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>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 static DEFINE_SPINLOCK(ptype_lock);
144 static DEFINE_SPINLOCK(offload_lock);
145 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
146 struct list_head ptype_all __read_mostly; /* Taps */
147 static struct list_head offload_base __read_mostly;
150 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
153 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base_head list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
168 DEFINE_RWLOCK(dev_base_lock);
169 EXPORT_SYMBOL(dev_base_lock);
171 /* protects napi_hash addition/deletion and napi_gen_id */
172 static DEFINE_SPINLOCK(napi_hash_lock);
174 static unsigned int napi_gen_id;
175 static DEFINE_HASHTABLE(napi_hash, 8);
177 seqcount_t devnet_rename_seq;
179 static inline void dev_base_seq_inc(struct net *net)
181 while (++net->dev_base_seq == 0);
184 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
186 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
188 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
191 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
193 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
196 static inline void rps_lock(struct softnet_data *sd)
199 spin_lock(&sd->input_pkt_queue.lock);
203 static inline void rps_unlock(struct softnet_data *sd)
206 spin_unlock(&sd->input_pkt_queue.lock);
210 /* Device list insertion */
211 static void list_netdevice(struct net_device *dev)
213 struct net *net = dev_net(dev);
217 write_lock_bh(&dev_base_lock);
218 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head_rcu(&dev->index_hlist,
221 dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
224 dev_base_seq_inc(net);
227 /* Device list removal
228 * caller must respect a RCU grace period before freeing/reusing dev
230 static void unlist_netdevice(struct net_device *dev)
234 /* Unlink dev from the device chain */
235 write_lock_bh(&dev_base_lock);
236 list_del_rcu(&dev->dev_list);
237 hlist_del_rcu(&dev->name_hlist);
238 hlist_del_rcu(&dev->index_hlist);
239 write_unlock_bh(&dev_base_lock);
241 dev_base_seq_inc(dev_net(dev));
248 static RAW_NOTIFIER_HEAD(netdev_chain);
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
255 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
256 EXPORT_PER_CPU_SYMBOL(softnet_data);
258 #ifdef CONFIG_LOCKDEP
260 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
261 * according to dev->type
263 static const unsigned short netdev_lock_type[] =
264 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
265 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
266 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
267 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
268 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
269 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
270 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
271 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
272 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
273 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
274 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
275 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
276 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
277 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
278 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
280 static const char *const netdev_lock_name[] =
281 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
282 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
283 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
284 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
285 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
286 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
287 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
288 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
289 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
290 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
291 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
292 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
293 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
294 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
295 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
297 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
304 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
305 if (netdev_lock_type[i] == dev_type)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type) - 1;
311 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
312 unsigned short dev_type)
316 i = netdev_lock_pos(dev_type);
317 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
318 netdev_lock_name[i]);
321 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
325 i = netdev_lock_pos(dev->type);
326 lockdep_set_class_and_name(&dev->addr_list_lock,
327 &netdev_addr_lock_key[i],
328 netdev_lock_name[i]);
331 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
332 unsigned short dev_type)
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 /*******************************************************************************
342 Protocol management and registration routines
344 *******************************************************************************/
347 * Add a protocol ID to the list. Now that the input handler is
348 * smarter we can dispense with all the messy stuff that used to be
351 * BEWARE!!! Protocol handlers, mangling input packets,
352 * MUST BE last in hash buckets and checking protocol handlers
353 * MUST start from promiscuous ptype_all chain in net_bh.
354 * It is true now, do not change it.
355 * Explanation follows: if protocol handler, mangling packet, will
356 * be the first on list, it is not able to sense, that packet
357 * is cloned and should be copied-on-write, so that it will
358 * change it and subsequent readers will get broken packet.
362 static inline struct list_head *ptype_head(const struct packet_type *pt)
364 if (pt->type == htons(ETH_P_ALL))
367 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
371 * dev_add_pack - add packet handler
372 * @pt: packet type declaration
374 * Add a protocol handler to the networking stack. The passed &packet_type
375 * is linked into kernel lists and may not be freed until it has been
376 * removed from the kernel lists.
378 * This call does not sleep therefore it can not
379 * guarantee all CPU's that are in middle of receiving packets
380 * will see the new packet type (until the next received packet).
383 void dev_add_pack(struct packet_type *pt)
385 struct list_head *head = ptype_head(pt);
387 spin_lock(&ptype_lock);
388 list_add_rcu(&pt->list, head);
389 spin_unlock(&ptype_lock);
391 EXPORT_SYMBOL(dev_add_pack);
394 * __dev_remove_pack - remove packet handler
395 * @pt: packet type declaration
397 * Remove a protocol handler that was previously added to the kernel
398 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
399 * from the kernel lists and can be freed or reused once this function
402 * The packet type might still be in use by receivers
403 * and must not be freed until after all the CPU's have gone
404 * through a quiescent state.
406 void __dev_remove_pack(struct packet_type *pt)
408 struct list_head *head = ptype_head(pt);
409 struct packet_type *pt1;
411 spin_lock(&ptype_lock);
413 list_for_each_entry(pt1, head, list) {
415 list_del_rcu(&pt->list);
420 pr_warn("dev_remove_pack: %p not found\n", pt);
422 spin_unlock(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
444 EXPORT_SYMBOL(dev_remove_pack);
448 * dev_add_offload - register offload handlers
449 * @po: protocol offload declaration
451 * Add protocol offload handlers to the networking stack. The passed
452 * &proto_offload is linked into kernel lists and may not be freed until
453 * it has been removed from the kernel lists.
455 * This call does not sleep therefore it can not
456 * guarantee all CPU's that are in middle of receiving packets
457 * will see the new offload handlers (until the next received packet).
459 void dev_add_offload(struct packet_offload *po)
461 struct list_head *head = &offload_base;
463 spin_lock(&offload_lock);
464 list_add_rcu(&po->list, head);
465 spin_unlock(&offload_lock);
467 EXPORT_SYMBOL(dev_add_offload);
470 * __dev_remove_offload - remove offload handler
471 * @po: packet offload declaration
473 * Remove a protocol offload handler that was previously added to the
474 * kernel offload handlers by dev_add_offload(). The passed &offload_type
475 * is removed from the kernel lists and can be freed or reused once this
478 * The packet type might still be in use by receivers
479 * and must not be freed until after all the CPU's have gone
480 * through a quiescent state.
482 void __dev_remove_offload(struct packet_offload *po)
484 struct list_head *head = &offload_base;
485 struct packet_offload *po1;
487 spin_lock(&offload_lock);
489 list_for_each_entry(po1, head, list) {
491 list_del_rcu(&po->list);
496 pr_warn("dev_remove_offload: %p not found\n", po);
498 spin_unlock(&offload_lock);
500 EXPORT_SYMBOL(__dev_remove_offload);
503 * dev_remove_offload - remove packet offload handler
504 * @po: packet offload declaration
506 * Remove a packet offload handler that was previously added to the kernel
507 * offload handlers by dev_add_offload(). The passed &offload_type is
508 * removed from the kernel lists and can be freed or reused once this
511 * This call sleeps to guarantee that no CPU is looking at the packet
514 void dev_remove_offload(struct packet_offload *po)
516 __dev_remove_offload(po);
520 EXPORT_SYMBOL(dev_remove_offload);
522 /******************************************************************************
524 Device Boot-time Settings Routines
526 *******************************************************************************/
528 /* Boot time configuration table */
529 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
532 * netdev_boot_setup_add - add new setup entry
533 * @name: name of the device
534 * @map: configured settings for the device
536 * Adds new setup entry to the dev_boot_setup list. The function
537 * returns 0 on error and 1 on success. This is a generic routine to
540 static int netdev_boot_setup_add(char *name, struct ifmap *map)
542 struct netdev_boot_setup *s;
546 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
547 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
548 memset(s[i].name, 0, sizeof(s[i].name));
549 strlcpy(s[i].name, name, IFNAMSIZ);
550 memcpy(&s[i].map, map, sizeof(s[i].map));
555 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
559 * netdev_boot_setup_check - check boot time settings
560 * @dev: the netdevice
562 * Check boot time settings for the device.
563 * The found settings are set for the device to be used
564 * later in the device probing.
565 * Returns 0 if no settings found, 1 if they are.
567 int netdev_boot_setup_check(struct net_device *dev)
569 struct netdev_boot_setup *s = dev_boot_setup;
572 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
573 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
574 !strcmp(dev->name, s[i].name)) {
575 dev->irq = s[i].map.irq;
576 dev->base_addr = s[i].map.base_addr;
577 dev->mem_start = s[i].map.mem_start;
578 dev->mem_end = s[i].map.mem_end;
584 EXPORT_SYMBOL(netdev_boot_setup_check);
588 * netdev_boot_base - get address from boot time settings
589 * @prefix: prefix for network device
590 * @unit: id for network device
592 * Check boot time settings for the base address of device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found.
597 unsigned long netdev_boot_base(const char *prefix, int unit)
599 const struct netdev_boot_setup *s = dev_boot_setup;
603 sprintf(name, "%s%d", prefix, unit);
606 * If device already registered then return base of 1
607 * to indicate not to probe for this interface
609 if (__dev_get_by_name(&init_net, name))
612 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
613 if (!strcmp(name, s[i].name))
614 return s[i].map.base_addr;
619 * Saves at boot time configured settings for any netdevice.
621 int __init netdev_boot_setup(char *str)
626 str = get_options(str, ARRAY_SIZE(ints), ints);
631 memset(&map, 0, sizeof(map));
635 map.base_addr = ints[2];
637 map.mem_start = ints[3];
639 map.mem_end = ints[4];
641 /* Add new entry to the list */
642 return netdev_boot_setup_add(str, &map);
645 __setup("netdev=", netdev_boot_setup);
647 /*******************************************************************************
649 Device Interface Subroutines
651 *******************************************************************************/
654 * __dev_get_by_name - find a device by its name
655 * @net: the applicable net namespace
656 * @name: name to find
658 * Find an interface by name. Must be called under RTNL semaphore
659 * or @dev_base_lock. If the name is found a pointer to the device
660 * is returned. If the name is not found then %NULL is returned. The
661 * reference counters are not incremented so the caller must be
662 * careful with locks.
665 struct net_device *__dev_get_by_name(struct net *net, const char *name)
667 struct net_device *dev;
668 struct hlist_head *head = dev_name_hash(net, name);
670 hlist_for_each_entry(dev, head, name_hlist)
671 if (!strncmp(dev->name, name, IFNAMSIZ))
676 EXPORT_SYMBOL(__dev_get_by_name);
679 * dev_get_by_name_rcu - find a device by its name
680 * @net: the applicable net namespace
681 * @name: name to find
683 * Find an interface by name.
684 * If the name is found a pointer to the device is returned.
685 * If the name is not found then %NULL is returned.
686 * The reference counters are not incremented so the caller must be
687 * careful with locks. The caller must hold RCU lock.
690 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
692 struct net_device *dev;
693 struct hlist_head *head = dev_name_hash(net, name);
695 hlist_for_each_entry_rcu(dev, head, name_hlist)
696 if (!strncmp(dev->name, name, IFNAMSIZ))
701 EXPORT_SYMBOL(dev_get_by_name_rcu);
704 * dev_get_by_name - find a device by its name
705 * @net: the applicable net namespace
706 * @name: name to find
708 * Find an interface by name. This can be called from any
709 * context and does its own locking. The returned handle has
710 * the usage count incremented and the caller must use dev_put() to
711 * release it when it is no longer needed. %NULL is returned if no
712 * matching device is found.
715 struct net_device *dev_get_by_name(struct net *net, const char *name)
717 struct net_device *dev;
720 dev = dev_get_by_name_rcu(net, name);
726 EXPORT_SYMBOL(dev_get_by_name);
729 * __dev_get_by_index - find a device by its ifindex
730 * @net: the applicable net namespace
731 * @ifindex: index of device
733 * Search for an interface by index. Returns %NULL if the device
734 * is not found or a pointer to the device. The device has not
735 * had its reference counter increased so the caller must be careful
736 * about locking. The caller must hold either the RTNL semaphore
740 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
742 struct net_device *dev;
743 struct hlist_head *head = dev_index_hash(net, ifindex);
745 hlist_for_each_entry(dev, head, index_hlist)
746 if (dev->ifindex == ifindex)
751 EXPORT_SYMBOL(__dev_get_by_index);
754 * dev_get_by_index_rcu - find a device by its ifindex
755 * @net: the applicable net namespace
756 * @ifindex: index of device
758 * Search for an interface by index. Returns %NULL if the device
759 * is not found or a pointer to the device. The device has not
760 * had its reference counter increased so the caller must be careful
761 * about locking. The caller must hold RCU lock.
764 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
766 struct net_device *dev;
767 struct hlist_head *head = dev_index_hash(net, ifindex);
769 hlist_for_each_entry_rcu(dev, head, index_hlist)
770 if (dev->ifindex == ifindex)
775 EXPORT_SYMBOL(dev_get_by_index_rcu);
779 * dev_get_by_index - find a device by its ifindex
780 * @net: the applicable net namespace
781 * @ifindex: index of device
783 * Search for an interface by index. Returns NULL if the device
784 * is not found or a pointer to the device. The device returned has
785 * had a reference added and the pointer is safe until the user calls
786 * dev_put to indicate they have finished with it.
789 struct net_device *dev_get_by_index(struct net *net, int ifindex)
791 struct net_device *dev;
794 dev = dev_get_by_index_rcu(net, ifindex);
800 EXPORT_SYMBOL(dev_get_by_index);
803 * netdev_get_name - get a netdevice name, knowing its ifindex.
804 * @net: network namespace
805 * @name: a pointer to the buffer where the name will be stored.
806 * @ifindex: the ifindex of the interface to get the name from.
808 * The use of raw_seqcount_begin() and cond_resched() before
809 * retrying is required as we want to give the writers a chance
810 * to complete when CONFIG_PREEMPT is not set.
812 int netdev_get_name(struct net *net, char *name, int ifindex)
814 struct net_device *dev;
818 seq = raw_seqcount_begin(&devnet_rename_seq);
820 dev = dev_get_by_index_rcu(net, ifindex);
826 strcpy(name, dev->name);
828 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
837 * dev_getbyhwaddr_rcu - find a device by its hardware address
838 * @net: the applicable net namespace
839 * @type: media type of device
840 * @ha: hardware address
842 * Search for an interface by MAC address. Returns NULL if the device
843 * is not found or a pointer to the device.
844 * The caller must hold RCU or RTNL.
845 * The returned device has not had its ref count increased
846 * and the caller must therefore be careful about locking
850 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
853 struct net_device *dev;
855 for_each_netdev_rcu(net, dev)
856 if (dev->type == type &&
857 !memcmp(dev->dev_addr, ha, dev->addr_len))
862 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
864 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
866 struct net_device *dev;
869 for_each_netdev(net, dev)
870 if (dev->type == type)
875 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
877 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
879 struct net_device *dev, *ret = NULL;
882 for_each_netdev_rcu(net, dev)
883 if (dev->type == type) {
891 EXPORT_SYMBOL(dev_getfirstbyhwtype);
894 * dev_get_by_flags_rcu - find any device with given flags
895 * @net: the applicable net namespace
896 * @if_flags: IFF_* values
897 * @mask: bitmask of bits in if_flags to check
899 * Search for any interface with the given flags. Returns NULL if a device
900 * is not found or a pointer to the device. Must be called inside
901 * rcu_read_lock(), and result refcount is unchanged.
904 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
907 struct net_device *dev, *ret;
910 for_each_netdev_rcu(net, dev) {
911 if (((dev->flags ^ if_flags) & mask) == 0) {
918 EXPORT_SYMBOL(dev_get_by_flags_rcu);
921 * dev_valid_name - check if name is okay for network device
924 * Network device names need to be valid file names to
925 * to allow sysfs to work. We also disallow any kind of
928 bool dev_valid_name(const char *name)
932 if (strlen(name) >= IFNAMSIZ)
934 if (!strcmp(name, ".") || !strcmp(name, ".."))
938 if (*name == '/' || isspace(*name))
944 EXPORT_SYMBOL(dev_valid_name);
947 * __dev_alloc_name - allocate a name for a device
948 * @net: network namespace to allocate the device name in
949 * @name: name format string
950 * @buf: scratch buffer and result name string
952 * Passed a format string - eg "lt%d" it will try and find a suitable
953 * id. It scans list of devices to build up a free map, then chooses
954 * the first empty slot. The caller must hold the dev_base or rtnl lock
955 * while allocating the name and adding the device in order to avoid
957 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
958 * Returns the number of the unit assigned or a negative errno code.
