2 * NET3 Protocol independent device support routines.
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.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
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>
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
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
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
51 * Rudi Cilibrasi : Pass the right thing to
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
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
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
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>
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>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.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>
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>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
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;
148 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
151 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
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.
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.
162 * See, for example usages, register_netdevice() and
163 * unregister_netdevice(), which must be called with the rtnl
166 DEFINE_RWLOCK(dev_base_lock);
167 EXPORT_SYMBOL(dev_base_lock);
169 seqcount_t devnet_rename_seq;
171 static inline void dev_base_seq_inc(struct net *net)
173 while (++net->dev_base_seq == 0);
176 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
178 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
180 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
183 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
185 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
188 static inline void rps_lock(struct softnet_data *sd)
191 spin_lock(&sd->input_pkt_queue.lock);
195 static inline void rps_unlock(struct softnet_data *sd)
198 spin_unlock(&sd->input_pkt_queue.lock);
202 /* Device list insertion */
203 static void list_netdevice(struct net_device *dev)
205 struct net *net = dev_net(dev);
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);
216 dev_base_seq_inc(net);
219 /* Device list removal
220 * caller must respect a RCU grace period before freeing/reusing dev
222 static void unlist_netdevice(struct net_device *dev)
226 /* Unlink dev from the device chain */
227 write_lock_bh(&dev_base_lock);
228 list_del_rcu(&dev->dev_list);
229 hlist_del_rcu(&dev->name_hlist);
230 hlist_del_rcu(&dev->index_hlist);
231 write_unlock_bh(&dev_base_lock);
233 dev_base_seq_inc(dev_net(dev));
240 static RAW_NOTIFIER_HEAD(netdev_chain);
243 * Device drivers call our routines to queue packets here. We empty the
244 * queue in the local softnet handler.
247 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
248 EXPORT_PER_CPU_SYMBOL(softnet_data);
250 #ifdef CONFIG_LOCKDEP
252 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
253 * according to dev->type
255 static const unsigned short netdev_lock_type[] =
256 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
257 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
258 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
259 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
260 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
261 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
262 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
263 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
264 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
265 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
266 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
267 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
268 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
269 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
270 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
272 static const char *const netdev_lock_name[] =
273 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
274 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
275 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
276 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
277 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
278 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
279 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
280 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
281 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
282 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
283 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
284 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
285 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
286 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
287 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
289 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
290 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
292 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
296 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
297 if (netdev_lock_type[i] == dev_type)
299 /* the last key is used by default */
300 return ARRAY_SIZE(netdev_lock_type) - 1;
303 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
304 unsigned short dev_type)
308 i = netdev_lock_pos(dev_type);
309 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
310 netdev_lock_name[i]);
313 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
317 i = netdev_lock_pos(dev->type);
318 lockdep_set_class_and_name(&dev->addr_list_lock,
319 &netdev_addr_lock_key[i],
320 netdev_lock_name[i]);
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 /*******************************************************************************
334 Protocol management and registration routines
336 *******************************************************************************/
339 * Add a protocol ID to the list. Now that the input handler is
340 * smarter we can dispense with all the messy stuff that used to be
343 * BEWARE!!! Protocol handlers, mangling input packets,
344 * MUST BE last in hash buckets and checking protocol handlers
345 * MUST start from promiscuous ptype_all chain in net_bh.
346 * It is true now, do not change it.
347 * Explanation follows: if protocol handler, mangling packet, will
348 * be the first on list, it is not able to sense, that packet
349 * is cloned and should be copied-on-write, so that it will
350 * change it and subsequent readers will get broken packet.
354 static inline struct list_head *ptype_head(const struct packet_type *pt)
356 if (pt->type == htons(ETH_P_ALL))
359 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
363 * dev_add_pack - add packet handler
364 * @pt: packet type declaration
366 * Add a protocol handler to the networking stack. The passed &packet_type
367 * is linked into kernel lists and may not be freed until it has been
368 * removed from the kernel lists.
370 * This call does not sleep therefore it can not
371 * guarantee all CPU's that are in middle of receiving packets
372 * will see the new packet type (until the next received packet).
375 void dev_add_pack(struct packet_type *pt)
377 struct list_head *head = ptype_head(pt);
379 spin_lock(&ptype_lock);
380 list_add_rcu(&pt->list, head);
381 spin_unlock(&ptype_lock);
383 EXPORT_SYMBOL(dev_add_pack);
386 * __dev_remove_pack - remove packet handler
387 * @pt: packet type declaration
389 * Remove a protocol handler that was previously added to the kernel
390 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
391 * from the kernel lists and can be freed or reused once this function
394 * The packet type might still be in use by receivers
395 * and must not be freed until after all the CPU's have gone
396 * through a quiescent state.
398 void __dev_remove_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
401 struct packet_type *pt1;
403 spin_lock(&ptype_lock);
405 list_for_each_entry(pt1, head, list) {
407 list_del_rcu(&pt->list);
412 pr_warn("dev_remove_pack: %p not found\n", pt);
414 spin_unlock(&ptype_lock);
416 EXPORT_SYMBOL(__dev_remove_pack);
419 * dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * This call sleeps to guarantee that no CPU is looking at the packet
430 void dev_remove_pack(struct packet_type *pt)
432 __dev_remove_pack(pt);
436 EXPORT_SYMBOL(dev_remove_pack);
440 * dev_add_offload - register offload handlers
441 * @po: protocol offload declaration
443 * Add protocol offload handlers to the networking stack. The passed
444 * &proto_offload is linked into kernel lists and may not be freed until
445 * it has been removed from the kernel lists.
447 * This call does not sleep therefore it can not
448 * guarantee all CPU's that are in middle of receiving packets
449 * will see the new offload handlers (until the next received packet).
451 void dev_add_offload(struct packet_offload *po)
453 struct list_head *head = &offload_base;
455 spin_lock(&offload_lock);
456 list_add_rcu(&po->list, head);
457 spin_unlock(&offload_lock);
459 EXPORT_SYMBOL(dev_add_offload);
462 * __dev_remove_offload - remove offload handler
463 * @po: packet offload declaration
465 * Remove a protocol offload handler that was previously added to the
466 * kernel offload handlers by dev_add_offload(). The passed &offload_type
467 * is removed from the kernel lists and can be freed or reused once this
470 * The packet type might still be in use by receivers
471 * and must not be freed until after all the CPU's have gone
472 * through a quiescent state.
474 void __dev_remove_offload(struct packet_offload *po)
476 struct list_head *head = &offload_base;
477 struct packet_offload *po1;
479 spin_lock(&offload_lock);
481 list_for_each_entry(po1, head, list) {
483 list_del_rcu(&po->list);
488 pr_warn("dev_remove_offload: %p not found\n", po);
490 spin_unlock(&offload_lock);
492 EXPORT_SYMBOL(__dev_remove_offload);
495 * dev_remove_offload - remove packet offload handler
496 * @po: packet offload declaration
498 * Remove a packet offload handler that was previously added to the kernel
499 * offload handlers by dev_add_offload(). The passed &offload_type is
500 * removed from the kernel lists and can be freed or reused once this
503 * This call sleeps to guarantee that no CPU is looking at the packet
506 void dev_remove_offload(struct packet_offload *po)
508 __dev_remove_offload(po);
512 EXPORT_SYMBOL(dev_remove_offload);
514 /******************************************************************************
516 Device Boot-time Settings Routines
518 *******************************************************************************/
520 /* Boot time configuration table */
521 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
524 * netdev_boot_setup_add - add new setup entry
525 * @name: name of the device
526 * @map: configured settings for the device
528 * Adds new setup entry to the dev_boot_setup list. The function
529 * returns 0 on error and 1 on success. This is a generic routine to
532 static int netdev_boot_setup_add(char *name, struct ifmap *map)
534 struct netdev_boot_setup *s;
538 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
539 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
540 memset(s[i].name, 0, sizeof(s[i].name));
541 strlcpy(s[i].name, name, IFNAMSIZ);
542 memcpy(&s[i].map, map, sizeof(s[i].map));
547 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
551 * netdev_boot_setup_check - check boot time settings
552 * @dev: the netdevice
554 * Check boot time settings for the device.
555 * The found settings are set for the device to be used
556 * later in the device probing.
557 * Returns 0 if no settings found, 1 if they are.
559 int netdev_boot_setup_check(struct net_device *dev)
561 struct netdev_boot_setup *s = dev_boot_setup;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
566 !strcmp(dev->name, s[i].name)) {
567 dev->irq = s[i].map.irq;
568 dev->base_addr = s[i].map.base_addr;
569 dev->mem_start = s[i].map.mem_start;
570 dev->mem_end = s[i].map.mem_end;
576 EXPORT_SYMBOL(netdev_boot_setup_check);
580 * netdev_boot_base - get address from boot time settings
581 * @prefix: prefix for network device
582 * @unit: id for network device
584 * Check boot time settings for the base address of device.
585 * The found settings are set for the device to be used
586 * later in the device probing.
587 * Returns 0 if no settings found.
589 unsigned long netdev_boot_base(const char *prefix, int unit)
591 const struct netdev_boot_setup *s = dev_boot_setup;
595 sprintf(name, "%s%d", prefix, unit);
598 * If device already registered then return base of 1
599 * to indicate not to probe for this interface
601 if (__dev_get_by_name(&init_net, name))
604 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
605 if (!strcmp(name, s[i].name))
606 return s[i].map.base_addr;
611 * Saves at boot time configured settings for any netdevice.
613 int __init netdev_boot_setup(char *str)
618 str = get_options(str, ARRAY_SIZE(ints), ints);
623 memset(&map, 0, sizeof(map));
627 map.base_addr = ints[2];
629 map.mem_start = ints[3];
631 map.mem_end = ints[4];
633 /* Add new entry to the list */
634 return netdev_boot_setup_add(str, &map);
637 __setup("netdev=", netdev_boot_setup);
639 /*******************************************************************************
641 Device Interface Subroutines
643 *******************************************************************************/
646 * __dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. Must be called under RTNL semaphore
651 * or @dev_base_lock. If the name is found a pointer to the device
652 * is returned. If the name is not found then %NULL is returned. The
653 * reference counters are not incremented so the caller must be
654 * careful with locks.
657 struct net_device *__dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
660 struct hlist_head *head = dev_name_hash(net, name);
662 hlist_for_each_entry(dev, head, name_hlist)
663 if (!strncmp(dev->name, name, IFNAMSIZ))
668 EXPORT_SYMBOL(__dev_get_by_name);
671 * dev_get_by_name_rcu - find a device by its name
672 * @net: the applicable net namespace
673 * @name: name to find
675 * Find an interface by name.
676 * If the name is found a pointer to the device is returned.
677 * If the name is not found then %NULL is returned.
678 * The reference counters are not incremented so the caller must be
679 * careful with locks. The caller must hold RCU lock.
682 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
684 struct net_device *dev;
685 struct hlist_head *head = dev_name_hash(net, name);
687 hlist_for_each_entry_rcu(dev, head, name_hlist)
688 if (!strncmp(dev->name, name, IFNAMSIZ))
693 EXPORT_SYMBOL(dev_get_by_name_rcu);
696 * dev_get_by_name - find a device by its name
697 * @net: the applicable net namespace
698 * @name: name to find
700 * Find an interface by name. This can be called from any
701 * context and does its own locking. The returned handle has
702 * the usage count incremented and the caller must use dev_put() to
703 * release it when it is no longer needed. %NULL is returned if no
704 * matching device is found.
707 struct net_device *dev_get_by_name(struct net *net, const char *name)
709 struct net_device *dev;
712 dev = dev_get_by_name_rcu(net, name);
718 EXPORT_SYMBOL(dev_get_by_name);
721 * __dev_get_by_index - find a device by its ifindex
722 * @net: the applicable net namespace
723 * @ifindex: index of device
725 * Search for an interface by index. Returns %NULL if the device
726 * is not found or a pointer to the device. The device has not
727 * had its reference counter increased so the caller must be careful
728 * about locking. The caller must hold either the RTNL semaphore
732 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
735 struct hlist_head *head = dev_index_hash(net, ifindex);
737 hlist_for_each_entry(dev, head, index_hlist)
738 if (dev->ifindex == ifindex)
743 EXPORT_SYMBOL(__dev_get_by_index);
746 * dev_get_by_index_rcu - find a device by its ifindex
747 * @net: the applicable net namespace
748 * @ifindex: index of device
750 * Search for an interface by index. Returns %NULL if the device
751 * is not found or a pointer to the device. The device has not
752 * had its reference counter increased so the caller must be careful
753 * about locking. The caller must hold RCU lock.
756 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
758 struct net_device *dev;
759 struct hlist_head *head = dev_index_hash(net, ifindex);
761 hlist_for_each_entry_rcu(dev, head, index_hlist)
762 if (dev->ifindex == ifindex)
767 EXPORT_SYMBOL(dev_get_by_index_rcu);
771 * dev_get_by_index - find a device by its ifindex
772 * @net: the applicable net namespace
773 * @ifindex: index of device
775 * Search for an interface by index. Returns NULL if the device
776 * is not found or a pointer to the device. The device returned has
777 * had a reference added and the pointer is safe until the user calls
778 * dev_put to indicate they have finished with it.
781 struct net_device *dev_get_by_index(struct net *net, int ifindex)
783 struct net_device *dev;
786 dev = dev_get_by_index_rcu(net, ifindex);
792 EXPORT_SYMBOL(dev_get_by_index);
795 * dev_getbyhwaddr_rcu - find a device by its hardware address
796 * @net: the applicable net namespace
797 * @type: media type of device
798 * @ha: hardware address
800 * Search for an interface by MAC address. Returns NULL if the device
801 * is not found or a pointer to the device.
802 * The caller must hold RCU or RTNL.
803 * The returned device has not had its ref count increased
804 * and the caller must therefore be careful about locking
808 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
811 struct net_device *dev;
813 for_each_netdev_rcu(net, dev)
814 if (dev->type == type &&
815 !memcmp(dev->dev_addr, ha, dev->addr_len))
820 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
822 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
824 struct net_device *dev;
827 for_each_netdev(net, dev)
828 if (dev->type == type)
833 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
835 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
837 struct net_device *dev, *ret = NULL;
840 for_each_netdev_rcu(net, dev)
841 if (dev->type == type) {
849 EXPORT_SYMBOL(dev_getfirstbyhwtype);
852 * dev_get_by_flags_rcu - find any device with given flags
853 * @net: the applicable net namespace
854 * @if_flags: IFF_* values
855 * @mask: bitmask of bits in if_flags to check
857 * Search for any interface with the given flags. Returns NULL if a device
858 * is not found or a pointer to the device. Must be called inside
859 * rcu_read_lock(), and result refcount is unchanged.
862 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
865 struct net_device *dev, *ret;
868 for_each_netdev_rcu(net, dev) {
869 if (((dev->flags ^ if_flags) & mask) == 0) {
876 EXPORT_SYMBOL(dev_get_by_flags_rcu);
879 * dev_valid_name - check if name is okay for network device
882 * Network device names need to be valid file names to
883 * to allow sysfs to work. We also disallow any kind of
886 bool dev_valid_name(const char *name)
890 if (strlen(name) >= IFNAMSIZ)
892 if (!strcmp(name, ".") || !strcmp(name, ".."))
896 if (*name == '/' || isspace(*name))
902 EXPORT_SYMBOL(dev_valid_name);
905 * __dev_alloc_name - allocate a name for a device
906 * @net: network namespace to allocate the device name in
907 * @name: name format string
908 * @buf: scratch buffer and result name string
910 * Passed a format string - eg "lt%d" it will try and find a suitable
911 * id. It scans list of devices to build up a free map, then chooses
912 * the first empty slot. The caller must hold the dev_base or rtnl lock
913 * while allocating the name and adding the device in order to avoid
915 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
916 * Returns the number of the unit assigned or a negative errno code.
