xref: /openbmc/linux/net/core/dev.c (revision 97da55fc)
1 /*
2  * 	NET3	Protocol independent device support routines.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  *	Derived from the non IP parts of dev.c 1.0.19
10  * 		Authors:	Ross Biro
11  *				Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *				Mark Evans, <evansmp@uhura.aston.ac.uk>
13  *
14  *	Additional Authors:
15  *		Florian la Roche <rzsfl@rz.uni-sb.de>
16  *		Alan Cox <gw4pts@gw4pts.ampr.org>
17  *		David Hinds <dahinds@users.sourceforge.net>
18  *		Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19  *		Adam Sulmicki <adam@cfar.umd.edu>
20  *              Pekka Riikonen <priikone@poesidon.pspt.fi>
21  *
22  *	Changes:
23  *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
24  *              			to 2 if register_netdev gets called
25  *              			before net_dev_init & also removed a
26  *              			few lines of code in the process.
27  *		Alan Cox	:	device private ioctl copies fields back.
28  *		Alan Cox	:	Transmit queue code does relevant
29  *					stunts to keep the queue safe.
30  *		Alan Cox	:	Fixed double lock.
31  *		Alan Cox	:	Fixed promisc NULL pointer trap
32  *		????????	:	Support the full private ioctl range
33  *		Alan Cox	:	Moved ioctl permission check into
34  *					drivers
35  *		Tim Kordas	:	SIOCADDMULTI/SIOCDELMULTI
36  *		Alan Cox	:	100 backlog just doesn't cut it when
37  *					you start doing multicast video 8)
38  *		Alan Cox	:	Rewrote net_bh and list manager.
39  *		Alan Cox	: 	Fix ETH_P_ALL echoback lengths.
40  *		Alan Cox	:	Took out transmit every packet pass
41  *					Saved a few bytes in the ioctl handler
42  *		Alan Cox	:	Network driver sets packet type before
43  *					calling netif_rx. Saves a function
44  *					call a packet.
45  *		Alan Cox	:	Hashed net_bh()
46  *		Richard Kooijman:	Timestamp fixes.
47  *		Alan Cox	:	Wrong field in SIOCGIFDSTADDR
48  *		Alan Cox	:	Device lock protection.
49  *		Alan Cox	: 	Fixed nasty side effect of device close
50  *					changes.
51  *		Rudi Cilibrasi	:	Pass the right thing to
52  *					set_mac_address()
53  *		Dave Miller	:	32bit quantity for the device lock to
54  *					make it work out on a Sparc.
55  *		Bjorn Ekwall	:	Added KERNELD hack.
56  *		Alan Cox	:	Cleaned up the backlog initialise.
57  *		Craig Metz	:	SIOCGIFCONF fix if space for under
58  *					1 device.
59  *	    Thomas Bogendoerfer :	Return ENODEV for dev_open, if there
60  *					is no device open function.
61  *		Andi Kleen	:	Fix error reporting for SIOCGIFCONF
62  *	    Michael Chastain	:	Fix signed/unsigned for SIOCGIFCONF
63  *		Cyrus Durgin	:	Cleaned for KMOD
64  *		Adam Sulmicki   :	Bug Fix : Network Device Unload
65  *					A network device unload needs to purge
66  *					the backlog queue.
67  *	Paul Rusty Russell	:	SIOCSIFNAME
68  *              Pekka Riikonen  :	Netdev boot-time settings code
69  *              Andrew Morton   :       Make unregister_netdevice wait
70  *              			indefinitely on dev->refcnt
71  * 		J Hadi Salim	:	- Backlog queue sampling
72  *				        - netif_rx() feedback
73  */
74 
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
98 #include <net/sock.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
101 #include <net/dst.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
120 #include <net/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 
133 #include "net-sysfs.h"
134 
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
137 
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
140 
141 static DEFINE_SPINLOCK(ptype_lock);
142 static DEFINE_SPINLOCK(offload_lock);
143 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
144 struct list_head ptype_all __read_mostly;	/* Taps */
145 static struct list_head offload_base __read_mostly;
146 
147 /*
148  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
149  * semaphore.
150  *
151  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
152  *
153  * Writers must hold the rtnl semaphore while they loop through the
154  * dev_base_head list, and hold dev_base_lock for writing when they do the
155  * actual updates.  This allows pure readers to access the list even
156  * while a writer is preparing to update it.
157  *
158  * To put it another way, dev_base_lock is held for writing only to
159  * protect against pure readers; the rtnl semaphore provides the
160  * protection against other writers.
161  *
162  * See, for example usages, register_netdevice() and
163  * unregister_netdevice(), which must be called with the rtnl
164  * semaphore held.
165  */
166 DEFINE_RWLOCK(dev_base_lock);
167 EXPORT_SYMBOL(dev_base_lock);
168 
169 seqcount_t devnet_rename_seq;
170 
171 static inline void dev_base_seq_inc(struct net *net)
172 {
173 	while (++net->dev_base_seq == 0);
174 }
175 
176 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
177 {
178 	unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
179 
180 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
181 }
182 
183 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
184 {
185 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
186 }
187 
188 static inline void rps_lock(struct softnet_data *sd)
189 {
190 #ifdef CONFIG_RPS
191 	spin_lock(&sd->input_pkt_queue.lock);
192 #endif
193 }
194 
195 static inline void rps_unlock(struct softnet_data *sd)
196 {
197 #ifdef CONFIG_RPS
198 	spin_unlock(&sd->input_pkt_queue.lock);
199 #endif
200 }
201 
202 /* Device list insertion */
203 static int list_netdevice(struct net_device *dev)
204 {
205 	struct net *net = dev_net(dev);
206 
207 	ASSERT_RTNL();
208 
209 	write_lock_bh(&dev_base_lock);
210 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
211 	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
212 	hlist_add_head_rcu(&dev->index_hlist,
213 			   dev_index_hash(net, dev->ifindex));
214 	write_unlock_bh(&dev_base_lock);
215 
216 	dev_base_seq_inc(net);
217 
218 	return 0;
219 }
220 
221 /* Device list removal
222  * caller must respect a RCU grace period before freeing/reusing dev
223  */
224 static void unlist_netdevice(struct net_device *dev)
225 {
226 	ASSERT_RTNL();
227 
228 	/* Unlink dev from the device chain */
229 	write_lock_bh(&dev_base_lock);
230 	list_del_rcu(&dev->dev_list);
231 	hlist_del_rcu(&dev->name_hlist);
232 	hlist_del_rcu(&dev->index_hlist);
233 	write_unlock_bh(&dev_base_lock);
234 
235 	dev_base_seq_inc(dev_net(dev));
236 }
237 
238 /*
239  *	Our notifier list
240  */
241 
242 static RAW_NOTIFIER_HEAD(netdev_chain);
243 
244 /*
245  *	Device drivers call our routines to queue packets here. We empty the
246  *	queue in the local softnet handler.
247  */
248 
249 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
250 EXPORT_PER_CPU_SYMBOL(softnet_data);
251 
252 #ifdef CONFIG_LOCKDEP
253 /*
254  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
255  * according to dev->type
256  */
257 static const unsigned short netdev_lock_type[] =
258 	{ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
259 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
260 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
261 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
262 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
263 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
264 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
265 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
266 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
267 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
268 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
269 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
270 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
271 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
272 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
273 
274 static const char *const netdev_lock_name[] =
275 	{"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
276 	 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
277 	 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
278 	 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
279 	 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
280 	 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
281 	 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
282 	 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
283 	 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
284 	 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
285 	 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
286 	 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
287 	 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
288 	 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
289 	 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
290 
291 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
292 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
293 
294 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
295 {
296 	int i;
297 
298 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
299 		if (netdev_lock_type[i] == dev_type)
300 			return i;
301 	/* the last key is used by default */
302 	return ARRAY_SIZE(netdev_lock_type) - 1;
303 }
304 
305 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
306 						 unsigned short dev_type)
307 {
308 	int i;
309 
310 	i = netdev_lock_pos(dev_type);
311 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
312 				   netdev_lock_name[i]);
313 }
314 
315 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
316 {
317 	int i;
318 
319 	i = netdev_lock_pos(dev->type);
320 	lockdep_set_class_and_name(&dev->addr_list_lock,
321 				   &netdev_addr_lock_key[i],
322 				   netdev_lock_name[i]);
323 }
324 #else
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 						 unsigned short dev_type)
327 {
328 }
329 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
330 {
331 }
332 #endif
333 
334 /*******************************************************************************
335 
336 		Protocol management and registration routines
337 
338 *******************************************************************************/
339 
340 /*
341  *	Add a protocol ID to the list. Now that the input handler is
342  *	smarter we can dispense with all the messy stuff that used to be
343  *	here.
344  *
345  *	BEWARE!!! Protocol handlers, mangling input packets,
346  *	MUST BE last in hash buckets and checking protocol handlers
347  *	MUST start from promiscuous ptype_all chain in net_bh.
348  *	It is true now, do not change it.
349  *	Explanation follows: if protocol handler, mangling packet, will
350  *	be the first on list, it is not able to sense, that packet
351  *	is cloned and should be copied-on-write, so that it will
352  *	change it and subsequent readers will get broken packet.
353  *							--ANK (980803)
354  */
355 
356 static inline struct list_head *ptype_head(const struct packet_type *pt)
357 {
358 	if (pt->type == htons(ETH_P_ALL))
359 		return &ptype_all;
360 	else
361 		return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
362 }
363 
364 /**
365  *	dev_add_pack - add packet handler
366  *	@pt: packet type declaration
367  *
368  *	Add a protocol handler to the networking stack. The passed &packet_type
369  *	is linked into kernel lists and may not be freed until it has been
370  *	removed from the kernel lists.
371  *
372  *	This call does not sleep therefore it can not
373  *	guarantee all CPU's that are in middle of receiving packets
374  *	will see the new packet type (until the next received packet).
375  */
376 
377 void dev_add_pack(struct packet_type *pt)
378 {
379 	struct list_head *head = ptype_head(pt);
380 
381 	spin_lock(&ptype_lock);
382 	list_add_rcu(&pt->list, head);
383 	spin_unlock(&ptype_lock);
384 }
385 EXPORT_SYMBOL(dev_add_pack);
386 
387 /**
388  *	__dev_remove_pack	 - remove packet handler
389  *	@pt: packet type declaration
390  *
391  *	Remove a protocol handler that was previously added to the kernel
392  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
393  *	from the kernel lists and can be freed or reused once this function
394  *	returns.
395  *
396  *      The packet type might still be in use by receivers
397  *	and must not be freed until after all the CPU's have gone
398  *	through a quiescent state.
399  */
400 void __dev_remove_pack(struct packet_type *pt)
401 {
402 	struct list_head *head = ptype_head(pt);
403 	struct packet_type *pt1;
404 
405 	spin_lock(&ptype_lock);
406 
407 	list_for_each_entry(pt1, head, list) {
408 		if (pt == pt1) {
409 			list_del_rcu(&pt->list);
410 			goto out;
411 		}
412 	}
413 
414 	pr_warn("dev_remove_pack: %p not found\n", pt);
415 out:
416 	spin_unlock(&ptype_lock);
417 }
418 EXPORT_SYMBOL(__dev_remove_pack);
419 
420 /**
421  *	dev_remove_pack	 - remove packet handler
422  *	@pt: packet type declaration
423  *
424  *	Remove a protocol handler that was previously added to the kernel
425  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
426  *	from the kernel lists and can be freed or reused once this function
427  *	returns.
428  *
429  *	This call sleeps to guarantee that no CPU is looking at the packet
430  *	type after return.
431  */
432 void dev_remove_pack(struct packet_type *pt)
433 {
434 	__dev_remove_pack(pt);
435 
436 	synchronize_net();
437 }
438 EXPORT_SYMBOL(dev_remove_pack);
439 
440 
441 /**
442  *	dev_add_offload - register offload handlers
443  *	@po: protocol offload declaration
444  *
445  *	Add protocol offload handlers to the networking stack. The passed
446  *	&proto_offload is linked into kernel lists and may not be freed until
447  *	it has been removed from the kernel lists.
448  *
449  *	This call does not sleep therefore it can not
450  *	guarantee all CPU's that are in middle of receiving packets
451  *	will see the new offload handlers (until the next received packet).
452  */
453 void dev_add_offload(struct packet_offload *po)
454 {
455 	struct list_head *head = &offload_base;
456 
457 	spin_lock(&offload_lock);
458 	list_add_rcu(&po->list, head);
459 	spin_unlock(&offload_lock);
460 }
461 EXPORT_SYMBOL(dev_add_offload);
462 
463 /**
464  *	__dev_remove_offload	 - remove offload handler
465  *	@po: packet offload declaration
466  *
467  *	Remove a protocol offload handler that was previously added to the
468  *	kernel offload handlers by dev_add_offload(). The passed &offload_type
469  *	is removed from the kernel lists and can be freed or reused once this
470  *	function returns.
471  *
472  *      The packet type might still be in use by receivers
473  *	and must not be freed until after all the CPU's have gone
474  *	through a quiescent state.
475  */
476 void __dev_remove_offload(struct packet_offload *po)
477 {
478 	struct list_head *head = &offload_base;
479 	struct packet_offload *po1;
480 
481 	spin_lock(&offload_lock);
482 
483 	list_for_each_entry(po1, head, list) {
484 		if (po == po1) {
485 			list_del_rcu(&po->list);
486 			goto out;
487 		}
488 	}
489 
490 	pr_warn("dev_remove_offload: %p not found\n", po);
491 out:
492 	spin_unlock(&offload_lock);
493 }
494 EXPORT_SYMBOL(__dev_remove_offload);
495 
496 /**
497  *	dev_remove_offload	 - remove packet offload handler
498  *	@po: packet offload declaration
499  *
500  *	Remove a packet offload handler that was previously added to the kernel
501  *	offload handlers by dev_add_offload(). The passed &offload_type is
502  *	removed from the kernel lists and can be freed or reused once this
503  *	function returns.
504  *
505  *	This call sleeps to guarantee that no CPU is looking at the packet
506  *	type after return.
507  */
508 void dev_remove_offload(struct packet_offload *po)
509 {
510 	__dev_remove_offload(po);
511 
512 	synchronize_net();
513 }
514 EXPORT_SYMBOL(dev_remove_offload);
515 
516 /******************************************************************************
517 
518 		      Device Boot-time Settings Routines
519 
520 *******************************************************************************/
521 
522 /* Boot time configuration table */
523 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
524 
525 /**
526  *	netdev_boot_setup_add	- add new setup entry
527  *	@name: name of the device
528  *	@map: configured settings for the device
529  *
530  *	Adds new setup entry to the dev_boot_setup list.  The function
531  *	returns 0 on error and 1 on success.  This is a generic routine to
532  *	all netdevices.
533  */
534 static int netdev_boot_setup_add(char *name, struct ifmap *map)
535 {
536 	struct netdev_boot_setup *s;
537 	int i;
538 
539 	s = dev_boot_setup;
540 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
541 		if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
542 			memset(s[i].name, 0, sizeof(s[i].name));
543 			strlcpy(s[i].name, name, IFNAMSIZ);
544 			memcpy(&s[i].map, map, sizeof(s[i].map));
545 			break;
546 		}
547 	}
548 
549 	return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
550 }
551 
552 /**
553  *	netdev_boot_setup_check	- check boot time settings
554  *	@dev: the netdevice
555  *
556  * 	Check boot time settings for the device.
557  *	The found settings are set for the device to be used
558  *	later in the device probing.
559  *	Returns 0 if no settings found, 1 if they are.
560  */
561 int netdev_boot_setup_check(struct net_device *dev)
562 {
563 	struct netdev_boot_setup *s = dev_boot_setup;
564 	int i;
565 
566 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
567 		if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
568 		    !strcmp(dev->name, s[i].name)) {
569 			dev->irq 	= s[i].map.irq;
570 			dev->base_addr 	= s[i].map.base_addr;
571 			dev->mem_start 	= s[i].map.mem_start;
572 			dev->mem_end 	= s[i].map.mem_end;
573 			return 1;
574 		}
575 	}
576 	return 0;
577 }
578 EXPORT_SYMBOL(netdev_boot_setup_check);
579 
580 
581 /**
582  *	netdev_boot_base	- get address from boot time settings
583  *	@prefix: prefix for network device
584  *	@unit: id for network device
585  *
586  * 	Check boot time settings for the base address of device.
587  *	The found settings are set for the device to be used
588  *	later in the device probing.
589  *	Returns 0 if no settings found.
590  */
591 unsigned long netdev_boot_base(const char *prefix, int unit)
592 {
593 	const struct netdev_boot_setup *s = dev_boot_setup;
594 	char name[IFNAMSIZ];
595 	int i;
596 
597 	sprintf(name, "%s%d", prefix, unit);
598 
599 	/*
600 	 * If device already registered then return base of 1
601 	 * to indicate not to probe for this interface
602 	 */
603 	if (__dev_get_by_name(&init_net, name))
604 		return 1;
605 
606 	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
607 		if (!strcmp(name, s[i].name))
608 			return s[i].map.base_addr;
609 	return 0;
610 }
611 
612 /*
613  * Saves at boot time configured settings for any netdevice.
614  */
615 int __init netdev_boot_setup(char *str)
616 {
617 	int ints[5];
618 	struct ifmap map;
619 
620 	str = get_options(str, ARRAY_SIZE(ints), ints);
621 	if (!str || !*str)
622 		return 0;
623 
624 	/* Save settings */
625 	memset(&map, 0, sizeof(map));
626 	if (ints[0] > 0)
627 		map.irq = ints[1];
628 	if (ints[0] > 1)
629 		map.base_addr = ints[2];
630 	if (ints[0] > 2)
631 		map.mem_start = ints[3];
632 	if (ints[0] > 3)
633 		map.mem_end = ints[4];
634 
635 	/* Add new entry to the list */
636 	return netdev_boot_setup_add(str, &map);
637 }
638 
639 __setup("netdev=", netdev_boot_setup);
640 
641 /*******************************************************************************
642 
643 			    Device Interface Subroutines
644 
645 *******************************************************************************/
646 
647 /**
648  *	__dev_get_by_name	- find a device by its name
649  *	@net: the applicable net namespace
650  *	@name: name to find
651  *
652  *	Find an interface by name. Must be called under RTNL semaphore
653  *	or @dev_base_lock. If the name is found a pointer to the device
654  *	is returned. If the name is not found then %NULL is returned. The
655  *	reference counters are not incremented so the caller must be
656  *	careful with locks.
657  */
658 
659 struct net_device *__dev_get_by_name(struct net *net, const char *name)
660 {
661 	struct net_device *dev;
662 	struct hlist_head *head = dev_name_hash(net, name);
663 
664 	hlist_for_each_entry(dev, head, name_hlist)
665 		if (!strncmp(dev->name, name, IFNAMSIZ))
666 			return dev;
667 
668 	return NULL;
669 }
670 EXPORT_SYMBOL(__dev_get_by_name);
671 
672 /**
673  *	dev_get_by_name_rcu	- find a device by its name
674  *	@net: the applicable net namespace
675  *	@name: name to find
676  *
677  *	Find an interface by name.
678  *	If the name is found a pointer to the device is returned.
679  * 	If the name is not found then %NULL is returned.
680  *	The reference counters are not incremented so the caller must be
681  *	careful with locks. The caller must hold RCU lock.
682  */
683 
684 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
685 {
686 	struct net_device *dev;
687 	struct hlist_head *head = dev_name_hash(net, name);
688 
689 	hlist_for_each_entry_rcu(dev, head, name_hlist)
690 		if (!strncmp(dev->name, name, IFNAMSIZ))
691 			return dev;
692 
693 	return NULL;
694 }
695 EXPORT_SYMBOL(dev_get_by_name_rcu);
696 
697 /**
698  *	dev_get_by_name		- find a device by its name
699  *	@net: the applicable net namespace
700  *	@name: name to find
701  *
702  *	Find an interface by name. This can be called from any
703  *	context and does its own locking. The returned handle has
704  *	the usage count incremented and the caller must use dev_put() to
705  *	release it when it is no longer needed. %NULL is returned if no
706  *	matching device is found.
707  */
708 
709 struct net_device *dev_get_by_name(struct net *net, const char *name)
710 {
711 	struct net_device *dev;
712 
713 	rcu_read_lock();
714 	dev = dev_get_by_name_rcu(net, name);
715 	if (dev)
716 		dev_hold(dev);
717 	rcu_read_unlock();
718 	return dev;
719 }
720 EXPORT_SYMBOL(dev_get_by_name);
721 
722 /**
723  *	__dev_get_by_index - find a device by its ifindex
724  *	@net: the applicable net namespace
725  *	@ifindex: index of device
726  *
727  *	Search for an interface by index. Returns %NULL if the device
728  *	is not found or a pointer to the device. The device has not
729  *	had its reference counter increased so the caller must be careful
730  *	about locking. The caller must hold either the RTNL semaphore
731  *	or @dev_base_lock.
732  */
733 
734 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
735 {
736 	struct net_device *dev;
737 	struct hlist_head *head = dev_index_hash(net, ifindex);
738 
739 	hlist_for_each_entry(dev, head, index_hlist)
740 		if (dev->ifindex == ifindex)
741 			return dev;
742 
743 	return NULL;
744 }
745 EXPORT_SYMBOL(__dev_get_by_index);
746 
747 /**
748  *	dev_get_by_index_rcu - find a device by its ifindex
749  *	@net: the applicable net namespace
750  *	@ifindex: index of device
751  *
752  *	Search for an interface by index. Returns %NULL if the device
753  *	is not found or a pointer to the device. The device has not
754  *	had its reference counter increased so the caller must be careful
755  *	about locking. The caller must hold RCU lock.
756  */
757 
758 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
759 {
760 	struct net_device *dev;
761 	struct hlist_head *head = dev_index_hash(net, ifindex);
762 
763 	hlist_for_each_entry_rcu(dev, head, index_hlist)
764 		if (dev->ifindex == ifindex)
765 			return dev;
766 
767 	return NULL;
768 }
769 EXPORT_SYMBOL(dev_get_by_index_rcu);
770 
771 
772 /**
773  *	dev_get_by_index - find a device by its ifindex
774  *	@net: the applicable net namespace
775  *	@ifindex: index of device
776  *
777  *	Search for an interface by index. Returns NULL if the device
778  *	is not found or a pointer to the device. The device returned has
779  *	had a reference added and the pointer is safe until the user calls
780  *	dev_put to indicate they have finished with it.
781  */
782 
783 struct net_device *dev_get_by_index(struct net *net, int ifindex)
784 {
785 	struct net_device *dev;
786 
787 	rcu_read_lock();
788 	dev = dev_get_by_index_rcu(net, ifindex);
789 	if (dev)
790 		dev_hold(dev);
791 	rcu_read_unlock();
792 	return dev;
793 }
794 EXPORT_SYMBOL(dev_get_by_index);
795 
796 /**
797  *	dev_getbyhwaddr_rcu - find a device by its hardware address
798  *	@net: the applicable net namespace
799  *	@type: media type of device
800  *	@ha: hardware address
801  *
802  *	Search for an interface by MAC address. Returns NULL if the device
803  *	is not found or a pointer to the device.
804  *	The caller must hold RCU or RTNL.
805  *	The returned device has not had its ref count increased
806  *	and the caller must therefore be careful about locking
807  *
808  */
809 
810 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
811 				       const char *ha)
812 {
813 	struct net_device *dev;
814 
815 	for_each_netdev_rcu(net, dev)
816 		if (dev->type == type &&
817 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
818 			return dev;
819 
820 	return NULL;
821 }
822 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
823 
824 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
825 {
826 	struct net_device *dev;
827 
828 	ASSERT_RTNL();
829 	for_each_netdev(net, dev)
830 		if (dev->type == type)
831 			return dev;
832 
833 	return NULL;
834 }
835 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
836 
837 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
838 {
839 	struct net_device *dev, *ret = NULL;
840 
841 	rcu_read_lock();
842 	for_each_netdev_rcu(net, dev)
843 		if (dev->type == type) {
844 			dev_hold(dev);
845 			ret = dev;
846 			break;
847 		}
848 	rcu_read_unlock();
849 	return ret;
850 }
851 EXPORT_SYMBOL(dev_getfirstbyhwtype);
852 
853 /**
854  *	dev_get_by_flags_rcu - find any device with given flags
855  *	@net: the applicable net namespace
856  *	@if_flags: IFF_* values
857  *	@mask: bitmask of bits in if_flags to check
858  *
859  *	Search for any interface with the given flags. Returns NULL if a device
860  *	is not found or a pointer to the device. Must be called inside
861  *	rcu_read_lock(), and result refcount is unchanged.
