xref: /linux/net/core/dev.c (revision 49545357bf7e134a4012d4652c2df5f78f4485af)
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_dst_cmp(p, skb);
6018 		diffs |= skb_metadata_differs(p, skb);
6019 		if (maclen == ETH_HLEN)
6020 			diffs |= compare_ether_header(skb_mac_header(p),
6021 						      skb_mac_header(skb));
6022 		else if (!diffs)
6023 			diffs = memcmp(skb_mac_header(p),
6024 				       skb_mac_header(skb),
6025 				       maclen);
6026 
6027 		diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
6028 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
6029 		if (!diffs) {
6030 			struct tc_skb_ext *skb_ext = skb_ext_find(skb, TC_SKB_EXT);
6031 			struct tc_skb_ext *p_ext = skb_ext_find(p, TC_SKB_EXT);
6032 
6033 			diffs |= (!!p_ext) ^ (!!skb_ext);
6034 			if (!diffs && unlikely(skb_ext))
6035 				diffs |= p_ext->chain ^ skb_ext->chain;
6036 		}
6037 #endif
6038 
6039 		NAPI_GRO_CB(p)->same_flow = !diffs;
6040 	}
6041 }
6042 
6043 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
6044 {
6045 	const struct skb_shared_info *pinfo = skb_shinfo(skb);
6046 	const skb_frag_t *frag0 = &pinfo->frags[0];
6047 
6048 	NAPI_GRO_CB(skb)->data_offset = 0;
6049 	NAPI_GRO_CB(skb)->frag0 = NULL;
6050 	NAPI_GRO_CB(skb)->frag0_len = 0;
6051 
6052 	if (!skb_headlen(skb) && pinfo->nr_frags &&
6053 	    !PageHighMem(skb_frag_page(frag0)) &&
6054 	    (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
6055 		NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
6056 		NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
6057 						    skb_frag_size(frag0),
6058 						    skb->end - skb->tail);
6059 	}
6060 }
6061 
6062 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
6063 {
6064 	struct skb_shared_info *pinfo = skb_shinfo(skb);
6065 
6066 	BUG_ON(skb->end - skb->tail < grow);
6067 
6068 	memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6069 
6070 	skb->data_len -= grow;
6071 	skb->tail += grow;
6072 
6073 	skb_frag_off_add(&pinfo->frags[0], grow);
6074 	skb_frag_size_sub(&pinfo->frags[0], grow);
6075 
6076 	if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6077 		skb_frag_unref(skb, 0);
6078 		memmove(pinfo->frags, pinfo->frags + 1,
6079 			--pinfo->nr_frags * sizeof(pinfo->frags[0]));
6080 	}
6081 }
6082 
6083 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6084 {
6085 	struct sk_buff *oldest;
6086 
6087 	oldest = list_last_entry(head, struct sk_buff, list);
6088 
6089 	/* We are called with head length >= MAX_GRO_SKBS, so this is
6090 	 * impossible.
6091 	 */
6092 	if (WARN_ON_ONCE(!oldest))
6093 		return;
6094 
6095 	/* Do not adjust napi->gro_hash[].count, caller is adding a new
6096 	 * SKB to the chain.
6097 	 */
6098 	skb_list_del_init(oldest);
6099 	napi_gro_complete(napi, oldest);
6100 }
6101 
6102 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6103 {
6104 	u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6105 	struct gro_list *gro_list = &napi->gro_hash[bucket];
6106 	struct list_head *head = &offload_base;
6107 	struct packet_offload *ptype;
6108 	__be16 type = skb->protocol;
6109 	struct sk_buff *pp = NULL;
6110 	enum gro_result ret;
6111 	int same_flow;
6112 	int grow;
6113 
6114 	if (netif_elide_gro(skb->dev))
6115 		goto normal;
6116 
6117 	gro_list_prepare(&gro_list->list, skb);
6118 
6119 	rcu_read_lock();
6120 	list_for_each_entry_rcu(ptype, head, list) {
6121 		if (ptype->type != type || !ptype->callbacks.gro_receive)
6122 			continue;
6123 
6124 		skb_set_network_header(skb, skb_gro_offset(skb));
6125 		skb_reset_mac_len(skb);
6126 		NAPI_GRO_CB(skb)->same_flow = 0;
6127 		NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6128 		NAPI_GRO_CB(skb)->free = 0;
6129 		NAPI_GRO_CB(skb)->encap_mark = 0;
6130 		NAPI_GRO_CB(skb)->recursion_counter = 0;
6131 		NAPI_GRO_CB(skb)->is_fou = 0;
6132 		NAPI_GRO_CB(skb)->is_atomic = 1;
6133 		NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6134 
6135 		/* Setup for GRO checksum validation */
6136 		switch (skb->ip_summed) {
6137 		case CHECKSUM_COMPLETE:
6138 			NAPI_GRO_CB(skb)->csum = skb->csum;
6139 			NAPI_GRO_CB(skb)->csum_valid = 1;
6140 			NAPI_GRO_CB(skb)->csum_cnt = 0;
6141 			break;
6142 		case CHECKSUM_UNNECESSARY:
6143 			NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6144 			NAPI_GRO_CB(skb)->csum_valid = 0;
6145 			break;
6146 		default:
6147 			NAPI_GRO_CB(skb)->csum_cnt = 0;
6148 			NAPI_GRO_CB(skb)->csum_valid = 0;
6149 		}
6150 
6151 		pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6152 					ipv6_gro_receive, inet_gro_receive,
6153 					&gro_list->list, skb);
6154 		break;
6155 	}
6156 	rcu_read_unlock();
6157 
6158 	if (&ptype->list == head)
6159 		goto normal;
6160 
6161 	if (PTR_ERR(pp) == -EINPROGRESS) {
6162 		ret = GRO_CONSUMED;
6163 		goto ok;
6164 	}
6165 
6166 	same_flow = NAPI_GRO_CB(skb)->same_flow;
6167 	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6168 
6169 	if (pp) {
6170 		skb_list_del_init(pp);
6171 		napi_gro_complete(napi, pp);
6172 		gro_list->count--;
6173 	}
6174 
6175 	if (same_flow)
6176 		goto ok;
6177 
6178 	if (NAPI_GRO_CB(skb)->flush)
6179 		goto normal;
6180 
6181 	if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6182 		gro_flush_oldest(napi, &gro_list->list);
6183 	else
6184 		gro_list->count++;
6185 
6186 	NAPI_GRO_CB(skb)->count = 1;
6187 	NAPI_GRO_CB(skb)->age = jiffies;
6188 	NAPI_GRO_CB(skb)->last = skb;
6189 	skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6190 	list_add(&skb->list, &gro_list->list);
6191 	ret = GRO_HELD;
6192 
6193 pull:
6194 	grow = skb_gro_offset(skb) - skb_headlen(skb);
6195 	if (grow > 0)
6196 		gro_pull_from_frag0(skb, grow);
6197 ok:
6198 	if (gro_list->count) {
6199 		if (!test_bit(bucket, &napi->gro_bitmask))
6200 			__set_bit(bucket, &napi->gro_bitmask);
6201 	} else if (test_bit(bucket, &napi->gro_bitmask)) {
6202 		__clear_bit(bucket, &napi->gro_bitmask);
6203 	}
6204 
6205 	return ret;
6206 
6207 normal:
6208 	ret = GRO_NORMAL;
6209 	goto pull;
6210 }
6211 
6212 struct packet_offload *gro_find_receive_by_type(__be16 type)
6213 {
6214 	struct list_head *offload_head = &offload_base;
6215 	struct packet_offload *ptype;
6216 
6217 	list_for_each_entry_rcu(ptype, offload_head, list) {
6218 		if (ptype->type != type || !ptype->callbacks.gro_receive)
6219 			continue;
6220 		return ptype;
6221 	}
6222 	return NULL;
6223 }
6224 EXPORT_SYMBOL(gro_find_receive_by_type);
6225 
6226 struct packet_offload *gro_find_complete_by_type(__be16 type)
6227 {
6228 	struct list_head *offload_head = &offload_base;
6229 	struct packet_offload *ptype;
6230 
6231 	list_for_each_entry_rcu(ptype, offload_head, list) {
6232 		if (ptype->type != type || !ptype->callbacks.gro_complete)
6233 			continue;
6234 		return ptype;
6235 	}
6236 	return NULL;
6237 }
6238 EXPORT_SYMBOL(gro_find_complete_by_type);
6239 
6240 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6241 				    struct sk_buff *skb,
6242 				    gro_result_t ret)
6243 {
6244 	switch (ret) {
6245 	case GRO_NORMAL:
6246 		gro_normal_one(napi, skb, 1);
6247 		break;
6248 
6249 	case GRO_MERGED_FREE:
6250 		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6251 			napi_skb_free_stolen_head(skb);
6252 		else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6253 			__kfree_skb(skb);
6254 		else
6255 			__kfree_skb_defer(skb);
6256 		break;
6257 
6258 	case GRO_HELD:
6259 	case GRO_MERGED:
6260 	case GRO_CONSUMED:
6261 		break;
6262 	}
6263 
6264 	return ret;
6265 }
6266 
6267 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6268 {
6269 	gro_result_t ret;
6270 
6271 	skb_mark_napi_id(skb, napi);
6272 	trace_napi_gro_receive_entry(skb);
6273 
6274 	skb_gro_reset_offset(skb, 0);
6275 
6276 	ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6277 	trace_napi_gro_receive_exit(ret);
6278 
6279 	return ret;
6280 }
6281 EXPORT_SYMBOL(napi_gro_receive);
6282 
6283 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6284 {
6285 	if (unlikely(skb->pfmemalloc)) {
6286 		consume_skb(skb);
6287 		return;
6288 	}
6289 	__skb_pull(skb, skb_headlen(skb));
6290 	/* restore the reserve we had after netdev_alloc_skb_ip_align() */
6291 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6292 	__vlan_hwaccel_clear_tag(skb);
6293 	skb->dev = napi->dev;
6294 	skb->skb_iif = 0;
6295 
6296 	/* eth_type_trans() assumes pkt_type is PACKET_HOST */
6297 	skb->pkt_type = PACKET_HOST;
6298 
6299 	skb->encapsulation = 0;
6300 	skb_shinfo(skb)->gso_type = 0;
6301 	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6302 	skb_ext_reset(skb);
6303 	nf_reset_ct(skb);
6304 
6305 	napi->skb = skb;
6306 }
6307 
6308 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6309 {
6310 	struct sk_buff *skb = napi->skb;
6311 
6312 	if (!skb) {
6313 		skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6314 		if (skb) {
6315 			napi->skb = skb;
6316 			skb_mark_napi_id(skb, napi);
6317 		}
6318 	}
6319 	return skb;
6320 }
6321 EXPORT_SYMBOL(napi_get_frags);
6322 
6323 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6324 				      struct sk_buff *skb,
6325 				      gro_result_t ret)
6326 {
6327 	switch (ret) {
6328 	case GRO_NORMAL:
6329 	case GRO_HELD:
6330 		__skb_push(skb, ETH_HLEN);
6331 		skb->protocol = eth_type_trans(skb, skb->dev);
6332 		if (ret == GRO_NORMAL)
6333 			gro_normal_one(napi, skb, 1);
6334 		break;
6335 
6336 	case GRO_MERGED_FREE:
6337 		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6338 			napi_skb_free_stolen_head(skb);
6339 		else
6340 			napi_reuse_skb(napi, skb);
6341 		break;
6342 
6343 	case GRO_MERGED:
6344 	case GRO_CONSUMED:
6345 		break;
6346 	}
6347 
6348 	return ret;
6349 }
6350 
6351 /* Upper GRO stack assumes network header starts at gro_offset=0
6352  * Drivers could call both napi_gro_frags() and napi_gro_receive()
6353  * We copy ethernet header into skb->data to have a common layout.
6354  */
6355 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6356 {
6357 	struct sk_buff *skb = napi->skb;
6358 	const struct ethhdr *eth;
6359 	unsigned int hlen = sizeof(*eth);
6360 
6361 	napi->skb = NULL;
6362 
6363 	skb_reset_mac_header(skb);
6364 	skb_gro_reset_offset(skb, hlen);
6365 
6366 	if (unlikely(skb_gro_header_hard(skb, hlen))) {
6367 		eth = skb_gro_header_slow(skb, hlen, 0);
6368 		if (unlikely(!eth)) {
6369 			net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6370 					     __func__, napi->dev->name);
6371 			napi_reuse_skb(napi, skb);
6372 			return NULL;
6373 		}
6374 	} else {
6375 		eth = (const struct ethhdr *)skb->data;
6376 		gro_pull_from_frag0(skb, hlen);
6377 		NAPI_GRO_CB(skb)->frag0 += hlen;
6378 		NAPI_GRO_CB(skb)->frag0_len -= hlen;
6379 	}
6380 	__skb_pull(skb, hlen);
6381 
6382 	/*
6383 	 * This works because the only protocols we care about don't require
6384 	 * special handling.
6385 	 * We'll fix it up properly in napi_frags_finish()
6386 	 */
6387 	skb->protocol = eth->h_proto;
6388 
6389 	return skb;
6390 }
6391 
6392 gro_result_t napi_gro_frags(struct napi_struct *napi)
6393 {
6394 	gro_result_t ret;
6395 	struct sk_buff *skb = napi_frags_skb(napi);
6396 
6397 	trace_napi_gro_frags_entry(skb);
6398 
6399 	ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6400 	trace_napi_gro_frags_exit(ret);
6401 
6402 	return ret;
6403 }
6404 EXPORT_SYMBOL(napi_gro_frags);
6405 
6406 /* Compute the checksum from gro_offset and return the folded value
6407  * after adding in any pseudo checksum.
6408  */
6409 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6410 {
6411 	__wsum wsum;
6412 	__sum16 sum;
6413 
6414 	wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6415 
6416 	/* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6417 	sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6418 	/* See comments in __skb_checksum_complete(). */
6419 	if (likely(!sum)) {
6420 		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6421 		    !skb->csum_complete_sw)
6422 			netdev_rx_csum_fault(skb->dev, skb);
6423 	}
6424 
6425 	NAPI_GRO_CB(skb)->csum = wsum;
6426 	NAPI_GRO_CB(skb)->csum_valid = 1;
6427 
6428 	return sum;
6429 }
6430 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6431 
6432 static void net_rps_send_ipi(struct softnet_data *remsd)
6433 {
6434 #ifdef CONFIG_RPS
6435 	while (remsd) {
6436 		struct softnet_data *next = remsd->rps_ipi_next;
6437 
6438 		if (cpu_online(remsd->cpu))
6439 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
6440 		remsd = next;
6441 	}
6442 #endif
6443 }
6444 
6445 /*
6446  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6447  * Note: called with local irq disabled, but exits with local irq enabled.
6448  */
6449 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6450 {
6451 #ifdef CONFIG_RPS
6452 	struct softnet_data *remsd = sd->rps_ipi_list;
6453 
6454 	if (remsd) {
6455 		sd->rps_ipi_list = NULL;
6456 
6457 		local_irq_enable();
6458 
6459 		/* Send pending IPI's to kick RPS processing on remote cpus. */
6460 		net_rps_send_ipi(remsd);
6461 	} else
6462 #endif
6463 		local_irq_enable();
6464 }
6465 
6466 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6467 {
6468 #ifdef CONFIG_RPS
6469 	return sd->rps_ipi_list != NULL;
6470 #else
6471 	return false;
6472 #endif
6473 }
6474 
6475 static int process_backlog(struct napi_struct *napi, int quota)
6476 {
6477 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6478 	bool again = true;
6479 	int work = 0;
6480 
6481 	/* Check if we have pending ipi, its better to send them now,
6482 	 * not waiting net_rx_action() end.
6483 	 */
6484 	if (sd_has_rps_ipi_waiting(sd)) {
6485 		local_irq_disable();
6486 		net_rps_action_and_irq_enable(sd);
6487 	}
6488 
6489 	napi->weight = dev_rx_weight;
6490 	while (again) {
6491 		struct sk_buff *skb;
6492 
6493 		while ((skb = __skb_dequeue(&sd->process_queue))) {
6494 			rcu_read_lock();
6495 			__netif_receive_skb(skb);
6496 			rcu_read_unlock();
6497 			input_queue_head_incr(sd);
6498 			if (++work >= quota)
6499 				return work;
6500 
6501 		}
6502 
6503 		local_irq_disable();
6504 		rps_lock(sd);
6505 		if (skb_queue_empty(&sd->input_pkt_queue)) {
6506 			/*
6507 			 * Inline a custom version of __napi_complete().
6508 			 * only current cpu owns and manipulates this napi,
6509 			 * and NAPI_STATE_SCHED is the only possible flag set
6510 			 * on backlog.
6511 			 * We can use a plain write instead of clear_bit(),
6512 			 * and we dont need an smp_mb() memory barrier.
6513 			 */
6514 			napi->state = 0;
6515 			again = false;
6516 		} else {
6517 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
6518 						   &sd->process_queue);
6519 		}
6520 		rps_unlock(sd);
6521 		local_irq_enable();
6522 	}
6523 
6524 	return work;
6525 }
6526 
6527 /**
6528  * __napi_schedule - schedule for receive
6529  * @n: entry to schedule
6530  *
6531  * The entry's receive function will be scheduled to run.
6532  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6533  */
6534 void __napi_schedule(struct napi_struct *n)
6535 {
6536 	unsigned long flags;
6537 
6538 	local_irq_save(flags);
6539 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
6540 	local_irq_restore(flags);
6541 }
6542 EXPORT_SYMBOL(__napi_schedule);
6543 
6544 /**
6545  *	napi_schedule_prep - check if napi can be scheduled
6546  *	@n: napi context
6547  *
6548  * Test if NAPI routine is already running, and if not mark
6549  * it as running.  This is used as a condition variable to
6550  * insure only one NAPI poll instance runs.  We also make
6551  * sure there is no pending NAPI disable.
6552  */
6553 bool napi_schedule_prep(struct napi_struct *n)
6554 {
6555 	unsigned long val, new;
6556 
6557 	do {
6558 		val = READ_ONCE(n->state);
6559 		if (unlikely(val & NAPIF_STATE_DISABLE))
6560 			return false;
6561 		new = val | NAPIF_STATE_SCHED;
6562 
6563 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
6564 		 * This was suggested by Alexander Duyck, as compiler
6565 		 * emits better code than :
6566 		 * if (val & NAPIF_STATE_SCHED)
6567 		 *     new |= NAPIF_STATE_MISSED;
6568 		 */
6569 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6570 						   NAPIF_STATE_MISSED;
6571 	} while (cmpxchg(&n->state, val, new) != val);
6572 
6573 	return !(val & NAPIF_STATE_SCHED);
6574 }
6575 EXPORT_SYMBOL(napi_schedule_prep);
6576 
6577 /**
6578  * __napi_schedule_irqoff - schedule for receive
6579  * @n: entry to schedule
6580  *
6581  * Variant of __napi_schedule() assuming hard irqs are masked.
6582  *
6583  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6584  * because the interrupt disabled assumption might not be true
6585  * due to force-threaded interrupts and spinlock substitution.
6586  */
6587 void __napi_schedule_irqoff(struct napi_struct *n)
6588 {
6589 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6590 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6591 	else
6592 		__napi_schedule(n);
6593 }
6594 EXPORT_SYMBOL(__napi_schedule_irqoff);
6595 
6596 bool napi_complete_done(struct napi_struct *n, int work_done)
6597 {
6598 	unsigned long flags, val, new, timeout = 0;
6599 	bool ret = true;
6600 
6601 	/*
6602 	 * 1) Don't let napi dequeue from the cpu poll list
6603 	 *    just in case its running on a different cpu.
6604 	 * 2) If we are busy polling, do nothing here, we have
6605 	 *    the guarantee we will be called later.
6606 	 */
6607 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6608 				 NAPIF_STATE_IN_BUSY_POLL)))
6609 		return false;
6610 
6611 	if (work_done) {
6612 		if (n->gro_bitmask)
6613 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6614 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6615 	}
6616 	if (n->defer_hard_irqs_count > 0) {
6617 		n->defer_hard_irqs_count--;
6618 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6619 		if (timeout)
6620 			ret = false;
6621 	}
6622 	if (n->gro_bitmask) {
6623 		/* When the NAPI instance uses a timeout and keeps postponing
6624 		 * it, we need to bound somehow the time packets are kept in
6625 		 * the GRO layer
6626 		 */
6627 		napi_gro_flush(n, !!timeout);
6628 	}
6629 
6630 	gro_normal_list(n);
6631 
6632 	if (unlikely(!list_empty(&n->poll_list))) {
6633 		/* If n->poll_list is not empty, we need to mask irqs */
6634 		local_irq_save(flags);
6635 		list_del_init(&n->poll_list);
6636 		local_irq_restore(flags);
6637 	}
6638 
6639 	do {
6640 		val = READ_ONCE(n->state);
6641 
6642 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6643 
6644 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6645 			      NAPIF_STATE_SCHED_THREADED |
6646 			      NAPIF_STATE_PREFER_BUSY_POLL);
6647 
6648 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6649 		 * because we will call napi->poll() one more time.
6650 		 * This C code was suggested by Alexander Duyck to help gcc.