961 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
965 const int max_netdevices = 8*PAGE_SIZE;
966 unsigned long *inuse;
967 struct net_device *d;
969 p = strnchr(name, IFNAMSIZ-1, '%');
972 * Verify the string as this thing may have come from
973 * the user. There must be either one "%d" and no other "%"
976 if (p[1] != 'd' || strchr(p + 2, '%'))
979 /* Use one page as a bit array of possible slots */
980 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
984 for_each_netdev(net, d) {
985 if (!sscanf(d->name, name, &i))
987 if (i < 0 || i >= max_netdevices)
990 /* avoid cases where sscanf is not exact inverse of printf */
991 snprintf(buf, IFNAMSIZ, name, i);
992 if (!strncmp(buf, d->name, IFNAMSIZ))
996 i = find_first_zero_bit(inuse, max_netdevices);
997 free_page((unsigned long) inuse);
1001 snprintf(buf, IFNAMSIZ, name, i);
1002 if (!__dev_get_by_name(net, buf))
1005 /* It is possible to run out of possible slots
1006 * when the name is long and there isn't enough space left
1007 * for the digits, or if all bits are used.
1013 * dev_alloc_name - allocate a name for a device
1015 * @name: name format string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 int dev_alloc_name(struct net_device *dev, const char *name)
1032 BUG_ON(!dev_net(dev));
1034 ret = __dev_alloc_name(net, name, buf);
1036 strlcpy(dev->name, buf, IFNAMSIZ);
1039 EXPORT_SYMBOL(dev_alloc_name);
1041 static int dev_alloc_name_ns(struct net *net,
1042 struct net_device *dev,
1048 ret = __dev_alloc_name(net, name, buf);
1050 strlcpy(dev->name, buf, IFNAMSIZ);
1054 static int dev_get_valid_name(struct net *net,
1055 struct net_device *dev,
1060 if (!dev_valid_name(name))
1063 if (strchr(name, '%'))
1064 return dev_alloc_name_ns(net, dev, name);
1065 else if (__dev_get_by_name(net, name))
1067 else if (dev->name != name)
1068 strlcpy(dev->name, name, IFNAMSIZ);
1074 * dev_change_name - change name of a device
1076 * @newname: name (or format string) must be at least IFNAMSIZ
1078 * Change name of a device, can pass format strings "eth%d".
1081 int dev_change_name(struct net_device *dev, const char *newname)
1083 char oldname[IFNAMSIZ];
1089 BUG_ON(!dev_net(dev));
1092 if (dev->flags & IFF_UP)
1095 write_seqcount_begin(&devnet_rename_seq);
1097 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1098 write_seqcount_end(&devnet_rename_seq);
1102 memcpy(oldname, dev->name, IFNAMSIZ);
1104 err = dev_get_valid_name(net, dev, newname);
1106 write_seqcount_end(&devnet_rename_seq);
1111 ret = device_rename(&dev->dev, dev->name);
1113 memcpy(dev->name, oldname, IFNAMSIZ);
1114 write_seqcount_end(&devnet_rename_seq);
1118 write_seqcount_end(&devnet_rename_seq);
1120 write_lock_bh(&dev_base_lock);
1121 hlist_del_rcu(&dev->name_hlist);
1122 write_unlock_bh(&dev_base_lock);
1126 write_lock_bh(&dev_base_lock);
1127 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1128 write_unlock_bh(&dev_base_lock);
1130 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1131 ret = notifier_to_errno(ret);
1134 /* err >= 0 after dev_alloc_name() or stores the first errno */
1137 write_seqcount_begin(&devnet_rename_seq);
1138 memcpy(dev->name, oldname, IFNAMSIZ);
1141 pr_err("%s: name change rollback failed: %d\n",
1150 * dev_set_alias - change ifalias of a device
1152 * @alias: name up to IFALIASZ
1153 * @len: limit of bytes to copy from info
1155 * Set ifalias for a device,
1157 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1163 if (len >= IFALIASZ)
1167 kfree(dev->ifalias);
1168 dev->ifalias = NULL;
1172 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1175 dev->ifalias = new_ifalias;
1177 strlcpy(dev->ifalias, alias, len+1);
1183 * netdev_features_change - device changes features
1184 * @dev: device to cause notification
1186 * Called to indicate a device has changed features.
1188 void netdev_features_change(struct net_device *dev)
1190 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1192 EXPORT_SYMBOL(netdev_features_change);
1195 * netdev_state_change - device changes state
1196 * @dev: device to cause notification
1198 * Called to indicate a device has changed state. This function calls
1199 * the notifier chains for netdev_chain and sends a NEWLINK message
1200 * to the routing socket.
1202 void netdev_state_change(struct net_device *dev)
1204 if (dev->flags & IFF_UP) {
1205 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1206 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1209 EXPORT_SYMBOL(netdev_state_change);
1212 * netdev_notify_peers - notify network peers about existence of @dev
1213 * @dev: network device
1215 * Generate traffic such that interested network peers are aware of
1216 * @dev, such as by generating a gratuitous ARP. This may be used when
1217 * a device wants to inform the rest of the network about some sort of
1218 * reconfiguration such as a failover event or virtual machine
1221 void netdev_notify_peers(struct net_device *dev)
1224 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1227 EXPORT_SYMBOL(netdev_notify_peers);
1229 static int __dev_open(struct net_device *dev)
1231 const struct net_device_ops *ops = dev->netdev_ops;
1236 if (!netif_device_present(dev))
1239 /* Block netpoll from trying to do any rx path servicing.
1240 * If we don't do this there is a chance ndo_poll_controller
1241 * or ndo_poll may be running while we open the device
1243 netpoll_rx_disable(dev);
1245 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1246 ret = notifier_to_errno(ret);
1250 set_bit(__LINK_STATE_START, &dev->state);
1252 if (ops->ndo_validate_addr)
1253 ret = ops->ndo_validate_addr(dev);
1255 if (!ret && ops->ndo_open)
1256 ret = ops->ndo_open(dev);
1258 netpoll_rx_enable(dev);
1261 clear_bit(__LINK_STATE_START, &dev->state);
1263 dev->flags |= IFF_UP;
1264 net_dmaengine_get();
1265 dev_set_rx_mode(dev);
1267 add_device_randomness(dev->dev_addr, dev->addr_len);
1274 * dev_open - prepare an interface for use.
1275 * @dev: device to open
1277 * Takes a device from down to up state. The device's private open
1278 * function is invoked and then the multicast lists are loaded. Finally
1279 * the device is moved into the up state and a %NETDEV_UP message is
1280 * sent to the netdev notifier chain.
1282 * Calling this function on an active interface is a nop. On a failure
1283 * a negative errno code is returned.
1285 int dev_open(struct net_device *dev)
1289 if (dev->flags & IFF_UP)
1292 ret = __dev_open(dev);
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1297 call_netdevice_notifiers(NETDEV_UP, dev);
1301 EXPORT_SYMBOL(dev_open);
1303 static int __dev_close_many(struct list_head *head)
1305 struct net_device *dev;
1310 list_for_each_entry(dev, head, unreg_list) {
1311 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1313 clear_bit(__LINK_STATE_START, &dev->state);
1315 /* Synchronize to scheduled poll. We cannot touch poll list, it
1316 * can be even on different cpu. So just clear netif_running().
1318 * dev->stop() will invoke napi_disable() on all of it's
1319 * napi_struct instances on this device.
1321 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1324 dev_deactivate_many(head);
1326 list_for_each_entry(dev, head, unreg_list) {
1327 const struct net_device_ops *ops = dev->netdev_ops;
1330 * Call the device specific close. This cannot fail.
1331 * Only if device is UP
1333 * We allow it to be called even after a DETACH hot-plug
1339 dev->flags &= ~IFF_UP;
1340 net_dmaengine_put();
1346 static int __dev_close(struct net_device *dev)
1351 /* Temporarily disable netpoll until the interface is down */
1352 netpoll_rx_disable(dev);
1354 list_add(&dev->unreg_list, &single);
1355 retval = __dev_close_many(&single);
1358 netpoll_rx_enable(dev);
1362 static int dev_close_many(struct list_head *head)
1364 struct net_device *dev, *tmp;
1365 LIST_HEAD(tmp_list);
1367 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1368 if (!(dev->flags & IFF_UP))
1369 list_move(&dev->unreg_list, &tmp_list);
1371 __dev_close_many(head);
1373 list_for_each_entry(dev, head, unreg_list) {
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1375 call_netdevice_notifiers(NETDEV_DOWN, dev);
1378 /* rollback_registered_many needs the complete original list */
1379 list_splice(&tmp_list, head);
1384 * dev_close - shutdown an interface.
1385 * @dev: device to shutdown
1387 * This function moves an active device into down state. A
1388 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1389 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1392 int dev_close(struct net_device *dev)
1394 if (dev->flags & IFF_UP) {
1397 /* Block netpoll rx while the interface is going down */
1398 netpoll_rx_disable(dev);
1400 list_add(&dev->unreg_list, &single);
1401 dev_close_many(&single);
1404 netpoll_rx_enable(dev);
1408 EXPORT_SYMBOL(dev_close);
1412 * dev_disable_lro - disable Large Receive Offload on a device
1415 * Disable Large Receive Offload (LRO) on a net device. Must be
1416 * called under RTNL. This is needed if received packets may be
1417 * forwarded to another interface.
1419 void dev_disable_lro(struct net_device *dev)
1422 * If we're trying to disable lro on a vlan device
1423 * use the underlying physical device instead
1425 if (is_vlan_dev(dev))
1426 dev = vlan_dev_real_dev(dev);
1428 dev->wanted_features &= ~NETIF_F_LRO;
1429 netdev_update_features(dev);
1431 if (unlikely(dev->features & NETIF_F_LRO))
1432 netdev_WARN(dev, "failed to disable LRO!\n");
1434 EXPORT_SYMBOL(dev_disable_lro);
1436 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1437 struct net_device *dev)
1439 struct netdev_notifier_info info;
1441 netdev_notifier_info_init(&info, dev);
1442 return nb->notifier_call(nb, val, &info);
1445 static int dev_boot_phase = 1;
1448 * register_netdevice_notifier - register a network notifier block
1451 * Register a notifier to be called when network device events occur.
1452 * The notifier passed is linked into the kernel structures and must
1453 * not be reused until it has been unregistered. A negative errno code
1454 * is returned on a failure.
1456 * When registered all registration and up events are replayed
1457 * to the new notifier to allow device to have a race free
1458 * view of the network device list.
1461 int register_netdevice_notifier(struct notifier_block *nb)
1463 struct net_device *dev;
1464 struct net_device *last;
1469 err = raw_notifier_chain_register(&netdev_chain, nb);
1475 for_each_netdev(net, dev) {
1476 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1477 err = notifier_to_errno(err);
1481 if (!(dev->flags & IFF_UP))
1484 call_netdevice_notifier(nb, NETDEV_UP, dev);
1495 for_each_netdev(net, dev) {
1499 if (dev->flags & IFF_UP) {
1500 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1502 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1504 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1509 raw_notifier_chain_unregister(&netdev_chain, nb);
1512 EXPORT_SYMBOL(register_netdevice_notifier);
1515 * unregister_netdevice_notifier - unregister a network notifier block
1518 * Unregister a notifier previously registered by
1519 * register_netdevice_notifier(). The notifier is unlinked into the
1520 * kernel structures and may then be reused. A negative errno code
1521 * is returned on a failure.
1523 * After unregistering unregister and down device events are synthesized
1524 * for all devices on the device list to the removed notifier to remove
1525 * the need for special case cleanup code.
1528 int unregister_netdevice_notifier(struct notifier_block *nb)
1530 struct net_device *dev;
1535 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1540 for_each_netdev(net, dev) {
1541 if (dev->flags & IFF_UP) {
1542 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1544 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1546 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1553 EXPORT_SYMBOL(unregister_netdevice_notifier);
1556 * call_netdevice_notifiers_info - call all network notifier blocks
1557 * @val: value passed unmodified to notifier function
1558 * @dev: net_device pointer passed unmodified to notifier function
1559 * @info: notifier information data
1561 * Call all network notifier blocks. Parameters and return value
1562 * are as for raw_notifier_call_chain().
1565 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1566 struct netdev_notifier_info *info)
1569 netdev_notifier_info_init(info, dev);
1570 return raw_notifier_call_chain(&netdev_chain, val, info);
1572 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1575 * call_netdevice_notifiers - call all network notifier blocks
1576 * @val: value passed unmodified to notifier function
1577 * @dev: net_device pointer passed unmodified to notifier function
1579 * Call all network notifier blocks. Parameters and return value
1580 * are as for raw_notifier_call_chain().
1583 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1585 struct netdev_notifier_info info;
1587 return call_netdevice_notifiers_info(val, dev, &info);
1589 EXPORT_SYMBOL(call_netdevice_notifiers);
1591 static struct static_key netstamp_needed __read_mostly;
1592 #ifdef HAVE_JUMP_LABEL
1593 /* We are not allowed to call static_key_slow_dec() from irq context
1594 * If net_disable_timestamp() is called from irq context, defer the
1595 * static_key_slow_dec() calls.
1597 static atomic_t netstamp_needed_deferred;
1600 void net_enable_timestamp(void)
1602 #ifdef HAVE_JUMP_LABEL
1603 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1607 static_key_slow_dec(&netstamp_needed);
1611 static_key_slow_inc(&netstamp_needed);
1613 EXPORT_SYMBOL(net_enable_timestamp);
1615 void net_disable_timestamp(void)
1617 #ifdef HAVE_JUMP_LABEL
1618 if (in_interrupt()) {
1619 atomic_inc(&netstamp_needed_deferred);
1623 static_key_slow_dec(&netstamp_needed);
1625 EXPORT_SYMBOL(net_disable_timestamp);
1627 static inline void net_timestamp_set(struct sk_buff *skb)
1629 skb->tstamp.tv64 = 0;
1630 if (static_key_false(&netstamp_needed))
1631 __net_timestamp(skb);
1634 #define net_timestamp_check(COND, SKB) \
1635 if (static_key_false(&netstamp_needed)) { \
1636 if ((COND) && !(SKB)->tstamp.tv64) \
1637 __net_timestamp(SKB); \
1640 static inline bool is_skb_forwardable(struct net_device *dev,
1641 struct sk_buff *skb)
1645 if (!(dev->flags & IFF_UP))
1648 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1649 if (skb->len <= len)
1652 /* if TSO is enabled, we don't care about the length as the packet
1653 * could be forwarded without being segmented before
1655 if (skb_is_gso(skb))
1662 * dev_forward_skb - loopback an skb to another netif
1664 * @dev: destination network device
1665 * @skb: buffer to forward
1668 * NET_RX_SUCCESS (no congestion)
1669 * NET_RX_DROP (packet was dropped, but freed)
1671 * dev_forward_skb can be used for injecting an skb from the
1672 * start_xmit function of one device into the receive queue
1673 * of another device.
1675 * The receiving device may be in another namespace, so
1676 * we have to clear all information in the skb that could
1677 * impact namespace isolation.
1679 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1681 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1682 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1683 atomic_long_inc(&dev->rx_dropped);
1689 if (unlikely(!is_skb_forwardable(dev, skb))) {
1690 atomic_long_inc(&dev->rx_dropped);
1694 skb_scrub_packet(skb);
1695 skb->protocol = eth_type_trans(skb, dev);
1697 /* eth_type_trans() can set pkt_type.
1698 * clear pkt_type _after_ calling eth_type_trans()
1700 skb->pkt_type = PACKET_HOST;
1702 return netif_rx(skb);
1704 EXPORT_SYMBOL_GPL(dev_forward_skb);
1706 static inline int deliver_skb(struct sk_buff *skb,
1707 struct packet_type *pt_prev,
1708 struct net_device *orig_dev)
1710 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1712 atomic_inc(&skb->users);
1713 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1716 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1718 if (!ptype->af_packet_priv || !skb->sk)
1721 if (ptype->id_match)
1722 return ptype->id_match(ptype, skb->sk);
1723 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1730 * Support routine. Sends outgoing frames to any network
1731 * taps currently in use.
