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>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
151 static int call_netdevice_notifiers_info(unsigned long val,
152 struct net_device *dev,
153 struct netdev_notifier_info *info);
156 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
159 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
161 * Writers must hold the rtnl semaphore while they loop through the
162 * dev_base_head list, and hold dev_base_lock for writing when they do the
163 * actual updates. This allows pure readers to access the list even
164 * while a writer is preparing to update it.
166 * To put it another way, dev_base_lock is held for writing only to
167 * protect against pure readers; the rtnl semaphore provides the
168 * protection against other writers.
170 * See, for example usages, register_netdevice() and
171 * unregister_netdevice(), which must be called with the rtnl
174 DEFINE_RWLOCK(dev_base_lock);
175 EXPORT_SYMBOL(dev_base_lock);
177 /* protects napi_hash addition/deletion and napi_gen_id */
178 static DEFINE_SPINLOCK(napi_hash_lock);
180 static unsigned int napi_gen_id;
181 static DEFINE_HASHTABLE(napi_hash, 8);
183 static seqcount_t devnet_rename_seq;
185 static inline void dev_base_seq_inc(struct net *net)
187 while (++net->dev_base_seq == 0);
190 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
192 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
194 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
197 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
199 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
202 static inline void rps_lock(struct softnet_data *sd)
205 spin_lock(&sd->input_pkt_queue.lock);
209 static inline void rps_unlock(struct softnet_data *sd)
212 spin_unlock(&sd->input_pkt_queue.lock);
216 /* Device list insertion */
217 static void list_netdevice(struct net_device *dev)
219 struct net *net = dev_net(dev);
223 write_lock_bh(&dev_base_lock);
224 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
225 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
226 hlist_add_head_rcu(&dev->index_hlist,
227 dev_index_hash(net, dev->ifindex));
228 write_unlock_bh(&dev_base_lock);
230 dev_base_seq_inc(net);
233 /* Device list removal
234 * caller must respect a RCU grace period before freeing/reusing dev
236 static void unlist_netdevice(struct net_device *dev)
240 /* Unlink dev from the device chain */
241 write_lock_bh(&dev_base_lock);
242 list_del_rcu(&dev->dev_list);
243 hlist_del_rcu(&dev->name_hlist);
244 hlist_del_rcu(&dev->index_hlist);
245 write_unlock_bh(&dev_base_lock);
247 dev_base_seq_inc(dev_net(dev));
254 static RAW_NOTIFIER_HEAD(netdev_chain);
257 * Device drivers call our routines to queue packets here. We empty the
258 * queue in the local softnet handler.
261 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
262 EXPORT_PER_CPU_SYMBOL(softnet_data);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type[] =
270 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
271 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
272 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
273 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
274 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
275 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
276 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
277 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
278 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
279 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
280 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
281 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
282 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
283 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
284 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
286 static const char *const netdev_lock_name[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
300 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
301 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
303 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
304 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
310 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
311 if (netdev_lock_type[i] == dev_type)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
318 unsigned short dev_type)
322 i = netdev_lock_pos(dev_type);
323 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
331 i = netdev_lock_pos(dev->type);
332 lockdep_set_class_and_name(&dev->addr_list_lock,
333 &netdev_addr_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
338 unsigned short dev_type)
341 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
368 static inline struct list_head *ptype_head(const struct packet_type *pt)
370 if (pt->type == htons(ETH_P_ALL))
373 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
391 struct list_head *head = ptype_head(pt);
393 spin_lock(&ptype_lock);
394 list_add_rcu(&pt->list, head);
395 spin_unlock(&ptype_lock);
397 EXPORT_SYMBOL(dev_add_pack);
400 * __dev_remove_pack - remove packet handler
401 * @pt: packet type declaration
403 * Remove a protocol handler that was previously added to the kernel
404 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
405 * from the kernel lists and can be freed or reused once this function
408 * The packet type might still be in use by receivers
409 * and must not be freed until after all the CPU's have gone
410 * through a quiescent state.
412 void __dev_remove_pack(struct packet_type *pt)
414 struct list_head *head = ptype_head(pt);
415 struct packet_type *pt1;
417 spin_lock(&ptype_lock);
419 list_for_each_entry(pt1, head, list) {
421 list_del_rcu(&pt->list);
426 pr_warn("dev_remove_pack: %p not found\n", pt);
428 spin_unlock(&ptype_lock);
430 EXPORT_SYMBOL(__dev_remove_pack);
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
441 * This call sleeps to guarantee that no CPU is looking at the packet
444 void dev_remove_pack(struct packet_type *pt)
446 __dev_remove_pack(pt);
450 EXPORT_SYMBOL(dev_remove_pack);
454 * dev_add_offload - register offload handlers
455 * @po: protocol offload declaration
457 * Add protocol offload handlers to the networking stack. The passed
458 * &proto_offload is linked into kernel lists and may not be freed until
459 * it has been removed from the kernel lists.
461 * This call does not sleep therefore it can not
462 * guarantee all CPU's that are in middle of receiving packets
463 * will see the new offload handlers (until the next received packet).
465 void dev_add_offload(struct packet_offload *po)
467 struct list_head *head = &offload_base;
469 spin_lock(&offload_lock);
470 list_add_rcu(&po->list, head);
471 spin_unlock(&offload_lock);
473 EXPORT_SYMBOL(dev_add_offload);
476 * __dev_remove_offload - remove offload handler
477 * @po: packet offload declaration
479 * Remove a protocol offload handler that was previously added to the
480 * kernel offload handlers by dev_add_offload(). The passed &offload_type
481 * is removed from the kernel lists and can be freed or reused once this
484 * The packet type might still be in use by receivers
485 * and must not be freed until after all the CPU's have gone
486 * through a quiescent state.
488 static void __dev_remove_offload(struct packet_offload *po)
490 struct list_head *head = &offload_base;
491 struct packet_offload *po1;
493 spin_lock(&offload_lock);
495 list_for_each_entry(po1, head, list) {
497 list_del_rcu(&po->list);
502 pr_warn("dev_remove_offload: %p not found\n", po);
504 spin_unlock(&offload_lock);
508 * dev_remove_offload - remove packet offload handler
509 * @po: packet offload declaration
511 * Remove a packet offload handler that was previously added to the kernel
512 * offload handlers by dev_add_offload(). The passed &offload_type is
513 * removed from the kernel lists and can be freed or reused once this
516 * This call sleeps to guarantee that no CPU is looking at the packet
519 void dev_remove_offload(struct packet_offload *po)
521 __dev_remove_offload(po);
525 EXPORT_SYMBOL(dev_remove_offload);
527 /******************************************************************************
529 Device Boot-time Settings Routines
531 *******************************************************************************/
533 /* Boot time configuration table */
534 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
537 * netdev_boot_setup_add - add new setup entry
538 * @name: name of the device
539 * @map: configured settings for the device
541 * Adds new setup entry to the dev_boot_setup list. The function
542 * returns 0 on error and 1 on success. This is a generic routine to
545 static int netdev_boot_setup_add(char *name, struct ifmap *map)
547 struct netdev_boot_setup *s;
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
552 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
553 memset(s[i].name, 0, sizeof(s[i].name));
554 strlcpy(s[i].name, name, IFNAMSIZ);
555 memcpy(&s[i].map, map, sizeof(s[i].map));
560 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
564 * netdev_boot_setup_check - check boot time settings
565 * @dev: the netdevice
567 * Check boot time settings for the device.
568 * The found settings are set for the device to be used
569 * later in the device probing.
570 * Returns 0 if no settings found, 1 if they are.
572 int netdev_boot_setup_check(struct net_device *dev)
574 struct netdev_boot_setup *s = dev_boot_setup;
577 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
578 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
579 !strcmp(dev->name, s[i].name)) {
580 dev->irq = s[i].map.irq;
581 dev->base_addr = s[i].map.base_addr;
582 dev->mem_start = s[i].map.mem_start;
583 dev->mem_end = s[i].map.mem_end;
589 EXPORT_SYMBOL(netdev_boot_setup_check);
593 * netdev_boot_base - get address from boot time settings
594 * @prefix: prefix for network device
595 * @unit: id for network device
597 * Check boot time settings for the base address of device.
598 * The found settings are set for the device to be used
599 * later in the device probing.
600 * Returns 0 if no settings found.
602 unsigned long netdev_boot_base(const char *prefix, int unit)
604 const struct netdev_boot_setup *s = dev_boot_setup;
608 sprintf(name, "%s%d", prefix, unit);
611 * If device already registered then return base of 1
612 * to indicate not to probe for this interface
614 if (__dev_get_by_name(&init_net, name))
617 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
618 if (!strcmp(name, s[i].name))
619 return s[i].map.base_addr;
624 * Saves at boot time configured settings for any netdevice.
626 int __init netdev_boot_setup(char *str)
631 str = get_options(str, ARRAY_SIZE(ints), ints);
636 memset(&map, 0, sizeof(map));
640 map.base_addr = ints[2];
642 map.mem_start = ints[3];
644 map.mem_end = ints[4];
646 /* Add new entry to the list */
647 return netdev_boot_setup_add(str, &map);
650 __setup("netdev=", netdev_boot_setup);
652 /*******************************************************************************
654 Device Interface Subroutines
656 *******************************************************************************/
659 * __dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. Must be called under RTNL semaphore
664 * or @dev_base_lock. If the name is found a pointer to the device
665 * is returned. If the name is not found then %NULL is returned. The
666 * reference counters are not incremented so the caller must be
667 * careful with locks.
670 struct net_device *__dev_get_by_name(struct net *net, const char *name)
672 struct net_device *dev;
673 struct hlist_head *head = dev_name_hash(net, name);
675 hlist_for_each_entry(dev, head, name_hlist)
676 if (!strncmp(dev->name, name, IFNAMSIZ))
681 EXPORT_SYMBOL(__dev_get_by_name);
684 * dev_get_by_name_rcu - find a device by its name
685 * @net: the applicable net namespace
686 * @name: name to find
688 * Find an interface by name.
689 * If the name is found a pointer to the device is returned.
690 * If the name is not found then %NULL is returned.
691 * The reference counters are not incremented so the caller must be
692 * careful with locks. The caller must hold RCU lock.
695 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
697 struct net_device *dev;
698 struct hlist_head *head = dev_name_hash(net, name);
700 hlist_for_each_entry_rcu(dev, head, name_hlist)
701 if (!strncmp(dev->name, name, IFNAMSIZ))
706 EXPORT_SYMBOL(dev_get_by_name_rcu);
709 * dev_get_by_name - find a device by its name
710 * @net: the applicable net namespace
711 * @name: name to find
713 * Find an interface by name. This can be called from any
714 * context and does its own locking. The returned handle has
715 * the usage count incremented and the caller must use dev_put() to
716 * release it when it is no longer needed. %NULL is returned if no
717 * matching device is found.
720 struct net_device *dev_get_by_name(struct net *net, const char *name)
722 struct net_device *dev;
725 dev = dev_get_by_name_rcu(net, name);
731 EXPORT_SYMBOL(dev_get_by_name);
734 * __dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns %NULL if the device
739 * is not found or a pointer to the device. The device has not
740 * had its reference counter increased so the caller must be careful
741 * about locking. The caller must hold either the RTNL semaphore
745 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
747 struct net_device *dev;
748 struct hlist_head *head = dev_index_hash(net, ifindex);
750 hlist_for_each_entry(dev, head, index_hlist)
751 if (dev->ifindex == ifindex)
756 EXPORT_SYMBOL(__dev_get_by_index);
759 * dev_get_by_index_rcu - find a device by its ifindex
760 * @net: the applicable net namespace
761 * @ifindex: index of device
763 * Search for an interface by index. Returns %NULL if the device
764 * is not found or a pointer to the device. The device has not
765 * had its reference counter increased so the caller must be careful
766 * about locking. The caller must hold RCU lock.
769 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
771 struct net_device *dev;
772 struct hlist_head *head = dev_index_hash(net, ifindex);
774 hlist_for_each_entry_rcu(dev, head, index_hlist)
775 if (dev->ifindex == ifindex)
780 EXPORT_SYMBOL(dev_get_by_index_rcu);
784 * dev_get_by_index - find a device by its ifindex
785 * @net: the applicable net namespace
786 * @ifindex: index of device
788 * Search for an interface by index. Returns NULL if the device
789 * is not found or a pointer to the device. The device returned has
790 * had a reference added and the pointer is safe until the user calls
791 * dev_put to indicate they have finished with it.
794 struct net_device *dev_get_by_index(struct net *net, int ifindex)
796 struct net_device *dev;
799 dev = dev_get_by_index_rcu(net, ifindex);
805 EXPORT_SYMBOL(dev_get_by_index);
808 * netdev_get_name - get a netdevice name, knowing its ifindex.
809 * @net: network namespace
810 * @name: a pointer to the buffer where the name will be stored.
811 * @ifindex: the ifindex of the interface to get the name from.
813 * The use of raw_seqcount_begin() and cond_resched() before
814 * retrying is required as we want to give the writers a chance
815 * to complete when CONFIG_PREEMPT is not set.
817 int netdev_get_name(struct net *net, char *name, int ifindex)
819 struct net_device *dev;
823 seq = raw_seqcount_begin(&devnet_rename_seq);
825 dev = dev_get_by_index_rcu(net, ifindex);
831 strcpy(name, dev->name);
833 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
842 * dev_getbyhwaddr_rcu - find a device by its hardware address
843 * @net: the applicable net namespace
844 * @type: media type of device
845 * @ha: hardware address
847 * Search for an interface by MAC address. Returns NULL if the device
848 * is not found or a pointer to the device.
849 * The caller must hold RCU or RTNL.
850 * The returned device has not had its ref count increased
851 * and the caller must therefore be careful about locking
855 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
858 struct net_device *dev;
860 for_each_netdev_rcu(net, dev)
861 if (dev->type == type &&
862 !memcmp(dev->dev_addr, ha, dev->addr_len))
867 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
869 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
871 struct net_device *dev;
874 for_each_netdev(net, dev)
875 if (dev->type == type)
880 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
882 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
884 struct net_device *dev, *ret = NULL;
887 for_each_netdev_rcu(net, dev)
888 if (dev->type == type) {
896 EXPORT_SYMBOL(dev_getfirstbyhwtype);
899 * dev_get_by_flags_rcu - find any device with given flags
900 * @net: the applicable net namespace
901 * @if_flags: IFF_* values
902 * @mask: bitmask of bits in if_flags to check
904 * Search for any interface with the given flags. Returns NULL if a device
905 * is not found or a pointer to the device. Must be called inside
906 * rcu_read_lock(), and result refcount is unchanged.
909 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
912 struct net_device *dev, *ret;
915 for_each_netdev_rcu(net, dev) {
916 if (((dev->flags ^ if_flags) & mask) == 0) {
923 EXPORT_SYMBOL(dev_get_by_flags_rcu);
926 * dev_valid_name - check if name is okay for network device
929 * Network device names need to be valid file names to
930 * to allow sysfs to work. We also disallow any kind of
933 bool dev_valid_name(const char *name)
937 if (strlen(name) >= IFNAMSIZ)
939 if (!strcmp(name, ".") || !strcmp(name, ".."))
943 if (*name == '/' || isspace(*name))
949 EXPORT_SYMBOL(dev_valid_name);
952 * __dev_alloc_name - allocate a name for a device
953 * @net: network namespace to allocate the device name in
954 * @name: name format string
955 * @buf: scratch buffer and result name string
957 * Passed a format string - eg "lt%d" it will try and find a suitable
958 * id. It scans list of devices to build up a free map, then chooses
959 * the first empty slot. The caller must hold the dev_base or rtnl lock
960 * while allocating the name and adding the device in order to avoid
962 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
963 * Returns the number of the unit assigned or a negative errno code.
966 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
970 const int max_netdevices = 8*PAGE_SIZE;
971 unsigned long *inuse;
972 struct net_device *d;
974 p = strnchr(name, IFNAMSIZ-1, '%');
977 * Verify the string as this thing may have come from
978 * the user. There must be either one "%d" and no other "%"
981 if (p[1] != 'd' || strchr(p + 2, '%'))
984 /* Use one page as a bit array of possible slots */
985 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
989 for_each_netdev(net, d) {
990 if (!sscanf(d->name, name, &i))
992 if (i < 0 || i >= max_netdevices)
995 /* avoid cases where sscanf is not exact inverse of printf */
996 snprintf(buf, IFNAMSIZ, name, i);
997 if (!strncmp(buf, d->name, IFNAMSIZ))
1001 i = find_first_zero_bit(inuse, max_netdevices);
1002 free_page((unsigned long) inuse);
1006 snprintf(buf, IFNAMSIZ, name, i);
1007 if (!__dev_get_by_name(net, buf))
1010 /* It is possible to run out of possible slots
1011 * when the name is long and there isn't enough space left
1012 * for the digits, or if all bits are used.
1018 * dev_alloc_name - allocate a name for a device
1020 * @name: name format string
1022 * Passed a format string - eg "lt%d" it will try and find a suitable
1023 * id. It scans list of devices to build up a free map, then chooses
1024 * the first empty slot. The caller must hold the dev_base or rtnl lock
1025 * while allocating the name and adding the device in order to avoid
1027 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1028 * Returns the number of the unit assigned or a negative errno code.
1031 int dev_alloc_name(struct net_device *dev, const char *name)
1037 BUG_ON(!dev_net(dev));
1039 ret = __dev_alloc_name(net, name, buf);
1041 strlcpy(dev->name, buf, IFNAMSIZ);
1044 EXPORT_SYMBOL(dev_alloc_name);
1046 static int dev_alloc_name_ns(struct net *net,
1047 struct net_device *dev,
1053 ret = __dev_alloc_name(net, name, buf);
1055 strlcpy(dev->name, buf, IFNAMSIZ);
1059 static int dev_get_valid_name(struct net *net,
1060 struct net_device *dev,
1065 if (!dev_valid_name(name))
1068 if (strchr(name, '%'))
1069 return dev_alloc_name_ns(net, dev, name);
1070 else if (__dev_get_by_name(net, name))
1072 else if (dev->name != name)
1073 strlcpy(dev->name, name, IFNAMSIZ);
1079 * dev_change_name - change name of a device
1081 * @newname: name (or format string) must be at least IFNAMSIZ
1083 * Change name of a device, can pass format strings "eth%d".