862  */
863 
864 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
865 				    unsigned short mask)
866 {
867 	struct net_device *dev, *ret;
868 
869 	ret = NULL;
870 	for_each_netdev_rcu(net, dev) {
871 		if (((dev->flags ^ if_flags) & mask) == 0) {
872 			ret = dev;
873 			break;
874 		}
875 	}
876 	return ret;
877 }
878 EXPORT_SYMBOL(dev_get_by_flags_rcu);
879 
880 /**
881  *	dev_valid_name - check if name is okay for network device
882  *	@name: name string
883  *
884  *	Network device names need to be valid file names to
885  *	to allow sysfs to work.  We also disallow any kind of
886  *	whitespace.
887  */
888 bool dev_valid_name(const char *name)
889 {
890 	if (*name == '\0')
891 		return false;
892 	if (strlen(name) >= IFNAMSIZ)
893 		return false;
894 	if (!strcmp(name, ".") || !strcmp(name, ".."))
895 		return false;
896 
897 	while (*name) {
898 		if (*name == '/' || isspace(*name))
899 			return false;
900 		name++;
901 	}
902 	return true;
903 }
904 EXPORT_SYMBOL(dev_valid_name);
905 
906 /**
907  *	__dev_alloc_name - allocate a name for a device
908  *	@net: network namespace to allocate the device name in
909  *	@name: name format string
910  *	@buf:  scratch buffer and result name string
911  *
912  *	Passed a format string - eg "lt%d" it will try and find a suitable
913  *	id. It scans list of devices to build up a free map, then chooses
914  *	the first empty slot. The caller must hold the dev_base or rtnl lock
915  *	while allocating the name and adding the device in order to avoid
916  *	duplicates.
917  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
918  *	Returns the number of the unit assigned or a negative errno code.
919  */
920 
921 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
922 {
923 	int i = 0;
924 	const char *p;
925 	const int max_netdevices = 8*PAGE_SIZE;
926 	unsigned long *inuse;
927 	struct net_device *d;
928 
929 	p = strnchr(name, IFNAMSIZ-1, '%');
930 	if (p) {
931 		/*
932 		 * Verify the string as this thing may have come from
933 		 * the user.  There must be either one "%d" and no other "%"
934 		 * characters.
935 		 */
936 		if (p[1] != 'd' || strchr(p + 2, '%'))
937 			return -EINVAL;
938 
939 		/* Use one page as a bit array of possible slots */
940 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
941 		if (!inuse)
942 			return -ENOMEM;
943 
944 		for_each_netdev(net, d) {
945 			if (!sscanf(d->name, name, &i))
946 				continue;
947 			if (i < 0 || i >= max_netdevices)
948 				continue;
949 
950 			/*  avoid cases where sscanf is not exact inverse of printf */
951 			snprintf(buf, IFNAMSIZ, name, i);
952 			if (!strncmp(buf, d->name, IFNAMSIZ))
953 				set_bit(i, inuse);
954 		}
955 
956 		i = find_first_zero_bit(inuse, max_netdevices);
957 		free_page((unsigned long) inuse);
958 	}
959 
960 	if (buf != name)
961 		snprintf(buf, IFNAMSIZ, name, i);
962 	if (!__dev_get_by_name(net, buf))
963 		return i;
964 
965 	/* It is possible to run out of possible slots
966 	 * when the name is long and there isn't enough space left
967 	 * for the digits, or if all bits are used.
968 	 */
969 	return -ENFILE;
970 }
971 
972 /**
973  *	dev_alloc_name - allocate a name for a device
974  *	@dev: device
975  *	@name: name format string
976  *
977  *	Passed a format string - eg "lt%d" it will try and find a suitable
978  *	id. It scans list of devices to build up a free map, then chooses
979  *	the first empty slot. The caller must hold the dev_base or rtnl lock
980  *	while allocating the name and adding the device in order to avoid
981  *	duplicates.
982  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
983  *	Returns the number of the unit assigned or a negative errno code.
984  */
985 
986 int dev_alloc_name(struct net_device *dev, const char *name)
987 {
988 	char buf[IFNAMSIZ];
989 	struct net *net;
990 	int ret;
991 
992 	BUG_ON(!dev_net(dev));
993 	net = dev_net(dev);
994 	ret = __dev_alloc_name(net, name, buf);
995 	if (ret >= 0)
996 		strlcpy(dev->name, buf, IFNAMSIZ);
997 	return ret;
998 }
999 EXPORT_SYMBOL(dev_alloc_name);
1000 
1001 static int dev_alloc_name_ns(struct net *net,
1002 			     struct net_device *dev,
1003 			     const char *name)
1004 {
1005 	char buf[IFNAMSIZ];
1006 	int ret;
1007 
1008 	ret = __dev_alloc_name(net, name, buf);
1009 	if (ret >= 0)
1010 		strlcpy(dev->name, buf, IFNAMSIZ);
1011 	return ret;
1012 }
1013 
1014 static int dev_get_valid_name(struct net *net,
1015 			      struct net_device *dev,
1016 			      const char *name)
1017 {
1018 	BUG_ON(!net);
1019 
1020 	if (!dev_valid_name(name))
1021 		return -EINVAL;
1022 
1023 	if (strchr(name, '%'))
1024 		return dev_alloc_name_ns(net, dev, name);
1025 	else if (__dev_get_by_name(net, name))
1026 		return -EEXIST;
1027 	else if (dev->name != name)
1028 		strlcpy(dev->name, name, IFNAMSIZ);
1029 
1030 	return 0;
1031 }
1032 
1033 /**
1034  *	dev_change_name - change name of a device
1035  *	@dev: device
1036  *	@newname: name (or format string) must be at least IFNAMSIZ
1037  *
1038  *	Change name of a device, can pass format strings "eth%d".
1039  *	for wildcarding.
1040  */
1041 int dev_change_name(struct net_device *dev, const char *newname)
1042 {
1043 	char oldname[IFNAMSIZ];
1044 	int err = 0;
1045 	int ret;
1046 	struct net *net;
1047 
1048 	ASSERT_RTNL();
1049 	BUG_ON(!dev_net(dev));
1050 
1051 	net = dev_net(dev);
1052 	if (dev->flags & IFF_UP)
1053 		return -EBUSY;
1054 
1055 	write_seqcount_begin(&devnet_rename_seq);
1056 
1057 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1058 		write_seqcount_end(&devnet_rename_seq);
1059 		return 0;
1060 	}
1061 
1062 	memcpy(oldname, dev->name, IFNAMSIZ);
1063 
1064 	err = dev_get_valid_name(net, dev, newname);
1065 	if (err < 0) {
1066 		write_seqcount_end(&devnet_rename_seq);
1067 		return err;
1068 	}
1069 
1070 rollback:
1071 	ret = device_rename(&dev->dev, dev->name);
1072 	if (ret) {
1073 		memcpy(dev->name, oldname, IFNAMSIZ);
1074 		write_seqcount_end(&devnet_rename_seq);
1075 		return ret;
1076 	}
1077 
1078 	write_seqcount_end(&devnet_rename_seq);
1079 
1080 	write_lock_bh(&dev_base_lock);
1081 	hlist_del_rcu(&dev->name_hlist);
1082 	write_unlock_bh(&dev_base_lock);
1083 
1084 	synchronize_rcu();
1085 
1086 	write_lock_bh(&dev_base_lock);
1087 	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1088 	write_unlock_bh(&dev_base_lock);
1089 
1090 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1091 	ret = notifier_to_errno(ret);
1092 
1093 	if (ret) {
1094 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1095 		if (err >= 0) {
1096 			err = ret;
1097 			write_seqcount_begin(&devnet_rename_seq);
1098 			memcpy(dev->name, oldname, IFNAMSIZ);
1099 			goto rollback;
1100 		} else {
1101 			pr_err("%s: name change rollback failed: %d\n",
1102 			       dev->name, ret);
1103 		}
1104 	}
1105 
1106 	return err;
1107 }
1108 
1109 /**
1110  *	dev_set_alias - change ifalias of a device
1111  *	@dev: device
1112  *	@alias: name up to IFALIASZ
1113  *	@len: limit of bytes to copy from info
1114  *
1115  *	Set ifalias for a device,
1116  */
1117 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1118 {
1119 	char *new_ifalias;
1120 
1121 	ASSERT_RTNL();
1122 
1123 	if (len >= IFALIASZ)
1124 		return -EINVAL;
1125 
1126 	if (!len) {
1127 		kfree(dev->ifalias);
1128 		dev->ifalias = NULL;
1129 		return 0;
1130 	}
1131 
1132 	new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1133 	if (!new_ifalias)
1134 		return -ENOMEM;
1135 	dev->ifalias = new_ifalias;
1136 
1137 	strlcpy(dev->ifalias, alias, len+1);
1138 	return len;
1139 }
1140 
1141 
1142 /**
1143  *	netdev_features_change - device changes features
1144  *	@dev: device to cause notification
1145  *
1146  *	Called to indicate a device has changed features.
1147  */
1148 void netdev_features_change(struct net_device *dev)
1149 {
1150 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1151 }
1152 EXPORT_SYMBOL(netdev_features_change);
1153 
1154 /**
1155  *	netdev_state_change - device changes state
1156  *	@dev: device to cause notification
1157  *
1158  *	Called to indicate a device has changed state. This function calls
1159  *	the notifier chains for netdev_chain and sends a NEWLINK message
1160  *	to the routing socket.
1161  */
1162 void netdev_state_change(struct net_device *dev)
1163 {
1164 	if (dev->flags & IFF_UP) {
1165 		call_netdevice_notifiers(NETDEV_CHANGE, dev);
1166 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1167 	}
1168 }
1169 EXPORT_SYMBOL(netdev_state_change);
1170 
1171 /**
1172  * 	netdev_notify_peers - notify network peers about existence of @dev
1173  * 	@dev: network device
1174  *
1175  * Generate traffic such that interested network peers are aware of
1176  * @dev, such as by generating a gratuitous ARP. This may be used when
1177  * a device wants to inform the rest of the network about some sort of
1178  * reconfiguration such as a failover event or virtual machine
1179  * migration.
1180  */
1181 void netdev_notify_peers(struct net_device *dev)
1182 {
1183 	rtnl_lock();
1184 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1185 	rtnl_unlock();
1186 }
1187 EXPORT_SYMBOL(netdev_notify_peers);
1188 
1189 static int __dev_open(struct net_device *dev)
1190 {
1191 	const struct net_device_ops *ops = dev->netdev_ops;
1192 	int ret;
1193 
1194 	ASSERT_RTNL();
1195 
1196 	if (!netif_device_present(dev))
1197 		return -ENODEV;
1198 
1199 	/* Block netpoll from trying to do any rx path servicing.
1200 	 * If we don't do this there is a chance ndo_poll_controller
1201 	 * or ndo_poll may be running while we open the device
1202 	 */
1203 	ret = netpoll_rx_disable(dev);
1204 	if (ret)
1205 		return ret;
1206 
1207 	ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1208 	ret = notifier_to_errno(ret);
1209 	if (ret)
1210 		return ret;
1211 
1212 	set_bit(__LINK_STATE_START, &dev->state);
1213 
1214 	if (ops->ndo_validate_addr)
1215 		ret = ops->ndo_validate_addr(dev);
1216 
1217 	if (!ret && ops->ndo_open)
1218 		ret = ops->ndo_open(dev);
1219 
1220 	netpoll_rx_enable(dev);
1221 
1222 	if (ret)
1223 		clear_bit(__LINK_STATE_START, &dev->state);
1224 	else {
1225 		dev->flags |= IFF_UP;
1226 		net_dmaengine_get();
1227 		dev_set_rx_mode(dev);
1228 		dev_activate(dev);
1229 		add_device_randomness(dev->dev_addr, dev->addr_len);
1230 	}
1231 
1232 	return ret;
1233 }
1234 
1235 /**
1236  *	dev_open	- prepare an interface for use.
1237  *	@dev:	device to open
1238  *
1239  *	Takes a device from down to up state. The device's private open
1240  *	function is invoked and then the multicast lists are loaded. Finally
1241  *	the device is moved into the up state and a %NETDEV_UP message is
1242  *	sent to the netdev notifier chain.
1243  *
1244  *	Calling this function on an active interface is a nop. On a failure
1245  *	a negative errno code is returned.
1246  */
1247 int dev_open(struct net_device *dev)
1248 {
1249 	int ret;
1250 
1251 	if (dev->flags & IFF_UP)
1252 		return 0;
1253 
1254 	ret = __dev_open(dev);
1255 	if (ret < 0)
1256 		return ret;
1257 
1258 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1259 	call_netdevice_notifiers(NETDEV_UP, dev);
1260 
1261 	return ret;
1262 }
1263 EXPORT_SYMBOL(dev_open);
1264 
1265 static int __dev_close_many(struct list_head *head)
1266 {
1267 	struct net_device *dev;
1268 
1269 	ASSERT_RTNL();
1270 	might_sleep();
1271 
1272 	list_for_each_entry(dev, head, unreg_list) {
1273 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1274 
1275 		clear_bit(__LINK_STATE_START, &dev->state);
1276 
1277 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1278 		 * can be even on different cpu. So just clear netif_running().
1279 		 *
1280 		 * dev->stop() will invoke napi_disable() on all of it's
1281 		 * napi_struct instances on this device.
1282 		 */
1283 		smp_mb__after_clear_bit(); /* Commit netif_running(). */
1284 	}
1285 
1286 	dev_deactivate_many(head);
1287 
1288 	list_for_each_entry(dev, head, unreg_list) {
1289 		const struct net_device_ops *ops = dev->netdev_ops;
1290 
1291 		/*
1292 		 *	Call the device specific close. This cannot fail.
1293 		 *	Only if device is UP
1294 		 *
1295 		 *	We allow it to be called even after a DETACH hot-plug
1296 		 *	event.
1297 		 */
1298 		if (ops->ndo_stop)
1299 			ops->ndo_stop(dev);
1300 
1301 		dev->flags &= ~IFF_UP;
1302 		net_dmaengine_put();
1303 	}
1304 
1305 	return 0;
1306 }
1307 
1308 static int __dev_close(struct net_device *dev)
1309 {
1310 	int retval;
1311 	LIST_HEAD(single);
1312 
1313 	/* Temporarily disable netpoll until the interface is down */
1314 	retval = netpoll_rx_disable(dev);
1315 	if (retval)
1316 		return retval;
1317 
1318 	list_add(&dev->unreg_list, &single);
1319 	retval = __dev_close_many(&single);
1320 	list_del(&single);
1321 
1322 	netpoll_rx_enable(dev);
1323 	return retval;
1324 }
1325 
1326 static int dev_close_many(struct list_head *head)
1327 {
1328 	struct net_device *dev, *tmp;
1329 	LIST_HEAD(tmp_list);
1330 
1331 	list_for_each_entry_safe(dev, tmp, head, unreg_list)
1332 		if (!(dev->flags & IFF_UP))
1333 			list_move(&dev->unreg_list, &tmp_list);
1334 
1335 	__dev_close_many(head);
1336 
1337 	list_for_each_entry(dev, head, unreg_list) {
1338 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1339 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1340 	}
1341 
1342 	/* rollback_registered_many needs the complete original list */
1343 	list_splice(&tmp_list, head);
1344 	return 0;
1345 }
1346 
1347 /**
1348  *	dev_close - shutdown an interface.
1349  *	@dev: device to shutdown
1350  *
1351  *	This function moves an active device into down state. A
1352  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1353  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1354  *	chain.
1355  */
1356 int dev_close(struct net_device *dev)
1357 {
1358 	int ret = 0;
1359 	if (dev->flags & IFF_UP) {
1360 		LIST_HEAD(single);
1361 
1362 		/* Block netpoll rx while the interface is going down */
1363 		ret = netpoll_rx_disable(dev);
1364 		if (ret)
1365 			return ret;
1366 
1367 		list_add(&dev->unreg_list, &single);
1368 		dev_close_many(&single);
1369 		list_del(&single);
1370 
1371 		netpoll_rx_enable(dev);
1372 	}
1373 	return ret;
1374 }
1375 EXPORT_SYMBOL(dev_close);
1376 
1377 
1378 /**
1379  *	dev_disable_lro - disable Large Receive Offload on a device
1380  *	@dev: device
1381  *
1382  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1383  *	called under RTNL.  This is needed if received packets may be
1384  *	forwarded to another interface.
1385  */
1386 void dev_disable_lro(struct net_device *dev)
1387 {
1388 	/*
1389 	 * If we're trying to disable lro on a vlan device
1390 	 * use the underlying physical device instead
1391 	 */
1392 	if (is_vlan_dev(dev))
1393 		dev = vlan_dev_real_dev(dev);
1394 
1395 	dev->wanted_features &= ~NETIF_F_LRO;
1396 	netdev_update_features(dev);
1397 
1398 	if (unlikely(dev->features & NETIF_F_LRO))
1399 		netdev_WARN(dev, "failed to disable LRO!\n");
1400 }
1401 EXPORT_SYMBOL(dev_disable_lro);
1402 
1403 
1404 static int dev_boot_phase = 1;
1405 
1406 /**
1407  *	register_netdevice_notifier - register a network notifier block
1408  *	@nb: notifier
1409  *
1410  *	Register a notifier to be called when network device events occur.
1411  *	The notifier passed is linked into the kernel structures and must
1412  *	not be reused until it has been unregistered. A negative errno code
1413  *	is returned on a failure.
1414  *
1415  * 	When registered all registration and up events are replayed
1416  *	to the new notifier to allow device to have a race free
1417  *	view of the network device list.
1418  */
1419 
1420 int register_netdevice_notifier(struct notifier_block *nb)
1421 {
1422 	struct net_device *dev;
1423 	struct net_device *last;
1424 	struct net *net;
1425 	int err;
1426 
1427 	rtnl_lock();
1428 	err = raw_notifier_chain_register(&netdev_chain, nb);
1429 	if (err)
1430 		goto unlock;
1431 	if (dev_boot_phase)
1432 		goto unlock;
1433 	for_each_net(net) {
1434 		for_each_netdev(net, dev) {
1435 			err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1436 			err = notifier_to_errno(err);
1437 			if (err)
1438 				goto rollback;
1439 
1440 			if (!(dev->flags & IFF_UP))
1441 				continue;
1442 
1443 			nb->notifier_call(nb, NETDEV_UP, dev);
1444 		}
1445 	}
1446 
1447 unlock:
1448 	rtnl_unlock();
1449 	return err;
1450 
1451 rollback:
1452 	last = dev;
1453 	for_each_net(net) {
1454 		for_each_netdev(net, dev) {
1455 			if (dev == last)
1456 				goto outroll;
1457 
1458 			if (dev->flags & IFF_UP) {
1459 				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1460 				nb->notifier_call(nb, NETDEV_DOWN, dev);
1461 			}
1462 			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1463 		}
1464 	}
1465 
1466 outroll:
1467 	raw_notifier_chain_unregister(&netdev_chain, nb);
1468 	goto unlock;
1469 }
1470 EXPORT_SYMBOL(register_netdevice_notifier);
1471 
1472 /**
1473  *	unregister_netdevice_notifier - unregister a network notifier block
1474  *	@nb: notifier
1475  *
1476  *	Unregister a notifier previously registered by
1477  *	register_netdevice_notifier(). The notifier is unlinked into the
1478  *	kernel structures and may then be reused. A negative errno code
1479  *	is returned on a failure.
1480  *
1481  * 	After unregistering unregister and down device events are synthesized
1482  *	for all devices on the device list to the removed notifier to remove
1483  *	the need for special case cleanup code.
1484  */
1485 
1486 int unregister_netdevice_notifier(struct notifier_block *nb)
1487 {
1488 	struct net_device *dev;
1489 	struct net *net;
1490 	int err;
1491 
1492 	rtnl_lock();
1493 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1494 	if (err)
1495 		goto unlock;
1496 
1497 	for_each_net(net) {
1498 		for_each_netdev(net, dev) {
1499 			if (dev->flags & IFF_UP) {
1500 				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1501 				nb->notifier_call(nb, NETDEV_DOWN, dev);
1502 			}
1503 			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1504 		}
1505 	}
1506 unlock:
1507 	rtnl_unlock();
1508 	return err;
1509 }
1510 EXPORT_SYMBOL(unregister_netdevice_notifier);
1511 
1512 /**
1513  *	call_netdevice_notifiers - call all network notifier blocks
1514  *      @val: value passed unmodified to notifier function
1515  *      @dev: net_device pointer passed unmodified to notifier function
1516  *
1517  *	Call all network notifier blocks.  Parameters and return value
1518  *	are as for raw_notifier_call_chain().
1519  */
1520 
1521 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1522 {
1523 	ASSERT_RTNL();
1524 	return raw_notifier_call_chain(&netdev_chain, val, dev);
1525 }
1526 EXPORT_SYMBOL(call_netdevice_notifiers);
1527 
1528 static struct static_key netstamp_needed __read_mostly;
1529 #ifdef HAVE_JUMP_LABEL
1530 /* We are not allowed to call static_key_slow_dec() from irq context
1531  * If net_disable_timestamp() is called from irq context, defer the
1532  * static_key_slow_dec() calls.
1533  */
1534 static atomic_t netstamp_needed_deferred;
1535 #endif
1536 
1537 void net_enable_timestamp(void)
1538 {
1539 #ifdef HAVE_JUMP_LABEL
1540 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1541 
1542 	if (deferred) {
1543 		while (--deferred)
1544 			static_key_slow_dec(&netstamp_needed);
1545 		return;
1546 	}
1547 #endif
1548 	WARN_ON(in_interrupt());
1549 	static_key_slow_inc(&netstamp_needed);
1550 }
1551 EXPORT_SYMBOL(net_enable_timestamp);
1552 
1553 void net_disable_timestamp(void)
1554 {
1555 #ifdef HAVE_JUMP_LABEL
1556 	if (in_interrupt()) {
1557 		atomic_inc(&netstamp_needed_deferred);
1558 		return;
1559 	}
1560 #endif
1561 	static_key_slow_dec(&netstamp_needed);
1562 }
1563 EXPORT_SYMBOL(net_disable_timestamp);
1564 
1565 static inline void net_timestamp_set(struct sk_buff *skb)
1566 {
1567 	skb->tstamp.tv64 = 0;
1568 	if (static_key_false(&netstamp_needed))
1569 		__net_timestamp(skb);
1570 }
1571 
1572 #define net_timestamp_check(COND, SKB)			\
1573 	if (static_key_false(&netstamp_needed)) {		\
1574 		if ((COND) && !(SKB)->tstamp.tv64)	\
1575 			__net_timestamp(SKB);		\
1576 	}						\
1577 
1578 static inline bool is_skb_forwardable(struct net_device *dev,
1579 				      struct sk_buff *skb)
1580 {
1581 	unsigned int len;
1582 
1583 	if (!(dev->flags & IFF_UP))
1584 		return false;
1585 
1586 	len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1587 	if (skb->len <= len)
1588 		return true;
1589 
1590 	/* if TSO is enabled, we don't care about the length as the packet
1591 	 * could be forwarded without being segmented before
1592 	 */
1593 	if (skb_is_gso(skb))
1594 		return true;
1595 
1596 	return false;
1597 }
1598 
1599 /**
1600  * dev_forward_skb - loopback an skb to another netif
1601  *
1602  * @dev: destination network device
1603  * @skb: buffer to forward
1604  *
1605  * return values:
1606  *	NET_RX_SUCCESS	(no congestion)
1607  *	NET_RX_DROP     (packet was dropped, but freed)
1608  *
1609  * dev_forward_skb can be used for injecting an skb from the
1610  * start_xmit function of one device into the receive queue
1611  * of another device.
1612  *
1613  * The receiving device may be in another namespace, so
1614  * we have to clear all information in the skb that could
1615  * impact namespace isolation.
1616  */
1617 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1618 {
1619 	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1620 		if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1621 			atomic_long_inc(&dev->rx_dropped);
1622 			kfree_skb(skb);
1623 			return NET_RX_DROP;
1624 		}
1625 	}
1626 
1627 	skb_orphan(skb);
1628 	nf_reset(skb);
1629 
1630 	if (unlikely(!is_skb_forwardable(dev, skb))) {
1631 		atomic_long_inc(&dev->rx_dropped);
1632 		kfree_skb(skb);
1633 		return NET_RX_DROP;
1634 	}
1635 	skb->skb_iif = 0;
1636 	skb->dev = dev;
1637 	skb_dst_drop(skb);
1638 	skb->tstamp.tv64 = 0;
1639 	skb->pkt_type = PACKET_HOST;
1640 	skb->protocol = eth_type_trans(skb, dev);
1641 	skb->mark = 0;
1642 	secpath_reset(skb);
1643 	nf_reset(skb);
1644 	return netif_rx(skb);
1645 }
1646 EXPORT_SYMBOL_GPL(dev_forward_skb);
1647 
1648 static inline int deliver_skb(struct sk_buff *skb,
1649 			      struct packet_type *pt_prev,
1650 			      struct net_device *orig_dev)
1651 {
1652 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1653 		return -ENOMEM;
1654 	atomic_inc(&skb->users);
1655 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1656 }
1657 
1658 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1659 {
1660 	if (!ptype->af_packet_priv || !skb->sk)
1661 		return false;
1662 
1663 	if (ptype->id_match)
1664 		return ptype->id_match(ptype, skb->sk);
1665 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1666 		return true;
1667 
1668 	return false;
1669 }
1670 
1671 /*
1672  *	Support routine. Sends outgoing frames to any network
1673  *	taps currently in use.