919 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
923 const int max_netdevices = 8*PAGE_SIZE;
924 unsigned long *inuse;
925 struct net_device *d;
927 p = strnchr(name, IFNAMSIZ-1, '%');
930 * Verify the string as this thing may have come from
931 * the user. There must be either one "%d" and no other "%"
934 if (p[1] != 'd' || strchr(p + 2, '%'))
937 /* Use one page as a bit array of possible slots */
938 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
942 for_each_netdev(net, d) {
943 if (!sscanf(d->name, name, &i))
945 if (i < 0 || i >= max_netdevices)
948 /* avoid cases where sscanf is not exact inverse of printf */
949 snprintf(buf, IFNAMSIZ, name, i);
950 if (!strncmp(buf, d->name, IFNAMSIZ))
954 i = find_first_zero_bit(inuse, max_netdevices);
955 free_page((unsigned long) inuse);
959 snprintf(buf, IFNAMSIZ, name, i);
960 if (!__dev_get_by_name(net, buf))
963 /* It is possible to run out of possible slots
964 * when the name is long and there isn't enough space left
965 * for the digits, or if all bits are used.
971 * dev_alloc_name - allocate a name for a device
973 * @name: name format string
975 * Passed a format string - eg "lt%d" it will try and find a suitable
976 * id. It scans list of devices to build up a free map, then chooses
977 * the first empty slot. The caller must hold the dev_base or rtnl lock
978 * while allocating the name and adding the device in order to avoid
980 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
981 * Returns the number of the unit assigned or a negative errno code.
984 int dev_alloc_name(struct net_device *dev, const char *name)
990 BUG_ON(!dev_net(dev));
992 ret = __dev_alloc_name(net, name, buf);
994 strlcpy(dev->name, buf, IFNAMSIZ);
997 EXPORT_SYMBOL(dev_alloc_name);
999 static int dev_alloc_name_ns(struct net *net,
1000 struct net_device *dev,
1006 ret = __dev_alloc_name(net, name, buf);
1008 strlcpy(dev->name, buf, IFNAMSIZ);
1012 static int dev_get_valid_name(struct net *net,
1013 struct net_device *dev,
1018 if (!dev_valid_name(name))
1021 if (strchr(name, '%'))
1022 return dev_alloc_name_ns(net, dev, name);
1023 else if (__dev_get_by_name(net, name))
1025 else if (dev->name != name)
1026 strlcpy(dev->name, name, IFNAMSIZ);
1032 * dev_change_name - change name of a device
1034 * @newname: name (or format string) must be at least IFNAMSIZ
1036 * Change name of a device, can pass format strings "eth%d".
1039 int dev_change_name(struct net_device *dev, const char *newname)
1041 char oldname[IFNAMSIZ];
1047 BUG_ON(!dev_net(dev));
1050 if (dev->flags & IFF_UP)
1053 write_seqcount_begin(&devnet_rename_seq);
1055 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1056 write_seqcount_end(&devnet_rename_seq);
1060 memcpy(oldname, dev->name, IFNAMSIZ);
1062 err = dev_get_valid_name(net, dev, newname);
1064 write_seqcount_end(&devnet_rename_seq);
1069 ret = device_rename(&dev->dev, dev->name);
1071 memcpy(dev->name, oldname, IFNAMSIZ);
1072 write_seqcount_end(&devnet_rename_seq);
1076 write_seqcount_end(&devnet_rename_seq);
1078 write_lock_bh(&dev_base_lock);
1079 hlist_del_rcu(&dev->name_hlist);
1080 write_unlock_bh(&dev_base_lock);
1084 write_lock_bh(&dev_base_lock);
1085 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1086 write_unlock_bh(&dev_base_lock);
1088 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1089 ret = notifier_to_errno(ret);
1092 /* err >= 0 after dev_alloc_name() or stores the first errno */
1095 write_seqcount_begin(&devnet_rename_seq);
1096 memcpy(dev->name, oldname, IFNAMSIZ);
1099 pr_err("%s: name change rollback failed: %d\n",
1108 * dev_set_alias - change ifalias of a device
1110 * @alias: name up to IFALIASZ
1111 * @len: limit of bytes to copy from info
1113 * Set ifalias for a device,
1115 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1121 if (len >= IFALIASZ)
1125 kfree(dev->ifalias);
1126 dev->ifalias = NULL;
1130 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1133 dev->ifalias = new_ifalias;
1135 strlcpy(dev->ifalias, alias, len+1);
1141 * netdev_features_change - device changes features
1142 * @dev: device to cause notification
1144 * Called to indicate a device has changed features.
1146 void netdev_features_change(struct net_device *dev)
1148 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1150 EXPORT_SYMBOL(netdev_features_change);
1153 * netdev_state_change - device changes state
1154 * @dev: device to cause notification
1156 * Called to indicate a device has changed state. This function calls
1157 * the notifier chains for netdev_chain and sends a NEWLINK message
1158 * to the routing socket.
1160 void netdev_state_change(struct net_device *dev)
1162 if (dev->flags & IFF_UP) {
1163 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1164 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1167 EXPORT_SYMBOL(netdev_state_change);
1170 * netdev_notify_peers - notify network peers about existence of @dev
1171 * @dev: network device
1173 * Generate traffic such that interested network peers are aware of
1174 * @dev, such as by generating a gratuitous ARP. This may be used when
1175 * a device wants to inform the rest of the network about some sort of
1176 * reconfiguration such as a failover event or virtual machine
1179 void netdev_notify_peers(struct net_device *dev)
1182 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1185 EXPORT_SYMBOL(netdev_notify_peers);
1187 static int __dev_open(struct net_device *dev)
1189 const struct net_device_ops *ops = dev->netdev_ops;
1194 if (!netif_device_present(dev))
1197 /* Block netpoll from trying to do any rx path servicing.
1198 * If we don't do this there is a chance ndo_poll_controller
1199 * or ndo_poll may be running while we open the device
1201 ret = netpoll_rx_disable(dev);
1205 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1206 ret = notifier_to_errno(ret);
1210 set_bit(__LINK_STATE_START, &dev->state);
1212 if (ops->ndo_validate_addr)
1213 ret = ops->ndo_validate_addr(dev);
1215 if (!ret && ops->ndo_open)
1216 ret = ops->ndo_open(dev);
1218 netpoll_rx_enable(dev);
1221 clear_bit(__LINK_STATE_START, &dev->state);
1223 dev->flags |= IFF_UP;
1224 net_dmaengine_get();
1225 dev_set_rx_mode(dev);
1227 add_device_randomness(dev->dev_addr, dev->addr_len);
1234 * dev_open - prepare an interface for use.
1235 * @dev: device to open
1237 * Takes a device from down to up state. The device's private open
1238 * function is invoked and then the multicast lists are loaded. Finally
1239 * the device is moved into the up state and a %NETDEV_UP message is
1240 * sent to the netdev notifier chain.
1242 * Calling this function on an active interface is a nop. On a failure
1243 * a negative errno code is returned.
1245 int dev_open(struct net_device *dev)
1249 if (dev->flags & IFF_UP)
1252 ret = __dev_open(dev);
1256 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1257 call_netdevice_notifiers(NETDEV_UP, dev);
1261 EXPORT_SYMBOL(dev_open);
1263 static int __dev_close_many(struct list_head *head)
1265 struct net_device *dev;
1270 list_for_each_entry(dev, head, unreg_list) {
1271 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1273 clear_bit(__LINK_STATE_START, &dev->state);
1275 /* Synchronize to scheduled poll. We cannot touch poll list, it
1276 * can be even on different cpu. So just clear netif_running().
1278 * dev->stop() will invoke napi_disable() on all of it's
1279 * napi_struct instances on this device.
1281 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1284 dev_deactivate_many(head);
1286 list_for_each_entry(dev, head, unreg_list) {
1287 const struct net_device_ops *ops = dev->netdev_ops;
1290 * Call the device specific close. This cannot fail.
1291 * Only if device is UP
1293 * We allow it to be called even after a DETACH hot-plug
1299 dev->flags &= ~IFF_UP;
1300 net_dmaengine_put();
1306 static int __dev_close(struct net_device *dev)
1311 /* Temporarily disable netpoll until the interface is down */
1312 retval = netpoll_rx_disable(dev);
1316 list_add(&dev->unreg_list, &single);
1317 retval = __dev_close_many(&single);
1320 netpoll_rx_enable(dev);
1324 static int dev_close_many(struct list_head *head)
1326 struct net_device *dev, *tmp;
1327 LIST_HEAD(tmp_list);
1329 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1330 if (!(dev->flags & IFF_UP))
1331 list_move(&dev->unreg_list, &tmp_list);
1333 __dev_close_many(head);
1335 list_for_each_entry(dev, head, unreg_list) {
1336 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1337 call_netdevice_notifiers(NETDEV_DOWN, dev);
1340 /* rollback_registered_many needs the complete original list */
1341 list_splice(&tmp_list, head);
1346 * dev_close - shutdown an interface.
1347 * @dev: device to shutdown
1349 * This function moves an active device into down state. A
1350 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1351 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1354 int dev_close(struct net_device *dev)
1357 if (dev->flags & IFF_UP) {
1360 /* Block netpoll rx while the interface is going down */
1361 ret = netpoll_rx_disable(dev);
1365 list_add(&dev->unreg_list, &single);
1366 dev_close_many(&single);
1369 netpoll_rx_enable(dev);
1373 EXPORT_SYMBOL(dev_close);
1377 * dev_disable_lro - disable Large Receive Offload on a device
1380 * Disable Large Receive Offload (LRO) on a net device. Must be
1381 * called under RTNL. This is needed if received packets may be
1382 * forwarded to another interface.
1384 void dev_disable_lro(struct net_device *dev)
1387 * If we're trying to disable lro on a vlan device
1388 * use the underlying physical device instead
1390 if (is_vlan_dev(dev))
1391 dev = vlan_dev_real_dev(dev);
1393 dev->wanted_features &= ~NETIF_F_LRO;
1394 netdev_update_features(dev);
1396 if (unlikely(dev->features & NETIF_F_LRO))
1397 netdev_WARN(dev, "failed to disable LRO!\n");
1399 EXPORT_SYMBOL(dev_disable_lro);
1402 static int dev_boot_phase = 1;
1405 * register_netdevice_notifier - register a network notifier block
1408 * Register a notifier to be called when network device events occur.
1409 * The notifier passed is linked into the kernel structures and must
1410 * not be reused until it has been unregistered. A negative errno code
1411 * is returned on a failure.
1413 * When registered all registration and up events are replayed
1414 * to the new notifier to allow device to have a race free
1415 * view of the network device list.
1418 int register_netdevice_notifier(struct notifier_block *nb)
1420 struct net_device *dev;
1421 struct net_device *last;
1426 err = raw_notifier_chain_register(&netdev_chain, nb);
1432 for_each_netdev(net, dev) {
1433 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1434 err = notifier_to_errno(err);
1438 if (!(dev->flags & IFF_UP))
1441 nb->notifier_call(nb, NETDEV_UP, dev);
1452 for_each_netdev(net, dev) {
1456 if (dev->flags & IFF_UP) {
1457 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1458 nb->notifier_call(nb, NETDEV_DOWN, dev);
1460 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1465 raw_notifier_chain_unregister(&netdev_chain, nb);
1468 EXPORT_SYMBOL(register_netdevice_notifier);
1471 * unregister_netdevice_notifier - unregister a network notifier block
1474 * Unregister a notifier previously registered by
1475 * register_netdevice_notifier(). The notifier is unlinked into the
1476 * kernel structures and may then be reused. A negative errno code
1477 * is returned on a failure.
1479 * After unregistering unregister and down device events are synthesized
1480 * for all devices on the device list to the removed notifier to remove
1481 * the need for special case cleanup code.
1484 int unregister_netdevice_notifier(struct notifier_block *nb)
1486 struct net_device *dev;
1491 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1496 for_each_netdev(net, dev) {
1497 if (dev->flags & IFF_UP) {
1498 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1499 nb->notifier_call(nb, NETDEV_DOWN, dev);
1501 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1508 EXPORT_SYMBOL(unregister_netdevice_notifier);
1511 * call_netdevice_notifiers - call all network notifier blocks
1512 * @val: value passed unmodified to notifier function
1513 * @dev: net_device pointer passed unmodified to notifier function
1515 * Call all network notifier blocks. Parameters and return value
1516 * are as for raw_notifier_call_chain().
1519 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1522 return raw_notifier_call_chain(&netdev_chain, val, dev);
1524 EXPORT_SYMBOL(call_netdevice_notifiers);
1526 static struct static_key netstamp_needed __read_mostly;
1527 #ifdef HAVE_JUMP_LABEL
1528 /* We are not allowed to call static_key_slow_dec() from irq context
1529 * If net_disable_timestamp() is called from irq context, defer the
1530 * static_key_slow_dec() calls.
1532 static atomic_t netstamp_needed_deferred;
1535 void net_enable_timestamp(void)
1537 #ifdef HAVE_JUMP_LABEL
1538 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1542 static_key_slow_dec(&netstamp_needed);
1546 static_key_slow_inc(&netstamp_needed);
1548 EXPORT_SYMBOL(net_enable_timestamp);
1550 void net_disable_timestamp(void)
1552 #ifdef HAVE_JUMP_LABEL
1553 if (in_interrupt()) {
1554 atomic_inc(&netstamp_needed_deferred);
1558 static_key_slow_dec(&netstamp_needed);
1560 EXPORT_SYMBOL(net_disable_timestamp);
1562 static inline void net_timestamp_set(struct sk_buff *skb)
1564 skb->tstamp.tv64 = 0;
1565 if (static_key_false(&netstamp_needed))
1566 __net_timestamp(skb);
1569 #define net_timestamp_check(COND, SKB) \
1570 if (static_key_false(&netstamp_needed)) { \
1571 if ((COND) && !(SKB)->tstamp.tv64) \
1572 __net_timestamp(SKB); \
1575 static inline bool is_skb_forwardable(struct net_device *dev,
1576 struct sk_buff *skb)
1580 if (!(dev->flags & IFF_UP))
1583 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1584 if (skb->len <= len)
1587 /* if TSO is enabled, we don't care about the length as the packet
1588 * could be forwarded without being segmented before
1590 if (skb_is_gso(skb))
1597 * dev_forward_skb - loopback an skb to another netif
1599 * @dev: destination network device
1600 * @skb: buffer to forward
1603 * NET_RX_SUCCESS (no congestion)
1604 * NET_RX_DROP (packet was dropped, but freed)
1606 * dev_forward_skb can be used for injecting an skb from the
1607 * start_xmit function of one device into the receive queue
1608 * of another device.
1610 * The receiving device may be in another namespace, so
1611 * we have to clear all information in the skb that could
1612 * impact namespace isolation.
1614 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1616 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1617 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1618 atomic_long_inc(&dev->rx_dropped);
1626 if (unlikely(!is_skb_forwardable(dev, skb))) {
1627 atomic_long_inc(&dev->rx_dropped);
1634 skb->tstamp.tv64 = 0;
1635 skb->pkt_type = PACKET_HOST;
1636 skb->protocol = eth_type_trans(skb, dev);
1640 nf_reset_trace(skb);
1641 return netif_rx(skb);
1643 EXPORT_SYMBOL_GPL(dev_forward_skb);
1645 static inline int deliver_skb(struct sk_buff *skb,
1646 struct packet_type *pt_prev,
1647 struct net_device *orig_dev)
1649 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1651 atomic_inc(&skb->users);
1652 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1655 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1657 if (!ptype->af_packet_priv || !skb->sk)
1660 if (ptype->id_match)
1661 return ptype->id_match(ptype, skb->sk);
1662 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1669 * Support routine. Sends outgoing frames to any network
1670 * taps currently in use.
1673 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1675 struct packet_type *ptype;
1676 struct sk_buff *skb2 = NULL;
1677 struct packet_type *pt_prev = NULL;
1680 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1681 /* Never send packets back to the socket
1682 * they originated from - MvS (miquels@drinkel.ow.org)
1684 if ((ptype->dev == dev || !ptype->dev) &&
1685 (!skb_loop_sk(ptype, skb))) {
1687 deliver_skb(skb2, pt_prev, skb->dev);
1692 skb2 = skb_clone(skb, GFP_ATOMIC);
1696 net_timestamp_set(skb2);
1698 /* skb->nh should be correctly
1699 set by sender, so that the second statement is
1700 just protection against buggy protocols.