1734 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1736 struct packet_type *ptype;
1737 struct sk_buff *skb2 = NULL;
1738 struct packet_type *pt_prev = NULL;
1741 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1742 /* Never send packets back to the socket
1743 * they originated from - MvS (miquels@drinkel.ow.org)
1745 if ((ptype->dev == dev || !ptype->dev) &&
1746 (!skb_loop_sk(ptype, skb))) {
1748 deliver_skb(skb2, pt_prev, skb->dev);
1753 skb2 = skb_clone(skb, GFP_ATOMIC);
1757 net_timestamp_set(skb2);
1759 /* skb->nh should be correctly
1760 set by sender, so that the second statement is
1761 just protection against buggy protocols.
1763 skb_reset_mac_header(skb2);
1765 if (skb_network_header(skb2) < skb2->data ||
1766 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1767 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1768 ntohs(skb2->protocol),
1770 skb_reset_network_header(skb2);
1773 skb2->transport_header = skb2->network_header;
1774 skb2->pkt_type = PACKET_OUTGOING;
1779 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1784 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1785 * @dev: Network device
1786 * @txq: number of queues available
1788 * If real_num_tx_queues is changed the tc mappings may no longer be
1789 * valid. To resolve this verify the tc mapping remains valid and if
1790 * not NULL the mapping. With no priorities mapping to this
1791 * offset/count pair it will no longer be used. In the worst case TC0
1792 * is invalid nothing can be done so disable priority mappings. If is
1793 * expected that drivers will fix this mapping if they can before
1794 * calling netif_set_real_num_tx_queues.
1796 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1799 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1801 /* If TC0 is invalidated disable TC mapping */
1802 if (tc->offset + tc->count > txq) {
1803 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1808 /* Invalidated prio to tc mappings set to TC0 */
1809 for (i = 1; i < TC_BITMASK + 1; i++) {
1810 int q = netdev_get_prio_tc_map(dev, i);
1812 tc = &dev->tc_to_txq[q];
1813 if (tc->offset + tc->count > txq) {
1814 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1816 netdev_set_prio_tc_map(dev, i, 0);
1822 static DEFINE_MUTEX(xps_map_mutex);
1823 #define xmap_dereference(P) \
1824 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1826 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1829 struct xps_map *map = NULL;
1833 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1835 for (pos = 0; map && pos < map->len; pos++) {
1836 if (map->queues[pos] == index) {
1838 map->queues[pos] = map->queues[--map->len];
1840 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1841 kfree_rcu(map, rcu);
1851 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1853 struct xps_dev_maps *dev_maps;
1855 bool active = false;
1857 mutex_lock(&xps_map_mutex);
1858 dev_maps = xmap_dereference(dev->xps_maps);
1863 for_each_possible_cpu(cpu) {
1864 for (i = index; i < dev->num_tx_queues; i++) {
1865 if (!remove_xps_queue(dev_maps, cpu, i))
1868 if (i == dev->num_tx_queues)
1873 RCU_INIT_POINTER(dev->xps_maps, NULL);
1874 kfree_rcu(dev_maps, rcu);
1877 for (i = index; i < dev->num_tx_queues; i++)
1878 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1882 mutex_unlock(&xps_map_mutex);
1885 static struct xps_map *expand_xps_map(struct xps_map *map,
1888 struct xps_map *new_map;
1889 int alloc_len = XPS_MIN_MAP_ALLOC;
1892 for (pos = 0; map && pos < map->len; pos++) {
1893 if (map->queues[pos] != index)
1898 /* Need to add queue to this CPU's existing map */
1900 if (pos < map->alloc_len)
1903 alloc_len = map->alloc_len * 2;
1906 /* Need to allocate new map to store queue on this CPU's map */
1907 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1912 for (i = 0; i < pos; i++)
1913 new_map->queues[i] = map->queues[i];
1914 new_map->alloc_len = alloc_len;
1920 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1922 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1923 struct xps_map *map, *new_map;
1924 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1925 int cpu, numa_node_id = -2;
1926 bool active = false;
1928 mutex_lock(&xps_map_mutex);
1930 dev_maps = xmap_dereference(dev->xps_maps);
1932 /* allocate memory for queue storage */
1933 for_each_online_cpu(cpu) {
1934 if (!cpumask_test_cpu(cpu, mask))
1938 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1939 if (!new_dev_maps) {
1940 mutex_unlock(&xps_map_mutex);
1944 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1947 map = expand_xps_map(map, cpu, index);
1951 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1955 goto out_no_new_maps;
1957 for_each_possible_cpu(cpu) {
1958 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1959 /* add queue to CPU maps */
1962 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1963 while ((pos < map->len) && (map->queues[pos] != index))
1966 if (pos == map->len)
1967 map->queues[map->len++] = index;
1969 if (numa_node_id == -2)
1970 numa_node_id = cpu_to_node(cpu);
1971 else if (numa_node_id != cpu_to_node(cpu))
1974 } else if (dev_maps) {
1975 /* fill in the new device map from the old device map */
1976 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1982 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1984 /* Cleanup old maps */
1986 for_each_possible_cpu(cpu) {
1987 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 if (map && map != new_map)
1990 kfree_rcu(map, rcu);
1993 kfree_rcu(dev_maps, rcu);
1996 dev_maps = new_dev_maps;
2000 /* update Tx queue numa node */
2001 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2002 (numa_node_id >= 0) ? numa_node_id :
2008 /* removes queue from unused CPUs */
2009 for_each_possible_cpu(cpu) {
2010 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2013 if (remove_xps_queue(dev_maps, cpu, index))
2017 /* free map if not active */
2019 RCU_INIT_POINTER(dev->xps_maps, NULL);
2020 kfree_rcu(dev_maps, rcu);
2024 mutex_unlock(&xps_map_mutex);
2028 /* remove any maps that we added */
2029 for_each_possible_cpu(cpu) {
2030 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2031 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2033 if (new_map && new_map != map)
2037 mutex_unlock(&xps_map_mutex);
2039 kfree(new_dev_maps);
2042 EXPORT_SYMBOL(netif_set_xps_queue);
2046 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2047 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2049 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2053 if (txq < 1 || txq > dev->num_tx_queues)
2056 if (dev->reg_state == NETREG_REGISTERED ||
2057 dev->reg_state == NETREG_UNREGISTERING) {
2060 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066 netif_setup_tc(dev, txq);
2068 if (txq < dev->real_num_tx_queues) {
2069 qdisc_reset_all_tx_gt(dev, txq);
2071 netif_reset_xps_queues_gt(dev, txq);
2076 dev->real_num_tx_queues = txq;
2079 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2083 * netif_set_real_num_rx_queues - set actual number of RX queues used
2084 * @dev: Network device
2085 * @rxq: Actual number of RX queues
2087 * This must be called either with the rtnl_lock held or before
2088 * registration of the net device. Returns 0 on success, or a
2089 * negative error code. If called before registration, it always
2092 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2096 if (rxq < 1 || rxq > dev->num_rx_queues)
2099 if (dev->reg_state == NETREG_REGISTERED) {
2102 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108 dev->real_num_rx_queues = rxq;
2111 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2115 * netif_get_num_default_rss_queues - default number of RSS queues
2117 * This routine should set an upper limit on the number of RSS queues
2118 * used by default by multiqueue devices.
2120 int netif_get_num_default_rss_queues(void)
2122 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2124 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2126 static inline void __netif_reschedule(struct Qdisc *q)
2128 struct softnet_data *sd;
2129 unsigned long flags;
2131 local_irq_save(flags);
2132 sd = &__get_cpu_var(softnet_data);
2133 q->next_sched = NULL;
2134 *sd->output_queue_tailp = q;
2135 sd->output_queue_tailp = &q->next_sched;
2136 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2137 local_irq_restore(flags);
2140 void __netif_schedule(struct Qdisc *q)
2142 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2143 __netif_reschedule(q);
2145 EXPORT_SYMBOL(__netif_schedule);
2147 void dev_kfree_skb_irq(struct sk_buff *skb)
2149 if (atomic_dec_and_test(&skb->users)) {
2150 struct softnet_data *sd;
2151 unsigned long flags;
2153 local_irq_save(flags);
2154 sd = &__get_cpu_var(softnet_data);
2155 skb->next = sd->completion_queue;
2156 sd->completion_queue = skb;
2157 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2158 local_irq_restore(flags);
2161 EXPORT_SYMBOL(dev_kfree_skb_irq);
2163 void dev_kfree_skb_any(struct sk_buff *skb)
2165 if (in_irq() || irqs_disabled())
2166 dev_kfree_skb_irq(skb);
2170 EXPORT_SYMBOL(dev_kfree_skb_any);
2174 * netif_device_detach - mark device as removed
2175 * @dev: network device
2177 * Mark device as removed from system and therefore no longer available.
2179 void netif_device_detach(struct net_device *dev)
2181 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2182 netif_running(dev)) {
2183 netif_tx_stop_all_queues(dev);
2186 EXPORT_SYMBOL(netif_device_detach);
2189 * netif_device_attach - mark device as attached
2190 * @dev: network device
2192 * Mark device as attached from system and restart if needed.
2194 void netif_device_attach(struct net_device *dev)
2196 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2197 netif_running(dev)) {
2198 netif_tx_wake_all_queues(dev);
2199 __netdev_watchdog_up(dev);
2202 EXPORT_SYMBOL(netif_device_attach);
2204 static void skb_warn_bad_offload(const struct sk_buff *skb)
2206 static const netdev_features_t null_features = 0;
2207 struct net_device *dev = skb->dev;
2208 const char *driver = "";
2210 if (!net_ratelimit())
2213 if (dev && dev->dev.parent)
2214 driver = dev_driver_string(dev->dev.parent);
2216 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2217 "gso_type=%d ip_summed=%d\n",
2218 driver, dev ? &dev->features : &null_features,
2219 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2220 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2221 skb_shinfo(skb)->gso_type, skb->ip_summed);
2225 * Invalidate hardware checksum when packet is to be mangled, and
2226 * complete checksum manually on outgoing path.
2228 int skb_checksum_help(struct sk_buff *skb)
2231 int ret = 0, offset;
2233 if (skb->ip_summed == CHECKSUM_COMPLETE)
2234 goto out_set_summed;
2236 if (unlikely(skb_shinfo(skb)->gso_size)) {
2237 skb_warn_bad_offload(skb);
2241 /* Before computing a checksum, we should make sure no frag could
2242 * be modified by an external entity : checksum could be wrong.
2244 if (skb_has_shared_frag(skb)) {
2245 ret = __skb_linearize(skb);
2250 offset = skb_checksum_start_offset(skb);
2251 BUG_ON(offset >= skb_headlen(skb));
2252 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2254 offset += skb->csum_offset;
2255 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2257 if (skb_cloned(skb) &&
2258 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2259 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2264 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2266 skb->ip_summed = CHECKSUM_NONE;
2270 EXPORT_SYMBOL(skb_checksum_help);
2272 __be16 skb_network_protocol(struct sk_buff *skb)
2274 __be16 type = skb->protocol;
2275 int vlan_depth = ETH_HLEN;
2277 /* Tunnel gso handlers can set protocol to ethernet. */
2278 if (type == htons(ETH_P_TEB)) {
2281 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2284 eth = (struct ethhdr *)skb_mac_header(skb);
2285 type = eth->h_proto;
2288 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2289 struct vlan_hdr *vh;
2291 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2294 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2295 type = vh->h_vlan_encapsulated_proto;
2296 vlan_depth += VLAN_HLEN;
2303 * skb_mac_gso_segment - mac layer segmentation handler.
2304 * @skb: buffer to segment
2305 * @features: features for the output path (see dev->features)
2307 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2308 netdev_features_t features)
2310 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2311 struct packet_offload *ptype;
2312 __be16 type = skb_network_protocol(skb);
2314 if (unlikely(!type))
2315 return ERR_PTR(-EINVAL);
2317 __skb_pull(skb, skb->mac_len);
2320 list_for_each_entry_rcu(ptype, &offload_base, list) {
2321 if (ptype->type == type && ptype->callbacks.gso_segment) {
2322 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2325 err = ptype->callbacks.gso_send_check(skb);
2326 segs = ERR_PTR(err);
2327 if (err || skb_gso_ok(skb, features))
2329 __skb_push(skb, (skb->data -
2330 skb_network_header(skb)));
2332 segs = ptype->callbacks.gso_segment(skb, features);
2338 __skb_push(skb, skb->data - skb_mac_header(skb));
2342 EXPORT_SYMBOL(skb_mac_gso_segment);
2345 /* openvswitch calls this on rx path, so we need a different check.
2347 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2350 return skb->ip_summed != CHECKSUM_PARTIAL;
2352 return skb->ip_summed == CHECKSUM_NONE;
2356 * __skb_gso_segment - Perform segmentation on skb.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2359 * @tx_path: whether it is called in TX path
2361 * This function segments the given skb and returns a list of segments.
2363 * It may return NULL if the skb requires no segmentation. This is
2364 * only possible when GSO is used for verifying header integrity.
2366 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2367 netdev_features_t features, bool tx_path)
2369 if (unlikely(skb_needs_check(skb, tx_path))) {
2372 skb_warn_bad_offload(skb);
2374 if (skb_header_cloned(skb) &&
2375 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2376 return ERR_PTR(err);
2379 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2380 skb_reset_mac_header(skb);
2381 skb_reset_mac_len(skb);
2383 return skb_mac_gso_segment(skb, features);
2385 EXPORT_SYMBOL(__skb_gso_segment);
2387 /* Take action when hardware reception checksum errors are detected. */
2389 void netdev_rx_csum_fault(struct net_device *dev)
2391 if (net_ratelimit()) {
2392 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2396 EXPORT_SYMBOL(netdev_rx_csum_fault);
2399 /* Actually, we should eliminate this check as soon as we know, that:
2400 * 1. IOMMU is present and allows to map all the memory.
2401 * 2. No high memory really exists on this machine.
2404 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2406 #ifdef CONFIG_HIGHMEM
2408 if (!(dev->features & NETIF_F_HIGHDMA)) {
2409 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2410 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2411 if (PageHighMem(skb_frag_page(frag)))
2416 if (PCI_DMA_BUS_IS_PHYS) {
2417 struct device *pdev = dev->dev.parent;
2421 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2422 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2423 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2424 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2433 void (*destructor)(struct sk_buff *skb);
2436 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2438 static void dev_gso_skb_destructor(struct sk_buff *skb)
2440 struct dev_gso_cb *cb;
2443 struct sk_buff *nskb = skb->next;
2445 skb->next = nskb->next;
2448 } while (skb->next);
2450 cb = DEV_GSO_CB(skb);
2452 cb->destructor(skb);
2456 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2457 * @skb: buffer to segment
2458 * @features: device features as applicable to this skb
2460 * This function segments the given skb and stores the list of segments
2463 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2465 struct sk_buff *segs;
2467 segs = skb_gso_segment(skb, features);
2469 /* Verifying header integrity only. */
2474 return PTR_ERR(segs);
2477 DEV_GSO_CB(skb)->destructor = skb->destructor;
2478 skb->destructor = dev_gso_skb_destructor;
2483 static netdev_features_t harmonize_features(struct sk_buff *skb,
2484 __be16 protocol, netdev_features_t features)
2486 if (skb->ip_summed != CHECKSUM_NONE &&
2487 !can_checksum_protocol(features, protocol)) {
2488 features &= ~NETIF_F_ALL_CSUM;
2489 } else if (illegal_highdma(skb->dev, skb)) {
2490 features &= ~NETIF_F_SG;
2496 netdev_features_t netif_skb_features(struct sk_buff *skb)
2498 __be16 protocol = skb->protocol;
2499 netdev_features_t features = skb->dev->features;
2501 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2502 features &= ~NETIF_F_GSO_MASK;
2504 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2505 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2506 protocol = veh->h_vlan_encapsulated_proto;
2507 } else if (!vlan_tx_tag_present(skb)) {
2508 return harmonize_features(skb, protocol, features);
2511 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2512 NETIF_F_HW_VLAN_STAG_TX);
2514 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2515 return harmonize_features(skb, protocol, features);
2517 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2518 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2519 NETIF_F_HW_VLAN_STAG_TX;
2520 return harmonize_features(skb, protocol, features);
2523 EXPORT_SYMBOL(netif_skb_features);
2526 * Returns true if either:
2527 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2528 * 2. skb is fragmented and the device does not support SG.