6651 		 */
6652 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6653 						    NAPIF_STATE_SCHED;
6654 	} while (cmpxchg(&n->state, val, new) != val);
6655 
6656 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6657 		__napi_schedule(n);
6658 		return false;
6659 	}
6660 
6661 	if (timeout)
6662 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6663 			      HRTIMER_MODE_REL_PINNED);
6664 	return ret;
6665 }
6666 EXPORT_SYMBOL(napi_complete_done);
6667 
6668 /* must be called under rcu_read_lock(), as we dont take a reference */
6669 static struct napi_struct *napi_by_id(unsigned int napi_id)
6670 {
6671 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6672 	struct napi_struct *napi;
6673 
6674 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6675 		if (napi->napi_id == napi_id)
6676 			return napi;
6677 
6678 	return NULL;
6679 }
6680 
6681 #if defined(CONFIG_NET_RX_BUSY_POLL)
6682 
6683 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6684 {
6685 	if (!skip_schedule) {
6686 		gro_normal_list(napi);
6687 		__napi_schedule(napi);
6688 		return;
6689 	}
6690 
6691 	if (napi->gro_bitmask) {
6692 		/* flush too old packets
6693 		 * If HZ < 1000, flush all packets.
6694 		 */
6695 		napi_gro_flush(napi, HZ >= 1000);
6696 	}
6697 
6698 	gro_normal_list(napi);
6699 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6700 }
6701 
6702 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6703 			   u16 budget)
6704 {
6705 	bool skip_schedule = false;
6706 	unsigned long timeout;
6707 	int rc;
6708 
6709 	/* Busy polling means there is a high chance device driver hard irq
6710 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6711 	 * set in napi_schedule_prep().
6712 	 * Since we are about to call napi->poll() once more, we can safely
6713 	 * clear NAPI_STATE_MISSED.
6714 	 *
6715 	 * Note: x86 could use a single "lock and ..." instruction
6716 	 * to perform these two clear_bit()
6717 	 */
6718 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6719 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6720 
6721 	local_bh_disable();
6722 
6723 	if (prefer_busy_poll) {
6724 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6725 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6726 		if (napi->defer_hard_irqs_count && timeout) {
6727 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6728 			skip_schedule = true;
6729 		}
6730 	}
6731 
6732 	/* All we really want here is to re-enable device interrupts.
6733 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6734 	 */
6735 	rc = napi->poll(napi, budget);
6736 	/* We can't gro_normal_list() here, because napi->poll() might have
6737 	 * rearmed the napi (napi_complete_done()) in which case it could
6738 	 * already be running on another CPU.
6739 	 */
6740 	trace_napi_poll(napi, rc, budget);
6741 	netpoll_poll_unlock(have_poll_lock);
6742 	if (rc == budget)
6743 		__busy_poll_stop(napi, skip_schedule);
6744 	local_bh_enable();
6745 }
6746 
6747 void napi_busy_loop(unsigned int napi_id,
6748 		    bool (*loop_end)(void *, unsigned long),
6749 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6750 {
6751 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6752 	int (*napi_poll)(struct napi_struct *napi, int budget);
6753 	void *have_poll_lock = NULL;
6754 	struct napi_struct *napi;
6755 
6756 restart:
6757 	napi_poll = NULL;
6758 
6759 	rcu_read_lock();
6760 
6761 	napi = napi_by_id(napi_id);
6762 	if (!napi)
6763 		goto out;
6764 
6765 	preempt_disable();
6766 	for (;;) {
6767 		int work = 0;
6768 
6769 		local_bh_disable();
6770 		if (!napi_poll) {
6771 			unsigned long val = READ_ONCE(napi->state);
6772 
6773 			/* If multiple threads are competing for this napi,
6774 			 * we avoid dirtying napi->state as much as we can.
6775 			 */
6776 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6777 				   NAPIF_STATE_IN_BUSY_POLL)) {
6778 				if (prefer_busy_poll)
6779 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6780 				goto count;
6781 			}
6782 			if (cmpxchg(&napi->state, val,
6783 				    val | NAPIF_STATE_IN_BUSY_POLL |
6784 					  NAPIF_STATE_SCHED) != val) {
6785 				if (prefer_busy_poll)
6786 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6787 				goto count;
6788 			}
6789 			have_poll_lock = netpoll_poll_lock(napi);
6790 			napi_poll = napi->poll;
6791 		}
6792 		work = napi_poll(napi, budget);
6793 		trace_napi_poll(napi, work, budget);
6794 		gro_normal_list(napi);
6795 count:
6796 		if (work > 0)
6797 			__NET_ADD_STATS(dev_net(napi->dev),
6798 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6799 		local_bh_enable();
6800 
6801 		if (!loop_end || loop_end(loop_end_arg, start_time))
6802 			break;
6803 
6804 		if (unlikely(need_resched())) {
6805 			if (napi_poll)
6806 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6807 			preempt_enable();
6808 			rcu_read_unlock();
6809 			cond_resched();
6810 			if (loop_end(loop_end_arg, start_time))
6811 				return;
6812 			goto restart;
6813 		}
6814 		cpu_relax();
6815 	}
6816 	if (napi_poll)
6817 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6818 	preempt_enable();
6819 out:
6820 	rcu_read_unlock();
6821 }
6822 EXPORT_SYMBOL(napi_busy_loop);
6823 
6824 #endif /* CONFIG_NET_RX_BUSY_POLL */
6825 
6826 static void napi_hash_add(struct napi_struct *napi)
6827 {
6828 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6829 		return;
6830 
6831 	spin_lock(&napi_hash_lock);
6832 
6833 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6834 	do {
6835 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6836 			napi_gen_id = MIN_NAPI_ID;
6837 	} while (napi_by_id(napi_gen_id));
6838 	napi->napi_id = napi_gen_id;
6839 
6840 	hlist_add_head_rcu(&napi->napi_hash_node,
6841 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6842 
6843 	spin_unlock(&napi_hash_lock);
6844 }
6845 
6846 /* Warning : caller is responsible to make sure rcu grace period
6847  * is respected before freeing memory containing @napi
6848  */
6849 static void napi_hash_del(struct napi_struct *napi)
6850 {
6851 	spin_lock(&napi_hash_lock);
6852 
6853 	hlist_del_init_rcu(&napi->napi_hash_node);
6854 
6855 	spin_unlock(&napi_hash_lock);
6856 }
6857 
6858 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6859 {
6860 	struct napi_struct *napi;
6861 
6862 	napi = container_of(timer, struct napi_struct, timer);
6863 
6864 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6865 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6866 	 */
6867 	if (!napi_disable_pending(napi) &&
6868 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6869 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6870 		__napi_schedule_irqoff(napi);
6871 	}
6872 
6873 	return HRTIMER_NORESTART;
6874 }
6875 
6876 static void init_gro_hash(struct napi_struct *napi)
6877 {
6878 	int i;
6879 
6880 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6881 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6882 		napi->gro_hash[i].count = 0;
6883 	}
6884 	napi->gro_bitmask = 0;
6885 }
6886 
6887 int dev_set_threaded(struct net_device *dev, bool threaded)
6888 {
6889 	struct napi_struct *napi;
6890 	int err = 0;
6891 
6892 	if (dev->threaded == threaded)
6893 		return 0;
6894 
6895 	if (threaded) {
6896 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6897 			if (!napi->thread) {
6898 				err = napi_kthread_create(napi);
6899 				if (err) {
6900 					threaded = false;
6901 					break;
6902 				}
6903 			}
6904 		}
6905 	}
6906 
6907 	dev->threaded = threaded;
6908 
6909 	/* Make sure kthread is created before THREADED bit
6910 	 * is set.
6911 	 */
6912 	smp_mb__before_atomic();
6913 
6914 	/* Setting/unsetting threaded mode on a napi might not immediately
6915 	 * take effect, if the current napi instance is actively being
6916 	 * polled. In this case, the switch between threaded mode and
6917 	 * softirq mode will happen in the next round of napi_schedule().
6918 	 * This should not cause hiccups/stalls to the live traffic.
6919 	 */
6920 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6921 		if (threaded)
6922 			set_bit(NAPI_STATE_THREADED, &napi->state);
6923 		else
6924 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6925 	}
6926 
6927 	return err;
6928 }
6929 EXPORT_SYMBOL(dev_set_threaded);
6930 
6931 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6932 		    int (*poll)(struct napi_struct *, int), int weight)
6933 {
6934 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6935 		return;
6936 
6937 	INIT_LIST_HEAD(&napi->poll_list);
6938 	INIT_HLIST_NODE(&napi->napi_hash_node);
6939 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6940 	napi->timer.function = napi_watchdog;
6941 	init_gro_hash(napi);
6942 	napi->skb = NULL;
6943 	INIT_LIST_HEAD(&napi->rx_list);
6944 	napi->rx_count = 0;
6945 	napi->poll = poll;
6946 	if (weight > NAPI_POLL_WEIGHT)
6947 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6948 				weight);
6949 	napi->weight = weight;
6950 	napi->dev = dev;
6951 #ifdef CONFIG_NETPOLL
6952 	napi->poll_owner = -1;
6953 #endif
6954 	set_bit(NAPI_STATE_SCHED, &napi->state);
6955 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6956 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6957 	napi_hash_add(napi);
6958 	/* Create kthread for this napi if dev->threaded is set.
6959 	 * Clear dev->threaded if kthread creation failed so that
6960 	 * threaded mode will not be enabled in napi_enable().
6961 	 */
6962 	if (dev->threaded && napi_kthread_create(napi))
6963 		dev->threaded = 0;
6964 }
6965 EXPORT_SYMBOL(netif_napi_add);
6966 
6967 void napi_disable(struct napi_struct *n)
6968 {
6969 	might_sleep();
6970 	set_bit(NAPI_STATE_DISABLE, &n->state);
6971 
6972 	while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6973 		msleep(1);
6974 	while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6975 		msleep(1);
6976 
6977 	hrtimer_cancel(&n->timer);
6978 
6979 	clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6980 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6981 	clear_bit(NAPI_STATE_THREADED, &n->state);
6982 }
6983 EXPORT_SYMBOL(napi_disable);
6984 
6985 /**
6986  *	napi_enable - enable NAPI scheduling
6987  *	@n: NAPI context
6988  *
6989  * Resume NAPI from being scheduled on this context.
6990  * Must be paired with napi_disable.
6991  */
6992 void napi_enable(struct napi_struct *n)
6993 {
6994 	BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
6995 	smp_mb__before_atomic();
6996 	clear_bit(NAPI_STATE_SCHED, &n->state);
6997 	clear_bit(NAPI_STATE_NPSVC, &n->state);
6998 	if (n->dev->threaded && n->thread)
6999 		set_bit(NAPI_STATE_THREADED, &n->state);
7000 }
7001 EXPORT_SYMBOL(napi_enable);
7002 
7003 static void flush_gro_hash(struct napi_struct *napi)
7004 {
7005 	int i;
7006 
7007 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
7008 		struct sk_buff *skb, *n;
7009 
7010 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
7011 			kfree_skb(skb);
7012 		napi->gro_hash[i].count = 0;
7013 	}
7014 }
7015 
7016 /* Must be called in process context */
7017 void __netif_napi_del(struct napi_struct *napi)
7018 {
7019 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
7020 		return;
7021 
7022 	napi_hash_del(napi);
7023 	list_del_rcu(&napi->dev_list);
7024 	napi_free_frags(napi);
7025 
7026 	flush_gro_hash(napi);
7027 	napi->gro_bitmask = 0;
7028 
7029 	if (napi->thread) {
7030 		kthread_stop(napi->thread);
7031 		napi->thread = NULL;
7032 	}
7033 }
7034 EXPORT_SYMBOL(__netif_napi_del);
7035 
7036 static int __napi_poll(struct napi_struct *n, bool *repoll)
7037 {
7038 	int work, weight;
7039 
7040 	weight = n->weight;
7041 
7042 	/* This NAPI_STATE_SCHED test is for avoiding a race
7043 	 * with netpoll's poll_napi().  Only the entity which
7044 	 * obtains the lock and sees NAPI_STATE_SCHED set will
7045 	 * actually make the ->poll() call.  Therefore we avoid
7046 	 * accidentally calling ->poll() when NAPI is not scheduled.
7047 	 */
7048 	work = 0;
7049 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
7050 		work = n->poll(n, weight);
7051 		trace_napi_poll(n, work, weight);
7052 	}
7053 
7054 	if (unlikely(work > weight))
7055 		pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7056 			    n->poll, work, weight);
7057 
7058 	if (likely(work < weight))
7059 		return work;
7060 
7061 	/* Drivers must not modify the NAPI state if they
7062 	 * consume the entire weight.  In such cases this code
7063 	 * still "owns" the NAPI instance and therefore can
7064 	 * move the instance around on the list at-will.
7065 	 */
7066 	if (unlikely(napi_disable_pending(n))) {
7067 		napi_complete(n);
7068 		return work;
7069 	}
7070 
7071 	/* The NAPI context has more processing work, but busy-polling
7072 	 * is preferred. Exit early.
7073 	 */
7074 	if (napi_prefer_busy_poll(n)) {
7075 		if (napi_complete_done(n, work)) {
7076 			/* If timeout is not set, we need to make sure
7077 			 * that the NAPI is re-scheduled.
7078 			 */
7079 			napi_schedule(n);
7080 		}
7081 		return work;
7082 	}
7083 
7084 	if (n->gro_bitmask) {
7085 		/* flush too old packets
7086 		 * If HZ < 1000, flush all packets.
7087 		 */
7088 		napi_gro_flush(n, HZ >= 1000);
7089 	}
7090 
7091 	gro_normal_list(n);
7092 
7093 	/* Some drivers may have called napi_schedule
7094 	 * prior to exhausting their budget.
7095 	 */
7096 	if (unlikely(!list_empty(&n->poll_list))) {
7097 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7098 			     n->dev ? n->dev->name : "backlog");
7099 		return work;
7100 	}
7101 
7102 	*repoll = true;
7103 
7104 	return work;
7105 }
7106 
7107 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7108 {
7109 	bool do_repoll = false;
7110 	void *have;
7111 	int work;
7112 
7113 	list_del_init(&n->poll_list);
7114 
7115 	have = netpoll_poll_lock(n);
7116 
7117 	work = __napi_poll(n, &do_repoll);
7118 
7119 	if (do_repoll)
7120 		list_add_tail(&n->poll_list, repoll);
7121 
7122 	netpoll_poll_unlock(have);
7123 
7124 	return work;
7125 }
7126 
7127 static int napi_thread_wait(struct napi_struct *napi)
7128 {
7129 	bool woken = false;
7130 
7131 	set_current_state(TASK_INTERRUPTIBLE);
7132 
7133 	while (!kthread_should_stop()) {
7134 		/* Testing SCHED_THREADED bit here to make sure the current
7135 		 * kthread owns this napi and could poll on this napi.
7136 		 * Testing SCHED bit is not enough because SCHED bit might be
7137 		 * set by some other busy poll thread or by napi_disable().
7138 		 */
7139 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7140 			WARN_ON(!list_empty(&napi->poll_list));
7141 			__set_current_state(TASK_RUNNING);
7142 			return 0;
7143 		}
7144 
7145 		schedule();
7146 		/* woken being true indicates this thread owns this napi. */
7147 		woken = true;
7148 		set_current_state(TASK_INTERRUPTIBLE);
7149 	}
7150 	__set_current_state(TASK_RUNNING);
7151 
7152 	return -1;
7153 }
7154 
7155 static int napi_threaded_poll(void *data)
7156 {
7157 	struct napi_struct *napi = data;
7158 	void *have;
7159 
7160 	while (!napi_thread_wait(napi)) {
7161 		for (;;) {
7162 			bool repoll = false;
7163 
7164 			local_bh_disable();
7165 
7166 			have = netpoll_poll_lock(napi);
7167 			__napi_poll(napi, &repoll);
7168 			netpoll_poll_unlock(have);
7169 
7170 			local_bh_enable();
7171 
7172 			if (!repoll)
7173 				break;
7174 
7175 			cond_resched();
7176 		}
7177 	}
7178 	return 0;
7179 }
7180 
7181 static __latent_entropy void net_rx_action(struct softirq_action *h)
7182 {
7183 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7184 	unsigned long time_limit = jiffies +
7185 		usecs_to_jiffies(netdev_budget_usecs);
7186 	int budget = netdev_budget;
7187 	LIST_HEAD(list);
7188 	LIST_HEAD(repoll);
7189 
7190 	local_irq_disable();
7191 	list_splice_init(&sd->poll_list, &list);
7192 	local_irq_enable();
7193 
7194 	for (;;) {
7195 		struct napi_struct *n;
7196 
7197 		if (list_empty(&list)) {
7198 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7199 				return;
7200 			break;
7201 		}
7202 
7203 		n = list_first_entry(&list, struct napi_struct, poll_list);
7204 		budget -= napi_poll(n, &repoll);
7205 
7206 		/* If softirq window is exhausted then punt.
7207 		 * Allow this to run for 2 jiffies since which will allow
7208 		 * an average latency of 1.5/HZ.
7209 		 */
7210 		if (unlikely(budget <= 0 ||
7211 			     time_after_eq(jiffies, time_limit))) {
7212 			sd->time_squeeze++;
7213 			break;
7214 		}
7215 	}
7216 
7217 	local_irq_disable();
7218 
7219 	list_splice_tail_init(&sd->poll_list, &list);
7220 	list_splice_tail(&repoll, &list);
7221 	list_splice(&list, &sd->poll_list);
7222 	if (!list_empty(&sd->poll_list))
7223 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
7224 
7225 	net_rps_action_and_irq_enable(sd);
7226 }
7227 
7228 struct netdev_adjacent {
7229 	struct net_device *dev;
7230 
7231 	/* upper master flag, there can only be one master device per list */
7232 	bool master;
7233 
7234 	/* lookup ignore flag */
7235 	bool ignore;
7236 
7237 	/* counter for the number of times this device was added to us */
7238 	u16 ref_nr;
7239 
7240 	/* private field for the users */
7241 	void *private;
7242 
7243 	struct list_head list;
7244 	struct rcu_head rcu;
7245 };
7246 
7247 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7248 						 struct list_head *adj_list)
7249 {
7250 	struct netdev_adjacent *adj;
7251 
7252 	list_for_each_entry(adj, adj_list, list) {
7253 		if (adj->dev == adj_dev)
7254 			return adj;
7255 	}
7256 	return NULL;
7257 }
7258 
7259 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7260 				    struct netdev_nested_priv *priv)
7261 {
7262 	struct net_device *dev = (struct net_device *)priv->data;
7263 
7264 	return upper_dev == dev;
7265 }
7266 
7267 /**
7268  * netdev_has_upper_dev - Check if device is linked to an upper device
7269  * @dev: device
7270  * @upper_dev: upper device to check
7271  *
7272  * Find out if a device is linked to specified upper device and return true
7273  * in case it is. Note that this checks only immediate upper device,
7274  * not through a complete stack of devices. The caller must hold the RTNL lock.
7275  */
7276 bool netdev_has_upper_dev(struct net_device *dev,
7277 			  struct net_device *upper_dev)
7278 {
7279 	struct netdev_nested_priv priv = {
7280 		.data = (void *)upper_dev,
7281 	};
7282 
7283 	ASSERT_RTNL();
7284 
7285 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7286 					     &priv);
7287 }
7288 EXPORT_SYMBOL(netdev_has_upper_dev);
7289 
7290 /**
7291  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7292  * @dev: device
7293  * @upper_dev: upper device to check
7294  *
7295  * Find out if a device is linked to specified upper device and return true
7296  * in case it is. Note that this checks the entire upper device chain.
7297  * The caller must hold rcu lock.
7298  */
7299 
7300 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7301 				  struct net_device *upper_dev)
7302 {
7303 	struct netdev_nested_priv priv = {
7304 		.data = (void *)upper_dev,
7305 	};
7306 
7307 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7308 					       &priv);
7309 }
7310 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7311 
7312 /**
7313  * netdev_has_any_upper_dev - Check if device is linked to some device
7314  * @dev: device
7315  *
7316  * Find out if a device is linked to an upper device and return true in case
7317  * it is. The caller must hold the RTNL lock.
7318  */
7319 bool netdev_has_any_upper_dev(struct net_device *dev)
7320 {
7321 	ASSERT_RTNL();
7322 
7323 	return !list_empty(&dev->adj_list.upper);
7324 }
7325 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7326 
7327 /**
7328  * netdev_master_upper_dev_get - Get master upper device
7329  * @dev: device
7330  *
7331  * Find a master upper device and return pointer to it or NULL in case
7332  * it's not there. The caller must hold the RTNL lock.
7333  */
7334 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7335 {
7336 	struct netdev_adjacent *upper;
7337 
7338 	ASSERT_RTNL();
7339 
7340 	if (list_empty(&dev->adj_list.upper))
7341 		return NULL;
7342 
7343 	upper = list_first_entry(&dev->adj_list.upper,
7344 				 struct netdev_adjacent, list);
7345 	if (likely(upper->master))
7346 		return upper->dev;
7347 	return NULL;
7348 }
7349 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7350 
7351 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7352 {
7353 	struct netdev_adjacent *upper;
7354 
7355 	ASSERT_RTNL();
7356 
7357 	if (list_empty(&dev->adj_list.upper))
7358 		return NULL;
7359 
7360 	upper = list_first_entry(&dev->adj_list.upper,
7361 				 struct netdev_adjacent, list);
7362 	if (likely(upper->master) && !upper->ignore)
7363 		return upper->dev;
7364 	return NULL;
7365 }
7366 
7367 /**
7368  * netdev_has_any_lower_dev - Check if device is linked to some device
7369  * @dev: device
7370  *
7371  * Find out if a device is linked to a lower device and return true in case
7372  * it is. The caller must hold the RTNL lock.
7373  */
7374 static bool netdev_has_any_lower_dev(struct net_device *dev)
7375 {
7376 	ASSERT_RTNL();
7377 
7378 	return !list_empty(&dev->adj_list.lower);
7379 }
7380 
7381 void *netdev_adjacent_get_private(struct list_head *adj_list)
7382 {
7383 	struct netdev_adjacent *adj;
7384 
7385 	adj = list_entry(adj_list, struct netdev_adjacent, list);
7386 
7387 	return adj->private;
7388 }
7389 EXPORT_SYMBOL(netdev_adjacent_get_private);
7390 
7391 /**
7392  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7393  * @dev: device
7394  * @iter: list_head ** of the current position
7395  *
7396  * Gets the next device from the dev's upper list, starting from iter
7397  * position. The caller must hold RCU read lock.