1702 skb_reset_mac_header(skb2);
1704 if (skb_network_header(skb2) < skb2->data ||
1705 skb2->network_header > skb2->tail) {
1706 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1707 ntohs(skb2->protocol),
1709 skb_reset_network_header(skb2);
1712 skb2->transport_header = skb2->network_header;
1713 skb2->pkt_type = PACKET_OUTGOING;
1718 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1723 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1724 * @dev: Network device
1725 * @txq: number of queues available
1727 * If real_num_tx_queues is changed the tc mappings may no longer be
1728 * valid. To resolve this verify the tc mapping remains valid and if
1729 * not NULL the mapping. With no priorities mapping to this
1730 * offset/count pair it will no longer be used. In the worst case TC0
1731 * is invalid nothing can be done so disable priority mappings. If is
1732 * expected that drivers will fix this mapping if they can before
1733 * calling netif_set_real_num_tx_queues.
1735 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1738 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1740 /* If TC0 is invalidated disable TC mapping */
1741 if (tc->offset + tc->count > txq) {
1742 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1747 /* Invalidated prio to tc mappings set to TC0 */
1748 for (i = 1; i < TC_BITMASK + 1; i++) {
1749 int q = netdev_get_prio_tc_map(dev, i);
1751 tc = &dev->tc_to_txq[q];
1752 if (tc->offset + tc->count > txq) {
1753 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1755 netdev_set_prio_tc_map(dev, i, 0);
1761 static DEFINE_MUTEX(xps_map_mutex);
1762 #define xmap_dereference(P) \
1763 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1765 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1768 struct xps_map *map = NULL;
1772 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1774 for (pos = 0; map && pos < map->len; pos++) {
1775 if (map->queues[pos] == index) {
1777 map->queues[pos] = map->queues[--map->len];
1779 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1780 kfree_rcu(map, rcu);
1790 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1792 struct xps_dev_maps *dev_maps;
1794 bool active = false;
1796 mutex_lock(&xps_map_mutex);
1797 dev_maps = xmap_dereference(dev->xps_maps);
1802 for_each_possible_cpu(cpu) {
1803 for (i = index; i < dev->num_tx_queues; i++) {
1804 if (!remove_xps_queue(dev_maps, cpu, i))
1807 if (i == dev->num_tx_queues)
1812 RCU_INIT_POINTER(dev->xps_maps, NULL);
1813 kfree_rcu(dev_maps, rcu);
1816 for (i = index; i < dev->num_tx_queues; i++)
1817 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1821 mutex_unlock(&xps_map_mutex);
1824 static struct xps_map *expand_xps_map(struct xps_map *map,
1827 struct xps_map *new_map;
1828 int alloc_len = XPS_MIN_MAP_ALLOC;
1831 for (pos = 0; map && pos < map->len; pos++) {
1832 if (map->queues[pos] != index)
1837 /* Need to add queue to this CPU's existing map */
1839 if (pos < map->alloc_len)
1842 alloc_len = map->alloc_len * 2;
1845 /* Need to allocate new map to store queue on this CPU's map */
1846 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1851 for (i = 0; i < pos; i++)
1852 new_map->queues[i] = map->queues[i];
1853 new_map->alloc_len = alloc_len;
1859 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1861 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1862 struct xps_map *map, *new_map;
1863 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1864 int cpu, numa_node_id = -2;
1865 bool active = false;
1867 mutex_lock(&xps_map_mutex);
1869 dev_maps = xmap_dereference(dev->xps_maps);
1871 /* allocate memory for queue storage */
1872 for_each_online_cpu(cpu) {
1873 if (!cpumask_test_cpu(cpu, mask))
1877 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1878 if (!new_dev_maps) {
1879 mutex_unlock(&xps_map_mutex);
1883 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1886 map = expand_xps_map(map, cpu, index);
1890 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1894 goto out_no_new_maps;
1896 for_each_possible_cpu(cpu) {
1897 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1898 /* add queue to CPU maps */
1901 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1902 while ((pos < map->len) && (map->queues[pos] != index))
1905 if (pos == map->len)
1906 map->queues[map->len++] = index;
1908 if (numa_node_id == -2)
1909 numa_node_id = cpu_to_node(cpu);
1910 else if (numa_node_id != cpu_to_node(cpu))
1913 } else if (dev_maps) {
1914 /* fill in the new device map from the old device map */
1915 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1916 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1921 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1923 /* Cleanup old maps */
1925 for_each_possible_cpu(cpu) {
1926 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1927 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1928 if (map && map != new_map)
1929 kfree_rcu(map, rcu);
1932 kfree_rcu(dev_maps, rcu);
1935 dev_maps = new_dev_maps;
1939 /* update Tx queue numa node */
1940 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1941 (numa_node_id >= 0) ? numa_node_id :
1947 /* removes queue from unused CPUs */
1948 for_each_possible_cpu(cpu) {
1949 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1952 if (remove_xps_queue(dev_maps, cpu, index))
1956 /* free map if not active */
1958 RCU_INIT_POINTER(dev->xps_maps, NULL);
1959 kfree_rcu(dev_maps, rcu);
1963 mutex_unlock(&xps_map_mutex);
1967 /* remove any maps that we added */
1968 for_each_possible_cpu(cpu) {
1969 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1970 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1972 if (new_map && new_map != map)
1976 mutex_unlock(&xps_map_mutex);
1978 kfree(new_dev_maps);
1981 EXPORT_SYMBOL(netif_set_xps_queue);
1985 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1986 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1988 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1992 if (txq < 1 || txq > dev->num_tx_queues)
1995 if (dev->reg_state == NETREG_REGISTERED ||
1996 dev->reg_state == NETREG_UNREGISTERING) {
1999 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2005 netif_setup_tc(dev, txq);
2007 if (txq < dev->real_num_tx_queues) {
2008 qdisc_reset_all_tx_gt(dev, txq);
2010 netif_reset_xps_queues_gt(dev, txq);
2015 dev->real_num_tx_queues = txq;
2018 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2022 * netif_set_real_num_rx_queues - set actual number of RX queues used
2023 * @dev: Network device
2024 * @rxq: Actual number of RX queues
2026 * This must be called either with the rtnl_lock held or before
2027 * registration of the net device. Returns 0 on success, or a
2028 * negative error code. If called before registration, it always
2031 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2035 if (rxq < 1 || rxq > dev->num_rx_queues)
2038 if (dev->reg_state == NETREG_REGISTERED) {
2041 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2047 dev->real_num_rx_queues = rxq;
2050 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2054 * netif_get_num_default_rss_queues - default number of RSS queues
2056 * This routine should set an upper limit on the number of RSS queues
2057 * used by default by multiqueue devices.
2059 int netif_get_num_default_rss_queues(void)
2061 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2063 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2065 static inline void __netif_reschedule(struct Qdisc *q)
2067 struct softnet_data *sd;
2068 unsigned long flags;
2070 local_irq_save(flags);
2071 sd = &__get_cpu_var(softnet_data);
2072 q->next_sched = NULL;
2073 *sd->output_queue_tailp = q;
2074 sd->output_queue_tailp = &q->next_sched;
2075 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2076 local_irq_restore(flags);
2079 void __netif_schedule(struct Qdisc *q)
2081 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2082 __netif_reschedule(q);
2084 EXPORT_SYMBOL(__netif_schedule);
2086 void dev_kfree_skb_irq(struct sk_buff *skb)
2088 if (atomic_dec_and_test(&skb->users)) {
2089 struct softnet_data *sd;
2090 unsigned long flags;
2092 local_irq_save(flags);
2093 sd = &__get_cpu_var(softnet_data);
2094 skb->next = sd->completion_queue;
2095 sd->completion_queue = skb;
2096 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2097 local_irq_restore(flags);
2100 EXPORT_SYMBOL(dev_kfree_skb_irq);
2102 void dev_kfree_skb_any(struct sk_buff *skb)
2104 if (in_irq() || irqs_disabled())
2105 dev_kfree_skb_irq(skb);
2109 EXPORT_SYMBOL(dev_kfree_skb_any);
2113 * netif_device_detach - mark device as removed
2114 * @dev: network device
2116 * Mark device as removed from system and therefore no longer available.
2118 void netif_device_detach(struct net_device *dev)
2120 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2121 netif_running(dev)) {
2122 netif_tx_stop_all_queues(dev);
2125 EXPORT_SYMBOL(netif_device_detach);
2128 * netif_device_attach - mark device as attached
2129 * @dev: network device
2131 * Mark device as attached from system and restart if needed.
2133 void netif_device_attach(struct net_device *dev)
2135 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2136 netif_running(dev)) {
2137 netif_tx_wake_all_queues(dev);
2138 __netdev_watchdog_up(dev);
2141 EXPORT_SYMBOL(netif_device_attach);
2143 static void skb_warn_bad_offload(const struct sk_buff *skb)
2145 static const netdev_features_t null_features = 0;
2146 struct net_device *dev = skb->dev;
2147 const char *driver = "";
2149 if (!net_ratelimit())
2152 if (dev && dev->dev.parent)
2153 driver = dev_driver_string(dev->dev.parent);
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);
2164 * Invalidate hardware checksum when packet is to be mangled, and
2165 * complete checksum manually on outgoing path.
2167 int skb_checksum_help(struct sk_buff *skb)
2170 int ret = 0, offset;
2172 if (skb->ip_summed == CHECKSUM_COMPLETE)
2173 goto out_set_summed;
2175 if (unlikely(skb_shinfo(skb)->gso_size)) {
2176 skb_warn_bad_offload(skb);
2180 /* Before computing a checksum, we should make sure no frag could
2181 * be modified by an external entity : checksum could be wrong.
2183 if (skb_has_shared_frag(skb)) {
2184 ret = __skb_linearize(skb);
2189 offset = skb_checksum_start_offset(skb);
2190 BUG_ON(offset >= skb_headlen(skb));
2191 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2193 offset += skb->csum_offset;
2194 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2196 if (skb_cloned(skb) &&
2197 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2198 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2203 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2205 skb->ip_summed = CHECKSUM_NONE;
2209 EXPORT_SYMBOL(skb_checksum_help);
2211 __be16 skb_network_protocol(struct sk_buff *skb)
2213 __be16 type = skb->protocol;
2214 int vlan_depth = ETH_HLEN;
2216 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2217 struct vlan_hdr *vh;
2219 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2222 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2223 type = vh->h_vlan_encapsulated_proto;
2224 vlan_depth += VLAN_HLEN;
2231 * skb_mac_gso_segment - mac layer segmentation handler.
2232 * @skb: buffer to segment
2233 * @features: features for the output path (see dev->features)
2235 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2236 netdev_features_t features)
2238 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2239 struct packet_offload *ptype;
2240 __be16 type = skb_network_protocol(skb);
2242 if (unlikely(!type))
2243 return ERR_PTR(-EINVAL);
2245 __skb_pull(skb, skb->mac_len);
2248 list_for_each_entry_rcu(ptype, &offload_base, list) {
2249 if (ptype->type == type && ptype->callbacks.gso_segment) {
2250 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2253 err = ptype->callbacks.gso_send_check(skb);
2254 segs = ERR_PTR(err);
2255 if (err || skb_gso_ok(skb, features))
2257 __skb_push(skb, (skb->data -
2258 skb_network_header(skb)));
2260 segs = ptype->callbacks.gso_segment(skb, features);
2266 __skb_push(skb, skb->data - skb_mac_header(skb));
2270 EXPORT_SYMBOL(skb_mac_gso_segment);
2273 /* openvswitch calls this on rx path, so we need a different check.
2275 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2278 return skb->ip_summed != CHECKSUM_PARTIAL;
2280 return skb->ip_summed == CHECKSUM_NONE;
2284 * __skb_gso_segment - Perform segmentation on skb.
2285 * @skb: buffer to segment
2286 * @features: features for the output path (see dev->features)
2287 * @tx_path: whether it is called in TX path
2289 * This function segments the given skb and returns a list of segments.
2291 * It may return NULL if the skb requires no segmentation. This is
2292 * only possible when GSO is used for verifying header integrity.
2294 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2295 netdev_features_t features, bool tx_path)
2297 if (unlikely(skb_needs_check(skb, tx_path))) {
2300 skb_warn_bad_offload(skb);
2302 if (skb_header_cloned(skb) &&
2303 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2304 return ERR_PTR(err);
2307 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2308 skb_reset_mac_header(skb);
2309 skb_reset_mac_len(skb);
2311 return skb_mac_gso_segment(skb, features);
2313 EXPORT_SYMBOL(__skb_gso_segment);
2315 /* Take action when hardware reception checksum errors are detected. */
2317 void netdev_rx_csum_fault(struct net_device *dev)
2319 if (net_ratelimit()) {
2320 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2324 EXPORT_SYMBOL(netdev_rx_csum_fault);
2327 /* Actually, we should eliminate this check as soon as we know, that:
2328 * 1. IOMMU is present and allows to map all the memory.
2329 * 2. No high memory really exists on this machine.
2332 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2334 #ifdef CONFIG_HIGHMEM
2336 if (!(dev->features & NETIF_F_HIGHDMA)) {
2337 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2338 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2339 if (PageHighMem(skb_frag_page(frag)))
2344 if (PCI_DMA_BUS_IS_PHYS) {
2345 struct device *pdev = dev->dev.parent;
2349 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2350 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2351 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2352 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2361 void (*destructor)(struct sk_buff *skb);
2364 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2366 static void dev_gso_skb_destructor(struct sk_buff *skb)
2368 struct dev_gso_cb *cb;
2371 struct sk_buff *nskb = skb->next;
2373 skb->next = nskb->next;
2376 } while (skb->next);
2378 cb = DEV_GSO_CB(skb);
2380 cb->destructor(skb);
2384 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2385 * @skb: buffer to segment
2386 * @features: device features as applicable to this skb
2388 * This function segments the given skb and stores the list of segments
2391 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2393 struct sk_buff *segs;
2395 segs = skb_gso_segment(skb, features);
2397 /* Verifying header integrity only. */
2402 return PTR_ERR(segs);
2405 DEV_GSO_CB(skb)->destructor = skb->destructor;
2406 skb->destructor = dev_gso_skb_destructor;
2411 static netdev_features_t harmonize_features(struct sk_buff *skb,
2412 __be16 protocol, netdev_features_t features)
2414 if (skb->ip_summed != CHECKSUM_NONE &&
2415 !can_checksum_protocol(features, protocol)) {
2416 features &= ~NETIF_F_ALL_CSUM;
2417 } else if (illegal_highdma(skb->dev, skb)) {
2418 features &= ~NETIF_F_SG;
2424 netdev_features_t netif_skb_features(struct sk_buff *skb)
2426 __be16 protocol = skb->protocol;
2427 netdev_features_t features = skb->dev->features;
2429 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2430 features &= ~NETIF_F_GSO_MASK;
2432 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2433 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2434 protocol = veh->h_vlan_encapsulated_proto;
2435 } else if (!vlan_tx_tag_present(skb)) {
2436 return harmonize_features(skb, protocol, features);
2439 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2440 NETIF_F_HW_VLAN_STAG_TX);
2442 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2443 return harmonize_features(skb, protocol, features);
2445 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2446 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2447 NETIF_F_HW_VLAN_STAG_TX;
2448 return harmonize_features(skb, protocol, features);
2451 EXPORT_SYMBOL(netif_skb_features);
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.
2458 static inline int skb_needs_linearize(struct sk_buff *skb,
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)));
2468 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2469 struct netdev_queue *txq)
2471 const struct net_device_ops *ops = dev->netdev_ops;
2472 int rc = NETDEV_TX_OK;
2473 unsigned int skb_len;
2475 if (likely(!skb->next)) {
2476 netdev_features_t features;
2479 * If device doesn't need skb->dst, release it right now while
2480 * its hot in this cpu cache
2482 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2485 features = netif_skb_features(skb);
2487 if (vlan_tx_tag_present(skb) &&
2488 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2489 skb = __vlan_put_tag(skb, skb->vlan_proto,
2490 vlan_tx_tag_get(skb));
2497 /* If encapsulation offload request, verify we are testing
2498 * hardware encapsulation features instead of standard
2499 * features for the netdev
2501 if (skb->encapsulation)
2502 features &= dev->hw_enc_features;
2504 if (netif_needs_gso(skb, features)) {
2505 if (unlikely(dev_gso_segment(skb, features)))
2510 if (skb_needs_linearize(skb, features) &&
2511 __skb_linearize(skb))
2514 /* If packet is not checksummed and device does not
2515 * support checksumming for this protocol, complete
2516 * checksumming here.