2530 static inline int skb_needs_linearize(struct sk_buff *skb,
2531 netdev_features_t features)
2533 return skb_is_nonlinear(skb) &&
2534 ((skb_has_frag_list(skb) &&
2535 !(features & NETIF_F_FRAGLIST)) ||
2536 (skb_shinfo(skb)->nr_frags &&
2537 !(features & NETIF_F_SG)));
2540 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2541 struct netdev_queue *txq)
2543 const struct net_device_ops *ops = dev->netdev_ops;
2544 int rc = NETDEV_TX_OK;
2545 unsigned int skb_len;
2547 if (likely(!skb->next)) {
2548 netdev_features_t features;
2551 * If device doesn't need skb->dst, release it right now while
2552 * its hot in this cpu cache
2554 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2557 features = netif_skb_features(skb);
2559 if (vlan_tx_tag_present(skb) &&
2560 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2561 skb = __vlan_put_tag(skb, skb->vlan_proto,
2562 vlan_tx_tag_get(skb));
2569 /* If encapsulation offload request, verify we are testing
2570 * hardware encapsulation features instead of standard
2571 * features for the netdev
2573 if (skb->encapsulation)
2574 features &= dev->hw_enc_features;
2576 if (netif_needs_gso(skb, features)) {
2577 if (unlikely(dev_gso_segment(skb, features)))
2582 if (skb_needs_linearize(skb, features) &&
2583 __skb_linearize(skb))
2586 /* If packet is not checksummed and device does not
2587 * support checksumming for this protocol, complete
2588 * checksumming here.
2590 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2591 if (skb->encapsulation)
2592 skb_set_inner_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2595 skb_set_transport_header(skb,
2596 skb_checksum_start_offset(skb));
2597 if (!(features & NETIF_F_ALL_CSUM) &&
2598 skb_checksum_help(skb))
2603 if (!list_empty(&ptype_all))
2604 dev_queue_xmit_nit(skb, dev);
2607 rc = ops->ndo_start_xmit(skb, dev);
2608 trace_net_dev_xmit(skb, rc, dev, skb_len);
2609 if (rc == NETDEV_TX_OK)
2610 txq_trans_update(txq);
2616 struct sk_buff *nskb = skb->next;
2618 skb->next = nskb->next;
2621 if (!list_empty(&ptype_all))
2622 dev_queue_xmit_nit(nskb, dev);
2624 skb_len = nskb->len;
2625 rc = ops->ndo_start_xmit(nskb, dev);
2626 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2627 if (unlikely(rc != NETDEV_TX_OK)) {
2628 if (rc & ~NETDEV_TX_MASK)
2629 goto out_kfree_gso_skb;
2630 nskb->next = skb->next;
2634 txq_trans_update(txq);
2635 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2636 return NETDEV_TX_BUSY;
2637 } while (skb->next);
2640 if (likely(skb->next == NULL)) {
2641 skb->destructor = DEV_GSO_CB(skb)->destructor;
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2784 * Queue a buffer for transmission to a network device. The caller must
2785 * have set the device and priority and built the buffer before calling
2786 * this function. The function can be called from an interrupt.
2788 * A negative errno code is returned on a failure. A success does not
2789 * guarantee the frame will be transmitted as it may be dropped due
2790 * to congestion or traffic shaping.
2792 * -----------------------------------------------------------------------------------
2793 * I notice this method can also return errors from the queue disciplines,
2794 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2797 * Regardless of the return value, the skb is consumed, so it is currently
2798 * difficult to retry a send to this method. (You can bump the ref count
2799 * before sending to hold a reference for retry if you are careful.)
2801 * When calling this method, interrupts MUST be enabled. This is because
2802 * the BH enable code must have IRQs enabled so that it will not deadlock.
2805 int dev_queue_xmit(struct sk_buff *skb)
2807 struct net_device *dev = skb->dev;
2808 struct netdev_queue *txq;
2812 skb_reset_mac_header(skb);
2814 /* Disable soft irqs for various locks below. Also
2815 * stops preemption for RCU.
2819 skb_update_prio(skb);
2821 txq = netdev_pick_tx(dev, skb);
2822 q = rcu_dereference_bh(txq->qdisc);
2824 #ifdef CONFIG_NET_CLS_ACT
2825 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2827 trace_net_dev_queue(skb);
2829 rc = __dev_xmit_skb(skb, q, dev, txq);
2833 /* The device has no queue. Common case for software devices:
2834 loopback, all the sorts of tunnels...
2836 Really, it is unlikely that netif_tx_lock protection is necessary
2837 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2839 However, it is possible, that they rely on protection
2842 Check this and shot the lock. It is not prone from deadlocks.
2843 Either shot noqueue qdisc, it is even simpler 8)
2845 if (dev->flags & IFF_UP) {
2846 int cpu = smp_processor_id(); /* ok because BHs are off */
2848 if (txq->xmit_lock_owner != cpu) {
2850 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2851 goto recursion_alert;
2853 HARD_TX_LOCK(dev, txq, cpu);
2855 if (!netif_xmit_stopped(txq)) {
2856 __this_cpu_inc(xmit_recursion);
2857 rc = dev_hard_start_xmit(skb, dev, txq);
2858 __this_cpu_dec(xmit_recursion);
2859 if (dev_xmit_complete(rc)) {
2860 HARD_TX_UNLOCK(dev, txq);
2864 HARD_TX_UNLOCK(dev, txq);
2865 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2868 /* Recursion is detected! It is possible,
2872 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2878 rcu_read_unlock_bh();
2883 rcu_read_unlock_bh();
2886 EXPORT_SYMBOL(dev_queue_xmit);
2889 /*=======================================================================
2891 =======================================================================*/
2893 int netdev_max_backlog __read_mostly = 1000;
2894 EXPORT_SYMBOL(netdev_max_backlog);
2896 int netdev_tstamp_prequeue __read_mostly = 1;
2897 int netdev_budget __read_mostly = 300;
2898 int weight_p __read_mostly = 64; /* old backlog weight */
2900 /* Called with irq disabled */
2901 static inline void ____napi_schedule(struct softnet_data *sd,
2902 struct napi_struct *napi)
2904 list_add_tail(&napi->poll_list, &sd->poll_list);
2905 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2910 /* One global table that all flow-based protocols share. */
2911 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2912 EXPORT_SYMBOL(rps_sock_flow_table);
2914 struct static_key rps_needed __read_mostly;
2916 static struct rps_dev_flow *
2917 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2918 struct rps_dev_flow *rflow, u16 next_cpu)
2920 if (next_cpu != RPS_NO_CPU) {
2921 #ifdef CONFIG_RFS_ACCEL
2922 struct netdev_rx_queue *rxqueue;
2923 struct rps_dev_flow_table *flow_table;
2924 struct rps_dev_flow *old_rflow;
2929 /* Should we steer this flow to a different hardware queue? */
2930 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2931 !(dev->features & NETIF_F_NTUPLE))
2933 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2934 if (rxq_index == skb_get_rx_queue(skb))
2937 rxqueue = dev->_rx + rxq_index;
2938 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2941 flow_id = skb->rxhash & flow_table->mask;
2942 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2943 rxq_index, flow_id);
2947 rflow = &flow_table->flows[flow_id];
2949 if (old_rflow->filter == rflow->filter)
2950 old_rflow->filter = RPS_NO_FILTER;
2954 per_cpu(softnet_data, next_cpu).input_queue_head;
2957 rflow->cpu = next_cpu;
2962 * get_rps_cpu is called from netif_receive_skb and returns the target
2963 * CPU from the RPS map of the receiving queue for a given skb.
2964 * rcu_read_lock must be held on entry.
2966 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2967 struct rps_dev_flow **rflowp)
2969 struct netdev_rx_queue *rxqueue;
2970 struct rps_map *map;
2971 struct rps_dev_flow_table *flow_table;
2972 struct rps_sock_flow_table *sock_flow_table;
2976 if (skb_rx_queue_recorded(skb)) {
2977 u16 index = skb_get_rx_queue(skb);
2978 if (unlikely(index >= dev->real_num_rx_queues)) {
2979 WARN_ONCE(dev->real_num_rx_queues > 1,
2980 "%s received packet on queue %u, but number "
2981 "of RX queues is %u\n",
2982 dev->name, index, dev->real_num_rx_queues);
2985 rxqueue = dev->_rx + index;
2989 map = rcu_dereference(rxqueue->rps_map);
2991 if (map->len == 1 &&
2992 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2993 tcpu = map->cpus[0];
2994 if (cpu_online(tcpu))
2998 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3002 skb_reset_network_header(skb);
3003 if (!skb_get_rxhash(skb))
3006 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3007 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3008 if (flow_table && sock_flow_table) {
3010 struct rps_dev_flow *rflow;
3012 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3015 next_cpu = sock_flow_table->ents[skb->rxhash &
3016 sock_flow_table->mask];
3019 * If the desired CPU (where last recvmsg was done) is
3020 * different from current CPU (one in the rx-queue flow
3021 * table entry), switch if one of the following holds:
3022 * - Current CPU is unset (equal to RPS_NO_CPU).
3023 * - Current CPU is offline.
3024 * - The current CPU's queue tail has advanced beyond the
3025 * last packet that was enqueued using this table entry.
3026 * This guarantees that all previous packets for the flow
3027 * have been dequeued, thus preserving in order delivery.
3029 if (unlikely(tcpu != next_cpu) &&
3030 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3031 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3032 rflow->last_qtail)) >= 0)) {
3034 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3037 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3045 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3047 if (cpu_online(tcpu)) {
3057 #ifdef CONFIG_RFS_ACCEL
3060 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3061 * @dev: Device on which the filter was set
3062 * @rxq_index: RX queue index
3063 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3064 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3066 * Drivers that implement ndo_rx_flow_steer() should periodically call
3067 * this function for each installed filter and remove the filters for
3068 * which it returns %true.
3070 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3071 u32 flow_id, u16 filter_id)
3073 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3074 struct rps_dev_flow_table *flow_table;
3075 struct rps_dev_flow *rflow;
3080 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3081 if (flow_table && flow_id <= flow_table->mask) {
3082 rflow = &flow_table->flows[flow_id];
3083 cpu = ACCESS_ONCE(rflow->cpu);
3084 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3085 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3086 rflow->last_qtail) <
3087 (int)(10 * flow_table->mask)))
3093 EXPORT_SYMBOL(rps_may_expire_flow);
3095 #endif /* CONFIG_RFS_ACCEL */
3097 /* Called from hardirq (IPI) context */
3098 static void rps_trigger_softirq(void *data)
3100 struct softnet_data *sd = data;
3102 ____napi_schedule(sd, &sd->backlog);
3106 #endif /* CONFIG_RPS */
3109 * Check if this softnet_data structure is another cpu one
3110 * If yes, queue it to our IPI list and return 1
3113 static int rps_ipi_queued(struct softnet_data *sd)
3116 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3119 sd->rps_ipi_next = mysd->rps_ipi_list;
3120 mysd->rps_ipi_list = sd;
3122 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3125 #endif /* CONFIG_RPS */
3129 #ifdef CONFIG_NET_FLOW_LIMIT
3130 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3133 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3135 #ifdef CONFIG_NET_FLOW_LIMIT
3136 struct sd_flow_limit *fl;
3137 struct softnet_data *sd;
3138 unsigned int old_flow, new_flow;
3140 if (qlen < (netdev_max_backlog >> 1))
3143 sd = &__get_cpu_var(softnet_data);
3146 fl = rcu_dereference(sd->flow_limit);
3148 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3149 old_flow = fl->history[fl->history_head];
3150 fl->history[fl->history_head] = new_flow;
3153 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3155 if (likely(fl->buckets[old_flow]))
3156 fl->buckets[old_flow]--;
3158 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3170 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3171 * queue (may be a remote CPU queue).
3173 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3174 unsigned int *qtail)
3176 struct softnet_data *sd;
3177 unsigned long flags;
3180 sd = &per_cpu(softnet_data, cpu);
3182 local_irq_save(flags);
3185 qlen = skb_queue_len(&sd->input_pkt_queue);
3186 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3187 if (skb_queue_len(&sd->input_pkt_queue)) {
3189 __skb_queue_tail(&sd->input_pkt_queue, skb);
3190 input_queue_tail_incr_save(sd, qtail);
3192 local_irq_restore(flags);
3193 return NET_RX_SUCCESS;
3196 /* Schedule NAPI for backlog device
3197 * We can use non atomic operation since we own the queue lock
3199 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3200 if (!rps_ipi_queued(sd))
3201 ____napi_schedule(sd, &sd->backlog);
3209 local_irq_restore(flags);
3211 atomic_long_inc(&skb->dev->rx_dropped);
3217 * netif_rx - post buffer to the network code
3218 * @skb: buffer to post
3220 * This function receives a packet from a device driver and queues it for
3221 * the upper (protocol) levels to process. It always succeeds. The buffer
3222 * may be dropped during processing for congestion control or by the
3226 * NET_RX_SUCCESS (no congestion)
3227 * NET_RX_DROP (packet was dropped)
3231 int netif_rx(struct sk_buff *skb)
3235 /* if netpoll wants it, pretend we never saw it */
3236 if (netpoll_rx(skb))
3239 net_timestamp_check(netdev_tstamp_prequeue, skb);
3241 trace_netif_rx(skb);
3243 if (static_key_false(&rps_needed)) {
3244 struct rps_dev_flow voidflow, *rflow = &voidflow;
3250 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3252 cpu = smp_processor_id();
3254 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3262 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3267 EXPORT_SYMBOL(netif_rx);
3269 int netif_rx_ni(struct sk_buff *skb)
3274 err = netif_rx(skb);
3275 if (local_softirq_pending())
3281 EXPORT_SYMBOL(netif_rx_ni);
3283 static void net_tx_action(struct softirq_action *h)
3285 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3287 if (sd->completion_queue) {
3288 struct sk_buff *clist;
3290 local_irq_disable();
3291 clist = sd->completion_queue;
3292 sd->completion_queue = NULL;
3296 struct sk_buff *skb = clist;
3297 clist = clist->next;
3299 WARN_ON(atomic_read(&skb->users));
3300 trace_kfree_skb(skb, net_tx_action);
3305 if (sd->output_queue) {
3308 local_irq_disable();
3309 head = sd->output_queue;
3310 sd->output_queue = NULL;
3311 sd->output_queue_tailp = &sd->output_queue;
3315 struct Qdisc *q = head;
3316 spinlock_t *root_lock;
3318 head = head->next_sched;
3320 root_lock = qdisc_lock(q);
3321 if (spin_trylock(root_lock)) {
3322 smp_mb__before_clear_bit();
3323 clear_bit(__QDISC_STATE_SCHED,
3326 spin_unlock(root_lock);
3328 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3330 __netif_reschedule(q);
3332 smp_mb__before_clear_bit();
3333 clear_bit(__QDISC_STATE_SCHED,
3341 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3342 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3343 /* This hook is defined here for ATM LANE */
3344 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3345 unsigned char *addr) __read_mostly;
3346 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3349 #ifdef CONFIG_NET_CLS_ACT
3350 /* TODO: Maybe we should just force sch_ingress to be compiled in
3351 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3352 * a compare and 2 stores extra right now if we dont have it on
3353 * but have CONFIG_NET_CLS_ACT
3354 * NOTE: This doesn't stop any functionality; if you dont have
3355 * the ingress scheduler, you just can't add policies on ingress.
3358 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3360 struct net_device *dev = skb->dev;
3361 u32 ttl = G_TC_RTTL(skb->tc_verd);
3362 int result = TC_ACT_OK;
3365 if (unlikely(MAX_RED_LOOP < ttl++)) {
3366 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3367 skb->skb_iif, dev->ifindex);
3371 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3372 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3375 if (q != &noop_qdisc) {
3376 spin_lock(qdisc_lock(q));
3377 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3378 result = qdisc_enqueue_root(skb, q);
3379 spin_unlock(qdisc_lock(q));
3385 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3386 struct packet_type **pt_prev,
3387 int *ret, struct net_device *orig_dev)
3389 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3391 if (!rxq || rxq->qdisc == &noop_qdisc)
3395 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3399 switch (ing_filter(skb, rxq)) {
3413 * netdev_rx_handler_register - register receive handler
3414 * @dev: device to register a handler for
3415 * @rx_handler: receive handler to register
3416 * @rx_handler_data: data pointer that is used by rx handler
3418 * Register a receive hander for a device. This handler will then be
3419 * called from __netif_receive_skb. A negative errno code is returned
3422 * The caller must hold the rtnl_mutex.