1086 int dev_change_name(struct net_device *dev, const char *newname)
1088 char oldname[IFNAMSIZ];
1094 BUG_ON(!dev_net(dev));
1097 if (dev->flags & IFF_UP)
1100 write_seqcount_begin(&devnet_rename_seq);
1102 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1103 write_seqcount_end(&devnet_rename_seq);
1107 memcpy(oldname, dev->name, IFNAMSIZ);
1109 err = dev_get_valid_name(net, dev, newname);
1111 write_seqcount_end(&devnet_rename_seq);
1116 ret = device_rename(&dev->dev, dev->name);
1118 memcpy(dev->name, oldname, IFNAMSIZ);
1119 write_seqcount_end(&devnet_rename_seq);
1123 write_seqcount_end(&devnet_rename_seq);
1125 netdev_adjacent_rename_links(dev, oldname);
1127 write_lock_bh(&dev_base_lock);
1128 hlist_del_rcu(&dev->name_hlist);
1129 write_unlock_bh(&dev_base_lock);
1133 write_lock_bh(&dev_base_lock);
1134 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1135 write_unlock_bh(&dev_base_lock);
1137 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1138 ret = notifier_to_errno(ret);
1141 /* err >= 0 after dev_alloc_name() or stores the first errno */
1144 write_seqcount_begin(&devnet_rename_seq);
1145 memcpy(dev->name, oldname, IFNAMSIZ);
1146 memcpy(oldname, newname, IFNAMSIZ);
1149 pr_err("%s: name change rollback failed: %d\n",
1158 * dev_set_alias - change ifalias of a device
1160 * @alias: name up to IFALIASZ
1161 * @len: limit of bytes to copy from info
1163 * Set ifalias for a device,
1165 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1171 if (len >= IFALIASZ)
1175 kfree(dev->ifalias);
1176 dev->ifalias = NULL;
1180 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1183 dev->ifalias = new_ifalias;
1185 strlcpy(dev->ifalias, alias, len+1);
1191 * netdev_features_change - device changes features
1192 * @dev: device to cause notification
1194 * Called to indicate a device has changed features.
1196 void netdev_features_change(struct net_device *dev)
1198 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1200 EXPORT_SYMBOL(netdev_features_change);
1203 * netdev_state_change - device changes state
1204 * @dev: device to cause notification
1206 * Called to indicate a device has changed state. This function calls
1207 * the notifier chains for netdev_chain and sends a NEWLINK message
1208 * to the routing socket.
1210 void netdev_state_change(struct net_device *dev)
1212 if (dev->flags & IFF_UP) {
1213 struct netdev_notifier_change_info change_info;
1215 change_info.flags_changed = 0;
1216 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1218 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1221 EXPORT_SYMBOL(netdev_state_change);
1224 * netdev_notify_peers - notify network peers about existence of @dev
1225 * @dev: network device
1227 * Generate traffic such that interested network peers are aware of
1228 * @dev, such as by generating a gratuitous ARP. This may be used when
1229 * a device wants to inform the rest of the network about some sort of
1230 * reconfiguration such as a failover event or virtual machine
1233 void netdev_notify_peers(struct net_device *dev)
1236 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1239 EXPORT_SYMBOL(netdev_notify_peers);
1241 static int __dev_open(struct net_device *dev)
1243 const struct net_device_ops *ops = dev->netdev_ops;
1248 if (!netif_device_present(dev))
1251 /* Block netpoll from trying to do any rx path servicing.
1252 * If we don't do this there is a chance ndo_poll_controller
1253 * or ndo_poll may be running while we open the device
1255 netpoll_rx_disable(dev);
1257 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1258 ret = notifier_to_errno(ret);
1262 set_bit(__LINK_STATE_START, &dev->state);
1264 if (ops->ndo_validate_addr)
1265 ret = ops->ndo_validate_addr(dev);
1267 if (!ret && ops->ndo_open)
1268 ret = ops->ndo_open(dev);
1270 netpoll_rx_enable(dev);
1273 clear_bit(__LINK_STATE_START, &dev->state);
1275 dev->flags |= IFF_UP;
1276 net_dmaengine_get();
1277 dev_set_rx_mode(dev);
1279 add_device_randomness(dev->dev_addr, dev->addr_len);
1286 * dev_open - prepare an interface for use.
1287 * @dev: device to open
1289 * Takes a device from down to up state. The device's private open
1290 * function is invoked and then the multicast lists are loaded. Finally
1291 * the device is moved into the up state and a %NETDEV_UP message is
1292 * sent to the netdev notifier chain.
1294 * Calling this function on an active interface is a nop. On a failure
1295 * a negative errno code is returned.
1297 int dev_open(struct net_device *dev)
1301 if (dev->flags & IFF_UP)
1304 ret = __dev_open(dev);
1308 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1309 call_netdevice_notifiers(NETDEV_UP, dev);
1313 EXPORT_SYMBOL(dev_open);
1315 static int __dev_close_many(struct list_head *head)
1317 struct net_device *dev;
1322 list_for_each_entry(dev, head, close_list) {
1323 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1325 clear_bit(__LINK_STATE_START, &dev->state);
1327 /* Synchronize to scheduled poll. We cannot touch poll list, it
1328 * can be even on different cpu. So just clear netif_running().
1330 * dev->stop() will invoke napi_disable() on all of it's
1331 * napi_struct instances on this device.
1333 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1336 dev_deactivate_many(head);
1338 list_for_each_entry(dev, head, close_list) {
1339 const struct net_device_ops *ops = dev->netdev_ops;
1342 * Call the device specific close. This cannot fail.
1343 * Only if device is UP
1345 * We allow it to be called even after a DETACH hot-plug
1351 dev->flags &= ~IFF_UP;
1352 net_dmaengine_put();
1358 static int __dev_close(struct net_device *dev)
1363 /* Temporarily disable netpoll until the interface is down */
1364 netpoll_rx_disable(dev);
1366 list_add(&dev->close_list, &single);
1367 retval = __dev_close_many(&single);
1370 netpoll_rx_enable(dev);
1374 static int dev_close_many(struct list_head *head)
1376 struct net_device *dev, *tmp;
1378 /* Remove the devices that don't need to be closed */
1379 list_for_each_entry_safe(dev, tmp, head, close_list)
1380 if (!(dev->flags & IFF_UP))
1381 list_del_init(&dev->close_list);
1383 __dev_close_many(head);
1385 list_for_each_entry_safe(dev, tmp, head, close_list) {
1386 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1387 call_netdevice_notifiers(NETDEV_DOWN, dev);
1388 list_del_init(&dev->close_list);
1395 * dev_close - shutdown an interface.
1396 * @dev: device to shutdown
1398 * This function moves an active device into down state. A
1399 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1400 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1403 int dev_close(struct net_device *dev)
1405 if (dev->flags & IFF_UP) {
1408 /* Block netpoll rx while the interface is going down */
1409 netpoll_rx_disable(dev);
1411 list_add(&dev->close_list, &single);
1412 dev_close_many(&single);
1415 netpoll_rx_enable(dev);
1419 EXPORT_SYMBOL(dev_close);
1423 * dev_disable_lro - disable Large Receive Offload on a device
1426 * Disable Large Receive Offload (LRO) on a net device. Must be
1427 * called under RTNL. This is needed if received packets may be
1428 * forwarded to another interface.
1430 void dev_disable_lro(struct net_device *dev)
1433 * If we're trying to disable lro on a vlan device
1434 * use the underlying physical device instead
1436 if (is_vlan_dev(dev))
1437 dev = vlan_dev_real_dev(dev);
1439 /* the same for macvlan devices */
1440 if (netif_is_macvlan(dev))
1441 dev = macvlan_dev_real_dev(dev);
1443 dev->wanted_features &= ~NETIF_F_LRO;
1444 netdev_update_features(dev);
1446 if (unlikely(dev->features & NETIF_F_LRO))
1447 netdev_WARN(dev, "failed to disable LRO!\n");
1449 EXPORT_SYMBOL(dev_disable_lro);
1451 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1452 struct net_device *dev)
1454 struct netdev_notifier_info info;
1456 netdev_notifier_info_init(&info, dev);
1457 return nb->notifier_call(nb, val, &info);
1460 static int dev_boot_phase = 1;
1463 * register_netdevice_notifier - register a network notifier block
1466 * Register a notifier to be called when network device events occur.
1467 * The notifier passed is linked into the kernel structures and must
1468 * not be reused until it has been unregistered. A negative errno code
1469 * is returned on a failure.
1471 * When registered all registration and up events are replayed
1472 * to the new notifier to allow device to have a race free
1473 * view of the network device list.
1476 int register_netdevice_notifier(struct notifier_block *nb)
1478 struct net_device *dev;
1479 struct net_device *last;
1484 err = raw_notifier_chain_register(&netdev_chain, nb);
1490 for_each_netdev(net, dev) {
1491 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1492 err = notifier_to_errno(err);
1496 if (!(dev->flags & IFF_UP))
1499 call_netdevice_notifier(nb, NETDEV_UP, dev);
1510 for_each_netdev(net, dev) {
1514 if (dev->flags & IFF_UP) {
1515 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1517 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1519 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1524 raw_notifier_chain_unregister(&netdev_chain, nb);
1527 EXPORT_SYMBOL(register_netdevice_notifier);
1530 * unregister_netdevice_notifier - unregister a network notifier block
1533 * Unregister a notifier previously registered by
1534 * register_netdevice_notifier(). The notifier is unlinked into the
1535 * kernel structures and may then be reused. A negative errno code
1536 * is returned on a failure.
1538 * After unregistering unregister and down device events are synthesized
1539 * for all devices on the device list to the removed notifier to remove
1540 * the need for special case cleanup code.
1543 int unregister_netdevice_notifier(struct notifier_block *nb)
1545 struct net_device *dev;
1550 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1555 for_each_netdev(net, dev) {
1556 if (dev->flags & IFF_UP) {
1557 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1559 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1561 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1568 EXPORT_SYMBOL(unregister_netdevice_notifier);
1571 * call_netdevice_notifiers_info - call all network notifier blocks
1572 * @val: value passed unmodified to notifier function
1573 * @dev: net_device pointer passed unmodified to notifier function
1574 * @info: notifier information data
1576 * Call all network notifier blocks. Parameters and return value
1577 * are as for raw_notifier_call_chain().
1580 static int call_netdevice_notifiers_info(unsigned long val,
1581 struct net_device *dev,
1582 struct netdev_notifier_info *info)
1585 netdev_notifier_info_init(info, dev);
1586 return raw_notifier_call_chain(&netdev_chain, val, info);
1590 * call_netdevice_notifiers - call all network notifier blocks
1591 * @val: value passed unmodified to notifier function
1592 * @dev: net_device pointer passed unmodified to notifier function
1594 * Call all network notifier blocks. Parameters and return value
1595 * are as for raw_notifier_call_chain().
1598 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1600 struct netdev_notifier_info info;
1602 return call_netdevice_notifiers_info(val, dev, &info);
1604 EXPORT_SYMBOL(call_netdevice_notifiers);
1606 static struct static_key netstamp_needed __read_mostly;
1607 #ifdef HAVE_JUMP_LABEL
1608 /* We are not allowed to call static_key_slow_dec() from irq context
1609 * If net_disable_timestamp() is called from irq context, defer the
1610 * static_key_slow_dec() calls.
1612 static atomic_t netstamp_needed_deferred;
1615 void net_enable_timestamp(void)
1617 #ifdef HAVE_JUMP_LABEL
1618 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1622 static_key_slow_dec(&netstamp_needed);
1626 static_key_slow_inc(&netstamp_needed);
1628 EXPORT_SYMBOL(net_enable_timestamp);
1630 void net_disable_timestamp(void)
1632 #ifdef HAVE_JUMP_LABEL
1633 if (in_interrupt()) {
1634 atomic_inc(&netstamp_needed_deferred);
1638 static_key_slow_dec(&netstamp_needed);
1640 EXPORT_SYMBOL(net_disable_timestamp);
1642 static inline void net_timestamp_set(struct sk_buff *skb)
1644 skb->tstamp.tv64 = 0;
1645 if (static_key_false(&netstamp_needed))
1646 __net_timestamp(skb);
1649 #define net_timestamp_check(COND, SKB) \
1650 if (static_key_false(&netstamp_needed)) { \
1651 if ((COND) && !(SKB)->tstamp.tv64) \
1652 __net_timestamp(SKB); \
1655 static inline bool is_skb_forwardable(struct net_device *dev,
1656 struct sk_buff *skb)
1660 if (!(dev->flags & IFF_UP))
1663 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1664 if (skb->len <= len)
1667 /* if TSO is enabled, we don't care about the length as the packet
1668 * could be forwarded without being segmented before
1670 if (skb_is_gso(skb))
1677 * dev_forward_skb - loopback an skb to another netif
1679 * @dev: destination network device
1680 * @skb: buffer to forward
1683 * NET_RX_SUCCESS (no congestion)
1684 * NET_RX_DROP (packet was dropped, but freed)
1686 * dev_forward_skb can be used for injecting an skb from the
1687 * start_xmit function of one device into the receive queue
1688 * of another device.
1690 * The receiving device may be in another namespace, so
1691 * we have to clear all information in the skb that could
1692 * impact namespace isolation.
1694 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1696 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1697 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1698 atomic_long_inc(&dev->rx_dropped);
1704 if (unlikely(!is_skb_forwardable(dev, skb))) {
1705 atomic_long_inc(&dev->rx_dropped);
1710 skb_scrub_packet(skb, true);
1711 skb->protocol = eth_type_trans(skb, dev);
1713 return netif_rx_internal(skb);
1715 EXPORT_SYMBOL_GPL(dev_forward_skb);
1717 static inline int deliver_skb(struct sk_buff *skb,
1718 struct packet_type *pt_prev,
1719 struct net_device *orig_dev)
1721 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1723 atomic_inc(&skb->users);
1724 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1727 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1729 if (!ptype->af_packet_priv || !skb->sk)
1732 if (ptype->id_match)
1733 return ptype->id_match(ptype, skb->sk);
1734 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1741 * Support routine. Sends outgoing frames to any network
1742 * taps currently in use.
1745 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1747 struct packet_type *ptype;
1748 struct sk_buff *skb2 = NULL;
1749 struct packet_type *pt_prev = NULL;
1752 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1753 /* Never send packets back to the socket
1754 * they originated from - MvS (miquels@drinkel.ow.org)
1756 if ((ptype->dev == dev || !ptype->dev) &&
1757 (!skb_loop_sk(ptype, skb))) {
1759 deliver_skb(skb2, pt_prev, skb->dev);
1764 skb2 = skb_clone(skb, GFP_ATOMIC);
1768 net_timestamp_set(skb2);
1770 /* skb->nh should be correctly
1771 set by sender, so that the second statement is
1772 just protection against buggy protocols.
1774 skb_reset_mac_header(skb2);
1776 if (skb_network_header(skb2) < skb2->data ||
1777 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1778 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1779 ntohs(skb2->protocol),
1781 skb_reset_network_header(skb2);
1784 skb2->transport_header = skb2->network_header;
1785 skb2->pkt_type = PACKET_OUTGOING;
1790 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1795 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1796 * @dev: Network device
1797 * @txq: number of queues available
1799 * If real_num_tx_queues is changed the tc mappings may no longer be
1800 * valid. To resolve this verify the tc mapping remains valid and if
1801 * not NULL the mapping. With no priorities mapping to this
1802 * offset/count pair it will no longer be used. In the worst case TC0
1803 * is invalid nothing can be done so disable priority mappings. If is
1804 * expected that drivers will fix this mapping if they can before
1805 * calling netif_set_real_num_tx_queues.