2518 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2519 if (skb->encapsulation)
2520 skb_set_inner_transport_header(skb,
2521 skb_checksum_start_offset(skb));
2523 skb_set_transport_header(skb,
2524 skb_checksum_start_offset(skb));
2525 if (!(features & NETIF_F_ALL_CSUM) &&
2526 skb_checksum_help(skb))
2531 if (!list_empty(&ptype_all))
2532 dev_queue_xmit_nit(skb, dev);
2535 rc = ops->ndo_start_xmit(skb, dev);
2536 trace_net_dev_xmit(skb, rc, dev, skb_len);
2537 if (rc == NETDEV_TX_OK)
2538 txq_trans_update(txq);
2544 struct sk_buff *nskb = skb->next;
2546 skb->next = nskb->next;
2549 if (!list_empty(&ptype_all))
2550 dev_queue_xmit_nit(nskb, dev);
2552 skb_len = nskb->len;
2553 rc = ops->ndo_start_xmit(nskb, dev);
2554 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2555 if (unlikely(rc != NETDEV_TX_OK)) {
2556 if (rc & ~NETDEV_TX_MASK)
2557 goto out_kfree_gso_skb;
2558 nskb->next = skb->next;
2562 txq_trans_update(txq);
2563 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2564 return NETDEV_TX_BUSY;
2565 } while (skb->next);
2568 if (likely(skb->next == NULL)) {
2569 skb->destructor = DEV_GSO_CB(skb)->destructor;
2579 static void qdisc_pkt_len_init(struct sk_buff *skb)
2581 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2583 qdisc_skb_cb(skb)->pkt_len = skb->len;
2585 /* To get more precise estimation of bytes sent on wire,
2586 * we add to pkt_len the headers size of all segments
2588 if (shinfo->gso_size) {
2589 unsigned int hdr_len;
2590 u16 gso_segs = shinfo->gso_segs;
2592 /* mac layer + network layer */
2593 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2595 /* + transport layer */
2596 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2597 hdr_len += tcp_hdrlen(skb);
2599 hdr_len += sizeof(struct udphdr);
2601 if (shinfo->gso_type & SKB_GSO_DODGY)
2602 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2605 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2609 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2610 struct net_device *dev,
2611 struct netdev_queue *txq)
2613 spinlock_t *root_lock = qdisc_lock(q);
2617 qdisc_pkt_len_init(skb);
2618 qdisc_calculate_pkt_len(skb, q);
2620 * Heuristic to force contended enqueues to serialize on a
2621 * separate lock before trying to get qdisc main lock.
2622 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2623 * and dequeue packets faster.
2625 contended = qdisc_is_running(q);
2626 if (unlikely(contended))
2627 spin_lock(&q->busylock);
2629 spin_lock(root_lock);
2630 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2633 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2634 qdisc_run_begin(q)) {
2636 * This is a work-conserving queue; there are no old skbs
2637 * waiting to be sent out; and the qdisc is not running -
2638 * xmit the skb directly.
2640 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2643 qdisc_bstats_update(q, skb);
2645 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2646 if (unlikely(contended)) {
2647 spin_unlock(&q->busylock);
2654 rc = NET_XMIT_SUCCESS;
2657 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2658 if (qdisc_run_begin(q)) {
2659 if (unlikely(contended)) {
2660 spin_unlock(&q->busylock);
2666 spin_unlock(root_lock);
2667 if (unlikely(contended))
2668 spin_unlock(&q->busylock);
2672 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2673 static void skb_update_prio(struct sk_buff *skb)
2675 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2677 if (!skb->priority && skb->sk && map) {
2678 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2680 if (prioidx < map->priomap_len)
2681 skb->priority = map->priomap[prioidx];
2685 #define skb_update_prio(skb)
2688 static DEFINE_PER_CPU(int, xmit_recursion);
2689 #define RECURSION_LIMIT 10
2692 * dev_loopback_xmit - loop back @skb
2693 * @skb: buffer to transmit
2695 int dev_loopback_xmit(struct sk_buff *skb)
2697 skb_reset_mac_header(skb);
2698 __skb_pull(skb, skb_network_offset(skb));
2699 skb->pkt_type = PACKET_LOOPBACK;
2700 skb->ip_summed = CHECKSUM_UNNECESSARY;
2701 WARN_ON(!skb_dst(skb));
2706 EXPORT_SYMBOL(dev_loopback_xmit);
2709 * dev_queue_xmit - transmit a buffer
2710 * @skb: buffer to transmit
2712 * Queue a buffer for transmission to a network device. The caller must
2713 * have set the device and priority and built the buffer before calling
2714 * this function. The function can be called from an interrupt.
2716 * A negative errno code is returned on a failure. A success does not
2717 * guarantee the frame will be transmitted as it may be dropped due
2718 * to congestion or traffic shaping.
2720 * -----------------------------------------------------------------------------------
2721 * I notice this method can also return errors from the queue disciplines,
2722 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2725 * Regardless of the return value, the skb is consumed, so it is currently
2726 * difficult to retry a send to this method. (You can bump the ref count
2727 * before sending to hold a reference for retry if you are careful.)
2729 * When calling this method, interrupts MUST be enabled. This is because
2730 * the BH enable code must have IRQs enabled so that it will not deadlock.
2733 int dev_queue_xmit(struct sk_buff *skb)
2735 struct net_device *dev = skb->dev;
2736 struct netdev_queue *txq;
2740 skb_reset_mac_header(skb);
2742 /* Disable soft irqs for various locks below. Also
2743 * stops preemption for RCU.
2747 skb_update_prio(skb);
2749 txq = netdev_pick_tx(dev, skb);
2750 q = rcu_dereference_bh(txq->qdisc);
2752 #ifdef CONFIG_NET_CLS_ACT
2753 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2755 trace_net_dev_queue(skb);
2757 rc = __dev_xmit_skb(skb, q, dev, txq);
2761 /* The device has no queue. Common case for software devices:
2762 loopback, all the sorts of tunnels...
2764 Really, it is unlikely that netif_tx_lock protection is necessary
2765 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2767 However, it is possible, that they rely on protection
2770 Check this and shot the lock. It is not prone from deadlocks.
2771 Either shot noqueue qdisc, it is even simpler 8)
2773 if (dev->flags & IFF_UP) {
2774 int cpu = smp_processor_id(); /* ok because BHs are off */
2776 if (txq->xmit_lock_owner != cpu) {
2778 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2779 goto recursion_alert;
2781 HARD_TX_LOCK(dev, txq, cpu);
2783 if (!netif_xmit_stopped(txq)) {
2784 __this_cpu_inc(xmit_recursion);
2785 rc = dev_hard_start_xmit(skb, dev, txq);
2786 __this_cpu_dec(xmit_recursion);
2787 if (dev_xmit_complete(rc)) {
2788 HARD_TX_UNLOCK(dev, txq);
2792 HARD_TX_UNLOCK(dev, txq);
2793 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2796 /* Recursion is detected! It is possible,
2800 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2806 rcu_read_unlock_bh();
2811 rcu_read_unlock_bh();
2814 EXPORT_SYMBOL(dev_queue_xmit);
2817 /*=======================================================================
2819 =======================================================================*/
2821 int netdev_max_backlog __read_mostly = 1000;
2822 EXPORT_SYMBOL(netdev_max_backlog);
2824 int netdev_tstamp_prequeue __read_mostly = 1;
2825 int netdev_budget __read_mostly = 300;
2826 int weight_p __read_mostly = 64; /* old backlog weight */
2828 /* Called with irq disabled */
2829 static inline void ____napi_schedule(struct softnet_data *sd,
2830 struct napi_struct *napi)
2832 list_add_tail(&napi->poll_list, &sd->poll_list);
2833 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2838 /* One global table that all flow-based protocols share. */
2839 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2840 EXPORT_SYMBOL(rps_sock_flow_table);
2842 struct static_key rps_needed __read_mostly;
2844 static struct rps_dev_flow *
2845 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2846 struct rps_dev_flow *rflow, u16 next_cpu)
2848 if (next_cpu != RPS_NO_CPU) {
2849 #ifdef CONFIG_RFS_ACCEL
2850 struct netdev_rx_queue *rxqueue;
2851 struct rps_dev_flow_table *flow_table;
2852 struct rps_dev_flow *old_rflow;
2857 /* Should we steer this flow to a different hardware queue? */
2858 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2859 !(dev->features & NETIF_F_NTUPLE))
2861 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2862 if (rxq_index == skb_get_rx_queue(skb))
2865 rxqueue = dev->_rx + rxq_index;
2866 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2869 flow_id = skb->rxhash & flow_table->mask;
2870 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2871 rxq_index, flow_id);
2875 rflow = &flow_table->flows[flow_id];
2877 if (old_rflow->filter == rflow->filter)
2878 old_rflow->filter = RPS_NO_FILTER;
2882 per_cpu(softnet_data, next_cpu).input_queue_head;
2885 rflow->cpu = next_cpu;
2890 * get_rps_cpu is called from netif_receive_skb and returns the target
2891 * CPU from the RPS map of the receiving queue for a given skb.
2892 * rcu_read_lock must be held on entry.
2894 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2895 struct rps_dev_flow **rflowp)
2897 struct netdev_rx_queue *rxqueue;
2898 struct rps_map *map;
2899 struct rps_dev_flow_table *flow_table;
2900 struct rps_sock_flow_table *sock_flow_table;
2904 if (skb_rx_queue_recorded(skb)) {
2905 u16 index = skb_get_rx_queue(skb);
2906 if (unlikely(index >= dev->real_num_rx_queues)) {
2907 WARN_ONCE(dev->real_num_rx_queues > 1,
2908 "%s received packet on queue %u, but number "
2909 "of RX queues is %u\n",
2910 dev->name, index, dev->real_num_rx_queues);
2913 rxqueue = dev->_rx + index;
2917 map = rcu_dereference(rxqueue->rps_map);
2919 if (map->len == 1 &&
2920 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2921 tcpu = map->cpus[0];
2922 if (cpu_online(tcpu))
2926 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2930 skb_reset_network_header(skb);
2931 if (!skb_get_rxhash(skb))
2934 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2935 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2936 if (flow_table && sock_flow_table) {
2938 struct rps_dev_flow *rflow;
2940 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2943 next_cpu = sock_flow_table->ents[skb->rxhash &
2944 sock_flow_table->mask];
2947 * If the desired CPU (where last recvmsg was done) is
2948 * different from current CPU (one in the rx-queue flow
2949 * table entry), switch if one of the following holds:
2950 * - Current CPU is unset (equal to RPS_NO_CPU).
2951 * - Current CPU is offline.
2952 * - The current CPU's queue tail has advanced beyond the
2953 * last packet that was enqueued using this table entry.
2954 * This guarantees that all previous packets for the flow
2955 * have been dequeued, thus preserving in order delivery.
2957 if (unlikely(tcpu != next_cpu) &&
2958 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2959 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2960 rflow->last_qtail)) >= 0)) {
2962 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2965 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2973 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2975 if (cpu_online(tcpu)) {
2985 #ifdef CONFIG_RFS_ACCEL
2988 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2989 * @dev: Device on which the filter was set
2990 * @rxq_index: RX queue index
2991 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2992 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2994 * Drivers that implement ndo_rx_flow_steer() should periodically call
2995 * this function for each installed filter and remove the filters for
2996 * which it returns %true.
2998 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2999 u32 flow_id, u16 filter_id)
3001 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3002 struct rps_dev_flow_table *flow_table;
3003 struct rps_dev_flow *rflow;
3008 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3009 if (flow_table && flow_id <= flow_table->mask) {
3010 rflow = &flow_table->flows[flow_id];
3011 cpu = ACCESS_ONCE(rflow->cpu);
3012 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3013 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3014 rflow->last_qtail) <
3015 (int)(10 * flow_table->mask)))
3021 EXPORT_SYMBOL(rps_may_expire_flow);
3023 #endif /* CONFIG_RFS_ACCEL */
3025 /* Called from hardirq (IPI) context */
3026 static void rps_trigger_softirq(void *data)
3028 struct softnet_data *sd = data;
3030 ____napi_schedule(sd, &sd->backlog);
3034 #endif /* CONFIG_RPS */
3037 * Check if this softnet_data structure is another cpu one
3038 * If yes, queue it to our IPI list and return 1
3041 static int rps_ipi_queued(struct softnet_data *sd)
3044 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3047 sd->rps_ipi_next = mysd->rps_ipi_list;
3048 mysd->rps_ipi_list = sd;
3050 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3053 #endif /* CONFIG_RPS */
3058 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3059 * queue (may be a remote CPU queue).
3061 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3062 unsigned int *qtail)
3064 struct softnet_data *sd;
3065 unsigned long flags;
3067 sd = &per_cpu(softnet_data, cpu);
3069 local_irq_save(flags);
3072 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
3073 if (skb_queue_len(&sd->input_pkt_queue)) {
3075 __skb_queue_tail(&sd->input_pkt_queue, skb);
3076 input_queue_tail_incr_save(sd, qtail);
3078 local_irq_restore(flags);
3079 return NET_RX_SUCCESS;
3082 /* Schedule NAPI for backlog device
3083 * We can use non atomic operation since we own the queue lock
3085 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3086 if (!rps_ipi_queued(sd))
3087 ____napi_schedule(sd, &sd->backlog);
3095 local_irq_restore(flags);
3097 atomic_long_inc(&skb->dev->rx_dropped);
3103 * netif_rx - post buffer to the network code
3104 * @skb: buffer to post
3106 * This function receives a packet from a device driver and queues it for
3107 * the upper (protocol) levels to process. It always succeeds. The buffer
3108 * may be dropped during processing for congestion control or by the
3112 * NET_RX_SUCCESS (no congestion)
3113 * NET_RX_DROP (packet was dropped)
3117 int netif_rx(struct sk_buff *skb)
3121 /* if netpoll wants it, pretend we never saw it */
3122 if (netpoll_rx(skb))
3125 net_timestamp_check(netdev_tstamp_prequeue, skb);
3127 trace_netif_rx(skb);
3129 if (static_key_false(&rps_needed)) {
3130 struct rps_dev_flow voidflow, *rflow = &voidflow;
3136 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3138 cpu = smp_processor_id();
3140 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3148 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3153 EXPORT_SYMBOL(netif_rx);
3155 int netif_rx_ni(struct sk_buff *skb)
3160 err = netif_rx(skb);
3161 if (local_softirq_pending())
3167 EXPORT_SYMBOL(netif_rx_ni);
3169 static void net_tx_action(struct softirq_action *h)
3171 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3173 if (sd->completion_queue) {
3174 struct sk_buff *clist;
3176 local_irq_disable();
3177 clist = sd->completion_queue;
3178 sd->completion_queue = NULL;
3182 struct sk_buff *skb = clist;
3183 clist = clist->next;
3185 WARN_ON(atomic_read(&skb->users));
3186 trace_kfree_skb(skb, net_tx_action);
3191 if (sd->output_queue) {
3194 local_irq_disable();
3195 head = sd->output_queue;
3196 sd->output_queue = NULL;
3197 sd->output_queue_tailp = &sd->output_queue;
3201 struct Qdisc *q = head;
3202 spinlock_t *root_lock;
3204 head = head->next_sched;
3206 root_lock = qdisc_lock(q);
3207 if (spin_trylock(root_lock)) {
3208 smp_mb__before_clear_bit();
3209 clear_bit(__QDISC_STATE_SCHED,
3212 spin_unlock(root_lock);
3214 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3216 __netif_reschedule(q);
3218 smp_mb__before_clear_bit();
3219 clear_bit(__QDISC_STATE_SCHED,
3227 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3228 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3229 /* This hook is defined here for ATM LANE */
3230 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3231 unsigned char *addr) __read_mostly;
3232 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3235 #ifdef CONFIG_NET_CLS_ACT
3236 /* TODO: Maybe we should just force sch_ingress to be compiled in
3237 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3238 * a compare and 2 stores extra right now if we dont have it on
3239 * but have CONFIG_NET_CLS_ACT
3240 * NOTE: This doesn't stop any functionality; if you dont have
3241 * the ingress scheduler, you just can't add policies on ingress.
