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3895 		qdisc_run(q);
3896 
3897 no_lock_out:
3898 		if (unlikely(to_free))
3899 			kfree_skb_list(to_free);
3900 		return rc;
3901 	}
3902 
3903 	/*
3904 	 * Heuristic to force contended enqueues to serialize on a
3905 	 * separate lock before trying to get qdisc main lock.
3906 	 * This permits qdisc->running owner to get the lock more
3907 	 * often and dequeue packets faster.
3908 	 */
3909 	contended = qdisc_is_running(q);
3910 	if (unlikely(contended))
3911 		spin_lock(&q->busylock);
3912 
3913 	spin_lock(root_lock);
3914 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3915 		__qdisc_drop(skb, &to_free);
3916 		rc = NET_XMIT_DROP;
3917 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3918 		   qdisc_run_begin(q)) {
3919 		/*
3920 		 * This is a work-conserving queue; there are no old skbs
3921 		 * waiting to be sent out; and the qdisc is not running -
3922 		 * xmit the skb directly.
3923 		 */
3924 
3925 		qdisc_bstats_update(q, skb);
3926 
3927 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3928 			if (unlikely(contended)) {
3929 				spin_unlock(&q->busylock);
3930 				contended = false;
3931 			}
3932 			__qdisc_run(q);
3933 		}
3934 
3935 		qdisc_run_end(q);
3936 		rc = NET_XMIT_SUCCESS;
3937 	} else {
3938 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3939 		if (qdisc_run_begin(q)) {
3940 			if (unlikely(contended)) {
3941 				spin_unlock(&q->busylock);
3942 				contended = false;
3943 			}
3944 			__qdisc_run(q);
3945 			qdisc_run_end(q);
3946 		}
3947 	}
3948 	spin_unlock(root_lock);
3949 	if (unlikely(to_free))
3950 		kfree_skb_list(to_free);
3951 	if (unlikely(contended))
3952 		spin_unlock(&q->busylock);
3953 	return rc;
3954 }
3955 
3956 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3957 static void skb_update_prio(struct sk_buff *skb)
3958 {
3959 	const struct netprio_map *map;
3960 	const struct sock *sk;
3961 	unsigned int prioidx;
3962 
3963 	if (skb->priority)
3964 		return;
3965 	map = rcu_dereference_bh(skb->dev->priomap);
3966 	if (!map)
3967 		return;
3968 	sk = skb_to_full_sk(skb);
3969 	if (!sk)
3970 		return;
3971 
3972 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3973 
3974 	if (prioidx < map->priomap_len)
3975 		skb->priority = map->priomap[prioidx];
3976 }
3977 #else
3978 #define skb_update_prio(skb)
3979 #endif
3980 
3981 /**
3982  *	dev_loopback_xmit - loop back @skb
3983  *	@net: network namespace this loopback is happening in
3984  *	@sk:  sk needed to be a netfilter okfn
3985  *	@skb: buffer to transmit
3986  */
3987 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3988 {
3989 	skb_reset_mac_header(skb);
3990 	__skb_pull(skb, skb_network_offset(skb));
3991 	skb->pkt_type = PACKET_LOOPBACK;
3992 	skb->ip_summed = CHECKSUM_UNNECESSARY;
3993 	WARN_ON(!skb_dst(skb));
3994 	skb_dst_force(skb);
3995 	netif_rx_ni(skb);
3996 	return 0;
3997 }
3998 EXPORT_SYMBOL(dev_loopback_xmit);
3999 
4000 #ifdef CONFIG_NET_EGRESS
4001 static struct sk_buff *
4002 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4003 {
4004 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
4005 	struct tcf_result cl_res;
4006 
4007 	if (!miniq)
4008 		return skb;
4009 
4010 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
4011 	qdisc_skb_cb(skb)->mru = 0;
4012 	qdisc_skb_cb(skb)->post_ct = false;
4013 	mini_qdisc_bstats_cpu_update(miniq, skb);
4014 
4015 	switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4016 	case TC_ACT_OK:
4017 	case TC_ACT_RECLASSIFY:
4018 		skb->tc_index = TC_H_MIN(cl_res.classid);
4019 		break;
4020 	case TC_ACT_SHOT:
4021 		mini_qdisc_qstats_cpu_drop(miniq);
4022 		*ret = NET_XMIT_DROP;
4023 		kfree_skb(skb);
4024 		return NULL;
4025 	case TC_ACT_STOLEN:
4026 	case TC_ACT_QUEUED:
4027 	case TC_ACT_TRAP:
4028 		*ret = NET_XMIT_SUCCESS;
4029 		consume_skb(skb);
4030 		return NULL;
4031 	case TC_ACT_REDIRECT:
4032 		/* No need to push/pop skb's mac_header here on egress! */
4033 		skb_do_redirect(skb);
4034 		*ret = NET_XMIT_SUCCESS;
4035 		return NULL;
4036 	default:
4037 		break;
4038 	}
4039 
4040 	return skb;
4041 }
4042 #endif /* CONFIG_NET_EGRESS */
4043 
4044 #ifdef CONFIG_XPS
4045 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4046 			       struct xps_dev_maps *dev_maps, unsigned int tci)
4047 {
4048 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
4049 	struct xps_map *map;
4050 	int queue_index = -1;
4051 
4052 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4053 		return queue_index;
4054 
4055 	tci *= dev_maps->num_tc;
4056 	tci += tc;
4057 
4058 	map = rcu_dereference(dev_maps->attr_map[tci]);
4059 	if (map) {
4060 		if (map->len == 1)
4061 			queue_index = map->queues[0];
4062 		else
4063 			queue_index = map->queues[reciprocal_scale(
4064 						skb_get_hash(skb), map->len)];
4065 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4066 			queue_index = -1;
4067 	}
4068 	return queue_index;
4069 }
4070 #endif
4071 
4072 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4073 			 struct sk_buff *skb)
4074 {
4075 #ifdef CONFIG_XPS
4076 	struct xps_dev_maps *dev_maps;
4077 	struct sock *sk = skb->sk;
4078 	int queue_index = -1;
4079 
4080 	if (!static_key_false(&xps_needed))
4081 		return -1;
4082 
4083 	rcu_read_lock();
4084 	if (!static_key_false(&xps_rxqs_needed))
4085 		goto get_cpus_map;
4086 
4087 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4088 	if (dev_maps) {
4089 		int tci = sk_rx_queue_get(sk);
4090 
4091 		if (tci >= 0)
4092 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4093 							  tci);
4094 	}
4095 
4096 get_cpus_map:
4097 	if (queue_index < 0) {
4098 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4099 		if (dev_maps) {
4100 			unsigned int tci = skb->sender_cpu - 1;
4101 
4102 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4103 							  tci);
4104 		}
4105 	}
4106 	rcu_read_unlock();
4107 
4108 	return queue_index;
4109 #else
4110 	return -1;
4111 #endif
4112 }
4113 
4114 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4115 		     struct net_device *sb_dev)
4116 {
4117 	return 0;
4118 }
4119 EXPORT_SYMBOL(dev_pick_tx_zero);
4120 
4121 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4122 		       struct net_device *sb_dev)
4123 {
4124 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4125 }
4126 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4127 
4128 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4129 		     struct net_device *sb_dev)
4130 {
4131 	struct sock *sk = skb->sk;
4132 	int queue_index = sk_tx_queue_get(sk);
4133 
4134 	sb_dev = sb_dev ? : dev;
4135 
4136 	if (queue_index < 0 || skb->ooo_okay ||
4137 	    queue_index >= dev->real_num_tx_queues) {
4138 		int new_index = get_xps_queue(dev, sb_dev, skb);
4139 
4140 		if (new_index < 0)
4141 			new_index = skb_tx_hash(dev, sb_dev, skb);
4142 
4143 		if (queue_index != new_index && sk &&
4144 		    sk_fullsock(sk) &&
4145 		    rcu_access_pointer(sk->sk_dst_cache))
4146 			sk_tx_queue_set(sk, new_index);
4147 
4148 		queue_index = new_index;
4149 	}
4150 
4151 	return queue_index;
4152 }
4153 EXPORT_SYMBOL(netdev_pick_tx);
4154 
4155 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4156 					 struct sk_buff *skb,
4157 					 struct net_device *sb_dev)
4158 {
4159 	int queue_index = 0;
4160 
4161 #ifdef CONFIG_XPS
4162 	u32 sender_cpu = skb->sender_cpu - 1;
4163 
4164 	if (sender_cpu >= (u32)NR_CPUS)
4165 		skb->sender_cpu = raw_smp_processor_id() + 1;
4166 #endif
4167 
4168 	if (dev->real_num_tx_queues != 1) {
4169 		const struct net_device_ops *ops = dev->netdev_ops;
4170 
4171 		if (ops->ndo_select_queue)
4172 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4173 		else
4174 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4175 
4176 		queue_index = netdev_cap_txqueue(dev, queue_index);
4177 	}
4178 
4179 	skb_set_queue_mapping(skb, queue_index);
4180 	return netdev_get_tx_queue(dev, queue_index);
4181 }
4182 
4183 /**
4184  *	__dev_queue_xmit - transmit a buffer
4185  *	@skb: buffer to transmit
4186  *	@sb_dev: suboordinate device used for L2 forwarding offload
4187  *
4188  *	Queue a buffer for transmission to a network device. The caller must
4189  *	have set the device and priority and built the buffer before calling
4190  *	this function. The function can be called from an interrupt.
4191  *
4192  *	A negative errno code is returned on a failure. A success does not
4193  *	guarantee the frame will be transmitted as it may be dropped due
4194  *	to congestion or traffic shaping.
4195  *
4196  * -----------------------------------------------------------------------------------
4197  *      I notice this method can also return errors from the queue disciplines,
4198  *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
4199  *      be positive.
4200  *
4201  *      Regardless of the return value, the skb is consumed, so it is currently
4202  *      difficult to retry a send to this method.  (You can bump the ref count
4203  *      before sending to hold a reference for retry if you are careful.)
4204  *
4205  *      When calling this method, interrupts MUST be enabled.  This is because
4206  *      the BH enable code must have IRQs enabled so that it will not deadlock.
4207  *          --BLG
4208  */
4209 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4210 {
4211 	struct net_device *dev = skb->dev;
4212 	struct netdev_queue *txq;
4213 	struct Qdisc *q;
4214 	int rc = -ENOMEM;
4215 	bool again = false;
4216 
4217 	skb_reset_mac_header(skb);
4218 
4219 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4220 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4221 
4222 	/* Disable soft irqs for various locks below. Also
4223 	 * stops preemption for RCU.
4224 	 */
4225 	rcu_read_lock_bh();
4226 
4227 	skb_update_prio(skb);
4228 
4229 	qdisc_pkt_len_init(skb);
4230 #ifdef CONFIG_NET_CLS_ACT
4231 	skb->tc_at_ingress = 0;
4232 # ifdef CONFIG_NET_EGRESS
4233 	if (static_branch_unlikely(&egress_needed_key)) {
4234 		skb = sch_handle_egress(skb, &rc, dev);
4235 		if (!skb)
4236 			goto out;
4237 	}
4238 # endif
4239 #endif
4240 	/* If device/qdisc don't need skb->dst, release it right now while
4241 	 * its hot in this cpu cache.
4242 	 */
4243 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4244 		skb_dst_drop(skb);
4245 	else
4246 		skb_dst_force(skb);
4247 
4248 	txq = netdev_core_pick_tx(dev, skb, sb_dev);
4249 	q = rcu_dereference_bh(txq->qdisc);
4250 
4251 	trace_net_dev_queue(skb);
4252 	if (q->enqueue) {
4253 		rc = __dev_xmit_skb(skb, q, dev, txq);
4254 		goto out;
4255 	}
4256 
4257 	/* The device has no queue. Common case for software devices:
4258 	 * loopback, all the sorts of tunnels...
4259 
4260 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4261 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4262 	 * counters.)
4263 	 * However, it is possible, that they rely on protection
4264 	 * made by us here.
4265 
4266 	 * Check this and shot the lock. It is not prone from deadlocks.
4267 	 *Either shot noqueue qdisc, it is even simpler 8)
4268 	 */
4269 	if (dev->flags & IFF_UP) {
4270 		int cpu = smp_processor_id(); /* ok because BHs are off */
4271 
4272 		if (txq->xmit_lock_owner != cpu) {
4273 			if (dev_xmit_recursion())
4274 				goto recursion_alert;
4275 
4276 			skb = validate_xmit_skb(skb, dev, &again);
4277 			if (!skb)
4278 				goto out;
4279 
4280 			PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4281 			HARD_TX_LOCK(dev, txq, cpu);
4282 
4283 			if (!netif_xmit_stopped(txq)) {
4284 				dev_xmit_recursion_inc();
4285 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4286 				dev_xmit_recursion_dec();
4287 				if (dev_xmit_complete(rc)) {
4288 					HARD_TX_UNLOCK(dev, txq);
4289 					goto out;
4290 				}
4291 			}
4292 			HARD_TX_UNLOCK(dev, txq);
4293 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4294 					     dev->name);
4295 		} else {
4296 			/* Recursion is detected! It is possible,
4297 			 * unfortunately
4298 			 */
4299 recursion_alert:
4300 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4301 					     dev->name);
4302 		}
4303 	}
4304 
4305 	rc = -ENETDOWN;
4306 	rcu_read_unlock_bh();
4307 
4308 	atomic_long_inc(&dev->tx_dropped);
4309 	kfree_skb_list(skb);
4310 	return rc;
4311 out:
4312 	rcu_read_unlock_bh();
4313 	return rc;
4314 }
4315 
4316 int dev_queue_xmit(struct sk_buff *skb)
4317 {
4318 	return __dev_queue_xmit(skb, NULL);
4319 }
4320 EXPORT_SYMBOL(dev_queue_xmit);
4321 
4322 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4323 {
4324 	return __dev_queue_xmit(skb, sb_dev);
4325 }
4326 EXPORT_SYMBOL(dev_queue_xmit_accel);
4327 
4328 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4329 {
4330 	struct net_device *dev = skb->dev;
4331 	struct sk_buff *orig_skb = skb;
4332 	struct netdev_queue *txq;
4333 	int ret = NETDEV_TX_BUSY;
4334 	bool again = false;
4335 
4336 	if (unlikely(!netif_running(dev) ||
4337 		     !netif_carrier_ok(dev)))
4338 		goto drop;
4339 
4340 	skb = validate_xmit_skb_list(skb, dev, &again);
4341 	if (skb != orig_skb)
4342 		goto drop;
4343 
4344 	skb_set_queue_mapping(skb, queue_id);
4345 	txq = skb_get_tx_queue(dev, skb);
4346 	PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4347 
4348 	local_bh_disable();
4349 
4350 	dev_xmit_recursion_inc();
4351 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4352 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4353 		ret = netdev_start_xmit(skb, dev, txq, false);
4354 	HARD_TX_UNLOCK(dev, txq);
4355 	dev_xmit_recursion_dec();
4356 
4357 	local_bh_enable();
4358 	return ret;
4359 drop:
4360 	atomic_long_inc(&dev->tx_dropped);
4361 	kfree_skb_list(skb);
4362 	return NET_XMIT_DROP;
4363 }
4364 EXPORT_SYMBOL(__dev_direct_xmit);
4365 
4366 /*************************************************************************
4367  *			Receiver routines
4368  *************************************************************************/
4369 
4370 int netdev_max_backlog __read_mostly = 1000;
4371 EXPORT_SYMBOL(netdev_max_backlog);
4372 
4373 int netdev_tstamp_prequeue __read_mostly = 1;
4374 int netdev_budget __read_mostly = 300;
4375 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4376 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4377 int weight_p __read_mostly = 64;           /* old backlog weight */
4378 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4379 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4380 int dev_rx_weight __read_mostly = 64;
4381 int dev_tx_weight __read_mostly = 64;
4382 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4383 int gro_normal_batch __read_mostly = 8;
4384 
4385 /* Called with irq disabled */
4386 static inline void ____napi_schedule(struct softnet_data *sd,
4387 				     struct napi_struct *napi)
4388 {
4389 	struct task_struct *thread;
4390 
4391 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4392 		/* Paired with smp_mb__before_atomic() in
4393 		 * napi_enable()/dev_set_threaded().
4394 		 * Use READ_ONCE() to guarantee a complete
4395 		 * read on napi->thread. Only call
4396 		 * wake_up_process() when it's not NULL.
4397 		 */
4398 		thread = READ_ONCE(napi->thread);
4399 		if (thread) {
4400 			/* Avoid doing set_bit() if the thread is in
4401 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4402 			 * makes sure to proceed with napi polling
4403 			 * if the thread is explicitly woken from here.
4404 			 */
4405 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4406 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4407 			wake_up_process(thread);
4408 			return;
4409 		}
4410 	}
4411 
4412 	list_add_tail(&napi->poll_list, &sd->poll_list);
4413 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4414 }
4415 
4416 #ifdef CONFIG_RPS
4417 
4418 /* One global table that all flow-based protocols share. */
4419 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4420 EXPORT_SYMBOL(rps_sock_flow_table);
4421 u32 rps_cpu_mask __read_mostly;
4422 EXPORT_SYMBOL(rps_cpu_mask);
4423 
4424 struct static_key_false rps_needed __read_mostly;
4425 EXPORT_SYMBOL(rps_needed);
4426 struct static_key_false rfs_needed __read_mostly;
4427 EXPORT_SYMBOL(rfs_needed);
4428 
4429 static struct rps_dev_flow *
4430 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4431 	    struct rps_dev_flow *rflow, u16 next_cpu)
4432 {
4433 	if (next_cpu < nr_cpu_ids) {
4434 #ifdef CONFIG_RFS_ACCEL
4435 		struct netdev_rx_queue *rxqueue;
4436 		struct rps_dev_flow_table *flow_table;
4437 		struct rps_dev_flow *old_rflow;
4438 		u32 flow_id;
4439 		u16 rxq_index;
4440 		int rc;
4441 
4442 		/* Should we steer this flow to a different hardware queue? */
4443 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4444 		    !(dev->features & NETIF_F_NTUPLE))
4445 			goto out;
4446 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4447 		if (rxq_index == skb_get_rx_queue(skb))
4448 			goto out;
4449 
4450 		rxqueue = dev->_rx + rxq_index;
4451 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4452 		if (!flow_table)
4453 			goto out;
4454 		flow_id = skb_get_hash(skb) & flow_table->mask;
4455 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4456 							rxq_index, flow_id);
4457 		if (rc < 0)
4458 			goto out;
4459 		old_rflow = rflow;
4460 		rflow = &flow_table->flows[flow_id];
4461 		rflow->filter = rc;
4462 		if (old_rflow->filter == rflow->filter)
4463 			old_rflow->filter = RPS_NO_FILTER;
4464 	out:
4465 #endif
4466 		rflow->last_qtail =
4467 			per_cpu(softnet_data, next_cpu).input_queue_head;
4468 	}
4469 
4470 	rflow->cpu = next_cpu;
4471 	return rflow;
4472 }
4473 
4474 /*
4475  * get_rps_cpu is called from netif_receive_skb and returns the target
4476  * CPU from the RPS map of the receiving queue for a given skb.
4477  * rcu_read_lock must be held on entry.
4478  */
4479 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4480 		       struct rps_dev_flow **rflowp)
4481 {
4482 	const struct rps_sock_flow_table *sock_flow_table;
4483 	struct netdev_rx_queue *rxqueue = dev->_rx;
4484 	struct rps_dev_flow_table *flow_table;
4485 	struct rps_map *map;
4486 	int cpu = -1;
4487 	u32 tcpu;
4488 	u32 hash;
4489 
4490 	if (skb_rx_queue_recorded(skb)) {
4491 		u16 index = skb_get_rx_queue(skb);
4492 
4493 		if (unlikely(index >= dev->real_num_rx_queues)) {
4494 			WARN_ONCE(dev->real_num_rx_queues > 1,
4495 				  "%s received packet on queue %u, but number "
4496 				  "of RX queues is %u\n",
4497 				  dev->name, index, dev->real_num_rx_queues);
4498 			goto done;
4499 		}
4500 		rxqueue += index;
4501 	}
4502 
4503 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4504 
4505 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4506 	map = rcu_dereference(rxqueue->rps_map);
4507 	if (!flow_table && !map)
4508 		goto done;
4509 
4510 	skb_reset_network_header(skb);
4511 	hash = skb_get_hash(skb);
4512 	if (!hash)
4513 		goto done;
4514 
4515 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4516 	if (flow_table && sock_flow_table) {
4517 		struct rps_dev_flow *rflow;
4518 		u32 next_cpu;
4519 		u32 ident;
4520 
4521 		/* First check into global flow table if there is a match */
4522 		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4523 		if ((ident ^ hash) & ~rps_cpu_mask)
4524 			goto try_rps;
4525 
4526 		next_cpu = ident & rps_cpu_mask;
4527 
4528 		/* OK, now we know there is a match,
4529 		 * we can look at the local (per receive queue) flow table
4530 		 */
4531 		rflow = &flow_table->flows[hash & flow_table->mask];
4532 		tcpu = rflow->cpu;
4533 
4534 		/*
4535 		 * If the desired CPU (where last recvmsg was done) is
4536 		 * different from current CPU (one in the rx-queue flow
4537 		 * table entry), switch if one of the following holds:
4538 		 *   - Current CPU is unset (>= nr_cpu_ids).
4539 		 *   - Current CPU is offline.
4540 		 *   - The current CPU's queue tail has advanced beyond the
4541 		 *     last packet that was enqueued using this table entry.
4542 		 *     This guarantees that all previous packets for the flow
4543 		 *     have been dequeued, thus preserving in order delivery.
4544 		 */
4545 		if (unlikely(tcpu != next_cpu) &&
4546 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4547 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4548 		      rflow->last_qtail)) >= 0)) {
4549 			tcpu = next_cpu;
4550 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4551 		}
4552 
4553 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4554 			*rflowp = rflow;
4555 			cpu = tcpu;
4556 			goto done;
4557 		}
4558 	}
4559 
4560 try_rps:
4561 
4562 	if (map) {
4563 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4564 		if (cpu_online(tcpu)) {
4565 			cpu = tcpu;
4566 			goto done;
4567 		}
4568 	}
4569 
4570 done:
4571 	return cpu;
4572 }
4573 
4574 #ifdef CONFIG_RFS_ACCEL
4575 
4576 /**
4577  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4578  * @dev: Device on which the filter was set
4579  * @rxq_index: RX queue index
4580  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4581  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4582  *
4583  * Drivers that implement ndo_rx_flow_steer() should periodically call
4584  * this function for each installed filter and remove the filters for
4585  * which it returns %true.
4586  */
4587 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4588 			 u32 flow_id, u16 filter_id)
4589 {
4590 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4591 	struct rps_dev_flow_table *flow_table;
4592 	struct rps_dev_flow *rflow;
4593 	bool expire = true;
4594 	unsigned int cpu;
4595 
4596 	rcu_read_lock();
4597 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4598 	if (flow_table && flow_id <= flow_table->mask) {
4599 		rflow = &flow_table->flows[flow_id];
4600 		cpu = READ_ONCE(rflow->cpu);
4601 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4602 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4603 			   rflow->last_qtail) <
4604 		     (int)(10 * flow_table->mask)))
4605 			expire = false;
4606 	}
4607 	rcu_read_unlock();
4608 	return expire;
4609 }
4610 EXPORT_SYMBOL(rps_may_expire_flow);
4611 
4612 #endif /* CONFIG_RFS_ACCEL */
4613 
4614 /* Called from hardirq (IPI) context */
4615 static void rps_trigger_softirq(void *data)
4616 {
4617 	struct softnet_data *sd = data;
4618 
4619 	____napi_schedule(sd, &sd->backlog);
4620 	sd->received_rps++;
4621 }
4622 
4623 #endif /* CONFIG_RPS */
4624 
4625 /*
4626  * Check if this softnet_data structure is another cpu one
4627  * If yes, queue it to our IPI list and return 1
4628  * If no, return 0
4629  */
4630 static int rps_ipi_queued(struct softnet_data *sd)
4631 {
4632 #ifdef CONFIG_RPS
4633 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4634 
4635 	if (sd != mysd) {
4636 		sd->rps_ipi_next = mysd->rps_ipi_list;
4637 		mysd->rps_ipi_list = sd;
4638 
4639 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4640 		return 1;
4641 	}
4642 #endif /* CONFIG_RPS */
4643 	return 0;
4644 }
4645 
4646 #ifdef CONFIG_NET_FLOW_LIMIT
4647 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4648 #endif
4649 
4650 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4651 {
4652 #ifdef CONFIG_NET_FLOW_LIMIT
4653 	struct sd_flow_limit *fl;
4654 	struct softnet_data *sd;
4655 	unsigned int old_flow, new_flow;
4656 
4657 	if (qlen < (netdev_max_backlog >> 1))
4658 		return false;
4659 
4660 	sd = this_cpu_ptr(&softnet_data);
4661 
4662 	rcu_read_lock();
4663 	fl = rcu_dereference(sd->flow_limit);
4664 	if (fl) {
4665 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4666 		old_flow = fl->history[fl->history_head];
4667 		fl->history[fl->history_head] = new_flow;
4668 
4669 		fl->history_head++;
4670 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4671 
4672 		if (likely(fl->buckets[old_flow]))
4673 			fl->buckets[old_flow]--;
4674 
4675 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4676 			fl->count++;
4677 			rcu_read_unlock();
4678 			return true;
4679 		}
4680 	}
4681 	rcu_read_unlock();
4682 #endif
4683 	return false;
4684 }
4685 
4686 /*
4687  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4688  * queue (may be a remote CPU queue).
4689  */
4690 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4691 			      unsigned int *qtail)
4692 {
4693 	struct softnet_data *sd;
4694 	unsigned long flags;
4695 	unsigned int qlen;
4696 
4697 	sd = &per_cpu(softnet_data, cpu);
4698 
4699 	local_irq_save(flags);
4700 
4701 	rps_lock(sd);
4702 	if (!netif_running(skb->dev))
4703 		goto drop;
4704 	qlen = skb_queue_len(&sd->input_pkt_queue);
4705 	if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4706 		if (qlen) {
4707 enqueue:
4708 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4709 			input_queue_tail_incr_save(sd, qtail);
4710 			rps_unlock(sd);
4711 			local_irq_restore(flags);
4712 			return NET_RX_SUCCESS;
4713 		}
4714 
4715 		/* Schedule NAPI for backlog device
4716 		 * We can use non atomic operation since we own the queue lock
4717 		 */
4718 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4719 			if (!rps_ipi_queued(sd))
4720 				____napi_schedule(sd, &sd->backlog);
4721 		}
4722 		goto enqueue;
4723 	}
4724 
4725 drop:
4726 	sd->dropped++;
4727 	rps_unlock(sd);
4728 
4729 	local_irq_restore(flags);
4730 
4731 	atomic_long_inc(&skb->dev->rx_dropped);
4732 	kfree_skb(skb);
4733 	return NET_RX_DROP;
4734 }
4735 
4736 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4737 {
4738 	struct net_device *dev = skb->dev;
4739 	struct netdev_rx_queue *rxqueue;
4740 
4741 	rxqueue = dev->_rx;
4742 
4743 	if (skb_rx_queue_recorded(skb)) {
4744 		u16 index = skb_get_rx_queue(skb);
4745 
4746 		if (unlikely(index >= dev->real_num_rx_queues)) {
4747 			WARN_ONCE(dev->real_num_rx_queues > 1,
4748 				  "%s received packet on queue %u, but number "
4749 				  "of RX queues is %u\n",
4750 				  dev->name, index, dev->real_num_rx_queues);
4751 
4752 			return rxqueue; /* Return first rxqueue */
4753 		}
4754 		rxqueue += index;
4755 	}
4756 	return rxqueue;
4757 }
4758 
4759 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4760 			     struct bpf_prog *xdp_prog)
4761 {
4762 	void *orig_data, *orig_data_end, *hard_start;
4763 	struct netdev_rx_queue *rxqueue;
4764 	bool orig_bcast, orig_host;
4765 	u32 mac_len, frame_sz;
4766 	__be16 orig_eth_type;
4767 	struct ethhdr *eth;
4768 	u32 metalen, act;
4769 	int off;
4770 
4771 	/* The XDP program wants to see the packet starting at the MAC
4772 	 * header.
4773 	 */
4774 	mac_len = skb->data - skb_mac_header(skb);
4775 	hard_start = skb->data - skb_headroom(skb);
4776 
4777 	/* SKB "head" area always have tailroom for skb_shared_info */
4778 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4779 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4780 
4781 	rxqueue = netif_get_rxqueue(skb);
4782 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4783 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4784 			 skb_headlen(skb) + mac_len, true);
4785 
4786 	orig_data_end = xdp->data_end;
4787 	orig_data = xdp->data;
4788 	eth = (struct ethhdr *)xdp->data;
4789 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4790 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4791 	orig_eth_type = eth->h_proto;
4792 
4793 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4794 
4795 	/* check if bpf_xdp_adjust_head was used */
4796 	off = xdp->data - orig_data;
4797 	if (off) {
4798 		if (off > 0)
4799 			__skb_pull(skb, off);
4800 		else if (off < 0)
4801 			__skb_push(skb, -off);
4802 
4803 		skb->mac_header += off;
4804 		skb_reset_network_header(skb);
4805 	}
4806 
4807 	/* check if bpf_xdp_adjust_tail was used */
4808 	off = xdp->data_end - orig_data_end;
4809 	if (off != 0) {
4810 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4811 		skb->len += off; /* positive on grow, negative on shrink */
4812 	}
4813 
4814 	/* check if XDP changed eth hdr such SKB needs update */
4815 	eth = (struct ethhdr *)xdp->data;
4816 	if ((orig_eth_type != eth->h_proto) ||
4817 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4818 						  skb->dev->dev_addr)) ||
4819 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4820 		__skb_push(skb, ETH_HLEN);
4821 		skb->pkt_type = PACKET_HOST;
4822 		skb->protocol = eth_type_trans(skb, skb->dev);
4823 	}
4824 
4825 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4826 	 * before calling us again on redirect path. We do not call do_redirect
4827 	 * as we leave that up to the caller.