1807 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1810 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1812 /* If TC0 is invalidated disable TC mapping */
1813 if (tc->offset + tc->count > txq) {
1814 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1819 /* Invalidated prio to tc mappings set to TC0 */
1820 for (i = 1; i < TC_BITMASK + 1; i++) {
1821 int q = netdev_get_prio_tc_map(dev, i);
1823 tc = &dev->tc_to_txq[q];
1824 if (tc->offset + tc->count > txq) {
1825 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1827 netdev_set_prio_tc_map(dev, i, 0);
1833 static DEFINE_MUTEX(xps_map_mutex);
1834 #define xmap_dereference(P) \
1835 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1837 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1840 struct xps_map *map = NULL;
1844 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1846 for (pos = 0; map && pos < map->len; pos++) {
1847 if (map->queues[pos] == index) {
1849 map->queues[pos] = map->queues[--map->len];
1851 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1852 kfree_rcu(map, rcu);
1862 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1864 struct xps_dev_maps *dev_maps;
1866 bool active = false;
1868 mutex_lock(&xps_map_mutex);
1869 dev_maps = xmap_dereference(dev->xps_maps);
1874 for_each_possible_cpu(cpu) {
1875 for (i = index; i < dev->num_tx_queues; i++) {
1876 if (!remove_xps_queue(dev_maps, cpu, i))
1879 if (i == dev->num_tx_queues)
1884 RCU_INIT_POINTER(dev->xps_maps, NULL);
1885 kfree_rcu(dev_maps, rcu);
1888 for (i = index; i < dev->num_tx_queues; i++)
1889 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1893 mutex_unlock(&xps_map_mutex);
1896 static struct xps_map *expand_xps_map(struct xps_map *map,
1899 struct xps_map *new_map;
1900 int alloc_len = XPS_MIN_MAP_ALLOC;
1903 for (pos = 0; map && pos < map->len; pos++) {
1904 if (map->queues[pos] != index)
1909 /* Need to add queue to this CPU's existing map */
1911 if (pos < map->alloc_len)
1914 alloc_len = map->alloc_len * 2;
1917 /* Need to allocate new map to store queue on this CPU's map */
1918 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1923 for (i = 0; i < pos; i++)
1924 new_map->queues[i] = map->queues[i];
1925 new_map->alloc_len = alloc_len;
1931 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1934 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1935 struct xps_map *map, *new_map;
1936 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1937 int cpu, numa_node_id = -2;
1938 bool active = false;
1940 mutex_lock(&xps_map_mutex);
1942 dev_maps = xmap_dereference(dev->xps_maps);
1944 /* allocate memory for queue storage */
1945 for_each_online_cpu(cpu) {
1946 if (!cpumask_test_cpu(cpu, mask))
1950 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1951 if (!new_dev_maps) {
1952 mutex_unlock(&xps_map_mutex);
1956 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1959 map = expand_xps_map(map, cpu, index);
1963 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1967 goto out_no_new_maps;
1969 for_each_possible_cpu(cpu) {
1970 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1971 /* add queue to CPU maps */
1974 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1975 while ((pos < map->len) && (map->queues[pos] != index))
1978 if (pos == map->len)
1979 map->queues[map->len++] = index;
1981 if (numa_node_id == -2)
1982 numa_node_id = cpu_to_node(cpu);
1983 else if (numa_node_id != cpu_to_node(cpu))
1986 } else if (dev_maps) {
1987 /* fill in the new device map from the old device map */
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1994 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1996 /* Cleanup old maps */
1998 for_each_possible_cpu(cpu) {
1999 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2000 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2001 if (map && map != new_map)
2002 kfree_rcu(map, rcu);
2005 kfree_rcu(dev_maps, rcu);
2008 dev_maps = new_dev_maps;
2012 /* update Tx queue numa node */
2013 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2014 (numa_node_id >= 0) ? numa_node_id :
2020 /* removes queue from unused CPUs */
2021 for_each_possible_cpu(cpu) {
2022 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2025 if (remove_xps_queue(dev_maps, cpu, index))
2029 /* free map if not active */
2031 RCU_INIT_POINTER(dev->xps_maps, NULL);
2032 kfree_rcu(dev_maps, rcu);
2036 mutex_unlock(&xps_map_mutex);
2040 /* remove any maps that we added */
2041 for_each_possible_cpu(cpu) {
2042 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2043 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2045 if (new_map && new_map != map)
2049 mutex_unlock(&xps_map_mutex);
2051 kfree(new_dev_maps);
2054 EXPORT_SYMBOL(netif_set_xps_queue);
2058 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2059 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2061 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2065 if (txq < 1 || txq > dev->num_tx_queues)
2068 if (dev->reg_state == NETREG_REGISTERED ||
2069 dev->reg_state == NETREG_UNREGISTERING) {
2072 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2078 netif_setup_tc(dev, txq);
2080 if (txq < dev->real_num_tx_queues) {
2081 qdisc_reset_all_tx_gt(dev, txq);
2083 netif_reset_xps_queues_gt(dev, txq);
2088 dev->real_num_tx_queues = txq;
2091 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2095 * netif_set_real_num_rx_queues - set actual number of RX queues used
2096 * @dev: Network device
2097 * @rxq: Actual number of RX queues
2099 * This must be called either with the rtnl_lock held or before
2100 * registration of the net device. Returns 0 on success, or a
2101 * negative error code. If called before registration, it always
2104 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2108 if (rxq < 1 || rxq > dev->num_rx_queues)
2111 if (dev->reg_state == NETREG_REGISTERED) {
2114 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2120 dev->real_num_rx_queues = rxq;
2123 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2127 * netif_get_num_default_rss_queues - default number of RSS queues
2129 * This routine should set an upper limit on the number of RSS queues
2130 * used by default by multiqueue devices.
2132 int netif_get_num_default_rss_queues(void)
2134 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2136 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2138 static inline void __netif_reschedule(struct Qdisc *q)
2140 struct softnet_data *sd;
2141 unsigned long flags;
2143 local_irq_save(flags);
2144 sd = &__get_cpu_var(softnet_data);
2145 q->next_sched = NULL;
2146 *sd->output_queue_tailp = q;
2147 sd->output_queue_tailp = &q->next_sched;
2148 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2149 local_irq_restore(flags);
2152 void __netif_schedule(struct Qdisc *q)
2154 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2155 __netif_reschedule(q);
2157 EXPORT_SYMBOL(__netif_schedule);
2159 struct dev_kfree_skb_cb {
2160 enum skb_free_reason reason;
2163 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2165 return (struct dev_kfree_skb_cb *)skb->cb;
2168 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2170 unsigned long flags;
2172 if (likely(atomic_read(&skb->users) == 1)) {
2174 atomic_set(&skb->users, 0);
2175 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2178 get_kfree_skb_cb(skb)->reason = reason;
2179 local_irq_save(flags);
2180 skb->next = __this_cpu_read(softnet_data.completion_queue);
2181 __this_cpu_write(softnet_data.completion_queue, skb);
2182 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2183 local_irq_restore(flags);
2185 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2187 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2189 if (in_irq() || irqs_disabled())
2190 __dev_kfree_skb_irq(skb, reason);
2194 EXPORT_SYMBOL(__dev_kfree_skb_any);
2198 * netif_device_detach - mark device as removed
2199 * @dev: network device
2201 * Mark device as removed from system and therefore no longer available.
2203 void netif_device_detach(struct net_device *dev)
2205 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2206 netif_running(dev)) {
2207 netif_tx_stop_all_queues(dev);
2210 EXPORT_SYMBOL(netif_device_detach);
2213 * netif_device_attach - mark device as attached
2214 * @dev: network device
2216 * Mark device as attached from system and restart if needed.
2218 void netif_device_attach(struct net_device *dev)
2220 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2221 netif_running(dev)) {
2222 netif_tx_wake_all_queues(dev);
2223 __netdev_watchdog_up(dev);
2226 EXPORT_SYMBOL(netif_device_attach);
2228 static void skb_warn_bad_offload(const struct sk_buff *skb)
2230 static const netdev_features_t null_features = 0;
2231 struct net_device *dev = skb->dev;
2232 const char *driver = "";
2234 if (!net_ratelimit())
2237 if (dev && dev->dev.parent)
2238 driver = dev_driver_string(dev->dev.parent);
2240 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2241 "gso_type=%d ip_summed=%d\n",
2242 driver, dev ? &dev->features : &null_features,
2243 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2244 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2245 skb_shinfo(skb)->gso_type, skb->ip_summed);
2249 * Invalidate hardware checksum when packet is to be mangled, and
2250 * complete checksum manually on outgoing path.
2252 int skb_checksum_help(struct sk_buff *skb)
2255 int ret = 0, offset;
2257 if (skb->ip_summed == CHECKSUM_COMPLETE)
2258 goto out_set_summed;
2260 if (unlikely(skb_shinfo(skb)->gso_size)) {
2261 skb_warn_bad_offload(skb);
2265 /* Before computing a checksum, we should make sure no frag could
2266 * be modified by an external entity : checksum could be wrong.
2268 if (skb_has_shared_frag(skb)) {
2269 ret = __skb_linearize(skb);
2274 offset = skb_checksum_start_offset(skb);
2275 BUG_ON(offset >= skb_headlen(skb));
2276 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2278 offset += skb->csum_offset;
2279 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2281 if (skb_cloned(skb) &&
2282 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2283 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2288 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2290 skb->ip_summed = CHECKSUM_NONE;
2294 EXPORT_SYMBOL(skb_checksum_help);
2296 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2298 __be16 type = skb->protocol;
2299 int vlan_depth = skb->mac_len;
2301 /* Tunnel gso handlers can set protocol to ethernet. */
2302 if (type == htons(ETH_P_TEB)) {
2305 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2308 eth = (struct ethhdr *)skb_mac_header(skb);
2309 type = eth->h_proto;
2312 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2313 struct vlan_hdr *vh;
2315 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2318 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2319 type = vh->h_vlan_encapsulated_proto;
2320 vlan_depth += VLAN_HLEN;
2323 *depth = vlan_depth;
2329 * skb_mac_gso_segment - mac layer segmentation handler.
2330 * @skb: buffer to segment
2331 * @features: features for the output path (see dev->features)
2333 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2334 netdev_features_t features)
2336 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2337 struct packet_offload *ptype;
2338 int vlan_depth = skb->mac_len;
2339 __be16 type = skb_network_protocol(skb, &vlan_depth);
2341 if (unlikely(!type))
2342 return ERR_PTR(-EINVAL);
2344 __skb_pull(skb, vlan_depth);
2347 list_for_each_entry_rcu(ptype, &offload_base, list) {
2348 if (ptype->type == type && ptype->callbacks.gso_segment) {
2349 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2352 err = ptype->callbacks.gso_send_check(skb);
2353 segs = ERR_PTR(err);
2354 if (err || skb_gso_ok(skb, features))
2356 __skb_push(skb, (skb->data -
2357 skb_network_header(skb)));
2359 segs = ptype->callbacks.gso_segment(skb, features);
2365 __skb_push(skb, skb->data - skb_mac_header(skb));
2369 EXPORT_SYMBOL(skb_mac_gso_segment);
2372 /* openvswitch calls this on rx path, so we need a different check.
2374 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2377 return skb->ip_summed != CHECKSUM_PARTIAL;
2379 return skb->ip_summed == CHECKSUM_NONE;
2383 * __skb_gso_segment - Perform segmentation on skb.
2384 * @skb: buffer to segment
2385 * @features: features for the output path (see dev->features)
2386 * @tx_path: whether it is called in TX path
2388 * This function segments the given skb and returns a list of segments.
2390 * It may return NULL if the skb requires no segmentation. This is
2391 * only possible when GSO is used for verifying header integrity.
2393 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2394 netdev_features_t features, bool tx_path)
2396 if (unlikely(skb_needs_check(skb, tx_path))) {
2399 skb_warn_bad_offload(skb);
2401 if (skb_header_cloned(skb) &&
2402 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2403 return ERR_PTR(err);
2406 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2407 SKB_GSO_CB(skb)->encap_level = 0;
2409 skb_reset_mac_header(skb);
2410 skb_reset_mac_len(skb);
2412 return skb_mac_gso_segment(skb, features);
2414 EXPORT_SYMBOL(__skb_gso_segment);
2416 /* Take action when hardware reception checksum errors are detected. */
2418 void netdev_rx_csum_fault(struct net_device *dev)
2420 if (net_ratelimit()) {
2421 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2425 EXPORT_SYMBOL(netdev_rx_csum_fault);
2428 /* Actually, we should eliminate this check as soon as we know, that:
2429 * 1. IOMMU is present and allows to map all the memory.
2430 * 2. No high memory really exists on this machine.
2433 static int illegal_highdma(const struct net_device *dev, struct sk_buff *skb)
2435 #ifdef CONFIG_HIGHMEM
2437 if (!(dev->features & NETIF_F_HIGHDMA)) {
2438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2440 if (PageHighMem(skb_frag_page(frag)))
2445 if (PCI_DMA_BUS_IS_PHYS) {
2446 struct device *pdev = dev->dev.parent;
2450 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2451 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2452 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2453 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2462 void (*destructor)(struct sk_buff *skb);
2465 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2467 static void dev_gso_skb_destructor(struct sk_buff *skb)
2469 struct dev_gso_cb *cb;
2471 kfree_skb_list(skb->next);
2474 cb = DEV_GSO_CB(skb);
2476 cb->destructor(skb);
2480 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2481 * @skb: buffer to segment
2482 * @features: device features as applicable to this skb
2484 * This function segments the given skb and stores the list of segments
2487 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2489 struct sk_buff *segs;
2491 segs = skb_gso_segment(skb, features);
2493 /* Verifying header integrity only. */
2498 return PTR_ERR(segs);
2501 DEV_GSO_CB(skb)->destructor = skb->destructor;
2502 skb->destructor = dev_gso_skb_destructor;
2507 static netdev_features_t harmonize_features(struct sk_buff *skb,
2508 const struct net_device *dev,
2509 netdev_features_t features)
2513 if (skb->ip_summed != CHECKSUM_NONE &&
2514 !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) {
2515 features &= ~NETIF_F_ALL_CSUM;
2516 } else if (illegal_highdma(dev, skb)) {
2517 features &= ~NETIF_F_SG;
2523 netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
2524 const struct net_device *dev)
2526 __be16 protocol = skb->protocol;
2527 netdev_features_t features = dev->features;
2529 if (skb_shinfo(skb)->gso_segs > dev->gso_max_segs)
2530 features &= ~NETIF_F_GSO_MASK;
2532 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2533 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2534 protocol = veh->h_vlan_encapsulated_proto;
2535 } else if (!vlan_tx_tag_present(skb)) {
2536 return harmonize_features(skb, dev, features);
2539 features &= (dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2540 NETIF_F_HW_VLAN_STAG_TX);
2542 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2543 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2544 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2545 NETIF_F_HW_VLAN_STAG_TX;
2547 return harmonize_features(skb, dev, features);
2549 EXPORT_SYMBOL(netif_skb_dev_features);
2551 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2552 struct netdev_queue *txq)
2554 const struct net_device_ops *ops = dev->netdev_ops;
2555 int rc = NETDEV_TX_OK;
2556 unsigned int skb_len;
2558 if (likely(!skb->next)) {
2559 netdev_features_t features;
2562 * If device doesn't need skb->dst, release it right now while
2563 * its hot in this cpu cache
2565 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2568 features = netif_skb_features(skb);
2570 if (vlan_tx_tag_present(skb) &&
2571 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2572 skb = __vlan_put_tag(skb, skb->vlan_proto,
2573 vlan_tx_tag_get(skb));
2580 /* If encapsulation offload request, verify we are testing
2581 * hardware encapsulation features instead of standard
2582 * features for the netdev
2584 if (skb->encapsulation)
2585 features &= dev->hw_enc_features;
2587 if (netif_needs_gso(skb, features)) {
2588 if (unlikely(dev_gso_segment(skb, features)))
2593 if (skb_needs_linearize(skb, features) &&
2594 __skb_linearize(skb))
2597 /* If packet is not checksummed and device does not
2598 * support checksumming for this protocol, complete
2599 * checksumming here.
2601 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2602 if (skb->encapsulation)
2603 skb_set_inner_transport_header(skb,
2604 skb_checksum_start_offset(skb));
2606 skb_set_transport_header(skb,
2607 skb_checksum_start_offset(skb));
2608 if (!(features & NETIF_F_ALL_CSUM) &&
2609 skb_checksum_help(skb))
2614 if (!list_empty(&ptype_all))
2615 dev_queue_xmit_nit(skb, dev);
2618 trace_net_dev_start_xmit(skb, dev);
2619 rc = ops->ndo_start_xmit(skb, dev);
2620 trace_net_dev_xmit(skb, rc, dev, skb_len);
2621 if (rc == NETDEV_TX_OK)
2622 txq_trans_update(txq);
2628 struct sk_buff *nskb = skb->next;
2630 skb->next = nskb->next;
2633 if (!list_empty(&ptype_all))
2634 dev_queue_xmit_nit(nskb, dev);
2636 skb_len = nskb->len;
2637 trace_net_dev_start_xmit(nskb, dev);
2638 rc = ops->ndo_start_xmit(nskb, dev);
2639 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2640 if (unlikely(rc != NETDEV_TX_OK)) {
2641 if (rc & ~NETDEV_TX_MASK)
2642 goto out_kfree_gso_skb;
2643 nskb->next = skb->next;
2647 txq_trans_update(txq);
2648 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2649 return NETDEV_TX_BUSY;
2650 } while (skb->next);
2653 if (likely(skb->next == NULL)) {
2654 skb->destructor = DEV_GSO_CB(skb)->destructor;
2663 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2665 static void qdisc_pkt_len_init(struct sk_buff *skb)
2667 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2669 qdisc_skb_cb(skb)->pkt_len = skb->len;
2671 /* To get more precise estimation of bytes sent on wire,
2672 * we add to pkt_len the headers size of all segments
2674 if (shinfo->gso_size) {
2675 unsigned int hdr_len;
2676 u16 gso_segs = shinfo->gso_segs;
2678 /* mac layer + network layer */
2679 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2681 /* + transport layer */
2682 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2683 hdr_len += tcp_hdrlen(skb);
2685 hdr_len += sizeof(struct udphdr);
2687 if (shinfo->gso_type & SKB_GSO_DODGY)
2688 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2691 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2695 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2696 struct net_device *dev,
2697 struct netdev_queue *txq)
2699 spinlock_t *root_lock = qdisc_lock(q);
2703 qdisc_pkt_len_init(skb);
2704 qdisc_calculate_pkt_len(skb, q);
2706 * Heuristic to force contended enqueues to serialize on a
2707 * separate lock before trying to get qdisc main lock.
2708 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2709 * and dequeue packets faster.
2711 contended = qdisc_is_running(q);
2712 if (unlikely(contended))
2713 spin_lock(&q->busylock);
2715 spin_lock(root_lock);
2716 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2719 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2720 qdisc_run_begin(q)) {
2722 * This is a work-conserving queue; there are no old skbs
2723 * waiting to be sent out; and the qdisc is not running -
2724 * xmit the skb directly.
2726 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2729 qdisc_bstats_update(q, skb);
2731 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2732 if (unlikely(contended)) {
2733 spin_unlock(&q->busylock);
2740 rc = NET_XMIT_SUCCESS;
2743 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2744 if (qdisc_run_begin(q)) {
2745 if (unlikely(contended)) {
2746 spin_unlock(&q->busylock);
2752 spin_unlock(root_lock);
2753 if (unlikely(contended))
2754 spin_unlock(&q->busylock);
2758 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2759 static void skb_update_prio(struct sk_buff *skb)
2761 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2763 if (!skb->priority && skb->sk && map) {
2764 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2766 if (prioidx < map->priomap_len)
2767 skb->priority = map->priomap[prioidx];
2771 #define skb_update_prio(skb)
2774 static DEFINE_PER_CPU(int, xmit_recursion);
2775 #define RECURSION_LIMIT 10
2778 * dev_loopback_xmit - loop back @skb
2779 * @skb: buffer to transmit
2781 int dev_loopback_xmit(struct sk_buff *skb)
2783 skb_reset_mac_header(skb);
2784 __skb_pull(skb, skb_network_offset(skb));
2785 skb->pkt_type = PACKET_LOOPBACK;
2786 skb->ip_summed = CHECKSUM_UNNECESSARY;
2787 WARN_ON(!skb_dst(skb));
2792 EXPORT_SYMBOL(dev_loopback_xmit);
2795 * __dev_queue_xmit - transmit a buffer
2796 * @skb: buffer to transmit
2797 * @accel_priv: private data used for L2 forwarding offload
2799 * Queue a buffer for transmission to a network device. The caller must
2800 * have set the device and priority and built the buffer before calling
2801 * this function. The function can be called from an interrupt.