1674  */
1675 
1676 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1677 {
1678 	struct packet_type *ptype;
1679 	struct sk_buff *skb2 = NULL;
1680 	struct packet_type *pt_prev = NULL;
1681 
1682 	rcu_read_lock();
1683 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
1684 		/* Never send packets back to the socket
1685 		 * they originated from - MvS (miquels@drinkel.ow.org)
1686 		 */
1687 		if ((ptype->dev == dev || !ptype->dev) &&
1688 		    (!skb_loop_sk(ptype, skb))) {
1689 			if (pt_prev) {
1690 				deliver_skb(skb2, pt_prev, skb->dev);
1691 				pt_prev = ptype;
1692 				continue;
1693 			}
1694 
1695 			skb2 = skb_clone(skb, GFP_ATOMIC);
1696 			if (!skb2)
1697 				break;
1698 
1699 			net_timestamp_set(skb2);
1700 
1701 			/* skb->nh should be correctly
1702 			   set by sender, so that the second statement is
1703 			   just protection against buggy protocols.
1704 			 */
1705 			skb_reset_mac_header(skb2);
1706 
1707 			if (skb_network_header(skb2) < skb2->data ||
1708 			    skb2->network_header > skb2->tail) {
1709 				net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1710 						     ntohs(skb2->protocol),
1711 						     dev->name);
1712 				skb_reset_network_header(skb2);
1713 			}
1714 
1715 			skb2->transport_header = skb2->network_header;
1716 			skb2->pkt_type = PACKET_OUTGOING;
1717 			pt_prev = ptype;
1718 		}
1719 	}
1720 	if (pt_prev)
1721 		pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1722 	rcu_read_unlock();
1723 }
1724 
1725 /**
1726  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1727  * @dev: Network device
1728  * @txq: number of queues available
1729  *
1730  * If real_num_tx_queues is changed the tc mappings may no longer be
1731  * valid. To resolve this verify the tc mapping remains valid and if
1732  * not NULL the mapping. With no priorities mapping to this
1733  * offset/count pair it will no longer be used. In the worst case TC0
1734  * is invalid nothing can be done so disable priority mappings. If is
1735  * expected that drivers will fix this mapping if they can before
1736  * calling netif_set_real_num_tx_queues.
1737  */
1738 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1739 {
1740 	int i;
1741 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1742 
1743 	/* If TC0 is invalidated disable TC mapping */
1744 	if (tc->offset + tc->count > txq) {
1745 		pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1746 		dev->num_tc = 0;
1747 		return;
1748 	}
1749 
1750 	/* Invalidated prio to tc mappings set to TC0 */
1751 	for (i = 1; i < TC_BITMASK + 1; i++) {
1752 		int q = netdev_get_prio_tc_map(dev, i);
1753 
1754 		tc = &dev->tc_to_txq[q];
1755 		if (tc->offset + tc->count > txq) {
1756 			pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1757 				i, q);
1758 			netdev_set_prio_tc_map(dev, i, 0);
1759 		}
1760 	}
1761 }
1762 
1763 #ifdef CONFIG_XPS
1764 static DEFINE_MUTEX(xps_map_mutex);
1765 #define xmap_dereference(P)		\
1766 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1767 
1768 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1769 					int cpu, u16 index)
1770 {
1771 	struct xps_map *map = NULL;
1772 	int pos;
1773 
1774 	if (dev_maps)
1775 		map = xmap_dereference(dev_maps->cpu_map[cpu]);
1776 
1777 	for (pos = 0; map && pos < map->len; pos++) {
1778 		if (map->queues[pos] == index) {
1779 			if (map->len > 1) {
1780 				map->queues[pos] = map->queues[--map->len];
1781 			} else {
1782 				RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1783 				kfree_rcu(map, rcu);
1784 				map = NULL;
1785 			}
1786 			break;
1787 		}
1788 	}
1789 
1790 	return map;
1791 }
1792 
1793 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1794 {
1795 	struct xps_dev_maps *dev_maps;
1796 	int cpu, i;
1797 	bool active = false;
1798 
1799 	mutex_lock(&xps_map_mutex);
1800 	dev_maps = xmap_dereference(dev->xps_maps);
1801 
1802 	if (!dev_maps)
1803 		goto out_no_maps;
1804 
1805 	for_each_possible_cpu(cpu) {
1806 		for (i = index; i < dev->num_tx_queues; i++) {
1807 			if (!remove_xps_queue(dev_maps, cpu, i))
1808 				break;
1809 		}
1810 		if (i == dev->num_tx_queues)
1811 			active = true;
1812 	}
1813 
1814 	if (!active) {
1815 		RCU_INIT_POINTER(dev->xps_maps, NULL);
1816 		kfree_rcu(dev_maps, rcu);
1817 	}
1818 
1819 	for (i = index; i < dev->num_tx_queues; i++)
1820 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1821 					     NUMA_NO_NODE);
1822 
1823 out_no_maps:
1824 	mutex_unlock(&xps_map_mutex);
1825 }
1826 
1827 static struct xps_map *expand_xps_map(struct xps_map *map,
1828 				      int cpu, u16 index)
1829 {
1830 	struct xps_map *new_map;
1831 	int alloc_len = XPS_MIN_MAP_ALLOC;
1832 	int i, pos;
1833 
1834 	for (pos = 0; map && pos < map->len; pos++) {
1835 		if (map->queues[pos] != index)
1836 			continue;
1837 		return map;
1838 	}
1839 
1840 	/* Need to add queue to this CPU's existing map */
1841 	if (map) {
1842 		if (pos < map->alloc_len)
1843 			return map;
1844 
1845 		alloc_len = map->alloc_len * 2;
1846 	}
1847 
1848 	/* Need to allocate new map to store queue on this CPU's map */
1849 	new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1850 			       cpu_to_node(cpu));
1851 	if (!new_map)
1852 		return NULL;
1853 
1854 	for (i = 0; i < pos; i++)
1855 		new_map->queues[i] = map->queues[i];
1856 	new_map->alloc_len = alloc_len;
1857 	new_map->len = pos;
1858 
1859 	return new_map;
1860 }
1861 
1862 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1863 {
1864 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1865 	struct xps_map *map, *new_map;
1866 	int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1867 	int cpu, numa_node_id = -2;
1868 	bool active = false;
1869 
1870 	mutex_lock(&xps_map_mutex);
1871 
1872 	dev_maps = xmap_dereference(dev->xps_maps);
1873 
1874 	/* allocate memory for queue storage */
1875 	for_each_online_cpu(cpu) {
1876 		if (!cpumask_test_cpu(cpu, mask))
1877 			continue;
1878 
1879 		if (!new_dev_maps)
1880 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1881 		if (!new_dev_maps) {
1882 			mutex_unlock(&xps_map_mutex);
1883 			return -ENOMEM;
1884 		}
1885 
1886 		map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1887 				 NULL;
1888 
1889 		map = expand_xps_map(map, cpu, index);
1890 		if (!map)
1891 			goto error;
1892 
1893 		RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1894 	}
1895 
1896 	if (!new_dev_maps)
1897 		goto out_no_new_maps;
1898 
1899 	for_each_possible_cpu(cpu) {
1900 		if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1901 			/* add queue to CPU maps */
1902 			int pos = 0;
1903 
1904 			map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1905 			while ((pos < map->len) && (map->queues[pos] != index))
1906 				pos++;
1907 
1908 			if (pos == map->len)
1909 				map->queues[map->len++] = index;
1910 #ifdef CONFIG_NUMA
1911 			if (numa_node_id == -2)
1912 				numa_node_id = cpu_to_node(cpu);
1913 			else if (numa_node_id != cpu_to_node(cpu))
1914 				numa_node_id = -1;
1915 #endif
1916 		} else if (dev_maps) {
1917 			/* fill in the new device map from the old device map */
1918 			map = xmap_dereference(dev_maps->cpu_map[cpu]);
1919 			RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1920 		}
1921 
1922 	}
1923 
1924 	rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1925 
1926 	/* Cleanup old maps */
1927 	if (dev_maps) {
1928 		for_each_possible_cpu(cpu) {
1929 			new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1930 			map = xmap_dereference(dev_maps->cpu_map[cpu]);
1931 			if (map && map != new_map)
1932 				kfree_rcu(map, rcu);
1933 		}
1934 
1935 		kfree_rcu(dev_maps, rcu);
1936 	}
1937 
1938 	dev_maps = new_dev_maps;
1939 	active = true;
1940 
1941 out_no_new_maps:
1942 	/* update Tx queue numa node */
1943 	netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1944 				     (numa_node_id >= 0) ? numa_node_id :
1945 				     NUMA_NO_NODE);
1946 
1947 	if (!dev_maps)
1948 		goto out_no_maps;
1949 
1950 	/* removes queue from unused CPUs */
1951 	for_each_possible_cpu(cpu) {
1952 		if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1953 			continue;
1954 
1955 		if (remove_xps_queue(dev_maps, cpu, index))
1956 			active = true;
1957 	}
1958 
1959 	/* free map if not active */
1960 	if (!active) {
1961 		RCU_INIT_POINTER(dev->xps_maps, NULL);
1962 		kfree_rcu(dev_maps, rcu);
1963 	}
1964 
1965 out_no_maps:
1966 	mutex_unlock(&xps_map_mutex);
1967 
1968 	return 0;
1969 error:
1970 	/* remove any maps that we added */
1971 	for_each_possible_cpu(cpu) {
1972 		new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1973 		map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1974 				 NULL;
1975 		if (new_map && new_map != map)
1976 			kfree(new_map);
1977 	}
1978 
1979 	mutex_unlock(&xps_map_mutex);
1980 
1981 	kfree(new_dev_maps);
1982 	return -ENOMEM;
1983 }
1984 EXPORT_SYMBOL(netif_set_xps_queue);
1985 
1986 #endif
1987 /*
1988  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1989  * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1990  */
1991 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1992 {
1993 	int rc;
1994 
1995 	if (txq < 1 || txq > dev->num_tx_queues)
1996 		return -EINVAL;
1997 
1998 	if (dev->reg_state == NETREG_REGISTERED ||
1999 	    dev->reg_state == NETREG_UNREGISTERING) {
2000 		ASSERT_RTNL();
2001 
2002 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2003 						  txq);
2004 		if (rc)
2005 			return rc;
2006 
2007 		if (dev->num_tc)
2008 			netif_setup_tc(dev, txq);
2009 
2010 		if (txq < dev->real_num_tx_queues) {
2011 			qdisc_reset_all_tx_gt(dev, txq);
2012 #ifdef CONFIG_XPS
2013 			netif_reset_xps_queues_gt(dev, txq);
2014 #endif
2015 		}
2016 	}
2017 
2018 	dev->real_num_tx_queues = txq;
2019 	return 0;
2020 }
2021 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2022 
2023 #ifdef CONFIG_RPS
2024 /**
2025  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2026  *	@dev: Network device
2027  *	@rxq: Actual number of RX queues
2028  *
2029  *	This must be called either with the rtnl_lock held or before
2030  *	registration of the net device.  Returns 0 on success, or a
2031  *	negative error code.  If called before registration, it always
2032  *	succeeds.
2033  */
2034 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2035 {
2036 	int rc;
2037 
2038 	if (rxq < 1 || rxq > dev->num_rx_queues)
2039 		return -EINVAL;
2040 
2041 	if (dev->reg_state == NETREG_REGISTERED) {
2042 		ASSERT_RTNL();
2043 
2044 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2045 						  rxq);
2046 		if (rc)
2047 			return rc;
2048 	}
2049 
2050 	dev->real_num_rx_queues = rxq;
2051 	return 0;
2052 }
2053 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2054 #endif
2055 
2056 /**
2057  * netif_get_num_default_rss_queues - default number of RSS queues
2058  *
2059  * This routine should set an upper limit on the number of RSS queues
2060  * used by default by multiqueue devices.
2061  */
2062 int netif_get_num_default_rss_queues(void)
2063 {
2064 	return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2065 }
2066 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2067 
2068 static inline void __netif_reschedule(struct Qdisc *q)
2069 {
2070 	struct softnet_data *sd;
2071 	unsigned long flags;
2072 
2073 	local_irq_save(flags);
2074 	sd = &__get_cpu_var(softnet_data);
2075 	q->next_sched = NULL;
2076 	*sd->output_queue_tailp = q;
2077 	sd->output_queue_tailp = &q->next_sched;
2078 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
2079 	local_irq_restore(flags);
2080 }
2081 
2082 void __netif_schedule(struct Qdisc *q)
2083 {
2084 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2085 		__netif_reschedule(q);
2086 }
2087 EXPORT_SYMBOL(__netif_schedule);
2088 
2089 void dev_kfree_skb_irq(struct sk_buff *skb)
2090 {
2091 	if (atomic_dec_and_test(&skb->users)) {
2092 		struct softnet_data *sd;
2093 		unsigned long flags;
2094 
2095 		local_irq_save(flags);
2096 		sd = &__get_cpu_var(softnet_data);
2097 		skb->next = sd->completion_queue;
2098 		sd->completion_queue = skb;
2099 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
2100 		local_irq_restore(flags);
2101 	}
2102 }
2103 EXPORT_SYMBOL(dev_kfree_skb_irq);
2104 
2105 void dev_kfree_skb_any(struct sk_buff *skb)
2106 {
2107 	if (in_irq() || irqs_disabled())
2108 		dev_kfree_skb_irq(skb);
2109 	else
2110 		dev_kfree_skb(skb);
2111 }
2112 EXPORT_SYMBOL(dev_kfree_skb_any);
2113 
2114 
2115 /**
2116  * netif_device_detach - mark device as removed
2117  * @dev: network device
2118  *
2119  * Mark device as removed from system and therefore no longer available.
2120  */
2121 void netif_device_detach(struct net_device *dev)
2122 {
2123 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2124 	    netif_running(dev)) {
2125 		netif_tx_stop_all_queues(dev);
2126 	}
2127 }
2128 EXPORT_SYMBOL(netif_device_detach);
2129 
2130 /**
2131  * netif_device_attach - mark device as attached
2132  * @dev: network device
2133  *
2134  * Mark device as attached from system and restart if needed.
2135  */
2136 void netif_device_attach(struct net_device *dev)
2137 {
2138 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2139 	    netif_running(dev)) {
2140 		netif_tx_wake_all_queues(dev);
2141 		__netdev_watchdog_up(dev);
2142 	}
2143 }
2144 EXPORT_SYMBOL(netif_device_attach);
2145 
2146 static void skb_warn_bad_offload(const struct sk_buff *skb)
2147 {
2148 	static const netdev_features_t null_features = 0;
2149 	struct net_device *dev = skb->dev;
2150 	const char *driver = "";
2151 
2152 	if (dev && dev->dev.parent)
2153 		driver = dev_driver_string(dev->dev.parent);
2154 
2155 	WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2156 	     "gso_type=%d ip_summed=%d\n",
2157 	     driver, dev ? &dev->features : &null_features,
2158 	     skb->sk ? &skb->sk->sk_route_caps : &null_features,
2159 	     skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2160 	     skb_shinfo(skb)->gso_type, skb->ip_summed);
2161 }
2162 
2163 /*
2164  * Invalidate hardware checksum when packet is to be mangled, and
2165  * complete checksum manually on outgoing path.
2166  */
2167 int skb_checksum_help(struct sk_buff *skb)
2168 {
2169 	__wsum csum;
2170 	int ret = 0, offset;
2171 
2172 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2173 		goto out_set_summed;
2174 
2175 	if (unlikely(skb_shinfo(skb)->gso_size)) {
2176 		skb_warn_bad_offload(skb);
2177 		return -EINVAL;
2178 	}
2179 
2180 	/* Before computing a checksum, we should make sure no frag could
2181 	 * be modified by an external entity : checksum could be wrong.
2182 	 */
2183 	if (skb_has_shared_frag(skb)) {
2184 		ret = __skb_linearize(skb);
2185 		if (ret)
2186 			goto out;
2187 	}
2188 
2189 	offset = skb_checksum_start_offset(skb);
2190 	BUG_ON(offset >= skb_headlen(skb));
2191 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
2192 
2193 	offset += skb->csum_offset;
2194 	BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2195 
2196 	if (skb_cloned(skb) &&
2197 	    !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2198 		ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2199 		if (ret)
2200 			goto out;
2201 	}
2202 
2203 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
2204 out_set_summed:
2205 	skb->ip_summed = CHECKSUM_NONE;
2206 out:
2207 	return ret;
2208 }
2209 EXPORT_SYMBOL(skb_checksum_help);
2210 
2211 /**
2212  *	skb_mac_gso_segment - mac layer segmentation handler.
2213  *	@skb: buffer to segment
2214  *	@features: features for the output path (see dev->features)
2215  */
2216 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2217 				    netdev_features_t features)
2218 {
2219 	struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2220 	struct packet_offload *ptype;
2221 	__be16 type = skb->protocol;
2222 
2223 	while (type == htons(ETH_P_8021Q)) {
2224 		int vlan_depth = ETH_HLEN;
2225 		struct vlan_hdr *vh;
2226 
2227 		if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2228 			return ERR_PTR(-EINVAL);
2229 
2230 		vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2231 		type = vh->h_vlan_encapsulated_proto;
2232 		vlan_depth += VLAN_HLEN;
2233 	}
2234 
2235 	__skb_pull(skb, skb->mac_len);
2236 
2237 	rcu_read_lock();
2238 	list_for_each_entry_rcu(ptype, &offload_base, list) {
2239 		if (ptype->type == type && ptype->callbacks.gso_segment) {
2240 			if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2241 				int err;
2242 
2243 				err = ptype->callbacks.gso_send_check(skb);
2244 				segs = ERR_PTR(err);
2245 				if (err || skb_gso_ok(skb, features))
2246 					break;
2247 				__skb_push(skb, (skb->data -
2248 						 skb_network_header(skb)));
2249 			}
2250 			segs = ptype->callbacks.gso_segment(skb, features);
2251 			break;
2252 		}
2253 	}
2254 	rcu_read_unlock();
2255 
2256 	__skb_push(skb, skb->data - skb_mac_header(skb));
2257 
2258 	return segs;
2259 }
2260 EXPORT_SYMBOL(skb_mac_gso_segment);
2261 
2262 
2263 /* openvswitch calls this on rx path, so we need a different check.
2264  */
2265 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2266 {
2267 	if (tx_path)
2268 		return skb->ip_summed != CHECKSUM_PARTIAL;
2269 	else
2270 		return skb->ip_summed == CHECKSUM_NONE;
2271 }
2272 
2273 /**
2274  *	__skb_gso_segment - Perform segmentation on skb.
2275  *	@skb: buffer to segment
2276  *	@features: features for the output path (see dev->features)
2277  *	@tx_path: whether it is called in TX path
2278  *
2279  *	This function segments the given skb and returns a list of segments.
2280  *
2281  *	It may return NULL if the skb requires no segmentation.  This is
2282  *	only possible when GSO is used for verifying header integrity.
2283  */
2284 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2285 				  netdev_features_t features, bool tx_path)
2286 {
2287 	if (unlikely(skb_needs_check(skb, tx_path))) {
2288 		int err;
2289 
2290 		skb_warn_bad_offload(skb);
2291 
2292 		if (skb_header_cloned(skb) &&
2293 		    (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2294 			return ERR_PTR(err);
2295 	}
2296 
2297 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2298 	skb_reset_mac_header(skb);
2299 	skb_reset_mac_len(skb);
2300 
2301 	return skb_mac_gso_segment(skb, features);
2302 }
2303 EXPORT_SYMBOL(__skb_gso_segment);
2304 
2305 /* Take action when hardware reception checksum errors are detected. */
2306 #ifdef CONFIG_BUG
2307 void netdev_rx_csum_fault(struct net_device *dev)
2308 {
2309 	if (net_ratelimit()) {
2310 		pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2311 		dump_stack();
2312 	}
2313 }
2314 EXPORT_SYMBOL(netdev_rx_csum_fault);
2315 #endif
2316 
2317 /* Actually, we should eliminate this check as soon as we know, that:
2318  * 1. IOMMU is present and allows to map all the memory.
2319  * 2. No high memory really exists on this machine.
2320  */
2321 
2322 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2323 {
2324 #ifdef CONFIG_HIGHMEM
2325 	int i;
2326 	if (!(dev->features & NETIF_F_HIGHDMA)) {
2327 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2328 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2329 			if (PageHighMem(skb_frag_page(frag)))
2330 				return 1;
2331 		}
2332 	}
2333 
2334 	if (PCI_DMA_BUS_IS_PHYS) {
2335 		struct device *pdev = dev->dev.parent;
2336 
2337 		if (!pdev)
2338 			return 0;
2339 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2340 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2341 			dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2342 			if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2343 				return 1;
2344 		}
2345 	}
2346 #endif
2347 	return 0;
2348 }
2349 
2350 struct dev_gso_cb {
2351 	void (*destructor)(struct sk_buff *skb);
2352 };
2353 
2354 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2355 
2356 static void dev_gso_skb_destructor(struct sk_buff *skb)
2357 {
2358 	struct dev_gso_cb *cb;
2359 
2360 	do {
2361 		struct sk_buff *nskb = skb->next;
2362 
2363 		skb->next = nskb->next;
2364 		nskb->next = NULL;
2365 		kfree_skb(nskb);
2366 	} while (skb->next);
2367 
2368 	cb = DEV_GSO_CB(skb);
2369 	if (cb->destructor)
2370 		cb->destructor(skb);
2371 }
2372 
2373 /**
2374  *	dev_gso_segment - Perform emulated hardware segmentation on skb.
2375  *	@skb: buffer to segment
2376  *	@features: device features as applicable to this skb
2377  *
2378  *	This function segments the given skb and stores the list of segments
2379  *	in skb->next.
2380  */
2381 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2382 {
2383 	struct sk_buff *segs;
2384 
2385 	segs = skb_gso_segment(skb, features);
2386 
2387 	/* Verifying header integrity only. */
2388 	if (!segs)
2389 		return 0;
2390 
2391 	if (IS_ERR(segs))
2392 		return PTR_ERR(segs);
2393 
2394 	skb->next = segs;
2395 	DEV_GSO_CB(skb)->destructor = skb->destructor;
2396 	skb->destructor = dev_gso_skb_destructor;
2397 
2398 	return 0;
2399 }
2400 
2401 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2402 {
2403 	return ((features & NETIF_F_GEN_CSUM) ||
2404 		((features & NETIF_F_V4_CSUM) &&
2405 		 protocol == htons(ETH_P_IP)) ||
2406 		((features & NETIF_F_V6_CSUM) &&
2407 		 protocol == htons(ETH_P_IPV6)) ||
2408 		((features & NETIF_F_FCOE_CRC) &&
2409 		 protocol == htons(ETH_P_FCOE)));
2410 }
2411 
2412 static netdev_features_t harmonize_features(struct sk_buff *skb,
2413 	__be16 protocol, netdev_features_t features)
2414 {
2415 	if (skb->ip_summed != CHECKSUM_NONE &&
2416 	    !can_checksum_protocol(features, protocol)) {
2417 		features &= ~NETIF_F_ALL_CSUM;
2418 		features &= ~NETIF_F_SG;
2419 	} else if (illegal_highdma(skb->dev, skb)) {
2420 		features &= ~NETIF_F_SG;
2421 	}
2422 
2423 	return features;
2424 }
2425 
2426 netdev_features_t netif_skb_features(struct sk_buff *skb)
2427 {
2428 	__be16 protocol = skb->protocol;
2429 	netdev_features_t features = skb->dev->features;
2430 
2431 	if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2432 		features &= ~NETIF_F_GSO_MASK;
2433 
2434 	if (protocol == htons(ETH_P_8021Q)) {
2435 		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2436 		protocol = veh->h_vlan_encapsulated_proto;
2437 	} else if (!vlan_tx_tag_present(skb)) {
2438 		return harmonize_features(skb, protocol, features);
2439 	}
2440 
2441 	features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2442 
2443 	if (protocol != htons(ETH_P_8021Q)) {
2444 		return harmonize_features(skb, protocol, features);
2445 	} else {
2446 		features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2447 				NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2448 		return harmonize_features(skb, protocol, features);
2449 	}
2450 }
2451 EXPORT_SYMBOL(netif_skb_features);
2452 
2453 /*
2454  * Returns true if either:
2455  *	1. skb has frag_list and the device doesn't support FRAGLIST, or
2456  *	2. skb is fragmented and the device does not support SG.