4828 	 *
4829 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4830 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4831 	 */
4832 	switch (act) {
4833 	case XDP_REDIRECT:
4834 	case XDP_TX:
4835 		__skb_push(skb, mac_len);
4836 		break;
4837 	case XDP_PASS:
4838 		metalen = xdp->data - xdp->data_meta;
4839 		if (metalen)
4840 			skb_metadata_set(skb, metalen);
4841 		break;
4842 	}
4843 
4844 	return act;
4845 }
4846 
4847 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4848 				     struct xdp_buff *xdp,
4849 				     struct bpf_prog *xdp_prog)
4850 {
4851 	u32 act = XDP_DROP;
4852 
4853 	/* Reinjected packets coming from act_mirred or similar should
4854 	 * not get XDP generic processing.
4855 	 */
4856 	if (skb_is_redirected(skb))
4857 		return XDP_PASS;
4858 
4859 	/* XDP packets must be linear and must have sufficient headroom
4860 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4861 	 * native XDP provides, thus we need to do it here as well.
4862 	 */
4863 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4864 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4865 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4866 		int troom = skb->tail + skb->data_len - skb->end;
4867 
4868 		/* In case we have to go down the path and also linearize,
4869 		 * then lets do the pskb_expand_head() work just once here.
4870 		 */
4871 		if (pskb_expand_head(skb,
4872 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4873 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4874 			goto do_drop;
4875 		if (skb_linearize(skb))
4876 			goto do_drop;
4877 	}
4878 
4879 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4880 	switch (act) {
4881 	case XDP_REDIRECT:
4882 	case XDP_TX:
4883 	case XDP_PASS:
4884 		break;
4885 	default:
4886 		bpf_warn_invalid_xdp_action(act);
4887 		fallthrough;
4888 	case XDP_ABORTED:
4889 		trace_xdp_exception(skb->dev, xdp_prog, act);
4890 		fallthrough;
4891 	case XDP_DROP:
4892 	do_drop:
4893 		kfree_skb(skb);
4894 		break;
4895 	}
4896 
4897 	return act;
4898 }
4899 
4900 /* When doing generic XDP we have to bypass the qdisc layer and the
4901  * network taps in order to match in-driver-XDP behavior.
4902  */
4903 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4904 {
4905 	struct net_device *dev = skb->dev;
4906 	struct netdev_queue *txq;
4907 	bool free_skb = true;
4908 	int cpu, rc;
4909 
4910 	txq = netdev_core_pick_tx(dev, skb, NULL);
4911 	cpu = smp_processor_id();
4912 	HARD_TX_LOCK(dev, txq, cpu);
4913 	if (!netif_xmit_stopped(txq)) {
4914 		rc = netdev_start_xmit(skb, dev, txq, 0);
4915 		if (dev_xmit_complete(rc))
4916 			free_skb = false;
4917 	}
4918 	HARD_TX_UNLOCK(dev, txq);
4919 	if (free_skb) {
4920 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4921 		kfree_skb(skb);
4922 	}
4923 }
4924 
4925 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4926 
4927 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4928 {
4929 	if (xdp_prog) {
4930 		struct xdp_buff xdp;
4931 		u32 act;
4932 		int err;
4933 
4934 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4935 		if (act != XDP_PASS) {
4936 			switch (act) {
4937 			case XDP_REDIRECT:
4938 				err = xdp_do_generic_redirect(skb->dev, skb,
4939 							      &xdp, xdp_prog);
4940 				if (err)
4941 					goto out_redir;
4942 				break;
4943 			case XDP_TX:
4944 				generic_xdp_tx(skb, xdp_prog);
4945 				break;
4946 			}
4947 			return XDP_DROP;
4948 		}
4949 	}
4950 	return XDP_PASS;
4951 out_redir:
4952 	kfree_skb(skb);
4953 	return XDP_DROP;
4954 }
4955 EXPORT_SYMBOL_GPL(do_xdp_generic);
4956 
4957 static int netif_rx_internal(struct sk_buff *skb)
4958 {
4959 	int ret;
4960 
4961 	net_timestamp_check(netdev_tstamp_prequeue, skb);
4962 
4963 	trace_netif_rx(skb);
4964 
4965 #ifdef CONFIG_RPS
4966 	if (static_branch_unlikely(&rps_needed)) {
4967 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4968 		int cpu;
4969 
4970 		preempt_disable();
4971 		rcu_read_lock();
4972 
4973 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4974 		if (cpu < 0)
4975 			cpu = smp_processor_id();
4976 
4977 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4978 
4979 		rcu_read_unlock();
4980 		preempt_enable();
4981 	} else
4982 #endif
4983 	{
4984 		unsigned int qtail;
4985 
4986 		ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4987 		put_cpu();
4988 	}
4989 	return ret;
4990 }
4991 
4992 /**
4993  *	netif_rx	-	post buffer to the network code
4994  *	@skb: buffer to post
4995  *
4996  *	This function receives a packet from a device driver and queues it for
4997  *	the upper (protocol) levels to process.  It always succeeds. The buffer
4998  *	may be dropped during processing for congestion control or by the
4999  *	protocol layers.
5000  *
5001  *	return values:
5002  *	NET_RX_SUCCESS	(no congestion)
5003  *	NET_RX_DROP     (packet was dropped)
5004  *
5005  */
5006 
5007 int netif_rx(struct sk_buff *skb)
5008 {
5009 	int ret;
5010 
5011 	trace_netif_rx_entry(skb);
5012 
5013 	ret = netif_rx_internal(skb);
5014 	trace_netif_rx_exit(ret);
5015 
5016 	return ret;
5017 }
5018 EXPORT_SYMBOL(netif_rx);
5019 
5020 int netif_rx_ni(struct sk_buff *skb)
5021 {
5022 	int err;
5023 
5024 	trace_netif_rx_ni_entry(skb);
5025 
5026 	preempt_disable();
5027 	err = netif_rx_internal(skb);
5028 	if (local_softirq_pending())
5029 		do_softirq();
5030 	preempt_enable();
5031 	trace_netif_rx_ni_exit(err);
5032 
5033 	return err;
5034 }
5035 EXPORT_SYMBOL(netif_rx_ni);
5036 
5037 int netif_rx_any_context(struct sk_buff *skb)
5038 {
5039 	/*
5040 	 * If invoked from contexts which do not invoke bottom half
5041 	 * processing either at return from interrupt or when softrqs are
5042 	 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
5043 	 * directly.
5044 	 */
5045 	if (in_interrupt())
5046 		return netif_rx(skb);
5047 	else
5048 		return netif_rx_ni(skb);
5049 }
5050 EXPORT_SYMBOL(netif_rx_any_context);
5051 
5052 static __latent_entropy void net_tx_action(struct softirq_action *h)
5053 {
5054 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5055 
5056 	if (sd->completion_queue) {
5057 		struct sk_buff *clist;
5058 
5059 		local_irq_disable();
5060 		clist = sd->completion_queue;
5061 		sd->completion_queue = NULL;
5062 		local_irq_enable();
5063 
5064 		while (clist) {
5065 			struct sk_buff *skb = clist;
5066 
5067 			clist = clist->next;
5068 
5069 			WARN_ON(refcount_read(&skb->users));
5070 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5071 				trace_consume_skb(skb);
5072 			else
5073 				trace_kfree_skb(skb, net_tx_action);
5074 
5075 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5076 				__kfree_skb(skb);
5077 			else
5078 				__kfree_skb_defer(skb);
5079 		}
5080 	}
5081 
5082 	if (sd->output_queue) {
5083 		struct Qdisc *head;
5084 
5085 		local_irq_disable();
5086 		head = sd->output_queue;
5087 		sd->output_queue = NULL;
5088 		sd->output_queue_tailp = &sd->output_queue;
5089 		local_irq_enable();
5090 
5091 		rcu_read_lock();
5092 
5093 		while (head) {
5094 			struct Qdisc *q = head;
5095 			spinlock_t *root_lock = NULL;
5096 
5097 			head = head->next_sched;
5098 
5099 			/* We need to make sure head->next_sched is read
5100 			 * before clearing __QDISC_STATE_SCHED
5101 			 */
5102 			smp_mb__before_atomic();
5103 
5104 			if (!(q->flags & TCQ_F_NOLOCK)) {
5105 				root_lock = qdisc_lock(q);
5106 				spin_lock(root_lock);
5107 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5108 						     &q->state))) {
5109 				/* There is a synchronize_net() between
5110 				 * STATE_DEACTIVATED flag being set and
5111 				 * qdisc_reset()/some_qdisc_is_busy() in
5112 				 * dev_deactivate(), so we can safely bail out
5113 				 * early here to avoid data race between
5114 				 * qdisc_deactivate() and some_qdisc_is_busy()
5115 				 * for lockless qdisc.
5116 				 */
5117 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5118 				continue;
5119 			}
5120 
5121 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5122 			qdisc_run(q);
5123 			if (root_lock)
5124 				spin_unlock(root_lock);
5125 		}
5126 
5127 		rcu_read_unlock();
5128 	}
5129 
5130 	xfrm_dev_backlog(sd);
5131 }
5132 
5133 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5134 /* This hook is defined here for ATM LANE */
5135 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5136 			     unsigned char *addr) __read_mostly;
5137 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5138 #endif
5139 
5140 static inline struct sk_buff *
5141 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5142 		   struct net_device *orig_dev, bool *another)
5143 {
5144 #ifdef CONFIG_NET_CLS_ACT
5145 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5146 	struct tcf_result cl_res;
5147 
5148 	/* If there's at least one ingress present somewhere (so
5149 	 * we get here via enabled static key), remaining devices
5150 	 * that are not configured with an ingress qdisc will bail
5151 	 * out here.
5152 	 */
5153 	if (!miniq)
5154 		return skb;
5155 
5156 	if (*pt_prev) {
5157 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5158 		*pt_prev = NULL;
5159 	}
5160 
5161 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5162 	qdisc_skb_cb(skb)->mru = 0;
5163 	qdisc_skb_cb(skb)->post_ct = false;
5164 	skb->tc_at_ingress = 1;
5165 	mini_qdisc_bstats_cpu_update(miniq, skb);
5166 
5167 	switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
5168 				     &cl_res, false)) {
5169 	case TC_ACT_OK:
5170 	case TC_ACT_RECLASSIFY:
5171 		skb->tc_index = TC_H_MIN(cl_res.classid);
5172 		break;
5173 	case TC_ACT_SHOT:
5174 		mini_qdisc_qstats_cpu_drop(miniq);
5175 		kfree_skb(skb);
5176 		return NULL;
5177 	case TC_ACT_STOLEN:
5178 	case TC_ACT_QUEUED:
5179 	case TC_ACT_TRAP:
5180 		consume_skb(skb);
5181 		return NULL;
5182 	case TC_ACT_REDIRECT:
5183 		/* skb_mac_header check was done by cls/act_bpf, so
5184 		 * we can safely push the L2 header back before
5185 		 * redirecting to another netdev
5186 		 */
5187 		__skb_push(skb, skb->mac_len);
5188 		if (skb_do_redirect(skb) == -EAGAIN) {
5189 			__skb_pull(skb, skb->mac_len);
5190 			*another = true;
5191 			break;
5192 		}
5193 		return NULL;
5194 	case TC_ACT_CONSUMED:
5195 		return NULL;
5196 	default:
5197 		break;
5198 	}
5199 #endif /* CONFIG_NET_CLS_ACT */
5200 	return skb;
5201 }
5202 
5203 /**
5204  *	netdev_is_rx_handler_busy - check if receive handler is registered
5205  *	@dev: device to check
5206  *
5207  *	Check if a receive handler is already registered for a given device.
5208  *	Return true if there one.
5209  *
5210  *	The caller must hold the rtnl_mutex.
5211  */
5212 bool netdev_is_rx_handler_busy(struct net_device *dev)
5213 {
5214 	ASSERT_RTNL();
5215 	return dev && rtnl_dereference(dev->rx_handler);
5216 }
5217 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5218 
5219 /**
5220  *	netdev_rx_handler_register - register receive handler
5221  *	@dev: device to register a handler for
5222  *	@rx_handler: receive handler to register
5223  *	@rx_handler_data: data pointer that is used by rx handler
5224  *
5225  *	Register a receive handler for a device. This handler will then be
5226  *	called from __netif_receive_skb. A negative errno code is returned
5227  *	on a failure.
5228  *
5229  *	The caller must hold the rtnl_mutex.
5230  *
5231  *	For a general description of rx_handler, see enum rx_handler_result.
5232  */
5233 int netdev_rx_handler_register(struct net_device *dev,
5234 			       rx_handler_func_t *rx_handler,
5235 			       void *rx_handler_data)
5236 {
5237 	if (netdev_is_rx_handler_busy(dev))
5238 		return -EBUSY;
5239 
5240 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5241 		return -EINVAL;
5242 
5243 	/* Note: rx_handler_data must be set before rx_handler */
5244 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5245 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5246 
5247 	return 0;
5248 }
5249 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5250 
5251 /**
5252  *	netdev_rx_handler_unregister - unregister receive handler
5253  *	@dev: device to unregister a handler from
5254  *
5255  *	Unregister a receive handler from a device.
5256  *
5257  *	The caller must hold the rtnl_mutex.
5258  */
5259 void netdev_rx_handler_unregister(struct net_device *dev)
5260 {
5261 
5262 	ASSERT_RTNL();
5263 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5264 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5265 	 * section has a guarantee to see a non NULL rx_handler_data
5266 	 * as well.
5267 	 */
5268 	synchronize_net();
5269 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5270 }
5271 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5272 
5273 /*
5274  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5275  * the special handling of PFMEMALLOC skbs.
5276  */
5277 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5278 {
5279 	switch (skb->protocol) {
5280 	case htons(ETH_P_ARP):
5281 	case htons(ETH_P_IP):
5282 	case htons(ETH_P_IPV6):
5283 	case htons(ETH_P_8021Q):
5284 	case htons(ETH_P_8021AD):
5285 		return true;
5286 	default:
5287 		return false;
5288 	}
5289 }
5290 
5291 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5292 			     int *ret, struct net_device *orig_dev)
5293 {
5294 	if (nf_hook_ingress_active(skb)) {
5295 		int ingress_retval;
5296 
5297 		if (*pt_prev) {
5298 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5299 			*pt_prev = NULL;
5300 		}
5301 
5302 		rcu_read_lock();
5303 		ingress_retval = nf_hook_ingress(skb);
5304 		rcu_read_unlock();
5305 		return ingress_retval;
5306 	}
5307 	return 0;
5308 }
5309 
5310 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5311 				    struct packet_type **ppt_prev)
5312 {
5313 	struct packet_type *ptype, *pt_prev;
5314 	rx_handler_func_t *rx_handler;
5315 	struct sk_buff *skb = *pskb;
5316 	struct net_device *orig_dev;
5317 	bool deliver_exact = false;
5318 	int ret = NET_RX_DROP;
5319 	__be16 type;
5320 
5321 	net_timestamp_check(!netdev_tstamp_prequeue, skb);
5322 
5323 	trace_netif_receive_skb(skb);
5324 
5325 	orig_dev = skb->dev;
5326 
5327 	skb_reset_network_header(skb);
5328 	if (!skb_transport_header_was_set(skb))
5329 		skb_reset_transport_header(skb);
5330 	skb_reset_mac_len(skb);
5331 
5332 	pt_prev = NULL;
5333 
5334 another_round:
5335 	skb->skb_iif = skb->dev->ifindex;
5336 
5337 	__this_cpu_inc(softnet_data.processed);
5338 
5339 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5340 		int ret2;
5341 
5342 		migrate_disable();
5343 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5344 		migrate_enable();
5345 
5346 		if (ret2 != XDP_PASS) {
5347 			ret = NET_RX_DROP;
5348 			goto out;
5349 		}
5350 	}
5351 
5352 	if (eth_type_vlan(skb->protocol)) {
5353 		skb = skb_vlan_untag(skb);
5354 		if (unlikely(!skb))
5355 			goto out;
5356 	}
5357 
5358 	if (skb_skip_tc_classify(skb))
5359 		goto skip_classify;
5360 
5361 	if (pfmemalloc)
5362 		goto skip_taps;
5363 
5364 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5365 		if (pt_prev)
5366 			ret = deliver_skb(skb, pt_prev, orig_dev);
5367 		pt_prev = ptype;
5368 	}
5369 
5370 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5371 		if (pt_prev)
5372 			ret = deliver_skb(skb, pt_prev, orig_dev);
5373 		pt_prev = ptype;
5374 	}
5375 
5376 skip_taps:
5377 #ifdef CONFIG_NET_INGRESS
5378 	if (static_branch_unlikely(&ingress_needed_key)) {
5379 		bool another = false;
5380 
5381 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5382 					 &another);
5383 		if (another)
5384 			goto another_round;
5385 		if (!skb)
5386 			goto out;
5387 
5388 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5389 			goto out;
5390 	}
5391 #endif
5392 	skb_reset_redirect(skb);
5393 skip_classify:
5394 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5395 		goto drop;
5396 
5397 	if (skb_vlan_tag_present(skb)) {
5398 		if (pt_prev) {
5399 			ret = deliver_skb(skb, pt_prev, orig_dev);
5400 			pt_prev = NULL;
5401 		}
5402 		if (vlan_do_receive(&skb))
5403 			goto another_round;
5404 		else if (unlikely(!skb))
5405 			goto out;
5406 	}
5407 
5408 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5409 	if (rx_handler) {
5410 		if (pt_prev) {
5411 			ret = deliver_skb(skb, pt_prev, orig_dev);
5412 			pt_prev = NULL;
5413 		}
5414 		switch (rx_handler(&skb)) {
5415 		case RX_HANDLER_CONSUMED:
5416 			ret = NET_RX_SUCCESS;
5417 			goto out;
5418 		case RX_HANDLER_ANOTHER:
5419 			goto another_round;
5420 		case RX_HANDLER_EXACT:
5421 			deliver_exact = true;
5422 			break;
5423 		case RX_HANDLER_PASS:
5424 			break;
5425 		default:
5426 			BUG();
5427 		}
5428 	}
5429 
5430 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5431 check_vlan_id:
5432 		if (skb_vlan_tag_get_id(skb)) {
5433 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5434 			 * find vlan device.
5435 			 */
5436 			skb->pkt_type = PACKET_OTHERHOST;
5437 		} else if (eth_type_vlan(skb->protocol)) {
5438 			/* Outer header is 802.1P with vlan 0, inner header is
5439 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5440 			 * not find vlan dev for vlan id 0.
5441 			 */
5442 			__vlan_hwaccel_clear_tag(skb);
5443 			skb = skb_vlan_untag(skb);
5444 			if (unlikely(!skb))
5445 				goto out;
5446 			if (vlan_do_receive(&skb))
5447 				/* After stripping off 802.1P header with vlan 0
5448 				 * vlan dev is found for inner header.
5449 				 */
5450 				goto another_round;
5451 			else if (unlikely(!skb))
5452 				goto out;
5453 			else
5454 				/* We have stripped outer 802.1P vlan 0 header.
5455 				 * But could not find vlan dev.
5456 				 * check again for vlan id to set OTHERHOST.
5457 				 */
5458 				goto check_vlan_id;
5459 		}
5460 		/* Note: we might in the future use prio bits
5461 		 * and set skb->priority like in vlan_do_receive()
5462 		 * For the time being, just ignore Priority Code Point
5463 		 */
5464 		__vlan_hwaccel_clear_tag(skb);
5465 	}
5466 
5467 	type = skb->protocol;
5468 
5469 	/* deliver only exact match when indicated */
5470 	if (likely(!deliver_exact)) {
5471 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5472 				       &ptype_base[ntohs(type) &
5473 						   PTYPE_HASH_MASK]);
5474 	}
5475 
5476 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5477 			       &orig_dev->ptype_specific);
5478 
5479 	if (unlikely(skb->dev != orig_dev)) {
5480 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5481 				       &skb->dev->ptype_specific);
5482 	}
5483 
5484 	if (pt_prev) {
5485 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5486 			goto drop;
5487 		*ppt_prev = pt_prev;
5488 	} else {
5489 drop:
5490 		if (!deliver_exact)
5491 			atomic_long_inc(&skb->dev->rx_dropped);
5492 		else
5493 			atomic_long_inc(&skb->dev->rx_nohandler);
5494 		kfree_skb(skb);
5495 		/* Jamal, now you will not able to escape explaining
5496 		 * me how you were going to use this. :-)
5497 		 */
5498 		ret = NET_RX_DROP;
5499 	}
5500 
5501 out:
5502 	/* The invariant here is that if *ppt_prev is not NULL
5503 	 * then skb should also be non-NULL.
5504 	 *
5505 	 * Apparently *ppt_prev assignment above holds this invariant due to
5506 	 * skb dereferencing near it.
5507 	 */
5508 	*pskb = skb;
5509 	return ret;
5510 }
5511 
5512 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5513 {
5514 	struct net_device *orig_dev = skb->dev;
5515 	struct packet_type *pt_prev = NULL;
5516 	int ret;
5517 
5518 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5519 	if (pt_prev)
5520 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5521 					 skb->dev, pt_prev, orig_dev);
5522 	return ret;
5523 }
5524 
5525 /**
5526  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5527  *	@skb: buffer to process
5528  *
5529  *	More direct receive version of netif_receive_skb().  It should
5530  *	only be used by callers that have a need to skip RPS and Generic XDP.
5531  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5532  *
5533  *	This function may only be called from softirq context and interrupts
5534  *	should be enabled.
5535  *
5536  *	Return values (usually ignored):
5537  *	NET_RX_SUCCESS: no congestion
5538  *	NET_RX_DROP: packet was dropped
5539  */
5540 int netif_receive_skb_core(struct sk_buff *skb)
5541 {
5542 	int ret;
5543 
5544 	rcu_read_lock();
5545 	ret = __netif_receive_skb_one_core(skb, false);
5546 	rcu_read_unlock();
5547 
5548 	return ret;
5549 }
5550 EXPORT_SYMBOL(netif_receive_skb_core);
5551 
5552 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5553 						  struct packet_type *pt_prev,
5554 						  struct net_device *orig_dev)
5555 {
5556 	struct sk_buff *skb, *next;
5557 
5558 	if (!pt_prev)
5559 		return;
5560 	if (list_empty(head))
5561 		return;
5562 	if (pt_prev->list_func != NULL)
5563 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5564 				   ip_list_rcv, head, pt_prev, orig_dev);
5565 	else
5566 		list_for_each_entry_safe(skb, next, head, list) {
5567 			skb_list_del_init(skb);
5568 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5569 		}
5570 }
5571 
5572 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5573 {
5574 	/* Fast-path assumptions:
5575 	 * - There is no RX handler.
5576 	 * - Only one packet_type matches.
5577 	 * If either of these fails, we will end up doing some per-packet
5578 	 * processing in-line, then handling the 'last ptype' for the whole
5579 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5580 	 * because the 'last ptype' must be constant across the sublist, and all
5581 	 * other ptypes are handled per-packet.
5582 	 */
5583 	/* Current (common) ptype of sublist */
5584 	struct packet_type *pt_curr = NULL;
5585 	/* Current (common) orig_dev of sublist */
5586 	struct net_device *od_curr = NULL;
5587 	struct list_head sublist;
5588 	struct sk_buff *skb, *next;
5589 
5590 	INIT_LIST_HEAD(&sublist);
5591 	list_for_each_entry_safe(skb, next, head, list) {
5592 		struct net_device *orig_dev = skb->dev;
5593 		struct packet_type *pt_prev = NULL;
5594 
5595 		skb_list_del_init(skb);
5596 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5597 		if (!pt_prev)
5598 			continue;
5599 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5600 			/* dispatch old sublist */
5601 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5602 			/* start new sublist */
5603 			INIT_LIST_HEAD(&sublist);
5604 			pt_curr = pt_prev;
5605 			od_curr = orig_dev;
5606 		}
5607 		list_add_tail(&skb->list, &sublist);
5608 	}
5609 
5610 	/* dispatch final sublist */
5611 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5612 }
5613 
5614 static int __netif_receive_skb(struct sk_buff *skb)
5615 {
5616 	int ret;
5617 
5618 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5619 		unsigned int noreclaim_flag;
5620 
5621 		/*
5622 		 * PFMEMALLOC skbs are special, they should
5623 		 * - be delivered to SOCK_MEMALLOC sockets only
5624 		 * - stay away from userspace
5625 		 * - have bounded memory usage
5626 		 *
5627 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5628 		 * context down to all allocation sites.
5629 		 */
5630 		noreclaim_flag = memalloc_noreclaim_save();
5631 		ret = __netif_receive_skb_one_core(skb, true);
5632 		memalloc_noreclaim_restore(noreclaim_flag);
5633 	} else
5634 		ret = __netif_receive_skb_one_core(skb, false);
5635 
5636 	return ret;
5637 }
5638 
5639 static void __netif_receive_skb_list(struct list_head *head)
5640 {
5641 	unsigned long noreclaim_flag = 0;
5642 	struct sk_buff *skb, *next;
5643 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5644 
5645 	list_for_each_entry_safe(skb, next, head, list) {
5646 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5647 			struct list_head sublist;
5648 
5649 			/* Handle the previous sublist */
5650 			list_cut_before(&sublist, head, &skb->list);
5651 			if (!list_empty(&sublist))
5652 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5653 			pfmemalloc = !pfmemalloc;
5654 			/* See comments in __netif_receive_skb */
5655 			if (pfmemalloc)
5656 				noreclaim_flag = memalloc_noreclaim_save();
5657 			else
5658 				memalloc_noreclaim_restore(noreclaim_flag);
5659 		}
5660 	}
5661 	/* Handle the remaining sublist */
5662 	if (!list_empty(head))
5663 		__netif_receive_skb_list_core(head, pfmemalloc);
5664 	/* Restore pflags */
5665 	if (pfmemalloc)
5666 		memalloc_noreclaim_restore(noreclaim_flag);
5667 }
5668 
5669 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5670 {
5671 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5672 	struct bpf_prog *new = xdp->prog;
5673 	int ret = 0;
5674 
5675 	switch (xdp->command) {
5676 	case XDP_SETUP_PROG:
5677 		rcu_assign_pointer(dev->xdp_prog, new);
5678 		if (old)
5679 			bpf_prog_put(old);
5680 
5681 		if (old && !new) {
5682 			static_branch_dec(&generic_xdp_needed_key);
5683 		} else if (new && !old) {
5684 			static_branch_inc(&generic_xdp_needed_key);
5685 			dev_disable_lro(dev);
5686 			dev_disable_gro_hw(dev);
5687 		}
5688 		break;
5689 
5690 	default:
5691 		ret = -EINVAL;
5692 		break;
5693 	}
5694 
5695 	return ret;
5696 }
5697 
5698 static int netif_receive_skb_internal(struct sk_buff *skb)
5699 {
5700 	int ret;
5701 
5702 	net_timestamp_check(netdev_tstamp_prequeue, skb);
5703 
5704 	if (skb_defer_rx_timestamp(skb))
5705 		return NET_RX_SUCCESS;
5706 
5707 	rcu_read_lock();
5708 #ifdef CONFIG_RPS
5709 	if (static_branch_unlikely(&rps_needed)) {
5710 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5711 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5712 
5713 		if (cpu >= 0) {
5714 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5715 			rcu_read_unlock();
5716 			return ret;
5717 		}
5718 	}
5719 #endif
5720 	ret = __netif_receive_skb(skb);
5721 	rcu_read_unlock();
5722 	return ret;
5723 }
5724 
5725 static void netif_receive_skb_list_internal(struct list_head *head)
5726 {
5727 	struct sk_buff *skb, *next;
5728 	struct list_head sublist;
5729 
5730 	INIT_LIST_HEAD(&sublist);
5731 	list_for_each_entry_safe(skb, next, head, list) {
5732 		net_timestamp_check(netdev_tstamp_prequeue, skb);
5733 		skb_list_del_init(skb);
5734 		if (!skb_defer_rx_timestamp(skb))
5735 			list_add_tail(&skb->list, &sublist);
5736 	}
5737 	list_splice_init(&sublist, head);
5738 
5739 	rcu_read_lock();
5740 #ifdef CONFIG_RPS
5741 	if (static_branch_unlikely(&rps_needed)) {
5742 		list_for_each_entry_safe(skb, next, head, list) {
5743 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5744 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5745 
5746 			if (cpu >= 0) {
5747 				/* Will be handled, remove from list */
5748 				skb_list_del_init(skb);
5749 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5750 			}
5751 		}
5752 	}
5753 #endif
5754 	__netif_receive_skb_list(head);
5755 	rcu_read_unlock();
5756 }
5757 
5758 /**
5759  *	netif_receive_skb - process receive buffer from network
5760  *	@skb: buffer to process
5761  *
5762  *	netif_receive_skb() is the main receive data processing function.
5763  *	It always succeeds. The buffer may be dropped during processing
5764  *	for congestion control or by the protocol layers.
5765  *
5766  *	This function may only be called from softirq context and interrupts
5767  *	should be enabled.
5768  *
5769  *	Return values (usually ignored):
5770  *	NET_RX_SUCCESS: no congestion
5771  *	NET_RX_DROP: packet was dropped
5772  */
5773 int netif_receive_skb(struct sk_buff *skb)
5774 {
5775 	int ret;
5776 
5777 	trace_netif_receive_skb_entry(skb);
5778 
5779 	ret = netif_receive_skb_internal(skb);
5780 	trace_netif_receive_skb_exit(ret);
5781 
5782 	return ret;
5783 }
5784 EXPORT_SYMBOL(netif_receive_skb);
5785 
5786 /**
5787  *	netif_receive_skb_list - process many receive buffers from network
5788  *	@head: list of skbs to process.
5789  *
5790  *	Since return value of netif_receive_skb() is normally ignored, and
5791  *	wouldn't be meaningful for a list, this function returns void.
5792  *
5793  *	This function may only be called from softirq context and interrupts
5794  *	should be enabled.
