2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
151 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
154 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
156 * Writers must hold the rtnl semaphore while they loop through the
157 * dev_base_head list, and hold dev_base_lock for writing when they do the
158 * actual updates. This allows pure readers to access the list even
159 * while a writer is preparing to update it.
161 * To put it another way, dev_base_lock is held for writing only to
162 * protect against pure readers; the rtnl semaphore provides the
163 * protection against other writers.
165 * See, for example usages, register_netdevice() and
166 * unregister_netdevice(), which must be called with the rtnl
169 DEFINE_RWLOCK(dev_base_lock);
170 EXPORT_SYMBOL(dev_base_lock);
172 /* protects napi_hash addition/deletion and napi_gen_id */
173 static DEFINE_SPINLOCK(napi_hash_lock);
175 static unsigned int napi_gen_id;
176 static DEFINE_HASHTABLE(napi_hash, 8);
178 static seqcount_t devnet_rename_seq;
180 static inline void dev_base_seq_inc(struct net *net)
182 while (++net->dev_base_seq == 0);
185 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
187 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
189 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
192 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
194 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
197 static inline void rps_lock(struct softnet_data *sd)
200 spin_lock(&sd->input_pkt_queue.lock);
204 static inline void rps_unlock(struct softnet_data *sd)
207 spin_unlock(&sd->input_pkt_queue.lock);
211 /* Device list insertion */
212 static void list_netdevice(struct net_device *dev)
214 struct net *net = dev_net(dev);
218 write_lock_bh(&dev_base_lock);
219 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
220 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
221 hlist_add_head_rcu(&dev->index_hlist,
222 dev_index_hash(net, dev->ifindex));
223 write_unlock_bh(&dev_base_lock);
225 dev_base_seq_inc(net);
228 /* Device list removal
229 * caller must respect a RCU grace period before freeing/reusing dev
231 static void unlist_netdevice(struct net_device *dev)
235 /* Unlink dev from the device chain */
236 write_lock_bh(&dev_base_lock);
237 list_del_rcu(&dev->dev_list);
238 hlist_del_rcu(&dev->name_hlist);
239 hlist_del_rcu(&dev->index_hlist);
240 write_unlock_bh(&dev_base_lock);
242 dev_base_seq_inc(dev_net(dev));
249 static RAW_NOTIFIER_HEAD(netdev_chain);
252 * Device drivers call our routines to queue packets here. We empty the
253 * queue in the local softnet handler.
256 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
257 EXPORT_PER_CPU_SYMBOL(softnet_data);
259 #ifdef CONFIG_LOCKDEP
261 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
262 * according to dev->type
264 static const unsigned short netdev_lock_type[] =
265 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
266 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
267 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
268 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
269 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
270 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
271 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
272 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
273 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
274 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
275 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
276 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
277 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
278 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
279 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
281 static const char *const netdev_lock_name[] =
282 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
283 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
284 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
285 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
286 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
287 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
288 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
289 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
290 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
291 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
292 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
293 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
294 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
295 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
296 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
298 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
299 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
305 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
306 if (netdev_lock_type[i] == dev_type)
308 /* the last key is used by default */
309 return ARRAY_SIZE(netdev_lock_type) - 1;
312 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
313 unsigned short dev_type)
317 i = netdev_lock_pos(dev_type);
318 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
319 netdev_lock_name[i]);
322 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
326 i = netdev_lock_pos(dev->type);
327 lockdep_set_class_and_name(&dev->addr_list_lock,
328 &netdev_addr_lock_key[i],
329 netdev_lock_name[i]);
332 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
333 unsigned short dev_type)
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 /*******************************************************************************
343 Protocol management and registration routines
345 *******************************************************************************/
348 * Add a protocol ID to the list. Now that the input handler is
349 * smarter we can dispense with all the messy stuff that used to be
352 * BEWARE!!! Protocol handlers, mangling input packets,
353 * MUST BE last in hash buckets and checking protocol handlers
354 * MUST start from promiscuous ptype_all chain in net_bh.
355 * It is true now, do not change it.
356 * Explanation follows: if protocol handler, mangling packet, will
357 * be the first on list, it is not able to sense, that packet
358 * is cloned and should be copied-on-write, so that it will
359 * change it and subsequent readers will get broken packet.