2457  */
2458 static inline int skb_needs_linearize(struct sk_buff *skb,
2459 				      int features)
2460 {
2461 	return skb_is_nonlinear(skb) &&
2462 			((skb_has_frag_list(skb) &&
2463 				!(features & NETIF_F_FRAGLIST)) ||
2464 			(skb_shinfo(skb)->nr_frags &&
2465 				!(features & NETIF_F_SG)));
2466 }
2467 
2468 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2469 			struct netdev_queue *txq)
2470 {
2471 	const struct net_device_ops *ops = dev->netdev_ops;
2472 	int rc = NETDEV_TX_OK;
2473 	unsigned int skb_len;
2474 
2475 	if (likely(!skb->next)) {
2476 		netdev_features_t features;
2477 
2478 		/*
2479 		 * If device doesn't need skb->dst, release it right now while
2480 		 * its hot in this cpu cache
2481 		 */
2482 		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2483 			skb_dst_drop(skb);
2484 
2485 		features = netif_skb_features(skb);
2486 
2487 		if (vlan_tx_tag_present(skb) &&
2488 		    !(features & NETIF_F_HW_VLAN_TX)) {
2489 			skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2490 			if (unlikely(!skb))
2491 				goto out;
2492 
2493 			skb->vlan_tci = 0;
2494 		}
2495 
2496 		/* If encapsulation offload request, verify we are testing
2497 		 * hardware encapsulation features instead of standard
2498 		 * features for the netdev
2499 		 */
2500 		if (skb->encapsulation)
2501 			features &= dev->hw_enc_features;
2502 
2503 		if (netif_needs_gso(skb, features)) {
2504 			if (unlikely(dev_gso_segment(skb, features)))
2505 				goto out_kfree_skb;
2506 			if (skb->next)
2507 				goto gso;
2508 		} else {
2509 			if (skb_needs_linearize(skb, features) &&
2510 			    __skb_linearize(skb))
2511 				goto out_kfree_skb;
2512 
2513 			/* If packet is not checksummed and device does not
2514 			 * support checksumming for this protocol, complete
2515 			 * checksumming here.
2516 			 */
2517 			if (skb->ip_summed == CHECKSUM_PARTIAL) {
2518 				if (skb->encapsulation)
2519 					skb_set_inner_transport_header(skb,
2520 						skb_checksum_start_offset(skb));
2521 				else
2522 					skb_set_transport_header(skb,
2523 						skb_checksum_start_offset(skb));
2524 				if (!(features & NETIF_F_ALL_CSUM) &&
2525 				     skb_checksum_help(skb))
2526 					goto out_kfree_skb;
2527 			}
2528 		}
2529 
2530 		if (!list_empty(&ptype_all))
2531 			dev_queue_xmit_nit(skb, dev);
2532 
2533 		skb_len = skb->len;
2534 		rc = ops->ndo_start_xmit(skb, dev);
2535 		trace_net_dev_xmit(skb, rc, dev, skb_len);
2536 		if (rc == NETDEV_TX_OK)
2537 			txq_trans_update(txq);
2538 		return rc;
2539 	}
2540 
2541 gso:
2542 	do {
2543 		struct sk_buff *nskb = skb->next;
2544 
2545 		skb->next = nskb->next;
2546 		nskb->next = NULL;
2547 
2548 		/*
2549 		 * If device doesn't need nskb->dst, release it right now while
2550 		 * its hot in this cpu cache
2551 		 */
2552 		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2553 			skb_dst_drop(nskb);
2554 
2555 		if (!list_empty(&ptype_all))
2556 			dev_queue_xmit_nit(nskb, dev);
2557 
2558 		skb_len = nskb->len;
2559 		rc = ops->ndo_start_xmit(nskb, dev);
2560 		trace_net_dev_xmit(nskb, rc, dev, skb_len);
2561 		if (unlikely(rc != NETDEV_TX_OK)) {
2562 			if (rc & ~NETDEV_TX_MASK)
2563 				goto out_kfree_gso_skb;
2564 			nskb->next = skb->next;
2565 			skb->next = nskb;
2566 			return rc;
2567 		}
2568 		txq_trans_update(txq);
2569 		if (unlikely(netif_xmit_stopped(txq) && skb->next))
2570 			return NETDEV_TX_BUSY;
2571 	} while (skb->next);
2572 
2573 out_kfree_gso_skb:
2574 	if (likely(skb->next == NULL))
2575 		skb->destructor = DEV_GSO_CB(skb)->destructor;
2576 out_kfree_skb:
2577 	kfree_skb(skb);
2578 out:
2579 	return rc;
2580 }
2581 
2582 static void qdisc_pkt_len_init(struct sk_buff *skb)
2583 {
2584 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
2585 
2586 	qdisc_skb_cb(skb)->pkt_len = skb->len;
2587 
2588 	/* To get more precise estimation of bytes sent on wire,
2589 	 * we add to pkt_len the headers size of all segments
2590 	 */
2591 	if (shinfo->gso_size)  {
2592 		unsigned int hdr_len;
2593 
2594 		/* mac layer + network layer */
2595 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2596 
2597 		/* + transport layer */
2598 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2599 			hdr_len += tcp_hdrlen(skb);
2600 		else
2601 			hdr_len += sizeof(struct udphdr);
2602 		qdisc_skb_cb(skb)->pkt_len += (shinfo->gso_segs - 1) * hdr_len;
2603 	}
2604 }
2605 
2606 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2607 				 struct net_device *dev,
2608 				 struct netdev_queue *txq)
2609 {
2610 	spinlock_t *root_lock = qdisc_lock(q);
2611 	bool contended;
2612 	int rc;
2613 
2614 	qdisc_pkt_len_init(skb);
2615 	qdisc_calculate_pkt_len(skb, q);
2616 	/*
2617 	 * Heuristic to force contended enqueues to serialize on a
2618 	 * separate lock before trying to get qdisc main lock.
2619 	 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2620 	 * and dequeue packets faster.
2621 	 */
2622 	contended = qdisc_is_running(q);
2623 	if (unlikely(contended))
2624 		spin_lock(&q->busylock);
2625 
2626 	spin_lock(root_lock);
2627 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2628 		kfree_skb(skb);
2629 		rc = NET_XMIT_DROP;
2630 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2631 		   qdisc_run_begin(q)) {
2632 		/*
2633 		 * This is a work-conserving queue; there are no old skbs
2634 		 * waiting to be sent out; and the qdisc is not running -
2635 		 * xmit the skb directly.
2636 		 */
2637 		if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2638 			skb_dst_force(skb);
2639 
2640 		qdisc_bstats_update(q, skb);
2641 
2642 		if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2643 			if (unlikely(contended)) {
2644 				spin_unlock(&q->busylock);
2645 				contended = false;
2646 			}
2647 			__qdisc_run(q);
2648 		} else
2649 			qdisc_run_end(q);
2650 
2651 		rc = NET_XMIT_SUCCESS;
2652 	} else {
2653 		skb_dst_force(skb);
2654 		rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2655 		if (qdisc_run_begin(q)) {
2656 			if (unlikely(contended)) {
2657 				spin_unlock(&q->busylock);
2658 				contended = false;
2659 			}
2660 			__qdisc_run(q);
2661 		}
2662 	}
2663 	spin_unlock(root_lock);
2664 	if (unlikely(contended))
2665 		spin_unlock(&q->busylock);
2666 	return rc;
2667 }
2668 
2669 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2670 static void skb_update_prio(struct sk_buff *skb)
2671 {
2672 	struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2673 
2674 	if (!skb->priority && skb->sk && map) {
2675 		unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2676 
2677 		if (prioidx < map->priomap_len)
2678 			skb->priority = map->priomap[prioidx];
2679 	}
2680 }
2681 #else
2682 #define skb_update_prio(skb)
2683 #endif
2684 
2685 static DEFINE_PER_CPU(int, xmit_recursion);
2686 #define RECURSION_LIMIT 10
2687 
2688 /**
2689  *	dev_loopback_xmit - loop back @skb
2690  *	@skb: buffer to transmit
2691  */
2692 int dev_loopback_xmit(struct sk_buff *skb)
2693 {
2694 	skb_reset_mac_header(skb);
2695 	__skb_pull(skb, skb_network_offset(skb));
2696 	skb->pkt_type = PACKET_LOOPBACK;
2697 	skb->ip_summed = CHECKSUM_UNNECESSARY;
2698 	WARN_ON(!skb_dst(skb));
2699 	skb_dst_force(skb);
2700 	netif_rx_ni(skb);
2701 	return 0;
2702 }
2703 EXPORT_SYMBOL(dev_loopback_xmit);
2704 
2705 /**
2706  *	dev_queue_xmit - transmit a buffer
2707  *	@skb: buffer to transmit
2708  *
2709  *	Queue a buffer for transmission to a network device. The caller must
2710  *	have set the device and priority and built the buffer before calling
2711  *	this function. The function can be called from an interrupt.
2712  *
2713  *	A negative errno code is returned on a failure. A success does not
2714  *	guarantee the frame will be transmitted as it may be dropped due
2715  *	to congestion or traffic shaping.
2716  *
2717  * -----------------------------------------------------------------------------------
2718  *      I notice this method can also return errors from the queue disciplines,
2719  *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
2720  *      be positive.
2721  *
2722  *      Regardless of the return value, the skb is consumed, so it is currently
2723  *      difficult to retry a send to this method.  (You can bump the ref count
2724  *      before sending to hold a reference for retry if you are careful.)
2725  *
2726  *      When calling this method, interrupts MUST be enabled.  This is because
2727  *      the BH enable code must have IRQs enabled so that it will not deadlock.
2728  *          --BLG
2729  */
2730 int dev_queue_xmit(struct sk_buff *skb)
2731 {
2732 	struct net_device *dev = skb->dev;
2733 	struct netdev_queue *txq;
2734 	struct Qdisc *q;
2735 	int rc = -ENOMEM;
2736 
2737 	skb_reset_mac_header(skb);
2738 
2739 	/* Disable soft irqs for various locks below. Also
2740 	 * stops preemption for RCU.
2741 	 */
2742 	rcu_read_lock_bh();
2743 
2744 	skb_update_prio(skb);
2745 
2746 	txq = netdev_pick_tx(dev, skb);
2747 	q = rcu_dereference_bh(txq->qdisc);
2748 
2749 #ifdef CONFIG_NET_CLS_ACT
2750 	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2751 #endif
2752 	trace_net_dev_queue(skb);
2753 	if (q->enqueue) {
2754 		rc = __dev_xmit_skb(skb, q, dev, txq);
2755 		goto out;
2756 	}
2757 
2758 	/* The device has no queue. Common case for software devices:
2759 	   loopback, all the sorts of tunnels...
2760 
2761 	   Really, it is unlikely that netif_tx_lock protection is necessary
2762 	   here.  (f.e. loopback and IP tunnels are clean ignoring statistics
2763 	   counters.)
2764 	   However, it is possible, that they rely on protection
2765 	   made by us here.
2766 
2767 	   Check this and shot the lock. It is not prone from deadlocks.
2768 	   Either shot noqueue qdisc, it is even simpler 8)
2769 	 */
2770 	if (dev->flags & IFF_UP) {
2771 		int cpu = smp_processor_id(); /* ok because BHs are off */
2772 
2773 		if (txq->xmit_lock_owner != cpu) {
2774 
2775 			if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2776 				goto recursion_alert;
2777 
2778 			HARD_TX_LOCK(dev, txq, cpu);
2779 
2780 			if (!netif_xmit_stopped(txq)) {
2781 				__this_cpu_inc(xmit_recursion);
2782 				rc = dev_hard_start_xmit(skb, dev, txq);
2783 				__this_cpu_dec(xmit_recursion);
2784 				if (dev_xmit_complete(rc)) {
2785 					HARD_TX_UNLOCK(dev, txq);
2786 					goto out;
2787 				}
2788 			}
2789 			HARD_TX_UNLOCK(dev, txq);
2790 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2791 					     dev->name);
2792 		} else {
2793 			/* Recursion is detected! It is possible,
2794 			 * unfortunately
2795 			 */
2796 recursion_alert:
2797 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2798 					     dev->name);
2799 		}
2800 	}
2801 
2802 	rc = -ENETDOWN;
2803 	rcu_read_unlock_bh();
2804 
2805 	kfree_skb(skb);
2806 	return rc;
2807 out:
2808 	rcu_read_unlock_bh();
2809 	return rc;
2810 }
2811 EXPORT_SYMBOL(dev_queue_xmit);
2812 
2813 
2814 /*=======================================================================
2815 			Receiver routines
2816   =======================================================================*/
2817 
2818 int netdev_max_backlog __read_mostly = 1000;
2819 EXPORT_SYMBOL(netdev_max_backlog);
2820 
2821 int netdev_tstamp_prequeue __read_mostly = 1;
2822 int netdev_budget __read_mostly = 300;
2823 int weight_p __read_mostly = 64;            /* old backlog weight */
2824 
2825 /* Called with irq disabled */
2826 static inline void ____napi_schedule(struct softnet_data *sd,
2827 				     struct napi_struct *napi)
2828 {
2829 	list_add_tail(&napi->poll_list, &sd->poll_list);
2830 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
2831 }
2832 
2833 #ifdef CONFIG_RPS
2834 
2835 /* One global table that all flow-based protocols share. */
2836 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2837 EXPORT_SYMBOL(rps_sock_flow_table);
2838 
2839 struct static_key rps_needed __read_mostly;
2840 
2841 static struct rps_dev_flow *
2842 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2843 	    struct rps_dev_flow *rflow, u16 next_cpu)
2844 {
2845 	if (next_cpu != RPS_NO_CPU) {
2846 #ifdef CONFIG_RFS_ACCEL
2847 		struct netdev_rx_queue *rxqueue;
2848 		struct rps_dev_flow_table *flow_table;
2849 		struct rps_dev_flow *old_rflow;
2850 		u32 flow_id;
2851 		u16 rxq_index;
2852 		int rc;
2853 
2854 		/* Should we steer this flow to a different hardware queue? */
2855 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2856 		    !(dev->features & NETIF_F_NTUPLE))
2857 			goto out;
2858 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2859 		if (rxq_index == skb_get_rx_queue(skb))
2860 			goto out;
2861 
2862 		rxqueue = dev->_rx + rxq_index;
2863 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
2864 		if (!flow_table)
2865 			goto out;
2866 		flow_id = skb->rxhash & flow_table->mask;
2867 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2868 							rxq_index, flow_id);
2869 		if (rc < 0)
2870 			goto out;
2871 		old_rflow = rflow;
2872 		rflow = &flow_table->flows[flow_id];
2873 		rflow->filter = rc;
2874 		if (old_rflow->filter == rflow->filter)
2875 			old_rflow->filter = RPS_NO_FILTER;
2876 	out:
2877 #endif
2878 		rflow->last_qtail =
2879 			per_cpu(softnet_data, next_cpu).input_queue_head;
2880 	}
2881 
2882 	rflow->cpu = next_cpu;
2883 	return rflow;
2884 }
2885 
2886 /*
2887  * get_rps_cpu is called from netif_receive_skb and returns the target
2888  * CPU from the RPS map of the receiving queue for a given skb.
2889  * rcu_read_lock must be held on entry.
2890  */
2891 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2892 		       struct rps_dev_flow **rflowp)
2893 {
2894 	struct netdev_rx_queue *rxqueue;
2895 	struct rps_map *map;
2896 	struct rps_dev_flow_table *flow_table;
2897 	struct rps_sock_flow_table *sock_flow_table;
2898 	int cpu = -1;
2899 	u16 tcpu;
2900 
2901 	if (skb_rx_queue_recorded(skb)) {
2902 		u16 index = skb_get_rx_queue(skb);
2903 		if (unlikely(index >= dev->real_num_rx_queues)) {
2904 			WARN_ONCE(dev->real_num_rx_queues > 1,
2905 				  "%s received packet on queue %u, but number "
2906 				  "of RX queues is %u\n",
2907 				  dev->name, index, dev->real_num_rx_queues);
2908 			goto done;
2909 		}
2910 		rxqueue = dev->_rx + index;
2911 	} else
2912 		rxqueue = dev->_rx;
2913 
2914 	map = rcu_dereference(rxqueue->rps_map);
2915 	if (map) {
2916 		if (map->len == 1 &&
2917 		    !rcu_access_pointer(rxqueue->rps_flow_table)) {
2918 			tcpu = map->cpus[0];
2919 			if (cpu_online(tcpu))
2920 				cpu = tcpu;
2921 			goto done;
2922 		}
2923 	} else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2924 		goto done;
2925 	}
2926 
2927 	skb_reset_network_header(skb);
2928 	if (!skb_get_rxhash(skb))
2929 		goto done;
2930 
2931 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
2932 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
2933 	if (flow_table && sock_flow_table) {
2934 		u16 next_cpu;
2935 		struct rps_dev_flow *rflow;
2936 
2937 		rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2938 		tcpu = rflow->cpu;
2939 
2940 		next_cpu = sock_flow_table->ents[skb->rxhash &
2941 		    sock_flow_table->mask];
2942 
2943 		/*
2944 		 * If the desired CPU (where last recvmsg was done) is
2945 		 * different from current CPU (one in the rx-queue flow
2946 		 * table entry), switch if one of the following holds:
2947 		 *   - Current CPU is unset (equal to RPS_NO_CPU).
2948 		 *   - Current CPU is offline.
2949 		 *   - The current CPU's queue tail has advanced beyond the
2950 		 *     last packet that was enqueued using this table entry.
2951 		 *     This guarantees that all previous packets for the flow
2952 		 *     have been dequeued, thus preserving in order delivery.
2953 		 */
2954 		if (unlikely(tcpu != next_cpu) &&
2955 		    (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2956 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2957 		      rflow->last_qtail)) >= 0)) {
2958 			tcpu = next_cpu;
2959 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2960 		}
2961 
2962 		if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2963 			*rflowp = rflow;
2964 			cpu = tcpu;
2965 			goto done;
2966 		}
2967 	}
2968 
2969 	if (map) {
2970 		tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2971 
2972 		if (cpu_online(tcpu)) {
2973 			cpu = tcpu;
2974 			goto done;
2975 		}
2976 	}
2977 
2978 done:
2979 	return cpu;
2980 }
2981 
2982 #ifdef CONFIG_RFS_ACCEL
2983 
2984 /**
2985  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2986  * @dev: Device on which the filter was set
2987  * @rxq_index: RX queue index
2988  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2989  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2990  *
2991  * Drivers that implement ndo_rx_flow_steer() should periodically call
2992  * this function for each installed filter and remove the filters for
2993  * which it returns %true.
2994  */
2995 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2996 			 u32 flow_id, u16 filter_id)
2997 {
2998 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2999 	struct rps_dev_flow_table *flow_table;
3000 	struct rps_dev_flow *rflow;
3001 	bool expire = true;
3002 	int cpu;
3003 
3004 	rcu_read_lock();
3005 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
3006 	if (flow_table && flow_id <= flow_table->mask) {
3007 		rflow = &flow_table->flows[flow_id];
3008 		cpu = ACCESS_ONCE(rflow->cpu);
3009 		if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3010 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3011 			   rflow->last_qtail) <
3012 		     (int)(10 * flow_table->mask)))
3013 			expire = false;
3014 	}
3015 	rcu_read_unlock();
3016 	return expire;
3017 }
3018 EXPORT_SYMBOL(rps_may_expire_flow);
3019 
3020 #endif /* CONFIG_RFS_ACCEL */
3021 
3022 /* Called from hardirq (IPI) context */
3023 static void rps_trigger_softirq(void *data)
3024 {
3025 	struct softnet_data *sd = data;
3026 
3027 	____napi_schedule(sd, &sd->backlog);
3028 	sd->received_rps++;
3029 }
3030 
3031 #endif /* CONFIG_RPS */
3032 
3033 /*
3034  * Check if this softnet_data structure is another cpu one
3035  * If yes, queue it to our IPI list and return 1
3036  * If no, return 0
3037  */
3038 static int rps_ipi_queued(struct softnet_data *sd)
3039 {
3040 #ifdef CONFIG_RPS
3041 	struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3042 
3043 	if (sd != mysd) {
3044 		sd->rps_ipi_next = mysd->rps_ipi_list;
3045 		mysd->rps_ipi_list = sd;
3046 
3047 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
3048 		return 1;
3049 	}
3050 #endif /* CONFIG_RPS */
3051 	return 0;
3052 }
3053 
3054 /*
3055  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3056  * queue (may be a remote CPU queue).
3057  */
3058 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3059 			      unsigned int *qtail)
3060 {
3061 	struct softnet_data *sd;
3062 	unsigned long flags;
3063 
3064 	sd = &per_cpu(softnet_data, cpu);
3065 
3066 	local_irq_save(flags);
3067 
3068 	rps_lock(sd);
3069 	if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
3070 		if (skb_queue_len(&sd->input_pkt_queue)) {
3071 enqueue:
3072 			__skb_queue_tail(&sd->input_pkt_queue, skb);
3073 			input_queue_tail_incr_save(sd, qtail);
3074 			rps_unlock(sd);
3075 			local_irq_restore(flags);
3076 			return NET_RX_SUCCESS;
3077 		}
3078 
3079 		/* Schedule NAPI for backlog device
3080 		 * We can use non atomic operation since we own the queue lock
3081 		 */
3082 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3083 			if (!rps_ipi_queued(sd))
3084 				____napi_schedule(sd, &sd->backlog);
3085 		}
3086 		goto enqueue;
3087 	}
3088 
3089 	sd->dropped++;
3090 	rps_unlock(sd);
3091 
3092 	local_irq_restore(flags);
3093 
3094 	atomic_long_inc(&skb->dev->rx_dropped);
3095 	kfree_skb(skb);
3096 	return NET_RX_DROP;
3097 }
3098 
3099 /**
3100  *	netif_rx	-	post buffer to the network code
3101  *	@skb: buffer to post
3102  *
3103  *	This function receives a packet from a device driver and queues it for
3104  *	the upper (protocol) levels to process.  It always succeeds. The buffer
3105  *	may be dropped during processing for congestion control or by the
3106  *	protocol layers.
3107  *
3108  *	return values:
3109  *	NET_RX_SUCCESS	(no congestion)
3110  *	NET_RX_DROP     (packet was dropped)
3111  *
3112  */
3113 
3114 int netif_rx(struct sk_buff *skb)
3115 {
3116 	int ret;
3117 
3118 	/* if netpoll wants it, pretend we never saw it */
3119 	if (netpoll_rx(skb))
3120 		return NET_RX_DROP;
3121 
3122 	net_timestamp_check(netdev_tstamp_prequeue, skb);
3123 
3124 	trace_netif_rx(skb);
3125 #ifdef CONFIG_RPS
3126 	if (static_key_false(&rps_needed)) {
3127 		struct rps_dev_flow voidflow, *rflow = &voidflow;
3128 		int cpu;
3129 
3130 		preempt_disable();
3131 		rcu_read_lock();
3132 
3133 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
3134 		if (cpu < 0)
3135 			cpu = smp_processor_id();
3136 
3137 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3138 
3139 		rcu_read_unlock();
3140 		preempt_enable();
3141 	} else
3142 #endif
3143 	{
3144 		unsigned int qtail;
3145 		ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3146 		put_cpu();
3147 	}
3148 	return ret;
3149 }
3150 EXPORT_SYMBOL(netif_rx);
3151 
3152 int netif_rx_ni(struct sk_buff *skb)
3153 {
3154 	int err;
3155 
3156 	preempt_disable();
3157 	err = netif_rx(skb);
3158 	if (local_softirq_pending())
3159 		do_softirq();
3160 	preempt_enable();
3161 
3162 	return err;
3163 }
3164 EXPORT_SYMBOL(netif_rx_ni);
3165 
3166 static void net_tx_action(struct softirq_action *h)
3167 {
3168 	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3169 
3170 	if (sd->completion_queue) {
3171 		struct sk_buff *clist;
3172 
3173 		local_irq_disable();
3174 		clist = sd->completion_queue;
3175 		sd->completion_queue = NULL;
3176 		local_irq_enable();
3177 
3178 		while (clist) {
3179 			struct sk_buff *skb = clist;
3180 			clist = clist->next;
3181 
3182 			WARN_ON(atomic_read(&skb->users));
3183 			trace_kfree_skb(skb, net_tx_action);
3184 			__kfree_skb(skb);
3185 		}
3186 	}
3187 
3188 	if (sd->output_queue) {
3189 		struct Qdisc *head;
3190 
3191 		local_irq_disable();
3192 		head = sd->output_queue;
3193 		sd->output_queue = NULL;
3194 		sd->output_queue_tailp = &sd->output_queue;
3195 		local_irq_enable();
3196 
3197 		while (head) {
3198 			struct Qdisc *q = head;
3199 			spinlock_t *root_lock;
3200 
3201 			head = head->next_sched;
3202 
3203 			root_lock = qdisc_lock(q);
3204 			if (spin_trylock(root_lock)) {
3205 				smp_mb__before_clear_bit();
3206 				clear_bit(__QDISC_STATE_SCHED,
3207 					  &q->state);
3208 				qdisc_run(q);
3209 				spin_unlock(root_lock);
3210 			} else {
3211 				if (!test_bit(__QDISC_STATE_DEACTIVATED,
3212 					      &q->state)) {
3213 					__netif_reschedule(q);
3214 				} else {
3215 					smp_mb__before_clear_bit();
3216 					clear_bit(__QDISC_STATE_SCHED,
3217 						  &q->state);
3218 				}
3219 			}
3220 		}
3221 	}
3222 }
3223 
3224 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3225     (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3226 /* This hook is defined here for ATM LANE */
3227 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3228 			     unsigned char *addr) __read_mostly;
3229 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3230 #endif
3231 
3232 #ifdef CONFIG_NET_CLS_ACT
3233 /* TODO: Maybe we should just force sch_ingress to be compiled in
3234  * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3235  * a compare and 2 stores extra right now if we dont have it on
3236  * but have CONFIG_NET_CLS_ACT
3237  * NOTE: This doesn't stop any functionality; if you dont have
3238  * the ingress scheduler, you just can't add policies on ingress.