5795  */
5796 void netif_receive_skb_list(struct list_head *head)
5797 {
5798 	struct sk_buff *skb;
5799 
5800 	if (list_empty(head))
5801 		return;
5802 	if (trace_netif_receive_skb_list_entry_enabled()) {
5803 		list_for_each_entry(skb, head, list)
5804 			trace_netif_receive_skb_list_entry(skb);
5805 	}
5806 	netif_receive_skb_list_internal(head);
5807 	trace_netif_receive_skb_list_exit(0);
5808 }
5809 EXPORT_SYMBOL(netif_receive_skb_list);
5810 
5811 static DEFINE_PER_CPU(struct work_struct, flush_works);
5812 
5813 /* Network device is going away, flush any packets still pending */
5814 static void flush_backlog(struct work_struct *work)
5815 {
5816 	struct sk_buff *skb, *tmp;
5817 	struct softnet_data *sd;
5818 
5819 	local_bh_disable();
5820 	sd = this_cpu_ptr(&softnet_data);
5821 
5822 	local_irq_disable();
5823 	rps_lock(sd);
5824 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5825 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5826 			__skb_unlink(skb, &sd->input_pkt_queue);
5827 			dev_kfree_skb_irq(skb);
5828 			input_queue_head_incr(sd);
5829 		}
5830 	}
5831 	rps_unlock(sd);
5832 	local_irq_enable();
5833 
5834 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5835 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5836 			__skb_unlink(skb, &sd->process_queue);
5837 			kfree_skb(skb);
5838 			input_queue_head_incr(sd);
5839 		}
5840 	}
5841 	local_bh_enable();
5842 }
5843 
5844 static bool flush_required(int cpu)
5845 {
5846 #if IS_ENABLED(CONFIG_RPS)
5847 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5848 	bool do_flush;
5849 
5850 	local_irq_disable();
5851 	rps_lock(sd);
5852 
5853 	/* as insertion into process_queue happens with the rps lock held,
5854 	 * process_queue access may race only with dequeue
5855 	 */
5856 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5857 		   !skb_queue_empty_lockless(&sd->process_queue);
5858 	rps_unlock(sd);
5859 	local_irq_enable();
5860 
5861 	return do_flush;
5862 #endif
5863 	/* without RPS we can't safely check input_pkt_queue: during a
5864 	 * concurrent remote skb_queue_splice() we can detect as empty both
5865 	 * input_pkt_queue and process_queue even if the latter could end-up
5866 	 * containing a lot of packets.
5867 	 */
5868 	return true;
5869 }
5870 
5871 static void flush_all_backlogs(void)
5872 {
5873 	static cpumask_t flush_cpus;
5874 	unsigned int cpu;
5875 
5876 	/* since we are under rtnl lock protection we can use static data
5877 	 * for the cpumask and avoid allocating on stack the possibly
5878 	 * large mask
5879 	 */
5880 	ASSERT_RTNL();
5881 
5882 	get_online_cpus();
5883 
5884 	cpumask_clear(&flush_cpus);
5885 	for_each_online_cpu(cpu) {
5886 		if (flush_required(cpu)) {
5887 			queue_work_on(cpu, system_highpri_wq,
5888 				      per_cpu_ptr(&flush_works, cpu));
5889 			cpumask_set_cpu(cpu, &flush_cpus);
5890 		}
5891 	}
5892 
5893 	/* we can have in flight packet[s] on the cpus we are not flushing,
5894 	 * synchronize_net() in unregister_netdevice_many() will take care of
5895 	 * them
5896 	 */
5897 	for_each_cpu(cpu, &flush_cpus)
5898 		flush_work(per_cpu_ptr(&flush_works, cpu));
5899 
5900 	put_online_cpus();
5901 }
5902 
5903 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5904 static void gro_normal_list(struct napi_struct *napi)
5905 {
5906 	if (!napi->rx_count)
5907 		return;
5908 	netif_receive_skb_list_internal(&napi->rx_list);
5909 	INIT_LIST_HEAD(&napi->rx_list);
5910 	napi->rx_count = 0;
5911 }
5912 
5913 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5914  * pass the whole batch up to the stack.
5915  */
5916 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5917 {
5918 	list_add_tail(&skb->list, &napi->rx_list);
5919 	napi->rx_count += segs;
5920 	if (napi->rx_count >= gro_normal_batch)
5921 		gro_normal_list(napi);
5922 }
5923 
5924 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5925 {
5926 	struct packet_offload *ptype;
5927 	__be16 type = skb->protocol;
5928 	struct list_head *head = &offload_base;
5929 	int err = -ENOENT;
5930 
5931 	BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5932 
5933 	if (NAPI_GRO_CB(skb)->count == 1) {
5934 		skb_shinfo(skb)->gso_size = 0;
5935 		goto out;
5936 	}
5937 
5938 	rcu_read_lock();
5939 	list_for_each_entry_rcu(ptype, head, list) {
5940 		if (ptype->type != type || !ptype->callbacks.gro_complete)
5941 			continue;
5942 
5943 		err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5944 					 ipv6_gro_complete, inet_gro_complete,
5945 					 skb, 0);
5946 		break;
5947 	}
5948 	rcu_read_unlock();
5949 
5950 	if (err) {
5951 		WARN_ON(&ptype->list == head);
5952 		kfree_skb(skb);
5953 		return NET_RX_SUCCESS;
5954 	}
5955 
5956 out:
5957 	gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5958 	return NET_RX_SUCCESS;
5959 }
5960 
5961 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5962 				   bool flush_old)
5963 {
5964 	struct list_head *head = &napi->gro_hash[index].list;
5965 	struct sk_buff *skb, *p;
5966 
5967 	list_for_each_entry_safe_reverse(skb, p, head, list) {
5968 		if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5969 			return;
5970 		skb_list_del_init(skb);
5971 		napi_gro_complete(napi, skb);
5972 		napi->gro_hash[index].count--;
5973 	}
5974 
5975 	if (!napi->gro_hash[index].count)
5976 		__clear_bit(index, &napi->gro_bitmask);
5977 }
5978 
5979 /* napi->gro_hash[].list contains packets ordered by age.
5980  * youngest packets at the head of it.
5981  * Complete skbs in reverse order to reduce latencies.
5982  */
5983 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5984 {
5985 	unsigned long bitmask = napi->gro_bitmask;
5986 	unsigned int i, base = ~0U;
5987 
5988 	while ((i = ffs(bitmask)) != 0) {
5989 		bitmask >>= i;
5990 		base += i;
5991 		__napi_gro_flush_chain(napi, base, flush_old);
5992 	}
5993 }
5994 EXPORT_SYMBOL(napi_gro_flush);
5995 
5996 static void gro_list_prepare(const struct list_head *head,
5997 			     const struct sk_buff *skb)
5998 {
5999 	unsigned int maclen = skb->dev->hard_header_len;
6000 	u32 hash = skb_get_hash_raw(skb);
6001 	struct sk_buff *p;
6002 
6003 	list_for_each_entry(p, head, list) {
6004 		unsigned long diffs;
6005 
6006 		NAPI_GRO_CB(p)->flush = 0;
6007 
6008 		if (hash != skb_get_hash_raw(p)) {
6009 			NAPI_GRO_CB(p)->same_flow = 0;
6010 			continue;
6011 		}
6012 
6013 		diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
6014 		diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
6015 		if (skb_vlan_tag_present(p))
6016 			diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
6017 		diffs |= skb_metadata_differs(p, skb);
6018 		if (maclen == ETH_HLEN)
6019 			diffs |= compare_ether_header(skb_mac_header(p),
6020 						      skb_mac_header(skb));
6021 		else if (!diffs)
6022 			diffs = memcmp(skb_mac_header(p),
6023 				       skb_mac_header(skb),
6024 				       maclen);
6025 
6026 		/* in most common scenarions _state is 0
6027 		 * otherwise we are already on some slower paths
6028 		 * either skip all the infrequent tests altogether or
6029 		 * avoid trying too hard to skip each of them individually
6030 		 */
6031 		if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
6032 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
6033 			struct tc_skb_ext *skb_ext;
6034 			struct tc_skb_ext *p_ext;
6035 #endif
6036 
6037 			diffs |= p->sk != skb->sk;
6038 			diffs |= skb_metadata_dst_cmp(p, skb);
6039 			diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
6040 
6041 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
6042 			skb_ext = skb_ext_find(skb, TC_SKB_EXT);
6043 			p_ext = skb_ext_find(p, TC_SKB_EXT);
6044 
6045 			diffs |= (!!p_ext) ^ (!!skb_ext);
6046 			if (!diffs && unlikely(skb_ext))
6047 				diffs |= p_ext->chain ^ skb_ext->chain;
6048 #endif
6049 		}
6050 
6051 		NAPI_GRO_CB(p)->same_flow = !diffs;
6052 	}
6053 }
6054 
6055 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
6056 {
6057 	const struct skb_shared_info *pinfo = skb_shinfo(skb);
6058 	const skb_frag_t *frag0 = &pinfo->frags[0];
6059 
6060 	NAPI_GRO_CB(skb)->data_offset = 0;
6061 	NAPI_GRO_CB(skb)->frag0 = NULL;
6062 	NAPI_GRO_CB(skb)->frag0_len = 0;
6063 
6064 	if (!skb_headlen(skb) && pinfo->nr_frags &&
6065 	    !PageHighMem(skb_frag_page(frag0)) &&
6066 	    (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
6067 		NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
6068 		NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
6069 						    skb_frag_size(frag0),
6070 						    skb->end - skb->tail);
6071 	}
6072 }
6073 
6074 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
6075 {
6076 	struct skb_shared_info *pinfo = skb_shinfo(skb);
6077 
6078 	BUG_ON(skb->end - skb->tail < grow);
6079 
6080 	memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6081 
6082 	skb->data_len -= grow;
6083 	skb->tail += grow;
6084 
6085 	skb_frag_off_add(&pinfo->frags[0], grow);
6086 	skb_frag_size_sub(&pinfo->frags[0], grow);
6087 
6088 	if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6089 		skb_frag_unref(skb, 0);
6090 		memmove(pinfo->frags, pinfo->frags + 1,
6091 			--pinfo->nr_frags * sizeof(pinfo->frags[0]));
6092 	}
6093 }
6094 
6095 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6096 {
6097 	struct sk_buff *oldest;
6098 
6099 	oldest = list_last_entry(head, struct sk_buff, list);
6100 
6101 	/* We are called with head length >= MAX_GRO_SKBS, so this is
6102 	 * impossible.
6103 	 */
6104 	if (WARN_ON_ONCE(!oldest))
6105 		return;
6106 
6107 	/* Do not adjust napi->gro_hash[].count, caller is adding a new
6108 	 * SKB to the chain.
6109 	 */
6110 	skb_list_del_init(oldest);
6111 	napi_gro_complete(napi, oldest);
6112 }
6113 
6114 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6115 {
6116 	u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6117 	struct gro_list *gro_list = &napi->gro_hash[bucket];
6118 	struct list_head *head = &offload_base;
6119 	struct packet_offload *ptype;
6120 	__be16 type = skb->protocol;
6121 	struct sk_buff *pp = NULL;
6122 	enum gro_result ret;
6123 	int same_flow;
6124 	int grow;
6125 
6126 	if (netif_elide_gro(skb->dev))
6127 		goto normal;
6128 
6129 	gro_list_prepare(&gro_list->list, skb);
6130 
6131 	rcu_read_lock();
6132 	list_for_each_entry_rcu(ptype, head, list) {
6133 		if (ptype->type != type || !ptype->callbacks.gro_receive)
6134 			continue;
6135 
6136 		skb_set_network_header(skb, skb_gro_offset(skb));
6137 		skb_reset_mac_len(skb);
6138 		NAPI_GRO_CB(skb)->same_flow = 0;
6139 		NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6140 		NAPI_GRO_CB(skb)->free = 0;
6141 		NAPI_GRO_CB(skb)->encap_mark = 0;
6142 		NAPI_GRO_CB(skb)->recursion_counter = 0;
6143 		NAPI_GRO_CB(skb)->is_fou = 0;
6144 		NAPI_GRO_CB(skb)->is_atomic = 1;
6145 		NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6146 
6147 		/* Setup for GRO checksum validation */
6148 		switch (skb->ip_summed) {
6149 		case CHECKSUM_COMPLETE:
6150 			NAPI_GRO_CB(skb)->csum = skb->csum;
6151 			NAPI_GRO_CB(skb)->csum_valid = 1;
6152 			NAPI_GRO_CB(skb)->csum_cnt = 0;
6153 			break;
6154 		case CHECKSUM_UNNECESSARY:
6155 			NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6156 			NAPI_GRO_CB(skb)->csum_valid = 0;
6157 			break;
6158 		default:
6159 			NAPI_GRO_CB(skb)->csum_cnt = 0;
6160 			NAPI_GRO_CB(skb)->csum_valid = 0;
6161 		}
6162 
6163 		pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6164 					ipv6_gro_receive, inet_gro_receive,
6165 					&gro_list->list, skb);
6166 		break;
6167 	}
6168 	rcu_read_unlock();
6169 
6170 	if (&ptype->list == head)
6171 		goto normal;
6172 
6173 	if (PTR_ERR(pp) == -EINPROGRESS) {
6174 		ret = GRO_CONSUMED;
6175 		goto ok;
6176 	}
6177 
6178 	same_flow = NAPI_GRO_CB(skb)->same_flow;
6179 	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6180 
6181 	if (pp) {
6182 		skb_list_del_init(pp);
6183 		napi_gro_complete(napi, pp);
6184 		gro_list->count--;
6185 	}
6186 
6187 	if (same_flow)
6188 		goto ok;
6189 
6190 	if (NAPI_GRO_CB(skb)->flush)
6191 		goto normal;
6192 
6193 	if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6194 		gro_flush_oldest(napi, &gro_list->list);
6195 	else
6196 		gro_list->count++;
6197 
6198 	NAPI_GRO_CB(skb)->count = 1;
6199 	NAPI_GRO_CB(skb)->age = jiffies;
6200 	NAPI_GRO_CB(skb)->last = skb;
6201 	skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6202 	list_add(&skb->list, &gro_list->list);
6203 	ret = GRO_HELD;
6204 
6205 pull:
6206 	grow = skb_gro_offset(skb) - skb_headlen(skb);
6207 	if (grow > 0)
6208 		gro_pull_from_frag0(skb, grow);
6209 ok:
6210 	if (gro_list->count) {
6211 		if (!test_bit(bucket, &napi->gro_bitmask))
6212 			__set_bit(bucket, &napi->gro_bitmask);
6213 	} else if (test_bit(bucket, &napi->gro_bitmask)) {
6214 		__clear_bit(bucket, &napi->gro_bitmask);
6215 	}
6216 
6217 	return ret;
6218 
6219 normal:
6220 	ret = GRO_NORMAL;
6221 	goto pull;
6222 }
6223 
6224 struct packet_offload *gro_find_receive_by_type(__be16 type)
6225 {
6226 	struct list_head *offload_head = &offload_base;
6227 	struct packet_offload *ptype;
6228 
6229 	list_for_each_entry_rcu(ptype, offload_head, list) {
6230 		if (ptype->type != type || !ptype->callbacks.gro_receive)
6231 			continue;
6232 		return ptype;
6233 	}
6234 	return NULL;
6235 }
6236 EXPORT_SYMBOL(gro_find_receive_by_type);
6237 
6238 struct packet_offload *gro_find_complete_by_type(__be16 type)
6239 {
6240 	struct list_head *offload_head = &offload_base;
6241 	struct packet_offload *ptype;
6242 
6243 	list_for_each_entry_rcu(ptype, offload_head, list) {
6244 		if (ptype->type != type || !ptype->callbacks.gro_complete)
6245 			continue;
6246 		return ptype;
6247 	}
6248 	return NULL;
6249 }
6250 EXPORT_SYMBOL(gro_find_complete_by_type);
6251 
6252 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6253 				    struct sk_buff *skb,
6254 				    gro_result_t ret)
6255 {
6256 	switch (ret) {
6257 	case GRO_NORMAL:
6258 		gro_normal_one(napi, skb, 1);
6259 		break;
6260 
6261 	case GRO_MERGED_FREE:
6262 		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6263 			napi_skb_free_stolen_head(skb);
6264 		else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6265 			__kfree_skb(skb);
6266 		else
6267 			__kfree_skb_defer(skb);
6268 		break;
6269 
6270 	case GRO_HELD:
6271 	case GRO_MERGED:
6272 	case GRO_CONSUMED:
6273 		break;
6274 	}
6275 
6276 	return ret;
6277 }
6278 
6279 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6280 {
6281 	gro_result_t ret;
6282 
6283 	skb_mark_napi_id(skb, napi);
6284 	trace_napi_gro_receive_entry(skb);
6285 
6286 	skb_gro_reset_offset(skb, 0);
6287 
6288 	ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6289 	trace_napi_gro_receive_exit(ret);
6290 
6291 	return ret;
6292 }
6293 EXPORT_SYMBOL(napi_gro_receive);
6294 
6295 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6296 {
6297 	if (unlikely(skb->pfmemalloc)) {
6298 		consume_skb(skb);
6299 		return;
6300 	}
6301 	__skb_pull(skb, skb_headlen(skb));
6302 	/* restore the reserve we had after netdev_alloc_skb_ip_align() */
6303 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6304 	__vlan_hwaccel_clear_tag(skb);
6305 	skb->dev = napi->dev;
6306 	skb->skb_iif = 0;
6307 
6308 	/* eth_type_trans() assumes pkt_type is PACKET_HOST */
6309 	skb->pkt_type = PACKET_HOST;
6310 
6311 	skb->encapsulation = 0;
6312 	skb_shinfo(skb)->gso_type = 0;
6313 	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6314 	if (unlikely(skb->slow_gro)) {
6315 		skb_orphan(skb);
6316 		skb_ext_reset(skb);
6317 		nf_reset_ct(skb);
6318 		skb->slow_gro = 0;
6319 	}
6320 
6321 	napi->skb = skb;
6322 }
6323 
6324 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6325 {
6326 	struct sk_buff *skb = napi->skb;
6327 
6328 	if (!skb) {
6329 		skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6330 		if (skb) {
6331 			napi->skb = skb;
6332 			skb_mark_napi_id(skb, napi);
6333 		}
6334 	}
6335 	return skb;
6336 }
6337 EXPORT_SYMBOL(napi_get_frags);
6338 
6339 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6340 				      struct sk_buff *skb,
6341 				      gro_result_t ret)
6342 {
6343 	switch (ret) {
6344 	case GRO_NORMAL:
6345 	case GRO_HELD:
6346 		__skb_push(skb, ETH_HLEN);
6347 		skb->protocol = eth_type_trans(skb, skb->dev);
6348 		if (ret == GRO_NORMAL)
6349 			gro_normal_one(napi, skb, 1);
6350 		break;
6351 
6352 	case GRO_MERGED_FREE:
6353 		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6354 			napi_skb_free_stolen_head(skb);
6355 		else
6356 			napi_reuse_skb(napi, skb);
6357 		break;
6358 
6359 	case GRO_MERGED:
6360 	case GRO_CONSUMED:
6361 		break;
6362 	}
6363 
6364 	return ret;
6365 }
6366 
6367 /* Upper GRO stack assumes network header starts at gro_offset=0
6368  * Drivers could call both napi_gro_frags() and napi_gro_receive()
6369  * We copy ethernet header into skb->data to have a common layout.
6370  */
6371 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6372 {
6373 	struct sk_buff *skb = napi->skb;
6374 	const struct ethhdr *eth;
6375 	unsigned int hlen = sizeof(*eth);
6376 
6377 	napi->skb = NULL;
6378 
6379 	skb_reset_mac_header(skb);
6380 	skb_gro_reset_offset(skb, hlen);
6381 
6382 	if (unlikely(skb_gro_header_hard(skb, hlen))) {
6383 		eth = skb_gro_header_slow(skb, hlen, 0);
6384 		if (unlikely(!eth)) {
6385 			net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6386 					     __func__, napi->dev->name);
6387 			napi_reuse_skb(napi, skb);
6388 			return NULL;
6389 		}
6390 	} else {
6391 		eth = (const struct ethhdr *)skb->data;
6392 		gro_pull_from_frag0(skb, hlen);
6393 		NAPI_GRO_CB(skb)->frag0 += hlen;
6394 		NAPI_GRO_CB(skb)->frag0_len -= hlen;
6395 	}
6396 	__skb_pull(skb, hlen);
6397 
6398 	/*
6399 	 * This works because the only protocols we care about don't require
6400 	 * special handling.
6401 	 * We'll fix it up properly in napi_frags_finish()
6402 	 */
6403 	skb->protocol = eth->h_proto;
6404 
6405 	return skb;
6406 }
6407 
6408 gro_result_t napi_gro_frags(struct napi_struct *napi)
6409 {
6410 	gro_result_t ret;
6411 	struct sk_buff *skb = napi_frags_skb(napi);
6412 
6413 	trace_napi_gro_frags_entry(skb);
6414 
6415 	ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6416 	trace_napi_gro_frags_exit(ret);
6417 
6418 	return ret;
6419 }
6420 EXPORT_SYMBOL(napi_gro_frags);
6421 
6422 /* Compute the checksum from gro_offset and return the folded value
6423  * after adding in any pseudo checksum.
6424  */
6425 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6426 {
6427 	__wsum wsum;
6428 	__sum16 sum;
6429 
6430 	wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6431 
6432 	/* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6433 	sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6434 	/* See comments in __skb_checksum_complete(). */
6435 	if (likely(!sum)) {
6436 		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6437 		    !skb->csum_complete_sw)
6438 			netdev_rx_csum_fault(skb->dev, skb);
6439 	}
6440 
6441 	NAPI_GRO_CB(skb)->csum = wsum;
6442 	NAPI_GRO_CB(skb)->csum_valid = 1;
6443 
6444 	return sum;
6445 }
6446 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6447 
6448 static void net_rps_send_ipi(struct softnet_data *remsd)
6449 {
6450 #ifdef CONFIG_RPS
6451 	while (remsd) {
6452 		struct softnet_data *next = remsd->rps_ipi_next;
6453 
6454 		if (cpu_online(remsd->cpu))
6455 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
6456 		remsd = next;
6457 	}
6458 #endif
6459 }
6460 
6461 /*
6462  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6463  * Note: called with local irq disabled, but exits with local irq enabled.
6464  */
6465 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6466 {
6467 #ifdef CONFIG_RPS
6468 	struct softnet_data *remsd = sd->rps_ipi_list;
6469 
6470 	if (remsd) {
6471 		sd->rps_ipi_list = NULL;
6472 
6473 		local_irq_enable();
6474 
6475 		/* Send pending IPI's to kick RPS processing on remote cpus. */
6476 		net_rps_send_ipi(remsd);
6477 	} else
6478 #endif
6479 		local_irq_enable();
6480 }
6481 
6482 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6483 {
6484 #ifdef CONFIG_RPS
6485 	return sd->rps_ipi_list != NULL;
6486 #else
6487 	return false;
6488 #endif
6489 }
6490 
6491 static int process_backlog(struct napi_struct *napi, int quota)
6492 {
6493 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6494 	bool again = true;
6495 	int work = 0;
6496 
6497 	/* Check if we have pending ipi, its better to send them now,
6498 	 * not waiting net_rx_action() end.
6499 	 */
6500 	if (sd_has_rps_ipi_waiting(sd)) {
6501 		local_irq_disable();
6502 		net_rps_action_and_irq_enable(sd);
6503 	}
6504 
6505 	napi->weight = dev_rx_weight;
6506 	while (again) {
6507 		struct sk_buff *skb;
6508 
6509 		while ((skb = __skb_dequeue(&sd->process_queue))) {
6510 			rcu_read_lock();
6511 			__netif_receive_skb(skb);
6512 			rcu_read_unlock();
6513 			input_queue_head_incr(sd);
6514 			if (++work >= quota)
6515 				return work;
6516 
6517 		}
6518 
6519 		local_irq_disable();
6520 		rps_lock(sd);
6521 		if (skb_queue_empty(&sd->input_pkt_queue)) {
6522 			/*
6523 			 * Inline a custom version of __napi_complete().
6524 			 * only current cpu owns and manipulates this napi,
6525 			 * and NAPI_STATE_SCHED is the only possible flag set
6526 			 * on backlog.
6527 			 * We can use a plain write instead of clear_bit(),
6528 			 * and we dont need an smp_mb() memory barrier.
6529 			 */
6530 			napi->state = 0;
6531 			again = false;
6532 		} else {
6533 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
6534 						   &sd->process_queue);
6535 		}
6536 		rps_unlock(sd);
6537 		local_irq_enable();
6538 	}
6539 
6540 	return work;
6541 }
6542 
6543 /**
6544  * __napi_schedule - schedule for receive
6545  * @n: entry to schedule
6546  *
6547  * The entry's receive function will be scheduled to run.
6548  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6549  */
6550 void __napi_schedule(struct napi_struct *n)
6551 {
6552 	unsigned long flags;
6553 
6554 	local_irq_save(flags);
6555 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
6556 	local_irq_restore(flags);
6557 }
6558 EXPORT_SYMBOL(__napi_schedule);
6559 
6560 /**
6561  *	napi_schedule_prep - check if napi can be scheduled
6562  *	@n: napi context
6563  *
6564  * Test if NAPI routine is already running, and if not mark
6565  * it as running.  This is used as a condition variable to
6566  * insure only one NAPI poll instance runs.  We also make
6567  * sure there is no pending NAPI disable.
6568  */
6569 bool napi_schedule_prep(struct napi_struct *n)
6570 {
6571 	unsigned long val, new;
6572 
6573 	do {
6574 		val = READ_ONCE(n->state);
6575 		if (unlikely(val & NAPIF_STATE_DISABLE))
6576 			return false;
6577 		new = val | NAPIF_STATE_SCHED;
6578 
6579 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
6580 		 * This was suggested by Alexander Duyck, as compiler
6581 		 * emits better code than :
6582 		 * if (val & NAPIF_STATE_SCHED)
6583 		 *     new |= NAPIF_STATE_MISSED;
6584 		 */
6585 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6586 						   NAPIF_STATE_MISSED;
6587 	} while (cmpxchg(&n->state, val, new) != val);
6588 
6589 	return !(val & NAPIF_STATE_SCHED);
6590 }
6591 EXPORT_SYMBOL(napi_schedule_prep);
6592 
6593 /**
6594  * __napi_schedule_irqoff - schedule for receive
6595  * @n: entry to schedule
6596  *
6597  * Variant of __napi_schedule() assuming hard irqs are masked.
6598  *
6599  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6600  * because the interrupt disabled assumption might not be true
6601  * due to force-threaded interrupts and spinlock substitution.
6602  */
6603 void __napi_schedule_irqoff(struct napi_struct *n)
6604 {
6605 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6606 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6607 	else
6608 		__napi_schedule(n);
6609 }
6610 EXPORT_SYMBOL(__napi_schedule_irqoff);
6611 
6612 bool napi_complete_done(struct napi_struct *n, int work_done)
6613 {
6614 	unsigned long flags, val, new, timeout = 0;
6615 	bool ret = true;
6616 
6617 	/*
6618 	 * 1) Don't let napi dequeue from the cpu poll list
6619 	 *    just in case its running on a different cpu.
6620 	 * 2) If we are busy polling, do nothing here, we have
6621 	 *    the guarantee we will be called later.
6622 	 */
6623 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6624 				 NAPIF_STATE_IN_BUSY_POLL)))
6625 		return false;
6626 
6627 	if (work_done) {
6628 		if (n->gro_bitmask)
6629 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6630 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6631 	}
6632 	if (n->defer_hard_irqs_count > 0) {
6633 		n->defer_hard_irqs_count--;
6634 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6635 		if (timeout)
6636 			ret = false;
6637 	}
6638 	if (n->gro_bitmask) {
6639 		/* When the NAPI instance uses a timeout and keeps postponing
6640 		 * it, we need to bound somehow the time packets are kept in
6641 		 * the GRO layer
6642 		 */
6643 		napi_gro_flush(n, !!timeout);
6644 	}
6645 
6646 	gro_normal_list(n);
6647 
6648 	if (unlikely(!list_empty(&n->poll_list))) {
6649 		/* If n->poll_list is not empty, we need to mask irqs */
6650 		local_irq_save(flags);
6651 		list_del_init(&n->poll_list);
6652 		local_irq_restore(flags);
6653 	}
6654 
6655 	do {
6656 		val = READ_ONCE(n->state);
6657 
6658 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6659 
6660 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6661 			      NAPIF_STATE_SCHED_THREADED |
6662 			      NAPIF_STATE_PREFER_BUSY_POLL);
6663 
6664 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6665 		 * because we will call napi->poll() one more time.
6666 		 * This C code was suggested by Alexander Duyck to help gcc.
6667 		 */
6668 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6669 						    NAPIF_STATE_SCHED;
6670 	} while (cmpxchg(&n->state, val, new) != val);
6671 
6672 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6673 		__napi_schedule(n);
6674 		return false;
6675 	}
6676 
6677 	if (timeout)
6678 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6679 			      HRTIMER_MODE_REL_PINNED);
6680 	return ret;
6681 }
6682 EXPORT_SYMBOL(napi_complete_done);
6683 
6684 /* must be called under rcu_read_lock(), as we dont take a reference */
6685 static struct napi_struct *napi_by_id(unsigned int napi_id)
6686 {
6687 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6688 	struct napi_struct *napi;
6689 
6690 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6691 		if (napi->napi_id == napi_id)
6692 			return napi;
6693 
6694 	return NULL;
6695 }
6696 
6697 #if defined(CONFIG_NET_RX_BUSY_POLL)
6698 
6699 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6700 {
6701 	if (!skip_schedule) {
6702 		gro_normal_list(napi);
6703 		__napi_schedule(napi);
6704 		return;
6705 	}
6706 
6707 	if (napi->gro_bitmask) {
6708 		/* flush too old packets
6709 		 * If HZ < 1000, flush all packets.
6710 		 */
6711 		napi_gro_flush(napi, HZ >= 1000);
6712 	}
6713 
6714 	gro_normal_list(napi);
6715 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6716 }
6717 
6718 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6719 			   u16 budget)
6720 {
6721 	bool skip_schedule = false;
6722 	unsigned long timeout;
6723 	int rc;
6724 
6725 	/* Busy polling means there is a high chance device driver hard irq
6726 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6727 	 * set in napi_schedule_prep().