2803 * A negative errno code is returned on a failure. A success does not
2804 * guarantee the frame will be transmitted as it may be dropped due
2805 * to congestion or traffic shaping.
2807 * -----------------------------------------------------------------------------------
2808 * I notice this method can also return errors from the queue disciplines,
2809 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2812 * Regardless of the return value, the skb is consumed, so it is currently
2813 * difficult to retry a send to this method. (You can bump the ref count
2814 * before sending to hold a reference for retry if you are careful.)
2816 * When calling this method, interrupts MUST be enabled. This is because
2817 * the BH enable code must have IRQs enabled so that it will not deadlock.
2820 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2822 struct net_device *dev = skb->dev;
2823 struct netdev_queue *txq;
2827 skb_reset_mac_header(skb);
2829 /* Disable soft irqs for various locks below. Also
2830 * stops preemption for RCU.
2834 skb_update_prio(skb);
2836 txq = netdev_pick_tx(dev, skb, accel_priv);
2837 q = rcu_dereference_bh(txq->qdisc);
2839 #ifdef CONFIG_NET_CLS_ACT
2840 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2842 trace_net_dev_queue(skb);
2844 rc = __dev_xmit_skb(skb, q, dev, txq);
2848 /* The device has no queue. Common case for software devices:
2849 loopback, all the sorts of tunnels...
2851 Really, it is unlikely that netif_tx_lock protection is necessary
2852 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2854 However, it is possible, that they rely on protection
2857 Check this and shot the lock. It is not prone from deadlocks.
2858 Either shot noqueue qdisc, it is even simpler 8)
2860 if (dev->flags & IFF_UP) {
2861 int cpu = smp_processor_id(); /* ok because BHs are off */
2863 if (txq->xmit_lock_owner != cpu) {
2865 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2866 goto recursion_alert;
2868 HARD_TX_LOCK(dev, txq, cpu);
2870 if (!netif_xmit_stopped(txq)) {
2871 __this_cpu_inc(xmit_recursion);
2872 rc = dev_hard_start_xmit(skb, dev, txq);
2873 __this_cpu_dec(xmit_recursion);
2874 if (dev_xmit_complete(rc)) {
2875 HARD_TX_UNLOCK(dev, txq);
2879 HARD_TX_UNLOCK(dev, txq);
2880 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2883 /* Recursion is detected! It is possible,
2887 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2893 rcu_read_unlock_bh();
2898 rcu_read_unlock_bh();
2902 int dev_queue_xmit(struct sk_buff *skb)
2904 return __dev_queue_xmit(skb, NULL);
2906 EXPORT_SYMBOL(dev_queue_xmit);
2908 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2910 return __dev_queue_xmit(skb, accel_priv);
2912 EXPORT_SYMBOL(dev_queue_xmit_accel);
2915 /*=======================================================================
2917 =======================================================================*/
2919 int netdev_max_backlog __read_mostly = 1000;
2920 EXPORT_SYMBOL(netdev_max_backlog);
2922 int netdev_tstamp_prequeue __read_mostly = 1;
2923 int netdev_budget __read_mostly = 300;
2924 int weight_p __read_mostly = 64; /* old backlog weight */
2926 /* Called with irq disabled */
2927 static inline void ____napi_schedule(struct softnet_data *sd,
2928 struct napi_struct *napi)
2930 list_add_tail(&napi->poll_list, &sd->poll_list);
2931 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2936 /* One global table that all flow-based protocols share. */
2937 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2938 EXPORT_SYMBOL(rps_sock_flow_table);
2940 struct static_key rps_needed __read_mostly;
2942 static struct rps_dev_flow *
2943 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2944 struct rps_dev_flow *rflow, u16 next_cpu)
2946 if (next_cpu != RPS_NO_CPU) {
2947 #ifdef CONFIG_RFS_ACCEL
2948 struct netdev_rx_queue *rxqueue;
2949 struct rps_dev_flow_table *flow_table;
2950 struct rps_dev_flow *old_rflow;
2955 /* Should we steer this flow to a different hardware queue? */
2956 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2957 !(dev->features & NETIF_F_NTUPLE))
2959 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2960 if (rxq_index == skb_get_rx_queue(skb))
2963 rxqueue = dev->_rx + rxq_index;
2964 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2967 flow_id = skb->rxhash & flow_table->mask;
2968 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2969 rxq_index, flow_id);
2973 rflow = &flow_table->flows[flow_id];
2975 if (old_rflow->filter == rflow->filter)
2976 old_rflow->filter = RPS_NO_FILTER;
2980 per_cpu(softnet_data, next_cpu).input_queue_head;
2983 rflow->cpu = next_cpu;
2988 * get_rps_cpu is called from netif_receive_skb and returns the target
2989 * CPU from the RPS map of the receiving queue for a given skb.
2990 * rcu_read_lock must be held on entry.
2992 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2993 struct rps_dev_flow **rflowp)
2995 struct netdev_rx_queue *rxqueue;
2996 struct rps_map *map;
2997 struct rps_dev_flow_table *flow_table;
2998 struct rps_sock_flow_table *sock_flow_table;
3002 if (skb_rx_queue_recorded(skb)) {
3003 u16 index = skb_get_rx_queue(skb);
3004 if (unlikely(index >= dev->real_num_rx_queues)) {
3005 WARN_ONCE(dev->real_num_rx_queues > 1,
3006 "%s received packet on queue %u, but number "
3007 "of RX queues is %u\n",
3008 dev->name, index, dev->real_num_rx_queues);
3011 rxqueue = dev->_rx + index;
3015 map = rcu_dereference(rxqueue->rps_map);
3017 if (map->len == 1 &&
3018 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3019 tcpu = map->cpus[0];
3020 if (cpu_online(tcpu))
3024 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3028 skb_reset_network_header(skb);
3029 if (!skb_get_hash(skb))
3032 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3033 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3034 if (flow_table && sock_flow_table) {
3036 struct rps_dev_flow *rflow;
3038 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3041 next_cpu = sock_flow_table->ents[skb->rxhash &
3042 sock_flow_table->mask];
3045 * If the desired CPU (where last recvmsg was done) is
3046 * different from current CPU (one in the rx-queue flow
3047 * table entry), switch if one of the following holds:
3048 * - Current CPU is unset (equal to RPS_NO_CPU).
3049 * - Current CPU is offline.
3050 * - The current CPU's queue tail has advanced beyond the
3051 * last packet that was enqueued using this table entry.
3052 * This guarantees that all previous packets for the flow
3053 * have been dequeued, thus preserving in order delivery.
3055 if (unlikely(tcpu != next_cpu) &&
3056 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3057 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3058 rflow->last_qtail)) >= 0)) {
3060 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3063 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3071 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3073 if (cpu_online(tcpu)) {
3083 #ifdef CONFIG_RFS_ACCEL
3086 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3087 * @dev: Device on which the filter was set
3088 * @rxq_index: RX queue index
3089 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3090 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3092 * Drivers that implement ndo_rx_flow_steer() should periodically call
3093 * this function for each installed filter and remove the filters for
3094 * which it returns %true.
3096 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3097 u32 flow_id, u16 filter_id)
3099 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3100 struct rps_dev_flow_table *flow_table;
3101 struct rps_dev_flow *rflow;
3106 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3107 if (flow_table && flow_id <= flow_table->mask) {
3108 rflow = &flow_table->flows[flow_id];
3109 cpu = ACCESS_ONCE(rflow->cpu);
3110 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3111 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3112 rflow->last_qtail) <
3113 (int)(10 * flow_table->mask)))
3119 EXPORT_SYMBOL(rps_may_expire_flow);
3121 #endif /* CONFIG_RFS_ACCEL */
3123 /* Called from hardirq (IPI) context */
3124 static void rps_trigger_softirq(void *data)
3126 struct softnet_data *sd = data;
3128 ____napi_schedule(sd, &sd->backlog);
3132 #endif /* CONFIG_RPS */
3135 * Check if this softnet_data structure is another cpu one
3136 * If yes, queue it to our IPI list and return 1
3139 static int rps_ipi_queued(struct softnet_data *sd)
3142 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3145 sd->rps_ipi_next = mysd->rps_ipi_list;
3146 mysd->rps_ipi_list = sd;
3148 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3151 #endif /* CONFIG_RPS */
3155 #ifdef CONFIG_NET_FLOW_LIMIT
3156 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3159 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3161 #ifdef CONFIG_NET_FLOW_LIMIT
3162 struct sd_flow_limit *fl;
3163 struct softnet_data *sd;
3164 unsigned int old_flow, new_flow;
3166 if (qlen < (netdev_max_backlog >> 1))
3169 sd = &__get_cpu_var(softnet_data);
3172 fl = rcu_dereference(sd->flow_limit);
3174 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3175 old_flow = fl->history[fl->history_head];
3176 fl->history[fl->history_head] = new_flow;
3179 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3181 if (likely(fl->buckets[old_flow]))
3182 fl->buckets[old_flow]--;
3184 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3196 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3197 * queue (may be a remote CPU queue).
3199 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3200 unsigned int *qtail)
3202 struct softnet_data *sd;
3203 unsigned long flags;
3206 sd = &per_cpu(softnet_data, cpu);
3208 local_irq_save(flags);
3211 qlen = skb_queue_len(&sd->input_pkt_queue);
3212 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3213 if (skb_queue_len(&sd->input_pkt_queue)) {
3215 __skb_queue_tail(&sd->input_pkt_queue, skb);
3216 input_queue_tail_incr_save(sd, qtail);
3218 local_irq_restore(flags);
3219 return NET_RX_SUCCESS;
3222 /* Schedule NAPI for backlog device
3223 * We can use non atomic operation since we own the queue lock
3225 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3226 if (!rps_ipi_queued(sd))
3227 ____napi_schedule(sd, &sd->backlog);
3235 local_irq_restore(flags);
3237 atomic_long_inc(&skb->dev->rx_dropped);
3242 static int netif_rx_internal(struct sk_buff *skb)
3246 /* if netpoll wants it, pretend we never saw it */
3247 if (netpoll_rx(skb))
3250 net_timestamp_check(netdev_tstamp_prequeue, skb);
3252 trace_netif_rx(skb);
3254 if (static_key_false(&rps_needed)) {
3255 struct rps_dev_flow voidflow, *rflow = &voidflow;
3261 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3263 cpu = smp_processor_id();
3265 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3273 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3280 * netif_rx - post buffer to the network code
3281 * @skb: buffer to post
3283 * This function receives a packet from a device driver and queues it for
3284 * the upper (protocol) levels to process. It always succeeds. The buffer
3285 * may be dropped during processing for congestion control or by the
3289 * NET_RX_SUCCESS (no congestion)
3290 * NET_RX_DROP (packet was dropped)
3294 int netif_rx(struct sk_buff *skb)
3296 trace_netif_rx_entry(skb);
3298 return netif_rx_internal(skb);
3300 EXPORT_SYMBOL(netif_rx);
3302 int netif_rx_ni(struct sk_buff *skb)
3306 trace_netif_rx_ni_entry(skb);
3309 err = netif_rx_internal(skb);
3310 if (local_softirq_pending())
3316 EXPORT_SYMBOL(netif_rx_ni);
3318 static void net_tx_action(struct softirq_action *h)
3320 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3322 if (sd->completion_queue) {
3323 struct sk_buff *clist;
3325 local_irq_disable();
3326 clist = sd->completion_queue;
3327 sd->completion_queue = NULL;
3331 struct sk_buff *skb = clist;
3332 clist = clist->next;
3334 WARN_ON(atomic_read(&skb->users));
3335 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3336 trace_consume_skb(skb);
3338 trace_kfree_skb(skb, net_tx_action);
3343 if (sd->output_queue) {
3346 local_irq_disable();
3347 head = sd->output_queue;
3348 sd->output_queue = NULL;
3349 sd->output_queue_tailp = &sd->output_queue;
3353 struct Qdisc *q = head;
3354 spinlock_t *root_lock;
3356 head = head->next_sched;
3358 root_lock = qdisc_lock(q);
3359 if (spin_trylock(root_lock)) {
3360 smp_mb__before_clear_bit();
3361 clear_bit(__QDISC_STATE_SCHED,
3364 spin_unlock(root_lock);
3366 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3368 __netif_reschedule(q);
3370 smp_mb__before_clear_bit();
3371 clear_bit(__QDISC_STATE_SCHED,
3379 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3380 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3381 /* This hook is defined here for ATM LANE */
3382 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3383 unsigned char *addr) __read_mostly;
3384 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3387 #ifdef CONFIG_NET_CLS_ACT
3388 /* TODO: Maybe we should just force sch_ingress to be compiled in
3389 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3390 * a compare and 2 stores extra right now if we dont have it on
3391 * but have CONFIG_NET_CLS_ACT
3392 * NOTE: This doesn't stop any functionality; if you dont have
3393 * the ingress scheduler, you just can't add policies on ingress.
3396 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3398 struct net_device *dev = skb->dev;
3399 u32 ttl = G_TC_RTTL(skb->tc_verd);
3400 int result = TC_ACT_OK;
3403 if (unlikely(MAX_RED_LOOP < ttl++)) {
3404 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3405 skb->skb_iif, dev->ifindex);
3409 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3410 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3413 if (q != &noop_qdisc) {
3414 spin_lock(qdisc_lock(q));
3415 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3416 result = qdisc_enqueue_root(skb, q);
3417 spin_unlock(qdisc_lock(q));
3423 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3424 struct packet_type **pt_prev,
3425 int *ret, struct net_device *orig_dev)
3427 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3429 if (!rxq || rxq->qdisc == &noop_qdisc)
3433 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3437 switch (ing_filter(skb, rxq)) {
3451 * netdev_rx_handler_register - register receive handler
3452 * @dev: device to register a handler for
3453 * @rx_handler: receive handler to register
3454 * @rx_handler_data: data pointer that is used by rx handler
3456 * Register a receive hander for a device. This handler will then be
3457 * called from __netif_receive_skb. A negative errno code is returned
3460 * The caller must hold the rtnl_mutex.
3462 * For a general description of rx_handler, see enum rx_handler_result.
3464 int netdev_rx_handler_register(struct net_device *dev,
3465 rx_handler_func_t *rx_handler,
3466 void *rx_handler_data)
3470 if (dev->rx_handler)
3473 /* Note: rx_handler_data must be set before rx_handler */
3474 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3475 rcu_assign_pointer(dev->rx_handler, rx_handler);
3479 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3482 * netdev_rx_handler_unregister - unregister receive handler
3483 * @dev: device to unregister a handler from
3485 * Unregister a receive handler from a device.
3487 * The caller must hold the rtnl_mutex.
3489 void netdev_rx_handler_unregister(struct net_device *dev)
3493 RCU_INIT_POINTER(dev->rx_handler, NULL);
3494 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3495 * section has a guarantee to see a non NULL rx_handler_data
3499 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3501 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3504 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3505 * the special handling of PFMEMALLOC skbs.
3507 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3509 switch (skb->protocol) {
3510 case __constant_htons(ETH_P_ARP):
3511 case __constant_htons(ETH_P_IP):
3512 case __constant_htons(ETH_P_IPV6):
3513 case __constant_htons(ETH_P_8021Q):
3514 case __constant_htons(ETH_P_8021AD):
3521 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3523 struct packet_type *ptype, *pt_prev;
3524 rx_handler_func_t *rx_handler;
3525 struct net_device *orig_dev;
3526 struct net_device *null_or_dev;
3527 bool deliver_exact = false;
3528 int ret = NET_RX_DROP;
3531 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3533 trace_netif_receive_skb(skb);
3535 /* if we've gotten here through NAPI, check netpoll */
3536 if (netpoll_receive_skb(skb))
3539 orig_dev = skb->dev;
3541 skb_reset_network_header(skb);
3542 if (!skb_transport_header_was_set(skb))
3543 skb_reset_transport_header(skb);
3544 skb_reset_mac_len(skb);
3551 skb->skb_iif = skb->dev->ifindex;
3553 __this_cpu_inc(softnet_data.processed);
3555 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3556 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3557 skb = vlan_untag(skb);
3562 #ifdef CONFIG_NET_CLS_ACT
3563 if (skb->tc_verd & TC_NCLS) {
3564 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3572 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3573 if (!ptype->dev || ptype->dev == skb->dev) {
3575 ret = deliver_skb(skb, pt_prev, orig_dev);
3581 #ifdef CONFIG_NET_CLS_ACT
3582 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3588 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3591 if (vlan_tx_tag_present(skb)) {
3593 ret = deliver_skb(skb, pt_prev, orig_dev);
3596 if (vlan_do_receive(&skb))
3598 else if (unlikely(!skb))
3602 rx_handler = rcu_dereference(skb->dev->rx_handler);
3605 ret = deliver_skb(skb, pt_prev, orig_dev);
3608 switch (rx_handler(&skb)) {
3609 case RX_HANDLER_CONSUMED:
3610 ret = NET_RX_SUCCESS;
3612 case RX_HANDLER_ANOTHER:
3614 case RX_HANDLER_EXACT:
3615 deliver_exact = true;
3616 case RX_HANDLER_PASS:
3623 if (unlikely(vlan_tx_tag_present(skb))) {
3624 if (vlan_tx_tag_get_id(skb))
3625 skb->pkt_type = PACKET_OTHERHOST;
3626 /* Note: we might in the future use prio bits
3627 * and set skb->priority like in vlan_do_receive()
3628 * For the time being, just ignore Priority Code Point
3633 /* deliver only exact match when indicated */
3634 null_or_dev = deliver_exact ? skb->dev : NULL;
3636 type = skb->protocol;
3637 list_for_each_entry_rcu(ptype,
3638 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3639 if (ptype->type == type &&
3640 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3641 ptype->dev == orig_dev)) {
3643 ret = deliver_skb(skb, pt_prev, orig_dev);
3649 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3652 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3655 atomic_long_inc(&skb->dev->rx_dropped);
3657 /* Jamal, now you will not able to escape explaining
3658 * me how you were going to use this. :-)
3669 static int __netif_receive_skb(struct sk_buff *skb)
3673 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3674 unsigned long pflags = current->flags;
3677 * PFMEMALLOC skbs are special, they should
3678 * - be delivered to SOCK_MEMALLOC sockets only
3679 * - stay away from userspace
3680 * - have bounded memory usage
3682 * Use PF_MEMALLOC as this saves us from propagating the allocation
3683 * context down to all allocation sites.