7398  */
7399 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7400 						 struct list_head **iter)
7401 {
7402 	struct netdev_adjacent *upper;
7403 
7404 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7405 
7406 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7407 
7408 	if (&upper->list == &dev->adj_list.upper)
7409 		return NULL;
7410 
7411 	*iter = &upper->list;
7412 
7413 	return upper->dev;
7414 }
7415 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7416 
7417 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7418 						  struct list_head **iter,
7419 						  bool *ignore)
7420 {
7421 	struct netdev_adjacent *upper;
7422 
7423 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7424 
7425 	if (&upper->list == &dev->adj_list.upper)
7426 		return NULL;
7427 
7428 	*iter = &upper->list;
7429 	*ignore = upper->ignore;
7430 
7431 	return upper->dev;
7432 }
7433 
7434 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7435 						    struct list_head **iter)
7436 {
7437 	struct netdev_adjacent *upper;
7438 
7439 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7440 
7441 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7442 
7443 	if (&upper->list == &dev->adj_list.upper)
7444 		return NULL;
7445 
7446 	*iter = &upper->list;
7447 
7448 	return upper->dev;
7449 }
7450 
7451 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7452 				       int (*fn)(struct net_device *dev,
7453 					 struct netdev_nested_priv *priv),
7454 				       struct netdev_nested_priv *priv)
7455 {
7456 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7457 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7458 	int ret, cur = 0;
7459 	bool ignore;
7460 
7461 	now = dev;
7462 	iter = &dev->adj_list.upper;
7463 
7464 	while (1) {
7465 		if (now != dev) {
7466 			ret = fn(now, priv);
7467 			if (ret)
7468 				return ret;
7469 		}
7470 
7471 		next = NULL;
7472 		while (1) {
7473 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
7474 			if (!udev)
7475 				break;
7476 			if (ignore)
7477 				continue;
7478 
7479 			next = udev;
7480 			niter = &udev->adj_list.upper;
7481 			dev_stack[cur] = now;
7482 			iter_stack[cur++] = iter;
7483 			break;
7484 		}
7485 
7486 		if (!next) {
7487 			if (!cur)
7488 				return 0;
7489 			next = dev_stack[--cur];
7490 			niter = iter_stack[cur];
7491 		}
7492 
7493 		now = next;
7494 		iter = niter;
7495 	}
7496 
7497 	return 0;
7498 }
7499 
7500 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7501 				  int (*fn)(struct net_device *dev,
7502 					    struct netdev_nested_priv *priv),
7503 				  struct netdev_nested_priv *priv)
7504 {
7505 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7506 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7507 	int ret, cur = 0;
7508 
7509 	now = dev;
7510 	iter = &dev->adj_list.upper;
7511 
7512 	while (1) {
7513 		if (now != dev) {
7514 			ret = fn(now, priv);
7515 			if (ret)
7516 				return ret;
7517 		}
7518 
7519 		next = NULL;
7520 		while (1) {
7521 			udev = netdev_next_upper_dev_rcu(now, &iter);
7522 			if (!udev)
7523 				break;
7524 
7525 			next = udev;
7526 			niter = &udev->adj_list.upper;
7527 			dev_stack[cur] = now;
7528 			iter_stack[cur++] = iter;
7529 			break;
7530 		}
7531 
7532 		if (!next) {
7533 			if (!cur)
7534 				return 0;
7535 			next = dev_stack[--cur];
7536 			niter = iter_stack[cur];
7537 		}
7538 
7539 		now = next;
7540 		iter = niter;
7541 	}
7542 
7543 	return 0;
7544 }
7545 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7546 
7547 static bool __netdev_has_upper_dev(struct net_device *dev,
7548 				   struct net_device *upper_dev)
7549 {
7550 	struct netdev_nested_priv priv = {
7551 		.flags = 0,
7552 		.data = (void *)upper_dev,
7553 	};
7554 
7555 	ASSERT_RTNL();
7556 
7557 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7558 					   &priv);
7559 }
7560 
7561 /**
7562  * netdev_lower_get_next_private - Get the next ->private from the
7563  *				   lower neighbour list
7564  * @dev: device
7565  * @iter: list_head ** of the current position
7566  *
7567  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7568  * list, starting from iter position. The caller must hold either hold the
7569  * RTNL lock or its own locking that guarantees that the neighbour lower
7570  * list will remain unchanged.
7571  */
7572 void *netdev_lower_get_next_private(struct net_device *dev,
7573 				    struct list_head **iter)
7574 {
7575 	struct netdev_adjacent *lower;
7576 
7577 	lower = list_entry(*iter, struct netdev_adjacent, list);
7578 
7579 	if (&lower->list == &dev->adj_list.lower)
7580 		return NULL;
7581 
7582 	*iter = lower->list.next;
7583 
7584 	return lower->private;
7585 }
7586 EXPORT_SYMBOL(netdev_lower_get_next_private);
7587 
7588 /**
7589  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7590  *				       lower neighbour list, RCU
7591  *				       variant
7592  * @dev: device
7593  * @iter: list_head ** of the current position
7594  *
7595  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7596  * list, starting from iter position. The caller must hold RCU read lock.
7597  */
7598 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7599 					struct list_head **iter)
7600 {
7601 	struct netdev_adjacent *lower;
7602 
7603 	WARN_ON_ONCE(!rcu_read_lock_held());
7604 
7605 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7606 
7607 	if (&lower->list == &dev->adj_list.lower)
7608 		return NULL;
7609 
7610 	*iter = &lower->list;
7611 
7612 	return lower->private;
7613 }
7614 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7615 
7616 /**
7617  * netdev_lower_get_next - Get the next device from the lower neighbour
7618  *                         list
7619  * @dev: device
7620  * @iter: list_head ** of the current position
7621  *
7622  * Gets the next netdev_adjacent from the dev's lower neighbour
7623  * list, starting from iter position. The caller must hold RTNL lock or
7624  * its own locking that guarantees that the neighbour lower
7625  * list will remain unchanged.
7626  */
7627 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7628 {
7629 	struct netdev_adjacent *lower;
7630 
7631 	lower = list_entry(*iter, struct netdev_adjacent, list);
7632 
7633 	if (&lower->list == &dev->adj_list.lower)
7634 		return NULL;
7635 
7636 	*iter = lower->list.next;
7637 
7638 	return lower->dev;
7639 }
7640 EXPORT_SYMBOL(netdev_lower_get_next);
7641 
7642 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7643 						struct list_head **iter)
7644 {
7645 	struct netdev_adjacent *lower;
7646 
7647 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7648 
7649 	if (&lower->list == &dev->adj_list.lower)
7650 		return NULL;
7651 
7652 	*iter = &lower->list;
7653 
7654 	return lower->dev;
7655 }
7656 
7657 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7658 						  struct list_head **iter,
7659 						  bool *ignore)
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 	*ignore = lower->ignore;
7670 
7671 	return lower->dev;
7672 }
7673 
7674 int netdev_walk_all_lower_dev(struct net_device *dev,
7675 			      int (*fn)(struct net_device *dev,
7676 					struct netdev_nested_priv *priv),
7677 			      struct netdev_nested_priv *priv)
7678 {
7679 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7680 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7681 	int ret, cur = 0;
7682 
7683 	now = dev;
7684 	iter = &dev->adj_list.lower;
7685 
7686 	while (1) {
7687 		if (now != dev) {
7688 			ret = fn(now, priv);
7689 			if (ret)
7690 				return ret;
7691 		}
7692 
7693 		next = NULL;
7694 		while (1) {
7695 			ldev = netdev_next_lower_dev(now, &iter);
7696 			if (!ldev)
7697 				break;
7698 
7699 			next = ldev;
7700 			niter = &ldev->adj_list.lower;
7701 			dev_stack[cur] = now;
7702 			iter_stack[cur++] = iter;
7703 			break;
7704 		}
7705 
7706 		if (!next) {
7707 			if (!cur)
7708 				return 0;
7709 			next = dev_stack[--cur];
7710 			niter = iter_stack[cur];
7711 		}
7712 
7713 		now = next;
7714 		iter = niter;
7715 	}
7716 
7717 	return 0;
7718 }
7719 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7720 
7721 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7722 				       int (*fn)(struct net_device *dev,
7723 					 struct netdev_nested_priv *priv),
7724 				       struct netdev_nested_priv *priv)
7725 {
7726 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7727 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7728 	int ret, cur = 0;
7729 	bool ignore;
7730 
7731 	now = dev;
7732 	iter = &dev->adj_list.lower;
7733 
7734 	while (1) {
7735 		if (now != dev) {
7736 			ret = fn(now, priv);
7737 			if (ret)
7738 				return ret;
7739 		}
7740 
7741 		next = NULL;
7742 		while (1) {
7743 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7744 			if (!ldev)
7745 				break;
7746 			if (ignore)
7747 				continue;
7748 
7749 			next = ldev;
7750 			niter = &ldev->adj_list.lower;
7751 			dev_stack[cur] = now;
7752 			iter_stack[cur++] = iter;
7753 			break;
7754 		}
7755 
7756 		if (!next) {
7757 			if (!cur)
7758 				return 0;
7759 			next = dev_stack[--cur];
7760 			niter = iter_stack[cur];
7761 		}
7762 
7763 		now = next;
7764 		iter = niter;
7765 	}
7766 
7767 	return 0;
7768 }
7769 
7770 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7771 					     struct list_head **iter)
7772 {
7773 	struct netdev_adjacent *lower;
7774 
7775 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7776 	if (&lower->list == &dev->adj_list.lower)
7777 		return NULL;
7778 
7779 	*iter = &lower->list;
7780 
7781 	return lower->dev;
7782 }
7783 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7784 
7785 static u8 __netdev_upper_depth(struct net_device *dev)
7786 {
7787 	struct net_device *udev;
7788 	struct list_head *iter;
7789 	u8 max_depth = 0;
7790 	bool ignore;
7791 
7792 	for (iter = &dev->adj_list.upper,
7793 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7794 	     udev;
7795 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7796 		if (ignore)
7797 			continue;
7798 		if (max_depth < udev->upper_level)
7799 			max_depth = udev->upper_level;
7800 	}
7801 
7802 	return max_depth;
7803 }
7804 
7805 static u8 __netdev_lower_depth(struct net_device *dev)
7806 {
7807 	struct net_device *ldev;
7808 	struct list_head *iter;
7809 	u8 max_depth = 0;
7810 	bool ignore;
7811 
7812 	for (iter = &dev->adj_list.lower,
7813 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7814 	     ldev;
7815 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7816 		if (ignore)
7817 			continue;
7818 		if (max_depth < ldev->lower_level)
7819 			max_depth = ldev->lower_level;
7820 	}
7821 
7822 	return max_depth;
7823 }
7824 
7825 static int __netdev_update_upper_level(struct net_device *dev,
7826 				       struct netdev_nested_priv *__unused)
7827 {
7828 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7829 	return 0;
7830 }
7831 
7832 static int __netdev_update_lower_level(struct net_device *dev,
7833 				       struct netdev_nested_priv *priv)
7834 {
7835 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7836 
7837 #ifdef CONFIG_LOCKDEP
7838 	if (!priv)
7839 		return 0;
7840 
7841 	if (priv->flags & NESTED_SYNC_IMM)
7842 		dev->nested_level = dev->lower_level - 1;
7843 	if (priv->flags & NESTED_SYNC_TODO)
7844 		net_unlink_todo(dev);
7845 #endif
7846 	return 0;
7847 }
7848 
7849 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7850 				  int (*fn)(struct net_device *dev,
7851 					    struct netdev_nested_priv *priv),
7852 				  struct netdev_nested_priv *priv)
7853 {
7854 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7855 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7856 	int ret, cur = 0;
7857 
7858 	now = dev;
7859 	iter = &dev->adj_list.lower;
7860 
7861 	while (1) {
7862 		if (now != dev) {
7863 			ret = fn(now, priv);
7864 			if (ret)
7865 				return ret;
7866 		}
7867 
7868 		next = NULL;
7869 		while (1) {
7870 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7871 			if (!ldev)
7872 				break;
7873 
7874 			next = ldev;
7875 			niter = &ldev->adj_list.lower;
7876 			dev_stack[cur] = now;
7877 			iter_stack[cur++] = iter;
7878 			break;
7879 		}
7880 
7881 		if (!next) {
7882 			if (!cur)
7883 				return 0;
7884 			next = dev_stack[--cur];
7885 			niter = iter_stack[cur];
7886 		}
7887 
7888 		now = next;
7889 		iter = niter;
7890 	}
7891 
7892 	return 0;
7893 }
7894 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7895 
7896 /**
7897  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7898  *				       lower neighbour list, RCU
7899  *				       variant
7900  * @dev: device
7901  *
7902  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7903  * list. The caller must hold RCU read lock.
7904  */
7905 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7906 {
7907 	struct netdev_adjacent *lower;
7908 
7909 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7910 			struct netdev_adjacent, list);
7911 	if (lower)
7912 		return lower->private;
7913 	return NULL;
7914 }
7915 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7916 
7917 /**
7918  * netdev_master_upper_dev_get_rcu - Get master upper device
7919  * @dev: device
7920  *
7921  * Find a master upper device and return pointer to it or NULL in case
7922  * it's not there. The caller must hold the RCU read lock.
7923  */
7924 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7925 {
7926 	struct netdev_adjacent *upper;
7927 
7928 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7929 				       struct netdev_adjacent, list);
7930 	if (upper && likely(upper->master))
7931 		return upper->dev;
7932 	return NULL;
7933 }
7934 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7935 
7936 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7937 			      struct net_device *adj_dev,
7938 			      struct list_head *dev_list)
7939 {
7940 	char linkname[IFNAMSIZ+7];
7941 
7942 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7943 		"upper_%s" : "lower_%s", adj_dev->name);
7944 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7945 				 linkname);
7946 }
7947 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7948 			       char *name,
7949 			       struct list_head *dev_list)
7950 {
7951 	char linkname[IFNAMSIZ+7];
7952 
7953 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7954 		"upper_%s" : "lower_%s", name);
7955 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7956 }
7957 
7958 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7959 						 struct net_device *adj_dev,
7960 						 struct list_head *dev_list)
7961 {
7962 	return (dev_list == &dev->adj_list.upper ||
7963 		dev_list == &dev->adj_list.lower) &&
7964 		net_eq(dev_net(dev), dev_net(adj_dev));
7965 }
7966 
7967 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7968 					struct net_device *adj_dev,
7969 					struct list_head *dev_list,
7970 					void *private, bool master)
7971 {
7972 	struct netdev_adjacent *adj;
7973 	int ret;
7974 
7975 	adj = __netdev_find_adj(adj_dev, dev_list);
7976 
7977 	if (adj) {
7978 		adj->ref_nr += 1;
7979 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7980 			 dev->name, adj_dev->name, adj->ref_nr);
7981 
7982 		return 0;
7983 	}
7984 
7985 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7986 	if (!adj)
7987 		return -ENOMEM;
7988 
7989 	adj->dev = adj_dev;
7990 	adj->master = master;
7991 	adj->ref_nr = 1;
7992 	adj->private = private;
7993 	adj->ignore = false;
7994 	dev_hold(adj_dev);
7995 
7996 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7997 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7998 
7999 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
8000 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
8001 		if (ret)
8002 			goto free_adj;
8003 	}
8004 
8005 	/* Ensure that master link is always the first item in list. */
8006 	if (master) {
8007 		ret = sysfs_create_link(&(dev->dev.kobj),
8008 					&(adj_dev->dev.kobj), "master");
8009 		if (ret)
8010 			goto remove_symlinks;
8011 
8012 		list_add_rcu(&adj->list, dev_list);
8013 	} else {
8014 		list_add_tail_rcu(&adj->list, dev_list);
8015 	}
8016 
8017 	return 0;
8018 
8019 remove_symlinks:
8020 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8021 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8022 free_adj:
8023 	kfree(adj);
8024 	dev_put(adj_dev);
8025 
8026 	return ret;
8027 }
8028 
8029 static void __netdev_adjacent_dev_remove(struct net_device *dev,
8030 					 struct net_device *adj_dev,
8031 					 u16 ref_nr,
8032 					 struct list_head *dev_list)
8033 {
8034 	struct netdev_adjacent *adj;
8035 
8036 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
8037 		 dev->name, adj_dev->name, ref_nr);
8038 
8039 	adj = __netdev_find_adj(adj_dev, dev_list);
8040 
8041 	if (!adj) {
8042 		pr_err("Adjacency does not exist for device %s from %s\n",
8043 		       dev->name, adj_dev->name);
8044 		WARN_ON(1);
8045 		return;
8046 	}
8047 
8048 	if (adj->ref_nr > ref_nr) {
8049 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8050 			 dev->name, adj_dev->name, ref_nr,
8051 			 adj->ref_nr - ref_nr);
8052 		adj->ref_nr -= ref_nr;
8053 		return;
8054 	}
8055 
8056 	if (adj->master)
8057 		sysfs_remove_link(&(dev->dev.kobj), "master");
8058 
8059 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8060 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8061 
8062 	list_del_rcu(&adj->list);
8063 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8064 		 adj_dev->name, dev->name, adj_dev->name);
8065 	dev_put(adj_dev);
8066 	kfree_rcu(adj, rcu);
8067 }
8068 
8069 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8070 					    struct net_device *upper_dev,
8071 					    struct list_head *up_list,
8072 					    struct list_head *down_list,
8073 					    void *private, bool master)
8074 {
8075 	int ret;
8076 
8077 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8078 					   private, master);
8079 	if (ret)
8080 		return ret;
8081 
8082 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8083 					   private, false);
8084 	if (ret) {
8085 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8086 		return ret;
8087 	}
8088 
8089 	return 0;
8090 }
8091 
8092 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8093 					       struct net_device *upper_dev,
8094 					       u16 ref_nr,
8095 					       struct list_head *up_list,
8096 					       struct list_head *down_list)
8097 {
8098 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8099 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8100 }
8101 
8102 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8103 						struct net_device *upper_dev,
8104 						void *private, bool master)
8105 {
8106 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8107 						&dev->adj_list.upper,
8108 						&upper_dev->adj_list.lower,
8109 						private, master);
8110 }
8111 
8112 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8113 						   struct net_device *upper_dev)
8114 {
8115 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8116 					   &dev->adj_list.upper,
8117 					   &upper_dev->adj_list.lower);
8118 }
8119 
8120 static int __netdev_upper_dev_link(struct net_device *dev,
8121 				   struct net_device *upper_dev, bool master,
8122 				   void *upper_priv, void *upper_info,
8123 				   struct netdev_nested_priv *priv,
8124 				   struct netlink_ext_ack *extack)
8125 {
8126 	struct netdev_notifier_changeupper_info changeupper_info = {
8127 		.info = {
8128 			.dev = dev,
8129 			.extack = extack,
8130 		},
8131 		.upper_dev = upper_dev,
8132 		.master = master,
8133 		.linking = true,
8134 		.upper_info = upper_info,
8135 	};
8136 	struct net_device *master_dev;
8137 	int ret = 0;
8138 
8139 	ASSERT_RTNL();
8140 
8141 	if (dev == upper_dev)
8142 		return -EBUSY;
8143 
8144 	/* To prevent loops, check if dev is not upper device to upper_dev. */
8145 	if (__netdev_has_upper_dev(upper_dev, dev))
8146 		return -EBUSY;
8147 
8148 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8149 		return -EMLINK;
8150 
8151 	if (!master) {
8152 		if (__netdev_has_upper_dev(dev, upper_dev))
8153 			return -EEXIST;
8154 	} else {
8155 		master_dev = __netdev_master_upper_dev_get(dev);
8156 		if (master_dev)
8157 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
8158 	}
8159 
8160 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8161 					    &changeupper_info.info);
8162 	ret = notifier_to_errno(ret);
8163 	if (ret)
8164 		return ret;
8165 
8166 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8167 						   master);
8168 	if (ret)
8169 		return ret;
8170 
8171 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8172 					    &changeupper_info.info);
8173 	ret = notifier_to_errno(ret);
8174 	if (ret)
8175 		goto rollback;
8176 
8177 	__netdev_update_upper_level(dev, NULL);
8178 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8179 
8180 	__netdev_update_lower_level(upper_dev, priv);
8181 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8182 				    priv);
8183 
8184 	return 0;
8185 
8186 rollback:
8187 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8188 
8189 	return ret;
8190 }
8191 
8192 /**
8193  * netdev_upper_dev_link - Add a link to the upper device
8194  * @dev: device
8195  * @upper_dev: new upper device
8196  * @extack: netlink extended ack
8197  *
8198  * Adds a link to device which is upper to this one. The caller must hold
8199  * the RTNL lock. On a failure a negative errno code is returned.
8200  * On success the reference counts are adjusted and the function
8201  * returns zero.
8202  */
8203 int netdev_upper_dev_link(struct net_device *dev,
8204 			  struct net_device *upper_dev,
8205 			  struct netlink_ext_ack *extack)
8206 {
8207 	struct netdev_nested_priv priv = {
8208 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8209 		.data = NULL,
8210 	};
8211 
8212 	return __netdev_upper_dev_link(dev, upper_dev, false,
8213 				       NULL, NULL, &priv, extack);
8214 }
8215 EXPORT_SYMBOL(netdev_upper_dev_link);
8216 
8217 /**
8218  * netdev_master_upper_dev_link - Add a master link to the upper device
8219  * @dev: device
8220  * @upper_dev: new upper device
8221  * @upper_priv: upper device private
8222  * @upper_info: upper info to be passed down via notifier
8223  * @extack: netlink extended ack
8224  *
8225  * Adds a link to device which is upper to this one. In this case, only
8226  * one master upper device can be linked, although other non-master devices
8227  * might be linked as well. The caller must hold the RTNL lock.