6728 	 * Since we are about to call napi->poll() once more, we can safely
6729 	 * clear NAPI_STATE_MISSED.
6730 	 *
6731 	 * Note: x86 could use a single "lock and ..." instruction
6732 	 * to perform these two clear_bit()
6733 	 */
6734 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6735 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6736 
6737 	local_bh_disable();
6738 
6739 	if (prefer_busy_poll) {
6740 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6741 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6742 		if (napi->defer_hard_irqs_count && timeout) {
6743 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6744 			skip_schedule = true;
6745 		}
6746 	}
6747 
6748 	/* All we really want here is to re-enable device interrupts.
6749 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6750 	 */
6751 	rc = napi->poll(napi, budget);
6752 	/* We can't gro_normal_list() here, because napi->poll() might have
6753 	 * rearmed the napi (napi_complete_done()) in which case it could
6754 	 * already be running on another CPU.
6755 	 */
6756 	trace_napi_poll(napi, rc, budget);
6757 	netpoll_poll_unlock(have_poll_lock);
6758 	if (rc == budget)
6759 		__busy_poll_stop(napi, skip_schedule);
6760 	local_bh_enable();
6761 }
6762 
6763 void napi_busy_loop(unsigned int napi_id,
6764 		    bool (*loop_end)(void *, unsigned long),
6765 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6766 {
6767 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6768 	int (*napi_poll)(struct napi_struct *napi, int budget);
6769 	void *have_poll_lock = NULL;
6770 	struct napi_struct *napi;
6771 
6772 restart:
6773 	napi_poll = NULL;
6774 
6775 	rcu_read_lock();
6776 
6777 	napi = napi_by_id(napi_id);
6778 	if (!napi)
6779 		goto out;
6780 
6781 	preempt_disable();
6782 	for (;;) {
6783 		int work = 0;
6784 
6785 		local_bh_disable();
6786 		if (!napi_poll) {
6787 			unsigned long val = READ_ONCE(napi->state);
6788 
6789 			/* If multiple threads are competing for this napi,
6790 			 * we avoid dirtying napi->state as much as we can.
6791 			 */
6792 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6793 				   NAPIF_STATE_IN_BUSY_POLL)) {
6794 				if (prefer_busy_poll)
6795 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6796 				goto count;
6797 			}
6798 			if (cmpxchg(&napi->state, val,
6799 				    val | NAPIF_STATE_IN_BUSY_POLL |
6800 					  NAPIF_STATE_SCHED) != val) {
6801 				if (prefer_busy_poll)
6802 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6803 				goto count;
6804 			}
6805 			have_poll_lock = netpoll_poll_lock(napi);
6806 			napi_poll = napi->poll;
6807 		}
6808 		work = napi_poll(napi, budget);
6809 		trace_napi_poll(napi, work, budget);
6810 		gro_normal_list(napi);
6811 count:
6812 		if (work > 0)
6813 			__NET_ADD_STATS(dev_net(napi->dev),
6814 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6815 		local_bh_enable();
6816 
6817 		if (!loop_end || loop_end(loop_end_arg, start_time))
6818 			break;
6819 
6820 		if (unlikely(need_resched())) {
6821 			if (napi_poll)
6822 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6823 			preempt_enable();
6824 			rcu_read_unlock();
6825 			cond_resched();
6826 			if (loop_end(loop_end_arg, start_time))
6827 				return;
6828 			goto restart;
6829 		}
6830 		cpu_relax();
6831 	}
6832 	if (napi_poll)
6833 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6834 	preempt_enable();
6835 out:
6836 	rcu_read_unlock();
6837 }
6838 EXPORT_SYMBOL(napi_busy_loop);
6839 
6840 #endif /* CONFIG_NET_RX_BUSY_POLL */
6841 
6842 static void napi_hash_add(struct napi_struct *napi)
6843 {
6844 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6845 		return;
6846 
6847 	spin_lock(&napi_hash_lock);
6848 
6849 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6850 	do {
6851 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6852 			napi_gen_id = MIN_NAPI_ID;
6853 	} while (napi_by_id(napi_gen_id));
6854 	napi->napi_id = napi_gen_id;
6855 
6856 	hlist_add_head_rcu(&napi->napi_hash_node,
6857 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6858 
6859 	spin_unlock(&napi_hash_lock);
6860 }
6861 
6862 /* Warning : caller is responsible to make sure rcu grace period
6863  * is respected before freeing memory containing @napi
6864  */
6865 static void napi_hash_del(struct napi_struct *napi)
6866 {
6867 	spin_lock(&napi_hash_lock);
6868 
6869 	hlist_del_init_rcu(&napi->napi_hash_node);
6870 
6871 	spin_unlock(&napi_hash_lock);
6872 }
6873 
6874 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6875 {
6876 	struct napi_struct *napi;
6877 
6878 	napi = container_of(timer, struct napi_struct, timer);
6879 
6880 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6881 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6882 	 */
6883 	if (!napi_disable_pending(napi) &&
6884 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6885 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6886 		__napi_schedule_irqoff(napi);
6887 	}
6888 
6889 	return HRTIMER_NORESTART;
6890 }
6891 
6892 static void init_gro_hash(struct napi_struct *napi)
6893 {
6894 	int i;
6895 
6896 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6897 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6898 		napi->gro_hash[i].count = 0;
6899 	}
6900 	napi->gro_bitmask = 0;
6901 }
6902 
6903 int dev_set_threaded(struct net_device *dev, bool threaded)
6904 {
6905 	struct napi_struct *napi;
6906 	int err = 0;
6907 
6908 	if (dev->threaded == threaded)
6909 		return 0;
6910 
6911 	if (threaded) {
6912 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6913 			if (!napi->thread) {
6914 				err = napi_kthread_create(napi);
6915 				if (err) {
6916 					threaded = false;
6917 					break;
6918 				}
6919 			}
6920 		}
6921 	}
6922 
6923 	dev->threaded = threaded;
6924 
6925 	/* Make sure kthread is created before THREADED bit
6926 	 * is set.
6927 	 */
6928 	smp_mb__before_atomic();
6929 
6930 	/* Setting/unsetting threaded mode on a napi might not immediately
6931 	 * take effect, if the current napi instance is actively being
6932 	 * polled. In this case, the switch between threaded mode and
6933 	 * softirq mode will happen in the next round of napi_schedule().
6934 	 * This should not cause hiccups/stalls to the live traffic.
6935 	 */
6936 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6937 		if (threaded)
6938 			set_bit(NAPI_STATE_THREADED, &napi->state);
6939 		else
6940 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6941 	}
6942 
6943 	return err;
6944 }
6945 EXPORT_SYMBOL(dev_set_threaded);
6946 
6947 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6948 		    int (*poll)(struct napi_struct *, int), int weight)
6949 {
6950 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6951 		return;
6952 
6953 	INIT_LIST_HEAD(&napi->poll_list);
6954 	INIT_HLIST_NODE(&napi->napi_hash_node);
6955 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6956 	napi->timer.function = napi_watchdog;
6957 	init_gro_hash(napi);
6958 	napi->skb = NULL;
6959 	INIT_LIST_HEAD(&napi->rx_list);
6960 	napi->rx_count = 0;
6961 	napi->poll = poll;
6962 	if (weight > NAPI_POLL_WEIGHT)
6963 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6964 				weight);
6965 	napi->weight = weight;
6966 	napi->dev = dev;
6967 #ifdef CONFIG_NETPOLL
6968 	napi->poll_owner = -1;
6969 #endif
6970 	set_bit(NAPI_STATE_SCHED, &napi->state);
6971 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6972 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6973 	napi_hash_add(napi);
6974 	/* Create kthread for this napi if dev->threaded is set.
6975 	 * Clear dev->threaded if kthread creation failed so that
6976 	 * threaded mode will not be enabled in napi_enable().
6977 	 */
6978 	if (dev->threaded && napi_kthread_create(napi))
6979 		dev->threaded = 0;
6980 }
6981 EXPORT_SYMBOL(netif_napi_add);
6982 
6983 void napi_disable(struct napi_struct *n)
6984 {
6985 	might_sleep();
6986 	set_bit(NAPI_STATE_DISABLE, &n->state);
6987 
6988 	while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6989 		msleep(1);
6990 	while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6991 		msleep(1);
6992 
6993 	hrtimer_cancel(&n->timer);
6994 
6995 	clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6996 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6997 	clear_bit(NAPI_STATE_THREADED, &n->state);
6998 }
6999 EXPORT_SYMBOL(napi_disable);
7000 
7001 /**
7002  *	napi_enable - enable NAPI scheduling
7003  *	@n: NAPI context
7004  *
7005  * Resume NAPI from being scheduled on this context.
7006  * Must be paired with napi_disable.
7007  */
7008 void napi_enable(struct napi_struct *n)
7009 {
7010 	BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
7011 	smp_mb__before_atomic();
7012 	clear_bit(NAPI_STATE_SCHED, &n->state);
7013 	clear_bit(NAPI_STATE_NPSVC, &n->state);
7014 	if (n->dev->threaded && n->thread)
7015 		set_bit(NAPI_STATE_THREADED, &n->state);
7016 }
7017 EXPORT_SYMBOL(napi_enable);
7018 
7019 static void flush_gro_hash(struct napi_struct *napi)
7020 {
7021 	int i;
7022 
7023 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
7024 		struct sk_buff *skb, *n;
7025 
7026 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
7027 			kfree_skb(skb);
7028 		napi->gro_hash[i].count = 0;
7029 	}
7030 }
7031 
7032 /* Must be called in process context */
7033 void __netif_napi_del(struct napi_struct *napi)
7034 {
7035 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
7036 		return;
7037 
7038 	napi_hash_del(napi);
7039 	list_del_rcu(&napi->dev_list);
7040 	napi_free_frags(napi);
7041 
7042 	flush_gro_hash(napi);
7043 	napi->gro_bitmask = 0;
7044 
7045 	if (napi->thread) {
7046 		kthread_stop(napi->thread);
7047 		napi->thread = NULL;
7048 	}
7049 }
7050 EXPORT_SYMBOL(__netif_napi_del);
7051 
7052 static int __napi_poll(struct napi_struct *n, bool *repoll)
7053 {
7054 	int work, weight;
7055 
7056 	weight = n->weight;
7057 
7058 	/* This NAPI_STATE_SCHED test is for avoiding a race
7059 	 * with netpoll's poll_napi().  Only the entity which
7060 	 * obtains the lock and sees NAPI_STATE_SCHED set will
7061 	 * actually make the ->poll() call.  Therefore we avoid
7062 	 * accidentally calling ->poll() when NAPI is not scheduled.
7063 	 */
7064 	work = 0;
7065 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
7066 		work = n->poll(n, weight);
7067 		trace_napi_poll(n, work, weight);
7068 	}
7069 
7070 	if (unlikely(work > weight))
7071 		pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7072 			    n->poll, work, weight);
7073 
7074 	if (likely(work < weight))
7075 		return work;
7076 
7077 	/* Drivers must not modify the NAPI state if they
7078 	 * consume the entire weight.  In such cases this code
7079 	 * still "owns" the NAPI instance and therefore can
7080 	 * move the instance around on the list at-will.
7081 	 */
7082 	if (unlikely(napi_disable_pending(n))) {
7083 		napi_complete(n);
7084 		return work;
7085 	}
7086 
7087 	/* The NAPI context has more processing work, but busy-polling
7088 	 * is preferred. Exit early.
7089 	 */
7090 	if (napi_prefer_busy_poll(n)) {
7091 		if (napi_complete_done(n, work)) {
7092 			/* If timeout is not set, we need to make sure
7093 			 * that the NAPI is re-scheduled.
7094 			 */
7095 			napi_schedule(n);
7096 		}
7097 		return work;
7098 	}
7099 
7100 	if (n->gro_bitmask) {
7101 		/* flush too old packets
7102 		 * If HZ < 1000, flush all packets.
7103 		 */
7104 		napi_gro_flush(n, HZ >= 1000);
7105 	}
7106 
7107 	gro_normal_list(n);
7108 
7109 	/* Some drivers may have called napi_schedule
7110 	 * prior to exhausting their budget.
7111 	 */
7112 	if (unlikely(!list_empty(&n->poll_list))) {
7113 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7114 			     n->dev ? n->dev->name : "backlog");
7115 		return work;
7116 	}
7117 
7118 	*repoll = true;
7119 
7120 	return work;
7121 }
7122 
7123 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7124 {
7125 	bool do_repoll = false;
7126 	void *have;
7127 	int work;
7128 
7129 	list_del_init(&n->poll_list);
7130 
7131 	have = netpoll_poll_lock(n);
7132 
7133 	work = __napi_poll(n, &do_repoll);
7134 
7135 	if (do_repoll)
7136 		list_add_tail(&n->poll_list, repoll);
7137 
7138 	netpoll_poll_unlock(have);
7139 
7140 	return work;
7141 }
7142 
7143 static int napi_thread_wait(struct napi_struct *napi)
7144 {
7145 	bool woken = false;
7146 
7147 	set_current_state(TASK_INTERRUPTIBLE);
7148 
7149 	while (!kthread_should_stop()) {
7150 		/* Testing SCHED_THREADED bit here to make sure the current
7151 		 * kthread owns this napi and could poll on this napi.
7152 		 * Testing SCHED bit is not enough because SCHED bit might be
7153 		 * set by some other busy poll thread or by napi_disable().
7154 		 */
7155 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7156 			WARN_ON(!list_empty(&napi->poll_list));
7157 			__set_current_state(TASK_RUNNING);
7158 			return 0;
7159 		}
7160 
7161 		schedule();
7162 		/* woken being true indicates this thread owns this napi. */
7163 		woken = true;
7164 		set_current_state(TASK_INTERRUPTIBLE);
7165 	}
7166 	__set_current_state(TASK_RUNNING);
7167 
7168 	return -1;
7169 }
7170 
7171 static int napi_threaded_poll(void *data)
7172 {
7173 	struct napi_struct *napi = data;
7174 	void *have;
7175 
7176 	while (!napi_thread_wait(napi)) {
7177 		for (;;) {
7178 			bool repoll = false;
7179 
7180 			local_bh_disable();
7181 
7182 			have = netpoll_poll_lock(napi);
7183 			__napi_poll(napi, &repoll);
7184 			netpoll_poll_unlock(have);
7185 
7186 			local_bh_enable();
7187 
7188 			if (!repoll)
7189 				break;
7190 
7191 			cond_resched();
7192 		}
7193 	}
7194 	return 0;
7195 }
7196 
7197 static __latent_entropy void net_rx_action(struct softirq_action *h)
7198 {
7199 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7200 	unsigned long time_limit = jiffies +
7201 		usecs_to_jiffies(netdev_budget_usecs);
7202 	int budget = netdev_budget;
7203 	LIST_HEAD(list);
7204 	LIST_HEAD(repoll);
7205 
7206 	local_irq_disable();
7207 	list_splice_init(&sd->poll_list, &list);
7208 	local_irq_enable();
7209 
7210 	for (;;) {
7211 		struct napi_struct *n;
7212 
7213 		if (list_empty(&list)) {
7214 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7215 				return;
7216 			break;
7217 		}
7218 
7219 		n = list_first_entry(&list, struct napi_struct, poll_list);
7220 		budget -= napi_poll(n, &repoll);
7221 
7222 		/* If softirq window is exhausted then punt.
7223 		 * Allow this to run for 2 jiffies since which will allow
7224 		 * an average latency of 1.5/HZ.
7225 		 */
7226 		if (unlikely(budget <= 0 ||
7227 			     time_after_eq(jiffies, time_limit))) {
7228 			sd->time_squeeze++;
7229 			break;
7230 		}
7231 	}
7232 
7233 	local_irq_disable();
7234 
7235 	list_splice_tail_init(&sd->poll_list, &list);
7236 	list_splice_tail(&repoll, &list);
7237 	list_splice(&list, &sd->poll_list);
7238 	if (!list_empty(&sd->poll_list))
7239 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
7240 
7241 	net_rps_action_and_irq_enable(sd);
7242 }
7243 
7244 struct netdev_adjacent {
7245 	struct net_device *dev;
7246 
7247 	/* upper master flag, there can only be one master device per list */
7248 	bool master;
7249 
7250 	/* lookup ignore flag */
7251 	bool ignore;
7252 
7253 	/* counter for the number of times this device was added to us */
7254 	u16 ref_nr;
7255 
7256 	/* private field for the users */
7257 	void *private;
7258 
7259 	struct list_head list;
7260 	struct rcu_head rcu;
7261 };
7262 
7263 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7264 						 struct list_head *adj_list)
7265 {
7266 	struct netdev_adjacent *adj;
7267 
7268 	list_for_each_entry(adj, adj_list, list) {
7269 		if (adj->dev == adj_dev)
7270 			return adj;
7271 	}
7272 	return NULL;
7273 }
7274 
7275 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7276 				    struct netdev_nested_priv *priv)
7277 {
7278 	struct net_device *dev = (struct net_device *)priv->data;
7279 
7280 	return upper_dev == dev;
7281 }
7282 
7283 /**
7284  * netdev_has_upper_dev - Check if device is linked to an upper device
7285  * @dev: device
7286  * @upper_dev: upper device to check
7287  *
7288  * Find out if a device is linked to specified upper device and return true
7289  * in case it is. Note that this checks only immediate upper device,
7290  * not through a complete stack of devices. The caller must hold the RTNL lock.
7291  */
7292 bool netdev_has_upper_dev(struct net_device *dev,
7293 			  struct net_device *upper_dev)
7294 {
7295 	struct netdev_nested_priv priv = {
7296 		.data = (void *)upper_dev,
7297 	};
7298 
7299 	ASSERT_RTNL();
7300 
7301 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7302 					     &priv);
7303 }
7304 EXPORT_SYMBOL(netdev_has_upper_dev);
7305 
7306 /**
7307  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7308  * @dev: device
7309  * @upper_dev: upper device to check
7310  *
7311  * Find out if a device is linked to specified upper device and return true
7312  * in case it is. Note that this checks the entire upper device chain.
7313  * The caller must hold rcu lock.
7314  */
7315 
7316 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7317 				  struct net_device *upper_dev)
7318 {
7319 	struct netdev_nested_priv priv = {
7320 		.data = (void *)upper_dev,
7321 	};
7322 
7323 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7324 					       &priv);
7325 }
7326 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7327 
7328 /**
7329  * netdev_has_any_upper_dev - Check if device is linked to some device
7330  * @dev: device
7331  *
7332  * Find out if a device is linked to an upper device and return true in case
7333  * it is. The caller must hold the RTNL lock.
7334  */
7335 bool netdev_has_any_upper_dev(struct net_device *dev)
7336 {
7337 	ASSERT_RTNL();
7338 
7339 	return !list_empty(&dev->adj_list.upper);
7340 }
7341 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7342 
7343 /**
7344  * netdev_master_upper_dev_get - Get master upper device
7345  * @dev: device
7346  *
7347  * Find a master upper device and return pointer to it or NULL in case
7348  * it's not there. The caller must hold the RTNL lock.
7349  */
7350 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7351 {
7352 	struct netdev_adjacent *upper;
7353 
7354 	ASSERT_RTNL();
7355 
7356 	if (list_empty(&dev->adj_list.upper))
7357 		return NULL;
7358 
7359 	upper = list_first_entry(&dev->adj_list.upper,
7360 				 struct netdev_adjacent, list);
7361 	if (likely(upper->master))
7362 		return upper->dev;
7363 	return NULL;
7364 }
7365 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7366 
7367 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7368 {
7369 	struct netdev_adjacent *upper;
7370 
7371 	ASSERT_RTNL();
7372 
7373 	if (list_empty(&dev->adj_list.upper))
7374 		return NULL;
7375 
7376 	upper = list_first_entry(&dev->adj_list.upper,
7377 				 struct netdev_adjacent, list);
7378 	if (likely(upper->master) && !upper->ignore)
7379 		return upper->dev;
7380 	return NULL;
7381 }
7382 
7383 /**
7384  * netdev_has_any_lower_dev - Check if device is linked to some device
7385  * @dev: device
7386  *
7387  * Find out if a device is linked to a lower device and return true in case
7388  * it is. The caller must hold the RTNL lock.
7389  */
7390 static bool netdev_has_any_lower_dev(struct net_device *dev)
7391 {
7392 	ASSERT_RTNL();
7393 
7394 	return !list_empty(&dev->adj_list.lower);
7395 }
7396 
7397 void *netdev_adjacent_get_private(struct list_head *adj_list)
7398 {
7399 	struct netdev_adjacent *adj;
7400 
7401 	adj = list_entry(adj_list, struct netdev_adjacent, list);
7402 
7403 	return adj->private;
7404 }
7405 EXPORT_SYMBOL(netdev_adjacent_get_private);
7406 
7407 /**
7408  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7409  * @dev: device
7410  * @iter: list_head ** of the current position
7411  *
7412  * Gets the next device from the dev's upper list, starting from iter
7413  * position. The caller must hold RCU read lock.
7414  */
7415 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7416 						 struct list_head **iter)
7417 {
7418 	struct netdev_adjacent *upper;
7419 
7420 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7421 
7422 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7423 
7424 	if (&upper->list == &dev->adj_list.upper)
7425 		return NULL;
7426 
7427 	*iter = &upper->list;
7428 
7429 	return upper->dev;
7430 }
7431 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7432 
7433 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7434 						  struct list_head **iter,
7435 						  bool *ignore)
7436 {
7437 	struct netdev_adjacent *upper;
7438 
7439 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7440 
7441 	if (&upper->list == &dev->adj_list.upper)
7442 		return NULL;
7443 
7444 	*iter = &upper->list;
7445 	*ignore = upper->ignore;
7446 
7447 	return upper->dev;
7448 }
7449 
7450 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7451 						    struct list_head **iter)
7452 {
7453 	struct netdev_adjacent *upper;
7454 
7455 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7456 
7457 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7458 
7459 	if (&upper->list == &dev->adj_list.upper)
7460 		return NULL;
7461 
7462 	*iter = &upper->list;
7463 
7464 	return upper->dev;
7465 }
7466 
7467 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7468 				       int (*fn)(struct net_device *dev,
7469 					 struct netdev_nested_priv *priv),
7470 				       struct netdev_nested_priv *priv)
7471 {
7472 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7473 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7474 	int ret, cur = 0;
7475 	bool ignore;
7476 
7477 	now = dev;
7478 	iter = &dev->adj_list.upper;
7479 
7480 	while (1) {
7481 		if (now != dev) {
7482 			ret = fn(now, priv);
7483 			if (ret)
7484 				return ret;
7485 		}
7486 
7487 		next = NULL;
7488 		while (1) {
7489 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
7490 			if (!udev)
7491 				break;
7492 			if (ignore)
7493 				continue;
7494 
7495 			next = udev;
7496 			niter = &udev->adj_list.upper;
7497 			dev_stack[cur] = now;
7498 			iter_stack[cur++] = iter;
7499 			break;
7500 		}
7501 
7502 		if (!next) {
7503 			if (!cur)
7504 				return 0;
7505 			next = dev_stack[--cur];
7506 			niter = iter_stack[cur];
7507 		}
7508 
7509 		now = next;
7510 		iter = niter;
7511 	}
7512 
7513 	return 0;
7514 }
7515 
7516 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7517 				  int (*fn)(struct net_device *dev,
7518 					    struct netdev_nested_priv *priv),
7519 				  struct netdev_nested_priv *priv)
7520 {
7521 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7522 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7523 	int ret, cur = 0;
7524 
7525 	now = dev;
7526 	iter = &dev->adj_list.upper;
7527 
7528 	while (1) {
7529 		if (now != dev) {
7530 			ret = fn(now, priv);
7531 			if (ret)
7532 				return ret;
7533 		}
7534 
7535 		next = NULL;
7536 		while (1) {
7537 			udev = netdev_next_upper_dev_rcu(now, &iter);
7538 			if (!udev)
7539 				break;
7540 
7541 			next = udev;
7542 			niter = &udev->adj_list.upper;
7543 			dev_stack[cur] = now;
7544 			iter_stack[cur++] = iter;
7545 			break;
7546 		}
7547 
7548 		if (!next) {
7549 			if (!cur)
7550 				return 0;
7551 			next = dev_stack[--cur];
7552 			niter = iter_stack[cur];
7553 		}
7554 
7555 		now = next;
7556 		iter = niter;
7557 	}
7558 
7559 	return 0;
7560 }
7561 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7562 
7563 static bool __netdev_has_upper_dev(struct net_device *dev,
7564 				   struct net_device *upper_dev)
7565 {
7566 	struct netdev_nested_priv priv = {
7567 		.flags = 0,
7568 		.data = (void *)upper_dev,
7569 	};
7570 
7571 	ASSERT_RTNL();
7572 
7573 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7574 					   &priv);
7575 }
7576 
7577 /**
7578  * netdev_lower_get_next_private - Get the next ->private from the
7579  *				   lower neighbour list
7580  * @dev: device
7581  * @iter: list_head ** of the current position
7582  *
7583  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7584  * list, starting from iter position. The caller must hold either hold the
7585  * RTNL lock or its own locking that guarantees that the neighbour lower
7586  * list will remain unchanged.
7587  */
7588 void *netdev_lower_get_next_private(struct net_device *dev,
7589 				    struct list_head **iter)
7590 {
7591 	struct netdev_adjacent *lower;
7592 
7593 	lower = list_entry(*iter, struct netdev_adjacent, list);
7594 
7595 	if (&lower->list == &dev->adj_list.lower)
7596 		return NULL;
7597 
7598 	*iter = lower->list.next;
7599 
7600 	return lower->private;
7601 }
7602 EXPORT_SYMBOL(netdev_lower_get_next_private);
7603 
7604 /**
7605  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7606  *				       lower neighbour list, RCU
7607  *				       variant
7608  * @dev: device
7609  * @iter: list_head ** of the current position
7610  *
7611  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7612  * list, starting from iter position. The caller must hold RCU read lock.
7613  */
7614 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7615 					struct list_head **iter)
7616 {
7617 	struct netdev_adjacent *lower;
7618 
7619 	WARN_ON_ONCE(!rcu_read_lock_held());
7620 
7621 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7622 
7623 	if (&lower->list == &dev->adj_list.lower)
7624 		return NULL;
7625 
7626 	*iter = &lower->list;
7627 
7628 	return lower->private;
7629 }
7630 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7631 
7632 /**
7633  * netdev_lower_get_next - Get the next device from the lower neighbour
7634  *                         list
7635  * @dev: device
7636  * @iter: list_head ** of the current position
7637  *
7638  * Gets the next netdev_adjacent from the dev's lower neighbour
7639  * list, starting from iter position. The caller must hold RTNL lock or
7640  * its own locking that guarantees that the neighbour lower
7641  * list will remain unchanged.