3239  *
3240  */
3241 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3242 {
3243 	struct net_device *dev = skb->dev;
3244 	u32 ttl = G_TC_RTTL(skb->tc_verd);
3245 	int result = TC_ACT_OK;
3246 	struct Qdisc *q;
3247 
3248 	if (unlikely(MAX_RED_LOOP < ttl++)) {
3249 		net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3250 				     skb->skb_iif, dev->ifindex);
3251 		return TC_ACT_SHOT;
3252 	}
3253 
3254 	skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3255 	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3256 
3257 	q = rxq->qdisc;
3258 	if (q != &noop_qdisc) {
3259 		spin_lock(qdisc_lock(q));
3260 		if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3261 			result = qdisc_enqueue_root(skb, q);
3262 		spin_unlock(qdisc_lock(q));
3263 	}
3264 
3265 	return result;
3266 }
3267 
3268 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3269 					 struct packet_type **pt_prev,
3270 					 int *ret, struct net_device *orig_dev)
3271 {
3272 	struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3273 
3274 	if (!rxq || rxq->qdisc == &noop_qdisc)
3275 		goto out;
3276 
3277 	if (*pt_prev) {
3278 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
3279 		*pt_prev = NULL;
3280 	}
3281 
3282 	switch (ing_filter(skb, rxq)) {
3283 	case TC_ACT_SHOT:
3284 	case TC_ACT_STOLEN:
3285 		kfree_skb(skb);
3286 		return NULL;
3287 	}
3288 
3289 out:
3290 	skb->tc_verd = 0;
3291 	return skb;
3292 }
3293 #endif
3294 
3295 /**
3296  *	netdev_rx_handler_register - register receive handler
3297  *	@dev: device to register a handler for
3298  *	@rx_handler: receive handler to register
3299  *	@rx_handler_data: data pointer that is used by rx handler
3300  *
3301  *	Register a receive hander for a device. This handler will then be
3302  *	called from __netif_receive_skb. A negative errno code is returned
3303  *	on a failure.
3304  *
3305  *	The caller must hold the rtnl_mutex.
3306  *
3307  *	For a general description of rx_handler, see enum rx_handler_result.
3308  */
3309 int netdev_rx_handler_register(struct net_device *dev,
3310 			       rx_handler_func_t *rx_handler,
3311 			       void *rx_handler_data)
3312 {
3313 	ASSERT_RTNL();
3314 
3315 	if (dev->rx_handler)
3316 		return -EBUSY;
3317 
3318 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3319 	rcu_assign_pointer(dev->rx_handler, rx_handler);
3320 
3321 	return 0;
3322 }
3323 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3324 
3325 /**
3326  *	netdev_rx_handler_unregister - unregister receive handler
3327  *	@dev: device to unregister a handler from
3328  *
3329  *	Unregister a receive hander from a device.
3330  *
3331  *	The caller must hold the rtnl_mutex.
3332  */
3333 void netdev_rx_handler_unregister(struct net_device *dev)
3334 {
3335 
3336 	ASSERT_RTNL();
3337 	RCU_INIT_POINTER(dev->rx_handler, NULL);
3338 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3339 }
3340 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3341 
3342 /*
3343  * Limit the use of PFMEMALLOC reserves to those protocols that implement
3344  * the special handling of PFMEMALLOC skbs.
3345  */
3346 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3347 {
3348 	switch (skb->protocol) {
3349 	case __constant_htons(ETH_P_ARP):
3350 	case __constant_htons(ETH_P_IP):
3351 	case __constant_htons(ETH_P_IPV6):
3352 	case __constant_htons(ETH_P_8021Q):
3353 		return true;
3354 	default:
3355 		return false;
3356 	}
3357 }
3358 
3359 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3360 {
3361 	struct packet_type *ptype, *pt_prev;
3362 	rx_handler_func_t *rx_handler;
3363 	struct net_device *orig_dev;
3364 	struct net_device *null_or_dev;
3365 	bool deliver_exact = false;
3366 	int ret = NET_RX_DROP;
3367 	__be16 type;
3368 
3369 	net_timestamp_check(!netdev_tstamp_prequeue, skb);
3370 
3371 	trace_netif_receive_skb(skb);
3372 
3373 	/* if we've gotten here through NAPI, check netpoll */
3374 	if (netpoll_receive_skb(skb))
3375 		goto out;
3376 
3377 	orig_dev = skb->dev;
3378 
3379 	skb_reset_network_header(skb);
3380 	if (!skb_transport_header_was_set(skb))
3381 		skb_reset_transport_header(skb);
3382 	skb_reset_mac_len(skb);
3383 
3384 	pt_prev = NULL;
3385 
3386 	rcu_read_lock();
3387 
3388 another_round:
3389 	skb->skb_iif = skb->dev->ifindex;
3390 
3391 	__this_cpu_inc(softnet_data.processed);
3392 
3393 	if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3394 		skb = vlan_untag(skb);
3395 		if (unlikely(!skb))
3396 			goto unlock;
3397 	}
3398 
3399 #ifdef CONFIG_NET_CLS_ACT
3400 	if (skb->tc_verd & TC_NCLS) {
3401 		skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3402 		goto ncls;
3403 	}
3404 #endif
3405 
3406 	if (pfmemalloc)
3407 		goto skip_taps;
3408 
3409 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
3410 		if (!ptype->dev || ptype->dev == skb->dev) {
3411 			if (pt_prev)
3412 				ret = deliver_skb(skb, pt_prev, orig_dev);
3413 			pt_prev = ptype;
3414 		}
3415 	}
3416 
3417 skip_taps:
3418 #ifdef CONFIG_NET_CLS_ACT
3419 	skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3420 	if (!skb)
3421 		goto unlock;
3422 ncls:
3423 #endif
3424 
3425 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3426 		goto drop;
3427 
3428 	if (vlan_tx_tag_present(skb)) {
3429 		if (pt_prev) {
3430 			ret = deliver_skb(skb, pt_prev, orig_dev);
3431 			pt_prev = NULL;
3432 		}
3433 		if (vlan_do_receive(&skb))
3434 			goto another_round;
3435 		else if (unlikely(!skb))
3436 			goto unlock;
3437 	}
3438 
3439 	rx_handler = rcu_dereference(skb->dev->rx_handler);
3440 	if (rx_handler) {
3441 		if (pt_prev) {
3442 			ret = deliver_skb(skb, pt_prev, orig_dev);
3443 			pt_prev = NULL;
3444 		}
3445 		switch (rx_handler(&skb)) {
3446 		case RX_HANDLER_CONSUMED:
3447 			ret = NET_RX_SUCCESS;
3448 			goto unlock;
3449 		case RX_HANDLER_ANOTHER:
3450 			goto another_round;
3451 		case RX_HANDLER_EXACT:
3452 			deliver_exact = true;
3453 		case RX_HANDLER_PASS:
3454 			break;
3455 		default:
3456 			BUG();
3457 		}
3458 	}
3459 
3460 	if (vlan_tx_nonzero_tag_present(skb))
3461 		skb->pkt_type = PACKET_OTHERHOST;
3462 
3463 	/* deliver only exact match when indicated */
3464 	null_or_dev = deliver_exact ? skb->dev : NULL;
3465 
3466 	type = skb->protocol;
3467 	list_for_each_entry_rcu(ptype,
3468 			&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3469 		if (ptype->type == type &&
3470 		    (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3471 		     ptype->dev == orig_dev)) {
3472 			if (pt_prev)
3473 				ret = deliver_skb(skb, pt_prev, orig_dev);
3474 			pt_prev = ptype;
3475 		}
3476 	}
3477 
3478 	if (pt_prev) {
3479 		if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3480 			goto drop;
3481 		else
3482 			ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3483 	} else {
3484 drop:
3485 		atomic_long_inc(&skb->dev->rx_dropped);
3486 		kfree_skb(skb);
3487 		/* Jamal, now you will not able to escape explaining
3488 		 * me how you were going to use this. :-)
3489 		 */
3490 		ret = NET_RX_DROP;
3491 	}
3492 
3493 unlock:
3494 	rcu_read_unlock();
3495 out:
3496 	return ret;
3497 }
3498 
3499 static int __netif_receive_skb(struct sk_buff *skb)
3500 {
3501 	int ret;
3502 
3503 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3504 		unsigned long pflags = current->flags;
3505 
3506 		/*
3507 		 * PFMEMALLOC skbs are special, they should
3508 		 * - be delivered to SOCK_MEMALLOC sockets only
3509 		 * - stay away from userspace
3510 		 * - have bounded memory usage
3511 		 *
3512 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
3513 		 * context down to all allocation sites.
3514 		 */
3515 		current->flags |= PF_MEMALLOC;
3516 		ret = __netif_receive_skb_core(skb, true);
3517 		tsk_restore_flags(current, pflags, PF_MEMALLOC);
3518 	} else
3519 		ret = __netif_receive_skb_core(skb, false);
3520 
3521 	return ret;
3522 }
3523 
3524 /**
3525  *	netif_receive_skb - process receive buffer from network
3526  *	@skb: buffer to process
3527  *
3528  *	netif_receive_skb() is the main receive data processing function.
3529  *	It always succeeds. The buffer may be dropped during processing
3530  *	for congestion control or by the protocol layers.
3531  *
3532  *	This function may only be called from softirq context and interrupts
3533  *	should be enabled.
3534  *
3535  *	Return values (usually ignored):
3536  *	NET_RX_SUCCESS: no congestion
3537  *	NET_RX_DROP: packet was dropped
3538  */
3539 int netif_receive_skb(struct sk_buff *skb)
3540 {
3541 	net_timestamp_check(netdev_tstamp_prequeue, skb);
3542 
3543 	if (skb_defer_rx_timestamp(skb))
3544 		return NET_RX_SUCCESS;
3545 
3546 #ifdef CONFIG_RPS
3547 	if (static_key_false(&rps_needed)) {
3548 		struct rps_dev_flow voidflow, *rflow = &voidflow;
3549 		int cpu, ret;
3550 
3551 		rcu_read_lock();
3552 
3553 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
3554 
3555 		if (cpu >= 0) {
3556 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3557 			rcu_read_unlock();
3558 			return ret;
3559 		}
3560 		rcu_read_unlock();
3561 	}
3562 #endif
3563 	return __netif_receive_skb(skb);
3564 }
3565 EXPORT_SYMBOL(netif_receive_skb);
3566 
3567 /* Network device is going away, flush any packets still pending
3568  * Called with irqs disabled.
3569  */
3570 static void flush_backlog(void *arg)
3571 {
3572 	struct net_device *dev = arg;
3573 	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3574 	struct sk_buff *skb, *tmp;
3575 
3576 	rps_lock(sd);
3577 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3578 		if (skb->dev == dev) {
3579 			__skb_unlink(skb, &sd->input_pkt_queue);
3580 			kfree_skb(skb);
3581 			input_queue_head_incr(sd);
3582 		}
3583 	}
3584 	rps_unlock(sd);
3585 
3586 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3587 		if (skb->dev == dev) {
3588 			__skb_unlink(skb, &sd->process_queue);
3589 			kfree_skb(skb);
3590 			input_queue_head_incr(sd);
3591 		}
3592 	}
3593 }
3594 
3595 static int napi_gro_complete(struct sk_buff *skb)
3596 {
3597 	struct packet_offload *ptype;
3598 	__be16 type = skb->protocol;
3599 	struct list_head *head = &offload_base;
3600 	int err = -ENOENT;
3601 
3602 	BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3603 
3604 	if (NAPI_GRO_CB(skb)->count == 1) {
3605 		skb_shinfo(skb)->gso_size = 0;
3606 		goto out;
3607 	}
3608 
3609 	rcu_read_lock();
3610 	list_for_each_entry_rcu(ptype, head, list) {
3611 		if (ptype->type != type || !ptype->callbacks.gro_complete)
3612 			continue;
3613 
3614 		err = ptype->callbacks.gro_complete(skb);
3615 		break;
3616 	}
3617 	rcu_read_unlock();
3618 
3619 	if (err) {
3620 		WARN_ON(&ptype->list == head);
3621 		kfree_skb(skb);
3622 		return NET_RX_SUCCESS;
3623 	}
3624 
3625 out:
3626 	return netif_receive_skb(skb);
3627 }
3628 
3629 /* napi->gro_list contains packets ordered by age.
3630  * youngest packets at the head of it.
3631  * Complete skbs in reverse order to reduce latencies.
3632  */
3633 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3634 {
3635 	struct sk_buff *skb, *prev = NULL;
3636 
3637 	/* scan list and build reverse chain */
3638 	for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3639 		skb->prev = prev;
3640 		prev = skb;
3641 	}
3642 
3643 	for (skb = prev; skb; skb = prev) {
3644 		skb->next = NULL;
3645 
3646 		if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3647 			return;
3648 
3649 		prev = skb->prev;
3650 		napi_gro_complete(skb);
3651 		napi->gro_count--;
3652 	}
3653 
3654 	napi->gro_list = NULL;
3655 }
3656 EXPORT_SYMBOL(napi_gro_flush);
3657 
3658 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3659 {
3660 	struct sk_buff *p;
3661 	unsigned int maclen = skb->dev->hard_header_len;
3662 
3663 	for (p = napi->gro_list; p; p = p->next) {
3664 		unsigned long diffs;
3665 
3666 		diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3667 		diffs |= p->vlan_tci ^ skb->vlan_tci;
3668 		if (maclen == ETH_HLEN)
3669 			diffs |= compare_ether_header(skb_mac_header(p),
3670 						      skb_gro_mac_header(skb));
3671 		else if (!diffs)
3672 			diffs = memcmp(skb_mac_header(p),
3673 				       skb_gro_mac_header(skb),
3674 				       maclen);
3675 		NAPI_GRO_CB(p)->same_flow = !diffs;
3676 		NAPI_GRO_CB(p)->flush = 0;
3677 	}
3678 }
3679 
3680 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3681 {
3682 	struct sk_buff **pp = NULL;
3683 	struct packet_offload *ptype;
3684 	__be16 type = skb->protocol;
3685 	struct list_head *head = &offload_base;
3686 	int same_flow;
3687 	enum gro_result ret;
3688 
3689 	if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3690 		goto normal;
3691 
3692 	if (skb_is_gso(skb) || skb_has_frag_list(skb))
3693 		goto normal;
3694 
3695 	gro_list_prepare(napi, skb);
3696 
3697 	rcu_read_lock();
3698 	list_for_each_entry_rcu(ptype, head, list) {
3699 		if (ptype->type != type || !ptype->callbacks.gro_receive)
3700 			continue;
3701 
3702 		skb_set_network_header(skb, skb_gro_offset(skb));
3703 		skb_reset_mac_len(skb);
3704 		NAPI_GRO_CB(skb)->same_flow = 0;
3705 		NAPI_GRO_CB(skb)->flush = 0;
3706 		NAPI_GRO_CB(skb)->free = 0;
3707 
3708 		pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3709 		break;
3710 	}
3711 	rcu_read_unlock();
3712 
3713 	if (&ptype->list == head)
3714 		goto normal;
3715 
3716 	same_flow = NAPI_GRO_CB(skb)->same_flow;
3717 	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3718 
3719 	if (pp) {
3720 		struct sk_buff *nskb = *pp;
3721 
3722 		*pp = nskb->next;
3723 		nskb->next = NULL;
3724 		napi_gro_complete(nskb);
3725 		napi->gro_count--;
3726 	}
3727 
3728 	if (same_flow)
3729 		goto ok;
3730 
3731 	if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3732 		goto normal;
3733 
3734 	napi->gro_count++;
3735 	NAPI_GRO_CB(skb)->count = 1;
3736 	NAPI_GRO_CB(skb)->age = jiffies;
3737 	skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3738 	skb->next = napi->gro_list;
3739 	napi->gro_list = skb;
3740 	ret = GRO_HELD;
3741 
3742 pull:
3743 	if (skb_headlen(skb) < skb_gro_offset(skb)) {
3744 		int grow = skb_gro_offset(skb) - skb_headlen(skb);
3745 
3746 		BUG_ON(skb->end - skb->tail < grow);
3747 
3748 		memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3749 
3750 		skb->tail += grow;
3751 		skb->data_len -= grow;
3752 
3753 		skb_shinfo(skb)->frags[0].page_offset += grow;
3754 		skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3755 
3756 		if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3757 			skb_frag_unref(skb, 0);
3758 			memmove(skb_shinfo(skb)->frags,
3759 				skb_shinfo(skb)->frags + 1,
3760 				--skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3761 		}
3762 	}
3763 
3764 ok:
3765 	return ret;
3766 
3767 normal:
3768 	ret = GRO_NORMAL;
3769 	goto pull;
3770 }
3771 
3772 
3773 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3774 {
3775 	switch (ret) {
3776 	case GRO_NORMAL:
3777 		if (netif_receive_skb(skb))
3778 			ret = GRO_DROP;
3779 		break;
3780 
3781 	case GRO_DROP:
3782 		kfree_skb(skb);
3783 		break;
3784 
3785 	case GRO_MERGED_FREE:
3786 		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3787 			kmem_cache_free(skbuff_head_cache, skb);
3788 		else
3789 			__kfree_skb(skb);
3790 		break;
3791 
3792 	case GRO_HELD:
3793 	case GRO_MERGED:
3794 		break;
3795 	}
3796 
3797 	return ret;
3798 }
3799 
3800 static void skb_gro_reset_offset(struct sk_buff *skb)
3801 {
3802 	const struct skb_shared_info *pinfo = skb_shinfo(skb);
3803 	const skb_frag_t *frag0 = &pinfo->frags[0];
3804 
3805 	NAPI_GRO_CB(skb)->data_offset = 0;
3806 	NAPI_GRO_CB(skb)->frag0 = NULL;
3807 	NAPI_GRO_CB(skb)->frag0_len = 0;
3808 
3809 	if (skb->mac_header == skb->tail &&
3810 	    pinfo->nr_frags &&
3811 	    !PageHighMem(skb_frag_page(frag0))) {
3812 		NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3813 		NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3814 	}
3815 }
3816 
3817 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3818 {
3819 	skb_gro_reset_offset(skb);
3820 
3821 	return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3822 }
3823 EXPORT_SYMBOL(napi_gro_receive);
3824 
3825 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3826 {
3827 	__skb_pull(skb, skb_headlen(skb));
3828 	/* restore the reserve we had after netdev_alloc_skb_ip_align() */
3829 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3830 	skb->vlan_tci = 0;
3831 	skb->dev = napi->dev;
3832 	skb->skb_iif = 0;
3833 
3834 	napi->skb = skb;
3835 }
3836 
3837 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3838 {
3839 	struct sk_buff *skb = napi->skb;
3840 
3841 	if (!skb) {
3842 		skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3843 		if (skb)
3844 			napi->skb = skb;
3845 	}
3846 	return skb;
3847 }
3848 EXPORT_SYMBOL(napi_get_frags);
3849 
3850 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3851 			       gro_result_t ret)
3852 {
3853 	switch (ret) {
3854 	case GRO_NORMAL:
3855 	case GRO_HELD:
3856 		skb->protocol = eth_type_trans(skb, skb->dev);
3857 
3858 		if (ret == GRO_HELD)
3859 			skb_gro_pull(skb, -ETH_HLEN);
3860 		else if (netif_receive_skb(skb))
3861 			ret = GRO_DROP;
3862 		break;
3863 
3864 	case GRO_DROP:
3865 	case GRO_MERGED_FREE:
3866 		napi_reuse_skb(napi, skb);
3867 		break;
3868 
3869 	case GRO_MERGED:
3870 		break;
3871 	}
3872 
3873 	return ret;
3874 }
3875 
3876 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3877 {
3878 	struct sk_buff *skb = napi->skb;
3879 	struct ethhdr *eth;
3880 	unsigned int hlen;
3881 	unsigned int off;
3882 
3883 	napi->skb = NULL;
3884 
3885 	skb_reset_mac_header(skb);
3886 	skb_gro_reset_offset(skb);
3887 
3888 	off = skb_gro_offset(skb);
3889 	hlen = off + sizeof(*eth);
3890 	eth = skb_gro_header_fast(skb, off);
3891 	if (skb_gro_header_hard(skb, hlen)) {
3892 		eth = skb_gro_header_slow(skb, hlen, off);
3893 		if (unlikely(!eth)) {
3894 			napi_reuse_skb(napi, skb);
3895 			skb = NULL;
3896 			goto out;
3897 		}
3898 	}
3899 
3900 	skb_gro_pull(skb, sizeof(*eth));
3901 
3902 	/*
3903 	 * This works because the only protocols we care about don't require
3904 	 * special handling.  We'll fix it up properly at the end.
3905 	 */
3906 	skb->protocol = eth->h_proto;
3907 
3908 out:
3909 	return skb;
3910 }
3911 
3912 gro_result_t napi_gro_frags(struct napi_struct *napi)
3913 {
3914 	struct sk_buff *skb = napi_frags_skb(napi);
3915 
3916 	if (!skb)
3917 		return GRO_DROP;
3918 
3919 	return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
3920 }
3921 EXPORT_SYMBOL(napi_gro_frags);
3922 
3923 /*
3924  * net_rps_action sends any pending IPI's for rps.
3925  * Note: called with local irq disabled, but exits with local irq enabled.
3926  */
3927 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3928 {
3929 #ifdef CONFIG_RPS
3930 	struct softnet_data *remsd = sd->rps_ipi_list;
3931 
3932 	if (remsd) {
3933 		sd->rps_ipi_list = NULL;
3934 
3935 		local_irq_enable();
3936 
3937 		/* Send pending IPI's to kick RPS processing on remote cpus. */
3938 		while (remsd) {
3939 			struct softnet_data *next = remsd->rps_ipi_next;
3940 
3941 			if (cpu_online(remsd->cpu))
3942 				__smp_call_function_single(remsd->cpu,
3943 							   &remsd->csd, 0);
3944 			remsd = next;
3945 		}
3946 	} else
3947 #endif
3948 		local_irq_enable();
3949 }
3950 
3951 static int process_backlog(struct napi_struct *napi, int quota)
3952 {
3953 	int work = 0;
3954 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3955 
3956 #ifdef CONFIG_RPS
3957 	/* Check if we have pending ipi, its better to send them now,
3958 	 * not waiting net_rx_action() end.
3959 	 */
3960 	if (sd->rps_ipi_list) {
3961 		local_irq_disable();
3962 		net_rps_action_and_irq_enable(sd);
3963 	}
3964 #endif
3965 	napi->weight = weight_p;
3966 	local_irq_disable();
3967 	while (work < quota) {
3968 		struct sk_buff *skb;
3969 		unsigned int qlen;
3970 
3971 		while ((skb = __skb_dequeue(&sd->process_queue))) {
3972 			local_irq_enable();
3973 			__netif_receive_skb(skb);
3974 			local_irq_disable();
3975 			input_queue_head_incr(sd);
3976 			if (++work >= quota) {
3977 				local_irq_enable();
3978 				return work;
3979 			}
3980 		}
3981 
3982 		rps_lock(sd);
3983 		qlen = skb_queue_len(&sd->input_pkt_queue);
3984 		if (qlen)
3985 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
3986 						   &sd->process_queue);
3987 
3988 		if (qlen < quota - work) {
3989 			/*
3990 			 * Inline a custom version of __napi_complete().