3685 current->flags |= PF_MEMALLOC;
3686 ret = __netif_receive_skb_core(skb, true);
3687 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3689 ret = __netif_receive_skb_core(skb, false);
3694 static int netif_receive_skb_internal(struct sk_buff *skb)
3696 net_timestamp_check(netdev_tstamp_prequeue, skb);
3698 if (skb_defer_rx_timestamp(skb))
3699 return NET_RX_SUCCESS;
3702 if (static_key_false(&rps_needed)) {
3703 struct rps_dev_flow voidflow, *rflow = &voidflow;
3708 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3711 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3718 return __netif_receive_skb(skb);
3722 * netif_receive_skb - process receive buffer from network
3723 * @skb: buffer to process
3725 * netif_receive_skb() is the main receive data processing function.
3726 * It always succeeds. The buffer may be dropped during processing
3727 * for congestion control or by the protocol layers.
3729 * This function may only be called from softirq context and interrupts
3730 * should be enabled.
3732 * Return values (usually ignored):
3733 * NET_RX_SUCCESS: no congestion
3734 * NET_RX_DROP: packet was dropped
3736 int netif_receive_skb(struct sk_buff *skb)
3738 trace_netif_receive_skb_entry(skb);
3740 return netif_receive_skb_internal(skb);
3742 EXPORT_SYMBOL(netif_receive_skb);
3744 /* Network device is going away, flush any packets still pending
3745 * Called with irqs disabled.
3747 static void flush_backlog(void *arg)
3749 struct net_device *dev = arg;
3750 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3751 struct sk_buff *skb, *tmp;
3754 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3755 if (skb->dev == dev) {
3756 __skb_unlink(skb, &sd->input_pkt_queue);
3758 input_queue_head_incr(sd);
3763 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3764 if (skb->dev == dev) {
3765 __skb_unlink(skb, &sd->process_queue);
3767 input_queue_head_incr(sd);
3772 static int napi_gro_complete(struct sk_buff *skb)
3774 struct packet_offload *ptype;
3775 __be16 type = skb->protocol;
3776 struct list_head *head = &offload_base;
3779 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3781 if (NAPI_GRO_CB(skb)->count == 1) {
3782 skb_shinfo(skb)->gso_size = 0;
3787 list_for_each_entry_rcu(ptype, head, list) {
3788 if (ptype->type != type || !ptype->callbacks.gro_complete)
3791 err = ptype->callbacks.gro_complete(skb, 0);
3797 WARN_ON(&ptype->list == head);
3799 return NET_RX_SUCCESS;
3803 return netif_receive_skb_internal(skb);
3806 /* napi->gro_list contains packets ordered by age.
3807 * youngest packets at the head of it.
3808 * Complete skbs in reverse order to reduce latencies.
3810 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3812 struct sk_buff *skb, *prev = NULL;
3814 /* scan list and build reverse chain */
3815 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3820 for (skb = prev; skb; skb = prev) {
3823 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3827 napi_gro_complete(skb);
3831 napi->gro_list = NULL;
3833 EXPORT_SYMBOL(napi_gro_flush);
3835 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3838 unsigned int maclen = skb->dev->hard_header_len;
3839 u32 hash = skb_get_hash_raw(skb);
3841 for (p = napi->gro_list; p; p = p->next) {
3842 unsigned long diffs;
3844 NAPI_GRO_CB(p)->flush = 0;
3846 if (hash != skb_get_hash_raw(p)) {
3847 NAPI_GRO_CB(p)->same_flow = 0;
3851 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3852 diffs |= p->vlan_tci ^ skb->vlan_tci;
3853 if (maclen == ETH_HLEN)
3854 diffs |= compare_ether_header(skb_mac_header(p),
3855 skb_gro_mac_header(skb));
3857 diffs = memcmp(skb_mac_header(p),
3858 skb_gro_mac_header(skb),
3860 NAPI_GRO_CB(p)->same_flow = !diffs;
3864 static void skb_gro_reset_offset(struct sk_buff *skb)
3866 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3867 const skb_frag_t *frag0 = &pinfo->frags[0];
3869 NAPI_GRO_CB(skb)->data_offset = 0;
3870 NAPI_GRO_CB(skb)->frag0 = NULL;
3871 NAPI_GRO_CB(skb)->frag0_len = 0;
3873 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3875 !PageHighMem(skb_frag_page(frag0))) {
3876 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3877 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3881 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3883 struct sk_buff **pp = NULL;
3884 struct packet_offload *ptype;
3885 __be16 type = skb->protocol;
3886 struct list_head *head = &offload_base;
3888 enum gro_result ret;
3890 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3893 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3896 skb_gro_reset_offset(skb);
3897 gro_list_prepare(napi, skb);
3898 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3901 list_for_each_entry_rcu(ptype, head, list) {
3902 if (ptype->type != type || !ptype->callbacks.gro_receive)
3905 skb_set_network_header(skb, skb_gro_offset(skb));
3906 skb_reset_mac_len(skb);
3907 NAPI_GRO_CB(skb)->same_flow = 0;
3908 NAPI_GRO_CB(skb)->flush = 0;
3909 NAPI_GRO_CB(skb)->free = 0;
3910 NAPI_GRO_CB(skb)->udp_mark = 0;
3912 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3917 if (&ptype->list == head)
3920 same_flow = NAPI_GRO_CB(skb)->same_flow;
3921 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3924 struct sk_buff *nskb = *pp;
3928 napi_gro_complete(nskb);
3935 if (NAPI_GRO_CB(skb)->flush)
3938 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3939 struct sk_buff *nskb = napi->gro_list;
3941 /* locate the end of the list to select the 'oldest' flow */
3942 while (nskb->next) {
3948 napi_gro_complete(nskb);
3952 NAPI_GRO_CB(skb)->count = 1;
3953 NAPI_GRO_CB(skb)->age = jiffies;
3954 NAPI_GRO_CB(skb)->last = skb;
3955 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3956 skb->next = napi->gro_list;
3957 napi->gro_list = skb;
3961 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3962 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3964 BUG_ON(skb->end - skb->tail < grow);
3966 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3969 skb->data_len -= grow;
3971 skb_shinfo(skb)->frags[0].page_offset += grow;
3972 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3974 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3975 skb_frag_unref(skb, 0);
3976 memmove(skb_shinfo(skb)->frags,
3977 skb_shinfo(skb)->frags + 1,
3978 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3990 struct packet_offload *gro_find_receive_by_type(__be16 type)
3992 struct list_head *offload_head = &offload_base;
3993 struct packet_offload *ptype;
3995 list_for_each_entry_rcu(ptype, offload_head, list) {
3996 if (ptype->type != type || !ptype->callbacks.gro_receive)
4002 EXPORT_SYMBOL(gro_find_receive_by_type);
4004 struct packet_offload *gro_find_complete_by_type(__be16 type)
4006 struct list_head *offload_head = &offload_base;
4007 struct packet_offload *ptype;
4009 list_for_each_entry_rcu(ptype, offload_head, list) {
4010 if (ptype->type != type || !ptype->callbacks.gro_complete)
4016 EXPORT_SYMBOL(gro_find_complete_by_type);
4018 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4022 if (netif_receive_skb_internal(skb))
4030 case GRO_MERGED_FREE:
4031 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4032 kmem_cache_free(skbuff_head_cache, skb);
4045 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4047 trace_napi_gro_receive_entry(skb);
4049 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4051 EXPORT_SYMBOL(napi_gro_receive);
4053 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4055 __skb_pull(skb, skb_headlen(skb));
4056 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4057 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4059 skb->dev = napi->dev;
4061 skb->encapsulation = 0;
4062 skb_shinfo(skb)->gso_type = 0;
4063 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4068 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4070 struct sk_buff *skb = napi->skb;
4073 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4078 EXPORT_SYMBOL(napi_get_frags);
4080 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4085 if (netif_receive_skb_internal(skb))
4090 case GRO_MERGED_FREE:
4091 napi_reuse_skb(napi, skb);
4102 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4104 struct sk_buff *skb = napi->skb;
4108 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4109 napi_reuse_skb(napi, skb);
4112 skb->protocol = eth_type_trans(skb, skb->dev);
4117 gro_result_t napi_gro_frags(struct napi_struct *napi)
4119 struct sk_buff *skb = napi_frags_skb(napi);
4124 trace_napi_gro_frags_entry(skb);
4126 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4128 EXPORT_SYMBOL(napi_gro_frags);
4131 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4132 * Note: called with local irq disabled, but exits with local irq enabled.
4134 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4137 struct softnet_data *remsd = sd->rps_ipi_list;
4140 sd->rps_ipi_list = NULL;
4144 /* Send pending IPI's to kick RPS processing on remote cpus. */
4146 struct softnet_data *next = remsd->rps_ipi_next;
4148 if (cpu_online(remsd->cpu))
4149 __smp_call_function_single(remsd->cpu,
4158 static int process_backlog(struct napi_struct *napi, int quota)
4161 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4164 /* Check if we have pending ipi, its better to send them now,
4165 * not waiting net_rx_action() end.
4167 if (sd->rps_ipi_list) {
4168 local_irq_disable();
4169 net_rps_action_and_irq_enable(sd);
4172 napi->weight = weight_p;
4173 local_irq_disable();
4174 while (work < quota) {
4175 struct sk_buff *skb;
4178 while ((skb = __skb_dequeue(&sd->process_queue))) {
4180 __netif_receive_skb(skb);
4181 local_irq_disable();
4182 input_queue_head_incr(sd);
4183 if (++work >= quota) {
4190 qlen = skb_queue_len(&sd->input_pkt_queue);
4192 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4193 &sd->process_queue);
4195 if (qlen < quota - work) {
4197 * Inline a custom version of __napi_complete().
4198 * only current cpu owns and manipulates this napi,
4199 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4200 * we can use a plain write instead of clear_bit(),
4201 * and we dont need an smp_mb() memory barrier.
4203 list_del(&napi->poll_list);
4206 quota = work + qlen;
4216 * __napi_schedule - schedule for receive
4217 * @n: entry to schedule
4219 * The entry's receive function will be scheduled to run
4221 void __napi_schedule(struct napi_struct *n)
4223 unsigned long flags;
4225 local_irq_save(flags);
4226 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4227 local_irq_restore(flags);
4229 EXPORT_SYMBOL(__napi_schedule);
4231 void __napi_complete(struct napi_struct *n)
4233 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4234 BUG_ON(n->gro_list);
4236 list_del(&n->poll_list);
4237 smp_mb__before_clear_bit();
4238 clear_bit(NAPI_STATE_SCHED, &n->state);
4240 EXPORT_SYMBOL(__napi_complete);
4242 void napi_complete(struct napi_struct *n)
4244 unsigned long flags;
4247 * don't let napi dequeue from the cpu poll list
4248 * just in case its running on a different cpu
4250 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4253 napi_gro_flush(n, false);
4254 local_irq_save(flags);
4256 local_irq_restore(flags);
4258 EXPORT_SYMBOL(napi_complete);
4260 /* must be called under rcu_read_lock(), as we dont take a reference */
4261 struct napi_struct *napi_by_id(unsigned int napi_id)
4263 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4264 struct napi_struct *napi;
4266 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4267 if (napi->napi_id == napi_id)
4272 EXPORT_SYMBOL_GPL(napi_by_id);
4274 void napi_hash_add(struct napi_struct *napi)
4276 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4278 spin_lock(&napi_hash_lock);
4280 /* 0 is not a valid id, we also skip an id that is taken
4281 * we expect both events to be extremely rare
4284 while (!napi->napi_id) {
4285 napi->napi_id = ++napi_gen_id;
4286 if (napi_by_id(napi->napi_id))
4290 hlist_add_head_rcu(&napi->napi_hash_node,
4291 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4293 spin_unlock(&napi_hash_lock);
4296 EXPORT_SYMBOL_GPL(napi_hash_add);
4298 /* Warning : caller is responsible to make sure rcu grace period
4299 * is respected before freeing memory containing @napi
4301 void napi_hash_del(struct napi_struct *napi)
4303 spin_lock(&napi_hash_lock);
4305 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4306 hlist_del_rcu(&napi->napi_hash_node);
4308 spin_unlock(&napi_hash_lock);
4310 EXPORT_SYMBOL_GPL(napi_hash_del);
4312 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4313 int (*poll)(struct napi_struct *, int), int weight)
4315 INIT_LIST_HEAD(&napi->poll_list);
4316 napi->gro_count = 0;
4317 napi->gro_list = NULL;
4320 if (weight > NAPI_POLL_WEIGHT)
4321 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4323 napi->weight = weight;
4324 list_add(&napi->dev_list, &dev->napi_list);
4326 #ifdef CONFIG_NETPOLL
4327 spin_lock_init(&napi->poll_lock);
4328 napi->poll_owner = -1;
4330 set_bit(NAPI_STATE_SCHED, &napi->state);
4332 EXPORT_SYMBOL(netif_napi_add);
4334 void netif_napi_del(struct napi_struct *napi)
4336 list_del_init(&napi->dev_list);
4337 napi_free_frags(napi);
4339 kfree_skb_list(napi->gro_list);
4340 napi->gro_list = NULL;
4341 napi->gro_count = 0;
4343 EXPORT_SYMBOL(netif_napi_del);
4345 static void net_rx_action(struct softirq_action *h)
4347 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4348 unsigned long time_limit = jiffies + 2;
4349 int budget = netdev_budget;
4352 local_irq_disable();
4354 while (!list_empty(&sd->poll_list)) {
4355 struct napi_struct *n;
4358 /* If softirq window is exhuasted then punt.
4359 * Allow this to run for 2 jiffies since which will allow
4360 * an average latency of 1.5/HZ.
4362 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4367 /* Even though interrupts have been re-enabled, this
4368 * access is safe because interrupts can only add new
4369 * entries to the tail of this list, and only ->poll()
4370 * calls can remove this head entry from the list.
4372 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4374 have = netpoll_poll_lock(n);
4378 /* This NAPI_STATE_SCHED test is for avoiding a race
4379 * with netpoll's poll_napi(). Only the entity which
4380 * obtains the lock and sees NAPI_STATE_SCHED set will
4381 * actually make the ->poll() call. Therefore we avoid
4382 * accidentally calling ->poll() when NAPI is not scheduled.
4385 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4386 work = n->poll(n, weight);
4390 WARN_ON_ONCE(work > weight);
4394 local_irq_disable();
4396 /* Drivers must not modify the NAPI state if they
4397 * consume the entire weight. In such cases this code
4398 * still "owns" the NAPI instance and therefore can
4399 * move the instance around on the list at-will.
4401 if (unlikely(work == weight)) {
4402 if (unlikely(napi_disable_pending(n))) {
4405 local_irq_disable();
4408 /* flush too old packets
4409 * If HZ < 1000, flush all packets.
4412 napi_gro_flush(n, HZ >= 1000);
4413 local_irq_disable();
4415 list_move_tail(&n->poll_list, &sd->poll_list);
4419 netpoll_poll_unlock(have);
4422 net_rps_action_and_irq_enable(sd);
4424 #ifdef CONFIG_NET_DMA
4426 * There may not be any more sk_buffs coming right now, so push
4427 * any pending DMA copies to hardware
4429 dma_issue_pending_all();
4436 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4440 struct netdev_adjacent {
4441 struct net_device *dev;
4443 /* upper master flag, there can only be one master device per list */
4446 /* counter for the number of times this device was added to us */
4449 /* private field for the users */
4452 struct list_head list;
4453 struct rcu_head rcu;
4456 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4457 struct net_device *adj_dev,
4458 struct list_head *adj_list)
4460 struct netdev_adjacent *adj;
4462 list_for_each_entry(adj, adj_list, list) {
4463 if (adj->dev == adj_dev)
4470 * netdev_has_upper_dev - Check if device is linked to an upper device
4472 * @upper_dev: upper device to check
4474 * Find out if a device is linked to specified upper device and return true
4475 * in case it is. Note that this checks only immediate upper device,
4476 * not through a complete stack of devices. The caller must hold the RTNL lock.
4478 bool netdev_has_upper_dev(struct net_device *dev,
4479 struct net_device *upper_dev)
4483 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4485 EXPORT_SYMBOL(netdev_has_upper_dev);
4488 * netdev_has_any_upper_dev - Check if device is linked to some device
4491 * Find out if a device is linked to an upper device and return true in case
4492 * it is. The caller must hold the RTNL lock.
4494 static bool netdev_has_any_upper_dev(struct net_device *dev)
4498 return !list_empty(&dev->all_adj_list.upper);
4502 * netdev_master_upper_dev_get - Get master upper device
4505 * Find a master upper device and return pointer to it or NULL in case
4506 * it's not there. The caller must hold the RTNL lock.
4508 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4510 struct netdev_adjacent *upper;
4514 if (list_empty(&dev->adj_list.upper))
4517 upper = list_first_entry(&dev->adj_list.upper,
4518 struct netdev_adjacent, list);
4519 if (likely(upper->master))
4523 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4525 void *netdev_adjacent_get_private(struct list_head *adj_list)
4527 struct netdev_adjacent *adj;
4529 adj = list_entry(adj_list, struct netdev_adjacent, list);
4531 return adj->private;
4533 EXPORT_SYMBOL(netdev_adjacent_get_private);
4536 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4538 * @iter: list_head ** of the current position
4540 * Gets the next device from the dev's upper list, starting from iter
4541 * position. The caller must hold RCU read lock.