7642  */
7643 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7644 {
7645 	struct netdev_adjacent *lower;
7646 
7647 	lower = list_entry(*iter, struct netdev_adjacent, list);
7648 
7649 	if (&lower->list == &dev->adj_list.lower)
7650 		return NULL;
7651 
7652 	*iter = lower->list.next;
7653 
7654 	return lower->dev;
7655 }
7656 EXPORT_SYMBOL(netdev_lower_get_next);
7657 
7658 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7659 						struct list_head **iter)
7660 {
7661 	struct netdev_adjacent *lower;
7662 
7663 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7664 
7665 	if (&lower->list == &dev->adj_list.lower)
7666 		return NULL;
7667 
7668 	*iter = &lower->list;
7669 
7670 	return lower->dev;
7671 }
7672 
7673 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7674 						  struct list_head **iter,
7675 						  bool *ignore)
7676 {
7677 	struct netdev_adjacent *lower;
7678 
7679 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7680 
7681 	if (&lower->list == &dev->adj_list.lower)
7682 		return NULL;
7683 
7684 	*iter = &lower->list;
7685 	*ignore = lower->ignore;
7686 
7687 	return lower->dev;
7688 }
7689 
7690 int netdev_walk_all_lower_dev(struct net_device *dev,
7691 			      int (*fn)(struct net_device *dev,
7692 					struct netdev_nested_priv *priv),
7693 			      struct netdev_nested_priv *priv)
7694 {
7695 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7696 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7697 	int ret, cur = 0;
7698 
7699 	now = dev;
7700 	iter = &dev->adj_list.lower;
7701 
7702 	while (1) {
7703 		if (now != dev) {
7704 			ret = fn(now, priv);
7705 			if (ret)
7706 				return ret;
7707 		}
7708 
7709 		next = NULL;
7710 		while (1) {
7711 			ldev = netdev_next_lower_dev(now, &iter);
7712 			if (!ldev)
7713 				break;
7714 
7715 			next = ldev;
7716 			niter = &ldev->adj_list.lower;
7717 			dev_stack[cur] = now;
7718 			iter_stack[cur++] = iter;
7719 			break;
7720 		}
7721 
7722 		if (!next) {
7723 			if (!cur)
7724 				return 0;
7725 			next = dev_stack[--cur];
7726 			niter = iter_stack[cur];
7727 		}
7728 
7729 		now = next;
7730 		iter = niter;
7731 	}
7732 
7733 	return 0;
7734 }
7735 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7736 
7737 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7738 				       int (*fn)(struct net_device *dev,
7739 					 struct netdev_nested_priv *priv),
7740 				       struct netdev_nested_priv *priv)
7741 {
7742 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7743 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7744 	int ret, cur = 0;
7745 	bool ignore;
7746 
7747 	now = dev;
7748 	iter = &dev->adj_list.lower;
7749 
7750 	while (1) {
7751 		if (now != dev) {
7752 			ret = fn(now, priv);
7753 			if (ret)
7754 				return ret;
7755 		}
7756 
7757 		next = NULL;
7758 		while (1) {
7759 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7760 			if (!ldev)
7761 				break;
7762 			if (ignore)
7763 				continue;
7764 
7765 			next = ldev;
7766 			niter = &ldev->adj_list.lower;
7767 			dev_stack[cur] = now;
7768 			iter_stack[cur++] = iter;
7769 			break;
7770 		}
7771 
7772 		if (!next) {
7773 			if (!cur)
7774 				return 0;
7775 			next = dev_stack[--cur];
7776 			niter = iter_stack[cur];
7777 		}
7778 
7779 		now = next;
7780 		iter = niter;
7781 	}
7782 
7783 	return 0;
7784 }
7785 
7786 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7787 					     struct list_head **iter)
7788 {
7789 	struct netdev_adjacent *lower;
7790 
7791 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7792 	if (&lower->list == &dev->adj_list.lower)
7793 		return NULL;
7794 
7795 	*iter = &lower->list;
7796 
7797 	return lower->dev;
7798 }
7799 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7800 
7801 static u8 __netdev_upper_depth(struct net_device *dev)
7802 {
7803 	struct net_device *udev;
7804 	struct list_head *iter;
7805 	u8 max_depth = 0;
7806 	bool ignore;
7807 
7808 	for (iter = &dev->adj_list.upper,
7809 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7810 	     udev;
7811 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7812 		if (ignore)
7813 			continue;
7814 		if (max_depth < udev->upper_level)
7815 			max_depth = udev->upper_level;
7816 	}
7817 
7818 	return max_depth;
7819 }
7820 
7821 static u8 __netdev_lower_depth(struct net_device *dev)
7822 {
7823 	struct net_device *ldev;
7824 	struct list_head *iter;
7825 	u8 max_depth = 0;
7826 	bool ignore;
7827 
7828 	for (iter = &dev->adj_list.lower,
7829 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7830 	     ldev;
7831 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7832 		if (ignore)
7833 			continue;
7834 		if (max_depth < ldev->lower_level)
7835 			max_depth = ldev->lower_level;
7836 	}
7837 
7838 	return max_depth;
7839 }
7840 
7841 static int __netdev_update_upper_level(struct net_device *dev,
7842 				       struct netdev_nested_priv *__unused)
7843 {
7844 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7845 	return 0;
7846 }
7847 
7848 static int __netdev_update_lower_level(struct net_device *dev,
7849 				       struct netdev_nested_priv *priv)
7850 {
7851 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7852 
7853 #ifdef CONFIG_LOCKDEP
7854 	if (!priv)
7855 		return 0;
7856 
7857 	if (priv->flags & NESTED_SYNC_IMM)
7858 		dev->nested_level = dev->lower_level - 1;
7859 	if (priv->flags & NESTED_SYNC_TODO)
7860 		net_unlink_todo(dev);
7861 #endif
7862 	return 0;
7863 }
7864 
7865 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7866 				  int (*fn)(struct net_device *dev,
7867 					    struct netdev_nested_priv *priv),
7868 				  struct netdev_nested_priv *priv)
7869 {
7870 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7871 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7872 	int ret, cur = 0;
7873 
7874 	now = dev;
7875 	iter = &dev->adj_list.lower;
7876 
7877 	while (1) {
7878 		if (now != dev) {
7879 			ret = fn(now, priv);
7880 			if (ret)
7881 				return ret;
7882 		}
7883 
7884 		next = NULL;
7885 		while (1) {
7886 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7887 			if (!ldev)
7888 				break;
7889 
7890 			next = ldev;
7891 			niter = &ldev->adj_list.lower;
7892 			dev_stack[cur] = now;
7893 			iter_stack[cur++] = iter;
7894 			break;
7895 		}
7896 
7897 		if (!next) {
7898 			if (!cur)
7899 				return 0;
7900 			next = dev_stack[--cur];
7901 			niter = iter_stack[cur];
7902 		}
7903 
7904 		now = next;
7905 		iter = niter;
7906 	}
7907 
7908 	return 0;
7909 }
7910 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7911 
7912 /**
7913  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7914  *				       lower neighbour list, RCU
7915  *				       variant
7916  * @dev: device
7917  *
7918  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7919  * list. The caller must hold RCU read lock.
7920  */
7921 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7922 {
7923 	struct netdev_adjacent *lower;
7924 
7925 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7926 			struct netdev_adjacent, list);
7927 	if (lower)
7928 		return lower->private;
7929 	return NULL;
7930 }
7931 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7932 
7933 /**
7934  * netdev_master_upper_dev_get_rcu - Get master upper device
7935  * @dev: device
7936  *
7937  * Find a master upper device and return pointer to it or NULL in case
7938  * it's not there. The caller must hold the RCU read lock.
7939  */
7940 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7941 {
7942 	struct netdev_adjacent *upper;
7943 
7944 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7945 				       struct netdev_adjacent, list);
7946 	if (upper && likely(upper->master))
7947 		return upper->dev;
7948 	return NULL;
7949 }
7950 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7951 
7952 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7953 			      struct net_device *adj_dev,
7954 			      struct list_head *dev_list)
7955 {
7956 	char linkname[IFNAMSIZ+7];
7957 
7958 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7959 		"upper_%s" : "lower_%s", adj_dev->name);
7960 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7961 				 linkname);
7962 }
7963 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7964 			       char *name,
7965 			       struct list_head *dev_list)
7966 {
7967 	char linkname[IFNAMSIZ+7];
7968 
7969 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7970 		"upper_%s" : "lower_%s", name);
7971 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7972 }
7973 
7974 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7975 						 struct net_device *adj_dev,
7976 						 struct list_head *dev_list)
7977 {
7978 	return (dev_list == &dev->adj_list.upper ||
7979 		dev_list == &dev->adj_list.lower) &&
7980 		net_eq(dev_net(dev), dev_net(adj_dev));
7981 }
7982 
7983 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7984 					struct net_device *adj_dev,
7985 					struct list_head *dev_list,
7986 					void *private, bool master)
7987 {
7988 	struct netdev_adjacent *adj;
7989 	int ret;
7990 
7991 	adj = __netdev_find_adj(adj_dev, dev_list);
7992 
7993 	if (adj) {
7994 		adj->ref_nr += 1;
7995 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7996 			 dev->name, adj_dev->name, adj->ref_nr);
7997 
7998 		return 0;
7999 	}
8000 
8001 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
8002 	if (!adj)
8003 		return -ENOMEM;
8004 
8005 	adj->dev = adj_dev;
8006 	adj->master = master;
8007 	adj->ref_nr = 1;
8008 	adj->private = private;
8009 	adj->ignore = false;
8010 	dev_hold(adj_dev);
8011 
8012 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
8013 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
8014 
8015 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
8016 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
8017 		if (ret)
8018 			goto free_adj;
8019 	}
8020 
8021 	/* Ensure that master link is always the first item in list. */
8022 	if (master) {
8023 		ret = sysfs_create_link(&(dev->dev.kobj),
8024 					&(adj_dev->dev.kobj), "master");
8025 		if (ret)
8026 			goto remove_symlinks;
8027 
8028 		list_add_rcu(&adj->list, dev_list);
8029 	} else {
8030 		list_add_tail_rcu(&adj->list, dev_list);
8031 	}
8032 
8033 	return 0;
8034 
8035 remove_symlinks:
8036 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8037 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8038 free_adj:
8039 	kfree(adj);
8040 	dev_put(adj_dev);
8041 
8042 	return ret;
8043 }
8044 
8045 static void __netdev_adjacent_dev_remove(struct net_device *dev,
8046 					 struct net_device *adj_dev,
8047 					 u16 ref_nr,
8048 					 struct list_head *dev_list)
8049 {
8050 	struct netdev_adjacent *adj;
8051 
8052 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
8053 		 dev->name, adj_dev->name, ref_nr);
8054 
8055 	adj = __netdev_find_adj(adj_dev, dev_list);
8056 
8057 	if (!adj) {
8058 		pr_err("Adjacency does not exist for device %s from %s\n",
8059 		       dev->name, adj_dev->name);
8060 		WARN_ON(1);
8061 		return;
8062 	}
8063 
8064 	if (adj->ref_nr > ref_nr) {
8065 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8066 			 dev->name, adj_dev->name, ref_nr,
8067 			 adj->ref_nr - ref_nr);
8068 		adj->ref_nr -= ref_nr;
8069 		return;
8070 	}
8071 
8072 	if (adj->master)
8073 		sysfs_remove_link(&(dev->dev.kobj), "master");
8074 
8075 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8076 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8077 
8078 	list_del_rcu(&adj->list);
8079 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8080 		 adj_dev->name, dev->name, adj_dev->name);
8081 	dev_put(adj_dev);
8082 	kfree_rcu(adj, rcu);
8083 }
8084 
8085 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8086 					    struct net_device *upper_dev,
8087 					    struct list_head *up_list,
8088 					    struct list_head *down_list,
8089 					    void *private, bool master)
8090 {
8091 	int ret;
8092 
8093 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8094 					   private, master);
8095 	if (ret)
8096 		return ret;
8097 
8098 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8099 					   private, false);
8100 	if (ret) {
8101 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8102 		return ret;
8103 	}
8104 
8105 	return 0;
8106 }
8107 
8108 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8109 					       struct net_device *upper_dev,
8110 					       u16 ref_nr,
8111 					       struct list_head *up_list,
8112 					       struct list_head *down_list)
8113 {
8114 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8115 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8116 }
8117 
8118 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8119 						struct net_device *upper_dev,
8120 						void *private, bool master)
8121 {
8122 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8123 						&dev->adj_list.upper,
8124 						&upper_dev->adj_list.lower,
8125 						private, master);
8126 }
8127 
8128 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8129 						   struct net_device *upper_dev)
8130 {
8131 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8132 					   &dev->adj_list.upper,
8133 					   &upper_dev->adj_list.lower);
8134 }
8135 
8136 static int __netdev_upper_dev_link(struct net_device *dev,
8137 				   struct net_device *upper_dev, bool master,
8138 				   void *upper_priv, void *upper_info,
8139 				   struct netdev_nested_priv *priv,
8140 				   struct netlink_ext_ack *extack)
8141 {
8142 	struct netdev_notifier_changeupper_info changeupper_info = {
8143 		.info = {
8144 			.dev = dev,
8145 			.extack = extack,
8146 		},
8147 		.upper_dev = upper_dev,
8148 		.master = master,
8149 		.linking = true,
8150 		.upper_info = upper_info,
8151 	};
8152 	struct net_device *master_dev;
8153 	int ret = 0;
8154 
8155 	ASSERT_RTNL();
8156 
8157 	if (dev == upper_dev)
8158 		return -EBUSY;
8159 
8160 	/* To prevent loops, check if dev is not upper device to upper_dev. */
8161 	if (__netdev_has_upper_dev(upper_dev, dev))
8162 		return -EBUSY;
8163 
8164 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8165 		return -EMLINK;
8166 
8167 	if (!master) {
8168 		if (__netdev_has_upper_dev(dev, upper_dev))
8169 			return -EEXIST;
8170 	} else {
8171 		master_dev = __netdev_master_upper_dev_get(dev);
8172 		if (master_dev)
8173 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
8174 	}
8175 
8176 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8177 					    &changeupper_info.info);
8178 	ret = notifier_to_errno(ret);
8179 	if (ret)
8180 		return ret;
8181 
8182 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8183 						   master);
8184 	if (ret)
8185 		return ret;
8186 
8187 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8188 					    &changeupper_info.info);
8189 	ret = notifier_to_errno(ret);
8190 	if (ret)
8191 		goto rollback;
8192 
8193 	__netdev_update_upper_level(dev, NULL);
8194 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8195 
8196 	__netdev_update_lower_level(upper_dev, priv);
8197 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8198 				    priv);
8199 
8200 	return 0;
8201 
8202 rollback:
8203 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8204 
8205 	return ret;
8206 }
8207 
8208 /**
8209  * netdev_upper_dev_link - Add a link to the upper device
8210  * @dev: device
8211  * @upper_dev: new upper device
8212  * @extack: netlink extended ack
8213  *
8214  * Adds a link to device which is upper to this one. The caller must hold
8215  * the RTNL lock. On a failure a negative errno code is returned.
8216  * On success the reference counts are adjusted and the function
8217  * returns zero.
8218  */
8219 int netdev_upper_dev_link(struct net_device *dev,
8220 			  struct net_device *upper_dev,
8221 			  struct netlink_ext_ack *extack)
8222 {
8223 	struct netdev_nested_priv priv = {
8224 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8225 		.data = NULL,
8226 	};
8227 
8228 	return __netdev_upper_dev_link(dev, upper_dev, false,
8229 				       NULL, NULL, &priv, extack);
8230 }
8231 EXPORT_SYMBOL(netdev_upper_dev_link);
8232 
8233 /**
8234  * netdev_master_upper_dev_link - Add a master link to the upper device
8235  * @dev: device
8236  * @upper_dev: new upper device
8237  * @upper_priv: upper device private
8238  * @upper_info: upper info to be passed down via notifier
8239  * @extack: netlink extended ack
8240  *
8241  * Adds a link to device which is upper to this one. In this case, only
8242  * one master upper device can be linked, although other non-master devices
8243  * might be linked as well. The caller must hold the RTNL lock.
8244  * On a failure a negative errno code is returned. On success the reference
8245  * counts are adjusted and the function returns zero.
8246  */
8247 int netdev_master_upper_dev_link(struct net_device *dev,
8248 				 struct net_device *upper_dev,
8249 				 void *upper_priv, void *upper_info,
8250 				 struct netlink_ext_ack *extack)
8251 {
8252 	struct netdev_nested_priv priv = {
8253 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8254 		.data = NULL,
8255 	};
8256 
8257 	return __netdev_upper_dev_link(dev, upper_dev, true,
8258 				       upper_priv, upper_info, &priv, extack);
8259 }
8260 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8261 
8262 static void __netdev_upper_dev_unlink(struct net_device *dev,
8263 				      struct net_device *upper_dev,
8264 				      struct netdev_nested_priv *priv)
8265 {
8266 	struct netdev_notifier_changeupper_info changeupper_info = {
8267 		.info = {
8268 			.dev = dev,
8269 		},
8270 		.upper_dev = upper_dev,
8271 		.linking = false,
8272 	};
8273 
8274 	ASSERT_RTNL();
8275 
8276 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8277 
8278 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8279 				      &changeupper_info.info);
8280 
8281 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8282 
8283 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8284 				      &changeupper_info.info);
8285 
8286 	__netdev_update_upper_level(dev, NULL);
8287 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8288 
8289 	__netdev_update_lower_level(upper_dev, priv);
8290 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8291 				    priv);
8292 }
8293 
8294 /**
8295  * netdev_upper_dev_unlink - Removes a link to upper device
8296  * @dev: device
8297  * @upper_dev: new upper device
8298  *
8299  * Removes a link to device which is upper to this one. The caller must hold
8300  * the RTNL lock.
8301  */
8302 void netdev_upper_dev_unlink(struct net_device *dev,
8303 			     struct net_device *upper_dev)
8304 {
8305 	struct netdev_nested_priv priv = {
8306 		.flags = NESTED_SYNC_TODO,
8307 		.data = NULL,
8308 	};
8309 
8310 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
8311 }
8312 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8313 
8314 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8315 				      struct net_device *lower_dev,
8316 				      bool val)
8317 {
8318 	struct netdev_adjacent *adj;
8319 
8320 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8321 	if (adj)
8322 		adj->ignore = val;
8323 
8324 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8325 	if (adj)
8326 		adj->ignore = val;
8327 }
8328 
8329 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8330 					struct net_device *lower_dev)
8331 {
8332 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8333 }
8334 
8335 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8336 				       struct net_device *lower_dev)
8337 {
8338 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8339 }
8340 
8341 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8342 				   struct net_device *new_dev,
8343 				   struct net_device *dev,
8344 				   struct netlink_ext_ack *extack)
8345 {
8346 	struct netdev_nested_priv priv = {
8347 		.flags = 0,
8348 		.data = NULL,
8349 	};
8350 	int err;
8351 
8352 	if (!new_dev)
8353 		return 0;
8354 
8355 	if (old_dev && new_dev != old_dev)
8356 		netdev_adjacent_dev_disable(dev, old_dev);
8357 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8358 				      extack);
8359 	if (err) {
8360 		if (old_dev && new_dev != old_dev)
8361 			netdev_adjacent_dev_enable(dev, old_dev);
8362 		return err;
8363 	}
8364 
8365 	return 0;
8366 }
8367 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8368 
8369 void netdev_adjacent_change_commit(struct net_device *old_dev,
8370 				   struct net_device *new_dev,
8371 				   struct net_device *dev)
8372 {
8373 	struct netdev_nested_priv priv = {
8374 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8375 		.data = NULL,
8376 	};
8377 
8378 	if (!new_dev || !old_dev)
8379 		return;
8380 
8381 	if (new_dev == old_dev)
8382 		return;
8383 
8384 	netdev_adjacent_dev_enable(dev, old_dev);
8385 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
8386 }
8387 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8388 
8389 void netdev_adjacent_change_abort(struct net_device *old_dev,
8390 				  struct net_device *new_dev,
8391 				  struct net_device *dev)
8392 {
8393 	struct netdev_nested_priv priv = {
8394 		.flags = 0,
8395 		.data = NULL,
8396 	};
8397 
8398 	if (!new_dev)
8399 		return;
8400 
8401 	if (old_dev && new_dev != old_dev)
8402 		netdev_adjacent_dev_enable(dev, old_dev);
8403 
8404 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
8405 }
8406 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8407 
8408 /**
8409  * netdev_bonding_info_change - Dispatch event about slave change
8410  * @dev: device
8411  * @bonding_info: info to dispatch
8412  *
8413  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8414  * The caller must hold the RTNL lock.
8415  */
8416 void netdev_bonding_info_change(struct net_device *dev,
8417 				struct netdev_bonding_info *bonding_info)
8418 {
8419 	struct netdev_notifier_bonding_info info = {
8420 		.info.dev = dev,
8421 	};
8422 
8423 	memcpy(&info.bonding_info, bonding_info,
8424 	       sizeof(struct netdev_bonding_info));
8425 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8426 				      &info.info);
8427 }
8428 EXPORT_SYMBOL(netdev_bonding_info_change);
8429 
8430 /**
8431  * netdev_get_xmit_slave - Get the xmit slave of master device
8432  * @dev: device
8433  * @skb: The packet
8434  * @all_slaves: assume all the slaves are active
8435  *
8436  * The reference counters are not incremented so the caller must be
8437  * careful with locks. The caller must hold RCU lock.
8438  * %NULL is returned if no slave is found.
8439  */
8440 
8441 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8442 					 struct sk_buff *skb,
8443 					 bool all_slaves)
8444 {
8445 	const struct net_device_ops *ops = dev->netdev_ops;
8446 
8447 	if (!ops->ndo_get_xmit_slave)
8448 		return NULL;
8449 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8450 }
8451 EXPORT_SYMBOL(netdev_get_xmit_slave);
8452 
8453 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8454 						  struct sock *sk)
8455 {
8456 	const struct net_device_ops *ops = dev->netdev_ops;
8457 
8458 	if (!ops->ndo_sk_get_lower_dev)
8459 		return NULL;
8460 	return ops->ndo_sk_get_lower_dev(dev, sk);
8461 }
8462 
8463 /**
8464  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8465  * @dev: device
8466  * @sk: the socket
8467  *
8468  * %NULL is returned if no lower device is found.
8469  */
8470 
8471 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8472 					    struct sock *sk)
8473 {
8474 	struct net_device *lower;
8475 
8476 	lower = netdev_sk_get_lower_dev(dev, sk);
8477 	while (lower) {
8478 		dev = lower;
8479 		lower = netdev_sk_get_lower_dev(dev, sk);
8480 	}
8481 
8482 	return dev;
8483 }
8484 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8485 
8486 static void netdev_adjacent_add_links(struct net_device *dev)
8487 {
8488 	struct netdev_adjacent *iter;
8489 
8490 	struct net *net = dev_net(dev);
8491 
8492 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8493 		if (!net_eq(net, dev_net(iter->dev)))
8494 			continue;
8495 		netdev_adjacent_sysfs_add(iter->dev, dev,
8496 					  &iter->dev->adj_list.lower);
8497 		netdev_adjacent_sysfs_add(dev, iter->dev,
8498 					  &dev->adj_list.upper);
8499 	}
8500 
8501 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8502 		if (!net_eq(net, dev_net(iter->dev)))
8503 			continue;
8504 		netdev_adjacent_sysfs_add(iter->dev, dev,
8505 					  &iter->dev->adj_list.upper);
8506 		netdev_adjacent_sysfs_add(dev, iter->dev,
8507 					  &dev->adj_list.lower);
8508 	}
8509 }
8510 
8511 static void netdev_adjacent_del_links(struct net_device *dev)
8512 {
8513 	struct netdev_adjacent *iter;
8514 
8515 	struct net *net = dev_net(dev);
8516 
8517 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8518 		if (!net_eq(net, dev_net(iter->dev)))
8519 			continue;
8520 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8521 					  &iter->dev->adj_list.lower);
8522 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8523 					  &dev->adj_list.upper);
8524 	}
8525 
8526 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8527 		if (!net_eq(net, dev_net(iter->dev)))
8528 			continue;
8529 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8530 					  &iter->dev->adj_list.upper);
8531 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8532 					  &dev->adj_list.lower);
8533 	}
8534 }
8535 
8536 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8537 {
8538 	struct netdev_adjacent *iter;
8539 
8540 	struct net *net = dev_net(dev);
8541 
8542 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8543 		if (!net_eq(net, dev_net(iter->dev)))
8544 			continue;
8545 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8546 					  &iter->dev->adj_list.lower);
8547 		netdev_adjacent_sysfs_add(iter->dev, dev,
8548 					  &iter->dev->adj_list.lower);
8549 	}
8550 
8551 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8552 		if (!net_eq(net, dev_net(iter->dev)))
8553 			continue;
8554 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8555 					  &iter->dev->adj_list.upper);
8556 		netdev_adjacent_sysfs_add(iter->dev, dev,
8557 					  &iter->dev->adj_list.upper);
8558 	}
8559 }
8560 
8561 void *netdev_lower_dev_get_private(struct net_device *dev,
8562 				   struct net_device *lower_dev)
8563 {
8564 	struct netdev_adjacent *lower;
8565 
8566 	if (!lower_dev)
8567 		return NULL;
8568 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8569 	if (!lower)
8570 		return NULL;
8571 
8572 	return lower->private;
8573 }
8574 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8575 
8576 
8577 /**
8578  * netdev_lower_state_changed - Dispatch event about lower device state change
8579  * @lower_dev: device
8580  * @lower_state_info: state to dispatch
8581  *
8582  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8583  * The caller must hold the RTNL lock.
8584  */
8585 void netdev_lower_state_changed(struct net_device *lower_dev,
8586 				void *lower_state_info)
8587 {
8588 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8589 		.info.dev = lower_dev,
8590 	};
8591 
8592 	ASSERT_RTNL();
8593 	changelowerstate_info.lower_state_info = lower_state_info;
8594 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8595 				      &changelowerstate_info.info);
8596 }
8597 EXPORT_SYMBOL(netdev_lower_state_changed);
8598 
8599 static void dev_change_rx_flags(struct net_device *dev, int flags)
8600 {
8601 	const struct net_device_ops *ops = dev->netdev_ops;
8602 
8603 	if (ops->ndo_change_rx_flags)
8604 		ops->ndo_change_rx_flags(dev, flags);
8605 }
8606 
8607 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8608 {
8609 	unsigned int old_flags = dev->flags;
8610 	kuid_t uid;
8611 	kgid_t gid;
8612 
8613 	ASSERT_RTNL();
8614 
8615 	dev->flags |= IFF_PROMISC;
8616 	dev->promiscuity += inc;
8617 	if (dev->promiscuity == 0) {
8618 		/*
8619 		 * Avoid overflow.
8620 		 * If inc causes overflow, untouch promisc and return error.
8621 		 */
8622 		if (inc < 0)
8623 			dev->flags &= ~IFF_PROMISC;
8624 		else {
8625 			dev->promiscuity -= inc;
8626 			pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8627 				dev->name);
8628 			return -EOVERFLOW;
8629 		}
8630 	}
8631 	if (dev->flags != old_flags) {
8632 		pr_info("device %s %s promiscuous mode\n",
8633 			dev->name,
8634 			dev->flags & IFF_PROMISC ? "entered" : "left");
8635 		if (audit_enabled) {
8636 			current_uid_gid(&uid, &gid);
8637 			audit_log(audit_context(), GFP_ATOMIC,
8638 				  AUDIT_ANOM_PROMISCUOUS,
8639 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8640 				  dev->name, (dev->flags & IFF_PROMISC),
8641 				  (old_flags & IFF_PROMISC),
8642 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8643 				  from_kuid(&init_user_ns, uid),
8644 				  from_kgid(&init_user_ns, gid),
8645 				  audit_get_sessionid(current));
8646 		}
8647 
8648 		dev_change_rx_flags(dev, IFF_PROMISC);
8649 	}
8650 	if (notify)
8651 		__dev_notify_flags(dev, old_flags, IFF_PROMISC);
8652 	return 0;
8653 }
8654 
8655 /**
8656  *	dev_set_promiscuity	- update promiscuity count on a device
8657  *	@dev: device
8658  *	@inc: modifier
8659  *
8660  *	Add or remove promiscuity from a device. While the count in the device
8661  *	remains above zero the interface remains promiscuous. Once it hits zero
8662  *	the device reverts back to normal filtering operation. A negative inc
8663  *	value is used to drop promiscuity on the device.
8664  *	Return 0 if successful or a negative errno code on error.
8665  */
8666 int dev_set_promiscuity(struct net_device *dev, int inc)
8667 {
8668 	unsigned int old_flags = dev->flags;
8669 	int err;
8670 
8671 	err = __dev_set_promiscuity(dev, inc, true);
8672 	if (err < 0)
8673 		return err;
8674 	if (dev->flags != old_flags)
8675 		dev_set_rx_mode(dev);
8676 	return err;
8677 }
8678 EXPORT_SYMBOL(dev_set_promiscuity);
8679 
8680 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8681 {
8682 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8683 
8684 	ASSERT_RTNL();
8685 
8686 	dev->flags |= IFF_ALLMULTI;
8687 	dev->allmulti += inc;
8688 	if (dev->allmulti == 0) {
8689 		/*
8690 		 * Avoid overflow.
8691 		 * If inc causes overflow, untouch allmulti and return error.
8692 		 */
8693 		if (inc < 0)
8694 			dev->flags &= ~IFF_ALLMULTI;
8695 		else {
8696 			dev->allmulti -= inc;
8697 			pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8698 				dev->name);
8699 			return -EOVERFLOW;
8700 		}
8701 	}
8702 	if (dev->flags ^ old_flags) {
8703 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8704 		dev_set_rx_mode(dev);
8705 		if (notify)
8706 			__dev_notify_flags(dev, old_flags,
8707 					   dev->gflags ^ old_gflags);
8708 	}
8709 	return 0;
8710 }
8711 
8712 /**
8713  *	dev_set_allmulti	- update allmulti count on a device
8714  *	@dev: device
8715  *	@inc: modifier
8716  *
8717  *	Add or remove reception of all multicast frames to a device. While the
8718  *	count in the device remains above zero the interface remains listening
8719  *	to all interfaces. Once it hits zero the device reverts back to normal
8720  *	filtering operation. A negative @inc value is used to drop the counter
8721  *	when releasing a resource needing all multicasts.
8722  *	Return 0 if successful or a negative errno code on error.
8723  */
8724 
8725 int dev_set_allmulti(struct net_device *dev, int inc)
8726 {
8727 	return __dev_set_allmulti(dev, inc, true);
8728 }
8729 EXPORT_SYMBOL(dev_set_allmulti);
8730 
8731 /*
8732  *	Upload unicast and multicast address lists to device and
8733  *	configure RX filtering. When the device doesn't support unicast
8734  *	filtering it is put in promiscuous mode while unicast addresses
8735  *	are present.
8736  */
8737 void __dev_set_rx_mode(struct net_device *dev)
8738 {
8739 	const struct net_device_ops *ops = dev->netdev_ops;
8740 
8741 	/* dev_open will call this function so the list will stay sane. */
8742 	if (!(dev->flags&IFF_UP))
8743 		return;
8744 
8745 	if (!netif_device_present(dev))
8746 		return;
8747 
8748 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8749 		/* Unicast addresses changes may only happen under the rtnl,
8750 		 * therefore calling __dev_set_promiscuity here is safe.
8751 		 */
8752 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8753 			__dev_set_promiscuity(dev, 1, false);
8754 			dev->uc_promisc = true;
8755 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8756 			__dev_set_promiscuity(dev, -1, false);
8757 			dev->uc_promisc = false;
8758 		}
8759 	}
8760 
8761 	if (ops->ndo_set_rx_mode)
8762 		ops->ndo_set_rx_mode(dev);
8763 }
8764 
8765 void dev_set_rx_mode(struct net_device *dev)
8766 {
8767 	netif_addr_lock_bh(dev);
8768 	__dev_set_rx_mode(dev);
8769 	netif_addr_unlock_bh(dev);
8770 }
8771 
8772 /**
8773  *	dev_get_flags - get flags reported to userspace
8774  *	@dev: device
8775  *
8776  *	Get the combination of flag bits exported through APIs to userspace.
8777  */
8778 unsigned int dev_get_flags(const struct net_device *dev)
8779 {
8780 	unsigned int flags;
8781 
8782 	flags = (dev->flags & ~(IFF_PROMISC |
8783 				IFF_ALLMULTI |
8784 				IFF_RUNNING |
8785 				IFF_LOWER_UP |
8786 				IFF_DORMANT)) |
8787 		(dev->gflags & (IFF_PROMISC |
8788 				IFF_ALLMULTI));
8789 
8790 	if (netif_running(dev)) {
8791 		if (netif_oper_up(dev))
8792 			flags |= IFF_RUNNING;
8793 		if (netif_carrier_ok(dev))
8794 			flags |= IFF_LOWER_UP;
8795 		if (netif_dormant(dev))
8796 			flags |= IFF_DORMANT;
8797 	}
8798 
8799 	return flags;
8800 }
8801 EXPORT_SYMBOL(dev_get_flags);
8802 
8803 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8804 		       struct netlink_ext_ack *extack)
8805 {
8806 	unsigned int old_flags = dev->flags;
8807 	int ret;
8808 
8809 	ASSERT_RTNL();
8810 
8811 	/*
8812 	 *	Set the flags on our device.