3424 * For a general description of rx_handler, see enum rx_handler_result.
3426 int netdev_rx_handler_register(struct net_device *dev,
3427 rx_handler_func_t *rx_handler,
3428 void *rx_handler_data)
3432 if (dev->rx_handler)
3435 /* Note: rx_handler_data must be set before rx_handler */
3436 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3437 rcu_assign_pointer(dev->rx_handler, rx_handler);
3441 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3444 * netdev_rx_handler_unregister - unregister receive handler
3445 * @dev: device to unregister a handler from
3447 * Unregister a receive handler from a device.
3449 * The caller must hold the rtnl_mutex.
3451 void netdev_rx_handler_unregister(struct net_device *dev)
3455 RCU_INIT_POINTER(dev->rx_handler, NULL);
3456 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3457 * section has a guarantee to see a non NULL rx_handler_data
3461 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3463 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3466 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3467 * the special handling of PFMEMALLOC skbs.
3469 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3471 switch (skb->protocol) {
3472 case __constant_htons(ETH_P_ARP):
3473 case __constant_htons(ETH_P_IP):
3474 case __constant_htons(ETH_P_IPV6):
3475 case __constant_htons(ETH_P_8021Q):
3476 case __constant_htons(ETH_P_8021AD):
3483 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3485 struct packet_type *ptype, *pt_prev;
3486 rx_handler_func_t *rx_handler;
3487 struct net_device *orig_dev;
3488 struct net_device *null_or_dev;
3489 bool deliver_exact = false;
3490 int ret = NET_RX_DROP;
3493 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3495 trace_netif_receive_skb(skb);
3497 /* if we've gotten here through NAPI, check netpoll */
3498 if (netpoll_receive_skb(skb))
3501 orig_dev = skb->dev;
3503 skb_reset_network_header(skb);
3504 if (!skb_transport_header_was_set(skb))
3505 skb_reset_transport_header(skb);
3506 skb_reset_mac_len(skb);
3513 skb->skb_iif = skb->dev->ifindex;
3515 __this_cpu_inc(softnet_data.processed);
3517 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3518 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3519 skb = vlan_untag(skb);
3524 #ifdef CONFIG_NET_CLS_ACT
3525 if (skb->tc_verd & TC_NCLS) {
3526 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3534 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3535 if (!ptype->dev || ptype->dev == skb->dev) {
3537 ret = deliver_skb(skb, pt_prev, orig_dev);
3543 #ifdef CONFIG_NET_CLS_ACT
3544 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3550 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3553 if (vlan_tx_tag_present(skb)) {
3555 ret = deliver_skb(skb, pt_prev, orig_dev);
3558 if (vlan_do_receive(&skb))
3560 else if (unlikely(!skb))
3564 rx_handler = rcu_dereference(skb->dev->rx_handler);
3567 ret = deliver_skb(skb, pt_prev, orig_dev);
3570 switch (rx_handler(&skb)) {
3571 case RX_HANDLER_CONSUMED:
3572 ret = NET_RX_SUCCESS;
3574 case RX_HANDLER_ANOTHER:
3576 case RX_HANDLER_EXACT:
3577 deliver_exact = true;
3578 case RX_HANDLER_PASS:
3585 if (vlan_tx_nonzero_tag_present(skb))
3586 skb->pkt_type = PACKET_OTHERHOST;
3588 /* deliver only exact match when indicated */
3589 null_or_dev = deliver_exact ? skb->dev : NULL;
3591 type = skb->protocol;
3592 list_for_each_entry_rcu(ptype,
3593 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3594 if (ptype->type == type &&
3595 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3596 ptype->dev == orig_dev)) {
3598 ret = deliver_skb(skb, pt_prev, orig_dev);
3604 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3607 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3610 atomic_long_inc(&skb->dev->rx_dropped);
3612 /* Jamal, now you will not able to escape explaining
3613 * me how you were going to use this. :-)
3624 static int __netif_receive_skb(struct sk_buff *skb)
3628 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3629 unsigned long pflags = current->flags;
3632 * PFMEMALLOC skbs are special, they should
3633 * - be delivered to SOCK_MEMALLOC sockets only
3634 * - stay away from userspace
3635 * - have bounded memory usage
3637 * Use PF_MEMALLOC as this saves us from propagating the allocation
3638 * context down to all allocation sites.
3640 current->flags |= PF_MEMALLOC;
3641 ret = __netif_receive_skb_core(skb, true);
3642 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3644 ret = __netif_receive_skb_core(skb, false);
3650 * netif_receive_skb - process receive buffer from network
3651 * @skb: buffer to process
3653 * netif_receive_skb() is the main receive data processing function.
3654 * It always succeeds. The buffer may be dropped during processing
3655 * for congestion control or by the protocol layers.
3657 * This function may only be called from softirq context and interrupts
3658 * should be enabled.
3660 * Return values (usually ignored):
3661 * NET_RX_SUCCESS: no congestion
3662 * NET_RX_DROP: packet was dropped
3664 int netif_receive_skb(struct sk_buff *skb)
3666 net_timestamp_check(netdev_tstamp_prequeue, skb);
3668 if (skb_defer_rx_timestamp(skb))
3669 return NET_RX_SUCCESS;
3672 if (static_key_false(&rps_needed)) {
3673 struct rps_dev_flow voidflow, *rflow = &voidflow;
3678 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3681 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3688 return __netif_receive_skb(skb);
3690 EXPORT_SYMBOL(netif_receive_skb);
3692 /* Network device is going away, flush any packets still pending
3693 * Called with irqs disabled.
3695 static void flush_backlog(void *arg)
3697 struct net_device *dev = arg;
3698 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3699 struct sk_buff *skb, *tmp;
3702 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3703 if (skb->dev == dev) {
3704 __skb_unlink(skb, &sd->input_pkt_queue);
3706 input_queue_head_incr(sd);
3711 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3712 if (skb->dev == dev) {
3713 __skb_unlink(skb, &sd->process_queue);
3715 input_queue_head_incr(sd);
3720 static int napi_gro_complete(struct sk_buff *skb)
3722 struct packet_offload *ptype;
3723 __be16 type = skb->protocol;
3724 struct list_head *head = &offload_base;
3727 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3729 if (NAPI_GRO_CB(skb)->count == 1) {
3730 skb_shinfo(skb)->gso_size = 0;
3735 list_for_each_entry_rcu(ptype, head, list) {
3736 if (ptype->type != type || !ptype->callbacks.gro_complete)
3739 err = ptype->callbacks.gro_complete(skb);
3745 WARN_ON(&ptype->list == head);
3747 return NET_RX_SUCCESS;
3751 return netif_receive_skb(skb);
3754 /* napi->gro_list contains packets ordered by age.
3755 * youngest packets at the head of it.
3756 * Complete skbs in reverse order to reduce latencies.
3758 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3760 struct sk_buff *skb, *prev = NULL;
3762 /* scan list and build reverse chain */
3763 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3768 for (skb = prev; skb; skb = prev) {
3771 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3775 napi_gro_complete(skb);
3779 napi->gro_list = NULL;
3781 EXPORT_SYMBOL(napi_gro_flush);
3783 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3786 unsigned int maclen = skb->dev->hard_header_len;
3788 for (p = napi->gro_list; p; p = p->next) {
3789 unsigned long diffs;
3791 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3792 diffs |= p->vlan_tci ^ skb->vlan_tci;
3793 if (maclen == ETH_HLEN)
3794 diffs |= compare_ether_header(skb_mac_header(p),
3795 skb_gro_mac_header(skb));
3797 diffs = memcmp(skb_mac_header(p),
3798 skb_gro_mac_header(skb),
3800 NAPI_GRO_CB(p)->same_flow = !diffs;
3801 NAPI_GRO_CB(p)->flush = 0;
3805 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3807 struct sk_buff **pp = NULL;
3808 struct packet_offload *ptype;
3809 __be16 type = skb->protocol;
3810 struct list_head *head = &offload_base;
3812 enum gro_result ret;
3814 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3817 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3820 gro_list_prepare(napi, skb);
3823 list_for_each_entry_rcu(ptype, head, list) {
3824 if (ptype->type != type || !ptype->callbacks.gro_receive)
3827 skb_set_network_header(skb, skb_gro_offset(skb));
3828 skb_reset_mac_len(skb);
3829 NAPI_GRO_CB(skb)->same_flow = 0;
3830 NAPI_GRO_CB(skb)->flush = 0;
3831 NAPI_GRO_CB(skb)->free = 0;
3833 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3838 if (&ptype->list == head)
3841 same_flow = NAPI_GRO_CB(skb)->same_flow;
3842 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3845 struct sk_buff *nskb = *pp;
3849 napi_gro_complete(nskb);
3856 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3860 NAPI_GRO_CB(skb)->count = 1;
3861 NAPI_GRO_CB(skb)->age = jiffies;
3862 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3863 skb->next = napi->gro_list;
3864 napi->gro_list = skb;
3868 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3869 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3871 BUG_ON(skb->end - skb->tail < grow);
3873 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3876 skb->data_len -= grow;
3878 skb_shinfo(skb)->frags[0].page_offset += grow;
3879 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3881 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3882 skb_frag_unref(skb, 0);
3883 memmove(skb_shinfo(skb)->frags,
3884 skb_shinfo(skb)->frags + 1,
3885 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3898 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3902 if (netif_receive_skb(skb))
3910 case GRO_MERGED_FREE:
3911 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3912 kmem_cache_free(skbuff_head_cache, skb);
3925 static void skb_gro_reset_offset(struct sk_buff *skb)
3927 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3928 const skb_frag_t *frag0 = &pinfo->frags[0];
3930 NAPI_GRO_CB(skb)->data_offset = 0;
3931 NAPI_GRO_CB(skb)->frag0 = NULL;
3932 NAPI_GRO_CB(skb)->frag0_len = 0;
3934 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3936 !PageHighMem(skb_frag_page(frag0))) {
3937 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3938 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3942 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3944 skb_gro_reset_offset(skb);
3946 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3948 EXPORT_SYMBOL(napi_gro_receive);
3950 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3952 __skb_pull(skb, skb_headlen(skb));
3953 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3954 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3956 skb->dev = napi->dev;
3962 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3964 struct sk_buff *skb = napi->skb;
3967 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3973 EXPORT_SYMBOL(napi_get_frags);
3975 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3981 skb->protocol = eth_type_trans(skb, skb->dev);
3983 if (ret == GRO_HELD)
3984 skb_gro_pull(skb, -ETH_HLEN);
3985 else if (netif_receive_skb(skb))
3990 case GRO_MERGED_FREE:
3991 napi_reuse_skb(napi, skb);
4001 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4003 struct sk_buff *skb = napi->skb;
4010 skb_reset_mac_header(skb);
4011 skb_gro_reset_offset(skb);
4013 off = skb_gro_offset(skb);
4014 hlen = off + sizeof(*eth);
4015 eth = skb_gro_header_fast(skb, off);
4016 if (skb_gro_header_hard(skb, hlen)) {
4017 eth = skb_gro_header_slow(skb, hlen, off);
4018 if (unlikely(!eth)) {
4019 napi_reuse_skb(napi, skb);
4025 skb_gro_pull(skb, sizeof(*eth));
4028 * This works because the only protocols we care about don't require
4029 * special handling. We'll fix it up properly at the end.
4031 skb->protocol = eth->h_proto;
4037 gro_result_t napi_gro_frags(struct napi_struct *napi)
4039 struct sk_buff *skb = napi_frags_skb(napi);
4044 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4046 EXPORT_SYMBOL(napi_gro_frags);
4049 * net_rps_action sends any pending IPI's for rps.
4050 * Note: called with local irq disabled, but exits with local irq enabled.
4052 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4055 struct softnet_data *remsd = sd->rps_ipi_list;
4058 sd->rps_ipi_list = NULL;
4062 /* Send pending IPI's to kick RPS processing on remote cpus. */
4064 struct softnet_data *next = remsd->rps_ipi_next;
4066 if (cpu_online(remsd->cpu))
4067 __smp_call_function_single(remsd->cpu,
4076 static int process_backlog(struct napi_struct *napi, int quota)
4079 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4082 /* Check if we have pending ipi, its better to send them now,
4083 * not waiting net_rx_action() end.
4085 if (sd->rps_ipi_list) {
4086 local_irq_disable();
4087 net_rps_action_and_irq_enable(sd);
4090 napi->weight = weight_p;
4091 local_irq_disable();
4092 while (work < quota) {
4093 struct sk_buff *skb;
4096 while ((skb = __skb_dequeue(&sd->process_queue))) {
4098 __netif_receive_skb(skb);
4099 local_irq_disable();
4100 input_queue_head_incr(sd);
4101 if (++work >= quota) {
4108 qlen = skb_queue_len(&sd->input_pkt_queue);
4110 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4111 &sd->process_queue);
4113 if (qlen < quota - work) {
4115 * Inline a custom version of __napi_complete().
4116 * only current cpu owns and manipulates this napi,
4117 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4118 * we can use a plain write instead of clear_bit(),
4119 * and we dont need an smp_mb() memory barrier.
4121 list_del(&napi->poll_list);
4124 quota = work + qlen;
4134 * __napi_schedule - schedule for receive
4135 * @n: entry to schedule
4137 * The entry's receive function will be scheduled to run
4139 void __napi_schedule(struct napi_struct *n)
4141 unsigned long flags;
4143 local_irq_save(flags);
4144 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4145 local_irq_restore(flags);
4147 EXPORT_SYMBOL(__napi_schedule);
4149 void __napi_complete(struct napi_struct *n)
4151 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4152 BUG_ON(n->gro_list);
4154 list_del(&n->poll_list);
4155 smp_mb__before_clear_bit();
4156 clear_bit(NAPI_STATE_SCHED, &n->state);
4158 EXPORT_SYMBOL(__napi_complete);
4160 void napi_complete(struct napi_struct *n)
4162 unsigned long flags;
4165 * don't let napi dequeue from the cpu poll list
4166 * just in case its running on a different cpu
4168 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4171 napi_gro_flush(n, false);
4172 local_irq_save(flags);
4174 local_irq_restore(flags);
4176 EXPORT_SYMBOL(napi_complete);
4178 /* must be called under rcu_read_lock(), as we dont take a reference */
4179 struct napi_struct *napi_by_id(unsigned int napi_id)
4181 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4182 struct napi_struct *napi;
4184 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4185 if (napi->napi_id == napi_id)
4190 EXPORT_SYMBOL_GPL(napi_by_id);
4192 void napi_hash_add(struct napi_struct *napi)
4194 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4196 spin_lock(&napi_hash_lock);
4198 /* 0 is not a valid id, we also skip an id that is taken
4199 * we expect both events to be extremely rare
4202 while (!napi->napi_id) {
4203 napi->napi_id = ++napi_gen_id;
4204 if (napi_by_id(napi->napi_id))
4208 hlist_add_head_rcu(&napi->napi_hash_node,
4209 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4211 spin_unlock(&napi_hash_lock);
4214 EXPORT_SYMBOL_GPL(napi_hash_add);
4216 /* Warning : caller is responsible to make sure rcu grace period
4217 * is respected before freeing memory containing @napi
4219 void napi_hash_del(struct napi_struct *napi)
4221 spin_lock(&napi_hash_lock);
4223 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4224 hlist_del_rcu(&napi->napi_hash_node);
4226 spin_unlock(&napi_hash_lock);
4228 EXPORT_SYMBOL_GPL(napi_hash_del);
4230 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4231 int (*poll)(struct napi_struct *, int), int weight)
4233 INIT_LIST_HEAD(&napi->poll_list);
4234 napi->gro_count = 0;
4235 napi->gro_list = NULL;
4238 if (weight > NAPI_POLL_WEIGHT)
4239 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4241 napi->weight = weight;
4242 list_add(&napi->dev_list, &dev->napi_list);
4244 #ifdef CONFIG_NETPOLL
4245 spin_lock_init(&napi->poll_lock);
4246 napi->poll_owner = -1;
4248 set_bit(NAPI_STATE_SCHED, &napi->state);
4250 EXPORT_SYMBOL(netif_napi_add);
4252 void netif_napi_del(struct napi_struct *napi)
4254 struct sk_buff *skb, *next;
4256 list_del_init(&napi->dev_list);
4257 napi_free_frags(napi);
4259 for (skb = napi->gro_list; skb; skb = next) {
4265 napi->gro_list = NULL;
4266 napi->gro_count = 0;
4268 EXPORT_SYMBOL(netif_napi_del);
4270 static void net_rx_action(struct softirq_action *h)
4272 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4273 unsigned long time_limit = jiffies + 2;
4274 int budget = netdev_budget;
4277 local_irq_disable();
4279 while (!list_empty(&sd->poll_list)) {
4280 struct napi_struct *n;
4283 /* If softirq window is exhuasted then punt.