4543 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4544 struct list_head **iter)
4546 struct netdev_adjacent *upper;
4548 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4550 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4552 if (&upper->list == &dev->all_adj_list.upper)
4555 *iter = &upper->list;
4559 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4562 * netdev_lower_get_next_private - Get the next ->private from the
4563 * lower neighbour list
4565 * @iter: list_head ** of the current position
4567 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4568 * list, starting from iter position. The caller must hold either hold the
4569 * RTNL lock or its own locking that guarantees that the neighbour lower
4570 * list will remain unchainged.
4572 void *netdev_lower_get_next_private(struct net_device *dev,
4573 struct list_head **iter)
4575 struct netdev_adjacent *lower;
4577 lower = list_entry(*iter, struct netdev_adjacent, list);
4579 if (&lower->list == &dev->adj_list.lower)
4583 *iter = lower->list.next;
4585 return lower->private;
4587 EXPORT_SYMBOL(netdev_lower_get_next_private);
4590 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4591 * lower neighbour list, RCU
4594 * @iter: list_head ** of the current position
4596 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4597 * list, starting from iter position. The caller must hold RCU read lock.
4599 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4600 struct list_head **iter)
4602 struct netdev_adjacent *lower;
4604 WARN_ON_ONCE(!rcu_read_lock_held());
4606 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4608 if (&lower->list == &dev->adj_list.lower)
4612 *iter = &lower->list;
4614 return lower->private;
4616 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4619 * netdev_lower_get_next - Get the next device from the lower neighbour
4622 * @iter: list_head ** of the current position
4624 * Gets the next netdev_adjacent from the dev's lower neighbour
4625 * list, starting from iter position. The caller must hold RTNL lock or
4626 * its own locking that guarantees that the neighbour lower
4627 * list will remain unchainged.
4629 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4631 struct netdev_adjacent *lower;
4633 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4635 if (&lower->list == &dev->adj_list.lower)
4638 *iter = &lower->list;
4642 EXPORT_SYMBOL(netdev_lower_get_next);
4645 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4646 * lower neighbour list, RCU
4650 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4651 * list. The caller must hold RCU read lock.
4653 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4655 struct netdev_adjacent *lower;
4657 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4658 struct netdev_adjacent, list);
4660 return lower->private;
4663 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4666 * netdev_master_upper_dev_get_rcu - Get master upper device
4669 * Find a master upper device and return pointer to it or NULL in case
4670 * it's not there. The caller must hold the RCU read lock.
4672 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4674 struct netdev_adjacent *upper;
4676 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4677 struct netdev_adjacent, list);
4678 if (upper && likely(upper->master))
4682 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4684 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4685 struct net_device *adj_dev,
4686 struct list_head *dev_list)
4688 char linkname[IFNAMSIZ+7];
4689 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4690 "upper_%s" : "lower_%s", adj_dev->name);
4691 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4694 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4696 struct list_head *dev_list)
4698 char linkname[IFNAMSIZ+7];
4699 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4700 "upper_%s" : "lower_%s", name);
4701 sysfs_remove_link(&(dev->dev.kobj), linkname);
4704 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4705 (dev_list == &dev->adj_list.upper || \
4706 dev_list == &dev->adj_list.lower)
4708 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4709 struct net_device *adj_dev,
4710 struct list_head *dev_list,
4711 void *private, bool master)
4713 struct netdev_adjacent *adj;
4716 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4723 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4728 adj->master = master;
4730 adj->private = private;
4733 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4734 adj_dev->name, dev->name, adj_dev->name);
4736 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4737 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4742 /* Ensure that master link is always the first item in list. */
4744 ret = sysfs_create_link(&(dev->dev.kobj),
4745 &(adj_dev->dev.kobj), "master");
4747 goto remove_symlinks;
4749 list_add_rcu(&adj->list, dev_list);
4751 list_add_tail_rcu(&adj->list, dev_list);
4757 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4758 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4766 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4767 struct net_device *adj_dev,
4768 struct list_head *dev_list)
4770 struct netdev_adjacent *adj;
4772 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4775 pr_err("tried to remove device %s from %s\n",
4776 dev->name, adj_dev->name);
4780 if (adj->ref_nr > 1) {
4781 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4788 sysfs_remove_link(&(dev->dev.kobj), "master");
4790 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4791 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4793 list_del_rcu(&adj->list);
4794 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4795 adj_dev->name, dev->name, adj_dev->name);
4797 kfree_rcu(adj, rcu);
4800 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4801 struct net_device *upper_dev,
4802 struct list_head *up_list,
4803 struct list_head *down_list,
4804 void *private, bool master)
4808 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4813 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4816 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4823 static int __netdev_adjacent_dev_link(struct net_device *dev,
4824 struct net_device *upper_dev)
4826 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4827 &dev->all_adj_list.upper,
4828 &upper_dev->all_adj_list.lower,
4832 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4833 struct net_device *upper_dev,
4834 struct list_head *up_list,
4835 struct list_head *down_list)
4837 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4838 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4841 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4842 struct net_device *upper_dev)
4844 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4845 &dev->all_adj_list.upper,
4846 &upper_dev->all_adj_list.lower);
4849 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4850 struct net_device *upper_dev,
4851 void *private, bool master)
4853 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4858 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4859 &dev->adj_list.upper,
4860 &upper_dev->adj_list.lower,
4863 __netdev_adjacent_dev_unlink(dev, upper_dev);
4870 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4871 struct net_device *upper_dev)
4873 __netdev_adjacent_dev_unlink(dev, upper_dev);
4874 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4875 &dev->adj_list.upper,
4876 &upper_dev->adj_list.lower);
4879 static int __netdev_upper_dev_link(struct net_device *dev,
4880 struct net_device *upper_dev, bool master,
4883 struct netdev_adjacent *i, *j, *to_i, *to_j;
4888 if (dev == upper_dev)
4891 /* To prevent loops, check if dev is not upper device to upper_dev. */
4892 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4895 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4898 if (master && netdev_master_upper_dev_get(dev))
4901 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4906 /* Now that we linked these devs, make all the upper_dev's
4907 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4908 * versa, and don't forget the devices itself. All of these
4909 * links are non-neighbours.
4911 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4912 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4913 pr_debug("Interlinking %s with %s, non-neighbour\n",
4914 i->dev->name, j->dev->name);
4915 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4921 /* add dev to every upper_dev's upper device */
4922 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4923 pr_debug("linking %s's upper device %s with %s\n",
4924 upper_dev->name, i->dev->name, dev->name);
4925 ret = __netdev_adjacent_dev_link(dev, i->dev);
4927 goto rollback_upper_mesh;
4930 /* add upper_dev to every dev's lower device */
4931 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4932 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4933 i->dev->name, upper_dev->name);
4934 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4936 goto rollback_lower_mesh;
4939 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4942 rollback_lower_mesh:
4944 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4947 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4952 rollback_upper_mesh:
4954 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4957 __netdev_adjacent_dev_unlink(dev, i->dev);
4965 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4966 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4967 if (i == to_i && j == to_j)
4969 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4975 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4981 * netdev_upper_dev_link - Add a link to the upper device
4983 * @upper_dev: new upper device
4985 * Adds a link to device which is upper to this one. The caller must hold
4986 * the RTNL lock. On a failure a negative errno code is returned.
4987 * On success the reference counts are adjusted and the function
4990 int netdev_upper_dev_link(struct net_device *dev,
4991 struct net_device *upper_dev)
4993 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4995 EXPORT_SYMBOL(netdev_upper_dev_link);
4998 * netdev_master_upper_dev_link - Add a master link to the upper device
5000 * @upper_dev: new upper device
5002 * Adds a link to device which is upper to this one. In this case, only
5003 * one master upper device can be linked, although other non-master devices
5004 * might be linked as well. The caller must hold the RTNL lock.
5005 * On a failure a negative errno code is returned. On success the reference
5006 * counts are adjusted and the function returns zero.
5008 int netdev_master_upper_dev_link(struct net_device *dev,
5009 struct net_device *upper_dev)
5011 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5013 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5015 int netdev_master_upper_dev_link_private(struct net_device *dev,
5016 struct net_device *upper_dev,
5019 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5021 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5024 * netdev_upper_dev_unlink - Removes a link to upper device
5026 * @upper_dev: new upper device
5028 * Removes a link to device which is upper to this one. The caller must hold
5031 void netdev_upper_dev_unlink(struct net_device *dev,
5032 struct net_device *upper_dev)
5034 struct netdev_adjacent *i, *j;
5037 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5039 /* Here is the tricky part. We must remove all dev's lower
5040 * devices from all upper_dev's upper devices and vice
5041 * versa, to maintain the graph relationship.
5043 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5044 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5045 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5047 /* remove also the devices itself from lower/upper device
5050 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5051 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5053 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5054 __netdev_adjacent_dev_unlink(dev, i->dev);
5056 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5058 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5060 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5062 struct netdev_adjacent *iter;
5064 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5065 netdev_adjacent_sysfs_del(iter->dev, oldname,
5066 &iter->dev->adj_list.lower);
5067 netdev_adjacent_sysfs_add(iter->dev, dev,
5068 &iter->dev->adj_list.lower);
5071 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5072 netdev_adjacent_sysfs_del(iter->dev, oldname,
5073 &iter->dev->adj_list.upper);
5074 netdev_adjacent_sysfs_add(iter->dev, dev,
5075 &iter->dev->adj_list.upper);
5079 void *netdev_lower_dev_get_private(struct net_device *dev,
5080 struct net_device *lower_dev)
5082 struct netdev_adjacent *lower;
5086 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5090 return lower->private;
5092 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5095 int dev_get_nest_level(struct net_device *dev,
5096 bool (*type_check)(struct net_device *dev))
5098 struct net_device *lower = NULL;
5099 struct list_head *iter;
5105 netdev_for_each_lower_dev(dev, lower, iter) {
5106 nest = dev_get_nest_level(lower, type_check);
5107 if (max_nest < nest)
5111 if (type_check(dev))
5116 EXPORT_SYMBOL(dev_get_nest_level);
5118 static void dev_change_rx_flags(struct net_device *dev, int flags)
5120 const struct net_device_ops *ops = dev->netdev_ops;
5122 if (ops->ndo_change_rx_flags)
5123 ops->ndo_change_rx_flags(dev, flags);
5126 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5128 unsigned int old_flags = dev->flags;
5134 dev->flags |= IFF_PROMISC;
5135 dev->promiscuity += inc;
5136 if (dev->promiscuity == 0) {
5139 * If inc causes overflow, untouch promisc and return error.
5142 dev->flags &= ~IFF_PROMISC;
5144 dev->promiscuity -= inc;
5145 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5150 if (dev->flags != old_flags) {
5151 pr_info("device %s %s promiscuous mode\n",
5153 dev->flags & IFF_PROMISC ? "entered" : "left");
5154 if (audit_enabled) {
5155 current_uid_gid(&uid, &gid);
5156 audit_log(current->audit_context, GFP_ATOMIC,
5157 AUDIT_ANOM_PROMISCUOUS,
5158 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5159 dev->name, (dev->flags & IFF_PROMISC),
5160 (old_flags & IFF_PROMISC),
5161 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5162 from_kuid(&init_user_ns, uid),
5163 from_kgid(&init_user_ns, gid),
5164 audit_get_sessionid(current));
5167 dev_change_rx_flags(dev, IFF_PROMISC);
5170 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5175 * dev_set_promiscuity - update promiscuity count on a device
5179 * Add or remove promiscuity from a device. While the count in the device
5180 * remains above zero the interface remains promiscuous. Once it hits zero
5181 * the device reverts back to normal filtering operation. A negative inc
5182 * value is used to drop promiscuity on the device.
5183 * Return 0 if successful or a negative errno code on error.
5185 int dev_set_promiscuity(struct net_device *dev, int inc)
5187 unsigned int old_flags = dev->flags;
5190 err = __dev_set_promiscuity(dev, inc, true);
5193 if (dev->flags != old_flags)
5194 dev_set_rx_mode(dev);
5197 EXPORT_SYMBOL(dev_set_promiscuity);
5199 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5201 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5205 dev->flags |= IFF_ALLMULTI;
5206 dev->allmulti += inc;
5207 if (dev->allmulti == 0) {
5210 * If inc causes overflow, untouch allmulti and return error.
5213 dev->flags &= ~IFF_ALLMULTI;
5215 dev->allmulti -= inc;
5216 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5221 if (dev->flags ^ old_flags) {
5222 dev_change_rx_flags(dev, IFF_ALLMULTI);
5223 dev_set_rx_mode(dev);
5225 __dev_notify_flags(dev, old_flags,
5226 dev->gflags ^ old_gflags);
5232 * dev_set_allmulti - update allmulti count on a device
5236 * Add or remove reception of all multicast frames to a device. While the
5237 * count in the device remains above zero the interface remains listening
5238 * to all interfaces. Once it hits zero the device reverts back to normal
5239 * filtering operation. A negative @inc value is used to drop the counter
5240 * when releasing a resource needing all multicasts.
5241 * Return 0 if successful or a negative errno code on error.
5244 int dev_set_allmulti(struct net_device *dev, int inc)
5246 return __dev_set_allmulti(dev, inc, true);
5248 EXPORT_SYMBOL(dev_set_allmulti);
5251 * Upload unicast and multicast address lists to device and
5252 * configure RX filtering. When the device doesn't support unicast
5253 * filtering it is put in promiscuous mode while unicast addresses
5256 void __dev_set_rx_mode(struct net_device *dev)
5258 const struct net_device_ops *ops = dev->netdev_ops;
5260 /* dev_open will call this function so the list will stay sane. */
5261 if (!(dev->flags&IFF_UP))
5264 if (!netif_device_present(dev))
5267 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5268 /* Unicast addresses changes may only happen under the rtnl,
5269 * therefore calling __dev_set_promiscuity here is safe.
5271 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5272 __dev_set_promiscuity(dev, 1, false);
5273 dev->uc_promisc = true;
5274 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5275 __dev_set_promiscuity(dev, -1, false);
5276 dev->uc_promisc = false;
5280 if (ops->ndo_set_rx_mode)
5281 ops->ndo_set_rx_mode(dev);
5284 void dev_set_rx_mode(struct net_device *dev)
5286 netif_addr_lock_bh(dev);
5287 __dev_set_rx_mode(dev);
5288 netif_addr_unlock_bh(dev);
5292 * dev_get_flags - get flags reported to userspace
5295 * Get the combination of flag bits exported through APIs to userspace.
5297 unsigned int dev_get_flags(const struct net_device *dev)
5301 flags = (dev->flags & ~(IFF_PROMISC |
5306 (dev->gflags & (IFF_PROMISC |
5309 if (netif_running(dev)) {
5310 if (netif_oper_up(dev))
5311 flags |= IFF_RUNNING;
5312 if (netif_carrier_ok(dev))
5313 flags |= IFF_LOWER_UP;
5314 if (netif_dormant(dev))
5315 flags |= IFF_DORMANT;
5320 EXPORT_SYMBOL(dev_get_flags);
5322 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5324 unsigned int old_flags = dev->flags;
5330 * Set the flags on our device.
5333 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5334 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5336 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5340 * Load in the correct multicast list now the flags have changed.
5343 if ((old_flags ^ flags) & IFF_MULTICAST)
5344 dev_change_rx_flags(dev, IFF_MULTICAST);
5346 dev_set_rx_mode(dev);
5349 * Have we downed the interface. We handle IFF_UP ourselves
5350 * according to user attempts to set it, rather than blindly
5355 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5356 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5359 dev_set_rx_mode(dev);
5362 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5363 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5364 unsigned int old_flags = dev->flags;
5366 dev->gflags ^= IFF_PROMISC;
5368 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5369 if (dev->flags != old_flags)
5370 dev_set_rx_mode(dev);
5373 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5374 is important. Some (broken) drivers set IFF_PROMISC, when
5375 IFF_ALLMULTI is requested not asking us and not reporting.
5377 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5378 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5380 dev->gflags ^= IFF_ALLMULTI;
5381 __dev_set_allmulti(dev, inc, false);
5387 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5388 unsigned int gchanges)
5390 unsigned int changes = dev->flags ^ old_flags;
5393 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5395 if (changes & IFF_UP) {
5396 if (dev->flags & IFF_UP)
5397 call_netdevice_notifiers(NETDEV_UP, dev);
5399 call_netdevice_notifiers(NETDEV_DOWN, dev);
5402 if (dev->flags & IFF_UP &&
5403 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5404 struct netdev_notifier_change_info change_info;
5406 change_info.flags_changed = changes;
5407 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5413 * dev_change_flags - change device settings
5415 * @flags: device state flags
5417 * Change settings on device based state flags. The flags are
5418 * in the userspace exported format.
5420 int dev_change_flags(struct net_device *dev, unsigned int flags)
5423 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5425 ret = __dev_change_flags(dev, flags);
5429 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5430 __dev_notify_flags(dev, old_flags, changes);
5433 EXPORT_SYMBOL(dev_change_flags);
5435 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5437 const struct net_device_ops *ops = dev->netdev_ops;
5439 if (ops->ndo_change_mtu)
5440 return ops->ndo_change_mtu(dev, new_mtu);
5447 * dev_set_mtu - Change maximum transfer unit
5449 * @new_mtu: new transfer unit
5451 * Change the maximum transfer size of the network device.
5453 int dev_set_mtu(struct net_device *dev, int new_mtu)
5457 if (new_mtu == dev->mtu)
5460 /* MTU must be positive. */
5464 if (!netif_device_present(dev))
5467 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5468 err = notifier_to_errno(err);
5472 orig_mtu = dev->mtu;
5473 err = __dev_set_mtu(dev, new_mtu);
5476 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5477 err = notifier_to_errno(err);
5479 /* setting mtu back and notifying everyone again,
5480 * so that they have a chance to revert changes.