3991 			 * only current cpu owns and manipulates this napi,
3992 			 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3993 			 * we can use a plain write instead of clear_bit(),
3994 			 * and we dont need an smp_mb() memory barrier.
3995 			 */
3996 			list_del(&napi->poll_list);
3997 			napi->state = 0;
3998 
3999 			quota = work + qlen;
4000 		}
4001 		rps_unlock(sd);
4002 	}
4003 	local_irq_enable();
4004 
4005 	return work;
4006 }
4007 
4008 /**
4009  * __napi_schedule - schedule for receive
4010  * @n: entry to schedule
4011  *
4012  * The entry's receive function will be scheduled to run
4013  */
4014 void __napi_schedule(struct napi_struct *n)
4015 {
4016 	unsigned long flags;
4017 
4018 	local_irq_save(flags);
4019 	____napi_schedule(&__get_cpu_var(softnet_data), n);
4020 	local_irq_restore(flags);
4021 }
4022 EXPORT_SYMBOL(__napi_schedule);
4023 
4024 void __napi_complete(struct napi_struct *n)
4025 {
4026 	BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4027 	BUG_ON(n->gro_list);
4028 
4029 	list_del(&n->poll_list);
4030 	smp_mb__before_clear_bit();
4031 	clear_bit(NAPI_STATE_SCHED, &n->state);
4032 }
4033 EXPORT_SYMBOL(__napi_complete);
4034 
4035 void napi_complete(struct napi_struct *n)
4036 {
4037 	unsigned long flags;
4038 
4039 	/*
4040 	 * don't let napi dequeue from the cpu poll list
4041 	 * just in case its running on a different cpu
4042 	 */
4043 	if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4044 		return;
4045 
4046 	napi_gro_flush(n, false);
4047 	local_irq_save(flags);
4048 	__napi_complete(n);
4049 	local_irq_restore(flags);
4050 }
4051 EXPORT_SYMBOL(napi_complete);
4052 
4053 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4054 		    int (*poll)(struct napi_struct *, int), int weight)
4055 {
4056 	INIT_LIST_HEAD(&napi->poll_list);
4057 	napi->gro_count = 0;
4058 	napi->gro_list = NULL;
4059 	napi->skb = NULL;
4060 	napi->poll = poll;
4061 	napi->weight = weight;
4062 	list_add(&napi->dev_list, &dev->napi_list);
4063 	napi->dev = dev;
4064 #ifdef CONFIG_NETPOLL
4065 	spin_lock_init(&napi->poll_lock);
4066 	napi->poll_owner = -1;
4067 #endif
4068 	set_bit(NAPI_STATE_SCHED, &napi->state);
4069 }
4070 EXPORT_SYMBOL(netif_napi_add);
4071 
4072 void netif_napi_del(struct napi_struct *napi)
4073 {
4074 	struct sk_buff *skb, *next;
4075 
4076 	list_del_init(&napi->dev_list);
4077 	napi_free_frags(napi);
4078 
4079 	for (skb = napi->gro_list; skb; skb = next) {
4080 		next = skb->next;
4081 		skb->next = NULL;
4082 		kfree_skb(skb);
4083 	}
4084 
4085 	napi->gro_list = NULL;
4086 	napi->gro_count = 0;
4087 }
4088 EXPORT_SYMBOL(netif_napi_del);
4089 
4090 static void net_rx_action(struct softirq_action *h)
4091 {
4092 	struct softnet_data *sd = &__get_cpu_var(softnet_data);
4093 	unsigned long time_limit = jiffies + 2;
4094 	int budget = netdev_budget;
4095 	void *have;
4096 
4097 	local_irq_disable();
4098 
4099 	while (!list_empty(&sd->poll_list)) {
4100 		struct napi_struct *n;
4101 		int work, weight;
4102 
4103 		/* If softirq window is exhuasted then punt.
4104 		 * Allow this to run for 2 jiffies since which will allow
4105 		 * an average latency of 1.5/HZ.
4106 		 */
4107 		if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4108 			goto softnet_break;
4109 
4110 		local_irq_enable();
4111 
4112 		/* Even though interrupts have been re-enabled, this
4113 		 * access is safe because interrupts can only add new
4114 		 * entries to the tail of this list, and only ->poll()
4115 		 * calls can remove this head entry from the list.
4116 		 */
4117 		n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4118 
4119 		have = netpoll_poll_lock(n);
4120 
4121 		weight = n->weight;
4122 
4123 		/* This NAPI_STATE_SCHED test is for avoiding a race
4124 		 * with netpoll's poll_napi().  Only the entity which
4125 		 * obtains the lock and sees NAPI_STATE_SCHED set will
4126 		 * actually make the ->poll() call.  Therefore we avoid
4127 		 * accidentally calling ->poll() when NAPI is not scheduled.
4128 		 */
4129 		work = 0;
4130 		if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4131 			work = n->poll(n, weight);
4132 			trace_napi_poll(n);
4133 		}
4134 
4135 		WARN_ON_ONCE(work > weight);
4136 
4137 		budget -= work;
4138 
4139 		local_irq_disable();
4140 
4141 		/* Drivers must not modify the NAPI state if they
4142 		 * consume the entire weight.  In such cases this code
4143 		 * still "owns" the NAPI instance and therefore can
4144 		 * move the instance around on the list at-will.
4145 		 */
4146 		if (unlikely(work == weight)) {
4147 			if (unlikely(napi_disable_pending(n))) {
4148 				local_irq_enable();
4149 				napi_complete(n);
4150 				local_irq_disable();
4151 			} else {
4152 				if (n->gro_list) {
4153 					/* flush too old packets
4154 					 * If HZ < 1000, flush all packets.
4155 					 */
4156 					local_irq_enable();
4157 					napi_gro_flush(n, HZ >= 1000);
4158 					local_irq_disable();
4159 				}
4160 				list_move_tail(&n->poll_list, &sd->poll_list);
4161 			}
4162 		}
4163 
4164 		netpoll_poll_unlock(have);
4165 	}
4166 out:
4167 	net_rps_action_and_irq_enable(sd);
4168 
4169 #ifdef CONFIG_NET_DMA
4170 	/*
4171 	 * There may not be any more sk_buffs coming right now, so push
4172 	 * any pending DMA copies to hardware
4173 	 */
4174 	dma_issue_pending_all();
4175 #endif
4176 
4177 	return;
4178 
4179 softnet_break:
4180 	sd->time_squeeze++;
4181 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4182 	goto out;
4183 }
4184 
4185 struct netdev_upper {
4186 	struct net_device *dev;
4187 	bool master;
4188 	struct list_head list;
4189 	struct rcu_head rcu;
4190 	struct list_head search_list;
4191 };
4192 
4193 static void __append_search_uppers(struct list_head *search_list,
4194 				   struct net_device *dev)
4195 {
4196 	struct netdev_upper *upper;
4197 
4198 	list_for_each_entry(upper, &dev->upper_dev_list, list) {
4199 		/* check if this upper is not already in search list */
4200 		if (list_empty(&upper->search_list))
4201 			list_add_tail(&upper->search_list, search_list);
4202 	}
4203 }
4204 
4205 static bool __netdev_search_upper_dev(struct net_device *dev,
4206 				      struct net_device *upper_dev)
4207 {
4208 	LIST_HEAD(search_list);
4209 	struct netdev_upper *upper;
4210 	struct netdev_upper *tmp;
4211 	bool ret = false;
4212 
4213 	__append_search_uppers(&search_list, dev);
4214 	list_for_each_entry(upper, &search_list, search_list) {
4215 		if (upper->dev == upper_dev) {
4216 			ret = true;
4217 			break;
4218 		}
4219 		__append_search_uppers(&search_list, upper->dev);
4220 	}
4221 	list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4222 		INIT_LIST_HEAD(&upper->search_list);
4223 	return ret;
4224 }
4225 
4226 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4227 						struct net_device *upper_dev)
4228 {
4229 	struct netdev_upper *upper;
4230 
4231 	list_for_each_entry(upper, &dev->upper_dev_list, list) {
4232 		if (upper->dev == upper_dev)
4233 			return upper;
4234 	}
4235 	return NULL;
4236 }
4237 
4238 /**
4239  * netdev_has_upper_dev - Check if device is linked to an upper device
4240  * @dev: device
4241  * @upper_dev: upper device to check
4242  *
4243  * Find out if a device is linked to specified upper device and return true
4244  * in case it is. Note that this checks only immediate upper device,
4245  * not through a complete stack of devices. The caller must hold the RTNL lock.
4246  */
4247 bool netdev_has_upper_dev(struct net_device *dev,
4248 			  struct net_device *upper_dev)
4249 {
4250 	ASSERT_RTNL();
4251 
4252 	return __netdev_find_upper(dev, upper_dev);
4253 }
4254 EXPORT_SYMBOL(netdev_has_upper_dev);
4255 
4256 /**
4257  * netdev_has_any_upper_dev - Check if device is linked to some device
4258  * @dev: device
4259  *
4260  * Find out if a device is linked to an upper device and return true in case
4261  * it is. The caller must hold the RTNL lock.
4262  */
4263 bool netdev_has_any_upper_dev(struct net_device *dev)
4264 {
4265 	ASSERT_RTNL();
4266 
4267 	return !list_empty(&dev->upper_dev_list);
4268 }
4269 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4270 
4271 /**
4272  * netdev_master_upper_dev_get - Get master upper device
4273  * @dev: device
4274  *
4275  * Find a master upper device and return pointer to it or NULL in case
4276  * it's not there. The caller must hold the RTNL lock.
4277  */
4278 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4279 {
4280 	struct netdev_upper *upper;
4281 
4282 	ASSERT_RTNL();
4283 
4284 	if (list_empty(&dev->upper_dev_list))
4285 		return NULL;
4286 
4287 	upper = list_first_entry(&dev->upper_dev_list,
4288 				 struct netdev_upper, list);
4289 	if (likely(upper->master))
4290 		return upper->dev;
4291 	return NULL;
4292 }
4293 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4294 
4295 /**
4296  * netdev_master_upper_dev_get_rcu - Get master upper device
4297  * @dev: device
4298  *
4299  * Find a master upper device and return pointer to it or NULL in case
4300  * it's not there. The caller must hold the RCU read lock.
4301  */
4302 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4303 {
4304 	struct netdev_upper *upper;
4305 
4306 	upper = list_first_or_null_rcu(&dev->upper_dev_list,
4307 				       struct netdev_upper, list);
4308 	if (upper && likely(upper->master))
4309 		return upper->dev;
4310 	return NULL;
4311 }
4312 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4313 
4314 static int __netdev_upper_dev_link(struct net_device *dev,
4315 				   struct net_device *upper_dev, bool master)
4316 {
4317 	struct netdev_upper *upper;
4318 
4319 	ASSERT_RTNL();
4320 
4321 	if (dev == upper_dev)
4322 		return -EBUSY;
4323 
4324 	/* To prevent loops, check if dev is not upper device to upper_dev. */
4325 	if (__netdev_search_upper_dev(upper_dev, dev))
4326 		return -EBUSY;
4327 
4328 	if (__netdev_find_upper(dev, upper_dev))
4329 		return -EEXIST;
4330 
4331 	if (master && netdev_master_upper_dev_get(dev))
4332 		return -EBUSY;
4333 
4334 	upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4335 	if (!upper)
4336 		return -ENOMEM;
4337 
4338 	upper->dev = upper_dev;
4339 	upper->master = master;
4340 	INIT_LIST_HEAD(&upper->search_list);
4341 
4342 	/* Ensure that master upper link is always the first item in list. */
4343 	if (master)
4344 		list_add_rcu(&upper->list, &dev->upper_dev_list);
4345 	else
4346 		list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4347 	dev_hold(upper_dev);
4348 
4349 	return 0;
4350 }
4351 
4352 /**
4353  * netdev_upper_dev_link - Add a link to the upper device
4354  * @dev: device
4355  * @upper_dev: new upper device
4356  *
4357  * Adds a link to device which is upper to this one. The caller must hold
4358  * the RTNL lock. On a failure a negative errno code is returned.
4359  * On success the reference counts are adjusted and the function
4360  * returns zero.
4361  */
4362 int netdev_upper_dev_link(struct net_device *dev,
4363 			  struct net_device *upper_dev)
4364 {
4365 	return __netdev_upper_dev_link(dev, upper_dev, false);
4366 }
4367 EXPORT_SYMBOL(netdev_upper_dev_link);
4368 
4369 /**
4370  * netdev_master_upper_dev_link - Add a master link to the upper device
4371  * @dev: device
4372  * @upper_dev: new upper device
4373  *
4374  * Adds a link to device which is upper to this one. In this case, only
4375  * one master upper device can be linked, although other non-master devices
4376  * might be linked as well. The caller must hold the RTNL lock.
4377  * On a failure a negative errno code is returned. On success the reference
4378  * counts are adjusted and the function returns zero.
4379  */
4380 int netdev_master_upper_dev_link(struct net_device *dev,
4381 				 struct net_device *upper_dev)
4382 {
4383 	return __netdev_upper_dev_link(dev, upper_dev, true);
4384 }
4385 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4386 
4387 /**
4388  * netdev_upper_dev_unlink - Removes a link to upper device
4389  * @dev: device
4390  * @upper_dev: new upper device
4391  *
4392  * Removes a link to device which is upper to this one. The caller must hold
4393  * the RTNL lock.
4394  */
4395 void netdev_upper_dev_unlink(struct net_device *dev,
4396 			     struct net_device *upper_dev)
4397 {
4398 	struct netdev_upper *upper;
4399 
4400 	ASSERT_RTNL();
4401 
4402 	upper = __netdev_find_upper(dev, upper_dev);
4403 	if (!upper)
4404 		return;
4405 	list_del_rcu(&upper->list);
4406 	dev_put(upper_dev);
4407 	kfree_rcu(upper, rcu);
4408 }
4409 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4410 
4411 static void dev_change_rx_flags(struct net_device *dev, int flags)
4412 {
4413 	const struct net_device_ops *ops = dev->netdev_ops;
4414 
4415 	if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4416 		ops->ndo_change_rx_flags(dev, flags);
4417 }
4418 
4419 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4420 {
4421 	unsigned int old_flags = dev->flags;
4422 	kuid_t uid;
4423 	kgid_t gid;
4424 
4425 	ASSERT_RTNL();
4426 
4427 	dev->flags |= IFF_PROMISC;
4428 	dev->promiscuity += inc;
4429 	if (dev->promiscuity == 0) {
4430 		/*
4431 		 * Avoid overflow.
4432 		 * If inc causes overflow, untouch promisc and return error.
4433 		 */
4434 		if (inc < 0)
4435 			dev->flags &= ~IFF_PROMISC;
4436 		else {
4437 			dev->promiscuity -= inc;
4438 			pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4439 				dev->name);
4440 			return -EOVERFLOW;
4441 		}
4442 	}
4443 	if (dev->flags != old_flags) {
4444 		pr_info("device %s %s promiscuous mode\n",
4445 			dev->name,
4446 			dev->flags & IFF_PROMISC ? "entered" : "left");
4447 		if (audit_enabled) {
4448 			current_uid_gid(&uid, &gid);
4449 			audit_log(current->audit_context, GFP_ATOMIC,
4450 				AUDIT_ANOM_PROMISCUOUS,
4451 				"dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4452 				dev->name, (dev->flags & IFF_PROMISC),
4453 				(old_flags & IFF_PROMISC),
4454 				from_kuid(&init_user_ns, audit_get_loginuid(current)),
4455 				from_kuid(&init_user_ns, uid),
4456 				from_kgid(&init_user_ns, gid),
4457 				audit_get_sessionid(current));
4458 		}
4459 
4460 		dev_change_rx_flags(dev, IFF_PROMISC);
4461 	}
4462 	return 0;
4463 }
4464 
4465 /**
4466  *	dev_set_promiscuity	- update promiscuity count on a device
4467  *	@dev: device
4468  *	@inc: modifier
4469  *
4470  *	Add or remove promiscuity from a device. While the count in the device
4471  *	remains above zero the interface remains promiscuous. Once it hits zero
4472  *	the device reverts back to normal filtering operation. A negative inc
4473  *	value is used to drop promiscuity on the device.
4474  *	Return 0 if successful or a negative errno code on error.
4475  */
4476 int dev_set_promiscuity(struct net_device *dev, int inc)
4477 {
4478 	unsigned int old_flags = dev->flags;
4479 	int err;
4480 
4481 	err = __dev_set_promiscuity(dev, inc);
4482 	if (err < 0)
4483 		return err;
4484 	if (dev->flags != old_flags)
4485 		dev_set_rx_mode(dev);
4486 	return err;
4487 }
4488 EXPORT_SYMBOL(dev_set_promiscuity);
4489 
4490 /**
4491  *	dev_set_allmulti	- update allmulti count on a device
4492  *	@dev: device
4493  *	@inc: modifier
4494  *
4495  *	Add or remove reception of all multicast frames to a device. While the
4496  *	count in the device remains above zero the interface remains listening
4497  *	to all interfaces. Once it hits zero the device reverts back to normal
4498  *	filtering operation. A negative @inc value is used to drop the counter
4499  *	when releasing a resource needing all multicasts.
4500  *	Return 0 if successful or a negative errno code on error.
4501  */
4502 
4503 int dev_set_allmulti(struct net_device *dev, int inc)
4504 {
4505 	unsigned int old_flags = dev->flags;
4506 
4507 	ASSERT_RTNL();
4508 
4509 	dev->flags |= IFF_ALLMULTI;
4510 	dev->allmulti += inc;
4511 	if (dev->allmulti == 0) {
4512 		/*
4513 		 * Avoid overflow.
4514 		 * If inc causes overflow, untouch allmulti and return error.
4515 		 */
4516 		if (inc < 0)
4517 			dev->flags &= ~IFF_ALLMULTI;
4518 		else {
4519 			dev->allmulti -= inc;
4520 			pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4521 				dev->name);
4522 			return -EOVERFLOW;
4523 		}
4524 	}
4525 	if (dev->flags ^ old_flags) {
4526 		dev_change_rx_flags(dev, IFF_ALLMULTI);
4527 		dev_set_rx_mode(dev);
4528 	}
4529 	return 0;
4530 }
4531 EXPORT_SYMBOL(dev_set_allmulti);
4532 
4533 /*
4534  *	Upload unicast and multicast address lists to device and
4535  *	configure RX filtering. When the device doesn't support unicast
4536  *	filtering it is put in promiscuous mode while unicast addresses
4537  *	are present.
4538  */
4539 void __dev_set_rx_mode(struct net_device *dev)
4540 {
4541 	const struct net_device_ops *ops = dev->netdev_ops;
4542 
4543 	/* dev_open will call this function so the list will stay sane. */
4544 	if (!(dev->flags&IFF_UP))
4545 		return;
4546 
4547 	if (!netif_device_present(dev))
4548 		return;
4549 
4550 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4551 		/* Unicast addresses changes may only happen under the rtnl,
4552 		 * therefore calling __dev_set_promiscuity here is safe.
4553 		 */
4554 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4555 			__dev_set_promiscuity(dev, 1);
4556 			dev->uc_promisc = true;
4557 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4558 			__dev_set_promiscuity(dev, -1);
4559 			dev->uc_promisc = false;
4560 		}
4561 	}
4562 
4563 	if (ops->ndo_set_rx_mode)
4564 		ops->ndo_set_rx_mode(dev);
4565 }
4566 
4567 void dev_set_rx_mode(struct net_device *dev)
4568 {
4569 	netif_addr_lock_bh(dev);
4570 	__dev_set_rx_mode(dev);
4571 	netif_addr_unlock_bh(dev);
4572 }
4573 
4574 /**
4575  *	dev_get_flags - get flags reported to userspace
4576  *	@dev: device
4577  *
4578  *	Get the combination of flag bits exported through APIs to userspace.
4579  */
4580 unsigned int dev_get_flags(const struct net_device *dev)
4581 {
4582 	unsigned int flags;
4583 
4584 	flags = (dev->flags & ~(IFF_PROMISC |
4585 				IFF_ALLMULTI |
4586 				IFF_RUNNING |
4587 				IFF_LOWER_UP |
4588 				IFF_DORMANT)) |
4589 		(dev->gflags & (IFF_PROMISC |
4590 				IFF_ALLMULTI));
4591 
4592 	if (netif_running(dev)) {
4593 		if (netif_oper_up(dev))
4594 			flags |= IFF_RUNNING;
4595 		if (netif_carrier_ok(dev))
4596 			flags |= IFF_LOWER_UP;
4597 		if (netif_dormant(dev))
4598 			flags |= IFF_DORMANT;
4599 	}
4600 
4601 	return flags;
4602 }
4603 EXPORT_SYMBOL(dev_get_flags);
4604 
4605 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4606 {
4607 	unsigned int old_flags = dev->flags;
4608 	int ret;
4609 
4610 	ASSERT_RTNL();
4611 
4612 	/*
4613 	 *	Set the flags on our device.
4614 	 */
4615 
4616 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4617 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4618 			       IFF_AUTOMEDIA)) |
4619 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4620 				    IFF_ALLMULTI));
4621 
4622 	/*
4623 	 *	Load in the correct multicast list now the flags have changed.
4624 	 */
4625 
4626 	if ((old_flags ^ flags) & IFF_MULTICAST)
4627 		dev_change_rx_flags(dev, IFF_MULTICAST);
4628 
4629 	dev_set_rx_mode(dev);
4630 
4631 	/*
4632 	 *	Have we downed the interface. We handle IFF_UP ourselves
4633 	 *	according to user attempts to set it, rather than blindly
4634 	 *	setting it.
4635 	 */
4636 
4637 	ret = 0;
4638 	if ((old_flags ^ flags) & IFF_UP) {	/* Bit is different  ? */
4639 		ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4640 
4641 		if (!ret)
4642 			dev_set_rx_mode(dev);
4643 	}
4644 
4645 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
4646 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
4647 
4648 		dev->gflags ^= IFF_PROMISC;
4649 		dev_set_promiscuity(dev, inc);
4650 	}
4651 
4652 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4653 	   is important. Some (broken) drivers set IFF_PROMISC, when
4654 	   IFF_ALLMULTI is requested not asking us and not reporting.
4655 	 */
4656 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4657 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4658 
4659 		dev->gflags ^= IFF_ALLMULTI;
4660 		dev_set_allmulti(dev, inc);
4661 	}
4662 
4663 	return ret;
4664 }
4665 
4666 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4667 {
4668 	unsigned int changes = dev->flags ^ old_flags;
4669 
4670 	if (changes & IFF_UP) {
4671 		if (dev->flags & IFF_UP)
4672 			call_netdevice_notifiers(NETDEV_UP, dev);
4673 		else
4674 			call_netdevice_notifiers(NETDEV_DOWN, dev);
4675 	}
4676 
4677 	if (dev->flags & IFF_UP &&
4678 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4679 		call_netdevice_notifiers(NETDEV_CHANGE, dev);
4680 }
4681 
4682 /**
4683  *	dev_change_flags - change device settings
4684  *	@dev: device
4685  *	@flags: device state flags
4686  *
4687  *	Change settings on device based state flags. The flags are
4688  *	in the userspace exported format.
4689  */
4690 int dev_change_flags(struct net_device *dev, unsigned int flags)
4691 {
4692 	int ret;
4693 	unsigned int changes, old_flags = dev->flags;
4694 
4695 	ret = __dev_change_flags(dev, flags);
4696 	if (ret < 0)
4697 		return ret;
4698 
4699 	changes = old_flags ^ dev->flags;
4700 	if (changes)
4701 		rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4702 
4703 	__dev_notify_flags(dev, old_flags);
4704 	return ret;
4705 }
4706 EXPORT_SYMBOL(dev_change_flags);
4707 
4708 /**
4709  *	dev_set_mtu - Change maximum transfer unit
4710  *	@dev: device
4711  *	@new_mtu: new transfer unit
4712  *
4713  *	Change the maximum transfer size of the network device.