4284 * Allow this to run for 2 jiffies since which will allow
4285 * an average latency of 1.5/HZ.
4287 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4292 /* Even though interrupts have been re-enabled, this
4293 * access is safe because interrupts can only add new
4294 * entries to the tail of this list, and only ->poll()
4295 * calls can remove this head entry from the list.
4297 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4299 have = netpoll_poll_lock(n);
4303 /* This NAPI_STATE_SCHED test is for avoiding a race
4304 * with netpoll's poll_napi(). Only the entity which
4305 * obtains the lock and sees NAPI_STATE_SCHED set will
4306 * actually make the ->poll() call. Therefore we avoid
4307 * accidentally calling ->poll() when NAPI is not scheduled.
4310 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4311 work = n->poll(n, weight);
4315 WARN_ON_ONCE(work > weight);
4319 local_irq_disable();
4321 /* Drivers must not modify the NAPI state if they
4322 * consume the entire weight. In such cases this code
4323 * still "owns" the NAPI instance and therefore can
4324 * move the instance around on the list at-will.
4326 if (unlikely(work == weight)) {
4327 if (unlikely(napi_disable_pending(n))) {
4330 local_irq_disable();
4333 /* flush too old packets
4334 * If HZ < 1000, flush all packets.
4337 napi_gro_flush(n, HZ >= 1000);
4338 local_irq_disable();
4340 list_move_tail(&n->poll_list, &sd->poll_list);
4344 netpoll_poll_unlock(have);
4347 net_rps_action_and_irq_enable(sd);
4349 #ifdef CONFIG_NET_DMA
4351 * There may not be any more sk_buffs coming right now, so push
4352 * any pending DMA copies to hardware
4354 dma_issue_pending_all();
4361 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4365 struct netdev_upper {
4366 struct net_device *dev;
4368 struct list_head list;
4369 struct rcu_head rcu;
4370 struct list_head search_list;
4373 static void __append_search_uppers(struct list_head *search_list,
4374 struct net_device *dev)
4376 struct netdev_upper *upper;
4378 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4379 /* check if this upper is not already in search list */
4380 if (list_empty(&upper->search_list))
4381 list_add_tail(&upper->search_list, search_list);
4385 static bool __netdev_search_upper_dev(struct net_device *dev,
4386 struct net_device *upper_dev)
4388 LIST_HEAD(search_list);
4389 struct netdev_upper *upper;
4390 struct netdev_upper *tmp;
4393 __append_search_uppers(&search_list, dev);
4394 list_for_each_entry(upper, &search_list, search_list) {
4395 if (upper->dev == upper_dev) {
4399 __append_search_uppers(&search_list, upper->dev);
4401 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4402 INIT_LIST_HEAD(&upper->search_list);
4406 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4407 struct net_device *upper_dev)
4409 struct netdev_upper *upper;
4411 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4412 if (upper->dev == upper_dev)
4419 * netdev_has_upper_dev - Check if device is linked to an upper device
4421 * @upper_dev: upper device to check
4423 * Find out if a device is linked to specified upper device and return true
4424 * in case it is. Note that this checks only immediate upper device,
4425 * not through a complete stack of devices. The caller must hold the RTNL lock.
4427 bool netdev_has_upper_dev(struct net_device *dev,
4428 struct net_device *upper_dev)
4432 return __netdev_find_upper(dev, upper_dev);
4434 EXPORT_SYMBOL(netdev_has_upper_dev);
4437 * netdev_has_any_upper_dev - Check if device is linked to some device
4440 * Find out if a device is linked to an upper device and return true in case
4441 * it is. The caller must hold the RTNL lock.
4443 bool netdev_has_any_upper_dev(struct net_device *dev)
4447 return !list_empty(&dev->upper_dev_list);
4449 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4452 * netdev_master_upper_dev_get - Get master upper device
4455 * Find a master upper device and return pointer to it or NULL in case
4456 * it's not there. The caller must hold the RTNL lock.
4458 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4460 struct netdev_upper *upper;
4464 if (list_empty(&dev->upper_dev_list))
4467 upper = list_first_entry(&dev->upper_dev_list,
4468 struct netdev_upper, list);
4469 if (likely(upper->master))
4473 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4476 * netdev_master_upper_dev_get_rcu - Get master upper device
4479 * Find a master upper device and return pointer to it or NULL in case
4480 * it's not there. The caller must hold the RCU read lock.
4482 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4484 struct netdev_upper *upper;
4486 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4487 struct netdev_upper, list);
4488 if (upper && likely(upper->master))
4492 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4494 static int __netdev_upper_dev_link(struct net_device *dev,
4495 struct net_device *upper_dev, bool master)
4497 struct netdev_upper *upper;
4501 if (dev == upper_dev)
4504 /* To prevent loops, check if dev is not upper device to upper_dev. */
4505 if (__netdev_search_upper_dev(upper_dev, dev))
4508 if (__netdev_find_upper(dev, upper_dev))
4511 if (master && netdev_master_upper_dev_get(dev))
4514 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4518 upper->dev = upper_dev;
4519 upper->master = master;
4520 INIT_LIST_HEAD(&upper->search_list);
4522 /* Ensure that master upper link is always the first item in list. */
4524 list_add_rcu(&upper->list, &dev->upper_dev_list);
4526 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4527 dev_hold(upper_dev);
4528 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4533 * netdev_upper_dev_link - Add a link to the upper device
4535 * @upper_dev: new upper device
4537 * Adds a link to device which is upper to this one. The caller must hold
4538 * the RTNL lock. On a failure a negative errno code is returned.
4539 * On success the reference counts are adjusted and the function
4542 int netdev_upper_dev_link(struct net_device *dev,
4543 struct net_device *upper_dev)
4545 return __netdev_upper_dev_link(dev, upper_dev, false);
4547 EXPORT_SYMBOL(netdev_upper_dev_link);
4550 * netdev_master_upper_dev_link - Add a master link to the upper device
4552 * @upper_dev: new upper device
4554 * Adds a link to device which is upper to this one. In this case, only
4555 * one master upper device can be linked, although other non-master devices
4556 * might be linked as well. The caller must hold the RTNL lock.
4557 * On a failure a negative errno code is returned. On success the reference
4558 * counts are adjusted and the function returns zero.
4560 int netdev_master_upper_dev_link(struct net_device *dev,
4561 struct net_device *upper_dev)
4563 return __netdev_upper_dev_link(dev, upper_dev, true);
4565 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4568 * netdev_upper_dev_unlink - Removes a link to upper device
4570 * @upper_dev: new upper device
4572 * Removes a link to device which is upper to this one. The caller must hold
4575 void netdev_upper_dev_unlink(struct net_device *dev,
4576 struct net_device *upper_dev)
4578 struct netdev_upper *upper;
4582 upper = __netdev_find_upper(dev, upper_dev);
4585 list_del_rcu(&upper->list);
4587 kfree_rcu(upper, rcu);
4588 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4590 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4592 static void dev_change_rx_flags(struct net_device *dev, int flags)
4594 const struct net_device_ops *ops = dev->netdev_ops;
4596 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4597 ops->ndo_change_rx_flags(dev, flags);
4600 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4602 unsigned int old_flags = dev->flags;
4608 dev->flags |= IFF_PROMISC;
4609 dev->promiscuity += inc;
4610 if (dev->promiscuity == 0) {
4613 * If inc causes overflow, untouch promisc and return error.
4616 dev->flags &= ~IFF_PROMISC;
4618 dev->promiscuity -= inc;
4619 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4624 if (dev->flags != old_flags) {
4625 pr_info("device %s %s promiscuous mode\n",
4627 dev->flags & IFF_PROMISC ? "entered" : "left");
4628 if (audit_enabled) {
4629 current_uid_gid(&uid, &gid);
4630 audit_log(current->audit_context, GFP_ATOMIC,
4631 AUDIT_ANOM_PROMISCUOUS,
4632 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4633 dev->name, (dev->flags & IFF_PROMISC),
4634 (old_flags & IFF_PROMISC),
4635 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4636 from_kuid(&init_user_ns, uid),
4637 from_kgid(&init_user_ns, gid),
4638 audit_get_sessionid(current));
4641 dev_change_rx_flags(dev, IFF_PROMISC);
4647 * dev_set_promiscuity - update promiscuity count on a device
4651 * Add or remove promiscuity from a device. While the count in the device
4652 * remains above zero the interface remains promiscuous. Once it hits zero
4653 * the device reverts back to normal filtering operation. A negative inc
4654 * value is used to drop promiscuity on the device.
4655 * Return 0 if successful or a negative errno code on error.
4657 int dev_set_promiscuity(struct net_device *dev, int inc)
4659 unsigned int old_flags = dev->flags;
4662 err = __dev_set_promiscuity(dev, inc);
4665 if (dev->flags != old_flags)
4666 dev_set_rx_mode(dev);
4669 EXPORT_SYMBOL(dev_set_promiscuity);
4672 * dev_set_allmulti - update allmulti count on a device
4676 * Add or remove reception of all multicast frames to a device. While the
4677 * count in the device remains above zero the interface remains listening
4678 * to all interfaces. Once it hits zero the device reverts back to normal
4679 * filtering operation. A negative @inc value is used to drop the counter
4680 * when releasing a resource needing all multicasts.
4681 * Return 0 if successful or a negative errno code on error.
4684 int dev_set_allmulti(struct net_device *dev, int inc)
4686 unsigned int old_flags = dev->flags;
4690 dev->flags |= IFF_ALLMULTI;
4691 dev->allmulti += inc;
4692 if (dev->allmulti == 0) {
4695 * If inc causes overflow, untouch allmulti and return error.
4698 dev->flags &= ~IFF_ALLMULTI;
4700 dev->allmulti -= inc;
4701 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4706 if (dev->flags ^ old_flags) {
4707 dev_change_rx_flags(dev, IFF_ALLMULTI);
4708 dev_set_rx_mode(dev);
4712 EXPORT_SYMBOL(dev_set_allmulti);
4715 * Upload unicast and multicast address lists to device and
4716 * configure RX filtering. When the device doesn't support unicast
4717 * filtering it is put in promiscuous mode while unicast addresses
4720 void __dev_set_rx_mode(struct net_device *dev)
4722 const struct net_device_ops *ops = dev->netdev_ops;
4724 /* dev_open will call this function so the list will stay sane. */
4725 if (!(dev->flags&IFF_UP))
4728 if (!netif_device_present(dev))
4731 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4732 /* Unicast addresses changes may only happen under the rtnl,
4733 * therefore calling __dev_set_promiscuity here is safe.
4735 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4736 __dev_set_promiscuity(dev, 1);
4737 dev->uc_promisc = true;
4738 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4739 __dev_set_promiscuity(dev, -1);
4740 dev->uc_promisc = false;
4744 if (ops->ndo_set_rx_mode)
4745 ops->ndo_set_rx_mode(dev);
4748 void dev_set_rx_mode(struct net_device *dev)
4750 netif_addr_lock_bh(dev);
4751 __dev_set_rx_mode(dev);
4752 netif_addr_unlock_bh(dev);
4756 * dev_get_flags - get flags reported to userspace
4759 * Get the combination of flag bits exported through APIs to userspace.
4761 unsigned int dev_get_flags(const struct net_device *dev)
4765 flags = (dev->flags & ~(IFF_PROMISC |
4770 (dev->gflags & (IFF_PROMISC |
4773 if (netif_running(dev)) {
4774 if (netif_oper_up(dev))
4775 flags |= IFF_RUNNING;
4776 if (netif_carrier_ok(dev))
4777 flags |= IFF_LOWER_UP;
4778 if (netif_dormant(dev))
4779 flags |= IFF_DORMANT;
4784 EXPORT_SYMBOL(dev_get_flags);
4786 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4788 unsigned int old_flags = dev->flags;
4794 * Set the flags on our device.
4797 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4798 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4800 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4804 * Load in the correct multicast list now the flags have changed.
4807 if ((old_flags ^ flags) & IFF_MULTICAST)
4808 dev_change_rx_flags(dev, IFF_MULTICAST);
4810 dev_set_rx_mode(dev);
4813 * Have we downed the interface. We handle IFF_UP ourselves
4814 * according to user attempts to set it, rather than blindly
4819 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4820 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4823 dev_set_rx_mode(dev);
4826 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4827 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4829 dev->gflags ^= IFF_PROMISC;
4830 dev_set_promiscuity(dev, inc);
4833 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4834 is important. Some (broken) drivers set IFF_PROMISC, when
4835 IFF_ALLMULTI is requested not asking us and not reporting.
4837 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4838 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4840 dev->gflags ^= IFF_ALLMULTI;
4841 dev_set_allmulti(dev, inc);
4847 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4849 unsigned int changes = dev->flags ^ old_flags;
4851 if (changes & IFF_UP) {
4852 if (dev->flags & IFF_UP)
4853 call_netdevice_notifiers(NETDEV_UP, dev);
4855 call_netdevice_notifiers(NETDEV_DOWN, dev);
4858 if (dev->flags & IFF_UP &&
4859 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
4860 struct netdev_notifier_change_info change_info;
4862 change_info.flags_changed = changes;
4863 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
4869 * dev_change_flags - change device settings
4871 * @flags: device state flags
4873 * Change settings on device based state flags. The flags are
4874 * in the userspace exported format.
4876 int dev_change_flags(struct net_device *dev, unsigned int flags)
4879 unsigned int changes, old_flags = dev->flags;
4881 ret = __dev_change_flags(dev, flags);
4885 changes = old_flags ^ dev->flags;
4887 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4889 __dev_notify_flags(dev, old_flags);
4892 EXPORT_SYMBOL(dev_change_flags);
4895 * dev_set_mtu - Change maximum transfer unit
4897 * @new_mtu: new transfer unit
4899 * Change the maximum transfer size of the network device.
4901 int dev_set_mtu(struct net_device *dev, int new_mtu)
4903 const struct net_device_ops *ops = dev->netdev_ops;
4906 if (new_mtu == dev->mtu)
4909 /* MTU must be positive. */
4913 if (!netif_device_present(dev))
4917 if (ops->ndo_change_mtu)
4918 err = ops->ndo_change_mtu(dev, new_mtu);
4923 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4926 EXPORT_SYMBOL(dev_set_mtu);
4929 * dev_set_group - Change group this device belongs to
4931 * @new_group: group this device should belong to
4933 void dev_set_group(struct net_device *dev, int new_group)
4935 dev->group = new_group;
4937 EXPORT_SYMBOL(dev_set_group);
4940 * dev_set_mac_address - Change Media Access Control Address
4944 * Change the hardware (MAC) address of the device
4946 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4948 const struct net_device_ops *ops = dev->netdev_ops;
4951 if (!ops->ndo_set_mac_address)
4953 if (sa->sa_family != dev->type)
4955 if (!netif_device_present(dev))
4957 err = ops->ndo_set_mac_address(dev, sa);
4960 dev->addr_assign_type = NET_ADDR_SET;
4961 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4962 add_device_randomness(dev->dev_addr, dev->addr_len);
4965 EXPORT_SYMBOL(dev_set_mac_address);
4968 * dev_change_carrier - Change device carrier
4970 * @new_carrier: new value
4972 * Change device carrier
4974 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4976 const struct net_device_ops *ops = dev->netdev_ops;
4978 if (!ops->ndo_change_carrier)
4980 if (!netif_device_present(dev))
4982 return ops->ndo_change_carrier(dev, new_carrier);
4984 EXPORT_SYMBOL(dev_change_carrier);
4987 * dev_new_index - allocate an ifindex
4988 * @net: the applicable net namespace
4990 * Returns a suitable unique value for a new device interface
4991 * number. The caller must hold the rtnl semaphore or the
4992 * dev_base_lock to be sure it remains unique.