8228  * On a failure a negative errno code is returned. On success the reference
8229  * counts are adjusted and the function returns zero.
8230  */
8231 int netdev_master_upper_dev_link(struct net_device *dev,
8232 				 struct net_device *upper_dev,
8233 				 void *upper_priv, void *upper_info,
8234 				 struct netlink_ext_ack *extack)
8235 {
8236 	struct netdev_nested_priv priv = {
8237 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8238 		.data = NULL,
8239 	};
8240 
8241 	return __netdev_upper_dev_link(dev, upper_dev, true,
8242 				       upper_priv, upper_info, &priv, extack);
8243 }
8244 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8245 
8246 static void __netdev_upper_dev_unlink(struct net_device *dev,
8247 				      struct net_device *upper_dev,
8248 				      struct netdev_nested_priv *priv)
8249 {
8250 	struct netdev_notifier_changeupper_info changeupper_info = {
8251 		.info = {
8252 			.dev = dev,
8253 		},
8254 		.upper_dev = upper_dev,
8255 		.linking = false,
8256 	};
8257 
8258 	ASSERT_RTNL();
8259 
8260 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8261 
8262 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8263 				      &changeupper_info.info);
8264 
8265 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8266 
8267 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8268 				      &changeupper_info.info);
8269 
8270 	__netdev_update_upper_level(dev, NULL);
8271 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8272 
8273 	__netdev_update_lower_level(upper_dev, priv);
8274 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8275 				    priv);
8276 }
8277 
8278 /**
8279  * netdev_upper_dev_unlink - Removes a link to upper device
8280  * @dev: device
8281  * @upper_dev: new upper device
8282  *
8283  * Removes a link to device which is upper to this one. The caller must hold
8284  * the RTNL lock.
8285  */
8286 void netdev_upper_dev_unlink(struct net_device *dev,
8287 			     struct net_device *upper_dev)
8288 {
8289 	struct netdev_nested_priv priv = {
8290 		.flags = NESTED_SYNC_TODO,
8291 		.data = NULL,
8292 	};
8293 
8294 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
8295 }
8296 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8297 
8298 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8299 				      struct net_device *lower_dev,
8300 				      bool val)
8301 {
8302 	struct netdev_adjacent *adj;
8303 
8304 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8305 	if (adj)
8306 		adj->ignore = val;
8307 
8308 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8309 	if (adj)
8310 		adj->ignore = val;
8311 }
8312 
8313 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8314 					struct net_device *lower_dev)
8315 {
8316 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8317 }
8318 
8319 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8320 				       struct net_device *lower_dev)
8321 {
8322 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8323 }
8324 
8325 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8326 				   struct net_device *new_dev,
8327 				   struct net_device *dev,
8328 				   struct netlink_ext_ack *extack)
8329 {
8330 	struct netdev_nested_priv priv = {
8331 		.flags = 0,
8332 		.data = NULL,
8333 	};
8334 	int err;
8335 
8336 	if (!new_dev)
8337 		return 0;
8338 
8339 	if (old_dev && new_dev != old_dev)
8340 		netdev_adjacent_dev_disable(dev, old_dev);
8341 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8342 				      extack);
8343 	if (err) {
8344 		if (old_dev && new_dev != old_dev)
8345 			netdev_adjacent_dev_enable(dev, old_dev);
8346 		return err;
8347 	}
8348 
8349 	return 0;
8350 }
8351 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8352 
8353 void netdev_adjacent_change_commit(struct net_device *old_dev,
8354 				   struct net_device *new_dev,
8355 				   struct net_device *dev)
8356 {
8357 	struct netdev_nested_priv priv = {
8358 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8359 		.data = NULL,
8360 	};
8361 
8362 	if (!new_dev || !old_dev)
8363 		return;
8364 
8365 	if (new_dev == old_dev)
8366 		return;
8367 
8368 	netdev_adjacent_dev_enable(dev, old_dev);
8369 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
8370 }
8371 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8372 
8373 void netdev_adjacent_change_abort(struct net_device *old_dev,
8374 				  struct net_device *new_dev,
8375 				  struct net_device *dev)
8376 {
8377 	struct netdev_nested_priv priv = {
8378 		.flags = 0,
8379 		.data = NULL,
8380 	};
8381 
8382 	if (!new_dev)
8383 		return;
8384 
8385 	if (old_dev && new_dev != old_dev)
8386 		netdev_adjacent_dev_enable(dev, old_dev);
8387 
8388 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
8389 }
8390 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8391 
8392 /**
8393  * netdev_bonding_info_change - Dispatch event about slave change
8394  * @dev: device
8395  * @bonding_info: info to dispatch
8396  *
8397  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8398  * The caller must hold the RTNL lock.
8399  */
8400 void netdev_bonding_info_change(struct net_device *dev,
8401 				struct netdev_bonding_info *bonding_info)
8402 {
8403 	struct netdev_notifier_bonding_info info = {
8404 		.info.dev = dev,
8405 	};
8406 
8407 	memcpy(&info.bonding_info, bonding_info,
8408 	       sizeof(struct netdev_bonding_info));
8409 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8410 				      &info.info);
8411 }
8412 EXPORT_SYMBOL(netdev_bonding_info_change);
8413 
8414 /**
8415  * netdev_get_xmit_slave - Get the xmit slave of master device
8416  * @dev: device
8417  * @skb: The packet
8418  * @all_slaves: assume all the slaves are active
8419  *
8420  * The reference counters are not incremented so the caller must be
8421  * careful with locks. The caller must hold RCU lock.
8422  * %NULL is returned if no slave is found.
8423  */
8424 
8425 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8426 					 struct sk_buff *skb,
8427 					 bool all_slaves)
8428 {
8429 	const struct net_device_ops *ops = dev->netdev_ops;
8430 
8431 	if (!ops->ndo_get_xmit_slave)
8432 		return NULL;
8433 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8434 }
8435 EXPORT_SYMBOL(netdev_get_xmit_slave);
8436 
8437 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8438 						  struct sock *sk)
8439 {
8440 	const struct net_device_ops *ops = dev->netdev_ops;
8441 
8442 	if (!ops->ndo_sk_get_lower_dev)
8443 		return NULL;
8444 	return ops->ndo_sk_get_lower_dev(dev, sk);
8445 }
8446 
8447 /**
8448  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8449  * @dev: device
8450  * @sk: the socket
8451  *
8452  * %NULL is returned if no lower device is found.
8453  */
8454 
8455 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8456 					    struct sock *sk)
8457 {
8458 	struct net_device *lower;
8459 
8460 	lower = netdev_sk_get_lower_dev(dev, sk);
8461 	while (lower) {
8462 		dev = lower;
8463 		lower = netdev_sk_get_lower_dev(dev, sk);
8464 	}
8465 
8466 	return dev;
8467 }
8468 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8469 
8470 static void netdev_adjacent_add_links(struct net_device *dev)
8471 {
8472 	struct netdev_adjacent *iter;
8473 
8474 	struct net *net = dev_net(dev);
8475 
8476 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8477 		if (!net_eq(net, dev_net(iter->dev)))
8478 			continue;
8479 		netdev_adjacent_sysfs_add(iter->dev, dev,
8480 					  &iter->dev->adj_list.lower);
8481 		netdev_adjacent_sysfs_add(dev, iter->dev,
8482 					  &dev->adj_list.upper);
8483 	}
8484 
8485 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8486 		if (!net_eq(net, dev_net(iter->dev)))
8487 			continue;
8488 		netdev_adjacent_sysfs_add(iter->dev, dev,
8489 					  &iter->dev->adj_list.upper);
8490 		netdev_adjacent_sysfs_add(dev, iter->dev,
8491 					  &dev->adj_list.lower);
8492 	}
8493 }
8494 
8495 static void netdev_adjacent_del_links(struct net_device *dev)
8496 {
8497 	struct netdev_adjacent *iter;
8498 
8499 	struct net *net = dev_net(dev);
8500 
8501 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8502 		if (!net_eq(net, dev_net(iter->dev)))
8503 			continue;
8504 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8505 					  &iter->dev->adj_list.lower);
8506 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8507 					  &dev->adj_list.upper);
8508 	}
8509 
8510 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8511 		if (!net_eq(net, dev_net(iter->dev)))
8512 			continue;
8513 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8514 					  &iter->dev->adj_list.upper);
8515 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8516 					  &dev->adj_list.lower);
8517 	}
8518 }
8519 
8520 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8521 {
8522 	struct netdev_adjacent *iter;
8523 
8524 	struct net *net = dev_net(dev);
8525 
8526 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8527 		if (!net_eq(net, dev_net(iter->dev)))
8528 			continue;
8529 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8530 					  &iter->dev->adj_list.lower);
8531 		netdev_adjacent_sysfs_add(iter->dev, dev,
8532 					  &iter->dev->adj_list.lower);
8533 	}
8534 
8535 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8536 		if (!net_eq(net, dev_net(iter->dev)))
8537 			continue;
8538 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8539 					  &iter->dev->adj_list.upper);
8540 		netdev_adjacent_sysfs_add(iter->dev, dev,
8541 					  &iter->dev->adj_list.upper);
8542 	}
8543 }
8544 
8545 void *netdev_lower_dev_get_private(struct net_device *dev,
8546 				   struct net_device *lower_dev)
8547 {
8548 	struct netdev_adjacent *lower;
8549 
8550 	if (!lower_dev)
8551 		return NULL;
8552 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8553 	if (!lower)
8554 		return NULL;
8555 
8556 	return lower->private;
8557 }
8558 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8559 
8560 
8561 /**
8562  * netdev_lower_state_changed - Dispatch event about lower device state change
8563  * @lower_dev: device
8564  * @lower_state_info: state to dispatch
8565  *
8566  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8567  * The caller must hold the RTNL lock.
8568  */
8569 void netdev_lower_state_changed(struct net_device *lower_dev,
8570 				void *lower_state_info)
8571 {
8572 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8573 		.info.dev = lower_dev,
8574 	};
8575 
8576 	ASSERT_RTNL();
8577 	changelowerstate_info.lower_state_info = lower_state_info;
8578 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8579 				      &changelowerstate_info.info);
8580 }
8581 EXPORT_SYMBOL(netdev_lower_state_changed);
8582 
8583 static void dev_change_rx_flags(struct net_device *dev, int flags)
8584 {
8585 	const struct net_device_ops *ops = dev->netdev_ops;
8586 
8587 	if (ops->ndo_change_rx_flags)
8588 		ops->ndo_change_rx_flags(dev, flags);
8589 }
8590 
8591 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8592 {
8593 	unsigned int old_flags = dev->flags;
8594 	kuid_t uid;
8595 	kgid_t gid;
8596 
8597 	ASSERT_RTNL();
8598 
8599 	dev->flags |= IFF_PROMISC;
8600 	dev->promiscuity += inc;
8601 	if (dev->promiscuity == 0) {
8602 		/*
8603 		 * Avoid overflow.
8604 		 * If inc causes overflow, untouch promisc and return error.
8605 		 */
8606 		if (inc < 0)
8607 			dev->flags &= ~IFF_PROMISC;
8608 		else {
8609 			dev->promiscuity -= inc;
8610 			pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8611 				dev->name);
8612 			return -EOVERFLOW;
8613 		}
8614 	}
8615 	if (dev->flags != old_flags) {
8616 		pr_info("device %s %s promiscuous mode\n",
8617 			dev->name,
8618 			dev->flags & IFF_PROMISC ? "entered" : "left");
8619 		if (audit_enabled) {
8620 			current_uid_gid(&uid, &gid);
8621 			audit_log(audit_context(), GFP_ATOMIC,
8622 				  AUDIT_ANOM_PROMISCUOUS,
8623 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8624 				  dev->name, (dev->flags & IFF_PROMISC),
8625 				  (old_flags & IFF_PROMISC),
8626 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8627 				  from_kuid(&init_user_ns, uid),
8628 				  from_kgid(&init_user_ns, gid),
8629 				  audit_get_sessionid(current));
8630 		}
8631 
8632 		dev_change_rx_flags(dev, IFF_PROMISC);
8633 	}
8634 	if (notify)
8635 		__dev_notify_flags(dev, old_flags, IFF_PROMISC);
8636 	return 0;
8637 }
8638 
8639 /**
8640  *	dev_set_promiscuity	- update promiscuity count on a device
8641  *	@dev: device
8642  *	@inc: modifier
8643  *
8644  *	Add or remove promiscuity from a device. While the count in the device
8645  *	remains above zero the interface remains promiscuous. Once it hits zero
8646  *	the device reverts back to normal filtering operation. A negative inc
8647  *	value is used to drop promiscuity on the device.
8648  *	Return 0 if successful or a negative errno code on error.
8649  */
8650 int dev_set_promiscuity(struct net_device *dev, int inc)
8651 {
8652 	unsigned int old_flags = dev->flags;
8653 	int err;
8654 
8655 	err = __dev_set_promiscuity(dev, inc, true);
8656 	if (err < 0)
8657 		return err;
8658 	if (dev->flags != old_flags)
8659 		dev_set_rx_mode(dev);
8660 	return err;
8661 }
8662 EXPORT_SYMBOL(dev_set_promiscuity);
8663 
8664 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8665 {
8666 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8667 
8668 	ASSERT_RTNL();
8669 
8670 	dev->flags |= IFF_ALLMULTI;
8671 	dev->allmulti += inc;
8672 	if (dev->allmulti == 0) {
8673 		/*
8674 		 * Avoid overflow.
8675 		 * If inc causes overflow, untouch allmulti and return error.
8676 		 */
8677 		if (inc < 0)
8678 			dev->flags &= ~IFF_ALLMULTI;
8679 		else {
8680 			dev->allmulti -= inc;
8681 			pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8682 				dev->name);
8683 			return -EOVERFLOW;
8684 		}
8685 	}
8686 	if (dev->flags ^ old_flags) {
8687 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8688 		dev_set_rx_mode(dev);
8689 		if (notify)
8690 			__dev_notify_flags(dev, old_flags,
8691 					   dev->gflags ^ old_gflags);
8692 	}
8693 	return 0;
8694 }
8695 
8696 /**
8697  *	dev_set_allmulti	- update allmulti count on a device
8698  *	@dev: device
8699  *	@inc: modifier
8700  *
8701  *	Add or remove reception of all multicast frames to a device. While the
8702  *	count in the device remains above zero the interface remains listening
8703  *	to all interfaces. Once it hits zero the device reverts back to normal
8704  *	filtering operation. A negative @inc value is used to drop the counter
8705  *	when releasing a resource needing all multicasts.
8706  *	Return 0 if successful or a negative errno code on error.
8707  */
8708 
8709 int dev_set_allmulti(struct net_device *dev, int inc)
8710 {
8711 	return __dev_set_allmulti(dev, inc, true);
8712 }
8713 EXPORT_SYMBOL(dev_set_allmulti);
8714 
8715 /*
8716  *	Upload unicast and multicast address lists to device and
8717  *	configure RX filtering. When the device doesn't support unicast
8718  *	filtering it is put in promiscuous mode while unicast addresses
8719  *	are present.
8720  */
8721 void __dev_set_rx_mode(struct net_device *dev)
8722 {
8723 	const struct net_device_ops *ops = dev->netdev_ops;
8724 
8725 	/* dev_open will call this function so the list will stay sane. */
8726 	if (!(dev->flags&IFF_UP))
8727 		return;
8728 
8729 	if (!netif_device_present(dev))
8730 		return;
8731 
8732 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8733 		/* Unicast addresses changes may only happen under the rtnl,
8734 		 * therefore calling __dev_set_promiscuity here is safe.
8735 		 */
8736 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8737 			__dev_set_promiscuity(dev, 1, false);
8738 			dev->uc_promisc = true;
8739 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8740 			__dev_set_promiscuity(dev, -1, false);
8741 			dev->uc_promisc = false;
8742 		}
8743 	}
8744 
8745 	if (ops->ndo_set_rx_mode)
8746 		ops->ndo_set_rx_mode(dev);
8747 }
8748 
8749 void dev_set_rx_mode(struct net_device *dev)
8750 {
8751 	netif_addr_lock_bh(dev);
8752 	__dev_set_rx_mode(dev);
8753 	netif_addr_unlock_bh(dev);
8754 }
8755 
8756 /**
8757  *	dev_get_flags - get flags reported to userspace
8758  *	@dev: device
8759  *
8760  *	Get the combination of flag bits exported through APIs to userspace.
8761  */
8762 unsigned int dev_get_flags(const struct net_device *dev)
8763 {
8764 	unsigned int flags;
8765 
8766 	flags = (dev->flags & ~(IFF_PROMISC |
8767 				IFF_ALLMULTI |
8768 				IFF_RUNNING |
8769 				IFF_LOWER_UP |
8770 				IFF_DORMANT)) |
8771 		(dev->gflags & (IFF_PROMISC |
8772 				IFF_ALLMULTI));
8773 
8774 	if (netif_running(dev)) {
8775 		if (netif_oper_up(dev))
8776 			flags |= IFF_RUNNING;
8777 		if (netif_carrier_ok(dev))
8778 			flags |= IFF_LOWER_UP;
8779 		if (netif_dormant(dev))
8780 			flags |= IFF_DORMANT;
8781 	}
8782 
8783 	return flags;
8784 }
8785 EXPORT_SYMBOL(dev_get_flags);
8786 
8787 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8788 		       struct netlink_ext_ack *extack)
8789 {
8790 	unsigned int old_flags = dev->flags;
8791 	int ret;
8792 
8793 	ASSERT_RTNL();
8794 
8795 	/*
8796 	 *	Set the flags on our device.
8797 	 */
8798 
8799 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8800 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8801 			       IFF_AUTOMEDIA)) |
8802 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8803 				    IFF_ALLMULTI));
8804 
8805 	/*
8806 	 *	Load in the correct multicast list now the flags have changed.
8807 	 */
8808 
8809 	if ((old_flags ^ flags) & IFF_MULTICAST)
8810 		dev_change_rx_flags(dev, IFF_MULTICAST);
8811 
8812 	dev_set_rx_mode(dev);
8813 
8814 	/*
8815 	 *	Have we downed the interface. We handle IFF_UP ourselves
8816 	 *	according to user attempts to set it, rather than blindly
8817 	 *	setting it.
8818 	 */
8819 
8820 	ret = 0;
8821 	if ((old_flags ^ flags) & IFF_UP) {
8822 		if (old_flags & IFF_UP)
8823 			__dev_close(dev);
8824 		else
8825 			ret = __dev_open(dev, extack);
8826 	}
8827 
8828 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8829 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8830 		unsigned int old_flags = dev->flags;
8831 
8832 		dev->gflags ^= IFF_PROMISC;
8833 
8834 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8835 			if (dev->flags != old_flags)
8836 				dev_set_rx_mode(dev);
8837 	}
8838 
8839 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8840 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8841 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8842 	 */
8843 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8844 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8845 
8846 		dev->gflags ^= IFF_ALLMULTI;
8847 		__dev_set_allmulti(dev, inc, false);
8848 	}
8849 
8850 	return ret;
8851 }
8852 
8853 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8854 			unsigned int gchanges)
8855 {
8856 	unsigned int changes = dev->flags ^ old_flags;
8857 
8858 	if (gchanges)
8859 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8860 
8861 	if (changes & IFF_UP) {
8862 		if (dev->flags & IFF_UP)
8863 			call_netdevice_notifiers(NETDEV_UP, dev);
8864 		else
8865 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8866 	}
8867 
8868 	if (dev->flags & IFF_UP &&
8869 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8870 		struct netdev_notifier_change_info change_info = {
8871 			.info = {
8872 				.dev = dev,
8873 			},
8874 			.flags_changed = changes,
8875 		};
8876 
8877 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8878 	}
8879 }
8880 
8881 /**
8882  *	dev_change_flags - change device settings
8883  *	@dev: device
8884  *	@flags: device state flags
8885  *	@extack: netlink extended ack
8886  *
8887  *	Change settings on device based state flags. The flags are
8888  *	in the userspace exported format.
8889  */
8890 int dev_change_flags(struct net_device *dev, unsigned int flags,
8891 		     struct netlink_ext_ack *extack)
8892 {
8893 	int ret;
8894 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8895 
8896 	ret = __dev_change_flags(dev, flags, extack);
8897 	if (ret < 0)
8898 		return ret;
8899 
8900 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8901 	__dev_notify_flags(dev, old_flags, changes);
8902 	return ret;
8903 }
8904 EXPORT_SYMBOL(dev_change_flags);
8905 
8906 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8907 {
8908 	const struct net_device_ops *ops = dev->netdev_ops;
8909 
8910 	if (ops->ndo_change_mtu)
8911 		return ops->ndo_change_mtu(dev, new_mtu);
8912 
8913 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8914 	WRITE_ONCE(dev->mtu, new_mtu);
8915 	return 0;
8916 }
8917 EXPORT_SYMBOL(__dev_set_mtu);
8918 
8919 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8920 		     struct netlink_ext_ack *extack)
8921 {
8922 	/* MTU must be positive, and in range */
8923 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8924 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8925 		return -EINVAL;
8926 	}
8927 
8928 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8929 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8930 		return -EINVAL;
8931 	}
8932 	return 0;
8933 }
8934 
8935 /**
8936  *	dev_set_mtu_ext - Change maximum transfer unit
8937  *	@dev: device
8938  *	@new_mtu: new transfer unit
8939  *	@extack: netlink extended ack
8940  *
8941  *	Change the maximum transfer size of the network device.