8813 	 */
8814 
8815 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8816 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8817 			       IFF_AUTOMEDIA)) |
8818 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8819 				    IFF_ALLMULTI));
8820 
8821 	/*
8822 	 *	Load in the correct multicast list now the flags have changed.
8823 	 */
8824 
8825 	if ((old_flags ^ flags) & IFF_MULTICAST)
8826 		dev_change_rx_flags(dev, IFF_MULTICAST);
8827 
8828 	dev_set_rx_mode(dev);
8829 
8830 	/*
8831 	 *	Have we downed the interface. We handle IFF_UP ourselves
8832 	 *	according to user attempts to set it, rather than blindly
8833 	 *	setting it.
8834 	 */
8835 
8836 	ret = 0;
8837 	if ((old_flags ^ flags) & IFF_UP) {
8838 		if (old_flags & IFF_UP)
8839 			__dev_close(dev);
8840 		else
8841 			ret = __dev_open(dev, extack);
8842 	}
8843 
8844 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8845 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8846 		unsigned int old_flags = dev->flags;
8847 
8848 		dev->gflags ^= IFF_PROMISC;
8849 
8850 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8851 			if (dev->flags != old_flags)
8852 				dev_set_rx_mode(dev);
8853 	}
8854 
8855 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8856 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8857 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8858 	 */
8859 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8860 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8861 
8862 		dev->gflags ^= IFF_ALLMULTI;
8863 		__dev_set_allmulti(dev, inc, false);
8864 	}
8865 
8866 	return ret;
8867 }
8868 
8869 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8870 			unsigned int gchanges)
8871 {
8872 	unsigned int changes = dev->flags ^ old_flags;
8873 
8874 	if (gchanges)
8875 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8876 
8877 	if (changes & IFF_UP) {
8878 		if (dev->flags & IFF_UP)
8879 			call_netdevice_notifiers(NETDEV_UP, dev);
8880 		else
8881 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8882 	}
8883 
8884 	if (dev->flags & IFF_UP &&
8885 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8886 		struct netdev_notifier_change_info change_info = {
8887 			.info = {
8888 				.dev = dev,
8889 			},
8890 			.flags_changed = changes,
8891 		};
8892 
8893 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8894 	}
8895 }
8896 
8897 /**
8898  *	dev_change_flags - change device settings
8899  *	@dev: device
8900  *	@flags: device state flags
8901  *	@extack: netlink extended ack
8902  *
8903  *	Change settings on device based state flags. The flags are
8904  *	in the userspace exported format.
8905  */
8906 int dev_change_flags(struct net_device *dev, unsigned int flags,
8907 		     struct netlink_ext_ack *extack)
8908 {
8909 	int ret;
8910 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8911 
8912 	ret = __dev_change_flags(dev, flags, extack);
8913 	if (ret < 0)
8914 		return ret;
8915 
8916 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8917 	__dev_notify_flags(dev, old_flags, changes);
8918 	return ret;
8919 }
8920 EXPORT_SYMBOL(dev_change_flags);
8921 
8922 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8923 {
8924 	const struct net_device_ops *ops = dev->netdev_ops;
8925 
8926 	if (ops->ndo_change_mtu)
8927 		return ops->ndo_change_mtu(dev, new_mtu);
8928 
8929 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8930 	WRITE_ONCE(dev->mtu, new_mtu);
8931 	return 0;
8932 }
8933 EXPORT_SYMBOL(__dev_set_mtu);
8934 
8935 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8936 		     struct netlink_ext_ack *extack)
8937 {
8938 	/* MTU must be positive, and in range */
8939 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8940 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8941 		return -EINVAL;
8942 	}
8943 
8944 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8945 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8946 		return -EINVAL;
8947 	}
8948 	return 0;
8949 }
8950 
8951 /**
8952  *	dev_set_mtu_ext - Change maximum transfer unit
8953  *	@dev: device
8954  *	@new_mtu: new transfer unit
8955  *	@extack: netlink extended ack
8956  *
8957  *	Change the maximum transfer size of the network device.
8958  */
8959 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8960 		    struct netlink_ext_ack *extack)
8961 {
8962 	int err, orig_mtu;
8963 
8964 	if (new_mtu == dev->mtu)
8965 		return 0;
8966 
8967 	err = dev_validate_mtu(dev, new_mtu, extack);
8968 	if (err)
8969 		return err;
8970 
8971 	if (!netif_device_present(dev))
8972 		return -ENODEV;
8973 
8974 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8975 	err = notifier_to_errno(err);
8976 	if (err)
8977 		return err;
8978 
8979 	orig_mtu = dev->mtu;
8980 	err = __dev_set_mtu(dev, new_mtu);
8981 
8982 	if (!err) {
8983 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8984 						   orig_mtu);
8985 		err = notifier_to_errno(err);
8986 		if (err) {
8987 			/* setting mtu back and notifying everyone again,
8988 			 * so that they have a chance to revert changes.
8989 			 */
8990 			__dev_set_mtu(dev, orig_mtu);
8991 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8992 						     new_mtu);
8993 		}
8994 	}
8995 	return err;
8996 }
8997 
8998 int dev_set_mtu(struct net_device *dev, int new_mtu)
8999 {
9000 	struct netlink_ext_ack extack;
9001 	int err;
9002 
9003 	memset(&extack, 0, sizeof(extack));
9004 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
9005 	if (err && extack._msg)
9006 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
9007 	return err;
9008 }
9009 EXPORT_SYMBOL(dev_set_mtu);
9010 
9011 /**
9012  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
9013  *	@dev: device
9014  *	@new_len: new tx queue length
9015  */
9016 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9017 {
9018 	unsigned int orig_len = dev->tx_queue_len;
9019 	int res;
9020 
9021 	if (new_len != (unsigned int)new_len)
9022 		return -ERANGE;
9023 
9024 	if (new_len != orig_len) {
9025 		dev->tx_queue_len = new_len;
9026 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9027 		res = notifier_to_errno(res);
9028 		if (res)
9029 			goto err_rollback;
9030 		res = dev_qdisc_change_tx_queue_len(dev);
9031 		if (res)
9032 			goto err_rollback;
9033 	}
9034 
9035 	return 0;
9036 
9037 err_rollback:
9038 	netdev_err(dev, "refused to change device tx_queue_len\n");
9039 	dev->tx_queue_len = orig_len;
9040 	return res;
9041 }
9042 
9043 /**
9044  *	dev_set_group - Change group this device belongs to
9045  *	@dev: device
9046  *	@new_group: group this device should belong to
9047  */
9048 void dev_set_group(struct net_device *dev, int new_group)
9049 {
9050 	dev->group = new_group;
9051 }
9052 EXPORT_SYMBOL(dev_set_group);
9053 
9054 /**
9055  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9056  *	@dev: device
9057  *	@addr: new address
9058  *	@extack: netlink extended ack
9059  */
9060 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9061 			      struct netlink_ext_ack *extack)
9062 {
9063 	struct netdev_notifier_pre_changeaddr_info info = {
9064 		.info.dev = dev,
9065 		.info.extack = extack,
9066 		.dev_addr = addr,
9067 	};
9068 	int rc;
9069 
9070 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9071 	return notifier_to_errno(rc);
9072 }
9073 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9074 
9075 /**
9076  *	dev_set_mac_address - Change Media Access Control Address
9077  *	@dev: device
9078  *	@sa: new address
9079  *	@extack: netlink extended ack
9080  *
9081  *	Change the hardware (MAC) address of the device
9082  */
9083 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9084 			struct netlink_ext_ack *extack)
9085 {
9086 	const struct net_device_ops *ops = dev->netdev_ops;
9087 	int err;
9088 
9089 	if (!ops->ndo_set_mac_address)
9090 		return -EOPNOTSUPP;
9091 	if (sa->sa_family != dev->type)
9092 		return -EINVAL;
9093 	if (!netif_device_present(dev))
9094 		return -ENODEV;
9095 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9096 	if (err)
9097 		return err;
9098 	err = ops->ndo_set_mac_address(dev, sa);
9099 	if (err)
9100 		return err;
9101 	dev->addr_assign_type = NET_ADDR_SET;
9102 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9103 	add_device_randomness(dev->dev_addr, dev->addr_len);
9104 	return 0;
9105 }
9106 EXPORT_SYMBOL(dev_set_mac_address);
9107 
9108 static DECLARE_RWSEM(dev_addr_sem);
9109 
9110 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9111 			     struct netlink_ext_ack *extack)
9112 {
9113 	int ret;
9114 
9115 	down_write(&dev_addr_sem);
9116 	ret = dev_set_mac_address(dev, sa, extack);
9117 	up_write(&dev_addr_sem);
9118 	return ret;
9119 }
9120 EXPORT_SYMBOL(dev_set_mac_address_user);
9121 
9122 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9123 {
9124 	size_t size = sizeof(sa->sa_data);
9125 	struct net_device *dev;
9126 	int ret = 0;
9127 
9128 	down_read(&dev_addr_sem);
9129 	rcu_read_lock();
9130 
9131 	dev = dev_get_by_name_rcu(net, dev_name);
9132 	if (!dev) {
9133 		ret = -ENODEV;
9134 		goto unlock;
9135 	}
9136 	if (!dev->addr_len)
9137 		memset(sa->sa_data, 0, size);
9138 	else
9139 		memcpy(sa->sa_data, dev->dev_addr,
9140 		       min_t(size_t, size, dev->addr_len));
9141 	sa->sa_family = dev->type;
9142 
9143 unlock:
9144 	rcu_read_unlock();
9145 	up_read(&dev_addr_sem);
9146 	return ret;
9147 }
9148 EXPORT_SYMBOL(dev_get_mac_address);
9149 
9150 /**
9151  *	dev_change_carrier - Change device carrier
9152  *	@dev: device
9153  *	@new_carrier: new value
9154  *
9155  *	Change device carrier
9156  */
9157 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9158 {
9159 	const struct net_device_ops *ops = dev->netdev_ops;
9160 
9161 	if (!ops->ndo_change_carrier)
9162 		return -EOPNOTSUPP;
9163 	if (!netif_device_present(dev))
9164 		return -ENODEV;
9165 	return ops->ndo_change_carrier(dev, new_carrier);
9166 }
9167 EXPORT_SYMBOL(dev_change_carrier);
9168 
9169 /**
9170  *	dev_get_phys_port_id - Get device physical port ID
9171  *	@dev: device
9172  *	@ppid: port ID
9173  *
9174  *	Get device physical port ID
9175  */
9176 int dev_get_phys_port_id(struct net_device *dev,
9177 			 struct netdev_phys_item_id *ppid)
9178 {
9179 	const struct net_device_ops *ops = dev->netdev_ops;
9180 
9181 	if (!ops->ndo_get_phys_port_id)
9182 		return -EOPNOTSUPP;
9183 	return ops->ndo_get_phys_port_id(dev, ppid);
9184 }
9185 EXPORT_SYMBOL(dev_get_phys_port_id);
9186 
9187 /**
9188  *	dev_get_phys_port_name - Get device physical port name
9189  *	@dev: device
9190  *	@name: port name
9191  *	@len: limit of bytes to copy to name
9192  *
9193  *	Get device physical port name
9194  */
9195 int dev_get_phys_port_name(struct net_device *dev,
9196 			   char *name, size_t len)
9197 {
9198 	const struct net_device_ops *ops = dev->netdev_ops;
9199 	int err;
9200 
9201 	if (ops->ndo_get_phys_port_name) {
9202 		err = ops->ndo_get_phys_port_name(dev, name, len);
9203 		if (err != -EOPNOTSUPP)
9204 			return err;
9205 	}
9206 	return devlink_compat_phys_port_name_get(dev, name, len);
9207 }
9208 EXPORT_SYMBOL(dev_get_phys_port_name);
9209 
9210 /**
9211  *	dev_get_port_parent_id - Get the device's port parent identifier
9212  *	@dev: network device
9213  *	@ppid: pointer to a storage for the port's parent identifier
9214  *	@recurse: allow/disallow recursion to lower devices
9215  *
9216  *	Get the devices's port parent identifier
9217  */
9218 int dev_get_port_parent_id(struct net_device *dev,
9219 			   struct netdev_phys_item_id *ppid,
9220 			   bool recurse)
9221 {
9222 	const struct net_device_ops *ops = dev->netdev_ops;
9223 	struct netdev_phys_item_id first = { };
9224 	struct net_device *lower_dev;
9225 	struct list_head *iter;
9226 	int err;
9227 
9228 	if (ops->ndo_get_port_parent_id) {
9229 		err = ops->ndo_get_port_parent_id(dev, ppid);
9230 		if (err != -EOPNOTSUPP)
9231 			return err;
9232 	}
9233 
9234 	err = devlink_compat_switch_id_get(dev, ppid);
9235 	if (!err || err != -EOPNOTSUPP)
9236 		return err;
9237 
9238 	if (!recurse)
9239 		return -EOPNOTSUPP;
9240 
9241 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
9242 		err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9243 		if (err)
9244 			break;
9245 		if (!first.id_len)
9246 			first = *ppid;
9247 		else if (memcmp(&first, ppid, sizeof(*ppid)))
9248 			return -EOPNOTSUPP;
9249 	}
9250 
9251 	return err;
9252 }
9253 EXPORT_SYMBOL(dev_get_port_parent_id);
9254 
9255 /**
9256  *	netdev_port_same_parent_id - Indicate if two network devices have
9257  *	the same port parent identifier
9258  *	@a: first network device
9259  *	@b: second network device
9260  */
9261 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9262 {
9263 	struct netdev_phys_item_id a_id = { };
9264 	struct netdev_phys_item_id b_id = { };
9265 
9266 	if (dev_get_port_parent_id(a, &a_id, true) ||
9267 	    dev_get_port_parent_id(b, &b_id, true))
9268 		return false;
9269 
9270 	return netdev_phys_item_id_same(&a_id, &b_id);
9271 }
9272 EXPORT_SYMBOL(netdev_port_same_parent_id);
9273 
9274 /**
9275  *	dev_change_proto_down - update protocol port state information
9276  *	@dev: device
9277  *	@proto_down: new value
9278  *
9279  *	This info can be used by switch drivers to set the phys state of the
9280  *	port.
9281  */
9282 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9283 {
9284 	const struct net_device_ops *ops = dev->netdev_ops;
9285 
9286 	if (!ops->ndo_change_proto_down)
9287 		return -EOPNOTSUPP;
9288 	if (!netif_device_present(dev))
9289 		return -ENODEV;
9290 	return ops->ndo_change_proto_down(dev, proto_down);
9291 }
9292 EXPORT_SYMBOL(dev_change_proto_down);
9293 
9294 /**
9295  *	dev_change_proto_down_generic - generic implementation for
9296  * 	ndo_change_proto_down that sets carrier according to
9297  * 	proto_down.
9298  *
9299  *	@dev: device
9300  *	@proto_down: new value
9301  */
9302 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9303 {
9304 	if (proto_down)
9305 		netif_carrier_off(dev);
9306 	else
9307 		netif_carrier_on(dev);
9308 	dev->proto_down = proto_down;
9309 	return 0;
9310 }
9311 EXPORT_SYMBOL(dev_change_proto_down_generic);
9312 
9313 /**
9314  *	dev_change_proto_down_reason - proto down reason
9315  *
9316  *	@dev: device
9317  *	@mask: proto down mask
9318  *	@value: proto down value
9319  */
9320 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9321 				  u32 value)
9322 {
9323 	int b;
9324 
9325 	if (!mask) {
9326 		dev->proto_down_reason = value;
9327 	} else {
9328 		for_each_set_bit(b, &mask, 32) {
9329 			if (value & (1 << b))
9330 				dev->proto_down_reason |= BIT(b);
9331 			else
9332 				dev->proto_down_reason &= ~BIT(b);
9333 		}
9334 	}
9335 }
9336 EXPORT_SYMBOL(dev_change_proto_down_reason);
9337 
9338 struct bpf_xdp_link {
9339 	struct bpf_link link;
9340 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9341 	int flags;
9342 };
9343 
9344 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9345 {
9346 	if (flags & XDP_FLAGS_HW_MODE)
9347 		return XDP_MODE_HW;
9348 	if (flags & XDP_FLAGS_DRV_MODE)
9349 		return XDP_MODE_DRV;
9350 	if (flags & XDP_FLAGS_SKB_MODE)
9351 		return XDP_MODE_SKB;
9352 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9353 }
9354 
9355 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9356 {
9357 	switch (mode) {
9358 	case XDP_MODE_SKB:
9359 		return generic_xdp_install;
9360 	case XDP_MODE_DRV:
9361 	case XDP_MODE_HW:
9362 		return dev->netdev_ops->ndo_bpf;
9363 	default:
9364 		return NULL;
9365 	}
9366 }
9367 
9368 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9369 					 enum bpf_xdp_mode mode)
9370 {
9371 	return dev->xdp_state[mode].link;
9372 }
9373 
9374 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9375 				     enum bpf_xdp_mode mode)
9376 {
9377 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9378 
9379 	if (link)
9380 		return link->link.prog;
9381 	return dev->xdp_state[mode].prog;
9382 }
9383 
9384 static u8 dev_xdp_prog_count(struct net_device *dev)
9385 {
9386 	u8 count = 0;
9387 	int i;
9388 
9389 	for (i = 0; i < __MAX_XDP_MODE; i++)
9390 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9391 			count++;
9392 	return count;
9393 }
9394 
9395 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9396 {
9397 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9398 
9399 	return prog ? prog->aux->id : 0;
9400 }
9401 
9402 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9403 			     struct bpf_xdp_link *link)
9404 {
9405 	dev->xdp_state[mode].link = link;
9406 	dev->xdp_state[mode].prog = NULL;
9407 }
9408 
9409 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9410 			     struct bpf_prog *prog)
9411 {
9412 	dev->xdp_state[mode].link = NULL;
9413 	dev->xdp_state[mode].prog = prog;
9414 }
9415 
9416 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9417 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9418 			   u32 flags, struct bpf_prog *prog)
9419 {
9420 	struct netdev_bpf xdp;
9421 	int err;
9422 
9423 	memset(&xdp, 0, sizeof(xdp));
9424 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9425 	xdp.extack = extack;
9426 	xdp.flags = flags;
9427 	xdp.prog = prog;
9428 
9429 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9430 	 * "moved" into driver), so they don't increment it on their own, but
9431 	 * they do decrement refcnt when program is detached or replaced.
9432 	 * Given net_device also owns link/prog, we need to bump refcnt here
9433 	 * to prevent drivers from underflowing it.
9434 	 */
9435 	if (prog)
9436 		bpf_prog_inc(prog);
9437 	err = bpf_op(dev, &xdp);
9438 	if (err) {
9439 		if (prog)
9440 			bpf_prog_put(prog);
9441 		return err;
9442 	}
9443 
9444 	if (mode != XDP_MODE_HW)
9445 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9446 
9447 	return 0;
9448 }
9449 
9450 static void dev_xdp_uninstall(struct net_device *dev)
9451 {
9452 	struct bpf_xdp_link *link;
9453 	struct bpf_prog *prog;
9454 	enum bpf_xdp_mode mode;
9455 	bpf_op_t bpf_op;
9456 
9457 	ASSERT_RTNL();
9458 
9459 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9460 		prog = dev_xdp_prog(dev, mode);
9461 		if (!prog)
9462 			continue;
9463 
9464 		bpf_op = dev_xdp_bpf_op(dev, mode);
9465 		if (!bpf_op)
9466 			continue;
9467 
9468 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9469 
9470 		/* auto-detach link from net device */
9471 		link = dev_xdp_link(dev, mode);
9472 		if (link)
9473 			link->dev = NULL;
9474 		else
9475 			bpf_prog_put(prog);
9476 
9477 		dev_xdp_set_link(dev, mode, NULL);
9478 	}
9479 }
9480 
9481 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9482 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9483 			  struct bpf_prog *old_prog, u32 flags)
9484 {
9485 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9486 	struct bpf_prog *cur_prog;
9487 	enum bpf_xdp_mode mode;
9488 	bpf_op_t bpf_op;
9489 	int err;
9490 
9491 	ASSERT_RTNL();
9492 
9493 	/* either link or prog attachment, never both */
9494 	if (link && (new_prog || old_prog))
9495 		return -EINVAL;
9496 	/* link supports only XDP mode flags */
9497 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9498 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9499 		return -EINVAL;
9500 	}
9501 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9502 	if (num_modes > 1) {
9503 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9504 		return -EINVAL;
9505 	}
9506 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9507 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9508 		NL_SET_ERR_MSG(extack,
9509 			       "More than one program loaded, unset mode is ambiguous");
9510 		return -EINVAL;
9511 	}
9512 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9513 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9514 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9515 		return -EINVAL;
9516 	}
9517 
9518 	mode = dev_xdp_mode(dev, flags);
9519 	/* can't replace attached link */
9520 	if (dev_xdp_link(dev, mode)) {
9521 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9522 		return -EBUSY;
9523 	}
9524 
9525 	cur_prog = dev_xdp_prog(dev, mode);
9526 	/* can't replace attached prog with link */
9527 	if (link && cur_prog) {
9528 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9529 		return -EBUSY;
9530 	}
9531 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9532 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9533 		return -EEXIST;
9534 	}
9535 
9536 	/* put effective new program into new_prog */
9537 	if (link)
9538 		new_prog = link->link.prog;
9539 
9540 	if (new_prog) {
9541 		bool offload = mode == XDP_MODE_HW;
9542 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9543 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9544 
9545 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9546 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9547 			return -EBUSY;
9548 		}
9549 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9550 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9551 			return -EEXIST;
9552 		}
9553 		if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9554 			NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9555 			return -EINVAL;
9556 		}
9557 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9558 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9559 			return -EINVAL;
9560 		}
9561 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9562 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9563 			return -EINVAL;
9564 		}
9565 	}
9566 
9567 	/* don't call drivers if the effective program didn't change */
9568 	if (new_prog != cur_prog) {
9569 		bpf_op = dev_xdp_bpf_op(dev, mode);
9570 		if (!bpf_op) {
9571 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9572 			return -EOPNOTSUPP;
9573 		}
9574 
9575 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9576 		if (err)
9577 			return err;
9578 	}
9579 
9580 	if (link)
9581 		dev_xdp_set_link(dev, mode, link);
9582 	else
9583 		dev_xdp_set_prog(dev, mode, new_prog);
9584 	if (cur_prog)
9585 		bpf_prog_put(cur_prog);
9586 
9587 	return 0;
9588 }
9589 
9590 static int dev_xdp_attach_link(struct net_device *dev,
9591 			       struct netlink_ext_ack *extack,
9592 			       struct bpf_xdp_link *link)
9593 {
9594 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9595 }
9596 
9597 static int dev_xdp_detach_link(struct net_device *dev,
9598 			       struct netlink_ext_ack *extack,
9599 			       struct bpf_xdp_link *link)
9600 {
9601 	enum bpf_xdp_mode mode;
9602 	bpf_op_t bpf_op;
9603 
9604 	ASSERT_RTNL();
9605 
9606 	mode = dev_xdp_mode(dev, link->flags);
9607 	if (dev_xdp_link(dev, mode) != link)
9608 		return -EINVAL;
9609 
9610 	bpf_op = dev_xdp_bpf_op(dev, mode);
9611 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9612 	dev_xdp_set_link(dev, mode, NULL);
9613 	return 0;
9614 }
9615 
9616 static void bpf_xdp_link_release(struct bpf_link *link)
9617 {
9618 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9619 
9620 	rtnl_lock();
9621 
9622 	/* if racing with net_device's tear down, xdp_link->dev might be
9623 	 * already NULL, in which case link was already auto-detached
9624 	 */
9625 	if (xdp_link->dev) {
9626 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9627 		xdp_link->dev = NULL;
9628 	}
9629 
9630 	rtnl_unlock();
9631 }
9632 
9633 static int bpf_xdp_link_detach(struct bpf_link *link)
9634 {
9635 	bpf_xdp_link_release(link);
9636 	return 0;
9637 }
9638 
9639 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9640 {
9641 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9642 
9643 	kfree(xdp_link);
9644 }
9645 
9646 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9647 				     struct seq_file *seq)
9648 {
9649 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9650 	u32 ifindex = 0;
9651 
9652 	rtnl_lock();
9653 	if (xdp_link->dev)
9654 		ifindex = xdp_link->dev->ifindex;
9655 	rtnl_unlock();
9656 
9657 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9658 }
9659 
9660 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9661 				       struct bpf_link_info *info)
9662 {
9663 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9664 	u32 ifindex = 0;
9665 
9666 	rtnl_lock();
9667 	if (xdp_link->dev)
9668 		ifindex = xdp_link->dev->ifindex;
9669 	rtnl_unlock();
9670 
9671 	info->xdp.ifindex = ifindex;
9672 	return 0;
9673 }
9674 
9675 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9676 			       struct bpf_prog *old_prog)
9677 {
9678 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9679 	enum bpf_xdp_mode mode;
9680 	bpf_op_t bpf_op;
9681 	int err = 0;
9682 
9683 	rtnl_lock();
9684 
9685 	/* link might have been auto-released already, so fail */
9686 	if (!xdp_link->dev) {
9687 		err = -ENOLINK;
9688 		goto out_unlock;
9689 	}
9690 
9691 	if (old_prog && link->prog != old_prog) {
9692 		err = -EPERM;
9693 		goto out_unlock;
9694 	}
9695 	old_prog = link->prog;
9696 	if (old_prog == new_prog) {
9697 		/* no-op, don't disturb drivers */
9698 		bpf_prog_put(new_prog);
9699 		goto out_unlock;
9700 	}
9701 
9702 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9703 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9704 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9705 			      xdp_link->flags, new_prog);
9706 	if (err)
9707 		goto out_unlock;
9708 
9709 	old_prog = xchg(&link->prog, new_prog);
9710 	bpf_prog_put(old_prog);
9711 
9712 out_unlock:
9713 	rtnl_unlock();
9714 	return err;
9715 }
9716 
9717 static const struct bpf_link_ops bpf_xdp_link_lops = {
9718 	.release = bpf_xdp_link_release,
9719 	.dealloc = bpf_xdp_link_dealloc,
9720 	.detach = bpf_xdp_link_detach,
9721 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9722 	.fill_link_info = bpf_xdp_link_fill_link_info,
9723 	.update_prog = bpf_xdp_link_update,
9724 };
9725 
9726 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9727 {
9728 	struct net *net = current->nsproxy->net_ns;
9729 	struct bpf_link_primer link_primer;
9730 	struct bpf_xdp_link *link;
9731 	struct net_device *dev;
9732 	int err, fd;
9733 
9734 	rtnl_lock();
9735 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9736 	if (!dev) {
9737 		rtnl_unlock();
9738 		return -EINVAL;
9739 	}
9740 
9741 	link = kzalloc(sizeof(*link), GFP_USER);
9742 	if (!link) {
9743 		err = -ENOMEM;
9744 		goto unlock;
9745 	}
9746 
9747 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9748 	link->dev = dev;
9749 	link->flags = attr->link_create.flags;
9750 
9751 	err = bpf_link_prime(&link->link, &link_primer);
9752 	if (err) {
9753 		kfree(link);
9754 		goto unlock;
9755 	}
9756 
9757 	err = dev_xdp_attach_link(dev, NULL, link);
9758 	rtnl_unlock();
9759 
9760 	if (err) {
9761 		link->dev = NULL;
9762 		bpf_link_cleanup(&link_primer);
9763 		goto out_put_dev;
9764 	}
9765 
9766 	fd = bpf_link_settle(&link_primer);
9767 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9768 	dev_put(dev);
9769 	return fd;
9770 
9771 unlock:
9772 	rtnl_unlock();
9773 
9774 out_put_dev:
9775 	dev_put(dev);
9776 	return err;
9777 }
9778 
9779 /**
9780  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9781  *	@dev: device
9782  *	@extack: netlink extended ack
9783  *	@fd: new program fd or negative value to clear
9784  *	@expected_fd: old program fd that userspace expects to replace or clear
9785  *	@flags: xdp-related flags
9786  *
9787  *	Set or clear a bpf program for a device
9788  */
9789 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9790 		      int fd, int expected_fd, u32 flags)
9791 {
9792 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9793 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9794 	int err;
9795 
9796 	ASSERT_RTNL();
9797 
9798 	if (fd >= 0) {
9799 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9800 						 mode != XDP_MODE_SKB);
9801 		if (IS_ERR(new_prog))
9802 			return PTR_ERR(new_prog);
9803 	}
9804 
9805 	if (expected_fd >= 0) {
9806 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9807 						 mode != XDP_MODE_SKB);
9808 		if (IS_ERR(old_prog)) {
9809 			err = PTR_ERR(old_prog);
9810 			old_prog = NULL;
9811 			goto err_out;
9812 		}
9813 	}
9814 
9815 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9816 
9817 err_out:
9818 	if (err && new_prog)
9819 		bpf_prog_put(new_prog);
9820 	if (old_prog)
9821 		bpf_prog_put(old_prog);
9822 	return err;
9823 }
9824 
9825 /**
9826  *	dev_new_index	-	allocate an ifindex
9827  *	@net: the applicable net namespace
9828  *
9829  *	Returns a suitable unique value for a new device interface
9830  *	number.  The caller must hold the rtnl semaphore or the
9831  *	dev_base_lock to be sure it remains unique.