4714  */
4715 int dev_set_mtu(struct net_device *dev, int new_mtu)
4716 {
4717 	const struct net_device_ops *ops = dev->netdev_ops;
4718 	int err;
4719 
4720 	if (new_mtu == dev->mtu)
4721 		return 0;
4722 
4723 	/*	MTU must be positive.	 */
4724 	if (new_mtu < 0)
4725 		return -EINVAL;
4726 
4727 	if (!netif_device_present(dev))
4728 		return -ENODEV;
4729 
4730 	err = 0;
4731 	if (ops->ndo_change_mtu)
4732 		err = ops->ndo_change_mtu(dev, new_mtu);
4733 	else
4734 		dev->mtu = new_mtu;
4735 
4736 	if (!err)
4737 		call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4738 	return err;
4739 }
4740 EXPORT_SYMBOL(dev_set_mtu);
4741 
4742 /**
4743  *	dev_set_group - Change group this device belongs to
4744  *	@dev: device
4745  *	@new_group: group this device should belong to
4746  */
4747 void dev_set_group(struct net_device *dev, int new_group)
4748 {
4749 	dev->group = new_group;
4750 }
4751 EXPORT_SYMBOL(dev_set_group);
4752 
4753 /**
4754  *	dev_set_mac_address - Change Media Access Control Address
4755  *	@dev: device
4756  *	@sa: new address
4757  *
4758  *	Change the hardware (MAC) address of the device
4759  */
4760 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4761 {
4762 	const struct net_device_ops *ops = dev->netdev_ops;
4763 	int err;
4764 
4765 	if (!ops->ndo_set_mac_address)
4766 		return -EOPNOTSUPP;
4767 	if (sa->sa_family != dev->type)
4768 		return -EINVAL;
4769 	if (!netif_device_present(dev))
4770 		return -ENODEV;
4771 	err = ops->ndo_set_mac_address(dev, sa);
4772 	if (err)
4773 		return err;
4774 	dev->addr_assign_type = NET_ADDR_SET;
4775 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4776 	add_device_randomness(dev->dev_addr, dev->addr_len);
4777 	return 0;
4778 }
4779 EXPORT_SYMBOL(dev_set_mac_address);
4780 
4781 /**
4782  *	dev_change_carrier - Change device carrier
4783  *	@dev: device
4784  *	@new_carrier: new value
4785  *
4786  *	Change device carrier
4787  */
4788 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4789 {
4790 	const struct net_device_ops *ops = dev->netdev_ops;
4791 
4792 	if (!ops->ndo_change_carrier)
4793 		return -EOPNOTSUPP;
4794 	if (!netif_device_present(dev))
4795 		return -ENODEV;
4796 	return ops->ndo_change_carrier(dev, new_carrier);
4797 }
4798 EXPORT_SYMBOL(dev_change_carrier);
4799 
4800 /**
4801  *	dev_new_index	-	allocate an ifindex
4802  *	@net: the applicable net namespace
4803  *
4804  *	Returns a suitable unique value for a new device interface
4805  *	number.  The caller must hold the rtnl semaphore or the
4806  *	dev_base_lock to be sure it remains unique.
4807  */
4808 static int dev_new_index(struct net *net)
4809 {
4810 	int ifindex = net->ifindex;
4811 	for (;;) {
4812 		if (++ifindex <= 0)
4813 			ifindex = 1;
4814 		if (!__dev_get_by_index(net, ifindex))
4815 			return net->ifindex = ifindex;
4816 	}
4817 }
4818 
4819 /* Delayed registration/unregisteration */
4820 static LIST_HEAD(net_todo_list);
4821 
4822 static void net_set_todo(struct net_device *dev)
4823 {
4824 	list_add_tail(&dev->todo_list, &net_todo_list);
4825 }
4826 
4827 static void rollback_registered_many(struct list_head *head)
4828 {
4829 	struct net_device *dev, *tmp;
4830 
4831 	BUG_ON(dev_boot_phase);
4832 	ASSERT_RTNL();
4833 
4834 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4835 		/* Some devices call without registering
4836 		 * for initialization unwind. Remove those
4837 		 * devices and proceed with the remaining.
4838 		 */
4839 		if (dev->reg_state == NETREG_UNINITIALIZED) {
4840 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4841 				 dev->name, dev);
4842 
4843 			WARN_ON(1);
4844 			list_del(&dev->unreg_list);
4845 			continue;
4846 		}
4847 		dev->dismantle = true;
4848 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
4849 	}
4850 
4851 	/* If device is running, close it first. */
4852 	dev_close_many(head);
4853 
4854 	list_for_each_entry(dev, head, unreg_list) {
4855 		/* And unlink it from device chain. */
4856 		unlist_netdevice(dev);
4857 
4858 		dev->reg_state = NETREG_UNREGISTERING;
4859 	}
4860 
4861 	synchronize_net();
4862 
4863 	list_for_each_entry(dev, head, unreg_list) {
4864 		/* Shutdown queueing discipline. */
4865 		dev_shutdown(dev);
4866 
4867 
4868 		/* Notify protocols, that we are about to destroy
4869 		   this device. They should clean all the things.
4870 		*/
4871 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4872 
4873 		if (!dev->rtnl_link_ops ||
4874 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4875 			rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4876 
4877 		/*
4878 		 *	Flush the unicast and multicast chains
4879 		 */
4880 		dev_uc_flush(dev);
4881 		dev_mc_flush(dev);
4882 
4883 		if (dev->netdev_ops->ndo_uninit)
4884 			dev->netdev_ops->ndo_uninit(dev);
4885 
4886 		/* Notifier chain MUST detach us all upper devices. */
4887 		WARN_ON(netdev_has_any_upper_dev(dev));
4888 
4889 		/* Remove entries from kobject tree */
4890 		netdev_unregister_kobject(dev);
4891 #ifdef CONFIG_XPS
4892 		/* Remove XPS queueing entries */
4893 		netif_reset_xps_queues_gt(dev, 0);
4894 #endif
4895 	}
4896 
4897 	synchronize_net();
4898 
4899 	list_for_each_entry(dev, head, unreg_list)
4900 		dev_put(dev);
4901 }
4902 
4903 static void rollback_registered(struct net_device *dev)
4904 {
4905 	LIST_HEAD(single);
4906 
4907 	list_add(&dev->unreg_list, &single);
4908 	rollback_registered_many(&single);
4909 	list_del(&single);
4910 }
4911 
4912 static netdev_features_t netdev_fix_features(struct net_device *dev,
4913 	netdev_features_t features)
4914 {
4915 	/* Fix illegal checksum combinations */
4916 	if ((features & NETIF_F_HW_CSUM) &&
4917 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4918 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
4919 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4920 	}
4921 
4922 	/* Fix illegal SG+CSUM combinations. */
4923 	if ((features & NETIF_F_SG) &&
4924 	    !(features & NETIF_F_ALL_CSUM)) {
4925 		netdev_dbg(dev,
4926 			"Dropping NETIF_F_SG since no checksum feature.\n");
4927 		features &= ~NETIF_F_SG;
4928 	}
4929 
4930 	/* TSO requires that SG is present as well. */
4931 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
4932 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
4933 		features &= ~NETIF_F_ALL_TSO;
4934 	}
4935 
4936 	/* TSO ECN requires that TSO is present as well. */
4937 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
4938 		features &= ~NETIF_F_TSO_ECN;
4939 
4940 	/* Software GSO depends on SG. */
4941 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
4942 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
4943 		features &= ~NETIF_F_GSO;
4944 	}
4945 
4946 	/* UFO needs SG and checksumming */
4947 	if (features & NETIF_F_UFO) {
4948 		/* maybe split UFO into V4 and V6? */
4949 		if (!((features & NETIF_F_GEN_CSUM) ||
4950 		    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
4951 			    == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4952 			netdev_dbg(dev,
4953 				"Dropping NETIF_F_UFO since no checksum offload features.\n");
4954 			features &= ~NETIF_F_UFO;
4955 		}
4956 
4957 		if (!(features & NETIF_F_SG)) {
4958 			netdev_dbg(dev,
4959 				"Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
4960 			features &= ~NETIF_F_UFO;
4961 		}
4962 	}
4963 
4964 	return features;
4965 }
4966 
4967 int __netdev_update_features(struct net_device *dev)
4968 {
4969 	netdev_features_t features;
4970 	int err = 0;
4971 
4972 	ASSERT_RTNL();
4973 
4974 	features = netdev_get_wanted_features(dev);
4975 
4976 	if (dev->netdev_ops->ndo_fix_features)
4977 		features = dev->netdev_ops->ndo_fix_features(dev, features);
4978 
4979 	/* driver might be less strict about feature dependencies */
4980 	features = netdev_fix_features(dev, features);
4981 
4982 	if (dev->features == features)
4983 		return 0;
4984 
4985 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
4986 		&dev->features, &features);
4987 
4988 	if (dev->netdev_ops->ndo_set_features)
4989 		err = dev->netdev_ops->ndo_set_features(dev, features);
4990 
4991 	if (unlikely(err < 0)) {
4992 		netdev_err(dev,
4993 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
4994 			err, &features, &dev->features);
4995 		return -1;
4996 	}
4997 
4998 	if (!err)
4999 		dev->features = features;
5000 
5001 	return 1;
5002 }
5003 
5004 /**
5005  *	netdev_update_features - recalculate device features
5006  *	@dev: the device to check
5007  *
5008  *	Recalculate dev->features set and send notifications if it
5009  *	has changed. Should be called after driver or hardware dependent
5010  *	conditions might have changed that influence the features.
5011  */
5012 void netdev_update_features(struct net_device *dev)
5013 {
5014 	if (__netdev_update_features(dev))
5015 		netdev_features_change(dev);
5016 }
5017 EXPORT_SYMBOL(netdev_update_features);
5018 
5019 /**
5020  *	netdev_change_features - recalculate device features
5021  *	@dev: the device to check
5022  *
5023  *	Recalculate dev->features set and send notifications even
5024  *	if they have not changed. Should be called instead of
5025  *	netdev_update_features() if also dev->vlan_features might
5026  *	have changed to allow the changes to be propagated to stacked
5027  *	VLAN devices.
5028  */
5029 void netdev_change_features(struct net_device *dev)
5030 {
5031 	__netdev_update_features(dev);
5032 	netdev_features_change(dev);
5033 }
5034 EXPORT_SYMBOL(netdev_change_features);
5035 
5036 /**
5037  *	netif_stacked_transfer_operstate -	transfer operstate
5038  *	@rootdev: the root or lower level device to transfer state from
5039  *	@dev: the device to transfer operstate to
5040  *
5041  *	Transfer operational state from root to device. This is normally
5042  *	called when a stacking relationship exists between the root
5043  *	device and the device(a leaf device).
5044  */
5045 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5046 					struct net_device *dev)
5047 {
5048 	if (rootdev->operstate == IF_OPER_DORMANT)
5049 		netif_dormant_on(dev);
5050 	else
5051 		netif_dormant_off(dev);
5052 
5053 	if (netif_carrier_ok(rootdev)) {
5054 		if (!netif_carrier_ok(dev))
5055 			netif_carrier_on(dev);
5056 	} else {
5057 		if (netif_carrier_ok(dev))
5058 			netif_carrier_off(dev);
5059 	}
5060 }
5061 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5062 
5063 #ifdef CONFIG_RPS
5064 static int netif_alloc_rx_queues(struct net_device *dev)
5065 {
5066 	unsigned int i, count = dev->num_rx_queues;
5067 	struct netdev_rx_queue *rx;
5068 
5069 	BUG_ON(count < 1);
5070 
5071 	rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5072 	if (!rx)
5073 		return -ENOMEM;
5074 
5075 	dev->_rx = rx;
5076 
5077 	for (i = 0; i < count; i++)
5078 		rx[i].dev = dev;
5079 	return 0;
5080 }
5081 #endif
5082 
5083 static void netdev_init_one_queue(struct net_device *dev,
5084 				  struct netdev_queue *queue, void *_unused)
5085 {
5086 	/* Initialize queue lock */
5087 	spin_lock_init(&queue->_xmit_lock);
5088 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5089 	queue->xmit_lock_owner = -1;
5090 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5091 	queue->dev = dev;
5092 #ifdef CONFIG_BQL
5093 	dql_init(&queue->dql, HZ);
5094 #endif
5095 }
5096 
5097 static int netif_alloc_netdev_queues(struct net_device *dev)
5098 {
5099 	unsigned int count = dev->num_tx_queues;
5100 	struct netdev_queue *tx;
5101 
5102 	BUG_ON(count < 1);
5103 
5104 	tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5105 	if (!tx)
5106 		return -ENOMEM;
5107 
5108 	dev->_tx = tx;
5109 
5110 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5111 	spin_lock_init(&dev->tx_global_lock);
5112 
5113 	return 0;
5114 }
5115 
5116 /**
5117  *	register_netdevice	- register a network device
5118  *	@dev: device to register
5119  *
5120  *	Take a completed network device structure and add it to the kernel
5121  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5122  *	chain. 0 is returned on success. A negative errno code is returned
5123  *	on a failure to set up the device, or if the name is a duplicate.
5124  *
5125  *	Callers must hold the rtnl semaphore. You may want
5126  *	register_netdev() instead of this.
5127  *
5128  *	BUGS:
5129  *	The locking appears insufficient to guarantee two parallel registers
5130  *	will not get the same name.
5131  */
5132 
5133 int register_netdevice(struct net_device *dev)
5134 {
5135 	int ret;
5136 	struct net *net = dev_net(dev);
5137 
5138 	BUG_ON(dev_boot_phase);
5139 	ASSERT_RTNL();
5140 
5141 	might_sleep();
5142 
5143 	/* When net_device's are persistent, this will be fatal. */
5144 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5145 	BUG_ON(!net);
5146 
5147 	spin_lock_init(&dev->addr_list_lock);
5148 	netdev_set_addr_lockdep_class(dev);
5149 
5150 	dev->iflink = -1;
5151 
5152 	ret = dev_get_valid_name(net, dev, dev->name);
5153 	if (ret < 0)
5154 		goto out;
5155 
5156 	/* Init, if this function is available */
5157 	if (dev->netdev_ops->ndo_init) {
5158 		ret = dev->netdev_ops->ndo_init(dev);
5159 		if (ret) {
5160 			if (ret > 0)
5161 				ret = -EIO;
5162 			goto out;
5163 		}
5164 	}
5165 
5166 	if (((dev->hw_features | dev->features) & NETIF_F_HW_VLAN_FILTER) &&
5167 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5168 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5169 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5170 		ret = -EINVAL;
5171 		goto err_uninit;
5172 	}
5173 
5174 	ret = -EBUSY;
5175 	if (!dev->ifindex)
5176 		dev->ifindex = dev_new_index(net);
5177 	else if (__dev_get_by_index(net, dev->ifindex))
5178 		goto err_uninit;
5179 
5180 	if (dev->iflink == -1)
5181 		dev->iflink = dev->ifindex;
5182 
5183 	/* Transfer changeable features to wanted_features and enable
5184 	 * software offloads (GSO and GRO).
5185 	 */
5186 	dev->hw_features |= NETIF_F_SOFT_FEATURES;
5187 	dev->features |= NETIF_F_SOFT_FEATURES;
5188 	dev->wanted_features = dev->features & dev->hw_features;
5189 
5190 	/* Turn on no cache copy if HW is doing checksum */
5191 	if (!(dev->flags & IFF_LOOPBACK)) {
5192 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
5193 		if (dev->features & NETIF_F_ALL_CSUM) {
5194 			dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5195 			dev->features |= NETIF_F_NOCACHE_COPY;
5196 		}
5197 	}
5198 
5199 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5200 	 */
5201 	dev->vlan_features |= NETIF_F_HIGHDMA;
5202 
5203 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5204 	ret = notifier_to_errno(ret);
5205 	if (ret)
5206 		goto err_uninit;
5207 
5208 	ret = netdev_register_kobject(dev);
5209 	if (ret)
5210 		goto err_uninit;
5211 	dev->reg_state = NETREG_REGISTERED;
5212 
5213 	__netdev_update_features(dev);
5214 
5215 	/*
5216 	 *	Default initial state at registry is that the
5217 	 *	device is present.
5218 	 */
5219 
5220 	set_bit(__LINK_STATE_PRESENT, &dev->state);
5221 
5222 	linkwatch_init_dev(dev);
5223 
5224 	dev_init_scheduler(dev);
5225 	dev_hold(dev);
5226 	list_netdevice(dev);
5227 	add_device_randomness(dev->dev_addr, dev->addr_len);
5228 
5229 	/* If the device has permanent device address, driver should
5230 	 * set dev_addr and also addr_assign_type should be set to
5231 	 * NET_ADDR_PERM (default value).
5232 	 */
5233 	if (dev->addr_assign_type == NET_ADDR_PERM)
5234 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5235 
5236 	/* Notify protocols, that a new device appeared. */
5237 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5238 	ret = notifier_to_errno(ret);
5239 	if (ret) {
5240 		rollback_registered(dev);
5241 		dev->reg_state = NETREG_UNREGISTERED;
5242 	}
5243 	/*
5244 	 *	Prevent userspace races by waiting until the network
5245 	 *	device is fully setup before sending notifications.
5246 	 */
5247 	if (!dev->rtnl_link_ops ||
5248 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5249 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5250 
5251 out:
5252 	return ret;
5253 
5254 err_uninit:
5255 	if (dev->netdev_ops->ndo_uninit)
5256 		dev->netdev_ops->ndo_uninit(dev);
5257 	goto out;
5258 }
5259 EXPORT_SYMBOL(register_netdevice);
5260 
5261 /**
5262  *	init_dummy_netdev	- init a dummy network device for NAPI
5263  *	@dev: device to init
5264  *
5265  *	This takes a network device structure and initialize the minimum
5266  *	amount of fields so it can be used to schedule NAPI polls without
5267  *	registering a full blown interface. This is to be used by drivers
5268  *	that need to tie several hardware interfaces to a single NAPI
5269  *	poll scheduler due to HW limitations.
5270  */
5271 int init_dummy_netdev(struct net_device *dev)
5272 {
5273 	/* Clear everything. Note we don't initialize spinlocks
5274 	 * are they aren't supposed to be taken by any of the
5275 	 * NAPI code and this dummy netdev is supposed to be
5276 	 * only ever used for NAPI polls
5277 	 */
5278 	memset(dev, 0, sizeof(struct net_device));
5279 
5280 	/* make sure we BUG if trying to hit standard
5281 	 * register/unregister code path
5282 	 */
5283 	dev->reg_state = NETREG_DUMMY;
5284 
5285 	/* NAPI wants this */
5286 	INIT_LIST_HEAD(&dev->napi_list);
5287 
5288 	/* a dummy interface is started by default */
5289 	set_bit(__LINK_STATE_PRESENT, &dev->state);
5290 	set_bit(__LINK_STATE_START, &dev->state);
5291 
5292 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
5293 	 * because users of this 'device' dont need to change
5294 	 * its refcount.
5295 	 */
5296 
5297 	return 0;
5298 }
5299 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5300 
5301 
5302 /**
5303  *	register_netdev	- register a network device
5304  *	@dev: device to register
5305  *
5306  *	Take a completed network device structure and add it to the kernel
5307  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5308  *	chain. 0 is returned on success. A negative errno code is returned
5309  *	on a failure to set up the device, or if the name is a duplicate.
5310  *
5311  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
5312  *	and expands the device name if you passed a format string to
5313  *	alloc_netdev.
5314  */
5315 int register_netdev(struct net_device *dev)
5316 {
5317 	int err;
5318 
5319 	rtnl_lock();
5320 	err = register_netdevice(dev);
5321 	rtnl_unlock();
5322 	return err;
5323 }
5324 EXPORT_SYMBOL(register_netdev);
5325 
5326 int netdev_refcnt_read(const struct net_device *dev)
5327 {
5328 	int i, refcnt = 0;
5329 
5330 	for_each_possible_cpu(i)
5331 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5332 	return refcnt;
5333 }
5334 EXPORT_SYMBOL(netdev_refcnt_read);
5335 
5336 /**
5337  * netdev_wait_allrefs - wait until all references are gone.
5338  * @dev: target net_device
5339  *
5340  * This is called when unregistering network devices.
5341  *
5342  * Any protocol or device that holds a reference should register
5343  * for netdevice notification, and cleanup and put back the
5344  * reference if they receive an UNREGISTER event.
5345  * We can get stuck here if buggy protocols don't correctly
5346  * call dev_put.
5347  */
5348 static void netdev_wait_allrefs(struct net_device *dev)
5349 {
5350 	unsigned long rebroadcast_time, warning_time;
5351 	int refcnt;
5352 
5353 	linkwatch_forget_dev(dev);
5354 
5355 	rebroadcast_time = warning_time = jiffies;
5356 	refcnt = netdev_refcnt_read(dev);
5357 
5358 	while (refcnt != 0) {
5359 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5360 			rtnl_lock();
5361 
5362 			/* Rebroadcast unregister notification */
5363 			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5364 
5365 			__rtnl_unlock();
5366 			rcu_barrier();
5367 			rtnl_lock();
5368 
5369 			call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5370 			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5371 				     &dev->state)) {
5372 				/* We must not have linkwatch events
5373 				 * pending on unregister. If this
5374 				 * happens, we simply run the queue
5375 				 * unscheduled, resulting in a noop
5376 				 * for this device.
5377 				 */
5378 				linkwatch_run_queue();
5379 			}
5380 
5381 			__rtnl_unlock();
5382 
5383 			rebroadcast_time = jiffies;
5384 		}
5385 
5386 		msleep(250);
5387 
5388 		refcnt = netdev_refcnt_read(dev);
5389 
5390 		if (time_after(jiffies, warning_time + 10 * HZ)) {
5391 			pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5392 				 dev->name, refcnt);
5393 			warning_time = jiffies;
5394 		}
5395 	}
5396 }
5397 
5398 /* The sequence is:
5399  *
5400  *	rtnl_lock();
5401  *	...
5402  *	register_netdevice(x1);
5403  *	register_netdevice(x2);
5404  *	...
5405  *	unregister_netdevice(y1);
5406  *	unregister_netdevice(y2);
5407  *      ...
5408  *	rtnl_unlock();
5409  *	free_netdev(y1);
5410  *	free_netdev(y2);
5411  *
5412  * We are invoked by rtnl_unlock().
5413  * This allows us to deal with problems:
5414  * 1) We can delete sysfs objects which invoke hotplug
5415  *    without deadlocking with linkwatch via keventd.
5416  * 2) Since we run with the RTNL semaphore not held, we can sleep
5417  *    safely in order to wait for the netdev refcnt to drop to zero.
5418  *
5419  * We must not return until all unregister events added during
5420  * the interval the lock was held have been completed.
5421  */
5422 void netdev_run_todo(void)
5423 {
5424 	struct list_head list;
5425 
5426 	/* Snapshot list, allow later requests */
5427 	list_replace_init(&net_todo_list, &list);
5428 
5429 	__rtnl_unlock();
5430 
5431 
5432 	/* Wait for rcu callbacks to finish before next phase */
5433 	if (!list_empty(&list))
5434 		rcu_barrier();
5435 
5436 	while (!list_empty(&list)) {
5437 		struct net_device *dev
5438 			= list_first_entry(&list, struct net_device, todo_list);
5439 		list_del(&dev->todo_list);
5440 
5441 		rtnl_lock();
5442 		call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5443 		__rtnl_unlock();
5444 
5445 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5446 			pr_err("network todo '%s' but state %d\n",
5447 			       dev->name, dev->reg_state);
5448 			dump_stack();
5449 			continue;
5450 		}
5451 
5452 		dev->reg_state = NETREG_UNREGISTERED;
5453 
5454 		on_each_cpu(flush_backlog, dev, 1);
5455 
5456 		netdev_wait_allrefs(dev);
5457 
5458 		/* paranoia */
5459 		BUG_ON(netdev_refcnt_read(dev));
5460 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
5461 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5462 		WARN_ON(dev->dn_ptr);
5463 
5464 		if (dev->destructor)
5465 			dev->destructor(dev);
5466 
5467 		/* Free network device */
5468 		kobject_put(&dev->dev.kobj);
5469 	}
5470 }
5471 
5472 /* Convert net_device_stats to rtnl_link_stats64.  They have the same
5473  * fields in the same order, with only the type differing.
5474  */
5475 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5476 			     const struct net_device_stats *netdev_stats)
5477 {
5478 #if BITS_PER_LONG == 64
5479 	BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5480 	memcpy(stats64, netdev_stats, sizeof(*stats64));
5481 #else
5482 	size_t i, n = sizeof(*stats64) / sizeof(u64);
5483 	const unsigned long *src = (const unsigned long *)netdev_stats;
5484 	u64 *dst = (u64 *)stats64;
5485 
5486 	BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5487 		     sizeof(*stats64) / sizeof(u64));
5488 	for (i = 0; i < n; i++)
5489 		dst[i] = src[i];
5490 #endif
5491 }
5492 EXPORT_SYMBOL(netdev_stats_to_stats64);
5493 
5494 /**
5495  *	dev_get_stats	- get network device statistics
5496  *	@dev: device to get statistics from
5497  *	@storage: place to store stats
5498  *
5499  *	Get network statistics from device. Return @storage.
5500  *	The device driver may provide its own method by setting
5501  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5502  *	otherwise the internal statistics structure is used.