8942  */
8943 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8944 		    struct netlink_ext_ack *extack)
8945 {
8946 	int err, orig_mtu;
8947 
8948 	if (new_mtu == dev->mtu)
8949 		return 0;
8950 
8951 	err = dev_validate_mtu(dev, new_mtu, extack);
8952 	if (err)
8953 		return err;
8954 
8955 	if (!netif_device_present(dev))
8956 		return -ENODEV;
8957 
8958 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8959 	err = notifier_to_errno(err);
8960 	if (err)
8961 		return err;
8962 
8963 	orig_mtu = dev->mtu;
8964 	err = __dev_set_mtu(dev, new_mtu);
8965 
8966 	if (!err) {
8967 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8968 						   orig_mtu);
8969 		err = notifier_to_errno(err);
8970 		if (err) {
8971 			/* setting mtu back and notifying everyone again,
8972 			 * so that they have a chance to revert changes.
8973 			 */
8974 			__dev_set_mtu(dev, orig_mtu);
8975 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8976 						     new_mtu);
8977 		}
8978 	}
8979 	return err;
8980 }
8981 
8982 int dev_set_mtu(struct net_device *dev, int new_mtu)
8983 {
8984 	struct netlink_ext_ack extack;
8985 	int err;
8986 
8987 	memset(&extack, 0, sizeof(extack));
8988 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8989 	if (err && extack._msg)
8990 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8991 	return err;
8992 }
8993 EXPORT_SYMBOL(dev_set_mtu);
8994 
8995 /**
8996  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8997  *	@dev: device
8998  *	@new_len: new tx queue length
8999  */
9000 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9001 {
9002 	unsigned int orig_len = dev->tx_queue_len;
9003 	int res;
9004 
9005 	if (new_len != (unsigned int)new_len)
9006 		return -ERANGE;
9007 
9008 	if (new_len != orig_len) {
9009 		dev->tx_queue_len = new_len;
9010 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9011 		res = notifier_to_errno(res);
9012 		if (res)
9013 			goto err_rollback;
9014 		res = dev_qdisc_change_tx_queue_len(dev);
9015 		if (res)
9016 			goto err_rollback;
9017 	}
9018 
9019 	return 0;
9020 
9021 err_rollback:
9022 	netdev_err(dev, "refused to change device tx_queue_len\n");
9023 	dev->tx_queue_len = orig_len;
9024 	return res;
9025 }
9026 
9027 /**
9028  *	dev_set_group - Change group this device belongs to
9029  *	@dev: device
9030  *	@new_group: group this device should belong to
9031  */
9032 void dev_set_group(struct net_device *dev, int new_group)
9033 {
9034 	dev->group = new_group;
9035 }
9036 EXPORT_SYMBOL(dev_set_group);
9037 
9038 /**
9039  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9040  *	@dev: device
9041  *	@addr: new address
9042  *	@extack: netlink extended ack
9043  */
9044 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9045 			      struct netlink_ext_ack *extack)
9046 {
9047 	struct netdev_notifier_pre_changeaddr_info info = {
9048 		.info.dev = dev,
9049 		.info.extack = extack,
9050 		.dev_addr = addr,
9051 	};
9052 	int rc;
9053 
9054 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9055 	return notifier_to_errno(rc);
9056 }
9057 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9058 
9059 /**
9060  *	dev_set_mac_address - Change Media Access Control Address
9061  *	@dev: device
9062  *	@sa: new address
9063  *	@extack: netlink extended ack
9064  *
9065  *	Change the hardware (MAC) address of the device
9066  */
9067 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9068 			struct netlink_ext_ack *extack)
9069 {
9070 	const struct net_device_ops *ops = dev->netdev_ops;
9071 	int err;
9072 
9073 	if (!ops->ndo_set_mac_address)
9074 		return -EOPNOTSUPP;
9075 	if (sa->sa_family != dev->type)
9076 		return -EINVAL;
9077 	if (!netif_device_present(dev))
9078 		return -ENODEV;
9079 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9080 	if (err)
9081 		return err;
9082 	err = ops->ndo_set_mac_address(dev, sa);
9083 	if (err)
9084 		return err;
9085 	dev->addr_assign_type = NET_ADDR_SET;
9086 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9087 	add_device_randomness(dev->dev_addr, dev->addr_len);
9088 	return 0;
9089 }
9090 EXPORT_SYMBOL(dev_set_mac_address);
9091 
9092 static DECLARE_RWSEM(dev_addr_sem);
9093 
9094 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9095 			     struct netlink_ext_ack *extack)
9096 {
9097 	int ret;
9098 
9099 	down_write(&dev_addr_sem);
9100 	ret = dev_set_mac_address(dev, sa, extack);
9101 	up_write(&dev_addr_sem);
9102 	return ret;
9103 }
9104 EXPORT_SYMBOL(dev_set_mac_address_user);
9105 
9106 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9107 {
9108 	size_t size = sizeof(sa->sa_data);
9109 	struct net_device *dev;
9110 	int ret = 0;
9111 
9112 	down_read(&dev_addr_sem);
9113 	rcu_read_lock();
9114 
9115 	dev = dev_get_by_name_rcu(net, dev_name);
9116 	if (!dev) {
9117 		ret = -ENODEV;
9118 		goto unlock;
9119 	}
9120 	if (!dev->addr_len)
9121 		memset(sa->sa_data, 0, size);
9122 	else
9123 		memcpy(sa->sa_data, dev->dev_addr,
9124 		       min_t(size_t, size, dev->addr_len));
9125 	sa->sa_family = dev->type;
9126 
9127 unlock:
9128 	rcu_read_unlock();
9129 	up_read(&dev_addr_sem);
9130 	return ret;
9131 }
9132 EXPORT_SYMBOL(dev_get_mac_address);
9133 
9134 /**
9135  *	dev_change_carrier - Change device carrier
9136  *	@dev: device
9137  *	@new_carrier: new value
9138  *
9139  *	Change device carrier
9140  */
9141 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9142 {
9143 	const struct net_device_ops *ops = dev->netdev_ops;
9144 
9145 	if (!ops->ndo_change_carrier)
9146 		return -EOPNOTSUPP;
9147 	if (!netif_device_present(dev))
9148 		return -ENODEV;
9149 	return ops->ndo_change_carrier(dev, new_carrier);
9150 }
9151 EXPORT_SYMBOL(dev_change_carrier);
9152 
9153 /**
9154  *	dev_get_phys_port_id - Get device physical port ID
9155  *	@dev: device
9156  *	@ppid: port ID
9157  *
9158  *	Get device physical port ID
9159  */
9160 int dev_get_phys_port_id(struct net_device *dev,
9161 			 struct netdev_phys_item_id *ppid)
9162 {
9163 	const struct net_device_ops *ops = dev->netdev_ops;
9164 
9165 	if (!ops->ndo_get_phys_port_id)
9166 		return -EOPNOTSUPP;
9167 	return ops->ndo_get_phys_port_id(dev, ppid);
9168 }
9169 EXPORT_SYMBOL(dev_get_phys_port_id);
9170 
9171 /**
9172  *	dev_get_phys_port_name - Get device physical port name
9173  *	@dev: device
9174  *	@name: port name
9175  *	@len: limit of bytes to copy to name
9176  *
9177  *	Get device physical port name
9178  */
9179 int dev_get_phys_port_name(struct net_device *dev,
9180 			   char *name, size_t len)
9181 {
9182 	const struct net_device_ops *ops = dev->netdev_ops;
9183 	int err;
9184 
9185 	if (ops->ndo_get_phys_port_name) {
9186 		err = ops->ndo_get_phys_port_name(dev, name, len);
9187 		if (err != -EOPNOTSUPP)
9188 			return err;
9189 	}
9190 	return devlink_compat_phys_port_name_get(dev, name, len);
9191 }
9192 EXPORT_SYMBOL(dev_get_phys_port_name);
9193 
9194 /**
9195  *	dev_get_port_parent_id - Get the device's port parent identifier
9196  *	@dev: network device
9197  *	@ppid: pointer to a storage for the port's parent identifier
9198  *	@recurse: allow/disallow recursion to lower devices
9199  *
9200  *	Get the devices's port parent identifier
9201  */
9202 int dev_get_port_parent_id(struct net_device *dev,
9203 			   struct netdev_phys_item_id *ppid,
9204 			   bool recurse)
9205 {
9206 	const struct net_device_ops *ops = dev->netdev_ops;
9207 	struct netdev_phys_item_id first = { };
9208 	struct net_device *lower_dev;
9209 	struct list_head *iter;
9210 	int err;
9211 
9212 	if (ops->ndo_get_port_parent_id) {
9213 		err = ops->ndo_get_port_parent_id(dev, ppid);
9214 		if (err != -EOPNOTSUPP)
9215 			return err;
9216 	}
9217 
9218 	err = devlink_compat_switch_id_get(dev, ppid);
9219 	if (!err || err != -EOPNOTSUPP)
9220 		return err;
9221 
9222 	if (!recurse)
9223 		return -EOPNOTSUPP;
9224 
9225 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
9226 		err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9227 		if (err)
9228 			break;
9229 		if (!first.id_len)
9230 			first = *ppid;
9231 		else if (memcmp(&first, ppid, sizeof(*ppid)))
9232 			return -EOPNOTSUPP;
9233 	}
9234 
9235 	return err;
9236 }
9237 EXPORT_SYMBOL(dev_get_port_parent_id);
9238 
9239 /**
9240  *	netdev_port_same_parent_id - Indicate if two network devices have
9241  *	the same port parent identifier
9242  *	@a: first network device
9243  *	@b: second network device
9244  */
9245 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9246 {
9247 	struct netdev_phys_item_id a_id = { };
9248 	struct netdev_phys_item_id b_id = { };
9249 
9250 	if (dev_get_port_parent_id(a, &a_id, true) ||
9251 	    dev_get_port_parent_id(b, &b_id, true))
9252 		return false;
9253 
9254 	return netdev_phys_item_id_same(&a_id, &b_id);
9255 }
9256 EXPORT_SYMBOL(netdev_port_same_parent_id);
9257 
9258 /**
9259  *	dev_change_proto_down - update protocol port state information
9260  *	@dev: device
9261  *	@proto_down: new value
9262  *
9263  *	This info can be used by switch drivers to set the phys state of the
9264  *	port.
9265  */
9266 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9267 {
9268 	const struct net_device_ops *ops = dev->netdev_ops;
9269 
9270 	if (!ops->ndo_change_proto_down)
9271 		return -EOPNOTSUPP;
9272 	if (!netif_device_present(dev))
9273 		return -ENODEV;
9274 	return ops->ndo_change_proto_down(dev, proto_down);
9275 }
9276 EXPORT_SYMBOL(dev_change_proto_down);
9277 
9278 /**
9279  *	dev_change_proto_down_generic - generic implementation for
9280  * 	ndo_change_proto_down that sets carrier according to
9281  * 	proto_down.
9282  *
9283  *	@dev: device
9284  *	@proto_down: new value
9285  */
9286 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9287 {
9288 	if (proto_down)
9289 		netif_carrier_off(dev);
9290 	else
9291 		netif_carrier_on(dev);
9292 	dev->proto_down = proto_down;
9293 	return 0;
9294 }
9295 EXPORT_SYMBOL(dev_change_proto_down_generic);
9296 
9297 /**
9298  *	dev_change_proto_down_reason - proto down reason
9299  *
9300  *	@dev: device
9301  *	@mask: proto down mask
9302  *	@value: proto down value
9303  */
9304 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9305 				  u32 value)
9306 {
9307 	int b;
9308 
9309 	if (!mask) {
9310 		dev->proto_down_reason = value;
9311 	} else {
9312 		for_each_set_bit(b, &mask, 32) {
9313 			if (value & (1 << b))
9314 				dev->proto_down_reason |= BIT(b);
9315 			else
9316 				dev->proto_down_reason &= ~BIT(b);
9317 		}
9318 	}
9319 }
9320 EXPORT_SYMBOL(dev_change_proto_down_reason);
9321 
9322 struct bpf_xdp_link {
9323 	struct bpf_link link;
9324 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9325 	int flags;
9326 };
9327 
9328 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9329 {
9330 	if (flags & XDP_FLAGS_HW_MODE)
9331 		return XDP_MODE_HW;
9332 	if (flags & XDP_FLAGS_DRV_MODE)
9333 		return XDP_MODE_DRV;
9334 	if (flags & XDP_FLAGS_SKB_MODE)
9335 		return XDP_MODE_SKB;
9336 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9337 }
9338 
9339 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9340 {
9341 	switch (mode) {
9342 	case XDP_MODE_SKB:
9343 		return generic_xdp_install;
9344 	case XDP_MODE_DRV:
9345 	case XDP_MODE_HW:
9346 		return dev->netdev_ops->ndo_bpf;
9347 	default:
9348 		return NULL;
9349 	}
9350 }
9351 
9352 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9353 					 enum bpf_xdp_mode mode)
9354 {
9355 	return dev->xdp_state[mode].link;
9356 }
9357 
9358 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9359 				     enum bpf_xdp_mode mode)
9360 {
9361 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9362 
9363 	if (link)
9364 		return link->link.prog;
9365 	return dev->xdp_state[mode].prog;
9366 }
9367 
9368 static u8 dev_xdp_prog_count(struct net_device *dev)
9369 {
9370 	u8 count = 0;
9371 	int i;
9372 
9373 	for (i = 0; i < __MAX_XDP_MODE; i++)
9374 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9375 			count++;
9376 	return count;
9377 }
9378 
9379 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9380 {
9381 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9382 
9383 	return prog ? prog->aux->id : 0;
9384 }
9385 
9386 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9387 			     struct bpf_xdp_link *link)
9388 {
9389 	dev->xdp_state[mode].link = link;
9390 	dev->xdp_state[mode].prog = NULL;
9391 }
9392 
9393 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9394 			     struct bpf_prog *prog)
9395 {
9396 	dev->xdp_state[mode].link = NULL;
9397 	dev->xdp_state[mode].prog = prog;
9398 }
9399 
9400 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9401 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9402 			   u32 flags, struct bpf_prog *prog)
9403 {
9404 	struct netdev_bpf xdp;
9405 	int err;
9406 
9407 	memset(&xdp, 0, sizeof(xdp));
9408 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9409 	xdp.extack = extack;
9410 	xdp.flags = flags;
9411 	xdp.prog = prog;
9412 
9413 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9414 	 * "moved" into driver), so they don't increment it on their own, but
9415 	 * they do decrement refcnt when program is detached or replaced.
9416 	 * Given net_device also owns link/prog, we need to bump refcnt here
9417 	 * to prevent drivers from underflowing it.
9418 	 */
9419 	if (prog)
9420 		bpf_prog_inc(prog);
9421 	err = bpf_op(dev, &xdp);
9422 	if (err) {
9423 		if (prog)
9424 			bpf_prog_put(prog);
9425 		return err;
9426 	}
9427 
9428 	if (mode != XDP_MODE_HW)
9429 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9430 
9431 	return 0;
9432 }
9433 
9434 static void dev_xdp_uninstall(struct net_device *dev)
9435 {
9436 	struct bpf_xdp_link *link;
9437 	struct bpf_prog *prog;
9438 	enum bpf_xdp_mode mode;
9439 	bpf_op_t bpf_op;
9440 
9441 	ASSERT_RTNL();
9442 
9443 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9444 		prog = dev_xdp_prog(dev, mode);
9445 		if (!prog)
9446 			continue;
9447 
9448 		bpf_op = dev_xdp_bpf_op(dev, mode);
9449 		if (!bpf_op)
9450 			continue;
9451 
9452 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9453 
9454 		/* auto-detach link from net device */
9455 		link = dev_xdp_link(dev, mode);
9456 		if (link)
9457 			link->dev = NULL;
9458 		else
9459 			bpf_prog_put(prog);
9460 
9461 		dev_xdp_set_link(dev, mode, NULL);
9462 	}
9463 }
9464 
9465 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9466 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9467 			  struct bpf_prog *old_prog, u32 flags)
9468 {
9469 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9470 	struct bpf_prog *cur_prog;
9471 	enum bpf_xdp_mode mode;
9472 	bpf_op_t bpf_op;
9473 	int err;
9474 
9475 	ASSERT_RTNL();
9476 
9477 	/* either link or prog attachment, never both */
9478 	if (link && (new_prog || old_prog))
9479 		return -EINVAL;
9480 	/* link supports only XDP mode flags */
9481 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9482 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9483 		return -EINVAL;
9484 	}
9485 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9486 	if (num_modes > 1) {
9487 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9488 		return -EINVAL;
9489 	}
9490 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9491 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9492 		NL_SET_ERR_MSG(extack,
9493 			       "More than one program loaded, unset mode is ambiguous");
9494 		return -EINVAL;
9495 	}
9496 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9497 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9498 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9499 		return -EINVAL;
9500 	}
9501 
9502 	mode = dev_xdp_mode(dev, flags);
9503 	/* can't replace attached link */
9504 	if (dev_xdp_link(dev, mode)) {
9505 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9506 		return -EBUSY;
9507 	}
9508 
9509 	cur_prog = dev_xdp_prog(dev, mode);
9510 	/* can't replace attached prog with link */
9511 	if (link && cur_prog) {
9512 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9513 		return -EBUSY;
9514 	}
9515 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9516 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9517 		return -EEXIST;
9518 	}
9519 
9520 	/* put effective new program into new_prog */
9521 	if (link)
9522 		new_prog = link->link.prog;
9523 
9524 	if (new_prog) {
9525 		bool offload = mode == XDP_MODE_HW;
9526 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9527 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9528 
9529 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9530 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9531 			return -EBUSY;
9532 		}
9533 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9534 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9535 			return -EEXIST;
9536 		}
9537 		if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9538 			NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9539 			return -EINVAL;
9540 		}
9541 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9542 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9543 			return -EINVAL;
9544 		}
9545 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9546 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9547 			return -EINVAL;
9548 		}
9549 	}
9550 
9551 	/* don't call drivers if the effective program didn't change */
9552 	if (new_prog != cur_prog) {
9553 		bpf_op = dev_xdp_bpf_op(dev, mode);
9554 		if (!bpf_op) {
9555 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9556 			return -EOPNOTSUPP;
9557 		}
9558 
9559 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9560 		if (err)
9561 			return err;
9562 	}
9563 
9564 	if (link)
9565 		dev_xdp_set_link(dev, mode, link);
9566 	else
9567 		dev_xdp_set_prog(dev, mode, new_prog);
9568 	if (cur_prog)
9569 		bpf_prog_put(cur_prog);
9570 
9571 	return 0;
9572 }
9573 
9574 static int dev_xdp_attach_link(struct net_device *dev,
9575 			       struct netlink_ext_ack *extack,
9576 			       struct bpf_xdp_link *link)
9577 {
9578 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9579 }
9580 
9581 static int dev_xdp_detach_link(struct net_device *dev,
9582 			       struct netlink_ext_ack *extack,
9583 			       struct bpf_xdp_link *link)
9584 {
9585 	enum bpf_xdp_mode mode;
9586 	bpf_op_t bpf_op;
9587 
9588 	ASSERT_RTNL();
9589 
9590 	mode = dev_xdp_mode(dev, link->flags);
9591 	if (dev_xdp_link(dev, mode) != link)
9592 		return -EINVAL;
9593 
9594 	bpf_op = dev_xdp_bpf_op(dev, mode);
9595 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9596 	dev_xdp_set_link(dev, mode, NULL);
9597 	return 0;
9598 }
9599 
9600 static void bpf_xdp_link_release(struct bpf_link *link)
9601 {
9602 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9603 
9604 	rtnl_lock();
9605 
9606 	/* if racing with net_device's tear down, xdp_link->dev might be
9607 	 * already NULL, in which case link was already auto-detached
9608 	 */
9609 	if (xdp_link->dev) {
9610 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9611 		xdp_link->dev = NULL;
9612 	}
9613 
9614 	rtnl_unlock();
9615 }
9616 
9617 static int bpf_xdp_link_detach(struct bpf_link *link)
9618 {
9619 	bpf_xdp_link_release(link);
9620 	return 0;
9621 }
9622 
9623 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9624 {
9625 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9626 
9627 	kfree(xdp_link);
9628 }
9629 
9630 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9631 				     struct seq_file *seq)
9632 {
9633 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9634 	u32 ifindex = 0;
9635 
9636 	rtnl_lock();
9637 	if (xdp_link->dev)
9638 		ifindex = xdp_link->dev->ifindex;
9639 	rtnl_unlock();
9640 
9641 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9642 }
9643 
9644 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9645 				       struct bpf_link_info *info)
9646 {
9647 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9648 	u32 ifindex = 0;
9649 
9650 	rtnl_lock();
9651 	if (xdp_link->dev)
9652 		ifindex = xdp_link->dev->ifindex;
9653 	rtnl_unlock();
9654 
9655 	info->xdp.ifindex = ifindex;
9656 	return 0;
9657 }
9658 
9659 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9660 			       struct bpf_prog *old_prog)
9661 {
9662 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9663 	enum bpf_xdp_mode mode;
9664 	bpf_op_t bpf_op;
9665 	int err = 0;
9666 
9667 	rtnl_lock();
9668 
9669 	/* link might have been auto-released already, so fail */
9670 	if (!xdp_link->dev) {
9671 		err = -ENOLINK;
9672 		goto out_unlock;
9673 	}
9674 
9675 	if (old_prog && link->prog != old_prog) {
9676 		err = -EPERM;
9677 		goto out_unlock;
9678 	}
9679 	old_prog = link->prog;
9680 	if (old_prog == new_prog) {
9681 		/* no-op, don't disturb drivers */
9682 		bpf_prog_put(new_prog);
9683 		goto out_unlock;
9684 	}
9685 
9686 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9687 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9688 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9689 			      xdp_link->flags, new_prog);
9690 	if (err)
9691 		goto out_unlock;
9692 
9693 	old_prog = xchg(&link->prog, new_prog);
9694 	bpf_prog_put(old_prog);
9695 
9696 out_unlock:
9697 	rtnl_unlock();
9698 	return err;
9699 }
9700 
9701 static const struct bpf_link_ops bpf_xdp_link_lops = {
9702 	.release = bpf_xdp_link_release,
9703 	.dealloc = bpf_xdp_link_dealloc,
9704 	.detach = bpf_xdp_link_detach,
9705 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9706 	.fill_link_info = bpf_xdp_link_fill_link_info,
9707 	.update_prog = bpf_xdp_link_update,
9708 };
9709 
9710 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9711 {
9712 	struct net *net = current->nsproxy->net_ns;
9713 	struct bpf_link_primer link_primer;
9714 	struct bpf_xdp_link *link;
9715 	struct net_device *dev;
9716 	int err, fd;
9717 
9718 	rtnl_lock();
9719 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9720 	if (!dev) {
9721 		rtnl_unlock();
9722 		return -EINVAL;
9723 	}
9724 
9725 	link = kzalloc(sizeof(*link), GFP_USER);
9726 	if (!link) {
9727 		err = -ENOMEM;
9728 		goto unlock;
9729 	}
9730 
9731 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9732 	link->dev = dev;
9733 	link->flags = attr->link_create.flags;
9734 
9735 	err = bpf_link_prime(&link->link, &link_primer);
9736 	if (err) {
9737 		kfree(link);
9738 		goto unlock;
9739 	}
9740 
9741 	err = dev_xdp_attach_link(dev, NULL, link);
9742 	rtnl_unlock();
9743 
9744 	if (err) {
9745 		link->dev = NULL;
9746 		bpf_link_cleanup(&link_primer);
9747 		goto out_put_dev;
9748 	}
9749 
9750 	fd = bpf_link_settle(&link_primer);
9751 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9752 	dev_put(dev);
9753 	return fd;
9754 
9755 unlock:
9756 	rtnl_unlock();
9757 
9758 out_put_dev:
9759 	dev_put(dev);
9760 	return err;
9761 }
9762 
9763 /**
9764  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9765  *	@dev: device
9766  *	@extack: netlink extended ack
9767  *	@fd: new program fd or negative value to clear
9768  *	@expected_fd: old program fd that userspace expects to replace or clear
9769  *	@flags: xdp-related flags
9770  *
9771  *	Set or clear a bpf program for a device
9772  */
9773 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9774 		      int fd, int expected_fd, u32 flags)
9775 {
9776 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9777 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9778 	int err;
9779 
9780 	ASSERT_RTNL();
9781 
9782 	if (fd >= 0) {
9783 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9784 						 mode != XDP_MODE_SKB);
9785 		if (IS_ERR(new_prog))
9786 			return PTR_ERR(new_prog);
9787 	}
9788 
9789 	if (expected_fd >= 0) {
9790 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9791 						 mode != XDP_MODE_SKB);
9792 		if (IS_ERR(old_prog)) {
9793 			err = PTR_ERR(old_prog);
9794 			old_prog = NULL;
9795 			goto err_out;
9796 		}
9797 	}
9798 
9799 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9800 
9801 err_out:
9802 	if (err && new_prog)
9803 		bpf_prog_put(new_prog);
9804 	if (old_prog)
9805 		bpf_prog_put(old_prog);
9806 	return err;
9807 }
9808 
9809 /**
9810  *	dev_new_index	-	allocate an ifindex
9811  *	@net: the applicable net namespace
9812  *
9813  *	Returns a suitable unique value for a new device interface
9814  *	number.  The caller must hold the rtnl semaphore or the
9815  *	dev_base_lock to be sure it remains unique.