3244 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3246 struct net_device *dev = skb->dev;
3247 u32 ttl = G_TC_RTTL(skb->tc_verd);
3248 int result = TC_ACT_OK;
3251 if (unlikely(MAX_RED_LOOP < ttl++)) {
3252 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3253 skb->skb_iif, dev->ifindex);
3257 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3258 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3261 if (q != &noop_qdisc) {
3262 spin_lock(qdisc_lock(q));
3263 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3264 result = qdisc_enqueue_root(skb, q);
3265 spin_unlock(qdisc_lock(q));
3271 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3272 struct packet_type **pt_prev,
3273 int *ret, struct net_device *orig_dev)
3275 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3277 if (!rxq || rxq->qdisc == &noop_qdisc)
3281 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3285 switch (ing_filter(skb, rxq)) {
3299 * netdev_rx_handler_register - register receive handler
3300 * @dev: device to register a handler for
3301 * @rx_handler: receive handler to register
3302 * @rx_handler_data: data pointer that is used by rx handler
3304 * Register a receive hander for a device. This handler will then be
3305 * called from __netif_receive_skb. A negative errno code is returned
3308 * The caller must hold the rtnl_mutex.
3310 * For a general description of rx_handler, see enum rx_handler_result.
3312 int netdev_rx_handler_register(struct net_device *dev,
3313 rx_handler_func_t *rx_handler,
3314 void *rx_handler_data)
3318 if (dev->rx_handler)
3321 /* Note: rx_handler_data must be set before rx_handler */
3322 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3323 rcu_assign_pointer(dev->rx_handler, rx_handler);
3327 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3330 * netdev_rx_handler_unregister - unregister receive handler
3331 * @dev: device to unregister a handler from
3333 * Unregister a receive handler from a device.
3335 * The caller must hold the rtnl_mutex.
3337 void netdev_rx_handler_unregister(struct net_device *dev)
3341 RCU_INIT_POINTER(dev->rx_handler, NULL);
3342 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3343 * section has a guarantee to see a non NULL rx_handler_data
3347 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3349 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3352 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3353 * the special handling of PFMEMALLOC skbs.
3355 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3357 switch (skb->protocol) {
3358 case __constant_htons(ETH_P_ARP):
3359 case __constant_htons(ETH_P_IP):
3360 case __constant_htons(ETH_P_IPV6):
3361 case __constant_htons(ETH_P_8021Q):
3362 case __constant_htons(ETH_P_8021AD):
3369 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3371 struct packet_type *ptype, *pt_prev;
3372 rx_handler_func_t *rx_handler;
3373 struct net_device *orig_dev;
3374 struct net_device *null_or_dev;
3375 bool deliver_exact = false;
3376 int ret = NET_RX_DROP;
3379 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3381 trace_netif_receive_skb(skb);
3383 /* if we've gotten here through NAPI, check netpoll */
3384 if (netpoll_receive_skb(skb))
3387 orig_dev = skb->dev;
3389 skb_reset_network_header(skb);
3390 if (!skb_transport_header_was_set(skb))
3391 skb_reset_transport_header(skb);
3392 skb_reset_mac_len(skb);
3399 skb->skb_iif = skb->dev->ifindex;
3401 __this_cpu_inc(softnet_data.processed);
3403 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3404 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3405 skb = vlan_untag(skb);
3410 #ifdef CONFIG_NET_CLS_ACT
3411 if (skb->tc_verd & TC_NCLS) {
3412 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3420 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3421 if (!ptype->dev || ptype->dev == skb->dev) {
3423 ret = deliver_skb(skb, pt_prev, orig_dev);
3429 #ifdef CONFIG_NET_CLS_ACT
3430 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3436 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3439 if (vlan_tx_tag_present(skb)) {
3441 ret = deliver_skb(skb, pt_prev, orig_dev);
3444 if (vlan_do_receive(&skb))
3446 else if (unlikely(!skb))
3450 rx_handler = rcu_dereference(skb->dev->rx_handler);
3453 ret = deliver_skb(skb, pt_prev, orig_dev);
3456 switch (rx_handler(&skb)) {
3457 case RX_HANDLER_CONSUMED:
3458 ret = NET_RX_SUCCESS;
3460 case RX_HANDLER_ANOTHER:
3462 case RX_HANDLER_EXACT:
3463 deliver_exact = true;
3464 case RX_HANDLER_PASS:
3471 if (vlan_tx_nonzero_tag_present(skb))
3472 skb->pkt_type = PACKET_OTHERHOST;
3474 /* deliver only exact match when indicated */
3475 null_or_dev = deliver_exact ? skb->dev : NULL;
3477 type = skb->protocol;
3478 list_for_each_entry_rcu(ptype,
3479 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3480 if (ptype->type == type &&
3481 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3482 ptype->dev == orig_dev)) {
3484 ret = deliver_skb(skb, pt_prev, orig_dev);
3490 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3493 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3496 atomic_long_inc(&skb->dev->rx_dropped);
3498 /* Jamal, now you will not able to escape explaining
3499 * me how you were going to use this. :-)
3510 static int __netif_receive_skb(struct sk_buff *skb)
3514 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3515 unsigned long pflags = current->flags;
3518 * PFMEMALLOC skbs are special, they should
3519 * - be delivered to SOCK_MEMALLOC sockets only
3520 * - stay away from userspace
3521 * - have bounded memory usage
3523 * Use PF_MEMALLOC as this saves us from propagating the allocation
3524 * context down to all allocation sites.
3526 current->flags |= PF_MEMALLOC;
3527 ret = __netif_receive_skb_core(skb, true);
3528 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3530 ret = __netif_receive_skb_core(skb, false);
3536 * netif_receive_skb - process receive buffer from network
3537 * @skb: buffer to process
3539 * netif_receive_skb() is the main receive data processing function.
3540 * It always succeeds. The buffer may be dropped during processing
3541 * for congestion control or by the protocol layers.
3543 * This function may only be called from softirq context and interrupts
3544 * should be enabled.
3546 * Return values (usually ignored):
3547 * NET_RX_SUCCESS: no congestion
3548 * NET_RX_DROP: packet was dropped
3550 int netif_receive_skb(struct sk_buff *skb)
3552 net_timestamp_check(netdev_tstamp_prequeue, skb);
3554 if (skb_defer_rx_timestamp(skb))
3555 return NET_RX_SUCCESS;
3558 if (static_key_false(&rps_needed)) {
3559 struct rps_dev_flow voidflow, *rflow = &voidflow;
3564 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3567 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3574 return __netif_receive_skb(skb);
3576 EXPORT_SYMBOL(netif_receive_skb);
3578 /* Network device is going away, flush any packets still pending
3579 * Called with irqs disabled.
3581 static void flush_backlog(void *arg)
3583 struct net_device *dev = arg;
3584 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3585 struct sk_buff *skb, *tmp;
3588 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3589 if (skb->dev == dev) {
3590 __skb_unlink(skb, &sd->input_pkt_queue);
3592 input_queue_head_incr(sd);
3597 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3598 if (skb->dev == dev) {
3599 __skb_unlink(skb, &sd->process_queue);
3601 input_queue_head_incr(sd);
3606 static int napi_gro_complete(struct sk_buff *skb)
3608 struct packet_offload *ptype;
3609 __be16 type = skb->protocol;
3610 struct list_head *head = &offload_base;
3613 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3615 if (NAPI_GRO_CB(skb)->count == 1) {
3616 skb_shinfo(skb)->gso_size = 0;
3621 list_for_each_entry_rcu(ptype, head, list) {
3622 if (ptype->type != type || !ptype->callbacks.gro_complete)
3625 err = ptype->callbacks.gro_complete(skb);
3631 WARN_ON(&ptype->list == head);
3633 return NET_RX_SUCCESS;
3637 return netif_receive_skb(skb);
3640 /* napi->gro_list contains packets ordered by age.
3641 * youngest packets at the head of it.
3642 * Complete skbs in reverse order to reduce latencies.
3644 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3646 struct sk_buff *skb, *prev = NULL;
3648 /* scan list and build reverse chain */
3649 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3654 for (skb = prev; skb; skb = prev) {
3657 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3661 napi_gro_complete(skb);
3665 napi->gro_list = NULL;
3667 EXPORT_SYMBOL(napi_gro_flush);
3669 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3672 unsigned int maclen = skb->dev->hard_header_len;
3674 for (p = napi->gro_list; p; p = p->next) {
3675 unsigned long diffs;
3677 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3678 diffs |= p->vlan_tci ^ skb->vlan_tci;
3679 if (maclen == ETH_HLEN)
3680 diffs |= compare_ether_header(skb_mac_header(p),
3681 skb_gro_mac_header(skb));
3683 diffs = memcmp(skb_mac_header(p),
3684 skb_gro_mac_header(skb),
3686 NAPI_GRO_CB(p)->same_flow = !diffs;
3687 NAPI_GRO_CB(p)->flush = 0;
3691 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3693 struct sk_buff **pp = NULL;
3694 struct packet_offload *ptype;
3695 __be16 type = skb->protocol;
3696 struct list_head *head = &offload_base;
3698 enum gro_result ret;
3700 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3703 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3706 gro_list_prepare(napi, skb);
3709 list_for_each_entry_rcu(ptype, head, list) {
3710 if (ptype->type != type || !ptype->callbacks.gro_receive)
3713 skb_set_network_header(skb, skb_gro_offset(skb));
3714 skb_reset_mac_len(skb);
3715 NAPI_GRO_CB(skb)->same_flow = 0;
3716 NAPI_GRO_CB(skb)->flush = 0;
3717 NAPI_GRO_CB(skb)->free = 0;
3719 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3724 if (&ptype->list == head)
3727 same_flow = NAPI_GRO_CB(skb)->same_flow;
3728 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3731 struct sk_buff *nskb = *pp;
3735 napi_gro_complete(nskb);
3742 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3746 NAPI_GRO_CB(skb)->count = 1;
3747 NAPI_GRO_CB(skb)->age = jiffies;
3748 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3749 skb->next = napi->gro_list;
3750 napi->gro_list = skb;
3754 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3755 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3757 BUG_ON(skb->end - skb->tail < grow);
3759 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3762 skb->data_len -= grow;
3764 skb_shinfo(skb)->frags[0].page_offset += grow;
3765 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3767 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3768 skb_frag_unref(skb, 0);
3769 memmove(skb_shinfo(skb)->frags,
3770 skb_shinfo(skb)->frags + 1,
3771 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3784 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3788 if (netif_receive_skb(skb))
3796 case GRO_MERGED_FREE:
3797 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3798 kmem_cache_free(skbuff_head_cache, skb);
3811 static void skb_gro_reset_offset(struct sk_buff *skb)
3813 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3814 const skb_frag_t *frag0 = &pinfo->frags[0];
3816 NAPI_GRO_CB(skb)->data_offset = 0;
3817 NAPI_GRO_CB(skb)->frag0 = NULL;
3818 NAPI_GRO_CB(skb)->frag0_len = 0;
3820 if (skb->mac_header == skb->tail &&
3822 !PageHighMem(skb_frag_page(frag0))) {
3823 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3824 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3828 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3830 skb_gro_reset_offset(skb);
3832 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3834 EXPORT_SYMBOL(napi_gro_receive);
3836 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3838 __skb_pull(skb, skb_headlen(skb));
3839 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3840 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3842 skb->dev = napi->dev;
3848 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3850 struct sk_buff *skb = napi->skb;
3853 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3859 EXPORT_SYMBOL(napi_get_frags);
3861 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3867 skb->protocol = eth_type_trans(skb, skb->dev);
3869 if (ret == GRO_HELD)
3870 skb_gro_pull(skb, -ETH_HLEN);
3871 else if (netif_receive_skb(skb))
3876 case GRO_MERGED_FREE:
3877 napi_reuse_skb(napi, skb);
3887 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3889 struct sk_buff *skb = napi->skb;
3896 skb_reset_mac_header(skb);
3897 skb_gro_reset_offset(skb);
3899 off = skb_gro_offset(skb);
3900 hlen = off + sizeof(*eth);
3901 eth = skb_gro_header_fast(skb, off);
3902 if (skb_gro_header_hard(skb, hlen)) {
3903 eth = skb_gro_header_slow(skb, hlen, off);
3904 if (unlikely(!eth)) {
3905 napi_reuse_skb(napi, skb);
3911 skb_gro_pull(skb, sizeof(*eth));
3914 * This works because the only protocols we care about don't require
3915 * special handling. We'll fix it up properly at the end.
3917 skb->protocol = eth->h_proto;
3923 gro_result_t napi_gro_frags(struct napi_struct *napi)
3925 struct sk_buff *skb = napi_frags_skb(napi);
3930 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
3932 EXPORT_SYMBOL(napi_gro_frags);
3935 * net_rps_action sends any pending IPI's for rps.
3936 * Note: called with local irq disabled, but exits with local irq enabled.
3938 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3941 struct softnet_data *remsd = sd->rps_ipi_list;
3944 sd->rps_ipi_list = NULL;
3948 /* Send pending IPI's to kick RPS processing on remote cpus. */
3950 struct softnet_data *next = remsd->rps_ipi_next;
3952 if (cpu_online(remsd->cpu))
3953 __smp_call_function_single(remsd->cpu,
3962 static int process_backlog(struct napi_struct *napi, int quota)
3965 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3968 /* Check if we have pending ipi, its better to send them now,
3969 * not waiting net_rx_action() end.
3971 if (sd->rps_ipi_list) {
3972 local_irq_disable();
3973 net_rps_action_and_irq_enable(sd);
3976 napi->weight = weight_p;
3977 local_irq_disable();
3978 while (work < quota) {
3979 struct sk_buff *skb;
3982 while ((skb = __skb_dequeue(&sd->process_queue))) {
3984 __netif_receive_skb(skb);
3985 local_irq_disable();
3986 input_queue_head_incr(sd);
3987 if (++work >= quota) {
3994 qlen = skb_queue_len(&sd->input_pkt_queue);
3996 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3997 &sd->process_queue);
3999 if (qlen < quota - work) {
4001 * Inline a custom version of __napi_complete().
4002 * only current cpu owns and manipulates this napi,
4003 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4004 * we can use a plain write instead of clear_bit(),
4005 * and we dont need an smp_mb() memory barrier.
4007 list_del(&napi->poll_list);
4010 quota = work + qlen;
4020 * __napi_schedule - schedule for receive
4021 * @n: entry to schedule
4023 * The entry's receive function will be scheduled to run
4025 void __napi_schedule(struct napi_struct *n)
4027 unsigned long flags;
4029 local_irq_save(flags);
4030 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4031 local_irq_restore(flags);
4033 EXPORT_SYMBOL(__napi_schedule);
4035 void __napi_complete(struct napi_struct *n)
4037 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4038 BUG_ON(n->gro_list);
4040 list_del(&n->poll_list);
4041 smp_mb__before_clear_bit();
4042 clear_bit(NAPI_STATE_SCHED, &n->state);
4044 EXPORT_SYMBOL(__napi_complete);
4046 void napi_complete(struct napi_struct *n)
4048 unsigned long flags;
4051 * don't let napi dequeue from the cpu poll list
4052 * just in case its running on a different cpu
4054 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4057 napi_gro_flush(n, false);
4058 local_irq_save(flags);
4060 local_irq_restore(flags);
4062 EXPORT_SYMBOL(napi_complete);
4064 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4065 int (*poll)(struct napi_struct *, int), int weight)
4067 INIT_LIST_HEAD(&napi->poll_list);
4068 napi->gro_count = 0;
4069 napi->gro_list = NULL;
4072 if (weight > NAPI_POLL_WEIGHT)
4073 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4075 napi->weight = weight;
4076 list_add(&napi->dev_list, &dev->napi_list);
4078 #ifdef CONFIG_NETPOLL
4079 spin_lock_init(&napi->poll_lock);
4080 napi->poll_owner = -1;
4082 set_bit(NAPI_STATE_SCHED, &napi->state);
4084 EXPORT_SYMBOL(netif_napi_add);
4086 void netif_napi_del(struct napi_struct *napi)
4088 struct sk_buff *skb, *next;
4090 list_del_init(&napi->dev_list);
4091 napi_free_frags(napi);
4093 for (skb = napi->gro_list; skb; skb = next) {
4099 napi->gro_list = NULL;
4100 napi->gro_count = 0;
4102 EXPORT_SYMBOL(netif_napi_del);
4104 static void net_rx_action(struct softirq_action *h)
4106 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4107 unsigned long time_limit = jiffies + 2;
4108 int budget = netdev_budget;
4111 local_irq_disable();
4113 while (!list_empty(&sd->poll_list)) {
4114 struct napi_struct *n;
4117 /* If softirq window is exhuasted then punt.