5482 __dev_set_mtu(dev, orig_mtu);
5483 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5488 EXPORT_SYMBOL(dev_set_mtu);
5491 * dev_set_group - Change group this device belongs to
5493 * @new_group: group this device should belong to
5495 void dev_set_group(struct net_device *dev, int new_group)
5497 dev->group = new_group;
5499 EXPORT_SYMBOL(dev_set_group);
5502 * dev_set_mac_address - Change Media Access Control Address
5506 * Change the hardware (MAC) address of the device
5508 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5510 const struct net_device_ops *ops = dev->netdev_ops;
5513 if (!ops->ndo_set_mac_address)
5515 if (sa->sa_family != dev->type)
5517 if (!netif_device_present(dev))
5519 err = ops->ndo_set_mac_address(dev, sa);
5522 dev->addr_assign_type = NET_ADDR_SET;
5523 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5524 add_device_randomness(dev->dev_addr, dev->addr_len);
5527 EXPORT_SYMBOL(dev_set_mac_address);
5530 * dev_change_carrier - Change device carrier
5532 * @new_carrier: new value
5534 * Change device carrier
5536 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5538 const struct net_device_ops *ops = dev->netdev_ops;
5540 if (!ops->ndo_change_carrier)
5542 if (!netif_device_present(dev))
5544 return ops->ndo_change_carrier(dev, new_carrier);
5546 EXPORT_SYMBOL(dev_change_carrier);
5549 * dev_get_phys_port_id - Get device physical port ID
5553 * Get device physical port ID
5555 int dev_get_phys_port_id(struct net_device *dev,
5556 struct netdev_phys_port_id *ppid)
5558 const struct net_device_ops *ops = dev->netdev_ops;
5560 if (!ops->ndo_get_phys_port_id)
5562 return ops->ndo_get_phys_port_id(dev, ppid);
5564 EXPORT_SYMBOL(dev_get_phys_port_id);
5567 * dev_new_index - allocate an ifindex
5568 * @net: the applicable net namespace
5570 * Returns a suitable unique value for a new device interface
5571 * number. The caller must hold the rtnl semaphore or the
5572 * dev_base_lock to be sure it remains unique.
5574 static int dev_new_index(struct net *net)
5576 int ifindex = net->ifindex;
5580 if (!__dev_get_by_index(net, ifindex))
5581 return net->ifindex = ifindex;
5585 /* Delayed registration/unregisteration */
5586 static LIST_HEAD(net_todo_list);
5587 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5589 static void net_set_todo(struct net_device *dev)
5591 list_add_tail(&dev->todo_list, &net_todo_list);
5592 dev_net(dev)->dev_unreg_count++;
5595 static void rollback_registered_many(struct list_head *head)
5597 struct net_device *dev, *tmp;
5598 LIST_HEAD(close_head);
5600 BUG_ON(dev_boot_phase);
5603 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5604 /* Some devices call without registering
5605 * for initialization unwind. Remove those
5606 * devices and proceed with the remaining.
5608 if (dev->reg_state == NETREG_UNINITIALIZED) {
5609 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5613 list_del(&dev->unreg_list);
5616 dev->dismantle = true;
5617 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5620 /* If device is running, close it first. */
5621 list_for_each_entry(dev, head, unreg_list)
5622 list_add_tail(&dev->close_list, &close_head);
5623 dev_close_many(&close_head);
5625 list_for_each_entry(dev, head, unreg_list) {
5626 /* And unlink it from device chain. */
5627 unlist_netdevice(dev);
5629 dev->reg_state = NETREG_UNREGISTERING;
5634 list_for_each_entry(dev, head, unreg_list) {
5635 /* Shutdown queueing discipline. */
5639 /* Notify protocols, that we are about to destroy
5640 this device. They should clean all the things.
5642 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5644 if (!dev->rtnl_link_ops ||
5645 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5646 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5649 * Flush the unicast and multicast chains
5654 if (dev->netdev_ops->ndo_uninit)
5655 dev->netdev_ops->ndo_uninit(dev);
5657 /* Notifier chain MUST detach us all upper devices. */
5658 WARN_ON(netdev_has_any_upper_dev(dev));
5660 /* Remove entries from kobject tree */
5661 netdev_unregister_kobject(dev);
5663 /* Remove XPS queueing entries */
5664 netif_reset_xps_queues_gt(dev, 0);
5670 list_for_each_entry(dev, head, unreg_list)
5674 static void rollback_registered(struct net_device *dev)
5678 list_add(&dev->unreg_list, &single);
5679 rollback_registered_many(&single);
5683 static netdev_features_t netdev_fix_features(struct net_device *dev,
5684 netdev_features_t features)
5686 /* Fix illegal checksum combinations */
5687 if ((features & NETIF_F_HW_CSUM) &&
5688 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5689 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5690 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5693 /* TSO requires that SG is present as well. */
5694 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5695 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5696 features &= ~NETIF_F_ALL_TSO;
5699 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5700 !(features & NETIF_F_IP_CSUM)) {
5701 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5702 features &= ~NETIF_F_TSO;
5703 features &= ~NETIF_F_TSO_ECN;
5706 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5707 !(features & NETIF_F_IPV6_CSUM)) {
5708 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5709 features &= ~NETIF_F_TSO6;
5712 /* TSO ECN requires that TSO is present as well. */
5713 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5714 features &= ~NETIF_F_TSO_ECN;
5716 /* Software GSO depends on SG. */
5717 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5718 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5719 features &= ~NETIF_F_GSO;
5722 /* UFO needs SG and checksumming */
5723 if (features & NETIF_F_UFO) {
5724 /* maybe split UFO into V4 and V6? */
5725 if (!((features & NETIF_F_GEN_CSUM) ||
5726 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5727 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5729 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5730 features &= ~NETIF_F_UFO;
5733 if (!(features & NETIF_F_SG)) {
5735 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5736 features &= ~NETIF_F_UFO;
5743 int __netdev_update_features(struct net_device *dev)
5745 netdev_features_t features;
5750 features = netdev_get_wanted_features(dev);
5752 if (dev->netdev_ops->ndo_fix_features)
5753 features = dev->netdev_ops->ndo_fix_features(dev, features);
5755 /* driver might be less strict about feature dependencies */
5756 features = netdev_fix_features(dev, features);
5758 if (dev->features == features)
5761 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5762 &dev->features, &features);
5764 if (dev->netdev_ops->ndo_set_features)
5765 err = dev->netdev_ops->ndo_set_features(dev, features);
5767 if (unlikely(err < 0)) {
5769 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5770 err, &features, &dev->features);
5775 dev->features = features;
5781 * netdev_update_features - recalculate device features
5782 * @dev: the device to check
5784 * Recalculate dev->features set and send notifications if it
5785 * has changed. Should be called after driver or hardware dependent
5786 * conditions might have changed that influence the features.
5788 void netdev_update_features(struct net_device *dev)
5790 if (__netdev_update_features(dev))
5791 netdev_features_change(dev);
5793 EXPORT_SYMBOL(netdev_update_features);
5796 * netdev_change_features - recalculate device features
5797 * @dev: the device to check
5799 * Recalculate dev->features set and send notifications even
5800 * if they have not changed. Should be called instead of
5801 * netdev_update_features() if also dev->vlan_features might
5802 * have changed to allow the changes to be propagated to stacked
5805 void netdev_change_features(struct net_device *dev)
5807 __netdev_update_features(dev);
5808 netdev_features_change(dev);
5810 EXPORT_SYMBOL(netdev_change_features);
5813 * netif_stacked_transfer_operstate - transfer operstate
5814 * @rootdev: the root or lower level device to transfer state from
5815 * @dev: the device to transfer operstate to
5817 * Transfer operational state from root to device. This is normally
5818 * called when a stacking relationship exists between the root
5819 * device and the device(a leaf device).
5821 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5822 struct net_device *dev)
5824 if (rootdev->operstate == IF_OPER_DORMANT)
5825 netif_dormant_on(dev);
5827 netif_dormant_off(dev);
5829 if (netif_carrier_ok(rootdev)) {
5830 if (!netif_carrier_ok(dev))
5831 netif_carrier_on(dev);
5833 if (netif_carrier_ok(dev))
5834 netif_carrier_off(dev);
5837 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5840 static int netif_alloc_rx_queues(struct net_device *dev)
5842 unsigned int i, count = dev->num_rx_queues;
5843 struct netdev_rx_queue *rx;
5847 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5853 for (i = 0; i < count; i++)
5859 static void netdev_init_one_queue(struct net_device *dev,
5860 struct netdev_queue *queue, void *_unused)
5862 /* Initialize queue lock */
5863 spin_lock_init(&queue->_xmit_lock);
5864 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5865 queue->xmit_lock_owner = -1;
5866 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5869 dql_init(&queue->dql, HZ);
5873 static void netif_free_tx_queues(struct net_device *dev)
5875 if (is_vmalloc_addr(dev->_tx))
5881 static int netif_alloc_netdev_queues(struct net_device *dev)
5883 unsigned int count = dev->num_tx_queues;
5884 struct netdev_queue *tx;
5885 size_t sz = count * sizeof(*tx);
5887 BUG_ON(count < 1 || count > 0xffff);
5889 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5897 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5898 spin_lock_init(&dev->tx_global_lock);
5904 * register_netdevice - register a network device
5905 * @dev: device to register
5907 * Take a completed network device structure and add it to the kernel
5908 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5909 * chain. 0 is returned on success. A negative errno code is returned
5910 * on a failure to set up the device, or if the name is a duplicate.
5912 * Callers must hold the rtnl semaphore. You may want
5913 * register_netdev() instead of this.
5916 * The locking appears insufficient to guarantee two parallel registers
5917 * will not get the same name.
5920 int register_netdevice(struct net_device *dev)
5923 struct net *net = dev_net(dev);
5925 BUG_ON(dev_boot_phase);
5930 /* When net_device's are persistent, this will be fatal. */
5931 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5934 spin_lock_init(&dev->addr_list_lock);
5935 netdev_set_addr_lockdep_class(dev);
5939 ret = dev_get_valid_name(net, dev, dev->name);
5943 /* Init, if this function is available */
5944 if (dev->netdev_ops->ndo_init) {
5945 ret = dev->netdev_ops->ndo_init(dev);
5953 if (((dev->hw_features | dev->features) &
5954 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5955 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5956 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5957 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5964 dev->ifindex = dev_new_index(net);
5965 else if (__dev_get_by_index(net, dev->ifindex))
5968 if (dev->iflink == -1)
5969 dev->iflink = dev->ifindex;
5971 /* Transfer changeable features to wanted_features and enable
5972 * software offloads (GSO and GRO).
5974 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5975 dev->features |= NETIF_F_SOFT_FEATURES;
5976 dev->wanted_features = dev->features & dev->hw_features;
5978 if (!(dev->flags & IFF_LOOPBACK)) {
5979 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5982 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5984 dev->vlan_features |= NETIF_F_HIGHDMA;
5986 /* Make NETIF_F_SG inheritable to tunnel devices.
5988 dev->hw_enc_features |= NETIF_F_SG;
5990 /* Make NETIF_F_SG inheritable to MPLS.
5992 dev->mpls_features |= NETIF_F_SG;
5994 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5995 ret = notifier_to_errno(ret);
5999 ret = netdev_register_kobject(dev);
6002 dev->reg_state = NETREG_REGISTERED;
6004 __netdev_update_features(dev);
6007 * Default initial state at registry is that the
6008 * device is present.
6011 set_bit(__LINK_STATE_PRESENT, &dev->state);
6013 linkwatch_init_dev(dev);
6015 dev_init_scheduler(dev);
6017 list_netdevice(dev);
6018 add_device_randomness(dev->dev_addr, dev->addr_len);
6020 /* If the device has permanent device address, driver should
6021 * set dev_addr and also addr_assign_type should be set to
6022 * NET_ADDR_PERM (default value).
6024 if (dev->addr_assign_type == NET_ADDR_PERM)
6025 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6027 /* Notify protocols, that a new device appeared. */
6028 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6029 ret = notifier_to_errno(ret);
6031 rollback_registered(dev);
6032 dev->reg_state = NETREG_UNREGISTERED;
6035 * Prevent userspace races by waiting until the network
6036 * device is fully setup before sending notifications.
6038 if (!dev->rtnl_link_ops ||
6039 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6040 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6046 if (dev->netdev_ops->ndo_uninit)
6047 dev->netdev_ops->ndo_uninit(dev);
6050 EXPORT_SYMBOL(register_netdevice);
6053 * init_dummy_netdev - init a dummy network device for NAPI
6054 * @dev: device to init
6056 * This takes a network device structure and initialize the minimum
6057 * amount of fields so it can be used to schedule NAPI polls without
6058 * registering a full blown interface. This is to be used by drivers
6059 * that need to tie several hardware interfaces to a single NAPI
6060 * poll scheduler due to HW limitations.
6062 int init_dummy_netdev(struct net_device *dev)
6064 /* Clear everything. Note we don't initialize spinlocks
6065 * are they aren't supposed to be taken by any of the
6066 * NAPI code and this dummy netdev is supposed to be
6067 * only ever used for NAPI polls
6069 memset(dev, 0, sizeof(struct net_device));
6071 /* make sure we BUG if trying to hit standard
6072 * register/unregister code path
6074 dev->reg_state = NETREG_DUMMY;
6076 /* NAPI wants this */
6077 INIT_LIST_HEAD(&dev->napi_list);
6079 /* a dummy interface is started by default */
6080 set_bit(__LINK_STATE_PRESENT, &dev->state);
6081 set_bit(__LINK_STATE_START, &dev->state);
6083 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6084 * because users of this 'device' dont need to change
6090 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6094 * register_netdev - register a network device
6095 * @dev: device to register
6097 * Take a completed network device structure and add it to the kernel
6098 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6099 * chain. 0 is returned on success. A negative errno code is returned
6100 * on a failure to set up the device, or if the name is a duplicate.
6102 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6103 * and expands the device name if you passed a format string to
6106 int register_netdev(struct net_device *dev)
6111 err = register_netdevice(dev);
6115 EXPORT_SYMBOL(register_netdev);
6117 int netdev_refcnt_read(const struct net_device *dev)
6121 for_each_possible_cpu(i)
6122 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6125 EXPORT_SYMBOL(netdev_refcnt_read);
6128 * netdev_wait_allrefs - wait until all references are gone.
6129 * @dev: target net_device
6131 * This is called when unregistering network devices.
6133 * Any protocol or device that holds a reference should register
6134 * for netdevice notification, and cleanup and put back the
6135 * reference if they receive an UNREGISTER event.
6136 * We can get stuck here if buggy protocols don't correctly
6139 static void netdev_wait_allrefs(struct net_device *dev)
6141 unsigned long rebroadcast_time, warning_time;
6144 linkwatch_forget_dev(dev);
6146 rebroadcast_time = warning_time = jiffies;
6147 refcnt = netdev_refcnt_read(dev);
6149 while (refcnt != 0) {
6150 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6153 /* Rebroadcast unregister notification */
6154 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6160 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6161 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6163 /* We must not have linkwatch events
6164 * pending on unregister. If this
6165 * happens, we simply run the queue
6166 * unscheduled, resulting in a noop
6169 linkwatch_run_queue();
6174 rebroadcast_time = jiffies;
6179 refcnt = netdev_refcnt_read(dev);
6181 if (time_after(jiffies, warning_time + 10 * HZ)) {
6182 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6184 warning_time = jiffies;
6193 * register_netdevice(x1);
6194 * register_netdevice(x2);
6196 * unregister_netdevice(y1);
6197 * unregister_netdevice(y2);
6203 * We are invoked by rtnl_unlock().
6204 * This allows us to deal with problems:
6205 * 1) We can delete sysfs objects which invoke hotplug
6206 * without deadlocking with linkwatch via keventd.
6207 * 2) Since we run with the RTNL semaphore not held, we can sleep
6208 * safely in order to wait for the netdev refcnt to drop to zero.
6210 * We must not return until all unregister events added during
6211 * the interval the lock was held have been completed.
6213 void netdev_run_todo(void)
6215 struct list_head list;
6217 /* Snapshot list, allow later requests */
6218 list_replace_init(&net_todo_list, &list);
6223 /* Wait for rcu callbacks to finish before next phase */
6224 if (!list_empty(&list))
6227 while (!list_empty(&list)) {
6228 struct net_device *dev
6229 = list_first_entry(&list, struct net_device, todo_list);
6230 list_del(&dev->todo_list);
6233 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6236 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6237 pr_err("network todo '%s' but state %d\n",
6238 dev->name, dev->reg_state);
6243 dev->reg_state = NETREG_UNREGISTERED;
6245 on_each_cpu(flush_backlog, dev, 1);
6247 netdev_wait_allrefs(dev);
6250 BUG_ON(netdev_refcnt_read(dev));
6251 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6252 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6253 WARN_ON(dev->dn_ptr);
6255 if (dev->destructor)
6256 dev->destructor(dev);
6258 /* Report a network device has been unregistered */
6260 dev_net(dev)->dev_unreg_count--;
6262 wake_up(&netdev_unregistering_wq);
6264 /* Free network device */
6265 kobject_put(&dev->dev.kobj);
6269 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6270 * fields in the same order, with only the type differing.
6272 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6273 const struct net_device_stats *netdev_stats)
6275 #if BITS_PER_LONG == 64
6276 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6277 memcpy(stats64, netdev_stats, sizeof(*stats64));
6279 size_t i, n = sizeof(*stats64) / sizeof(u64);
6280 const unsigned long *src = (const unsigned long *)netdev_stats;
6281 u64 *dst = (u64 *)stats64;
6283 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6284 sizeof(*stats64) / sizeof(u64));
6285 for (i = 0; i < n; i++)
6289 EXPORT_SYMBOL(netdev_stats_to_stats64);
6292 * dev_get_stats - get network device statistics
6293 * @dev: device to get statistics from
6294 * @storage: place to store stats
6296 * Get network statistics from device. Return @storage.
6297 * The device driver may provide its own method by setting
6298 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6299 * otherwise the internal statistics structure is used.