9816  */
9817 static int dev_new_index(struct net *net)
9818 {
9819 	int ifindex = net->ifindex;
9820 
9821 	for (;;) {
9822 		if (++ifindex <= 0)
9823 			ifindex = 1;
9824 		if (!__dev_get_by_index(net, ifindex))
9825 			return net->ifindex = ifindex;
9826 	}
9827 }
9828 
9829 /* Delayed registration/unregisteration */
9830 static LIST_HEAD(net_todo_list);
9831 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9832 
9833 static void net_set_todo(struct net_device *dev)
9834 {
9835 	list_add_tail(&dev->todo_list, &net_todo_list);
9836 	dev_net(dev)->dev_unreg_count++;
9837 }
9838 
9839 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9840 	struct net_device *upper, netdev_features_t features)
9841 {
9842 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9843 	netdev_features_t feature;
9844 	int feature_bit;
9845 
9846 	for_each_netdev_feature(upper_disables, feature_bit) {
9847 		feature = __NETIF_F_BIT(feature_bit);
9848 		if (!(upper->wanted_features & feature)
9849 		    && (features & feature)) {
9850 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9851 				   &feature, upper->name);
9852 			features &= ~feature;
9853 		}
9854 	}
9855 
9856 	return features;
9857 }
9858 
9859 static void netdev_sync_lower_features(struct net_device *upper,
9860 	struct net_device *lower, netdev_features_t features)
9861 {
9862 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9863 	netdev_features_t feature;
9864 	int feature_bit;
9865 
9866 	for_each_netdev_feature(upper_disables, feature_bit) {
9867 		feature = __NETIF_F_BIT(feature_bit);
9868 		if (!(features & feature) && (lower->features & feature)) {
9869 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9870 				   &feature, lower->name);
9871 			lower->wanted_features &= ~feature;
9872 			__netdev_update_features(lower);
9873 
9874 			if (unlikely(lower->features & feature))
9875 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9876 					    &feature, lower->name);
9877 			else
9878 				netdev_features_change(lower);
9879 		}
9880 	}
9881 }
9882 
9883 static netdev_features_t netdev_fix_features(struct net_device *dev,
9884 	netdev_features_t features)
9885 {
9886 	/* Fix illegal checksum combinations */
9887 	if ((features & NETIF_F_HW_CSUM) &&
9888 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9889 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9890 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9891 	}
9892 
9893 	/* TSO requires that SG is present as well. */
9894 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9895 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9896 		features &= ~NETIF_F_ALL_TSO;
9897 	}
9898 
9899 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9900 					!(features & NETIF_F_IP_CSUM)) {
9901 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9902 		features &= ~NETIF_F_TSO;
9903 		features &= ~NETIF_F_TSO_ECN;
9904 	}
9905 
9906 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9907 					 !(features & NETIF_F_IPV6_CSUM)) {
9908 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9909 		features &= ~NETIF_F_TSO6;
9910 	}
9911 
9912 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9913 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9914 		features &= ~NETIF_F_TSO_MANGLEID;
9915 
9916 	/* TSO ECN requires that TSO is present as well. */
9917 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9918 		features &= ~NETIF_F_TSO_ECN;
9919 
9920 	/* Software GSO depends on SG. */
9921 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9922 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9923 		features &= ~NETIF_F_GSO;
9924 	}
9925 
9926 	/* GSO partial features require GSO partial be set */
9927 	if ((features & dev->gso_partial_features) &&
9928 	    !(features & NETIF_F_GSO_PARTIAL)) {
9929 		netdev_dbg(dev,
9930 			   "Dropping partially supported GSO features since no GSO partial.\n");
9931 		features &= ~dev->gso_partial_features;
9932 	}
9933 
9934 	if (!(features & NETIF_F_RXCSUM)) {
9935 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9936 		 * successfully merged by hardware must also have the
9937 		 * checksum verified by hardware.  If the user does not
9938 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9939 		 */
9940 		if (features & NETIF_F_GRO_HW) {
9941 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9942 			features &= ~NETIF_F_GRO_HW;
9943 		}
9944 	}
9945 
9946 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9947 	if (features & NETIF_F_RXFCS) {
9948 		if (features & NETIF_F_LRO) {
9949 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9950 			features &= ~NETIF_F_LRO;
9951 		}
9952 
9953 		if (features & NETIF_F_GRO_HW) {
9954 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9955 			features &= ~NETIF_F_GRO_HW;
9956 		}
9957 	}
9958 
9959 	if (features & NETIF_F_HW_TLS_TX) {
9960 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9961 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9962 		bool hw_csum = features & NETIF_F_HW_CSUM;
9963 
9964 		if (!ip_csum && !hw_csum) {
9965 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9966 			features &= ~NETIF_F_HW_TLS_TX;
9967 		}
9968 	}
9969 
9970 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9971 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9972 		features &= ~NETIF_F_HW_TLS_RX;
9973 	}
9974 
9975 	return features;
9976 }
9977 
9978 int __netdev_update_features(struct net_device *dev)
9979 {
9980 	struct net_device *upper, *lower;
9981 	netdev_features_t features;
9982 	struct list_head *iter;
9983 	int err = -1;
9984 
9985 	ASSERT_RTNL();
9986 
9987 	features = netdev_get_wanted_features(dev);
9988 
9989 	if (dev->netdev_ops->ndo_fix_features)
9990 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9991 
9992 	/* driver might be less strict about feature dependencies */
9993 	features = netdev_fix_features(dev, features);
9994 
9995 	/* some features can't be enabled if they're off on an upper device */
9996 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9997 		features = netdev_sync_upper_features(dev, upper, features);
9998 
9999 	if (dev->features == features)
10000 		goto sync_lower;
10001 
10002 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10003 		&dev->features, &features);
10004 
10005 	if (dev->netdev_ops->ndo_set_features)
10006 		err = dev->netdev_ops->ndo_set_features(dev, features);
10007 	else
10008 		err = 0;
10009 
10010 	if (unlikely(err < 0)) {
10011 		netdev_err(dev,
10012 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
10013 			err, &features, &dev->features);
10014 		/* return non-0 since some features might have changed and
10015 		 * it's better to fire a spurious notification than miss it
10016 		 */
10017 		return -1;
10018 	}
10019 
10020 sync_lower:
10021 	/* some features must be disabled on lower devices when disabled
10022 	 * on an upper device (think: bonding master or bridge)
10023 	 */
10024 	netdev_for_each_lower_dev(dev, lower, iter)
10025 		netdev_sync_lower_features(dev, lower, features);
10026 
10027 	if (!err) {
10028 		netdev_features_t diff = features ^ dev->features;
10029 
10030 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10031 			/* udp_tunnel_{get,drop}_rx_info both need
10032 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10033 			 * device, or they won't do anything.
10034 			 * Thus we need to update dev->features
10035 			 * *before* calling udp_tunnel_get_rx_info,
10036 			 * but *after* calling udp_tunnel_drop_rx_info.
10037 			 */
10038 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10039 				dev->features = features;
10040 				udp_tunnel_get_rx_info(dev);
10041 			} else {
10042 				udp_tunnel_drop_rx_info(dev);
10043 			}
10044 		}
10045 
10046 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10047 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10048 				dev->features = features;
10049 				err |= vlan_get_rx_ctag_filter_info(dev);
10050 			} else {
10051 				vlan_drop_rx_ctag_filter_info(dev);
10052 			}
10053 		}
10054 
10055 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10056 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10057 				dev->features = features;
10058 				err |= vlan_get_rx_stag_filter_info(dev);
10059 			} else {
10060 				vlan_drop_rx_stag_filter_info(dev);
10061 			}
10062 		}
10063 
10064 		dev->features = features;
10065 	}
10066 
10067 	return err < 0 ? 0 : 1;
10068 }
10069 
10070 /**
10071  *	netdev_update_features - recalculate device features
10072  *	@dev: the device to check
10073  *
10074  *	Recalculate dev->features set and send notifications if it
10075  *	has changed. Should be called after driver or hardware dependent
10076  *	conditions might have changed that influence the features.
10077  */
10078 void netdev_update_features(struct net_device *dev)
10079 {
10080 	if (__netdev_update_features(dev))
10081 		netdev_features_change(dev);
10082 }
10083 EXPORT_SYMBOL(netdev_update_features);
10084 
10085 /**
10086  *	netdev_change_features - recalculate device features
10087  *	@dev: the device to check
10088  *
10089  *	Recalculate dev->features set and send notifications even
10090  *	if they have not changed. Should be called instead of
10091  *	netdev_update_features() if also dev->vlan_features might
10092  *	have changed to allow the changes to be propagated to stacked
10093  *	VLAN devices.
10094  */
10095 void netdev_change_features(struct net_device *dev)
10096 {
10097 	__netdev_update_features(dev);
10098 	netdev_features_change(dev);
10099 }
10100 EXPORT_SYMBOL(netdev_change_features);
10101 
10102 /**
10103  *	netif_stacked_transfer_operstate -	transfer operstate
10104  *	@rootdev: the root or lower level device to transfer state from
10105  *	@dev: the device to transfer operstate to
10106  *
10107  *	Transfer operational state from root to device. This is normally
10108  *	called when a stacking relationship exists between the root
10109  *	device and the device(a leaf device).
10110  */
10111 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10112 					struct net_device *dev)
10113 {
10114 	if (rootdev->operstate == IF_OPER_DORMANT)
10115 		netif_dormant_on(dev);
10116 	else
10117 		netif_dormant_off(dev);
10118 
10119 	if (rootdev->operstate == IF_OPER_TESTING)
10120 		netif_testing_on(dev);
10121 	else
10122 		netif_testing_off(dev);
10123 
10124 	if (netif_carrier_ok(rootdev))
10125 		netif_carrier_on(dev);
10126 	else
10127 		netif_carrier_off(dev);
10128 }
10129 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10130 
10131 static int netif_alloc_rx_queues(struct net_device *dev)
10132 {
10133 	unsigned int i, count = dev->num_rx_queues;
10134 	struct netdev_rx_queue *rx;
10135 	size_t sz = count * sizeof(*rx);
10136 	int err = 0;
10137 
10138 	BUG_ON(count < 1);
10139 
10140 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10141 	if (!rx)
10142 		return -ENOMEM;
10143 
10144 	dev->_rx = rx;
10145 
10146 	for (i = 0; i < count; i++) {
10147 		rx[i].dev = dev;
10148 
10149 		/* XDP RX-queue setup */
10150 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10151 		if (err < 0)
10152 			goto err_rxq_info;
10153 	}
10154 	return 0;
10155 
10156 err_rxq_info:
10157 	/* Rollback successful reg's and free other resources */
10158 	while (i--)
10159 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10160 	kvfree(dev->_rx);
10161 	dev->_rx = NULL;
10162 	return err;
10163 }
10164 
10165 static void netif_free_rx_queues(struct net_device *dev)
10166 {
10167 	unsigned int i, count = dev->num_rx_queues;
10168 
10169 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10170 	if (!dev->_rx)
10171 		return;
10172 
10173 	for (i = 0; i < count; i++)
10174 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10175 
10176 	kvfree(dev->_rx);
10177 }
10178 
10179 static void netdev_init_one_queue(struct net_device *dev,
10180 				  struct netdev_queue *queue, void *_unused)
10181 {
10182 	/* Initialize queue lock */
10183 	spin_lock_init(&queue->_xmit_lock);
10184 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10185 	queue->xmit_lock_owner = -1;
10186 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10187 	queue->dev = dev;
10188 #ifdef CONFIG_BQL
10189 	dql_init(&queue->dql, HZ);
10190 #endif
10191 }
10192 
10193 static void netif_free_tx_queues(struct net_device *dev)
10194 {
10195 	kvfree(dev->_tx);
10196 }
10197 
10198 static int netif_alloc_netdev_queues(struct net_device *dev)
10199 {
10200 	unsigned int count = dev->num_tx_queues;
10201 	struct netdev_queue *tx;
10202 	size_t sz = count * sizeof(*tx);
10203 
10204 	if (count < 1 || count > 0xffff)
10205 		return -EINVAL;
10206 
10207 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10208 	if (!tx)
10209 		return -ENOMEM;
10210 
10211 	dev->_tx = tx;
10212 
10213 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10214 	spin_lock_init(&dev->tx_global_lock);
10215 
10216 	return 0;
10217 }
10218 
10219 void netif_tx_stop_all_queues(struct net_device *dev)
10220 {
10221 	unsigned int i;
10222 
10223 	for (i = 0; i < dev->num_tx_queues; i++) {
10224 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10225 
10226 		netif_tx_stop_queue(txq);
10227 	}
10228 }
10229 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10230 
10231 /**
10232  *	register_netdevice	- register a network device
10233  *	@dev: device to register
10234  *
10235  *	Take a completed network device structure and add it to the kernel
10236  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10237  *	chain. 0 is returned on success. A negative errno code is returned
10238  *	on a failure to set up the device, or if the name is a duplicate.
10239  *
10240  *	Callers must hold the rtnl semaphore. You may want
10241  *	register_netdev() instead of this.
10242  *
10243  *	BUGS:
10244  *	The locking appears insufficient to guarantee two parallel registers
10245  *	will not get the same name.
10246  */
10247 
10248 int register_netdevice(struct net_device *dev)
10249 {
10250 	int ret;
10251 	struct net *net = dev_net(dev);
10252 
10253 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10254 		     NETDEV_FEATURE_COUNT);
10255 	BUG_ON(dev_boot_phase);
10256 	ASSERT_RTNL();
10257 
10258 	might_sleep();
10259 
10260 	/* When net_device's are persistent, this will be fatal. */
10261 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10262 	BUG_ON(!net);
10263 
10264 	ret = ethtool_check_ops(dev->ethtool_ops);
10265 	if (ret)
10266 		return ret;
10267 
10268 	spin_lock_init(&dev->addr_list_lock);
10269 	netdev_set_addr_lockdep_class(dev);
10270 
10271 	ret = dev_get_valid_name(net, dev, dev->name);
10272 	if (ret < 0)
10273 		goto out;
10274 
10275 	ret = -ENOMEM;
10276 	dev->name_node = netdev_name_node_head_alloc(dev);
10277 	if (!dev->name_node)
10278 		goto out;
10279 
10280 	/* Init, if this function is available */
10281 	if (dev->netdev_ops->ndo_init) {
10282 		ret = dev->netdev_ops->ndo_init(dev);
10283 		if (ret) {
10284 			if (ret > 0)
10285 				ret = -EIO;
10286 			goto err_free_name;
10287 		}
10288 	}
10289 
10290 	if (((dev->hw_features | dev->features) &
10291 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
10292 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10293 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10294 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10295 		ret = -EINVAL;
10296 		goto err_uninit;
10297 	}
10298 
10299 	ret = -EBUSY;
10300 	if (!dev->ifindex)
10301 		dev->ifindex = dev_new_index(net);
10302 	else if (__dev_get_by_index(net, dev->ifindex))
10303 		goto err_uninit;
10304 
10305 	/* Transfer changeable features to wanted_features and enable
10306 	 * software offloads (GSO and GRO).
10307 	 */
10308 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10309 	dev->features |= NETIF_F_SOFT_FEATURES;
10310 
10311 	if (dev->udp_tunnel_nic_info) {
10312 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10313 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10314 	}
10315 
10316 	dev->wanted_features = dev->features & dev->hw_features;
10317 
10318 	if (!(dev->flags & IFF_LOOPBACK))
10319 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10320 
10321 	/* If IPv4 TCP segmentation offload is supported we should also
10322 	 * allow the device to enable segmenting the frame with the option
10323 	 * of ignoring a static IP ID value.  This doesn't enable the
10324 	 * feature itself but allows the user to enable it later.
10325 	 */
10326 	if (dev->hw_features & NETIF_F_TSO)
10327 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10328 	if (dev->vlan_features & NETIF_F_TSO)
10329 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10330 	if (dev->mpls_features & NETIF_F_TSO)
10331 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10332 	if (dev->hw_enc_features & NETIF_F_TSO)
10333 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10334 
10335 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10336 	 */
10337 	dev->vlan_features |= NETIF_F_HIGHDMA;
10338 
10339 	/* Make NETIF_F_SG inheritable to tunnel devices.
10340 	 */
10341 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10342 
10343 	/* Make NETIF_F_SG inheritable to MPLS.
10344 	 */
10345 	dev->mpls_features |= NETIF_F_SG;
10346 
10347 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10348 	ret = notifier_to_errno(ret);
10349 	if (ret)
10350 		goto err_uninit;
10351 
10352 	ret = netdev_register_kobject(dev);
10353 	if (ret) {
10354 		dev->reg_state = NETREG_UNREGISTERED;
10355 		goto err_uninit;
10356 	}
10357 	dev->reg_state = NETREG_REGISTERED;
10358 
10359 	__netdev_update_features(dev);
10360 
10361 	/*
10362 	 *	Default initial state at registry is that the
10363 	 *	device is present.
10364 	 */
10365 
10366 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10367 
10368 	linkwatch_init_dev(dev);
10369 
10370 	dev_init_scheduler(dev);
10371 	dev_hold(dev);
10372 	list_netdevice(dev);
10373 	add_device_randomness(dev->dev_addr, dev->addr_len);
10374 
10375 	/* If the device has permanent device address, driver should
10376 	 * set dev_addr and also addr_assign_type should be set to
10377 	 * NET_ADDR_PERM (default value).
10378 	 */
10379 	if (dev->addr_assign_type == NET_ADDR_PERM)
10380 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10381 
10382 	/* Notify protocols, that a new device appeared. */
10383 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10384 	ret = notifier_to_errno(ret);
10385 	if (ret) {
10386 		/* Expect explicit free_netdev() on failure */
10387 		dev->needs_free_netdev = false;
10388 		unregister_netdevice_queue(dev, NULL);
10389 		goto out;
10390 	}
10391 	/*
10392 	 *	Prevent userspace races by waiting until the network
10393 	 *	device is fully setup before sending notifications.