4118 * Allow this to run for 2 jiffies since which will allow
4119 * an average latency of 1.5/HZ.
4121 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4126 /* Even though interrupts have been re-enabled, this
4127 * access is safe because interrupts can only add new
4128 * entries to the tail of this list, and only ->poll()
4129 * calls can remove this head entry from the list.
4131 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4133 have = netpoll_poll_lock(n);
4137 /* This NAPI_STATE_SCHED test is for avoiding a race
4138 * with netpoll's poll_napi(). Only the entity which
4139 * obtains the lock and sees NAPI_STATE_SCHED set will
4140 * actually make the ->poll() call. Therefore we avoid
4141 * accidentally calling ->poll() when NAPI is not scheduled.
4144 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4145 work = n->poll(n, weight);
4149 WARN_ON_ONCE(work > weight);
4153 local_irq_disable();
4155 /* Drivers must not modify the NAPI state if they
4156 * consume the entire weight. In such cases this code
4157 * still "owns" the NAPI instance and therefore can
4158 * move the instance around on the list at-will.
4160 if (unlikely(work == weight)) {
4161 if (unlikely(napi_disable_pending(n))) {
4164 local_irq_disable();
4167 /* flush too old packets
4168 * If HZ < 1000, flush all packets.
4171 napi_gro_flush(n, HZ >= 1000);
4172 local_irq_disable();
4174 list_move_tail(&n->poll_list, &sd->poll_list);
4178 netpoll_poll_unlock(have);
4181 net_rps_action_and_irq_enable(sd);
4183 #ifdef CONFIG_NET_DMA
4185 * There may not be any more sk_buffs coming right now, so push
4186 * any pending DMA copies to hardware
4188 dma_issue_pending_all();
4195 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4199 struct netdev_upper {
4200 struct net_device *dev;
4202 struct list_head list;
4203 struct rcu_head rcu;
4204 struct list_head search_list;
4207 static void __append_search_uppers(struct list_head *search_list,
4208 struct net_device *dev)
4210 struct netdev_upper *upper;
4212 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4213 /* check if this upper is not already in search list */
4214 if (list_empty(&upper->search_list))
4215 list_add_tail(&upper->search_list, search_list);
4219 static bool __netdev_search_upper_dev(struct net_device *dev,
4220 struct net_device *upper_dev)
4222 LIST_HEAD(search_list);
4223 struct netdev_upper *upper;
4224 struct netdev_upper *tmp;
4227 __append_search_uppers(&search_list, dev);
4228 list_for_each_entry(upper, &search_list, search_list) {
4229 if (upper->dev == upper_dev) {
4233 __append_search_uppers(&search_list, upper->dev);
4235 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4236 INIT_LIST_HEAD(&upper->search_list);
4240 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4241 struct net_device *upper_dev)
4243 struct netdev_upper *upper;
4245 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4246 if (upper->dev == upper_dev)
4253 * netdev_has_upper_dev - Check if device is linked to an upper device
4255 * @upper_dev: upper device to check
4257 * Find out if a device is linked to specified upper device and return true
4258 * in case it is. Note that this checks only immediate upper device,
4259 * not through a complete stack of devices. The caller must hold the RTNL lock.
4261 bool netdev_has_upper_dev(struct net_device *dev,
4262 struct net_device *upper_dev)
4266 return __netdev_find_upper(dev, upper_dev);
4268 EXPORT_SYMBOL(netdev_has_upper_dev);
4271 * netdev_has_any_upper_dev - Check if device is linked to some device
4274 * Find out if a device is linked to an upper device and return true in case
4275 * it is. The caller must hold the RTNL lock.
4277 bool netdev_has_any_upper_dev(struct net_device *dev)
4281 return !list_empty(&dev->upper_dev_list);
4283 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4286 * netdev_master_upper_dev_get - Get master upper device
4289 * Find a master upper device and return pointer to it or NULL in case
4290 * it's not there. The caller must hold the RTNL lock.
4292 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4294 struct netdev_upper *upper;
4298 if (list_empty(&dev->upper_dev_list))
4301 upper = list_first_entry(&dev->upper_dev_list,
4302 struct netdev_upper, list);
4303 if (likely(upper->master))
4307 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4310 * netdev_master_upper_dev_get_rcu - Get master upper device
4313 * Find a master upper device and return pointer to it or NULL in case
4314 * it's not there. The caller must hold the RCU read lock.
4316 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4318 struct netdev_upper *upper;
4320 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4321 struct netdev_upper, list);
4322 if (upper && likely(upper->master))
4326 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4328 static int __netdev_upper_dev_link(struct net_device *dev,
4329 struct net_device *upper_dev, bool master)
4331 struct netdev_upper *upper;
4335 if (dev == upper_dev)
4338 /* To prevent loops, check if dev is not upper device to upper_dev. */
4339 if (__netdev_search_upper_dev(upper_dev, dev))
4342 if (__netdev_find_upper(dev, upper_dev))
4345 if (master && netdev_master_upper_dev_get(dev))
4348 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4352 upper->dev = upper_dev;
4353 upper->master = master;
4354 INIT_LIST_HEAD(&upper->search_list);
4356 /* Ensure that master upper link is always the first item in list. */
4358 list_add_rcu(&upper->list, &dev->upper_dev_list);
4360 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4361 dev_hold(upper_dev);
4367 * netdev_upper_dev_link - Add a link to the upper device
4369 * @upper_dev: new upper device
4371 * Adds a link to device which is upper to this one. The caller must hold
4372 * the RTNL lock. On a failure a negative errno code is returned.
4373 * On success the reference counts are adjusted and the function
4376 int netdev_upper_dev_link(struct net_device *dev,
4377 struct net_device *upper_dev)
4379 return __netdev_upper_dev_link(dev, upper_dev, false);
4381 EXPORT_SYMBOL(netdev_upper_dev_link);
4384 * netdev_master_upper_dev_link - Add a master link to the upper device
4386 * @upper_dev: new upper device
4388 * Adds a link to device which is upper to this one. In this case, only
4389 * one master upper device can be linked, although other non-master devices
4390 * might be linked as well. The caller must hold the RTNL lock.
4391 * On a failure a negative errno code is returned. On success the reference
4392 * counts are adjusted and the function returns zero.
4394 int netdev_master_upper_dev_link(struct net_device *dev,
4395 struct net_device *upper_dev)
4397 return __netdev_upper_dev_link(dev, upper_dev, true);
4399 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4402 * netdev_upper_dev_unlink - Removes a link to upper device
4404 * @upper_dev: new upper device
4406 * Removes a link to device which is upper to this one. The caller must hold
4409 void netdev_upper_dev_unlink(struct net_device *dev,
4410 struct net_device *upper_dev)
4412 struct netdev_upper *upper;
4416 upper = __netdev_find_upper(dev, upper_dev);
4419 list_del_rcu(&upper->list);
4421 kfree_rcu(upper, rcu);
4423 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4425 static void dev_change_rx_flags(struct net_device *dev, int flags)
4427 const struct net_device_ops *ops = dev->netdev_ops;
4429 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4430 ops->ndo_change_rx_flags(dev, flags);
4433 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4435 unsigned int old_flags = dev->flags;
4441 dev->flags |= IFF_PROMISC;
4442 dev->promiscuity += inc;
4443 if (dev->promiscuity == 0) {
4446 * If inc causes overflow, untouch promisc and return error.
4449 dev->flags &= ~IFF_PROMISC;
4451 dev->promiscuity -= inc;
4452 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4457 if (dev->flags != old_flags) {
4458 pr_info("device %s %s promiscuous mode\n",
4460 dev->flags & IFF_PROMISC ? "entered" : "left");
4461 if (audit_enabled) {
4462 current_uid_gid(&uid, &gid);
4463 audit_log(current->audit_context, GFP_ATOMIC,
4464 AUDIT_ANOM_PROMISCUOUS,
4465 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4466 dev->name, (dev->flags & IFF_PROMISC),
4467 (old_flags & IFF_PROMISC),
4468 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4469 from_kuid(&init_user_ns, uid),
4470 from_kgid(&init_user_ns, gid),
4471 audit_get_sessionid(current));
4474 dev_change_rx_flags(dev, IFF_PROMISC);
4480 * dev_set_promiscuity - update promiscuity count on a device
4484 * Add or remove promiscuity from a device. While the count in the device
4485 * remains above zero the interface remains promiscuous. Once it hits zero
4486 * the device reverts back to normal filtering operation. A negative inc
4487 * value is used to drop promiscuity on the device.
4488 * Return 0 if successful or a negative errno code on error.
4490 int dev_set_promiscuity(struct net_device *dev, int inc)
4492 unsigned int old_flags = dev->flags;
4495 err = __dev_set_promiscuity(dev, inc);
4498 if (dev->flags != old_flags)
4499 dev_set_rx_mode(dev);
4502 EXPORT_SYMBOL(dev_set_promiscuity);
4505 * dev_set_allmulti - update allmulti count on a device
4509 * Add or remove reception of all multicast frames to a device. While the
4510 * count in the device remains above zero the interface remains listening
4511 * to all interfaces. Once it hits zero the device reverts back to normal
4512 * filtering operation. A negative @inc value is used to drop the counter
4513 * when releasing a resource needing all multicasts.
4514 * Return 0 if successful or a negative errno code on error.
4517 int dev_set_allmulti(struct net_device *dev, int inc)
4519 unsigned int old_flags = dev->flags;
4523 dev->flags |= IFF_ALLMULTI;
4524 dev->allmulti += inc;
4525 if (dev->allmulti == 0) {
4528 * If inc causes overflow, untouch allmulti and return error.
4531 dev->flags &= ~IFF_ALLMULTI;
4533 dev->allmulti -= inc;
4534 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4539 if (dev->flags ^ old_flags) {
4540 dev_change_rx_flags(dev, IFF_ALLMULTI);
4541 dev_set_rx_mode(dev);
4545 EXPORT_SYMBOL(dev_set_allmulti);
4548 * Upload unicast and multicast address lists to device and
4549 * configure RX filtering. When the device doesn't support unicast
4550 * filtering it is put in promiscuous mode while unicast addresses
4553 void __dev_set_rx_mode(struct net_device *dev)
4555 const struct net_device_ops *ops = dev->netdev_ops;
4557 /* dev_open will call this function so the list will stay sane. */
4558 if (!(dev->flags&IFF_UP))
4561 if (!netif_device_present(dev))
4564 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4565 /* Unicast addresses changes may only happen under the rtnl,
4566 * therefore calling __dev_set_promiscuity here is safe.
4568 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4569 __dev_set_promiscuity(dev, 1);
4570 dev->uc_promisc = true;
4571 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4572 __dev_set_promiscuity(dev, -1);
4573 dev->uc_promisc = false;
4577 if (ops->ndo_set_rx_mode)
4578 ops->ndo_set_rx_mode(dev);
4581 void dev_set_rx_mode(struct net_device *dev)
4583 netif_addr_lock_bh(dev);
4584 __dev_set_rx_mode(dev);
4585 netif_addr_unlock_bh(dev);
4589 * dev_get_flags - get flags reported to userspace
4592 * Get the combination of flag bits exported through APIs to userspace.
4594 unsigned int dev_get_flags(const struct net_device *dev)
4598 flags = (dev->flags & ~(IFF_PROMISC |
4603 (dev->gflags & (IFF_PROMISC |
4606 if (netif_running(dev)) {
4607 if (netif_oper_up(dev))
4608 flags |= IFF_RUNNING;
4609 if (netif_carrier_ok(dev))
4610 flags |= IFF_LOWER_UP;
4611 if (netif_dormant(dev))
4612 flags |= IFF_DORMANT;
4617 EXPORT_SYMBOL(dev_get_flags);
4619 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4621 unsigned int old_flags = dev->flags;
4627 * Set the flags on our device.
4630 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4631 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4633 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4637 * Load in the correct multicast list now the flags have changed.
4640 if ((old_flags ^ flags) & IFF_MULTICAST)
4641 dev_change_rx_flags(dev, IFF_MULTICAST);
4643 dev_set_rx_mode(dev);
4646 * Have we downed the interface. We handle IFF_UP ourselves
4647 * according to user attempts to set it, rather than blindly
4652 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4653 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4656 dev_set_rx_mode(dev);
4659 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4660 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4662 dev->gflags ^= IFF_PROMISC;
4663 dev_set_promiscuity(dev, inc);
4666 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4667 is important. Some (broken) drivers set IFF_PROMISC, when
4668 IFF_ALLMULTI is requested not asking us and not reporting.
4670 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4671 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4673 dev->gflags ^= IFF_ALLMULTI;
4674 dev_set_allmulti(dev, inc);
4680 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4682 unsigned int changes = dev->flags ^ old_flags;
4684 if (changes & IFF_UP) {
4685 if (dev->flags & IFF_UP)
4686 call_netdevice_notifiers(NETDEV_UP, dev);
4688 call_netdevice_notifiers(NETDEV_DOWN, dev);
4691 if (dev->flags & IFF_UP &&
4692 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4693 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4697 * dev_change_flags - change device settings
4699 * @flags: device state flags
4701 * Change settings on device based state flags. The flags are
4702 * in the userspace exported format.
4704 int dev_change_flags(struct net_device *dev, unsigned int flags)
4707 unsigned int changes, old_flags = dev->flags;
4709 ret = __dev_change_flags(dev, flags);
4713 changes = old_flags ^ dev->flags;
4715 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4717 __dev_notify_flags(dev, old_flags);
4720 EXPORT_SYMBOL(dev_change_flags);
4723 * dev_set_mtu - Change maximum transfer unit
4725 * @new_mtu: new transfer unit
4727 * Change the maximum transfer size of the network device.
4729 int dev_set_mtu(struct net_device *dev, int new_mtu)
4731 const struct net_device_ops *ops = dev->netdev_ops;
4734 if (new_mtu == dev->mtu)
4737 /* MTU must be positive. */
4741 if (!netif_device_present(dev))
4745 if (ops->ndo_change_mtu)
4746 err = ops->ndo_change_mtu(dev, new_mtu);
4751 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4754 EXPORT_SYMBOL(dev_set_mtu);
4757 * dev_set_group - Change group this device belongs to
4759 * @new_group: group this device should belong to
4761 void dev_set_group(struct net_device *dev, int new_group)
4763 dev->group = new_group;
4765 EXPORT_SYMBOL(dev_set_group);
4768 * dev_set_mac_address - Change Media Access Control Address
4772 * Change the hardware (MAC) address of the device
4774 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4776 const struct net_device_ops *ops = dev->netdev_ops;
4779 if (!ops->ndo_set_mac_address)
4781 if (sa->sa_family != dev->type)
4783 if (!netif_device_present(dev))
4785 err = ops->ndo_set_mac_address(dev, sa);
4788 dev->addr_assign_type = NET_ADDR_SET;
4789 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4790 add_device_randomness(dev->dev_addr, dev->addr_len);
4793 EXPORT_SYMBOL(dev_set_mac_address);
4796 * dev_change_carrier - Change device carrier
4798 * @new_carrier: new value
4800 * Change device carrier
4802 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4804 const struct net_device_ops *ops = dev->netdev_ops;
4806 if (!ops->ndo_change_carrier)
4808 if (!netif_device_present(dev))
4810 return ops->ndo_change_carrier(dev, new_carrier);
4812 EXPORT_SYMBOL(dev_change_carrier);
4815 * dev_new_index - allocate an ifindex
4816 * @net: the applicable net namespace
4818 * Returns a suitable unique value for a new device interface
4819 * number. The caller must hold the rtnl semaphore or the
4820 * dev_base_lock to be sure it remains unique.