9832  */
9833 static int dev_new_index(struct net *net)
9834 {
9835 	int ifindex = net->ifindex;
9836 
9837 	for (;;) {
9838 		if (++ifindex <= 0)
9839 			ifindex = 1;
9840 		if (!__dev_get_by_index(net, ifindex))
9841 			return net->ifindex = ifindex;
9842 	}
9843 }
9844 
9845 /* Delayed registration/unregisteration */
9846 static LIST_HEAD(net_todo_list);
9847 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9848 
9849 static void net_set_todo(struct net_device *dev)
9850 {
9851 	list_add_tail(&dev->todo_list, &net_todo_list);
9852 	dev_net(dev)->dev_unreg_count++;
9853 }
9854 
9855 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9856 	struct net_device *upper, netdev_features_t features)
9857 {
9858 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9859 	netdev_features_t feature;
9860 	int feature_bit;
9861 
9862 	for_each_netdev_feature(upper_disables, feature_bit) {
9863 		feature = __NETIF_F_BIT(feature_bit);
9864 		if (!(upper->wanted_features & feature)
9865 		    && (features & feature)) {
9866 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9867 				   &feature, upper->name);
9868 			features &= ~feature;
9869 		}
9870 	}
9871 
9872 	return features;
9873 }
9874 
9875 static void netdev_sync_lower_features(struct net_device *upper,
9876 	struct net_device *lower, netdev_features_t features)
9877 {
9878 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9879 	netdev_features_t feature;
9880 	int feature_bit;
9881 
9882 	for_each_netdev_feature(upper_disables, feature_bit) {
9883 		feature = __NETIF_F_BIT(feature_bit);
9884 		if (!(features & feature) && (lower->features & feature)) {
9885 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9886 				   &feature, lower->name);
9887 			lower->wanted_features &= ~feature;
9888 			__netdev_update_features(lower);
9889 
9890 			if (unlikely(lower->features & feature))
9891 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9892 					    &feature, lower->name);
9893 			else
9894 				netdev_features_change(lower);
9895 		}
9896 	}
9897 }
9898 
9899 static netdev_features_t netdev_fix_features(struct net_device *dev,
9900 	netdev_features_t features)
9901 {
9902 	/* Fix illegal checksum combinations */
9903 	if ((features & NETIF_F_HW_CSUM) &&
9904 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9905 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9906 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9907 	}
9908 
9909 	/* TSO requires that SG is present as well. */
9910 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9911 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9912 		features &= ~NETIF_F_ALL_TSO;
9913 	}
9914 
9915 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9916 					!(features & NETIF_F_IP_CSUM)) {
9917 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9918 		features &= ~NETIF_F_TSO;
9919 		features &= ~NETIF_F_TSO_ECN;
9920 	}
9921 
9922 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9923 					 !(features & NETIF_F_IPV6_CSUM)) {
9924 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9925 		features &= ~NETIF_F_TSO6;
9926 	}
9927 
9928 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9929 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9930 		features &= ~NETIF_F_TSO_MANGLEID;
9931 
9932 	/* TSO ECN requires that TSO is present as well. */
9933 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9934 		features &= ~NETIF_F_TSO_ECN;
9935 
9936 	/* Software GSO depends on SG. */
9937 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9938 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9939 		features &= ~NETIF_F_GSO;
9940 	}
9941 
9942 	/* GSO partial features require GSO partial be set */
9943 	if ((features & dev->gso_partial_features) &&
9944 	    !(features & NETIF_F_GSO_PARTIAL)) {
9945 		netdev_dbg(dev,
9946 			   "Dropping partially supported GSO features since no GSO partial.\n");
9947 		features &= ~dev->gso_partial_features;
9948 	}
9949 
9950 	if (!(features & NETIF_F_RXCSUM)) {
9951 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9952 		 * successfully merged by hardware must also have the
9953 		 * checksum verified by hardware.  If the user does not
9954 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9955 		 */
9956 		if (features & NETIF_F_GRO_HW) {
9957 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9958 			features &= ~NETIF_F_GRO_HW;
9959 		}
9960 	}
9961 
9962 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9963 	if (features & NETIF_F_RXFCS) {
9964 		if (features & NETIF_F_LRO) {
9965 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9966 			features &= ~NETIF_F_LRO;
9967 		}
9968 
9969 		if (features & NETIF_F_GRO_HW) {
9970 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9971 			features &= ~NETIF_F_GRO_HW;
9972 		}
9973 	}
9974 
9975 	if (features & NETIF_F_HW_TLS_TX) {
9976 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9977 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9978 		bool hw_csum = features & NETIF_F_HW_CSUM;
9979 
9980 		if (!ip_csum && !hw_csum) {
9981 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9982 			features &= ~NETIF_F_HW_TLS_TX;
9983 		}
9984 	}
9985 
9986 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9987 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9988 		features &= ~NETIF_F_HW_TLS_RX;
9989 	}
9990 
9991 	return features;
9992 }
9993 
9994 int __netdev_update_features(struct net_device *dev)
9995 {
9996 	struct net_device *upper, *lower;
9997 	netdev_features_t features;
9998 	struct list_head *iter;
9999 	int err = -1;
10000 
10001 	ASSERT_RTNL();
10002 
10003 	features = netdev_get_wanted_features(dev);
10004 
10005 	if (dev->netdev_ops->ndo_fix_features)
10006 		features = dev->netdev_ops->ndo_fix_features(dev, features);
10007 
10008 	/* driver might be less strict about feature dependencies */
10009 	features = netdev_fix_features(dev, features);
10010 
10011 	/* some features can't be enabled if they're off on an upper device */
10012 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
10013 		features = netdev_sync_upper_features(dev, upper, features);
10014 
10015 	if (dev->features == features)
10016 		goto sync_lower;
10017 
10018 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10019 		&dev->features, &features);
10020 
10021 	if (dev->netdev_ops->ndo_set_features)
10022 		err = dev->netdev_ops->ndo_set_features(dev, features);
10023 	else
10024 		err = 0;
10025 
10026 	if (unlikely(err < 0)) {
10027 		netdev_err(dev,
10028 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
10029 			err, &features, &dev->features);
10030 		/* return non-0 since some features might have changed and
10031 		 * it's better to fire a spurious notification than miss it
10032 		 */
10033 		return -1;
10034 	}
10035 
10036 sync_lower:
10037 	/* some features must be disabled on lower devices when disabled
10038 	 * on an upper device (think: bonding master or bridge)
10039 	 */
10040 	netdev_for_each_lower_dev(dev, lower, iter)
10041 		netdev_sync_lower_features(dev, lower, features);
10042 
10043 	if (!err) {
10044 		netdev_features_t diff = features ^ dev->features;
10045 
10046 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10047 			/* udp_tunnel_{get,drop}_rx_info both need
10048 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10049 			 * device, or they won't do anything.
10050 			 * Thus we need to update dev->features
10051 			 * *before* calling udp_tunnel_get_rx_info,
10052 			 * but *after* calling udp_tunnel_drop_rx_info.
10053 			 */
10054 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10055 				dev->features = features;
10056 				udp_tunnel_get_rx_info(dev);
10057 			} else {
10058 				udp_tunnel_drop_rx_info(dev);
10059 			}
10060 		}
10061 
10062 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10063 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10064 				dev->features = features;
10065 				err |= vlan_get_rx_ctag_filter_info(dev);
10066 			} else {
10067 				vlan_drop_rx_ctag_filter_info(dev);
10068 			}
10069 		}
10070 
10071 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10072 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10073 				dev->features = features;
10074 				err |= vlan_get_rx_stag_filter_info(dev);
10075 			} else {
10076 				vlan_drop_rx_stag_filter_info(dev);
10077 			}
10078 		}
10079 
10080 		dev->features = features;
10081 	}
10082 
10083 	return err < 0 ? 0 : 1;
10084 }
10085 
10086 /**
10087  *	netdev_update_features - recalculate device features
10088  *	@dev: the device to check
10089  *
10090  *	Recalculate dev->features set and send notifications if it
10091  *	has changed. Should be called after driver or hardware dependent
10092  *	conditions might have changed that influence the features.
10093  */
10094 void netdev_update_features(struct net_device *dev)
10095 {
10096 	if (__netdev_update_features(dev))
10097 		netdev_features_change(dev);
10098 }
10099 EXPORT_SYMBOL(netdev_update_features);
10100 
10101 /**
10102  *	netdev_change_features - recalculate device features
10103  *	@dev: the device to check
10104  *
10105  *	Recalculate dev->features set and send notifications even
10106  *	if they have not changed. Should be called instead of
10107  *	netdev_update_features() if also dev->vlan_features might
10108  *	have changed to allow the changes to be propagated to stacked
10109  *	VLAN devices.
10110  */
10111 void netdev_change_features(struct net_device *dev)
10112 {
10113 	__netdev_update_features(dev);
10114 	netdev_features_change(dev);
10115 }
10116 EXPORT_SYMBOL(netdev_change_features);
10117 
10118 /**
10119  *	netif_stacked_transfer_operstate -	transfer operstate
10120  *	@rootdev: the root or lower level device to transfer state from
10121  *	@dev: the device to transfer operstate to
10122  *
10123  *	Transfer operational state from root to device. This is normally
10124  *	called when a stacking relationship exists between the root
10125  *	device and the device(a leaf device).
10126  */
10127 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10128 					struct net_device *dev)
10129 {
10130 	if (rootdev->operstate == IF_OPER_DORMANT)
10131 		netif_dormant_on(dev);
10132 	else
10133 		netif_dormant_off(dev);
10134 
10135 	if (rootdev->operstate == IF_OPER_TESTING)
10136 		netif_testing_on(dev);
10137 	else
10138 		netif_testing_off(dev);
10139 
10140 	if (netif_carrier_ok(rootdev))
10141 		netif_carrier_on(dev);
10142 	else
10143 		netif_carrier_off(dev);
10144 }
10145 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10146 
10147 static int netif_alloc_rx_queues(struct net_device *dev)
10148 {
10149 	unsigned int i, count = dev->num_rx_queues;
10150 	struct netdev_rx_queue *rx;
10151 	size_t sz = count * sizeof(*rx);
10152 	int err = 0;
10153 
10154 	BUG_ON(count < 1);
10155 
10156 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10157 	if (!rx)
10158 		return -ENOMEM;
10159 
10160 	dev->_rx = rx;
10161 
10162 	for (i = 0; i < count; i++) {
10163 		rx[i].dev = dev;
10164 
10165 		/* XDP RX-queue setup */
10166 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10167 		if (err < 0)
10168 			goto err_rxq_info;
10169 	}
10170 	return 0;
10171 
10172 err_rxq_info:
10173 	/* Rollback successful reg's and free other resources */
10174 	while (i--)
10175 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10176 	kvfree(dev->_rx);
10177 	dev->_rx = NULL;
10178 	return err;
10179 }
10180 
10181 static void netif_free_rx_queues(struct net_device *dev)
10182 {
10183 	unsigned int i, count = dev->num_rx_queues;
10184 
10185 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10186 	if (!dev->_rx)
10187 		return;
10188 
10189 	for (i = 0; i < count; i++)
10190 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10191 
10192 	kvfree(dev->_rx);
10193 }
10194 
10195 static void netdev_init_one_queue(struct net_device *dev,
10196 				  struct netdev_queue *queue, void *_unused)
10197 {
10198 	/* Initialize queue lock */
10199 	spin_lock_init(&queue->_xmit_lock);
10200 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10201 	queue->xmit_lock_owner = -1;
10202 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10203 	queue->dev = dev;
10204 #ifdef CONFIG_BQL
10205 	dql_init(&queue->dql, HZ);
10206 #endif
10207 }
10208 
10209 static void netif_free_tx_queues(struct net_device *dev)
10210 {
10211 	kvfree(dev->_tx);
10212 }
10213 
10214 static int netif_alloc_netdev_queues(struct net_device *dev)
10215 {
10216 	unsigned int count = dev->num_tx_queues;
10217 	struct netdev_queue *tx;
10218 	size_t sz = count * sizeof(*tx);
10219 
10220 	if (count < 1 || count > 0xffff)
10221 		return -EINVAL;
10222 
10223 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10224 	if (!tx)
10225 		return -ENOMEM;
10226 
10227 	dev->_tx = tx;
10228 
10229 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10230 	spin_lock_init(&dev->tx_global_lock);
10231 
10232 	return 0;
10233 }
10234 
10235 void netif_tx_stop_all_queues(struct net_device *dev)
10236 {
10237 	unsigned int i;
10238 
10239 	for (i = 0; i < dev->num_tx_queues; i++) {
10240 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10241 
10242 		netif_tx_stop_queue(txq);
10243 	}
10244 }
10245 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10246 
10247 /**
10248  *	register_netdevice	- register a network device
10249  *	@dev: device to register
10250  *
10251  *	Take a completed network device structure and add it to the kernel
10252  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10253  *	chain. 0 is returned on success. A negative errno code is returned
10254  *	on a failure to set up the device, or if the name is a duplicate.
10255  *
10256  *	Callers must hold the rtnl semaphore. You may want
10257  *	register_netdev() instead of this.
10258  *
10259  *	BUGS:
10260  *	The locking appears insufficient to guarantee two parallel registers
10261  *	will not get the same name.
10262  */
10263 
10264 int register_netdevice(struct net_device *dev)
10265 {
10266 	int ret;
10267 	struct net *net = dev_net(dev);
10268 
10269 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10270 		     NETDEV_FEATURE_COUNT);
10271 	BUG_ON(dev_boot_phase);
10272 	ASSERT_RTNL();
10273 
10274 	might_sleep();
10275 
10276 	/* When net_device's are persistent, this will be fatal. */
10277 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10278 	BUG_ON(!net);
10279 
10280 	ret = ethtool_check_ops(dev->ethtool_ops);
10281 	if (ret)
10282 		return ret;
10283 
10284 	spin_lock_init(&dev->addr_list_lock);
10285 	netdev_set_addr_lockdep_class(dev);
10286 
10287 	ret = dev_get_valid_name(net, dev, dev->name);
10288 	if (ret < 0)
10289 		goto out;
10290 
10291 	ret = -ENOMEM;
10292 	dev->name_node = netdev_name_node_head_alloc(dev);
10293 	if (!dev->name_node)
10294 		goto out;
10295 
10296 	/* Init, if this function is available */
10297 	if (dev->netdev_ops->ndo_init) {
10298 		ret = dev->netdev_ops->ndo_init(dev);
10299 		if (ret) {
10300 			if (ret > 0)
10301 				ret = -EIO;
10302 			goto err_free_name;
10303 		}
10304 	}
10305 
10306 	if (((dev->hw_features | dev->features) &
10307 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
10308 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10309 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10310 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10311 		ret = -EINVAL;
10312 		goto err_uninit;
10313 	}
10314 
10315 	ret = -EBUSY;
10316 	if (!dev->ifindex)
10317 		dev->ifindex = dev_new_index(net);
10318 	else if (__dev_get_by_index(net, dev->ifindex))
10319 		goto err_uninit;
10320 
10321 	/* Transfer changeable features to wanted_features and enable
10322 	 * software offloads (GSO and GRO).
10323 	 */
10324 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10325 	dev->features |= NETIF_F_SOFT_FEATURES;
10326 
10327 	if (dev->udp_tunnel_nic_info) {
10328 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10329 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10330 	}
10331 
10332 	dev->wanted_features = dev->features & dev->hw_features;
10333 
10334 	if (!(dev->flags & IFF_LOOPBACK))
10335 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10336 
10337 	/* If IPv4 TCP segmentation offload is supported we should also
10338 	 * allow the device to enable segmenting the frame with the option
10339 	 * of ignoring a static IP ID value.  This doesn't enable the
10340 	 * feature itself but allows the user to enable it later.
10341 	 */
10342 	if (dev->hw_features & NETIF_F_TSO)
10343 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10344 	if (dev->vlan_features & NETIF_F_TSO)
10345 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10346 	if (dev->mpls_features & NETIF_F_TSO)
10347 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10348 	if (dev->hw_enc_features & NETIF_F_TSO)
10349 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10350 
10351 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10352 	 */
10353 	dev->vlan_features |= NETIF_F_HIGHDMA;
10354 
10355 	/* Make NETIF_F_SG inheritable to tunnel devices.
10356 	 */
10357 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10358 
10359 	/* Make NETIF_F_SG inheritable to MPLS.
10360 	 */
10361 	dev->mpls_features |= NETIF_F_SG;
10362 
10363 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10364 	ret = notifier_to_errno(ret);
10365 	if (ret)
10366 		goto err_uninit;
10367 
10368 	ret = netdev_register_kobject(dev);
10369 	if (ret) {
10370 		dev->reg_state = NETREG_UNREGISTERED;
10371 		goto err_uninit;
10372 	}
10373 	dev->reg_state = NETREG_REGISTERED;
10374 
10375 	__netdev_update_features(dev);
10376 
10377 	/*
10378 	 *	Default initial state at registry is that the
10379 	 *	device is present.
10380 	 */
10381 
10382 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10383 
10384 	linkwatch_init_dev(dev);
10385 
10386 	dev_init_scheduler(dev);
10387 	dev_hold(dev);
10388 	list_netdevice(dev);
10389 	add_device_randomness(dev->dev_addr, dev->addr_len);
10390 
10391 	/* If the device has permanent device address, driver should
10392 	 * set dev_addr and also addr_assign_type should be set to
10393 	 * NET_ADDR_PERM (default value).
10394 	 */
10395 	if (dev->addr_assign_type == NET_ADDR_PERM)
10396 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10397 
10398 	/* Notify protocols, that a new device appeared. */
10399 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10400 	ret = notifier_to_errno(ret);
10401 	if (ret) {
10402 		/* Expect explicit free_netdev() on failure */
10403 		dev->needs_free_netdev = false;
10404 		unregister_netdevice_queue(dev, NULL);
10405 		goto out;
10406 	}
10407 	/*
10408 	 *	Prevent userspace races by waiting until the network
10409 	 *	device is fully setup before sending notifications.
10410 	 */
10411 	if (!dev->rtnl_link_ops ||
10412 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10413 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10414 
10415 out:
10416 	return ret;
10417 
10418 err_uninit:
10419 	if (dev->netdev_ops->ndo_uninit)
10420 		dev->netdev_ops->ndo_uninit(dev);
10421 	if (dev->priv_destructor)
10422 		dev->priv_destructor(dev);
10423 err_free_name:
10424 	netdev_name_node_free(dev->name_node);
10425 	goto out;
10426 }
10427 EXPORT_SYMBOL(register_netdevice);
10428 
10429 /**
10430  *	init_dummy_netdev	- init a dummy network device for NAPI
10431  *	@dev: device to init
10432  *
10433  *	This takes a network device structure and initialize the minimum
10434  *	amount of fields so it can be used to schedule NAPI polls without
10435  *	registering a full blown interface. This is to be used by drivers
10436  *	that need to tie several hardware interfaces to a single NAPI
10437  *	poll scheduler due to HW limitations.
10438  */
10439 int init_dummy_netdev(struct net_device *dev)
10440 {
10441 	/* Clear everything. Note we don't initialize spinlocks
10442 	 * are they aren't supposed to be taken by any of the
10443 	 * NAPI code and this dummy netdev is supposed to be
10444 	 * only ever used for NAPI polls
10445 	 */
10446 	memset(dev, 0, sizeof(struct net_device));
10447 
10448 	/* make sure we BUG if trying to hit standard
10449 	 * register/unregister code path
10450 	 */
10451 	dev->reg_state = NETREG_DUMMY;
10452 
10453 	/* NAPI wants this */
10454 	INIT_LIST_HEAD(&dev->napi_list);
10455 
10456 	/* a dummy interface is started by default */
10457 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10458 	set_bit(__LINK_STATE_START, &dev->state);
10459 
10460 	/* napi_busy_loop stats accounting wants this */
10461 	dev_net_set(dev, &init_net);
10462 
10463 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10464 	 * because users of this 'device' dont need to change
10465 	 * its refcount.
10466 	 */
10467 
10468 	return 0;
10469 }
10470 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10471 
10472 
10473 /**
10474  *	register_netdev	- register a network device
10475  *	@dev: device to register
10476  *
10477  *	Take a completed network device structure and add it to the kernel
10478  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10479  *	chain. 0 is returned on success. A negative errno code is returned
10480  *	on a failure to set up the device, or if the name is a duplicate.
10481  *
10482  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10483  *	and expands the device name if you passed a format string to
10484  *	alloc_netdev.
10485  */
10486 int register_netdev(struct net_device *dev)
10487 {
10488 	int err;
10489 
10490 	if (rtnl_lock_killable())
10491 		return -EINTR;
10492 	err = register_netdevice(dev);
10493 	rtnl_unlock();
10494 	return err;
10495 }
10496 EXPORT_SYMBOL(register_netdev);
10497 
10498 int netdev_refcnt_read(const struct net_device *dev)
10499 {
10500 #ifdef CONFIG_PCPU_DEV_REFCNT
10501 	int i, refcnt = 0;
10502 
10503 	for_each_possible_cpu(i)
10504 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10505 	return refcnt;
10506 #else
10507 	return refcount_read(&dev->dev_refcnt);
10508 #endif
10509 }
10510 EXPORT_SYMBOL(netdev_refcnt_read);
10511 
10512 int netdev_unregister_timeout_secs __read_mostly = 10;
10513 
10514 #define WAIT_REFS_MIN_MSECS 1
10515 #define WAIT_REFS_MAX_MSECS 250
10516 /**
10517  * netdev_wait_allrefs - wait until all references are gone.
10518  * @dev: target net_device
10519  *
10520  * This is called when unregistering network devices.
10521  *
10522  * Any protocol or device that holds a reference should register
10523  * for netdevice notification, and cleanup and put back the
10524  * reference if they receive an UNREGISTER event.
10525  * We can get stuck here if buggy protocols don't correctly
10526  * call dev_put.
10527  */
10528 static void netdev_wait_allrefs(struct net_device *dev)
10529 {
10530 	unsigned long rebroadcast_time, warning_time;
10531 	int wait = 0, refcnt;
10532 
10533 	linkwatch_forget_dev(dev);
10534 
10535 	rebroadcast_time = warning_time = jiffies;
10536 	refcnt = netdev_refcnt_read(dev);
10537 
10538 	while (refcnt != 1) {
10539 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10540 			rtnl_lock();
10541 
10542 			/* Rebroadcast unregister notification */
10543 			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10544 
10545 			__rtnl_unlock();
10546 			rcu_barrier();
10547 			rtnl_lock();
10548 
10549 			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10550 				     &dev->state)) {
10551 				/* We must not have linkwatch events
10552 				 * pending on unregister. If this
10553 				 * happens, we simply run the queue
10554 				 * unscheduled, resulting in a noop
10555 				 * for this device.
10556 				 */
10557 				linkwatch_run_queue();
10558 			}
10559 
10560 			__rtnl_unlock();
10561 
10562 			rebroadcast_time = jiffies;
10563 		}
10564 
10565 		if (!wait) {
10566 			rcu_barrier();
10567 			wait = WAIT_REFS_MIN_MSECS;
10568 		} else {
10569 			msleep(wait);
10570 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10571 		}
10572 
10573 		refcnt = netdev_refcnt_read(dev);
10574 
10575 		if (refcnt != 1 &&
10576 		    time_after(jiffies, warning_time +
10577 			       netdev_unregister_timeout_secs * HZ)) {
10578 			pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10579 				 dev->name, refcnt);
10580 			warning_time = jiffies;
10581 		}
10582 	}
10583 }
10584 
10585 /* The sequence is:
10586  *
10587  *	rtnl_lock();
10588  *	...
10589  *	register_netdevice(x1);
10590  *	register_netdevice(x2);
10591  *	...
10592  *	unregister_netdevice(y1);
10593  *	unregister_netdevice(y2);
10594  *      ...
10595  *	rtnl_unlock();
10596  *	free_netdev(y1);
10597  *	free_netdev(y2);
10598  *
10599  * We are invoked by rtnl_unlock().
10600  * This allows us to deal with problems:
10601  * 1) We can delete sysfs objects which invoke hotplug
10602  *    without deadlocking with linkwatch via keventd.
10603  * 2) Since we run with the RTNL semaphore not held, we can sleep
10604  *    safely in order to wait for the netdev refcnt to drop to zero.
10605  *
10606  * We must not return until all unregister events added during
10607  * the interval the lock was held have been completed.
10608  */
10609 void netdev_run_todo(void)
10610 {
10611 	struct list_head list;
10612 #ifdef CONFIG_LOCKDEP
10613 	struct list_head unlink_list;
10614 
10615 	list_replace_init(&net_unlink_list, &unlink_list);
10616 
10617 	while (!list_empty(&unlink_list)) {
10618 		struct net_device *dev = list_first_entry(&unlink_list,
10619 							  struct net_device,
10620 							  unlink_list);
10621 		list_del_init(&dev->unlink_list);
10622 		dev->nested_level = dev->lower_level - 1;
10623 	}
10624 #endif
10625 
10626 	/* Snapshot list, allow later requests */
10627 	list_replace_init(&net_todo_list, &list);
10628 
10629 	__rtnl_unlock();
10630 
10631 
10632 	/* Wait for rcu callbacks to finish before next phase */
10633 	if (!list_empty(&list))
10634 		rcu_barrier();
10635 
10636 	while (!list_empty(&list)) {
10637 		struct net_device *dev
10638 			= list_first_entry(&list, struct net_device, todo_list);
10639 		list_del(&dev->todo_list);
10640 
10641 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10642 			pr_err("network todo '%s' but state %d\n",
10643 			       dev->name, dev->reg_state);
10644 			dump_stack();
10645 			continue;
10646 		}
10647 
10648 		dev->reg_state = NETREG_UNREGISTERED;
10649 
10650 		netdev_wait_allrefs(dev);
10651 
10652 		/* paranoia */
10653 		BUG_ON(netdev_refcnt_read(dev) != 1);
10654 		BUG_ON(!list_empty(&dev->ptype_all));
10655 		BUG_ON(!list_empty(&dev->ptype_specific));
10656 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10657 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10658 #if IS_ENABLED(CONFIG_DECNET)
10659 		WARN_ON(dev->dn_ptr);
10660 #endif
10661 		if (dev->priv_destructor)
10662 			dev->priv_destructor(dev);
10663 		if (dev->needs_free_netdev)
10664 			free_netdev(dev);
10665 
10666 		/* Report a network device has been unregistered */
10667 		rtnl_lock();
10668 		dev_net(dev)->dev_unreg_count--;
10669 		__rtnl_unlock();
10670 		wake_up(&netdev_unregistering_wq);
10671 
10672 		/* Free network device */
10673 		kobject_put(&dev->dev.kobj);
10674 	}
10675 }
10676 
10677 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10678  * all the same fields in the same order as net_device_stats, with only
10679  * the type differing, but rtnl_link_stats64 may have additional fields
10680  * at the end for newer counters.
10681  */
10682 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10683 			     const struct net_device_stats *netdev_stats)
10684 {
10685 #if BITS_PER_LONG == 64
10686 	BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10687 	memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10688 	/* zero out counters that only exist in rtnl_link_stats64 */
10689 	memset((char *)stats64 + sizeof(*netdev_stats), 0,
10690 	       sizeof(*stats64) - sizeof(*netdev_stats));
10691 #else
10692 	size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10693 	const unsigned long *src = (const unsigned long *)netdev_stats;
10694 	u64 *dst = (u64 *)stats64;
10695 
10696 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10697 	for (i = 0; i < n; i++)
10698 		dst[i] = src[i];
10699 	/* zero out counters that only exist in rtnl_link_stats64 */
10700 	memset((char *)stats64 + n * sizeof(u64), 0,
10701 	       sizeof(*stats64) - n * sizeof(u64));
10702 #endif
10703 }
10704 EXPORT_SYMBOL(netdev_stats_to_stats64);
10705 
10706 /**
10707  *	dev_get_stats	- get network device statistics
10708  *	@dev: device to get statistics from
10709  *	@storage: place to store stats
10710  *
10711  *	Get network statistics from device. Return @storage.
10712  *	The device driver may provide its own method by setting
10713  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10714  *	otherwise the internal statistics structure is used.
10715  */
10716 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10717 					struct rtnl_link_stats64 *storage)
10718 {
10719 	const struct net_device_ops *ops = dev->netdev_ops;
10720 
10721 	if (ops->ndo_get_stats64) {
10722 		memset(storage, 0, sizeof(*storage));
10723 		ops->ndo_get_stats64(dev, storage);
10724 	} else if (ops->ndo_get_stats) {
10725 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10726 	} else {
10727 		netdev_stats_to_stats64(storage, &dev->stats);
10728 	}
10729 	storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10730 	storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10731 	storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10732 	return storage;
10733 }
10734 EXPORT_SYMBOL(dev_get_stats);
10735 
10736 /**
10737  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10738  *	@s: place to store stats
10739  *	@netstats: per-cpu network stats to read from
10740  *
10741  *	Read per-cpu network statistics and populate the related fields in @s.
10742  */
10743 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10744 			   const struct pcpu_sw_netstats __percpu *netstats)
10745 {
10746 	int cpu;
10747 
10748 	for_each_possible_cpu(cpu) {
10749 		const struct pcpu_sw_netstats *stats;
10750 		struct pcpu_sw_netstats tmp;
10751 		unsigned int start;
10752 
10753 		stats = per_cpu_ptr(netstats, cpu);
10754 		do {
10755 			start = u64_stats_fetch_begin_irq(&stats->syncp);
10756 			tmp.rx_packets = stats->rx_packets;
10757 			tmp.rx_bytes   = stats->rx_bytes;
10758 			tmp.tx_packets = stats->tx_packets;
10759 			tmp.tx_bytes   = stats->tx_bytes;
10760 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10761 
10762 		s->rx_packets += tmp.rx_packets;
10763 		s->rx_bytes   += tmp.rx_bytes;
10764 		s->tx_packets += tmp.tx_packets;
10765 		s->tx_bytes   += tmp.tx_bytes;
10766 	}
10767 }
10768 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10769 
10770 /**
10771  *	dev_get_tstats64 - ndo_get_stats64 implementation
10772  *	@dev: device to get statistics from
10773  *	@s: place to store stats
10774  *
10775  *	Populate @s from dev->stats and dev->tstats. Can be used as
10776  *	ndo_get_stats64() callback.