10394 	 */
10395 	if (!dev->rtnl_link_ops ||
10396 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10397 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10398 
10399 out:
10400 	return ret;
10401 
10402 err_uninit:
10403 	if (dev->netdev_ops->ndo_uninit)
10404 		dev->netdev_ops->ndo_uninit(dev);
10405 	if (dev->priv_destructor)
10406 		dev->priv_destructor(dev);
10407 err_free_name:
10408 	netdev_name_node_free(dev->name_node);
10409 	goto out;
10410 }
10411 EXPORT_SYMBOL(register_netdevice);
10412 
10413 /**
10414  *	init_dummy_netdev	- init a dummy network device for NAPI
10415  *	@dev: device to init
10416  *
10417  *	This takes a network device structure and initialize the minimum
10418  *	amount of fields so it can be used to schedule NAPI polls without
10419  *	registering a full blown interface. This is to be used by drivers
10420  *	that need to tie several hardware interfaces to a single NAPI
10421  *	poll scheduler due to HW limitations.
10422  */
10423 int init_dummy_netdev(struct net_device *dev)
10424 {
10425 	/* Clear everything. Note we don't initialize spinlocks
10426 	 * are they aren't supposed to be taken by any of the
10427 	 * NAPI code and this dummy netdev is supposed to be
10428 	 * only ever used for NAPI polls
10429 	 */
10430 	memset(dev, 0, sizeof(struct net_device));
10431 
10432 	/* make sure we BUG if trying to hit standard
10433 	 * register/unregister code path
10434 	 */
10435 	dev->reg_state = NETREG_DUMMY;
10436 
10437 	/* NAPI wants this */
10438 	INIT_LIST_HEAD(&dev->napi_list);
10439 
10440 	/* a dummy interface is started by default */
10441 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10442 	set_bit(__LINK_STATE_START, &dev->state);
10443 
10444 	/* napi_busy_loop stats accounting wants this */
10445 	dev_net_set(dev, &init_net);
10446 
10447 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10448 	 * because users of this 'device' dont need to change
10449 	 * its refcount.
10450 	 */
10451 
10452 	return 0;
10453 }
10454 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10455 
10456 
10457 /**
10458  *	register_netdev	- register a network device
10459  *	@dev: device to register
10460  *
10461  *	Take a completed network device structure and add it to the kernel
10462  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10463  *	chain. 0 is returned on success. A negative errno code is returned
10464  *	on a failure to set up the device, or if the name is a duplicate.
10465  *
10466  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10467  *	and expands the device name if you passed a format string to
10468  *	alloc_netdev.
10469  */
10470 int register_netdev(struct net_device *dev)
10471 {
10472 	int err;
10473 
10474 	if (rtnl_lock_killable())
10475 		return -EINTR;
10476 	err = register_netdevice(dev);
10477 	rtnl_unlock();
10478 	return err;
10479 }
10480 EXPORT_SYMBOL(register_netdev);
10481 
10482 int netdev_refcnt_read(const struct net_device *dev)
10483 {
10484 #ifdef CONFIG_PCPU_DEV_REFCNT
10485 	int i, refcnt = 0;
10486 
10487 	for_each_possible_cpu(i)
10488 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10489 	return refcnt;
10490 #else
10491 	return refcount_read(&dev->dev_refcnt);
10492 #endif
10493 }
10494 EXPORT_SYMBOL(netdev_refcnt_read);
10495 
10496 int netdev_unregister_timeout_secs __read_mostly = 10;
10497 
10498 #define WAIT_REFS_MIN_MSECS 1
10499 #define WAIT_REFS_MAX_MSECS 250
10500 /**
10501  * netdev_wait_allrefs - wait until all references are gone.
10502  * @dev: target net_device
10503  *
10504  * This is called when unregistering network devices.
10505  *
10506  * Any protocol or device that holds a reference should register
10507  * for netdevice notification, and cleanup and put back the
10508  * reference if they receive an UNREGISTER event.
10509  * We can get stuck here if buggy protocols don't correctly
10510  * call dev_put.
10511  */
10512 static void netdev_wait_allrefs(struct net_device *dev)
10513 {
10514 	unsigned long rebroadcast_time, warning_time;
10515 	int wait = 0, refcnt;
10516 
10517 	linkwatch_forget_dev(dev);
10518 
10519 	rebroadcast_time = warning_time = jiffies;
10520 	refcnt = netdev_refcnt_read(dev);
10521 
10522 	while (refcnt != 1) {
10523 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10524 			rtnl_lock();
10525 
10526 			/* Rebroadcast unregister notification */
10527 			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10528 
10529 			__rtnl_unlock();
10530 			rcu_barrier();
10531 			rtnl_lock();
10532 
10533 			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10534 				     &dev->state)) {
10535 				/* We must not have linkwatch events
10536 				 * pending on unregister. If this
10537 				 * happens, we simply run the queue
10538 				 * unscheduled, resulting in a noop
10539 				 * for this device.
10540 				 */
10541 				linkwatch_run_queue();
10542 			}
10543 
10544 			__rtnl_unlock();
10545 
10546 			rebroadcast_time = jiffies;
10547 		}
10548 
10549 		if (!wait) {
10550 			rcu_barrier();
10551 			wait = WAIT_REFS_MIN_MSECS;
10552 		} else {
10553 			msleep(wait);
10554 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10555 		}
10556 
10557 		refcnt = netdev_refcnt_read(dev);
10558 
10559 		if (refcnt != 1 &&
10560 		    time_after(jiffies, warning_time +
10561 			       netdev_unregister_timeout_secs * HZ)) {
10562 			pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10563 				 dev->name, refcnt);
10564 			warning_time = jiffies;
10565 		}
10566 	}
10567 }
10568 
10569 /* The sequence is:
10570  *
10571  *	rtnl_lock();
10572  *	...
10573  *	register_netdevice(x1);
10574  *	register_netdevice(x2);
10575  *	...
10576  *	unregister_netdevice(y1);
10577  *	unregister_netdevice(y2);
10578  *      ...
10579  *	rtnl_unlock();
10580  *	free_netdev(y1);
10581  *	free_netdev(y2);
10582  *
10583  * We are invoked by rtnl_unlock().
10584  * This allows us to deal with problems:
10585  * 1) We can delete sysfs objects which invoke hotplug
10586  *    without deadlocking with linkwatch via keventd.
10587  * 2) Since we run with the RTNL semaphore not held, we can sleep
10588  *    safely in order to wait for the netdev refcnt to drop to zero.
10589  *
10590  * We must not return until all unregister events added during
10591  * the interval the lock was held have been completed.
10592  */
10593 void netdev_run_todo(void)
10594 {
10595 	struct list_head list;
10596 #ifdef CONFIG_LOCKDEP
10597 	struct list_head unlink_list;
10598 
10599 	list_replace_init(&net_unlink_list, &unlink_list);
10600 
10601 	while (!list_empty(&unlink_list)) {
10602 		struct net_device *dev = list_first_entry(&unlink_list,
10603 							  struct net_device,
10604 							  unlink_list);
10605 		list_del_init(&dev->unlink_list);
10606 		dev->nested_level = dev->lower_level - 1;
10607 	}
10608 #endif
10609 
10610 	/* Snapshot list, allow later requests */
10611 	list_replace_init(&net_todo_list, &list);
10612 
10613 	__rtnl_unlock();
10614 
10615 
10616 	/* Wait for rcu callbacks to finish before next phase */
10617 	if (!list_empty(&list))
10618 		rcu_barrier();
10619 
10620 	while (!list_empty(&list)) {
10621 		struct net_device *dev
10622 			= list_first_entry(&list, struct net_device, todo_list);
10623 		list_del(&dev->todo_list);
10624 
10625 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10626 			pr_err("network todo '%s' but state %d\n",
10627 			       dev->name, dev->reg_state);
10628 			dump_stack();
10629 			continue;
10630 		}
10631 
10632 		dev->reg_state = NETREG_UNREGISTERED;
10633 
10634 		netdev_wait_allrefs(dev);
10635 
10636 		/* paranoia */
10637 		BUG_ON(netdev_refcnt_read(dev) != 1);
10638 		BUG_ON(!list_empty(&dev->ptype_all));
10639 		BUG_ON(!list_empty(&dev->ptype_specific));
10640 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10641 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10642 #if IS_ENABLED(CONFIG_DECNET)
10643 		WARN_ON(dev->dn_ptr);
10644 #endif
10645 		if (dev->priv_destructor)
10646 			dev->priv_destructor(dev);
10647 		if (dev->needs_free_netdev)
10648 			free_netdev(dev);
10649 
10650 		/* Report a network device has been unregistered */
10651 		rtnl_lock();
10652 		dev_net(dev)->dev_unreg_count--;
10653 		__rtnl_unlock();
10654 		wake_up(&netdev_unregistering_wq);
10655 
10656 		/* Free network device */
10657 		kobject_put(&dev->dev.kobj);
10658 	}
10659 }
10660 
10661 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10662  * all the same fields in the same order as net_device_stats, with only
10663  * the type differing, but rtnl_link_stats64 may have additional fields
10664  * at the end for newer counters.
10665  */
10666 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10667 			     const struct net_device_stats *netdev_stats)
10668 {
10669 #if BITS_PER_LONG == 64
10670 	BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10671 	memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10672 	/* zero out counters that only exist in rtnl_link_stats64 */
10673 	memset((char *)stats64 + sizeof(*netdev_stats), 0,
10674 	       sizeof(*stats64) - sizeof(*netdev_stats));
10675 #else
10676 	size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10677 	const unsigned long *src = (const unsigned long *)netdev_stats;
10678 	u64 *dst = (u64 *)stats64;
10679 
10680 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10681 	for (i = 0; i < n; i++)
10682 		dst[i] = src[i];
10683 	/* zero out counters that only exist in rtnl_link_stats64 */
10684 	memset((char *)stats64 + n * sizeof(u64), 0,
10685 	       sizeof(*stats64) - n * sizeof(u64));
10686 #endif
10687 }
10688 EXPORT_SYMBOL(netdev_stats_to_stats64);
10689 
10690 /**
10691  *	dev_get_stats	- get network device statistics
10692  *	@dev: device to get statistics from
10693  *	@storage: place to store stats
10694  *
10695  *	Get network statistics from device. Return @storage.
10696  *	The device driver may provide its own method by setting
10697  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10698  *	otherwise the internal statistics structure is used.
10699  */
10700 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10701 					struct rtnl_link_stats64 *storage)
10702 {
10703 	const struct net_device_ops *ops = dev->netdev_ops;
10704 
10705 	if (ops->ndo_get_stats64) {
10706 		memset(storage, 0, sizeof(*storage));
10707 		ops->ndo_get_stats64(dev, storage);
10708 	} else if (ops->ndo_get_stats) {
10709 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10710 	} else {
10711 		netdev_stats_to_stats64(storage, &dev->stats);
10712 	}
10713 	storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10714 	storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10715 	storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10716 	return storage;
10717 }
10718 EXPORT_SYMBOL(dev_get_stats);
10719 
10720 /**
10721  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10722  *	@s: place to store stats
10723  *	@netstats: per-cpu network stats to read from
10724  *
10725  *	Read per-cpu network statistics and populate the related fields in @s.
10726  */
10727 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10728 			   const struct pcpu_sw_netstats __percpu *netstats)
10729 {
10730 	int cpu;
10731 
10732 	for_each_possible_cpu(cpu) {
10733 		const struct pcpu_sw_netstats *stats;
10734 		struct pcpu_sw_netstats tmp;
10735 		unsigned int start;
10736 
10737 		stats = per_cpu_ptr(netstats, cpu);
10738 		do {
10739 			start = u64_stats_fetch_begin_irq(&stats->syncp);
10740 			tmp.rx_packets = stats->rx_packets;
10741 			tmp.rx_bytes   = stats->rx_bytes;
10742 			tmp.tx_packets = stats->tx_packets;
10743 			tmp.tx_bytes   = stats->tx_bytes;
10744 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10745 
10746 		s->rx_packets += tmp.rx_packets;
10747 		s->rx_bytes   += tmp.rx_bytes;
10748 		s->tx_packets += tmp.tx_packets;
10749 		s->tx_bytes   += tmp.tx_bytes;
10750 	}
10751 }
10752 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10753 
10754 /**
10755  *	dev_get_tstats64 - ndo_get_stats64 implementation
10756  *	@dev: device to get statistics from
10757  *	@s: place to store stats
10758  *
10759  *	Populate @s from dev->stats and dev->tstats. Can be used as
10760  *	ndo_get_stats64() callback.
10761  */
10762 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10763 {
10764 	netdev_stats_to_stats64(s, &dev->stats);
10765 	dev_fetch_sw_netstats(s, dev->tstats);
10766 }
10767 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10768 
10769 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10770 {
10771 	struct netdev_queue *queue = dev_ingress_queue(dev);
10772 
10773 #ifdef CONFIG_NET_CLS_ACT
10774 	if (queue)
10775 		return queue;
10776 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10777 	if (!queue)
10778 		return NULL;
10779 	netdev_init_one_queue(dev, queue, NULL);
10780 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10781 	queue->qdisc_sleeping = &noop_qdisc;
10782 	rcu_assign_pointer(dev->ingress_queue, queue);
10783 #endif
10784 	return queue;
10785 }
10786 
10787 static const struct ethtool_ops default_ethtool_ops;
10788 
10789 void netdev_set_default_ethtool_ops(struct net_device *dev,
10790 				    const struct ethtool_ops *ops)
10791 {
10792 	if (dev->ethtool_ops == &default_ethtool_ops)
10793 		dev->ethtool_ops = ops;
10794 }
10795 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10796 
10797 void netdev_freemem(struct net_device *dev)
10798 {
10799 	char *addr = (char *)dev - dev->padded;
10800 
10801 	kvfree(addr);
10802 }
10803 
10804 /**
10805  * alloc_netdev_mqs - allocate network device
10806  * @sizeof_priv: size of private data to allocate space for
10807  * @name: device name format string
10808  * @name_assign_type: origin of device name
10809  * @setup: callback to initialize device
10810  * @txqs: the number of TX subqueues to allocate
10811  * @rxqs: the number of RX subqueues to allocate
10812  *
10813  * Allocates a struct net_device with private data area for driver use
10814  * and performs basic initialization.  Also allocates subqueue structs
10815  * for each queue on the device.
10816  */
10817 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10818 		unsigned char name_assign_type,
10819 		void (*setup)(struct net_device *),
10820 		unsigned int txqs, unsigned int rxqs)
10821 {
10822 	struct net_device *dev;
10823 	unsigned int alloc_size;
10824 	struct net_device *p;
10825 
10826 	BUG_ON(strlen(name) >= sizeof(dev->name));
10827 
10828 	if (txqs < 1) {
10829 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10830 		return NULL;
10831 	}
10832 
10833 	if (rxqs < 1) {
10834 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10835 		return NULL;
10836 	}
10837 
10838 	alloc_size = sizeof(struct net_device);
10839 	if (sizeof_priv) {
10840 		/* ensure 32-byte alignment of private area */
10841 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10842 		alloc_size += sizeof_priv;
10843 	}
10844 	/* ensure 32-byte alignment of whole construct */
10845 	alloc_size += NETDEV_ALIGN - 1;
10846 
10847 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10848 	if (!p)
10849 		return NULL;
10850 
10851 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10852 	dev->padded = (char *)dev - (char *)p;
10853 
10854 #ifdef CONFIG_PCPU_DEV_REFCNT
10855 	dev->pcpu_refcnt = alloc_percpu(int);
10856 	if (!dev->pcpu_refcnt)
10857 		goto free_dev;
10858 	dev_hold(dev);
10859 #else
10860 	refcount_set(&dev->dev_refcnt, 1);
10861 #endif
10862 
10863 	if (dev_addr_init(dev))
10864 		goto free_pcpu;
10865 
10866 	dev_mc_init(dev);
10867 	dev_uc_init(dev);
10868 
10869 	dev_net_set(dev, &init_net);
10870 
10871 	dev->gso_max_size = GSO_MAX_SIZE;
10872 	dev->gso_max_segs = GSO_MAX_SEGS;
10873 	dev->upper_level = 1;
10874 	dev->lower_level = 1;
10875 #ifdef CONFIG_LOCKDEP
10876 	dev->nested_level = 0;
10877 	INIT_LIST_HEAD(&dev->unlink_list);
10878 #endif
10879 
10880 	INIT_LIST_HEAD(&dev->napi_list);
10881 	INIT_LIST_HEAD(&dev->unreg_list);
10882 	INIT_LIST_HEAD(&dev->close_list);
10883 	INIT_LIST_HEAD(&dev->link_watch_list);
10884 	INIT_LIST_HEAD(&dev->adj_list.upper);
10885 	INIT_LIST_HEAD(&dev->adj_list.lower);
10886 	INIT_LIST_HEAD(&dev->ptype_all);
10887 	INIT_LIST_HEAD(&dev->ptype_specific);
10888 	INIT_LIST_HEAD(&dev->net_notifier_list);
10889 #ifdef CONFIG_NET_SCHED
10890 	hash_init(dev->qdisc_hash);
10891 #endif
10892 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10893 	setup(dev);
10894 
10895 	if (!dev->tx_queue_len) {
10896 		dev->priv_flags |= IFF_NO_QUEUE;
10897 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10898 	}
10899 
10900 	dev->num_tx_queues = txqs;
10901 	dev->real_num_tx_queues = txqs;
10902 	if (netif_alloc_netdev_queues(dev))
10903 		goto free_all;
10904 
10905 	dev->num_rx_queues = rxqs;
10906 	dev->real_num_rx_queues = rxqs;
10907 	if (netif_alloc_rx_queues(dev))
10908 		goto free_all;
10909 
10910 	strcpy(dev->name, name);
10911 	dev->name_assign_type = name_assign_type;
10912 	dev->group = INIT_NETDEV_GROUP;
10913 	if (!dev->ethtool_ops)
10914 		dev->ethtool_ops = &default_ethtool_ops;
10915 
10916 	nf_hook_ingress_init(dev);
10917 
10918 	return dev;
10919 
10920 free_all:
10921 	free_netdev(dev);
10922 	return NULL;
10923 
10924 free_pcpu:
10925 #ifdef CONFIG_PCPU_DEV_REFCNT
10926 	free_percpu(dev->pcpu_refcnt);
10927 free_dev:
10928 #endif
10929 	netdev_freemem(dev);
10930 	return NULL;
10931 }
10932 EXPORT_SYMBOL(alloc_netdev_mqs);
10933 
10934 /**
10935  * free_netdev - free network device
10936  * @dev: device
10937  *
10938  * This function does the last stage of destroying an allocated device
10939  * interface. The reference to the device object is released. If this
10940  * is the last reference then it will be freed.Must be called in process
10941  * context.
10942  */
10943 void free_netdev(struct net_device *dev)
10944 {
10945 	struct napi_struct *p, *n;
10946 
10947 	might_sleep();
10948 
10949 	/* When called immediately after register_netdevice() failed the unwind
10950 	 * handling may still be dismantling the device. Handle that case by
10951 	 * deferring the free.
10952 	 */
10953 	if (dev->reg_state == NETREG_UNREGISTERING) {
10954 		ASSERT_RTNL();
10955 		dev->needs_free_netdev = true;
10956 		return;
10957 	}
10958 
10959 	netif_free_tx_queues(dev);
10960 	netif_free_rx_queues(dev);
10961 
10962 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10963 
10964 	/* Flush device addresses */
10965 	dev_addr_flush(dev);
10966 
10967 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10968 		netif_napi_del(p);
10969 
10970 #ifdef CONFIG_PCPU_DEV_REFCNT
10971 	free_percpu(dev->pcpu_refcnt);
10972 	dev->pcpu_refcnt = NULL;
10973 #endif
10974 	free_percpu(dev->xdp_bulkq);
10975 	dev->xdp_bulkq = NULL;
10976 
10977 	/*  Compatibility with error handling in drivers */
10978 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10979 		netdev_freemem(dev);
10980 		return;
10981 	}
10982 
10983 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10984 	dev->reg_state = NETREG_RELEASED;
10985 
10986 	/* will free via device release */
10987 	put_device(&dev->dev);
10988 }
10989 EXPORT_SYMBOL(free_netdev);
10990 
10991 /**
10992  *	synchronize_net -  Synchronize with packet receive processing
10993  *
10994  *	Wait for packets currently being received to be done.
10995  *	Does not block later packets from starting.
10996  */
10997 void synchronize_net(void)
10998 {
10999 	might_sleep();
11000 	if (rtnl_is_locked())
11001 		synchronize_rcu_expedited();
11002 	else
11003 		synchronize_rcu();
11004 }
11005 EXPORT_SYMBOL(synchronize_net);
11006 
11007 /**
11008  *	unregister_netdevice_queue - remove device from the kernel
11009  *	@dev: device
11010  *	@head: list
11011  *
11012  *	This function shuts down a device interface and removes it
11013  *	from the kernel tables.
11014  *	If head not NULL, device is queued to be unregistered later.
11015  *
11016  *	Callers must hold the rtnl semaphore.  You may want
11017  *	unregister_netdev() instead of this.
11018  */
11019 
11020 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11021 {
11022 	ASSERT_RTNL();
11023 
11024 	if (head) {
11025 		list_move_tail(&dev->unreg_list, head);
11026 	} else {
11027 		LIST_HEAD(single);
11028 
11029 		list_add(&dev->unreg_list, &single);
11030 		unregister_netdevice_many(&single);
11031 	}
11032 }
11033 EXPORT_SYMBOL(unregister_netdevice_queue);
11034 
11035 /**
11036  *	unregister_netdevice_many - unregister many devices
11037  *	@head: list of devices
11038  *
11039  *  Note: As most callers use a stack allocated list_head,
11040  *  we force a list_del() to make sure stack wont be corrupted later.