4822 static int dev_new_index(struct net *net)
4824 int ifindex = net->ifindex;
4828 if (!__dev_get_by_index(net, ifindex))
4829 return net->ifindex = ifindex;
4833 /* Delayed registration/unregisteration */
4834 static LIST_HEAD(net_todo_list);
4836 static void net_set_todo(struct net_device *dev)
4838 list_add_tail(&dev->todo_list, &net_todo_list);
4841 static void rollback_registered_many(struct list_head *head)
4843 struct net_device *dev, *tmp;
4845 BUG_ON(dev_boot_phase);
4848 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4849 /* Some devices call without registering
4850 * for initialization unwind. Remove those
4851 * devices and proceed with the remaining.
4853 if (dev->reg_state == NETREG_UNINITIALIZED) {
4854 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4858 list_del(&dev->unreg_list);
4861 dev->dismantle = true;
4862 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4865 /* If device is running, close it first. */
4866 dev_close_many(head);
4868 list_for_each_entry(dev, head, unreg_list) {
4869 /* And unlink it from device chain. */
4870 unlist_netdevice(dev);
4872 dev->reg_state = NETREG_UNREGISTERING;
4877 list_for_each_entry(dev, head, unreg_list) {
4878 /* Shutdown queueing discipline. */
4882 /* Notify protocols, that we are about to destroy
4883 this device. They should clean all the things.
4885 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4887 if (!dev->rtnl_link_ops ||
4888 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4889 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4892 * Flush the unicast and multicast chains
4897 if (dev->netdev_ops->ndo_uninit)
4898 dev->netdev_ops->ndo_uninit(dev);
4900 /* Notifier chain MUST detach us all upper devices. */
4901 WARN_ON(netdev_has_any_upper_dev(dev));
4903 /* Remove entries from kobject tree */
4904 netdev_unregister_kobject(dev);
4906 /* Remove XPS queueing entries */
4907 netif_reset_xps_queues_gt(dev, 0);
4913 list_for_each_entry(dev, head, unreg_list)
4917 static void rollback_registered(struct net_device *dev)
4921 list_add(&dev->unreg_list, &single);
4922 rollback_registered_many(&single);
4926 static netdev_features_t netdev_fix_features(struct net_device *dev,
4927 netdev_features_t features)
4929 /* Fix illegal checksum combinations */
4930 if ((features & NETIF_F_HW_CSUM) &&
4931 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4932 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
4933 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4936 /* TSO requires that SG is present as well. */
4937 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
4938 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
4939 features &= ~NETIF_F_ALL_TSO;
4942 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
4943 !(features & NETIF_F_IP_CSUM)) {
4944 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
4945 features &= ~NETIF_F_TSO;
4946 features &= ~NETIF_F_TSO_ECN;
4949 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
4950 !(features & NETIF_F_IPV6_CSUM)) {
4951 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
4952 features &= ~NETIF_F_TSO6;
4955 /* TSO ECN requires that TSO is present as well. */
4956 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
4957 features &= ~NETIF_F_TSO_ECN;
4959 /* Software GSO depends on SG. */
4960 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
4961 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
4962 features &= ~NETIF_F_GSO;
4965 /* UFO needs SG and checksumming */
4966 if (features & NETIF_F_UFO) {
4967 /* maybe split UFO into V4 and V6? */
4968 if (!((features & NETIF_F_GEN_CSUM) ||
4969 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
4970 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4972 "Dropping NETIF_F_UFO since no checksum offload features.\n");
4973 features &= ~NETIF_F_UFO;
4976 if (!(features & NETIF_F_SG)) {
4978 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
4979 features &= ~NETIF_F_UFO;
4986 int __netdev_update_features(struct net_device *dev)
4988 netdev_features_t features;
4993 features = netdev_get_wanted_features(dev);
4995 if (dev->netdev_ops->ndo_fix_features)
4996 features = dev->netdev_ops->ndo_fix_features(dev, features);
4998 /* driver might be less strict about feature dependencies */
4999 features = netdev_fix_features(dev, features);
5001 if (dev->features == features)
5004 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5005 &dev->features, &features);
5007 if (dev->netdev_ops->ndo_set_features)
5008 err = dev->netdev_ops->ndo_set_features(dev, features);
5010 if (unlikely(err < 0)) {
5012 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5013 err, &features, &dev->features);
5018 dev->features = features;
5024 * netdev_update_features - recalculate device features
5025 * @dev: the device to check
5027 * Recalculate dev->features set and send notifications if it
5028 * has changed. Should be called after driver or hardware dependent
5029 * conditions might have changed that influence the features.
5031 void netdev_update_features(struct net_device *dev)
5033 if (__netdev_update_features(dev))
5034 netdev_features_change(dev);
5036 EXPORT_SYMBOL(netdev_update_features);
5039 * netdev_change_features - recalculate device features
5040 * @dev: the device to check
5042 * Recalculate dev->features set and send notifications even
5043 * if they have not changed. Should be called instead of
5044 * netdev_update_features() if also dev->vlan_features might
5045 * have changed to allow the changes to be propagated to stacked
5048 void netdev_change_features(struct net_device *dev)
5050 __netdev_update_features(dev);
5051 netdev_features_change(dev);
5053 EXPORT_SYMBOL(netdev_change_features);
5056 * netif_stacked_transfer_operstate - transfer operstate
5057 * @rootdev: the root or lower level device to transfer state from
5058 * @dev: the device to transfer operstate to
5060 * Transfer operational state from root to device. This is normally
5061 * called when a stacking relationship exists between the root
5062 * device and the device(a leaf device).
5064 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5065 struct net_device *dev)
5067 if (rootdev->operstate == IF_OPER_DORMANT)
5068 netif_dormant_on(dev);
5070 netif_dormant_off(dev);
5072 if (netif_carrier_ok(rootdev)) {
5073 if (!netif_carrier_ok(dev))
5074 netif_carrier_on(dev);
5076 if (netif_carrier_ok(dev))
5077 netif_carrier_off(dev);
5080 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5083 static int netif_alloc_rx_queues(struct net_device *dev)
5085 unsigned int i, count = dev->num_rx_queues;
5086 struct netdev_rx_queue *rx;
5090 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5096 for (i = 0; i < count; i++)
5102 static void netdev_init_one_queue(struct net_device *dev,
5103 struct netdev_queue *queue, void *_unused)
5105 /* Initialize queue lock */
5106 spin_lock_init(&queue->_xmit_lock);
5107 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5108 queue->xmit_lock_owner = -1;
5109 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5112 dql_init(&queue->dql, HZ);
5116 static int netif_alloc_netdev_queues(struct net_device *dev)
5118 unsigned int count = dev->num_tx_queues;
5119 struct netdev_queue *tx;
5123 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5129 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5130 spin_lock_init(&dev->tx_global_lock);
5136 * register_netdevice - register a network device
5137 * @dev: device to register
5139 * Take a completed network device structure and add it to the kernel
5140 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5141 * chain. 0 is returned on success. A negative errno code is returned
5142 * on a failure to set up the device, or if the name is a duplicate.
5144 * Callers must hold the rtnl semaphore. You may want
5145 * register_netdev() instead of this.
5148 * The locking appears insufficient to guarantee two parallel registers
5149 * will not get the same name.
5152 int register_netdevice(struct net_device *dev)
5155 struct net *net = dev_net(dev);
5157 BUG_ON(dev_boot_phase);
5162 /* When net_device's are persistent, this will be fatal. */
5163 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5166 spin_lock_init(&dev->addr_list_lock);
5167 netdev_set_addr_lockdep_class(dev);
5171 ret = dev_get_valid_name(net, dev, dev->name);
5175 /* Init, if this function is available */
5176 if (dev->netdev_ops->ndo_init) {
5177 ret = dev->netdev_ops->ndo_init(dev);
5185 if (((dev->hw_features | dev->features) &
5186 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5187 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5188 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5189 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5196 dev->ifindex = dev_new_index(net);
5197 else if (__dev_get_by_index(net, dev->ifindex))
5200 if (dev->iflink == -1)
5201 dev->iflink = dev->ifindex;
5203 /* Transfer changeable features to wanted_features and enable
5204 * software offloads (GSO and GRO).
5206 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5207 dev->features |= NETIF_F_SOFT_FEATURES;
5208 dev->wanted_features = dev->features & dev->hw_features;
5210 /* Turn on no cache copy if HW is doing checksum */
5211 if (!(dev->flags & IFF_LOOPBACK)) {
5212 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5213 if (dev->features & NETIF_F_ALL_CSUM) {
5214 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5215 dev->features |= NETIF_F_NOCACHE_COPY;
5219 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5221 dev->vlan_features |= NETIF_F_HIGHDMA;
5223 /* Make NETIF_F_SG inheritable to tunnel devices.
5225 dev->hw_enc_features |= NETIF_F_SG;
5227 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5228 ret = notifier_to_errno(ret);
5232 ret = netdev_register_kobject(dev);
5235 dev->reg_state = NETREG_REGISTERED;
5237 __netdev_update_features(dev);
5240 * Default initial state at registry is that the
5241 * device is present.
5244 set_bit(__LINK_STATE_PRESENT, &dev->state);
5246 linkwatch_init_dev(dev);
5248 dev_init_scheduler(dev);
5250 list_netdevice(dev);
5251 add_device_randomness(dev->dev_addr, dev->addr_len);
5253 /* If the device has permanent device address, driver should
5254 * set dev_addr and also addr_assign_type should be set to
5255 * NET_ADDR_PERM (default value).
5257 if (dev->addr_assign_type == NET_ADDR_PERM)
5258 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5260 /* Notify protocols, that a new device appeared. */
5261 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5262 ret = notifier_to_errno(ret);
5264 rollback_registered(dev);
5265 dev->reg_state = NETREG_UNREGISTERED;
5268 * Prevent userspace races by waiting until the network
5269 * device is fully setup before sending notifications.
5271 if (!dev->rtnl_link_ops ||
5272 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5273 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5279 if (dev->netdev_ops->ndo_uninit)
5280 dev->netdev_ops->ndo_uninit(dev);
5283 EXPORT_SYMBOL(register_netdevice);
5286 * init_dummy_netdev - init a dummy network device for NAPI
5287 * @dev: device to init
5289 * This takes a network device structure and initialize the minimum
5290 * amount of fields so it can be used to schedule NAPI polls without
5291 * registering a full blown interface. This is to be used by drivers
5292 * that need to tie several hardware interfaces to a single NAPI
5293 * poll scheduler due to HW limitations.
5295 int init_dummy_netdev(struct net_device *dev)
5297 /* Clear everything. Note we don't initialize spinlocks
5298 * are they aren't supposed to be taken by any of the
5299 * NAPI code and this dummy netdev is supposed to be
5300 * only ever used for NAPI polls
5302 memset(dev, 0, sizeof(struct net_device));
5304 /* make sure we BUG if trying to hit standard
5305 * register/unregister code path
5307 dev->reg_state = NETREG_DUMMY;
5309 /* NAPI wants this */
5310 INIT_LIST_HEAD(&dev->napi_list);
5312 /* a dummy interface is started by default */
5313 set_bit(__LINK_STATE_PRESENT, &dev->state);
5314 set_bit(__LINK_STATE_START, &dev->state);
5316 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5317 * because users of this 'device' dont need to change
5323 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5327 * register_netdev - register a network device
5328 * @dev: device to register
5330 * Take a completed network device structure and add it to the kernel
5331 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5332 * chain. 0 is returned on success. A negative errno code is returned
5333 * on a failure to set up the device, or if the name is a duplicate.
5335 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5336 * and expands the device name if you passed a format string to
5339 int register_netdev(struct net_device *dev)
5344 err = register_netdevice(dev);
5348 EXPORT_SYMBOL(register_netdev);
5350 int netdev_refcnt_read(const struct net_device *dev)
5354 for_each_possible_cpu(i)
5355 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5358 EXPORT_SYMBOL(netdev_refcnt_read);
5361 * netdev_wait_allrefs - wait until all references are gone.
5362 * @dev: target net_device
5364 * This is called when unregistering network devices.
5366 * Any protocol or device that holds a reference should register
5367 * for netdevice notification, and cleanup and put back the
5368 * reference if they receive an UNREGISTER event.
5369 * We can get stuck here if buggy protocols don't correctly
5372 static void netdev_wait_allrefs(struct net_device *dev)
5374 unsigned long rebroadcast_time, warning_time;
5377 linkwatch_forget_dev(dev);
5379 rebroadcast_time = warning_time = jiffies;
5380 refcnt = netdev_refcnt_read(dev);
5382 while (refcnt != 0) {
5383 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5386 /* Rebroadcast unregister notification */
5387 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5393 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5394 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5396 /* We must not have linkwatch events
5397 * pending on unregister. If this
5398 * happens, we simply run the queue
5399 * unscheduled, resulting in a noop
5402 linkwatch_run_queue();
5407 rebroadcast_time = jiffies;
5412 refcnt = netdev_refcnt_read(dev);
5414 if (time_after(jiffies, warning_time + 10 * HZ)) {
5415 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5417 warning_time = jiffies;
5426 * register_netdevice(x1);
5427 * register_netdevice(x2);
5429 * unregister_netdevice(y1);
5430 * unregister_netdevice(y2);
5436 * We are invoked by rtnl_unlock().
5437 * This allows us to deal with problems:
5438 * 1) We can delete sysfs objects which invoke hotplug
5439 * without deadlocking with linkwatch via keventd.
5440 * 2) Since we run with the RTNL semaphore not held, we can sleep
5441 * safely in order to wait for the netdev refcnt to drop to zero.
5443 * We must not return until all unregister events added during
5444 * the interval the lock was held have been completed.
5446 void netdev_run_todo(void)
5448 struct list_head list;
5450 /* Snapshot list, allow later requests */
5451 list_replace_init(&net_todo_list, &list);
5456 /* Wait for rcu callbacks to finish before next phase */
5457 if (!list_empty(&list))
5460 while (!list_empty(&list)) {
5461 struct net_device *dev
5462 = list_first_entry(&list, struct net_device, todo_list);
5463 list_del(&dev->todo_list);
5466 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5469 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5470 pr_err("network todo '%s' but state %d\n",
5471 dev->name, dev->reg_state);
5476 dev->reg_state = NETREG_UNREGISTERED;
5478 on_each_cpu(flush_backlog, dev, 1);
5480 netdev_wait_allrefs(dev);
5483 BUG_ON(netdev_refcnt_read(dev));
5484 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5485 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5486 WARN_ON(dev->dn_ptr);
5488 if (dev->destructor)
5489 dev->destructor(dev);
5491 /* Free network device */
5492 kobject_put(&dev->dev.kobj);
5496 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5497 * fields in the same order, with only the type differing.
5499 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5500 const struct net_device_stats *netdev_stats)
5502 #if BITS_PER_LONG == 64
5503 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5504 memcpy(stats64, netdev_stats, sizeof(*stats64));
5506 size_t i, n = sizeof(*stats64) / sizeof(u64);
5507 const unsigned long *src = (const unsigned long *)netdev_stats;
5508 u64 *dst = (u64 *)stats64;
5510 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5511 sizeof(*stats64) / sizeof(u64));
5512 for (i = 0; i < n; i++)
5516 EXPORT_SYMBOL(netdev_stats_to_stats64);
5519 * dev_get_stats - get network device statistics
5520 * @dev: device to get statistics from
5521 * @storage: place to store stats
5523 * Get network statistics from device. Return @storage.
5524 * The device driver may provide its own method by setting
5525 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5526 * otherwise the internal statistics structure is used.
5528 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5529 struct rtnl_link_stats64 *storage)
5531 const struct net_device_ops *ops = dev->netdev_ops;
5533 if (ops->ndo_get_stats64) {
5534 memset(storage, 0, sizeof(*storage));
5535 ops->ndo_get_stats64(dev, storage);
5536 } else if (ops->ndo_get_stats) {
5537 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5539 netdev_stats_to_stats64(storage, &dev->stats);
5541 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5544 EXPORT_SYMBOL(dev_get_stats);
5546 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5548 struct netdev_queue *queue = dev_ingress_queue(dev);
5550 #ifdef CONFIG_NET_CLS_ACT
5553 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5556 netdev_init_one_queue(dev, queue, NULL);
5557 queue->qdisc = &noop_qdisc;
5558 queue->qdisc_sleeping = &noop_qdisc;
5559 rcu_assign_pointer(dev->ingress_queue, queue);
5564 static const struct ethtool_ops default_ethtool_ops;
5566 void netdev_set_default_ethtool_ops(struct net_device *dev,
5567 const struct ethtool_ops *ops)
5569 if (dev->ethtool_ops == &default_ethtool_ops)
5570 dev->ethtool_ops = ops;
5572 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5575 * alloc_netdev_mqs - allocate network device
5576 * @sizeof_priv: size of private data to allocate space for
5577 * @name: device name format string
5578 * @setup: callback to initialize device
5579 * @txqs: the number of TX subqueues to allocate
5580 * @rxqs: the number of RX subqueues to allocate
5582 * Allocates a struct net_device with private data area for driver use
5583 * and performs basic initialization. Also allocates subquue structs
5584 * for each queue on the device.