5503  */
5504 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5505 					struct rtnl_link_stats64 *storage)
5506 {
5507 	const struct net_device_ops *ops = dev->netdev_ops;
5508 
5509 	if (ops->ndo_get_stats64) {
5510 		memset(storage, 0, sizeof(*storage));
5511 		ops->ndo_get_stats64(dev, storage);
5512 	} else if (ops->ndo_get_stats) {
5513 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5514 	} else {
5515 		netdev_stats_to_stats64(storage, &dev->stats);
5516 	}
5517 	storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5518 	return storage;
5519 }
5520 EXPORT_SYMBOL(dev_get_stats);
5521 
5522 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5523 {
5524 	struct netdev_queue *queue = dev_ingress_queue(dev);
5525 
5526 #ifdef CONFIG_NET_CLS_ACT
5527 	if (queue)
5528 		return queue;
5529 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5530 	if (!queue)
5531 		return NULL;
5532 	netdev_init_one_queue(dev, queue, NULL);
5533 	queue->qdisc = &noop_qdisc;
5534 	queue->qdisc_sleeping = &noop_qdisc;
5535 	rcu_assign_pointer(dev->ingress_queue, queue);
5536 #endif
5537 	return queue;
5538 }
5539 
5540 static const struct ethtool_ops default_ethtool_ops;
5541 
5542 void netdev_set_default_ethtool_ops(struct net_device *dev,
5543 				    const struct ethtool_ops *ops)
5544 {
5545 	if (dev->ethtool_ops == &default_ethtool_ops)
5546 		dev->ethtool_ops = ops;
5547 }
5548 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5549 
5550 /**
5551  *	alloc_netdev_mqs - allocate network device
5552  *	@sizeof_priv:	size of private data to allocate space for
5553  *	@name:		device name format string
5554  *	@setup:		callback to initialize device
5555  *	@txqs:		the number of TX subqueues to allocate
5556  *	@rxqs:		the number of RX subqueues to allocate
5557  *
5558  *	Allocates a struct net_device with private data area for driver use
5559  *	and performs basic initialization.  Also allocates subquue structs
5560  *	for each queue on the device.
5561  */
5562 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5563 		void (*setup)(struct net_device *),
5564 		unsigned int txqs, unsigned int rxqs)
5565 {
5566 	struct net_device *dev;
5567 	size_t alloc_size;
5568 	struct net_device *p;
5569 
5570 	BUG_ON(strlen(name) >= sizeof(dev->name));
5571 
5572 	if (txqs < 1) {
5573 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5574 		return NULL;
5575 	}
5576 
5577 #ifdef CONFIG_RPS
5578 	if (rxqs < 1) {
5579 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5580 		return NULL;
5581 	}
5582 #endif
5583 
5584 	alloc_size = sizeof(struct net_device);
5585 	if (sizeof_priv) {
5586 		/* ensure 32-byte alignment of private area */
5587 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5588 		alloc_size += sizeof_priv;
5589 	}
5590 	/* ensure 32-byte alignment of whole construct */
5591 	alloc_size += NETDEV_ALIGN - 1;
5592 
5593 	p = kzalloc(alloc_size, GFP_KERNEL);
5594 	if (!p)
5595 		return NULL;
5596 
5597 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
5598 	dev->padded = (char *)dev - (char *)p;
5599 
5600 	dev->pcpu_refcnt = alloc_percpu(int);
5601 	if (!dev->pcpu_refcnt)
5602 		goto free_p;
5603 
5604 	if (dev_addr_init(dev))
5605 		goto free_pcpu;
5606 
5607 	dev_mc_init(dev);
5608 	dev_uc_init(dev);
5609 
5610 	dev_net_set(dev, &init_net);
5611 
5612 	dev->gso_max_size = GSO_MAX_SIZE;
5613 	dev->gso_max_segs = GSO_MAX_SEGS;
5614 
5615 	INIT_LIST_HEAD(&dev->napi_list);
5616 	INIT_LIST_HEAD(&dev->unreg_list);
5617 	INIT_LIST_HEAD(&dev->link_watch_list);
5618 	INIT_LIST_HEAD(&dev->upper_dev_list);
5619 	dev->priv_flags = IFF_XMIT_DST_RELEASE;
5620 	setup(dev);
5621 
5622 	dev->num_tx_queues = txqs;
5623 	dev->real_num_tx_queues = txqs;
5624 	if (netif_alloc_netdev_queues(dev))
5625 		goto free_all;
5626 
5627 #ifdef CONFIG_RPS
5628 	dev->num_rx_queues = rxqs;
5629 	dev->real_num_rx_queues = rxqs;
5630 	if (netif_alloc_rx_queues(dev))
5631 		goto free_all;
5632 #endif
5633 
5634 	strcpy(dev->name, name);
5635 	dev->group = INIT_NETDEV_GROUP;
5636 	if (!dev->ethtool_ops)
5637 		dev->ethtool_ops = &default_ethtool_ops;
5638 	return dev;
5639 
5640 free_all:
5641 	free_netdev(dev);
5642 	return NULL;
5643 
5644 free_pcpu:
5645 	free_percpu(dev->pcpu_refcnt);
5646 	kfree(dev->_tx);
5647 #ifdef CONFIG_RPS
5648 	kfree(dev->_rx);
5649 #endif
5650 
5651 free_p:
5652 	kfree(p);
5653 	return NULL;
5654 }
5655 EXPORT_SYMBOL(alloc_netdev_mqs);
5656 
5657 /**
5658  *	free_netdev - free network device
5659  *	@dev: device
5660  *
5661  *	This function does the last stage of destroying an allocated device
5662  * 	interface. The reference to the device object is released.
5663  *	If this is the last reference then it will be freed.
5664  */
5665 void free_netdev(struct net_device *dev)
5666 {
5667 	struct napi_struct *p, *n;
5668 
5669 	release_net(dev_net(dev));
5670 
5671 	kfree(dev->_tx);
5672 #ifdef CONFIG_RPS
5673 	kfree(dev->_rx);
5674 #endif
5675 
5676 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5677 
5678 	/* Flush device addresses */
5679 	dev_addr_flush(dev);
5680 
5681 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5682 		netif_napi_del(p);
5683 
5684 	free_percpu(dev->pcpu_refcnt);
5685 	dev->pcpu_refcnt = NULL;
5686 
5687 	/*  Compatibility with error handling in drivers */
5688 	if (dev->reg_state == NETREG_UNINITIALIZED) {
5689 		kfree((char *)dev - dev->padded);
5690 		return;
5691 	}
5692 
5693 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5694 	dev->reg_state = NETREG_RELEASED;
5695 
5696 	/* will free via device release */
5697 	put_device(&dev->dev);
5698 }
5699 EXPORT_SYMBOL(free_netdev);
5700 
5701 /**
5702  *	synchronize_net -  Synchronize with packet receive processing
5703  *
5704  *	Wait for packets currently being received to be done.
5705  *	Does not block later packets from starting.
5706  */
5707 void synchronize_net(void)
5708 {
5709 	might_sleep();
5710 	if (rtnl_is_locked())
5711 		synchronize_rcu_expedited();
5712 	else
5713 		synchronize_rcu();
5714 }
5715 EXPORT_SYMBOL(synchronize_net);
5716 
5717 /**
5718  *	unregister_netdevice_queue - remove device from the kernel
5719  *	@dev: device
5720  *	@head: list
5721  *
5722  *	This function shuts down a device interface and removes it
5723  *	from the kernel tables.
5724  *	If head not NULL, device is queued to be unregistered later.
5725  *
5726  *	Callers must hold the rtnl semaphore.  You may want
5727  *	unregister_netdev() instead of this.
5728  */
5729 
5730 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5731 {
5732 	ASSERT_RTNL();
5733 
5734 	if (head) {
5735 		list_move_tail(&dev->unreg_list, head);
5736 	} else {
5737 		rollback_registered(dev);
5738 		/* Finish processing unregister after unlock */
5739 		net_set_todo(dev);
5740 	}
5741 }
5742 EXPORT_SYMBOL(unregister_netdevice_queue);
5743 
5744 /**
5745  *	unregister_netdevice_many - unregister many devices
5746  *	@head: list of devices
5747  */
5748 void unregister_netdevice_many(struct list_head *head)
5749 {
5750 	struct net_device *dev;
5751 
5752 	if (!list_empty(head)) {
5753 		rollback_registered_many(head);
5754 		list_for_each_entry(dev, head, unreg_list)
5755 			net_set_todo(dev);
5756 	}
5757 }
5758 EXPORT_SYMBOL(unregister_netdevice_many);
5759 
5760 /**
5761  *	unregister_netdev - remove device from the kernel
5762  *	@dev: device
5763  *
5764  *	This function shuts down a device interface and removes it
5765  *	from the kernel tables.
5766  *
5767  *	This is just a wrapper for unregister_netdevice that takes
5768  *	the rtnl semaphore.  In general you want to use this and not
5769  *	unregister_netdevice.
5770  */
5771 void unregister_netdev(struct net_device *dev)
5772 {
5773 	rtnl_lock();
5774 	unregister_netdevice(dev);
5775 	rtnl_unlock();
5776 }
5777 EXPORT_SYMBOL(unregister_netdev);
5778 
5779 /**
5780  *	dev_change_net_namespace - move device to different nethost namespace
5781  *	@dev: device
5782  *	@net: network namespace
5783  *	@pat: If not NULL name pattern to try if the current device name
5784  *	      is already taken in the destination network namespace.
5785  *
5786  *	This function shuts down a device interface and moves it
5787  *	to a new network namespace. On success 0 is returned, on
5788  *	a failure a netagive errno code is returned.
5789  *
5790  *	Callers must hold the rtnl semaphore.
5791  */
5792 
5793 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5794 {
5795 	int err;
5796 
5797 	ASSERT_RTNL();
5798 
5799 	/* Don't allow namespace local devices to be moved. */
5800 	err = -EINVAL;
5801 	if (dev->features & NETIF_F_NETNS_LOCAL)
5802 		goto out;
5803 
5804 	/* Ensure the device has been registrered */
5805 	if (dev->reg_state != NETREG_REGISTERED)
5806 		goto out;
5807 
5808 	/* Get out if there is nothing todo */
5809 	err = 0;
5810 	if (net_eq(dev_net(dev), net))
5811 		goto out;
5812 
5813 	/* Pick the destination device name, and ensure
5814 	 * we can use it in the destination network namespace.
5815 	 */
5816 	err = -EEXIST;
5817 	if (__dev_get_by_name(net, dev->name)) {
5818 		/* We get here if we can't use the current device name */
5819 		if (!pat)
5820 			goto out;
5821 		if (dev_get_valid_name(net, dev, pat) < 0)
5822 			goto out;
5823 	}
5824 
5825 	/*
5826 	 * And now a mini version of register_netdevice unregister_netdevice.
5827 	 */
5828 
5829 	/* If device is running close it first. */
5830 	dev_close(dev);
5831 
5832 	/* And unlink it from device chain */
5833 	err = -ENODEV;
5834 	unlist_netdevice(dev);
5835 
5836 	synchronize_net();
5837 
5838 	/* Shutdown queueing discipline. */
5839 	dev_shutdown(dev);
5840 
5841 	/* Notify protocols, that we are about to destroy
5842 	   this device. They should clean all the things.
5843 
5844 	   Note that dev->reg_state stays at NETREG_REGISTERED.
5845 	   This is wanted because this way 8021q and macvlan know
5846 	   the device is just moving and can keep their slaves up.
5847 	*/
5848 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5849 	rcu_barrier();
5850 	call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5851 	rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5852 
5853 	/*
5854 	 *	Flush the unicast and multicast chains
5855 	 */
5856 	dev_uc_flush(dev);
5857 	dev_mc_flush(dev);
5858 
5859 	/* Send a netdev-removed uevent to the old namespace */
5860 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
5861 
5862 	/* Actually switch the network namespace */
5863 	dev_net_set(dev, net);
5864 
5865 	/* If there is an ifindex conflict assign a new one */
5866 	if (__dev_get_by_index(net, dev->ifindex)) {
5867 		int iflink = (dev->iflink == dev->ifindex);
5868 		dev->ifindex = dev_new_index(net);
5869 		if (iflink)
5870 			dev->iflink = dev->ifindex;
5871 	}
5872 
5873 	/* Send a netdev-add uevent to the new namespace */
5874 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
5875 
5876 	/* Fixup kobjects */
5877 	err = device_rename(&dev->dev, dev->name);
5878 	WARN_ON(err);
5879 
5880 	/* Add the device back in the hashes */
5881 	list_netdevice(dev);
5882 
5883 	/* Notify protocols, that a new device appeared. */
5884 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
5885 
5886 	/*
5887 	 *	Prevent userspace races by waiting until the network
5888 	 *	device is fully setup before sending notifications.
5889 	 */
5890 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5891 
5892 	synchronize_net();
5893 	err = 0;
5894 out:
5895 	return err;
5896 }
5897 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5898 
5899 static int dev_cpu_callback(struct notifier_block *nfb,
5900 			    unsigned long action,
5901 			    void *ocpu)
5902 {
5903 	struct sk_buff **list_skb;
5904 	struct sk_buff *skb;
5905 	unsigned int cpu, oldcpu = (unsigned long)ocpu;
5906 	struct softnet_data *sd, *oldsd;
5907 
5908 	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5909 		return NOTIFY_OK;
5910 
5911 	local_irq_disable();
5912 	cpu = smp_processor_id();
5913 	sd = &per_cpu(softnet_data, cpu);
5914 	oldsd = &per_cpu(softnet_data, oldcpu);
5915 
5916 	/* Find end of our completion_queue. */
5917 	list_skb = &sd->completion_queue;
5918 	while (*list_skb)
5919 		list_skb = &(*list_skb)->next;
5920 	/* Append completion queue from offline CPU. */
5921 	*list_skb = oldsd->completion_queue;
5922 	oldsd->completion_queue = NULL;
5923 
5924 	/* Append output queue from offline CPU. */
5925 	if (oldsd->output_queue) {
5926 		*sd->output_queue_tailp = oldsd->output_queue;
5927 		sd->output_queue_tailp = oldsd->output_queue_tailp;
5928 		oldsd->output_queue = NULL;
5929 		oldsd->output_queue_tailp = &oldsd->output_queue;
5930 	}
5931 	/* Append NAPI poll list from offline CPU. */
5932 	if (!list_empty(&oldsd->poll_list)) {
5933 		list_splice_init(&oldsd->poll_list, &sd->poll_list);
5934 		raise_softirq_irqoff(NET_RX_SOFTIRQ);
5935 	}
5936 
5937 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
5938 	local_irq_enable();
5939 
5940 	/* Process offline CPU's input_pkt_queue */
5941 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5942 		netif_rx(skb);
5943 		input_queue_head_incr(oldsd);
5944 	}
5945 	while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5946 		netif_rx(skb);
5947 		input_queue_head_incr(oldsd);
5948 	}
5949 
5950 	return NOTIFY_OK;
5951 }
5952 
5953 
5954 /**
5955  *	netdev_increment_features - increment feature set by one
5956  *	@all: current feature set
5957  *	@one: new feature set
5958  *	@mask: mask feature set
5959  *
5960  *	Computes a new feature set after adding a device with feature set
5961  *	@one to the master device with current feature set @all.  Will not
5962  *	enable anything that is off in @mask. Returns the new feature set.
5963  */
5964 netdev_features_t netdev_increment_features(netdev_features_t all,
5965 	netdev_features_t one, netdev_features_t mask)
5966 {
5967 	if (mask & NETIF_F_GEN_CSUM)
5968 		mask |= NETIF_F_ALL_CSUM;
5969 	mask |= NETIF_F_VLAN_CHALLENGED;
5970 
5971 	all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
5972 	all &= one | ~NETIF_F_ALL_FOR_ALL;
5973 
5974 	/* If one device supports hw checksumming, set for all. */
5975 	if (all & NETIF_F_GEN_CSUM)
5976 		all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
5977 
5978 	return all;
5979 }
5980 EXPORT_SYMBOL(netdev_increment_features);
5981 
5982 static struct hlist_head *netdev_create_hash(void)
5983 {
5984 	int i;
5985 	struct hlist_head *hash;
5986 
5987 	hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5988 	if (hash != NULL)
5989 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
5990 			INIT_HLIST_HEAD(&hash[i]);
5991 
5992 	return hash;
5993 }
5994 
5995 /* Initialize per network namespace state */
5996 static int __net_init netdev_init(struct net *net)
5997 {
5998 	if (net != &init_net)
5999 		INIT_LIST_HEAD(&net->dev_base_head);
6000 
6001 	net->dev_name_head = netdev_create_hash();
6002 	if (net->dev_name_head == NULL)
6003 		goto err_name;
6004 
6005 	net->dev_index_head = netdev_create_hash();
6006 	if (net->dev_index_head == NULL)
6007 		goto err_idx;
6008 
6009 	return 0;
6010 
6011 err_idx:
6012 	kfree(net->dev_name_head);
6013 err_name:
6014 	return -ENOMEM;
6015 }
6016 
6017 /**
6018  *	netdev_drivername - network driver for the device
6019  *	@dev: network device
6020  *
6021  *	Determine network driver for device.
6022  */
6023 const char *netdev_drivername(const struct net_device *dev)
6024 {
6025 	const struct device_driver *driver;
6026 	const struct device *parent;
6027 	const char *empty = "";
6028 
6029 	parent = dev->dev.parent;
6030 	if (!parent)
6031 		return empty;
6032 
6033 	driver = parent->driver;
6034 	if (driver && driver->name)
6035 		return driver->name;
6036 	return empty;
6037 }
6038 
6039 static int __netdev_printk(const char *level, const struct net_device *dev,
6040 			   struct va_format *vaf)
6041 {
6042 	int r;
6043 
6044 	if (dev && dev->dev.parent) {
6045 		r = dev_printk_emit(level[1] - '0',
6046 				    dev->dev.parent,
6047 				    "%s %s %s: %pV",
6048 				    dev_driver_string(dev->dev.parent),
6049 				    dev_name(dev->dev.parent),
6050 				    netdev_name(dev), vaf);
6051 	} else if (dev) {
6052 		r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6053 	} else {
6054 		r = printk("%s(NULL net_device): %pV", level, vaf);
6055 	}
6056 
6057 	return r;
6058 }
6059 
6060 int netdev_printk(const char *level, const struct net_device *dev,
6061 		  const char *format, ...)
6062 {
6063 	struct va_format vaf;
6064 	va_list args;
6065 	int r;
6066 
6067 	va_start(args, format);
6068 
6069 	vaf.fmt = format;
6070 	vaf.va = &args;
6071 
6072 	r = __netdev_printk(level, dev, &vaf);
6073 
6074 	va_end(args);
6075 
6076 	return r;
6077 }
6078 EXPORT_SYMBOL(netdev_printk);
6079 
6080 #define define_netdev_printk_level(func, level)			\
6081 int func(const struct net_device *dev, const char *fmt, ...)	\
6082 {								\
6083 	int r;							\
6084 	struct va_format vaf;					\
6085 	va_list args;						\
6086 								\
6087 	va_start(args, fmt);					\
6088 								\
6089 	vaf.fmt = fmt;						\
6090 	vaf.va = &args;						\
6091 								\
6092 	r = __netdev_printk(level, dev, &vaf);			\
6093 								\
6094 	va_end(args);						\
6095 								\
6096 	return r;						\
6097 }								\
6098 EXPORT_SYMBOL(func);
6099 
6100 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6101 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6102 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6103 define_netdev_printk_level(netdev_err, KERN_ERR);
6104 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6105 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6106 define_netdev_printk_level(netdev_info, KERN_INFO);
6107 
6108 static void __net_exit netdev_exit(struct net *net)
6109 {
6110 	kfree(net->dev_name_head);
6111 	kfree(net->dev_index_head);
6112 }
6113 
6114 static struct pernet_operations __net_initdata netdev_net_ops = {
6115 	.init = netdev_init,
6116 	.exit = netdev_exit,
6117 };
6118 
6119 static void __net_exit default_device_exit(struct net *net)
6120 {
6121 	struct net_device *dev, *aux;
6122 	/*
6123 	 * Push all migratable network devices back to the
6124 	 * initial network namespace
6125 	 */
6126 	rtnl_lock();
6127 	for_each_netdev_safe(net, dev, aux) {
6128 		int err;
6129 		char fb_name[IFNAMSIZ];
6130 
6131 		/* Ignore unmoveable devices (i.e. loopback) */
6132 		if (dev->features & NETIF_F_NETNS_LOCAL)
6133 			continue;
6134 
6135 		/* Leave virtual devices for the generic cleanup */
6136 		if (dev->rtnl_link_ops)
6137 			continue;
6138 
6139 		/* Push remaining network devices to init_net */
6140 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6141 		err = dev_change_net_namespace(dev, &init_net, fb_name);
6142 		if (err) {
6143 			pr_emerg("%s: failed to move %s to init_net: %d\n",
6144 				 __func__, dev->name, err);
6145 			BUG();
6146 		}
6147 	}
6148 	rtnl_unlock();
6149 }
6150 
6151 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6152 {
6153 	/* At exit all network devices most be removed from a network
6154 	 * namespace.  Do this in the reverse order of registration.
6155 	 * Do this across as many network namespaces as possible to
6156 	 * improve batching efficiency.
6157 	 */
6158 	struct net_device *dev;
6159 	struct net *net;
6160 	LIST_HEAD(dev_kill_list);
6161 
6162 	rtnl_lock();
6163 	list_for_each_entry(net, net_list, exit_list) {
6164 		for_each_netdev_reverse(net, dev) {
6165 			if (dev->rtnl_link_ops)
6166 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6167 			else
6168 				unregister_netdevice_queue(dev, &dev_kill_list);
6169 		}
6170 	}
6171 	unregister_netdevice_many(&dev_kill_list);
6172 	list_del(&dev_kill_list);
6173 	rtnl_unlock();
6174 }
6175 
6176 static struct pernet_operations __net_initdata default_device_ops = {
6177 	.exit = default_device_exit,
6178 	.exit_batch = default_device_exit_batch,
6179 };
6180 
6181 /*
6182  *	Initialize the DEV module. At boot time this walks the device list and
6183  *	unhooks any devices that fail to initialise (normally hardware not
6184  *	present) and leaves us with a valid list of present and active devices.
6185  *
6186  */
6187 
6188 /*
6189  *       This is called single threaded during boot, so no need
6190  *       to take the rtnl semaphore.
6191  */
6192 static int __init net_dev_init(void)
6193 {
6194 	int i, rc = -ENOMEM;
6195 
6196 	BUG_ON(!dev_boot_phase);
6197 
6198 	if (dev_proc_init())
6199 		goto out;
6200 
6201 	if (netdev_kobject_init())
6202 		goto out;
6203 
6204 	INIT_LIST_HEAD(&ptype_all);
6205 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
6206 		INIT_LIST_HEAD(&ptype_base[i]);
6207 
6208 	INIT_LIST_HEAD(&offload_base);
6209 
6210 	if (register_pernet_subsys(&netdev_net_ops))
6211 		goto out;
6212 
6213 	/*
6214 	 *	Initialise the packet receive queues.
6215 	 */
6216 
6217 	for_each_possible_cpu(i) {
6218 		struct softnet_data *sd = &per_cpu(softnet_data, i);
6219 
6220 		memset(sd, 0, sizeof(*sd));
6221 		skb_queue_head_init(&sd->input_pkt_queue);
6222 		skb_queue_head_init(&sd->process_queue);
6223 		sd->completion_queue = NULL;
6224 		INIT_LIST_HEAD(&sd->poll_list);
6225 		sd->output_queue = NULL;
6226 		sd->output_queue_tailp = &sd->output_queue;
6227 #ifdef CONFIG_RPS
6228 		sd->csd.func = rps_trigger_softirq;
6229 		sd->csd.info = sd;
6230 		sd->csd.flags = 0;
6231 		sd->cpu = i;
6232 #endif
6233 
6234 		sd->backlog.poll = process_backlog;
6235 		sd->backlog.weight = weight_p;
6236 		sd->backlog.gro_list = NULL;
6237 		sd->backlog.gro_count = 0;
6238 	}
6239 
6240 	dev_boot_phase = 0;
6241 
6242 	/* The loopback device is special if any other network devices
6243 	 * is present in a network namespace the loopback device must
6244 	 * be present. Since we now dynamically allocate and free the
6245 	 * loopback device ensure this invariant is maintained by
6246 	 * keeping the loopback device as the first device on the
6247 	 * list of network devices.  Ensuring the loopback devices
6248 	 * is the first device that appears and the last network device
6249 	 * that disappears.
6250 	 */
6251 	if (register_pernet_device(&loopback_net_ops))
6252 		goto out;
6253 
6254 	if (register_pernet_device(&default_device_ops))
6255 		goto out;
6256 
6257 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6258 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6259 
6260 	hotcpu_notifier(dev_cpu_callback, 0);
6261 	dst_init();
6262 	rc = 0;
6263 out:
6264 	return rc;
6265 }
6266 
6267 subsys_initcall(net_dev_init);
6268