6301 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6302 struct rtnl_link_stats64 *storage)
6304 const struct net_device_ops *ops = dev->netdev_ops;
6306 if (ops->ndo_get_stats64) {
6307 memset(storage, 0, sizeof(*storage));
6308 ops->ndo_get_stats64(dev, storage);
6309 } else if (ops->ndo_get_stats) {
6310 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6312 netdev_stats_to_stats64(storage, &dev->stats);
6314 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6317 EXPORT_SYMBOL(dev_get_stats);
6319 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6321 struct netdev_queue *queue = dev_ingress_queue(dev);
6323 #ifdef CONFIG_NET_CLS_ACT
6326 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6329 netdev_init_one_queue(dev, queue, NULL);
6330 queue->qdisc = &noop_qdisc;
6331 queue->qdisc_sleeping = &noop_qdisc;
6332 rcu_assign_pointer(dev->ingress_queue, queue);
6337 static const struct ethtool_ops default_ethtool_ops;
6339 void netdev_set_default_ethtool_ops(struct net_device *dev,
6340 const struct ethtool_ops *ops)
6342 if (dev->ethtool_ops == &default_ethtool_ops)
6343 dev->ethtool_ops = ops;
6345 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6347 void netdev_freemem(struct net_device *dev)
6349 char *addr = (char *)dev - dev->padded;
6351 if (is_vmalloc_addr(addr))
6358 * alloc_netdev_mqs - allocate network device
6359 * @sizeof_priv: size of private data to allocate space for
6360 * @name: device name format string
6361 * @setup: callback to initialize device
6362 * @txqs: the number of TX subqueues to allocate
6363 * @rxqs: the number of RX subqueues to allocate
6365 * Allocates a struct net_device with private data area for driver use
6366 * and performs basic initialization. Also allocates subqueue structs
6367 * for each queue on the device.
6369 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6370 void (*setup)(struct net_device *),
6371 unsigned int txqs, unsigned int rxqs)
6373 struct net_device *dev;
6375 struct net_device *p;
6377 BUG_ON(strlen(name) >= sizeof(dev->name));
6380 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6386 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6391 alloc_size = sizeof(struct net_device);
6393 /* ensure 32-byte alignment of private area */
6394 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6395 alloc_size += sizeof_priv;
6397 /* ensure 32-byte alignment of whole construct */
6398 alloc_size += NETDEV_ALIGN - 1;
6400 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6402 p = vzalloc(alloc_size);
6406 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6407 dev->padded = (char *)dev - (char *)p;
6409 dev->pcpu_refcnt = alloc_percpu(int);
6410 if (!dev->pcpu_refcnt)
6413 if (dev_addr_init(dev))
6419 dev_net_set(dev, &init_net);
6421 dev->gso_max_size = GSO_MAX_SIZE;
6422 dev->gso_max_segs = GSO_MAX_SEGS;
6424 INIT_LIST_HEAD(&dev->napi_list);
6425 INIT_LIST_HEAD(&dev->unreg_list);
6426 INIT_LIST_HEAD(&dev->close_list);
6427 INIT_LIST_HEAD(&dev->link_watch_list);
6428 INIT_LIST_HEAD(&dev->adj_list.upper);
6429 INIT_LIST_HEAD(&dev->adj_list.lower);
6430 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6431 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6432 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6435 dev->num_tx_queues = txqs;
6436 dev->real_num_tx_queues = txqs;
6437 if (netif_alloc_netdev_queues(dev))
6441 dev->num_rx_queues = rxqs;
6442 dev->real_num_rx_queues = rxqs;
6443 if (netif_alloc_rx_queues(dev))
6447 strcpy(dev->name, name);
6448 dev->group = INIT_NETDEV_GROUP;
6449 if (!dev->ethtool_ops)
6450 dev->ethtool_ops = &default_ethtool_ops;
6458 free_percpu(dev->pcpu_refcnt);
6459 netif_free_tx_queues(dev);
6465 netdev_freemem(dev);
6468 EXPORT_SYMBOL(alloc_netdev_mqs);
6471 * free_netdev - free network device
6474 * This function does the last stage of destroying an allocated device
6475 * interface. The reference to the device object is released.
6476 * If this is the last reference then it will be freed.
6478 void free_netdev(struct net_device *dev)
6480 struct napi_struct *p, *n;
6482 release_net(dev_net(dev));
6484 netif_free_tx_queues(dev);
6489 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6491 /* Flush device addresses */
6492 dev_addr_flush(dev);
6494 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6497 free_percpu(dev->pcpu_refcnt);
6498 dev->pcpu_refcnt = NULL;
6500 /* Compatibility with error handling in drivers */
6501 if (dev->reg_state == NETREG_UNINITIALIZED) {
6502 netdev_freemem(dev);
6506 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6507 dev->reg_state = NETREG_RELEASED;
6509 /* will free via device release */
6510 put_device(&dev->dev);
6512 EXPORT_SYMBOL(free_netdev);
6515 * synchronize_net - Synchronize with packet receive processing
6517 * Wait for packets currently being received to be done.
6518 * Does not block later packets from starting.
6520 void synchronize_net(void)
6523 if (rtnl_is_locked())
6524 synchronize_rcu_expedited();
6528 EXPORT_SYMBOL(synchronize_net);
6531 * unregister_netdevice_queue - remove device from the kernel
6535 * This function shuts down a device interface and removes it
6536 * from the kernel tables.
6537 * If head not NULL, device is queued to be unregistered later.
6539 * Callers must hold the rtnl semaphore. You may want
6540 * unregister_netdev() instead of this.
6543 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6548 list_move_tail(&dev->unreg_list, head);
6550 rollback_registered(dev);
6551 /* Finish processing unregister after unlock */
6555 EXPORT_SYMBOL(unregister_netdevice_queue);
6558 * unregister_netdevice_many - unregister many devices
6559 * @head: list of devices
6561 * Note: As most callers use a stack allocated list_head,
6562 * we force a list_del() to make sure stack wont be corrupted later.
6564 void unregister_netdevice_many(struct list_head *head)
6566 struct net_device *dev;
6568 if (!list_empty(head)) {
6569 rollback_registered_many(head);
6570 list_for_each_entry(dev, head, unreg_list)
6575 EXPORT_SYMBOL(unregister_netdevice_many);
6578 * unregister_netdev - remove device from the kernel
6581 * This function shuts down a device interface and removes it
6582 * from the kernel tables.
6584 * This is just a wrapper for unregister_netdevice that takes
6585 * the rtnl semaphore. In general you want to use this and not
6586 * unregister_netdevice.
6588 void unregister_netdev(struct net_device *dev)
6591 unregister_netdevice(dev);
6594 EXPORT_SYMBOL(unregister_netdev);
6597 * dev_change_net_namespace - move device to different nethost namespace
6599 * @net: network namespace
6600 * @pat: If not NULL name pattern to try if the current device name
6601 * is already taken in the destination network namespace.
6603 * This function shuts down a device interface and moves it
6604 * to a new network namespace. On success 0 is returned, on
6605 * a failure a netagive errno code is returned.
6607 * Callers must hold the rtnl semaphore.
6610 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6616 /* Don't allow namespace local devices to be moved. */
6618 if (dev->features & NETIF_F_NETNS_LOCAL)
6621 /* Ensure the device has been registrered */
6622 if (dev->reg_state != NETREG_REGISTERED)
6625 /* Get out if there is nothing todo */
6627 if (net_eq(dev_net(dev), net))
6630 /* Pick the destination device name, and ensure
6631 * we can use it in the destination network namespace.
6634 if (__dev_get_by_name(net, dev->name)) {
6635 /* We get here if we can't use the current device name */
6638 if (dev_get_valid_name(net, dev, pat) < 0)
6643 * And now a mini version of register_netdevice unregister_netdevice.
6646 /* If device is running close it first. */
6649 /* And unlink it from device chain */
6651 unlist_netdevice(dev);
6655 /* Shutdown queueing discipline. */
6658 /* Notify protocols, that we are about to destroy
6659 this device. They should clean all the things.
6661 Note that dev->reg_state stays at NETREG_REGISTERED.
6662 This is wanted because this way 8021q and macvlan know
6663 the device is just moving and can keep their slaves up.
6665 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6667 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6668 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6671 * Flush the unicast and multicast chains
6676 /* Send a netdev-removed uevent to the old namespace */
6677 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6679 /* Actually switch the network namespace */
6680 dev_net_set(dev, net);
6682 /* If there is an ifindex conflict assign a new one */
6683 if (__dev_get_by_index(net, dev->ifindex)) {
6684 int iflink = (dev->iflink == dev->ifindex);
6685 dev->ifindex = dev_new_index(net);
6687 dev->iflink = dev->ifindex;
6690 /* Send a netdev-add uevent to the new namespace */
6691 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6693 /* Fixup kobjects */
6694 err = device_rename(&dev->dev, dev->name);
6697 /* Add the device back in the hashes */
6698 list_netdevice(dev);
6700 /* Notify protocols, that a new device appeared. */
6701 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6704 * Prevent userspace races by waiting until the network
6705 * device is fully setup before sending notifications.
6707 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6714 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6716 static int dev_cpu_callback(struct notifier_block *nfb,
6717 unsigned long action,
6720 struct sk_buff **list_skb;
6721 struct sk_buff *skb;
6722 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6723 struct softnet_data *sd, *oldsd;
6725 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6728 local_irq_disable();
6729 cpu = smp_processor_id();
6730 sd = &per_cpu(softnet_data, cpu);
6731 oldsd = &per_cpu(softnet_data, oldcpu);
6733 /* Find end of our completion_queue. */
6734 list_skb = &sd->completion_queue;
6736 list_skb = &(*list_skb)->next;
6737 /* Append completion queue from offline CPU. */
6738 *list_skb = oldsd->completion_queue;
6739 oldsd->completion_queue = NULL;
6741 /* Append output queue from offline CPU. */
6742 if (oldsd->output_queue) {
6743 *sd->output_queue_tailp = oldsd->output_queue;
6744 sd->output_queue_tailp = oldsd->output_queue_tailp;
6745 oldsd->output_queue = NULL;
6746 oldsd->output_queue_tailp = &oldsd->output_queue;
6748 /* Append NAPI poll list from offline CPU. */
6749 if (!list_empty(&oldsd->poll_list)) {
6750 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6751 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6754 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6757 /* Process offline CPU's input_pkt_queue */
6758 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6759 netif_rx_internal(skb);
6760 input_queue_head_incr(oldsd);
6762 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6763 netif_rx_internal(skb);
6764 input_queue_head_incr(oldsd);
6772 * netdev_increment_features - increment feature set by one
6773 * @all: current feature set
6774 * @one: new feature set
6775 * @mask: mask feature set
6777 * Computes a new feature set after adding a device with feature set
6778 * @one to the master device with current feature set @all. Will not
6779 * enable anything that is off in @mask. Returns the new feature set.
6781 netdev_features_t netdev_increment_features(netdev_features_t all,
6782 netdev_features_t one, netdev_features_t mask)
6784 if (mask & NETIF_F_GEN_CSUM)
6785 mask |= NETIF_F_ALL_CSUM;
6786 mask |= NETIF_F_VLAN_CHALLENGED;
6788 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6789 all &= one | ~NETIF_F_ALL_FOR_ALL;
6791 /* If one device supports hw checksumming, set for all. */
6792 if (all & NETIF_F_GEN_CSUM)
6793 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6797 EXPORT_SYMBOL(netdev_increment_features);
6799 static struct hlist_head * __net_init netdev_create_hash(void)
6802 struct hlist_head *hash;
6804 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6806 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6807 INIT_HLIST_HEAD(&hash[i]);
6812 /* Initialize per network namespace state */
6813 static int __net_init netdev_init(struct net *net)
6815 if (net != &init_net)
6816 INIT_LIST_HEAD(&net->dev_base_head);
6818 net->dev_name_head = netdev_create_hash();
6819 if (net->dev_name_head == NULL)
6822 net->dev_index_head = netdev_create_hash();
6823 if (net->dev_index_head == NULL)
6829 kfree(net->dev_name_head);
6835 * netdev_drivername - network driver for the device
6836 * @dev: network device
6838 * Determine network driver for device.
6840 const char *netdev_drivername(const struct net_device *dev)
6842 const struct device_driver *driver;
6843 const struct device *parent;
6844 const char *empty = "";
6846 parent = dev->dev.parent;
6850 driver = parent->driver;
6851 if (driver && driver->name)
6852 return driver->name;
6856 static int __netdev_printk(const char *level, const struct net_device *dev,
6857 struct va_format *vaf)
6861 if (dev && dev->dev.parent) {
6862 r = dev_printk_emit(level[1] - '0',
6865 dev_driver_string(dev->dev.parent),
6866 dev_name(dev->dev.parent),
6867 netdev_name(dev), vaf);
6869 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6871 r = printk("%s(NULL net_device): %pV", level, vaf);
6877 int netdev_printk(const char *level, const struct net_device *dev,
6878 const char *format, ...)
6880 struct va_format vaf;
6884 va_start(args, format);
6889 r = __netdev_printk(level, dev, &vaf);
6895 EXPORT_SYMBOL(netdev_printk);
6897 #define define_netdev_printk_level(func, level) \
6898 int func(const struct net_device *dev, const char *fmt, ...) \
6901 struct va_format vaf; \
6904 va_start(args, fmt); \
6909 r = __netdev_printk(level, dev, &vaf); \
6915 EXPORT_SYMBOL(func);
6917 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6918 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6919 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6920 define_netdev_printk_level(netdev_err, KERN_ERR);
6921 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6922 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6923 define_netdev_printk_level(netdev_info, KERN_INFO);
6925 static void __net_exit netdev_exit(struct net *net)
6927 kfree(net->dev_name_head);
6928 kfree(net->dev_index_head);
6931 static struct pernet_operations __net_initdata netdev_net_ops = {
6932 .init = netdev_init,
6933 .exit = netdev_exit,
6936 static void __net_exit default_device_exit(struct net *net)
6938 struct net_device *dev, *aux;
6940 * Push all migratable network devices back to the
6941 * initial network namespace
6944 for_each_netdev_safe(net, dev, aux) {
6946 char fb_name[IFNAMSIZ];
6948 /* Ignore unmoveable devices (i.e. loopback) */
6949 if (dev->features & NETIF_F_NETNS_LOCAL)
6952 /* Leave virtual devices for the generic cleanup */
6953 if (dev->rtnl_link_ops)
6956 /* Push remaining network devices to init_net */
6957 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6958 err = dev_change_net_namespace(dev, &init_net, fb_name);
6960 pr_emerg("%s: failed to move %s to init_net: %d\n",
6961 __func__, dev->name, err);
6968 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6970 /* Return with the rtnl_lock held when there are no network
6971 * devices unregistering in any network namespace in net_list.
6978 prepare_to_wait(&netdev_unregistering_wq, &wait,
6979 TASK_UNINTERRUPTIBLE);
6980 unregistering = false;
6982 list_for_each_entry(net, net_list, exit_list) {
6983 if (net->dev_unreg_count > 0) {
6984 unregistering = true;
6993 finish_wait(&netdev_unregistering_wq, &wait);
6996 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6998 /* At exit all network devices most be removed from a network
6999 * namespace. Do this in the reverse order of registration.
7000 * Do this across as many network namespaces as possible to
7001 * improve batching efficiency.
7003 struct net_device *dev;
7005 LIST_HEAD(dev_kill_list);
7007 /* To prevent network device cleanup code from dereferencing
7008 * loopback devices or network devices that have been freed
7009 * wait here for all pending unregistrations to complete,
7010 * before unregistring the loopback device and allowing the
7011 * network namespace be freed.
7013 * The netdev todo list containing all network devices
7014 * unregistrations that happen in default_device_exit_batch
7015 * will run in the rtnl_unlock() at the end of
7016 * default_device_exit_batch.
7018 rtnl_lock_unregistering(net_list);
7019 list_for_each_entry(net, net_list, exit_list) {
7020 for_each_netdev_reverse(net, dev) {
7021 if (dev->rtnl_link_ops)
7022 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7024 unregister_netdevice_queue(dev, &dev_kill_list);
7027 unregister_netdevice_many(&dev_kill_list);
7031 static struct pernet_operations __net_initdata default_device_ops = {
7032 .exit = default_device_exit,
7033 .exit_batch = default_device_exit_batch,
7037 * Initialize the DEV module. At boot time this walks the device list and
7038 * unhooks any devices that fail to initialise (normally hardware not
7039 * present) and leaves us with a valid list of present and active devices.
7044 * This is called single threaded during boot, so no need
7045 * to take the rtnl semaphore.
7047 static int __init net_dev_init(void)
7049 int i, rc = -ENOMEM;
7051 BUG_ON(!dev_boot_phase);
7053 if (dev_proc_init())
7056 if (netdev_kobject_init())
7059 INIT_LIST_HEAD(&ptype_all);
7060 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7061 INIT_LIST_HEAD(&ptype_base[i]);
7063 INIT_LIST_HEAD(&offload_base);
7065 if (register_pernet_subsys(&netdev_net_ops))
7069 * Initialise the packet receive queues.
7072 for_each_possible_cpu(i) {
7073 struct softnet_data *sd = &per_cpu(softnet_data, i);
7075 skb_queue_head_init(&sd->input_pkt_queue);
7076 skb_queue_head_init(&sd->process_queue);
7077 INIT_LIST_HEAD(&sd->poll_list);
7078 sd->output_queue_tailp = &sd->output_queue;
7080 sd->csd.func = rps_trigger_softirq;
7085 sd->backlog.poll = process_backlog;
7086 sd->backlog.weight = weight_p;
7091 /* The loopback device is special if any other network devices
7092 * is present in a network namespace the loopback device must
7093 * be present. Since we now dynamically allocate and free the
7094 * loopback device ensure this invariant is maintained by
7095 * keeping the loopback device as the first device on the
7096 * list of network devices. Ensuring the loopback devices
7097 * is the first device that appears and the last network device
7100 if (register_pernet_device(&loopback_net_ops))
7103 if (register_pernet_device(&default_device_ops))
7106 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7107 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7109 hotcpu_notifier(dev_cpu_callback, 0);
7116 subsys_initcall(net_dev_init);