4994 static int dev_new_index(struct net *net)
4996 int ifindex = net->ifindex;
5000 if (!__dev_get_by_index(net, ifindex))
5001 return net->ifindex = ifindex;
5005 /* Delayed registration/unregisteration */
5006 static LIST_HEAD(net_todo_list);
5008 static void net_set_todo(struct net_device *dev)
5010 list_add_tail(&dev->todo_list, &net_todo_list);
5013 static void rollback_registered_many(struct list_head *head)
5015 struct net_device *dev, *tmp;
5017 BUG_ON(dev_boot_phase);
5020 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5021 /* Some devices call without registering
5022 * for initialization unwind. Remove those
5023 * devices and proceed with the remaining.
5025 if (dev->reg_state == NETREG_UNINITIALIZED) {
5026 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5030 list_del(&dev->unreg_list);
5033 dev->dismantle = true;
5034 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5037 /* If device is running, close it first. */
5038 dev_close_many(head);
5040 list_for_each_entry(dev, head, unreg_list) {
5041 /* And unlink it from device chain. */
5042 unlist_netdevice(dev);
5044 dev->reg_state = NETREG_UNREGISTERING;
5049 list_for_each_entry(dev, head, unreg_list) {
5050 /* Shutdown queueing discipline. */
5054 /* Notify protocols, that we are about to destroy
5055 this device. They should clean all the things.
5057 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5059 if (!dev->rtnl_link_ops ||
5060 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5061 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5064 * Flush the unicast and multicast chains
5069 if (dev->netdev_ops->ndo_uninit)
5070 dev->netdev_ops->ndo_uninit(dev);
5072 /* Notifier chain MUST detach us all upper devices. */
5073 WARN_ON(netdev_has_any_upper_dev(dev));
5075 /* Remove entries from kobject tree */
5076 netdev_unregister_kobject(dev);
5078 /* Remove XPS queueing entries */
5079 netif_reset_xps_queues_gt(dev, 0);
5085 list_for_each_entry(dev, head, unreg_list)
5089 static void rollback_registered(struct net_device *dev)
5093 list_add(&dev->unreg_list, &single);
5094 rollback_registered_many(&single);
5098 static netdev_features_t netdev_fix_features(struct net_device *dev,
5099 netdev_features_t features)
5101 /* Fix illegal checksum combinations */
5102 if ((features & NETIF_F_HW_CSUM) &&
5103 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5104 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5105 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5108 /* TSO requires that SG is present as well. */
5109 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5110 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5111 features &= ~NETIF_F_ALL_TSO;
5114 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5115 !(features & NETIF_F_IP_CSUM)) {
5116 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5117 features &= ~NETIF_F_TSO;
5118 features &= ~NETIF_F_TSO_ECN;
5121 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5122 !(features & NETIF_F_IPV6_CSUM)) {
5123 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5124 features &= ~NETIF_F_TSO6;
5127 /* TSO ECN requires that TSO is present as well. */
5128 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5129 features &= ~NETIF_F_TSO_ECN;
5131 /* Software GSO depends on SG. */
5132 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5133 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5134 features &= ~NETIF_F_GSO;
5137 /* UFO needs SG and checksumming */
5138 if (features & NETIF_F_UFO) {
5139 /* maybe split UFO into V4 and V6? */
5140 if (!((features & NETIF_F_GEN_CSUM) ||
5141 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5142 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5144 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5145 features &= ~NETIF_F_UFO;
5148 if (!(features & NETIF_F_SG)) {
5150 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5151 features &= ~NETIF_F_UFO;
5158 int __netdev_update_features(struct net_device *dev)
5160 netdev_features_t features;
5165 features = netdev_get_wanted_features(dev);
5167 if (dev->netdev_ops->ndo_fix_features)
5168 features = dev->netdev_ops->ndo_fix_features(dev, features);
5170 /* driver might be less strict about feature dependencies */
5171 features = netdev_fix_features(dev, features);
5173 if (dev->features == features)
5176 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5177 &dev->features, &features);
5179 if (dev->netdev_ops->ndo_set_features)
5180 err = dev->netdev_ops->ndo_set_features(dev, features);
5182 if (unlikely(err < 0)) {
5184 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5185 err, &features, &dev->features);
5190 dev->features = features;
5196 * netdev_update_features - recalculate device features
5197 * @dev: the device to check
5199 * Recalculate dev->features set and send notifications if it
5200 * has changed. Should be called after driver or hardware dependent
5201 * conditions might have changed that influence the features.
5203 void netdev_update_features(struct net_device *dev)
5205 if (__netdev_update_features(dev))
5206 netdev_features_change(dev);
5208 EXPORT_SYMBOL(netdev_update_features);
5211 * netdev_change_features - recalculate device features
5212 * @dev: the device to check
5214 * Recalculate dev->features set and send notifications even
5215 * if they have not changed. Should be called instead of
5216 * netdev_update_features() if also dev->vlan_features might
5217 * have changed to allow the changes to be propagated to stacked
5220 void netdev_change_features(struct net_device *dev)
5222 __netdev_update_features(dev);
5223 netdev_features_change(dev);
5225 EXPORT_SYMBOL(netdev_change_features);
5228 * netif_stacked_transfer_operstate - transfer operstate
5229 * @rootdev: the root or lower level device to transfer state from
5230 * @dev: the device to transfer operstate to
5232 * Transfer operational state from root to device. This is normally
5233 * called when a stacking relationship exists between the root
5234 * device and the device(a leaf device).
5236 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5237 struct net_device *dev)
5239 if (rootdev->operstate == IF_OPER_DORMANT)
5240 netif_dormant_on(dev);
5242 netif_dormant_off(dev);
5244 if (netif_carrier_ok(rootdev)) {
5245 if (!netif_carrier_ok(dev))
5246 netif_carrier_on(dev);
5248 if (netif_carrier_ok(dev))
5249 netif_carrier_off(dev);
5252 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5255 static int netif_alloc_rx_queues(struct net_device *dev)
5257 unsigned int i, count = dev->num_rx_queues;
5258 struct netdev_rx_queue *rx;
5262 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5268 for (i = 0; i < count; i++)
5274 static void netdev_init_one_queue(struct net_device *dev,
5275 struct netdev_queue *queue, void *_unused)
5277 /* Initialize queue lock */
5278 spin_lock_init(&queue->_xmit_lock);
5279 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5280 queue->xmit_lock_owner = -1;
5281 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5284 dql_init(&queue->dql, HZ);
5288 static void netif_free_tx_queues(struct net_device *dev)
5290 if (is_vmalloc_addr(dev->_tx))
5296 static int netif_alloc_netdev_queues(struct net_device *dev)
5298 unsigned int count = dev->num_tx_queues;
5299 struct netdev_queue *tx;
5300 size_t sz = count * sizeof(*tx);
5302 BUG_ON(count < 1 || count > 0xffff);
5304 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5312 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5313 spin_lock_init(&dev->tx_global_lock);
5319 * register_netdevice - register a network device
5320 * @dev: device to register
5322 * Take a completed network device structure and add it to the kernel
5323 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5324 * chain. 0 is returned on success. A negative errno code is returned
5325 * on a failure to set up the device, or if the name is a duplicate.
5327 * Callers must hold the rtnl semaphore. You may want
5328 * register_netdev() instead of this.
5331 * The locking appears insufficient to guarantee two parallel registers
5332 * will not get the same name.
5335 int register_netdevice(struct net_device *dev)
5338 struct net *net = dev_net(dev);
5340 BUG_ON(dev_boot_phase);
5345 /* When net_device's are persistent, this will be fatal. */
5346 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5349 spin_lock_init(&dev->addr_list_lock);
5350 netdev_set_addr_lockdep_class(dev);
5354 ret = dev_get_valid_name(net, dev, dev->name);
5358 /* Init, if this function is available */
5359 if (dev->netdev_ops->ndo_init) {
5360 ret = dev->netdev_ops->ndo_init(dev);
5368 if (((dev->hw_features | dev->features) &
5369 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5370 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5371 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5372 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5379 dev->ifindex = dev_new_index(net);
5380 else if (__dev_get_by_index(net, dev->ifindex))
5383 if (dev->iflink == -1)
5384 dev->iflink = dev->ifindex;
5386 /* Transfer changeable features to wanted_features and enable
5387 * software offloads (GSO and GRO).
5389 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5390 dev->features |= NETIF_F_SOFT_FEATURES;
5391 dev->wanted_features = dev->features & dev->hw_features;
5393 /* Turn on no cache copy if HW is doing checksum */
5394 if (!(dev->flags & IFF_LOOPBACK)) {
5395 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5396 if (dev->features & NETIF_F_ALL_CSUM) {
5397 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5398 dev->features |= NETIF_F_NOCACHE_COPY;
5402 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5404 dev->vlan_features |= NETIF_F_HIGHDMA;
5406 /* Make NETIF_F_SG inheritable to tunnel devices.
5408 dev->hw_enc_features |= NETIF_F_SG;
5410 /* Make NETIF_F_SG inheritable to MPLS.
5412 dev->mpls_features |= NETIF_F_SG;
5414 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5415 ret = notifier_to_errno(ret);
5419 ret = netdev_register_kobject(dev);
5422 dev->reg_state = NETREG_REGISTERED;
5424 __netdev_update_features(dev);
5427 * Default initial state at registry is that the
5428 * device is present.
5431 set_bit(__LINK_STATE_PRESENT, &dev->state);
5433 linkwatch_init_dev(dev);
5435 dev_init_scheduler(dev);
5437 list_netdevice(dev);
5438 add_device_randomness(dev->dev_addr, dev->addr_len);
5440 /* If the device has permanent device address, driver should
5441 * set dev_addr and also addr_assign_type should be set to
5442 * NET_ADDR_PERM (default value).
5444 if (dev->addr_assign_type == NET_ADDR_PERM)
5445 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5447 /* Notify protocols, that a new device appeared. */
5448 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5449 ret = notifier_to_errno(ret);
5451 rollback_registered(dev);
5452 dev->reg_state = NETREG_UNREGISTERED;
5455 * Prevent userspace races by waiting until the network
5456 * device is fully setup before sending notifications.
5458 if (!dev->rtnl_link_ops ||
5459 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5460 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5466 if (dev->netdev_ops->ndo_uninit)
5467 dev->netdev_ops->ndo_uninit(dev);
5470 EXPORT_SYMBOL(register_netdevice);
5473 * init_dummy_netdev - init a dummy network device for NAPI
5474 * @dev: device to init
5476 * This takes a network device structure and initialize the minimum
5477 * amount of fields so it can be used to schedule NAPI polls without
5478 * registering a full blown interface. This is to be used by drivers
5479 * that need to tie several hardware interfaces to a single NAPI
5480 * poll scheduler due to HW limitations.
5482 int init_dummy_netdev(struct net_device *dev)
5484 /* Clear everything. Note we don't initialize spinlocks
5485 * are they aren't supposed to be taken by any of the
5486 * NAPI code and this dummy netdev is supposed to be
5487 * only ever used for NAPI polls
5489 memset(dev, 0, sizeof(struct net_device));
5491 /* make sure we BUG if trying to hit standard
5492 * register/unregister code path
5494 dev->reg_state = NETREG_DUMMY;
5496 /* NAPI wants this */
5497 INIT_LIST_HEAD(&dev->napi_list);
5499 /* a dummy interface is started by default */
5500 set_bit(__LINK_STATE_PRESENT, &dev->state);
5501 set_bit(__LINK_STATE_START, &dev->state);
5503 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5504 * because users of this 'device' dont need to change
5510 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5514 * register_netdev - register a network device
5515 * @dev: device to register
5517 * Take a completed network device structure and add it to the kernel
5518 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5519 * chain. 0 is returned on success. A negative errno code is returned
5520 * on a failure to set up the device, or if the name is a duplicate.
5522 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5523 * and expands the device name if you passed a format string to
5526 int register_netdev(struct net_device *dev)
5531 err = register_netdevice(dev);
5535 EXPORT_SYMBOL(register_netdev);
5537 int netdev_refcnt_read(const struct net_device *dev)
5541 for_each_possible_cpu(i)
5542 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5545 EXPORT_SYMBOL(netdev_refcnt_read);
5548 * netdev_wait_allrefs - wait until all references are gone.
5549 * @dev: target net_device
5551 * This is called when unregistering network devices.
5553 * Any protocol or device that holds a reference should register
5554 * for netdevice notification, and cleanup and put back the
5555 * reference if they receive an UNREGISTER event.
5556 * We can get stuck here if buggy protocols don't correctly
5559 static void netdev_wait_allrefs(struct net_device *dev)
5561 unsigned long rebroadcast_time, warning_time;
5564 linkwatch_forget_dev(dev);
5566 rebroadcast_time = warning_time = jiffies;
5567 refcnt = netdev_refcnt_read(dev);
5569 while (refcnt != 0) {
5570 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5573 /* Rebroadcast unregister notification */
5574 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5580 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5581 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5583 /* We must not have linkwatch events
5584 * pending on unregister. If this
5585 * happens, we simply run the queue
5586 * unscheduled, resulting in a noop
5589 linkwatch_run_queue();
5594 rebroadcast_time = jiffies;
5599 refcnt = netdev_refcnt_read(dev);
5601 if (time_after(jiffies, warning_time + 10 * HZ)) {
5602 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5604 warning_time = jiffies;
5613 * register_netdevice(x1);
5614 * register_netdevice(x2);
5616 * unregister_netdevice(y1);
5617 * unregister_netdevice(y2);
5623 * We are invoked by rtnl_unlock().
5624 * This allows us to deal with problems:
5625 * 1) We can delete sysfs objects which invoke hotplug
5626 * without deadlocking with linkwatch via keventd.
5627 * 2) Since we run with the RTNL semaphore not held, we can sleep
5628 * safely in order to wait for the netdev refcnt to drop to zero.
5630 * We must not return until all unregister events added during
5631 * the interval the lock was held have been completed.
5633 void netdev_run_todo(void)
5635 struct list_head list;
5637 /* Snapshot list, allow later requests */
5638 list_replace_init(&net_todo_list, &list);
5643 /* Wait for rcu callbacks to finish before next phase */
5644 if (!list_empty(&list))
5647 while (!list_empty(&list)) {
5648 struct net_device *dev
5649 = list_first_entry(&list, struct net_device, todo_list);
5650 list_del(&dev->todo_list);
5653 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5656 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5657 pr_err("network todo '%s' but state %d\n",
5658 dev->name, dev->reg_state);
5663 dev->reg_state = NETREG_UNREGISTERED;
5665 on_each_cpu(flush_backlog, dev, 1);
5667 netdev_wait_allrefs(dev);
5670 BUG_ON(netdev_refcnt_read(dev));
5671 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5672 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5673 WARN_ON(dev->dn_ptr);
5675 if (dev->destructor)
5676 dev->destructor(dev);
5678 /* Free network device */
5679 kobject_put(&dev->dev.kobj);
5683 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5684 * fields in the same order, with only the type differing.
5686 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5687 const struct net_device_stats *netdev_stats)
5689 #if BITS_PER_LONG == 64
5690 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5691 memcpy(stats64, netdev_stats, sizeof(*stats64));
5693 size_t i, n = sizeof(*stats64) / sizeof(u64);
5694 const unsigned long *src = (const unsigned long *)netdev_stats;
5695 u64 *dst = (u64 *)stats64;
5697 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5698 sizeof(*stats64) / sizeof(u64));
5699 for (i = 0; i < n; i++)
5703 EXPORT_SYMBOL(netdev_stats_to_stats64);
5706 * dev_get_stats - get network device statistics
5707 * @dev: device to get statistics from
5708 * @storage: place to store stats
5710 * Get network statistics from device. Return @storage.
5711 * The device driver may provide its own method by setting
5712 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5713 * otherwise the internal statistics structure is used.