11041  */
11042 void unregister_netdevice_many(struct list_head *head)
11043 {
11044 	struct net_device *dev, *tmp;
11045 	LIST_HEAD(close_head);
11046 
11047 	BUG_ON(dev_boot_phase);
11048 	ASSERT_RTNL();
11049 
11050 	if (list_empty(head))
11051 		return;
11052 
11053 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11054 		/* Some devices call without registering
11055 		 * for initialization unwind. Remove those
11056 		 * devices and proceed with the remaining.
11057 		 */
11058 		if (dev->reg_state == NETREG_UNINITIALIZED) {
11059 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11060 				 dev->name, dev);
11061 
11062 			WARN_ON(1);
11063 			list_del(&dev->unreg_list);
11064 			continue;
11065 		}
11066 		dev->dismantle = true;
11067 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
11068 	}
11069 
11070 	/* If device is running, close it first. */
11071 	list_for_each_entry(dev, head, unreg_list)
11072 		list_add_tail(&dev->close_list, &close_head);
11073 	dev_close_many(&close_head, true);
11074 
11075 	list_for_each_entry(dev, head, unreg_list) {
11076 		/* And unlink it from device chain. */
11077 		unlist_netdevice(dev);
11078 
11079 		dev->reg_state = NETREG_UNREGISTERING;
11080 	}
11081 	flush_all_backlogs();
11082 
11083 	synchronize_net();
11084 
11085 	list_for_each_entry(dev, head, unreg_list) {
11086 		struct sk_buff *skb = NULL;
11087 
11088 		/* Shutdown queueing discipline. */
11089 		dev_shutdown(dev);
11090 
11091 		dev_xdp_uninstall(dev);
11092 
11093 		/* Notify protocols, that we are about to destroy
11094 		 * this device. They should clean all the things.
11095 		 */
11096 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11097 
11098 		if (!dev->rtnl_link_ops ||
11099 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11100 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11101 						     GFP_KERNEL, NULL, 0);
11102 
11103 		/*
11104 		 *	Flush the unicast and multicast chains
11105 		 */
11106 		dev_uc_flush(dev);
11107 		dev_mc_flush(dev);
11108 
11109 		netdev_name_node_alt_flush(dev);
11110 		netdev_name_node_free(dev->name_node);
11111 
11112 		if (dev->netdev_ops->ndo_uninit)
11113 			dev->netdev_ops->ndo_uninit(dev);
11114 
11115 		if (skb)
11116 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11117 
11118 		/* Notifier chain MUST detach us all upper devices. */
11119 		WARN_ON(netdev_has_any_upper_dev(dev));
11120 		WARN_ON(netdev_has_any_lower_dev(dev));
11121 
11122 		/* Remove entries from kobject tree */
11123 		netdev_unregister_kobject(dev);
11124 #ifdef CONFIG_XPS
11125 		/* Remove XPS queueing entries */
11126 		netif_reset_xps_queues_gt(dev, 0);
11127 #endif
11128 	}
11129 
11130 	synchronize_net();
11131 
11132 	list_for_each_entry(dev, head, unreg_list) {
11133 		dev_put(dev);
11134 		net_set_todo(dev);
11135 	}
11136 
11137 	list_del(head);
11138 }
11139 EXPORT_SYMBOL(unregister_netdevice_many);
11140 
11141 /**
11142  *	unregister_netdev - remove device from the kernel
11143  *	@dev: device
11144  *
11145  *	This function shuts down a device interface and removes it
11146  *	from the kernel tables.
11147  *
11148  *	This is just a wrapper for unregister_netdevice that takes
11149  *	the rtnl semaphore.  In general you want to use this and not
11150  *	unregister_netdevice.
11151  */
11152 void unregister_netdev(struct net_device *dev)
11153 {
11154 	rtnl_lock();
11155 	unregister_netdevice(dev);
11156 	rtnl_unlock();
11157 }
11158 EXPORT_SYMBOL(unregister_netdev);
11159 
11160 /**
11161  *	__dev_change_net_namespace - move device to different nethost namespace
11162  *	@dev: device
11163  *	@net: network namespace
11164  *	@pat: If not NULL name pattern to try if the current device name
11165  *	      is already taken in the destination network namespace.
11166  *	@new_ifindex: If not zero, specifies device index in the target
11167  *	              namespace.
11168  *
11169  *	This function shuts down a device interface and moves it
11170  *	to a new network namespace. On success 0 is returned, on
11171  *	a failure a netagive errno code is returned.
11172  *
11173  *	Callers must hold the rtnl semaphore.
11174  */
11175 
11176 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11177 			       const char *pat, int new_ifindex)
11178 {
11179 	struct net *net_old = dev_net(dev);
11180 	int err, new_nsid;
11181 
11182 	ASSERT_RTNL();
11183 
11184 	/* Don't allow namespace local devices to be moved. */
11185 	err = -EINVAL;
11186 	if (dev->features & NETIF_F_NETNS_LOCAL)
11187 		goto out;
11188 
11189 	/* Ensure the device has been registrered */
11190 	if (dev->reg_state != NETREG_REGISTERED)
11191 		goto out;
11192 
11193 	/* Get out if there is nothing todo */
11194 	err = 0;
11195 	if (net_eq(net_old, net))
11196 		goto out;
11197 
11198 	/* Pick the destination device name, and ensure
11199 	 * we can use it in the destination network namespace.
11200 	 */
11201 	err = -EEXIST;
11202 	if (__dev_get_by_name(net, dev->name)) {
11203 		/* We get here if we can't use the current device name */
11204 		if (!pat)
11205 			goto out;
11206 		err = dev_get_valid_name(net, dev, pat);
11207 		if (err < 0)
11208 			goto out;
11209 	}
11210 
11211 	/* Check that new_ifindex isn't used yet. */
11212 	err = -EBUSY;
11213 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11214 		goto out;
11215 
11216 	/*
11217 	 * And now a mini version of register_netdevice unregister_netdevice.
11218 	 */
11219 
11220 	/* If device is running close it first. */
11221 	dev_close(dev);
11222 
11223 	/* And unlink it from device chain */
11224 	unlist_netdevice(dev);
11225 
11226 	synchronize_net();
11227 
11228 	/* Shutdown queueing discipline. */
11229 	dev_shutdown(dev);
11230 
11231 	/* Notify protocols, that we are about to destroy
11232 	 * this device. They should clean all the things.
11233 	 *
11234 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
11235 	 * This is wanted because this way 8021q and macvlan know
11236 	 * the device is just moving and can keep their slaves up.
11237 	 */
11238 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11239 	rcu_barrier();
11240 
11241 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11242 	/* If there is an ifindex conflict assign a new one */
11243 	if (!new_ifindex) {
11244 		if (__dev_get_by_index(net, dev->ifindex))
11245 			new_ifindex = dev_new_index(net);
11246 		else
11247 			new_ifindex = dev->ifindex;
11248 	}
11249 
11250 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11251 			    new_ifindex);
11252 
11253 	/*
11254 	 *	Flush the unicast and multicast chains
11255 	 */
11256 	dev_uc_flush(dev);
11257 	dev_mc_flush(dev);
11258 
11259 	/* Send a netdev-removed uevent to the old namespace */
11260 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11261 	netdev_adjacent_del_links(dev);
11262 
11263 	/* Move per-net netdevice notifiers that are following the netdevice */
11264 	move_netdevice_notifiers_dev_net(dev, net);
11265 
11266 	/* Actually switch the network namespace */
11267 	dev_net_set(dev, net);
11268 	dev->ifindex = new_ifindex;
11269 
11270 	/* Send a netdev-add uevent to the new namespace */
11271 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11272 	netdev_adjacent_add_links(dev);
11273 
11274 	/* Fixup kobjects */
11275 	err = device_rename(&dev->dev, dev->name);
11276 	WARN_ON(err);
11277 
11278 	/* Adapt owner in case owning user namespace of target network
11279 	 * namespace is different from the original one.
11280 	 */
11281 	err = netdev_change_owner(dev, net_old, net);
11282 	WARN_ON(err);
11283 
11284 	/* Add the device back in the hashes */
11285 	list_netdevice(dev);
11286 
11287 	/* Notify protocols, that a new device appeared. */
11288 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11289 
11290 	/*
11291 	 *	Prevent userspace races by waiting until the network
11292 	 *	device is fully setup before sending notifications.
11293 	 */
11294 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11295 
11296 	synchronize_net();
11297 	err = 0;
11298 out:
11299 	return err;
11300 }
11301 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11302 
11303 static int dev_cpu_dead(unsigned int oldcpu)
11304 {
11305 	struct sk_buff **list_skb;
11306 	struct sk_buff *skb;
11307 	unsigned int cpu;
11308 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11309 
11310 	local_irq_disable();
11311 	cpu = smp_processor_id();
11312 	sd = &per_cpu(softnet_data, cpu);
11313 	oldsd = &per_cpu(softnet_data, oldcpu);
11314 
11315 	/* Find end of our completion_queue. */
11316 	list_skb = &sd->completion_queue;
11317 	while (*list_skb)
11318 		list_skb = &(*list_skb)->next;
11319 	/* Append completion queue from offline CPU. */
11320 	*list_skb = oldsd->completion_queue;
11321 	oldsd->completion_queue = NULL;
11322 
11323 	/* Append output queue from offline CPU. */
11324 	if (oldsd->output_queue) {
11325 		*sd->output_queue_tailp = oldsd->output_queue;
11326 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11327 		oldsd->output_queue = NULL;
11328 		oldsd->output_queue_tailp = &oldsd->output_queue;
11329 	}
11330 	/* Append NAPI poll list from offline CPU, with one exception :
11331 	 * process_backlog() must be called by cpu owning percpu backlog.
11332 	 * We properly handle process_queue & input_pkt_queue later.
11333 	 */
11334 	while (!list_empty(&oldsd->poll_list)) {
11335 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11336 							    struct napi_struct,
11337 							    poll_list);
11338 
11339 		list_del_init(&napi->poll_list);
11340 		if (napi->poll == process_backlog)
11341 			napi->state = 0;
11342 		else
11343 			____napi_schedule(sd, napi);
11344 	}
11345 
11346 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11347 	local_irq_enable();
11348 
11349 #ifdef CONFIG_RPS
11350 	remsd = oldsd->rps_ipi_list;
11351 	oldsd->rps_ipi_list = NULL;
11352 #endif
11353 	/* send out pending IPI's on offline CPU */
11354 	net_rps_send_ipi(remsd);
11355 
11356 	/* Process offline CPU's input_pkt_queue */
11357 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11358 		netif_rx_ni(skb);
11359 		input_queue_head_incr(oldsd);
11360 	}
11361 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11362 		netif_rx_ni(skb);
11363 		input_queue_head_incr(oldsd);
11364 	}
11365 
11366 	return 0;
11367 }
11368 
11369 /**
11370  *	netdev_increment_features - increment feature set by one
11371  *	@all: current feature set
11372  *	@one: new feature set
11373  *	@mask: mask feature set
11374  *
11375  *	Computes a new feature set after adding a device with feature set
11376  *	@one to the master device with current feature set @all.  Will not
11377  *	enable anything that is off in @mask. Returns the new feature set.
11378  */
11379 netdev_features_t netdev_increment_features(netdev_features_t all,
11380 	netdev_features_t one, netdev_features_t mask)
11381 {
11382 	if (mask & NETIF_F_HW_CSUM)
11383 		mask |= NETIF_F_CSUM_MASK;
11384 	mask |= NETIF_F_VLAN_CHALLENGED;
11385 
11386 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11387 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11388 
11389 	/* If one device supports hw checksumming, set for all. */
11390 	if (all & NETIF_F_HW_CSUM)
11391 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11392 
11393 	return all;
11394 }
11395 EXPORT_SYMBOL(netdev_increment_features);
11396 
11397 static struct hlist_head * __net_init netdev_create_hash(void)
11398 {
11399 	int i;
11400 	struct hlist_head *hash;
11401 
11402 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11403 	if (hash != NULL)
11404 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11405 			INIT_HLIST_HEAD(&hash[i]);
11406 
11407 	return hash;
11408 }
11409 
11410 /* Initialize per network namespace state */
11411 static int __net_init netdev_init(struct net *net)
11412 {
11413 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11414 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11415 
11416 	if (net != &init_net)
11417 		INIT_LIST_HEAD(&net->dev_base_head);
11418 
11419 	net->dev_name_head = netdev_create_hash();
11420 	if (net->dev_name_head == NULL)
11421 		goto err_name;
11422 
11423 	net->dev_index_head = netdev_create_hash();
11424 	if (net->dev_index_head == NULL)
11425 		goto err_idx;
11426 
11427 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11428 
11429 	return 0;
11430 
11431 err_idx:
11432 	kfree(net->dev_name_head);
11433 err_name:
11434 	return -ENOMEM;
11435 }
11436 
11437 /**
11438  *	netdev_drivername - network driver for the device
11439  *	@dev: network device
11440  *
11441  *	Determine network driver for device.
11442  */
11443 const char *netdev_drivername(const struct net_device *dev)
11444 {
11445 	const struct device_driver *driver;
11446 	const struct device *parent;
11447 	const char *empty = "";
11448 
11449 	parent = dev->dev.parent;
11450 	if (!parent)
11451 		return empty;
11452 
11453 	driver = parent->driver;
11454 	if (driver && driver->name)
11455 		return driver->name;
11456 	return empty;
11457 }
11458 
11459 static void __netdev_printk(const char *level, const struct net_device *dev,
11460 			    struct va_format *vaf)
11461 {
11462 	if (dev && dev->dev.parent) {
11463 		dev_printk_emit(level[1] - '0',
11464 				dev->dev.parent,
11465 				"%s %s %s%s: %pV",
11466 				dev_driver_string(dev->dev.parent),
11467 				dev_name(dev->dev.parent),
11468 				netdev_name(dev), netdev_reg_state(dev),
11469 				vaf);
11470 	} else if (dev) {
11471 		printk("%s%s%s: %pV",
11472 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11473 	} else {
11474 		printk("%s(NULL net_device): %pV", level, vaf);
11475 	}
11476 }
11477 
11478 void netdev_printk(const char *level, const struct net_device *dev,
11479 		   const char *format, ...)
11480 {
11481 	struct va_format vaf;
11482 	va_list args;
11483 
11484 	va_start(args, format);
11485 
11486 	vaf.fmt = format;
11487 	vaf.va = &args;
11488 
11489 	__netdev_printk(level, dev, &vaf);
11490 
11491 	va_end(args);
11492 }
11493 EXPORT_SYMBOL(netdev_printk);
11494 
11495 #define define_netdev_printk_level(func, level)			\
11496 void func(const struct net_device *dev, const char *fmt, ...)	\
11497 {								\
11498 	struct va_format vaf;					\
11499 	va_list args;						\
11500 								\
11501 	va_start(args, fmt);					\
11502 								\
11503 	vaf.fmt = fmt;						\
11504 	vaf.va = &args;						\
11505 								\
11506 	__netdev_printk(level, dev, &vaf);			\
11507 								\
11508 	va_end(args);						\
11509 }								\
11510 EXPORT_SYMBOL(func);
11511 
11512 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11513 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11514 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11515 define_netdev_printk_level(netdev_err, KERN_ERR);
11516 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11517 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11518 define_netdev_printk_level(netdev_info, KERN_INFO);
11519 
11520 static void __net_exit netdev_exit(struct net *net)
11521 {
11522 	kfree(net->dev_name_head);
11523 	kfree(net->dev_index_head);
11524 	if (net != &init_net)
11525 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11526 }
11527 
11528 static struct pernet_operations __net_initdata netdev_net_ops = {
11529 	.init = netdev_init,
11530 	.exit = netdev_exit,
11531 };
11532 
11533 static void __net_exit default_device_exit(struct net *net)
11534 {
11535 	struct net_device *dev, *aux;
11536 	/*
11537 	 * Push all migratable network devices back to the
11538 	 * initial network namespace
11539 	 */
11540 	rtnl_lock();
11541 	for_each_netdev_safe(net, dev, aux) {
11542 		int err;
11543 		char fb_name[IFNAMSIZ];
11544 
11545 		/* Ignore unmoveable devices (i.e. loopback) */
11546 		if (dev->features & NETIF_F_NETNS_LOCAL)
11547 			continue;
11548 
11549 		/* Leave virtual devices for the generic cleanup */
11550 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11551 			continue;
11552 
11553 		/* Push remaining network devices to init_net */
11554 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11555 		if (__dev_get_by_name(&init_net, fb_name))
11556 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11557 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11558 		if (err) {
11559 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11560 				 __func__, dev->name, err);
11561 			BUG();
11562 		}
11563 	}
11564 	rtnl_unlock();
11565 }
11566 
11567 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11568 {
11569 	/* Return with the rtnl_lock held when there are no network
11570 	 * devices unregistering in any network namespace in net_list.
11571 	 */
11572 	struct net *net;
11573 	bool unregistering;
11574 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
11575 
11576 	add_wait_queue(&netdev_unregistering_wq, &wait);
11577 	for (;;) {
11578 		unregistering = false;
11579 		rtnl_lock();
11580 		list_for_each_entry(net, net_list, exit_list) {
11581 			if (net->dev_unreg_count > 0) {
11582 				unregistering = true;
11583 				break;
11584 			}
11585 		}
11586 		if (!unregistering)
11587 			break;
11588 		__rtnl_unlock();
11589 
11590 		wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11591 	}
11592 	remove_wait_queue(&netdev_unregistering_wq, &wait);
11593 }
11594 
11595 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11596 {
11597 	/* At exit all network devices most be removed from a network
11598 	 * namespace.  Do this in the reverse order of registration.
11599 	 * Do this across as many network namespaces as possible to
11600 	 * improve batching efficiency.
11601 	 */
11602 	struct net_device *dev;
11603 	struct net *net;
11604 	LIST_HEAD(dev_kill_list);
11605 
11606 	/* To prevent network device cleanup code from dereferencing
11607 	 * loopback devices or network devices that have been freed
11608 	 * wait here for all pending unregistrations to complete,
11609 	 * before unregistring the loopback device and allowing the
11610 	 * network namespace be freed.
11611 	 *
11612 	 * The netdev todo list containing all network devices
11613 	 * unregistrations that happen in default_device_exit_batch
11614 	 * will run in the rtnl_unlock() at the end of
11615 	 * default_device_exit_batch.
11616 	 */
11617 	rtnl_lock_unregistering(net_list);
11618 	list_for_each_entry(net, net_list, exit_list) {
11619 		for_each_netdev_reverse(net, dev) {
11620 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11621 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11622 			else
11623 				unregister_netdevice_queue(dev, &dev_kill_list);
11624 		}
11625 	}
11626 	unregister_netdevice_many(&dev_kill_list);
11627 	rtnl_unlock();
11628 }
11629 
11630 static struct pernet_operations __net_initdata default_device_ops = {
11631 	.exit = default_device_exit,
11632 	.exit_batch = default_device_exit_batch,
11633 };
11634 
11635 /*
11636  *	Initialize the DEV module. At boot time this walks the device list and
11637  *	unhooks any devices that fail to initialise (normally hardware not
11638  *	present) and leaves us with a valid list of present and active devices.
11639  *
11640  */
11641 
11642 /*
11643  *       This is called single threaded during boot, so no need
11644  *       to take the rtnl semaphore.
11645  */
11646 static int __init net_dev_init(void)
11647 {
11648 	int i, rc = -ENOMEM;
11649 
11650 	BUG_ON(!dev_boot_phase);
11651 
11652 	if (dev_proc_init())
11653 		goto out;
11654 
11655 	if (netdev_kobject_init())
11656 		goto out;
11657 
11658 	INIT_LIST_HEAD(&ptype_all);
11659 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11660 		INIT_LIST_HEAD(&ptype_base[i]);
11661 
11662 	INIT_LIST_HEAD(&offload_base);
11663 
11664 	if (register_pernet_subsys(&netdev_net_ops))
11665 		goto out;
11666 
11667 	/*
11668 	 *	Initialise the packet receive queues.
11669 	 */
11670 
11671 	for_each_possible_cpu(i) {
11672 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11673 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11674 
11675 		INIT_WORK(flush, flush_backlog);
11676 
11677 		skb_queue_head_init(&sd->input_pkt_queue);
11678 		skb_queue_head_init(&sd->process_queue);
11679 #ifdef CONFIG_XFRM_OFFLOAD
11680 		skb_queue_head_init(&sd->xfrm_backlog);
11681 #endif
11682 		INIT_LIST_HEAD(&sd->poll_list);
11683 		sd->output_queue_tailp = &sd->output_queue;
11684 #ifdef CONFIG_RPS
11685 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11686 		sd->cpu = i;
11687 #endif
11688 
11689 		init_gro_hash(&sd->backlog);
11690 		sd->backlog.poll = process_backlog;
11691 		sd->backlog.weight = weight_p;
11692 	}
11693 
11694 	dev_boot_phase = 0;
11695 
11696 	/* The loopback device is special if any other network devices
11697 	 * is present in a network namespace the loopback device must
11698 	 * be present. Since we now dynamically allocate and free the
11699 	 * loopback device ensure this invariant is maintained by
11700 	 * keeping the loopback device as the first device on the
11701 	 * list of network devices.  Ensuring the loopback devices
11702 	 * is the first device that appears and the last network device
11703 	 * that disappears.
11704 	 */
11705 	if (register_pernet_device(&loopback_net_ops))
11706 		goto out;
11707 
11708 	if (register_pernet_device(&default_device_ops))
11709 		goto out;
11710 
11711 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11712 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11713 
11714 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11715 				       NULL, dev_cpu_dead);
11716 	WARN_ON(rc < 0);
11717 	rc = 0;
11718 out:
11719 	return rc;
11720 }
11721 
11722 subsys_initcall(net_dev_init);
11723