5586 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5587 void (*setup)(struct net_device *),
5588 unsigned int txqs, unsigned int rxqs)
5590 struct net_device *dev;
5592 struct net_device *p;
5594 BUG_ON(strlen(name) >= sizeof(dev->name));
5597 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5603 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5608 alloc_size = sizeof(struct net_device);
5610 /* ensure 32-byte alignment of private area */
5611 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5612 alloc_size += sizeof_priv;
5614 /* ensure 32-byte alignment of whole construct */
5615 alloc_size += NETDEV_ALIGN - 1;
5617 p = kzalloc(alloc_size, GFP_KERNEL);
5621 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5622 dev->padded = (char *)dev - (char *)p;
5624 dev->pcpu_refcnt = alloc_percpu(int);
5625 if (!dev->pcpu_refcnt)
5628 if (dev_addr_init(dev))
5634 dev_net_set(dev, &init_net);
5636 dev->gso_max_size = GSO_MAX_SIZE;
5637 dev->gso_max_segs = GSO_MAX_SEGS;
5639 INIT_LIST_HEAD(&dev->napi_list);
5640 INIT_LIST_HEAD(&dev->unreg_list);
5641 INIT_LIST_HEAD(&dev->link_watch_list);
5642 INIT_LIST_HEAD(&dev->upper_dev_list);
5643 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5646 dev->num_tx_queues = txqs;
5647 dev->real_num_tx_queues = txqs;
5648 if (netif_alloc_netdev_queues(dev))
5652 dev->num_rx_queues = rxqs;
5653 dev->real_num_rx_queues = rxqs;
5654 if (netif_alloc_rx_queues(dev))
5658 strcpy(dev->name, name);
5659 dev->group = INIT_NETDEV_GROUP;
5660 if (!dev->ethtool_ops)
5661 dev->ethtool_ops = &default_ethtool_ops;
5669 free_percpu(dev->pcpu_refcnt);
5679 EXPORT_SYMBOL(alloc_netdev_mqs);
5682 * free_netdev - free network device
5685 * This function does the last stage of destroying an allocated device
5686 * interface. The reference to the device object is released.
5687 * If this is the last reference then it will be freed.
5689 void free_netdev(struct net_device *dev)
5691 struct napi_struct *p, *n;
5693 release_net(dev_net(dev));
5700 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5702 /* Flush device addresses */
5703 dev_addr_flush(dev);
5705 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5708 free_percpu(dev->pcpu_refcnt);
5709 dev->pcpu_refcnt = NULL;
5711 /* Compatibility with error handling in drivers */
5712 if (dev->reg_state == NETREG_UNINITIALIZED) {
5713 kfree((char *)dev - dev->padded);
5717 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5718 dev->reg_state = NETREG_RELEASED;
5720 /* will free via device release */
5721 put_device(&dev->dev);
5723 EXPORT_SYMBOL(free_netdev);
5726 * synchronize_net - Synchronize with packet receive processing
5728 * Wait for packets currently being received to be done.
5729 * Does not block later packets from starting.
5731 void synchronize_net(void)
5734 if (rtnl_is_locked())
5735 synchronize_rcu_expedited();
5739 EXPORT_SYMBOL(synchronize_net);
5742 * unregister_netdevice_queue - remove device from the kernel
5746 * This function shuts down a device interface and removes it
5747 * from the kernel tables.
5748 * If head not NULL, device is queued to be unregistered later.
5750 * Callers must hold the rtnl semaphore. You may want
5751 * unregister_netdev() instead of this.
5754 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5759 list_move_tail(&dev->unreg_list, head);
5761 rollback_registered(dev);
5762 /* Finish processing unregister after unlock */
5766 EXPORT_SYMBOL(unregister_netdevice_queue);
5769 * unregister_netdevice_many - unregister many devices
5770 * @head: list of devices
5772 void unregister_netdevice_many(struct list_head *head)
5774 struct net_device *dev;
5776 if (!list_empty(head)) {
5777 rollback_registered_many(head);
5778 list_for_each_entry(dev, head, unreg_list)
5782 EXPORT_SYMBOL(unregister_netdevice_many);
5785 * unregister_netdev - remove device from the kernel
5788 * This function shuts down a device interface and removes it
5789 * from the kernel tables.
5791 * This is just a wrapper for unregister_netdevice that takes
5792 * the rtnl semaphore. In general you want to use this and not
5793 * unregister_netdevice.
5795 void unregister_netdev(struct net_device *dev)
5798 unregister_netdevice(dev);
5801 EXPORT_SYMBOL(unregister_netdev);
5804 * dev_change_net_namespace - move device to different nethost namespace
5806 * @net: network namespace
5807 * @pat: If not NULL name pattern to try if the current device name
5808 * is already taken in the destination network namespace.
5810 * This function shuts down a device interface and moves it
5811 * to a new network namespace. On success 0 is returned, on
5812 * a failure a netagive errno code is returned.
5814 * Callers must hold the rtnl semaphore.
5817 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5823 /* Don't allow namespace local devices to be moved. */
5825 if (dev->features & NETIF_F_NETNS_LOCAL)
5828 /* Ensure the device has been registrered */
5829 if (dev->reg_state != NETREG_REGISTERED)
5832 /* Get out if there is nothing todo */
5834 if (net_eq(dev_net(dev), net))
5837 /* Pick the destination device name, and ensure
5838 * we can use it in the destination network namespace.
5841 if (__dev_get_by_name(net, dev->name)) {
5842 /* We get here if we can't use the current device name */
5845 if (dev_get_valid_name(net, dev, pat) < 0)
5850 * And now a mini version of register_netdevice unregister_netdevice.
5853 /* If device is running close it first. */
5856 /* And unlink it from device chain */
5858 unlist_netdevice(dev);
5862 /* Shutdown queueing discipline. */
5865 /* Notify protocols, that we are about to destroy
5866 this device. They should clean all the things.
5868 Note that dev->reg_state stays at NETREG_REGISTERED.
5869 This is wanted because this way 8021q and macvlan know
5870 the device is just moving and can keep their slaves up.
5872 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5874 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5875 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5878 * Flush the unicast and multicast chains
5883 /* Send a netdev-removed uevent to the old namespace */
5884 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
5886 /* Actually switch the network namespace */
5887 dev_net_set(dev, net);
5889 /* If there is an ifindex conflict assign a new one */
5890 if (__dev_get_by_index(net, dev->ifindex)) {
5891 int iflink = (dev->iflink == dev->ifindex);
5892 dev->ifindex = dev_new_index(net);
5894 dev->iflink = dev->ifindex;
5897 /* Send a netdev-add uevent to the new namespace */
5898 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
5900 /* Fixup kobjects */
5901 err = device_rename(&dev->dev, dev->name);
5904 /* Add the device back in the hashes */
5905 list_netdevice(dev);
5907 /* Notify protocols, that a new device appeared. */
5908 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5911 * Prevent userspace races by waiting until the network
5912 * device is fully setup before sending notifications.
5914 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5921 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5923 static int dev_cpu_callback(struct notifier_block *nfb,
5924 unsigned long action,
5927 struct sk_buff **list_skb;
5928 struct sk_buff *skb;
5929 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5930 struct softnet_data *sd, *oldsd;
5932 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5935 local_irq_disable();
5936 cpu = smp_processor_id();
5937 sd = &per_cpu(softnet_data, cpu);
5938 oldsd = &per_cpu(softnet_data, oldcpu);
5940 /* Find end of our completion_queue. */
5941 list_skb = &sd->completion_queue;
5943 list_skb = &(*list_skb)->next;
5944 /* Append completion queue from offline CPU. */
5945 *list_skb = oldsd->completion_queue;
5946 oldsd->completion_queue = NULL;
5948 /* Append output queue from offline CPU. */
5949 if (oldsd->output_queue) {
5950 *sd->output_queue_tailp = oldsd->output_queue;
5951 sd->output_queue_tailp = oldsd->output_queue_tailp;
5952 oldsd->output_queue = NULL;
5953 oldsd->output_queue_tailp = &oldsd->output_queue;
5955 /* Append NAPI poll list from offline CPU. */
5956 if (!list_empty(&oldsd->poll_list)) {
5957 list_splice_init(&oldsd->poll_list, &sd->poll_list);
5958 raise_softirq_irqoff(NET_RX_SOFTIRQ);
5961 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5964 /* Process offline CPU's input_pkt_queue */
5965 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5967 input_queue_head_incr(oldsd);
5969 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5971 input_queue_head_incr(oldsd);
5979 * netdev_increment_features - increment feature set by one
5980 * @all: current feature set
5981 * @one: new feature set
5982 * @mask: mask feature set
5984 * Computes a new feature set after adding a device with feature set
5985 * @one to the master device with current feature set @all. Will not
5986 * enable anything that is off in @mask. Returns the new feature set.
5988 netdev_features_t netdev_increment_features(netdev_features_t all,
5989 netdev_features_t one, netdev_features_t mask)
5991 if (mask & NETIF_F_GEN_CSUM)
5992 mask |= NETIF_F_ALL_CSUM;
5993 mask |= NETIF_F_VLAN_CHALLENGED;
5995 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
5996 all &= one | ~NETIF_F_ALL_FOR_ALL;
5998 /* If one device supports hw checksumming, set for all. */
5999 if (all & NETIF_F_GEN_CSUM)
6000 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6004 EXPORT_SYMBOL(netdev_increment_features);
6006 static struct hlist_head *netdev_create_hash(void)
6009 struct hlist_head *hash;
6011 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6013 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6014 INIT_HLIST_HEAD(&hash[i]);
6019 /* Initialize per network namespace state */
6020 static int __net_init netdev_init(struct net *net)
6022 if (net != &init_net)
6023 INIT_LIST_HEAD(&net->dev_base_head);
6025 net->dev_name_head = netdev_create_hash();
6026 if (net->dev_name_head == NULL)
6029 net->dev_index_head = netdev_create_hash();
6030 if (net->dev_index_head == NULL)
6036 kfree(net->dev_name_head);
6042 * netdev_drivername - network driver for the device
6043 * @dev: network device
6045 * Determine network driver for device.
6047 const char *netdev_drivername(const struct net_device *dev)
6049 const struct device_driver *driver;
6050 const struct device *parent;
6051 const char *empty = "";
6053 parent = dev->dev.parent;
6057 driver = parent->driver;
6058 if (driver && driver->name)
6059 return driver->name;
6063 static int __netdev_printk(const char *level, const struct net_device *dev,
6064 struct va_format *vaf)
6068 if (dev && dev->dev.parent) {
6069 r = dev_printk_emit(level[1] - '0',
6072 dev_driver_string(dev->dev.parent),
6073 dev_name(dev->dev.parent),
6074 netdev_name(dev), vaf);
6076 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6078 r = printk("%s(NULL net_device): %pV", level, vaf);
6084 int netdev_printk(const char *level, const struct net_device *dev,
6085 const char *format, ...)
6087 struct va_format vaf;
6091 va_start(args, format);
6096 r = __netdev_printk(level, dev, &vaf);
6102 EXPORT_SYMBOL(netdev_printk);
6104 #define define_netdev_printk_level(func, level) \
6105 int func(const struct net_device *dev, const char *fmt, ...) \
6108 struct va_format vaf; \
6111 va_start(args, fmt); \
6116 r = __netdev_printk(level, dev, &vaf); \
6122 EXPORT_SYMBOL(func);
6124 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6125 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6126 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6127 define_netdev_printk_level(netdev_err, KERN_ERR);
6128 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6129 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6130 define_netdev_printk_level(netdev_info, KERN_INFO);
6132 static void __net_exit netdev_exit(struct net *net)
6134 kfree(net->dev_name_head);
6135 kfree(net->dev_index_head);
6138 static struct pernet_operations __net_initdata netdev_net_ops = {
6139 .init = netdev_init,
6140 .exit = netdev_exit,
6143 static void __net_exit default_device_exit(struct net *net)
6145 struct net_device *dev, *aux;
6147 * Push all migratable network devices back to the
6148 * initial network namespace
6151 for_each_netdev_safe(net, dev, aux) {
6153 char fb_name[IFNAMSIZ];
6155 /* Ignore unmoveable devices (i.e. loopback) */
6156 if (dev->features & NETIF_F_NETNS_LOCAL)
6159 /* Leave virtual devices for the generic cleanup */
6160 if (dev->rtnl_link_ops)
6163 /* Push remaining network devices to init_net */
6164 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6165 err = dev_change_net_namespace(dev, &init_net, fb_name);
6167 pr_emerg("%s: failed to move %s to init_net: %d\n",
6168 __func__, dev->name, err);
6175 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6177 /* At exit all network devices most be removed from a network
6178 * namespace. Do this in the reverse order of registration.
6179 * Do this across as many network namespaces as possible to
6180 * improve batching efficiency.
6182 struct net_device *dev;
6184 LIST_HEAD(dev_kill_list);
6187 list_for_each_entry(net, net_list, exit_list) {
6188 for_each_netdev_reverse(net, dev) {
6189 if (dev->rtnl_link_ops)
6190 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6192 unregister_netdevice_queue(dev, &dev_kill_list);
6195 unregister_netdevice_many(&dev_kill_list);
6196 list_del(&dev_kill_list);
6200 static struct pernet_operations __net_initdata default_device_ops = {
6201 .exit = default_device_exit,
6202 .exit_batch = default_device_exit_batch,
6206 * Initialize the DEV module. At boot time this walks the device list and
6207 * unhooks any devices that fail to initialise (normally hardware not
6208 * present) and leaves us with a valid list of present and active devices.
6213 * This is called single threaded during boot, so no need
6214 * to take the rtnl semaphore.
6216 static int __init net_dev_init(void)
6218 int i, rc = -ENOMEM;
6220 BUG_ON(!dev_boot_phase);
6222 if (dev_proc_init())
6225 if (netdev_kobject_init())
6228 INIT_LIST_HEAD(&ptype_all);
6229 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6230 INIT_LIST_HEAD(&ptype_base[i]);
6232 INIT_LIST_HEAD(&offload_base);
6234 if (register_pernet_subsys(&netdev_net_ops))
6238 * Initialise the packet receive queues.
6241 for_each_possible_cpu(i) {
6242 struct softnet_data *sd = &per_cpu(softnet_data, i);
6244 memset(sd, 0, sizeof(*sd));
6245 skb_queue_head_init(&sd->input_pkt_queue);
6246 skb_queue_head_init(&sd->process_queue);
6247 sd->completion_queue = NULL;
6248 INIT_LIST_HEAD(&sd->poll_list);
6249 sd->output_queue = NULL;
6250 sd->output_queue_tailp = &sd->output_queue;
6252 sd->csd.func = rps_trigger_softirq;
6258 sd->backlog.poll = process_backlog;
6259 sd->backlog.weight = weight_p;
6260 sd->backlog.gro_list = NULL;
6261 sd->backlog.gro_count = 0;
6266 /* The loopback device is special if any other network devices
6267 * is present in a network namespace the loopback device must
6268 * be present. Since we now dynamically allocate and free the
6269 * loopback device ensure this invariant is maintained by
6270 * keeping the loopback device as the first device on the
6271 * list of network devices. Ensuring the loopback devices
6272 * is the first device that appears and the last network device
6275 if (register_pernet_device(&loopback_net_ops))
6278 if (register_pernet_device(&default_device_ops))
6281 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6282 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6284 hotcpu_notifier(dev_cpu_callback, 0);
6291 subsys_initcall(net_dev_init);