5715 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5716 struct rtnl_link_stats64 *storage)
5718 const struct net_device_ops *ops = dev->netdev_ops;
5720 if (ops->ndo_get_stats64) {
5721 memset(storage, 0, sizeof(*storage));
5722 ops->ndo_get_stats64(dev, storage);
5723 } else if (ops->ndo_get_stats) {
5724 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5726 netdev_stats_to_stats64(storage, &dev->stats);
5728 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5731 EXPORT_SYMBOL(dev_get_stats);
5733 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5735 struct netdev_queue *queue = dev_ingress_queue(dev);
5737 #ifdef CONFIG_NET_CLS_ACT
5740 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5743 netdev_init_one_queue(dev, queue, NULL);
5744 queue->qdisc = &noop_qdisc;
5745 queue->qdisc_sleeping = &noop_qdisc;
5746 rcu_assign_pointer(dev->ingress_queue, queue);
5751 static const struct ethtool_ops default_ethtool_ops;
5753 void netdev_set_default_ethtool_ops(struct net_device *dev,
5754 const struct ethtool_ops *ops)
5756 if (dev->ethtool_ops == &default_ethtool_ops)
5757 dev->ethtool_ops = ops;
5759 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5762 * alloc_netdev_mqs - allocate network device
5763 * @sizeof_priv: size of private data to allocate space for
5764 * @name: device name format string
5765 * @setup: callback to initialize device
5766 * @txqs: the number of TX subqueues to allocate
5767 * @rxqs: the number of RX subqueues to allocate
5769 * Allocates a struct net_device with private data area for driver use
5770 * and performs basic initialization. Also allocates subquue structs
5771 * for each queue on the device.
5773 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5774 void (*setup)(struct net_device *),
5775 unsigned int txqs, unsigned int rxqs)
5777 struct net_device *dev;
5779 struct net_device *p;
5781 BUG_ON(strlen(name) >= sizeof(dev->name));
5784 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5790 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5795 alloc_size = sizeof(struct net_device);
5797 /* ensure 32-byte alignment of private area */
5798 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5799 alloc_size += sizeof_priv;
5801 /* ensure 32-byte alignment of whole construct */
5802 alloc_size += NETDEV_ALIGN - 1;
5804 p = kzalloc(alloc_size, GFP_KERNEL);
5808 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5809 dev->padded = (char *)dev - (char *)p;
5811 dev->pcpu_refcnt = alloc_percpu(int);
5812 if (!dev->pcpu_refcnt)
5815 if (dev_addr_init(dev))
5821 dev_net_set(dev, &init_net);
5823 dev->gso_max_size = GSO_MAX_SIZE;
5824 dev->gso_max_segs = GSO_MAX_SEGS;
5826 INIT_LIST_HEAD(&dev->napi_list);
5827 INIT_LIST_HEAD(&dev->unreg_list);
5828 INIT_LIST_HEAD(&dev->link_watch_list);
5829 INIT_LIST_HEAD(&dev->upper_dev_list);
5830 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5833 dev->num_tx_queues = txqs;
5834 dev->real_num_tx_queues = txqs;
5835 if (netif_alloc_netdev_queues(dev))
5839 dev->num_rx_queues = rxqs;
5840 dev->real_num_rx_queues = rxqs;
5841 if (netif_alloc_rx_queues(dev))
5845 strcpy(dev->name, name);
5846 dev->group = INIT_NETDEV_GROUP;
5847 if (!dev->ethtool_ops)
5848 dev->ethtool_ops = &default_ethtool_ops;
5856 free_percpu(dev->pcpu_refcnt);
5857 netif_free_tx_queues(dev);
5866 EXPORT_SYMBOL(alloc_netdev_mqs);
5869 * free_netdev - free network device
5872 * This function does the last stage of destroying an allocated device
5873 * interface. The reference to the device object is released.
5874 * If this is the last reference then it will be freed.
5876 void free_netdev(struct net_device *dev)
5878 struct napi_struct *p, *n;
5880 release_net(dev_net(dev));
5882 netif_free_tx_queues(dev);
5887 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5889 /* Flush device addresses */
5890 dev_addr_flush(dev);
5892 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5895 free_percpu(dev->pcpu_refcnt);
5896 dev->pcpu_refcnt = NULL;
5898 /* Compatibility with error handling in drivers */
5899 if (dev->reg_state == NETREG_UNINITIALIZED) {
5900 kfree((char *)dev - dev->padded);
5904 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5905 dev->reg_state = NETREG_RELEASED;
5907 /* will free via device release */
5908 put_device(&dev->dev);
5910 EXPORT_SYMBOL(free_netdev);
5913 * synchronize_net - Synchronize with packet receive processing
5915 * Wait for packets currently being received to be done.
5916 * Does not block later packets from starting.
5918 void synchronize_net(void)
5921 if (rtnl_is_locked())
5922 synchronize_rcu_expedited();
5926 EXPORT_SYMBOL(synchronize_net);
5929 * unregister_netdevice_queue - remove device from the kernel
5933 * This function shuts down a device interface and removes it
5934 * from the kernel tables.
5935 * If head not NULL, device is queued to be unregistered later.
5937 * Callers must hold the rtnl semaphore. You may want
5938 * unregister_netdev() instead of this.
5941 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5946 list_move_tail(&dev->unreg_list, head);
5948 rollback_registered(dev);
5949 /* Finish processing unregister after unlock */
5953 EXPORT_SYMBOL(unregister_netdevice_queue);
5956 * unregister_netdevice_many - unregister many devices
5957 * @head: list of devices
5959 void unregister_netdevice_many(struct list_head *head)
5961 struct net_device *dev;
5963 if (!list_empty(head)) {
5964 rollback_registered_many(head);
5965 list_for_each_entry(dev, head, unreg_list)
5969 EXPORT_SYMBOL(unregister_netdevice_many);
5972 * unregister_netdev - remove device from the kernel
5975 * This function shuts down a device interface and removes it
5976 * from the kernel tables.
5978 * This is just a wrapper for unregister_netdevice that takes
5979 * the rtnl semaphore. In general you want to use this and not
5980 * unregister_netdevice.
5982 void unregister_netdev(struct net_device *dev)
5985 unregister_netdevice(dev);
5988 EXPORT_SYMBOL(unregister_netdev);
5991 * dev_change_net_namespace - move device to different nethost namespace
5993 * @net: network namespace
5994 * @pat: If not NULL name pattern to try if the current device name
5995 * is already taken in the destination network namespace.
5997 * This function shuts down a device interface and moves it
5998 * to a new network namespace. On success 0 is returned, on
5999 * a failure a netagive errno code is returned.
6001 * Callers must hold the rtnl semaphore.
6004 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6010 /* Don't allow namespace local devices to be moved. */
6012 if (dev->features & NETIF_F_NETNS_LOCAL)
6015 /* Ensure the device has been registrered */
6016 if (dev->reg_state != NETREG_REGISTERED)
6019 /* Get out if there is nothing todo */
6021 if (net_eq(dev_net(dev), net))
6024 /* Pick the destination device name, and ensure
6025 * we can use it in the destination network namespace.
6028 if (__dev_get_by_name(net, dev->name)) {
6029 /* We get here if we can't use the current device name */
6032 if (dev_get_valid_name(net, dev, pat) < 0)
6037 * And now a mini version of register_netdevice unregister_netdevice.
6040 /* If device is running close it first. */
6043 /* And unlink it from device chain */
6045 unlist_netdevice(dev);
6049 /* Shutdown queueing discipline. */
6052 /* Notify protocols, that we are about to destroy
6053 this device. They should clean all the things.
6055 Note that dev->reg_state stays at NETREG_REGISTERED.
6056 This is wanted because this way 8021q and macvlan know
6057 the device is just moving and can keep their slaves up.
6059 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6061 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6062 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6065 * Flush the unicast and multicast chains
6070 /* Send a netdev-removed uevent to the old namespace */
6071 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6073 /* Actually switch the network namespace */
6074 dev_net_set(dev, net);
6076 /* If there is an ifindex conflict assign a new one */
6077 if (__dev_get_by_index(net, dev->ifindex)) {
6078 int iflink = (dev->iflink == dev->ifindex);
6079 dev->ifindex = dev_new_index(net);
6081 dev->iflink = dev->ifindex;
6084 /* Send a netdev-add uevent to the new namespace */
6085 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6087 /* Fixup kobjects */
6088 err = device_rename(&dev->dev, dev->name);
6091 /* Add the device back in the hashes */
6092 list_netdevice(dev);
6094 /* Notify protocols, that a new device appeared. */
6095 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6098 * Prevent userspace races by waiting until the network
6099 * device is fully setup before sending notifications.
6101 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6108 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6110 static int dev_cpu_callback(struct notifier_block *nfb,
6111 unsigned long action,
6114 struct sk_buff **list_skb;
6115 struct sk_buff *skb;
6116 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6117 struct softnet_data *sd, *oldsd;
6119 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6122 local_irq_disable();
6123 cpu = smp_processor_id();
6124 sd = &per_cpu(softnet_data, cpu);
6125 oldsd = &per_cpu(softnet_data, oldcpu);
6127 /* Find end of our completion_queue. */
6128 list_skb = &sd->completion_queue;
6130 list_skb = &(*list_skb)->next;
6131 /* Append completion queue from offline CPU. */
6132 *list_skb = oldsd->completion_queue;
6133 oldsd->completion_queue = NULL;
6135 /* Append output queue from offline CPU. */
6136 if (oldsd->output_queue) {
6137 *sd->output_queue_tailp = oldsd->output_queue;
6138 sd->output_queue_tailp = oldsd->output_queue_tailp;
6139 oldsd->output_queue = NULL;
6140 oldsd->output_queue_tailp = &oldsd->output_queue;
6142 /* Append NAPI poll list from offline CPU. */
6143 if (!list_empty(&oldsd->poll_list)) {
6144 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6145 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6148 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6151 /* Process offline CPU's input_pkt_queue */
6152 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6154 input_queue_head_incr(oldsd);
6156 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6158 input_queue_head_incr(oldsd);
6166 * netdev_increment_features - increment feature set by one
6167 * @all: current feature set
6168 * @one: new feature set
6169 * @mask: mask feature set
6171 * Computes a new feature set after adding a device with feature set
6172 * @one to the master device with current feature set @all. Will not
6173 * enable anything that is off in @mask. Returns the new feature set.
6175 netdev_features_t netdev_increment_features(netdev_features_t all,
6176 netdev_features_t one, netdev_features_t mask)
6178 if (mask & NETIF_F_GEN_CSUM)
6179 mask |= NETIF_F_ALL_CSUM;
6180 mask |= NETIF_F_VLAN_CHALLENGED;
6182 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6183 all &= one | ~NETIF_F_ALL_FOR_ALL;
6185 /* If one device supports hw checksumming, set for all. */
6186 if (all & NETIF_F_GEN_CSUM)
6187 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6191 EXPORT_SYMBOL(netdev_increment_features);
6193 static struct hlist_head * __net_init netdev_create_hash(void)
6196 struct hlist_head *hash;
6198 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6200 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6201 INIT_HLIST_HEAD(&hash[i]);
6206 /* Initialize per network namespace state */
6207 static int __net_init netdev_init(struct net *net)
6209 if (net != &init_net)
6210 INIT_LIST_HEAD(&net->dev_base_head);
6212 net->dev_name_head = netdev_create_hash();
6213 if (net->dev_name_head == NULL)
6216 net->dev_index_head = netdev_create_hash();
6217 if (net->dev_index_head == NULL)
6223 kfree(net->dev_name_head);
6229 * netdev_drivername - network driver for the device
6230 * @dev: network device
6232 * Determine network driver for device.
6234 const char *netdev_drivername(const struct net_device *dev)
6236 const struct device_driver *driver;
6237 const struct device *parent;
6238 const char *empty = "";
6240 parent = dev->dev.parent;
6244 driver = parent->driver;
6245 if (driver && driver->name)
6246 return driver->name;
6250 static int __netdev_printk(const char *level, const struct net_device *dev,
6251 struct va_format *vaf)
6255 if (dev && dev->dev.parent) {
6256 r = dev_printk_emit(level[1] - '0',
6259 dev_driver_string(dev->dev.parent),
6260 dev_name(dev->dev.parent),
6261 netdev_name(dev), vaf);
6263 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6265 r = printk("%s(NULL net_device): %pV", level, vaf);
6271 int netdev_printk(const char *level, const struct net_device *dev,
6272 const char *format, ...)
6274 struct va_format vaf;
6278 va_start(args, format);
6283 r = __netdev_printk(level, dev, &vaf);
6289 EXPORT_SYMBOL(netdev_printk);
6291 #define define_netdev_printk_level(func, level) \
6292 int func(const struct net_device *dev, const char *fmt, ...) \
6295 struct va_format vaf; \
6298 va_start(args, fmt); \
6303 r = __netdev_printk(level, dev, &vaf); \
6309 EXPORT_SYMBOL(func);
6311 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6312 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6313 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6314 define_netdev_printk_level(netdev_err, KERN_ERR);
6315 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6316 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6317 define_netdev_printk_level(netdev_info, KERN_INFO);
6319 static void __net_exit netdev_exit(struct net *net)
6321 kfree(net->dev_name_head);
6322 kfree(net->dev_index_head);
6325 static struct pernet_operations __net_initdata netdev_net_ops = {
6326 .init = netdev_init,
6327 .exit = netdev_exit,
6330 static void __net_exit default_device_exit(struct net *net)
6332 struct net_device *dev, *aux;
6334 * Push all migratable network devices back to the
6335 * initial network namespace
6338 for_each_netdev_safe(net, dev, aux) {
6340 char fb_name[IFNAMSIZ];
6342 /* Ignore unmoveable devices (i.e. loopback) */
6343 if (dev->features & NETIF_F_NETNS_LOCAL)
6346 /* Leave virtual devices for the generic cleanup */
6347 if (dev->rtnl_link_ops)
6350 /* Push remaining network devices to init_net */
6351 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6352 err = dev_change_net_namespace(dev, &init_net, fb_name);
6354 pr_emerg("%s: failed to move %s to init_net: %d\n",
6355 __func__, dev->name, err);
6362 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6364 /* At exit all network devices most be removed from a network
6365 * namespace. Do this in the reverse order of registration.
6366 * Do this across as many network namespaces as possible to
6367 * improve batching efficiency.
6369 struct net_device *dev;
6371 LIST_HEAD(dev_kill_list);
6374 list_for_each_entry(net, net_list, exit_list) {
6375 for_each_netdev_reverse(net, dev) {
6376 if (dev->rtnl_link_ops)
6377 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6379 unregister_netdevice_queue(dev, &dev_kill_list);
6382 unregister_netdevice_many(&dev_kill_list);
6383 list_del(&dev_kill_list);
6387 static struct pernet_operations __net_initdata default_device_ops = {
6388 .exit = default_device_exit,
6389 .exit_batch = default_device_exit_batch,
6393 * Initialize the DEV module. At boot time this walks the device list and
6394 * unhooks any devices that fail to initialise (normally hardware not
6395 * present) and leaves us with a valid list of present and active devices.
6400 * This is called single threaded during boot, so no need
6401 * to take the rtnl semaphore.
6403 static int __init net_dev_init(void)
6405 int i, rc = -ENOMEM;
6407 BUG_ON(!dev_boot_phase);
6409 if (dev_proc_init())
6412 if (netdev_kobject_init())
6415 INIT_LIST_HEAD(&ptype_all);
6416 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6417 INIT_LIST_HEAD(&ptype_base[i]);
6419 INIT_LIST_HEAD(&offload_base);
6421 if (register_pernet_subsys(&netdev_net_ops))
6425 * Initialise the packet receive queues.
6428 for_each_possible_cpu(i) {
6429 struct softnet_data *sd = &per_cpu(softnet_data, i);
6431 memset(sd, 0, sizeof(*sd));
6432 skb_queue_head_init(&sd->input_pkt_queue);
6433 skb_queue_head_init(&sd->process_queue);
6434 sd->completion_queue = NULL;
6435 INIT_LIST_HEAD(&sd->poll_list);
6436 sd->output_queue = NULL;
6437 sd->output_queue_tailp = &sd->output_queue;
6439 sd->csd.func = rps_trigger_softirq;
6445 sd->backlog.poll = process_backlog;
6446 sd->backlog.weight = weight_p;
6447 sd->backlog.gro_list = NULL;
6448 sd->backlog.gro_count = 0;
6450 #ifdef CONFIG_NET_FLOW_LIMIT
6451 sd->flow_limit = NULL;
6457 /* The loopback device is special if any other network devices
6458 * is present in a network namespace the loopback device must
6459 * be present. Since we now dynamically allocate and free the
6460 * loopback device ensure this invariant is maintained by
6461 * keeping the loopback device as the first device on the
6462 * list of network devices. Ensuring the loopback devices
6463 * is the first device that appears and the last network device
6466 if (register_pernet_device(&loopback_net_ops))
6469 if (register_pernet_device(&default_device_ops))
6472 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6473 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6475 hotcpu_notifier(dev_cpu_callback, 0);
6482 subsys_initcall(net_dev_init);