10777  */
10778 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10779 {
10780 	netdev_stats_to_stats64(s, &dev->stats);
10781 	dev_fetch_sw_netstats(s, dev->tstats);
10782 }
10783 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10784 
10785 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10786 {
10787 	struct netdev_queue *queue = dev_ingress_queue(dev);
10788 
10789 #ifdef CONFIG_NET_CLS_ACT
10790 	if (queue)
10791 		return queue;
10792 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10793 	if (!queue)
10794 		return NULL;
10795 	netdev_init_one_queue(dev, queue, NULL);
10796 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10797 	queue->qdisc_sleeping = &noop_qdisc;
10798 	rcu_assign_pointer(dev->ingress_queue, queue);
10799 #endif
10800 	return queue;
10801 }
10802 
10803 static const struct ethtool_ops default_ethtool_ops;
10804 
10805 void netdev_set_default_ethtool_ops(struct net_device *dev,
10806 				    const struct ethtool_ops *ops)
10807 {
10808 	if (dev->ethtool_ops == &default_ethtool_ops)
10809 		dev->ethtool_ops = ops;
10810 }
10811 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10812 
10813 void netdev_freemem(struct net_device *dev)
10814 {
10815 	char *addr = (char *)dev - dev->padded;
10816 
10817 	kvfree(addr);
10818 }
10819 
10820 /**
10821  * alloc_netdev_mqs - allocate network device
10822  * @sizeof_priv: size of private data to allocate space for
10823  * @name: device name format string
10824  * @name_assign_type: origin of device name
10825  * @setup: callback to initialize device
10826  * @txqs: the number of TX subqueues to allocate
10827  * @rxqs: the number of RX subqueues to allocate
10828  *
10829  * Allocates a struct net_device with private data area for driver use
10830  * and performs basic initialization.  Also allocates subqueue structs
10831  * for each queue on the device.
10832  */
10833 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10834 		unsigned char name_assign_type,
10835 		void (*setup)(struct net_device *),
10836 		unsigned int txqs, unsigned int rxqs)
10837 {
10838 	struct net_device *dev;
10839 	unsigned int alloc_size;
10840 	struct net_device *p;
10841 
10842 	BUG_ON(strlen(name) >= sizeof(dev->name));
10843 
10844 	if (txqs < 1) {
10845 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10846 		return NULL;
10847 	}
10848 
10849 	if (rxqs < 1) {
10850 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10851 		return NULL;
10852 	}
10853 
10854 	alloc_size = sizeof(struct net_device);
10855 	if (sizeof_priv) {
10856 		/* ensure 32-byte alignment of private area */
10857 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10858 		alloc_size += sizeof_priv;
10859 	}
10860 	/* ensure 32-byte alignment of whole construct */
10861 	alloc_size += NETDEV_ALIGN - 1;
10862 
10863 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10864 	if (!p)
10865 		return NULL;
10866 
10867 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10868 	dev->padded = (char *)dev - (char *)p;
10869 
10870 #ifdef CONFIG_PCPU_DEV_REFCNT
10871 	dev->pcpu_refcnt = alloc_percpu(int);
10872 	if (!dev->pcpu_refcnt)
10873 		goto free_dev;
10874 	dev_hold(dev);
10875 #else
10876 	refcount_set(&dev->dev_refcnt, 1);
10877 #endif
10878 
10879 	if (dev_addr_init(dev))
10880 		goto free_pcpu;
10881 
10882 	dev_mc_init(dev);
10883 	dev_uc_init(dev);
10884 
10885 	dev_net_set(dev, &init_net);
10886 
10887 	dev->gso_max_size = GSO_MAX_SIZE;
10888 	dev->gso_max_segs = GSO_MAX_SEGS;
10889 	dev->upper_level = 1;
10890 	dev->lower_level = 1;
10891 #ifdef CONFIG_LOCKDEP
10892 	dev->nested_level = 0;
10893 	INIT_LIST_HEAD(&dev->unlink_list);
10894 #endif
10895 
10896 	INIT_LIST_HEAD(&dev->napi_list);
10897 	INIT_LIST_HEAD(&dev->unreg_list);
10898 	INIT_LIST_HEAD(&dev->close_list);
10899 	INIT_LIST_HEAD(&dev->link_watch_list);
10900 	INIT_LIST_HEAD(&dev->adj_list.upper);
10901 	INIT_LIST_HEAD(&dev->adj_list.lower);
10902 	INIT_LIST_HEAD(&dev->ptype_all);
10903 	INIT_LIST_HEAD(&dev->ptype_specific);
10904 	INIT_LIST_HEAD(&dev->net_notifier_list);
10905 #ifdef CONFIG_NET_SCHED
10906 	hash_init(dev->qdisc_hash);
10907 #endif
10908 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10909 	setup(dev);
10910 
10911 	if (!dev->tx_queue_len) {
10912 		dev->priv_flags |= IFF_NO_QUEUE;
10913 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10914 	}
10915 
10916 	dev->num_tx_queues = txqs;
10917 	dev->real_num_tx_queues = txqs;
10918 	if (netif_alloc_netdev_queues(dev))
10919 		goto free_all;
10920 
10921 	dev->num_rx_queues = rxqs;
10922 	dev->real_num_rx_queues = rxqs;
10923 	if (netif_alloc_rx_queues(dev))
10924 		goto free_all;
10925 
10926 	strcpy(dev->name, name);
10927 	dev->name_assign_type = name_assign_type;
10928 	dev->group = INIT_NETDEV_GROUP;
10929 	if (!dev->ethtool_ops)
10930 		dev->ethtool_ops = &default_ethtool_ops;
10931 
10932 	nf_hook_ingress_init(dev);
10933 
10934 	return dev;
10935 
10936 free_all:
10937 	free_netdev(dev);
10938 	return NULL;
10939 
10940 free_pcpu:
10941 #ifdef CONFIG_PCPU_DEV_REFCNT
10942 	free_percpu(dev->pcpu_refcnt);
10943 free_dev:
10944 #endif
10945 	netdev_freemem(dev);
10946 	return NULL;
10947 }
10948 EXPORT_SYMBOL(alloc_netdev_mqs);
10949 
10950 /**
10951  * free_netdev - free network device
10952  * @dev: device
10953  *
10954  * This function does the last stage of destroying an allocated device
10955  * interface. The reference to the device object is released. If this
10956  * is the last reference then it will be freed.Must be called in process
10957  * context.
10958  */
10959 void free_netdev(struct net_device *dev)
10960 {
10961 	struct napi_struct *p, *n;
10962 
10963 	might_sleep();
10964 
10965 	/* When called immediately after register_netdevice() failed the unwind
10966 	 * handling may still be dismantling the device. Handle that case by
10967 	 * deferring the free.
10968 	 */
10969 	if (dev->reg_state == NETREG_UNREGISTERING) {
10970 		ASSERT_RTNL();
10971 		dev->needs_free_netdev = true;
10972 		return;
10973 	}
10974 
10975 	netif_free_tx_queues(dev);
10976 	netif_free_rx_queues(dev);
10977 
10978 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10979 
10980 	/* Flush device addresses */
10981 	dev_addr_flush(dev);
10982 
10983 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10984 		netif_napi_del(p);
10985 
10986 #ifdef CONFIG_PCPU_DEV_REFCNT
10987 	free_percpu(dev->pcpu_refcnt);
10988 	dev->pcpu_refcnt = NULL;
10989 #endif
10990 	free_percpu(dev->xdp_bulkq);
10991 	dev->xdp_bulkq = NULL;
10992 
10993 	/*  Compatibility with error handling in drivers */
10994 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10995 		netdev_freemem(dev);
10996 		return;
10997 	}
10998 
10999 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
11000 	dev->reg_state = NETREG_RELEASED;
11001 
11002 	/* will free via device release */
11003 	put_device(&dev->dev);
11004 }
11005 EXPORT_SYMBOL(free_netdev);
11006 
11007 /**
11008  *	synchronize_net -  Synchronize with packet receive processing
11009  *
11010  *	Wait for packets currently being received to be done.
11011  *	Does not block later packets from starting.
11012  */
11013 void synchronize_net(void)
11014 {
11015 	might_sleep();
11016 	if (rtnl_is_locked())
11017 		synchronize_rcu_expedited();
11018 	else
11019 		synchronize_rcu();
11020 }
11021 EXPORT_SYMBOL(synchronize_net);
11022 
11023 /**
11024  *	unregister_netdevice_queue - remove device from the kernel
11025  *	@dev: device
11026  *	@head: list
11027  *
11028  *	This function shuts down a device interface and removes it
11029  *	from the kernel tables.
11030  *	If head not NULL, device is queued to be unregistered later.
11031  *
11032  *	Callers must hold the rtnl semaphore.  You may want
11033  *	unregister_netdev() instead of this.
11034  */
11035 
11036 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11037 {
11038 	ASSERT_RTNL();
11039 
11040 	if (head) {
11041 		list_move_tail(&dev->unreg_list, head);
11042 	} else {
11043 		LIST_HEAD(single);
11044 
11045 		list_add(&dev->unreg_list, &single);
11046 		unregister_netdevice_many(&single);
11047 	}
11048 }
11049 EXPORT_SYMBOL(unregister_netdevice_queue);
11050 
11051 /**
11052  *	unregister_netdevice_many - unregister many devices
11053  *	@head: list of devices
11054  *
11055  *  Note: As most callers use a stack allocated list_head,
11056  *  we force a list_del() to make sure stack wont be corrupted later.
11057  */
11058 void unregister_netdevice_many(struct list_head *head)
11059 {
11060 	struct net_device *dev, *tmp;
11061 	LIST_HEAD(close_head);
11062 
11063 	BUG_ON(dev_boot_phase);
11064 	ASSERT_RTNL();
11065 
11066 	if (list_empty(head))
11067 		return;
11068 
11069 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11070 		/* Some devices call without registering
11071 		 * for initialization unwind. Remove those
11072 		 * devices and proceed with the remaining.
11073 		 */
11074 		if (dev->reg_state == NETREG_UNINITIALIZED) {
11075 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11076 				 dev->name, dev);
11077 
11078 			WARN_ON(1);
11079 			list_del(&dev->unreg_list);
11080 			continue;
11081 		}
11082 		dev->dismantle = true;
11083 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
11084 	}
11085 
11086 	/* If device is running, close it first. */
11087 	list_for_each_entry(dev, head, unreg_list)
11088 		list_add_tail(&dev->close_list, &close_head);
11089 	dev_close_many(&close_head, true);
11090 
11091 	list_for_each_entry(dev, head, unreg_list) {
11092 		/* And unlink it from device chain. */
11093 		unlist_netdevice(dev);
11094 
11095 		dev->reg_state = NETREG_UNREGISTERING;
11096 	}
11097 	flush_all_backlogs();
11098 
11099 	synchronize_net();
11100 
11101 	list_for_each_entry(dev, head, unreg_list) {
11102 		struct sk_buff *skb = NULL;
11103 
11104 		/* Shutdown queueing discipline. */
11105 		dev_shutdown(dev);
11106 
11107 		dev_xdp_uninstall(dev);
11108 
11109 		/* Notify protocols, that we are about to destroy
11110 		 * this device. They should clean all the things.
11111 		 */
11112 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11113 
11114 		if (!dev->rtnl_link_ops ||
11115 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11116 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11117 						     GFP_KERNEL, NULL, 0);
11118 
11119 		/*
11120 		 *	Flush the unicast and multicast chains
11121 		 */
11122 		dev_uc_flush(dev);
11123 		dev_mc_flush(dev);
11124 
11125 		netdev_name_node_alt_flush(dev);
11126 		netdev_name_node_free(dev->name_node);
11127 
11128 		if (dev->netdev_ops->ndo_uninit)
11129 			dev->netdev_ops->ndo_uninit(dev);
11130 
11131 		if (skb)
11132 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11133 
11134 		/* Notifier chain MUST detach us all upper devices. */
11135 		WARN_ON(netdev_has_any_upper_dev(dev));
11136 		WARN_ON(netdev_has_any_lower_dev(dev));
11137 
11138 		/* Remove entries from kobject tree */
11139 		netdev_unregister_kobject(dev);
11140 #ifdef CONFIG_XPS
11141 		/* Remove XPS queueing entries */
11142 		netif_reset_xps_queues_gt(dev, 0);
11143 #endif
11144 	}
11145 
11146 	synchronize_net();
11147 
11148 	list_for_each_entry(dev, head, unreg_list) {
11149 		dev_put(dev);
11150 		net_set_todo(dev);
11151 	}
11152 
11153 	list_del(head);
11154 }
11155 EXPORT_SYMBOL(unregister_netdevice_many);
11156 
11157 /**
11158  *	unregister_netdev - remove device from the kernel
11159  *	@dev: device
11160  *
11161  *	This function shuts down a device interface and removes it
11162  *	from the kernel tables.
11163  *
11164  *	This is just a wrapper for unregister_netdevice that takes
11165  *	the rtnl semaphore.  In general you want to use this and not
11166  *	unregister_netdevice.
11167  */
11168 void unregister_netdev(struct net_device *dev)
11169 {
11170 	rtnl_lock();
11171 	unregister_netdevice(dev);
11172 	rtnl_unlock();
11173 }
11174 EXPORT_SYMBOL(unregister_netdev);
11175 
11176 /**
11177  *	__dev_change_net_namespace - move device to different nethost namespace
11178  *	@dev: device
11179  *	@net: network namespace
11180  *	@pat: If not NULL name pattern to try if the current device name
11181  *	      is already taken in the destination network namespace.
11182  *	@new_ifindex: If not zero, specifies device index in the target
11183  *	              namespace.
11184  *
11185  *	This function shuts down a device interface and moves it
11186  *	to a new network namespace. On success 0 is returned, on
11187  *	a failure a netagive errno code is returned.
11188  *
11189  *	Callers must hold the rtnl semaphore.
11190  */
11191 
11192 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11193 			       const char *pat, int new_ifindex)
11194 {
11195 	struct net *net_old = dev_net(dev);
11196 	int err, new_nsid;
11197 
11198 	ASSERT_RTNL();
11199 
11200 	/* Don't allow namespace local devices to be moved. */
11201 	err = -EINVAL;
11202 	if (dev->features & NETIF_F_NETNS_LOCAL)
11203 		goto out;
11204 
11205 	/* Ensure the device has been registrered */
11206 	if (dev->reg_state != NETREG_REGISTERED)
11207 		goto out;
11208 
11209 	/* Get out if there is nothing todo */
11210 	err = 0;
11211 	if (net_eq(net_old, net))
11212 		goto out;
11213 
11214 	/* Pick the destination device name, and ensure
11215 	 * we can use it in the destination network namespace.
11216 	 */
11217 	err = -EEXIST;
11218 	if (__dev_get_by_name(net, dev->name)) {
11219 		/* We get here if we can't use the current device name */
11220 		if (!pat)
11221 			goto out;
11222 		err = dev_get_valid_name(net, dev, pat);
11223 		if (err < 0)
11224 			goto out;
11225 	}
11226 
11227 	/* Check that new_ifindex isn't used yet. */
11228 	err = -EBUSY;
11229 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11230 		goto out;
11231 
11232 	/*
11233 	 * And now a mini version of register_netdevice unregister_netdevice.
11234 	 */
11235 
11236 	/* If device is running close it first. */
11237 	dev_close(dev);
11238 
11239 	/* And unlink it from device chain */
11240 	unlist_netdevice(dev);
11241 
11242 	synchronize_net();
11243 
11244 	/* Shutdown queueing discipline. */
11245 	dev_shutdown(dev);
11246 
11247 	/* Notify protocols, that we are about to destroy
11248 	 * this device. They should clean all the things.
11249 	 *
11250 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
11251 	 * This is wanted because this way 8021q and macvlan know
11252 	 * the device is just moving and can keep their slaves up.
11253 	 */
11254 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11255 	rcu_barrier();
11256 
11257 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11258 	/* If there is an ifindex conflict assign a new one */
11259 	if (!new_ifindex) {
11260 		if (__dev_get_by_index(net, dev->ifindex))
11261 			new_ifindex = dev_new_index(net);
11262 		else
11263 			new_ifindex = dev->ifindex;
11264 	}
11265 
11266 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11267 			    new_ifindex);
11268 
11269 	/*
11270 	 *	Flush the unicast and multicast chains
11271 	 */
11272 	dev_uc_flush(dev);
11273 	dev_mc_flush(dev);
11274 
11275 	/* Send a netdev-removed uevent to the old namespace */
11276 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11277 	netdev_adjacent_del_links(dev);
11278 
11279 	/* Move per-net netdevice notifiers that are following the netdevice */
11280 	move_netdevice_notifiers_dev_net(dev, net);
11281 
11282 	/* Actually switch the network namespace */
11283 	dev_net_set(dev, net);
11284 	dev->ifindex = new_ifindex;
11285 
11286 	/* Send a netdev-add uevent to the new namespace */
11287 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11288 	netdev_adjacent_add_links(dev);
11289 
11290 	/* Fixup kobjects */
11291 	err = device_rename(&dev->dev, dev->name);
11292 	WARN_ON(err);
11293 
11294 	/* Adapt owner in case owning user namespace of target network
11295 	 * namespace is different from the original one.
11296 	 */
11297 	err = netdev_change_owner(dev, net_old, net);
11298 	WARN_ON(err);
11299 
11300 	/* Add the device back in the hashes */
11301 	list_netdevice(dev);
11302 
11303 	/* Notify protocols, that a new device appeared. */
11304 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11305 
11306 	/*
11307 	 *	Prevent userspace races by waiting until the network
11308 	 *	device is fully setup before sending notifications.
11309 	 */
11310 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11311 
11312 	synchronize_net();
11313 	err = 0;
11314 out:
11315 	return err;
11316 }
11317 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11318 
11319 static int dev_cpu_dead(unsigned int oldcpu)
11320 {
11321 	struct sk_buff **list_skb;
11322 	struct sk_buff *skb;
11323 	unsigned int cpu;
11324 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11325 
11326 	local_irq_disable();
11327 	cpu = smp_processor_id();
11328 	sd = &per_cpu(softnet_data, cpu);
11329 	oldsd = &per_cpu(softnet_data, oldcpu);
11330 
11331 	/* Find end of our completion_queue. */
11332 	list_skb = &sd->completion_queue;
11333 	while (*list_skb)
11334 		list_skb = &(*list_skb)->next;
11335 	/* Append completion queue from offline CPU. */
11336 	*list_skb = oldsd->completion_queue;
11337 	oldsd->completion_queue = NULL;
11338 
11339 	/* Append output queue from offline CPU. */
11340 	if (oldsd->output_queue) {
11341 		*sd->output_queue_tailp = oldsd->output_queue;
11342 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11343 		oldsd->output_queue = NULL;
11344 		oldsd->output_queue_tailp = &oldsd->output_queue;
11345 	}
11346 	/* Append NAPI poll list from offline CPU, with one exception :
11347 	 * process_backlog() must be called by cpu owning percpu backlog.
11348 	 * We properly handle process_queue & input_pkt_queue later.
11349 	 */
11350 	while (!list_empty(&oldsd->poll_list)) {
11351 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11352 							    struct napi_struct,
11353 							    poll_list);
11354 
11355 		list_del_init(&napi->poll_list);
11356 		if (napi->poll == process_backlog)
11357 			napi->state = 0;
11358 		else
11359 			____napi_schedule(sd, napi);
11360 	}
11361 
11362 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11363 	local_irq_enable();
11364 
11365 #ifdef CONFIG_RPS
11366 	remsd = oldsd->rps_ipi_list;
11367 	oldsd->rps_ipi_list = NULL;
11368 #endif
11369 	/* send out pending IPI's on offline CPU */
11370 	net_rps_send_ipi(remsd);
11371 
11372 	/* Process offline CPU's input_pkt_queue */
11373 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11374 		netif_rx_ni(skb);
11375 		input_queue_head_incr(oldsd);
11376 	}
11377 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11378 		netif_rx_ni(skb);
11379 		input_queue_head_incr(oldsd);
11380 	}
11381 
11382 	return 0;
11383 }
11384 
11385 /**
11386  *	netdev_increment_features - increment feature set by one
11387  *	@all: current feature set
11388  *	@one: new feature set
11389  *	@mask: mask feature set
11390  *
11391  *	Computes a new feature set after adding a device with feature set
11392  *	@one to the master device with current feature set @all.  Will not
11393  *	enable anything that is off in @mask. Returns the new feature set.
11394  */
11395 netdev_features_t netdev_increment_features(netdev_features_t all,
11396 	netdev_features_t one, netdev_features_t mask)
11397 {
11398 	if (mask & NETIF_F_HW_CSUM)
11399 		mask |= NETIF_F_CSUM_MASK;
11400 	mask |= NETIF_F_VLAN_CHALLENGED;
11401 
11402 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11403 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11404 
11405 	/* If one device supports hw checksumming, set for all. */
11406 	if (all & NETIF_F_HW_CSUM)
11407 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11408 
11409 	return all;
11410 }
11411 EXPORT_SYMBOL(netdev_increment_features);
11412 
11413 static struct hlist_head * __net_init netdev_create_hash(void)
11414 {
11415 	int i;
11416 	struct hlist_head *hash;
11417 
11418 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11419 	if (hash != NULL)
11420 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11421 			INIT_HLIST_HEAD(&hash[i]);
11422 
11423 	return hash;
11424 }
11425 
11426 /* Initialize per network namespace state */
11427 static int __net_init netdev_init(struct net *net)
11428 {
11429 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11430 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11431 
11432 	if (net != &init_net)
11433 		INIT_LIST_HEAD(&net->dev_base_head);
11434 
11435 	net->dev_name_head = netdev_create_hash();
11436 	if (net->dev_name_head == NULL)
11437 		goto err_name;
11438 
11439 	net->dev_index_head = netdev_create_hash();
11440 	if (net->dev_index_head == NULL)
11441 		goto err_idx;
11442 
11443 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11444 
11445 	return 0;
11446 
11447 err_idx:
11448 	kfree(net->dev_name_head);
11449 err_name:
11450 	return -ENOMEM;
11451 }
11452 
11453 /**
11454  *	netdev_drivername - network driver for the device
11455  *	@dev: network device
11456  *
11457  *	Determine network driver for device.
11458  */
11459 const char *netdev_drivername(const struct net_device *dev)
11460 {
11461 	const struct device_driver *driver;
11462 	const struct device *parent;
11463 	const char *empty = "";
11464 
11465 	parent = dev->dev.parent;
11466 	if (!parent)
11467 		return empty;
11468 
11469 	driver = parent->driver;
11470 	if (driver && driver->name)
11471 		return driver->name;
11472 	return empty;
11473 }
11474 
11475 static void __netdev_printk(const char *level, const struct net_device *dev,
11476 			    struct va_format *vaf)
11477 {
11478 	if (dev && dev->dev.parent) {
11479 		dev_printk_emit(level[1] - '0',
11480 				dev->dev.parent,
11481 				"%s %s %s%s: %pV",
11482 				dev_driver_string(dev->dev.parent),
11483 				dev_name(dev->dev.parent),
11484 				netdev_name(dev), netdev_reg_state(dev),
11485 				vaf);
11486 	} else if (dev) {
11487 		printk("%s%s%s: %pV",
11488 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11489 	} else {
11490 		printk("%s(NULL net_device): %pV", level, vaf);
11491 	}
11492 }
11493 
11494 void netdev_printk(const char *level, const struct net_device *dev,
11495 		   const char *format, ...)
11496 {
11497 	struct va_format vaf;
11498 	va_list args;
11499 
11500 	va_start(args, format);
11501 
11502 	vaf.fmt = format;
11503 	vaf.va = &args;
11504 
11505 	__netdev_printk(level, dev, &vaf);
11506 
11507 	va_end(args);
11508 }
11509 EXPORT_SYMBOL(netdev_printk);
11510 
11511 #define define_netdev_printk_level(func, level)			\
11512 void func(const struct net_device *dev, const char *fmt, ...)	\
11513 {								\
11514 	struct va_format vaf;					\
11515 	va_list args;						\
11516 								\
11517 	va_start(args, fmt);					\
11518 								\
11519 	vaf.fmt = fmt;						\
11520 	vaf.va = &args;						\
11521 								\
11522 	__netdev_printk(level, dev, &vaf);			\
11523 								\
11524 	va_end(args);						\
11525 }								\
11526 EXPORT_SYMBOL(func);
11527 
11528 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11529 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11530 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11531 define_netdev_printk_level(netdev_err, KERN_ERR);
11532 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11533 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11534 define_netdev_printk_level(netdev_info, KERN_INFO);
11535 
11536 static void __net_exit netdev_exit(struct net *net)
11537 {
11538 	kfree(net->dev_name_head);
11539 	kfree(net->dev_index_head);
11540 	if (net != &init_net)
11541 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11542 }
11543 
11544 static struct pernet_operations __net_initdata netdev_net_ops = {
11545 	.init = netdev_init,
11546 	.exit = netdev_exit,
11547 };
11548 
11549 static void __net_exit default_device_exit(struct net *net)
11550 {
11551 	struct net_device *dev, *aux;
11552 	/*
11553 	 * Push all migratable network devices back to the
11554 	 * initial network namespace
11555 	 */
11556 	rtnl_lock();
11557 	for_each_netdev_safe(net, dev, aux) {
11558 		int err;
11559 		char fb_name[IFNAMSIZ];
11560 
11561 		/* Ignore unmoveable devices (i.e. loopback) */
11562 		if (dev->features & NETIF_F_NETNS_LOCAL)
11563 			continue;
11564 
11565 		/* Leave virtual devices for the generic cleanup */
11566 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11567 			continue;
11568 
11569 		/* Push remaining network devices to init_net */
11570 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11571 		if (__dev_get_by_name(&init_net, fb_name))
11572 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11573 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11574 		if (err) {
11575 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11576 				 __func__, dev->name, err);
11577 			BUG();
11578 		}
11579 	}
11580 	rtnl_unlock();
11581 }
11582 
11583 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11584 {
11585 	/* Return with the rtnl_lock held when there are no network
11586 	 * devices unregistering in any network namespace in net_list.
11587 	 */
11588 	struct net *net;
11589 	bool unregistering;
11590 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
11591 
11592 	add_wait_queue(&netdev_unregistering_wq, &wait);
11593 	for (;;) {
11594 		unregistering = false;
11595 		rtnl_lock();
11596 		list_for_each_entry(net, net_list, exit_list) {
11597 			if (net->dev_unreg_count > 0) {
11598 				unregistering = true;
11599 				break;
11600 			}
11601 		}
11602 		if (!unregistering)
11603 			break;
11604 		__rtnl_unlock();
11605 
11606 		wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11607 	}
11608 	remove_wait_queue(&netdev_unregistering_wq, &wait);
11609 }
11610 
11611 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11612 {
11613 	/* At exit all network devices most be removed from a network
11614 	 * namespace.  Do this in the reverse order of registration.
11615 	 * Do this across as many network namespaces as possible to
11616 	 * improve batching efficiency.
11617 	 */
11618 	struct net_device *dev;
11619 	struct net *net;
11620 	LIST_HEAD(dev_kill_list);
11621 
11622 	/* To prevent network device cleanup code from dereferencing
11623 	 * loopback devices or network devices that have been freed
11624 	 * wait here for all pending unregistrations to complete,
11625 	 * before unregistring the loopback device and allowing the
11626 	 * network namespace be freed.
11627 	 *
11628 	 * The netdev todo list containing all network devices
11629 	 * unregistrations that happen in default_device_exit_batch
11630 	 * will run in the rtnl_unlock() at the end of
11631 	 * default_device_exit_batch.
11632 	 */
11633 	rtnl_lock_unregistering(net_list);
11634 	list_for_each_entry(net, net_list, exit_list) {
11635 		for_each_netdev_reverse(net, dev) {
11636 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11637 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11638 			else
11639 				unregister_netdevice_queue(dev, &dev_kill_list);
11640 		}
11641 	}
11642 	unregister_netdevice_many(&dev_kill_list);
11643 	rtnl_unlock();
11644 }
11645 
11646 static struct pernet_operations __net_initdata default_device_ops = {
11647 	.exit = default_device_exit,
11648 	.exit_batch = default_device_exit_batch,
11649 };
11650 
11651 /*
11652  *	Initialize the DEV module. At boot time this walks the device list and
11653  *	unhooks any devices that fail to initialise (normally hardware not
11654  *	present) and leaves us with a valid list of present and active devices.
11655  *
11656  */
11657 
11658 /*
11659  *       This is called single threaded during boot, so no need
11660  *       to take the rtnl semaphore.
11661  */
11662 static int __init net_dev_init(void)
11663 {
11664 	int i, rc = -ENOMEM;
11665 
11666 	BUG_ON(!dev_boot_phase);
11667 
11668 	if (dev_proc_init())
11669 		goto out;
11670 
11671 	if (netdev_kobject_init())
11672 		goto out;
11673 
11674 	INIT_LIST_HEAD(&ptype_all);
11675 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11676 		INIT_LIST_HEAD(&ptype_base[i]);
11677 
11678 	INIT_LIST_HEAD(&offload_base);
11679 
11680 	if (register_pernet_subsys(&netdev_net_ops))
11681 		goto out;
11682 
11683 	/*
11684 	 *	Initialise the packet receive queues.
11685 	 */
11686 
11687 	for_each_possible_cpu(i) {
11688 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11689 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11690 
11691 		INIT_WORK(flush, flush_backlog);
11692 
11693 		skb_queue_head_init(&sd->input_pkt_queue);
11694 		skb_queue_head_init(&sd->process_queue);
11695 #ifdef CONFIG_XFRM_OFFLOAD
11696 		skb_queue_head_init(&sd->xfrm_backlog);
11697 #endif
11698 		INIT_LIST_HEAD(&sd->poll_list);
11699 		sd->output_queue_tailp = &sd->output_queue;
11700 #ifdef CONFIG_RPS
11701 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11702 		sd->cpu = i;
11703 #endif
11704 
11705 		init_gro_hash(&sd->backlog);
11706 		sd->backlog.poll = process_backlog;
11707 		sd->backlog.weight = weight_p;
11708 	}
11709 
11710 	dev_boot_phase = 0;
11711 
11712 	/* The loopback device is special if any other network devices
11713 	 * is present in a network namespace the loopback device must
11714 	 * be present. Since we now dynamically allocate and free the
11715 	 * loopback device ensure this invariant is maintained by
11716 	 * keeping the loopback device as the first device on the
11717 	 * list of network devices.  Ensuring the loopback devices
11718 	 * is the first device that appears and the last network device
11719 	 * that disappears.
11720 	 */
11721 	if (register_pernet_device(&loopback_net_ops))
11722 		goto out;
11723 
11724 	if (register_pernet_device(&default_device_ops))
11725 		goto out;
11726 
11727 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11728 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11729 
11730 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11731 				       NULL, dev_cpu_dead);
11732 	WARN_ON(rc < 0);
11733 	rc = 0;
11734 out:
11735 	return rc;
11736 }
11737 
11738 subsys_initcall(net_dev_init);
11739