363 static inline struct list_head *ptype_head(const struct packet_type *pt)
365 if (pt->type == htons(ETH_P_ALL))
368 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
372 * dev_add_pack - add packet handler
373 * @pt: packet type declaration
375 * Add a protocol handler to the networking stack. The passed &packet_type
376 * is linked into kernel lists and may not be freed until it has been
377 * removed from the kernel lists.
379 * This call does not sleep therefore it can not
380 * guarantee all CPU's that are in middle of receiving packets
381 * will see the new packet type (until the next received packet).
384 void dev_add_pack(struct packet_type *pt)
386 struct list_head *head = ptype_head(pt);
388 spin_lock(&ptype_lock);
389 list_add_rcu(&pt->list, head);
390 spin_unlock(&ptype_lock);
392 EXPORT_SYMBOL(dev_add_pack);
395 * __dev_remove_pack - remove packet handler
396 * @pt: packet type declaration
398 * Remove a protocol handler that was previously added to the kernel
399 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
400 * from the kernel lists and can be freed or reused once this function
403 * The packet type might still be in use by receivers
404 * and must not be freed until after all the CPU's have gone
405 * through a quiescent state.
407 void __dev_remove_pack(struct packet_type *pt)
409 struct list_head *head = ptype_head(pt);
410 struct packet_type *pt1;
412 spin_lock(&ptype_lock);
414 list_for_each_entry(pt1, head, list) {
416 list_del_rcu(&pt->list);
421 pr_warn("dev_remove_pack: %p not found\n", pt);
423 spin_unlock(&ptype_lock);
425 EXPORT_SYMBOL(__dev_remove_pack);
428 * dev_remove_pack - remove packet handler
429 * @pt: packet type declaration
431 * Remove a protocol handler that was previously added to the kernel
432 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
433 * from the kernel lists and can be freed or reused once this function
436 * This call sleeps to guarantee that no CPU is looking at the packet
439 void dev_remove_pack(struct packet_type *pt)
441 __dev_remove_pack(pt);
445 EXPORT_SYMBOL(dev_remove_pack);
449 * dev_add_offload - register offload handlers
450 * @po: protocol offload declaration
452 * Add protocol offload handlers to the networking stack. The passed
453 * &proto_offload is linked into kernel lists and may not be freed until
454 * it has been removed from the kernel lists.
456 * This call does not sleep therefore it can not
457 * guarantee all CPU's that are in middle of receiving packets
458 * will see the new offload handlers (until the next received packet).
460 void dev_add_offload(struct packet_offload *po)
462 struct list_head *head = &offload_base;
464 spin_lock(&offload_lock);
465 list_add_rcu(&po->list, head);
466 spin_unlock(&offload_lock);
468 EXPORT_SYMBOL(dev_add_offload);
471 * __dev_remove_offload - remove offload handler
472 * @po: packet offload declaration
474 * Remove a protocol offload handler that was previously added to the
475 * kernel offload handlers by dev_add_offload(). The passed &offload_type
476 * is removed from the kernel lists and can be freed or reused once this
479 * The packet type might still be in use by receivers
480 * and must not be freed until after all the CPU's have gone
481 * through a quiescent state.
483 void __dev_remove_offload(struct packet_offload *po)
485 struct list_head *head = &offload_base;
486 struct packet_offload *po1;
488 spin_lock(&offload_lock);
490 list_for_each_entry(po1, head, list) {
492 list_del_rcu(&po->list);
497 pr_warn("dev_remove_offload: %p not found\n", po);
499 spin_unlock(&offload_lock);
501 EXPORT_SYMBOL(__dev_remove_offload);
504 * dev_remove_offload - remove packet offload handler
505 * @po: packet offload declaration
507 * Remove a packet offload handler that was previously added to the kernel
508 * offload handlers by dev_add_offload(). The passed &offload_type is
509 * removed from the kernel lists and can be freed or reused once this
512 * This call sleeps to guarantee that no CPU is looking at the packet
515 void dev_remove_offload(struct packet_offload *po)
517 __dev_remove_offload(po);
521 EXPORT_SYMBOL(dev_remove_offload);
523 /******************************************************************************
525 Device Boot-time Settings Routines
527 *******************************************************************************/
529 /* Boot time configuration table */
530 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
533 * netdev_boot_setup_add - add new setup entry
534 * @name: name of the device
535 * @map: configured settings for the device
537 * Adds new setup entry to the dev_boot_setup list. The function
538 * returns 0 on error and 1 on success. This is a generic routine to
541 static int netdev_boot_setup_add(char *name, struct ifmap *map)
543 struct netdev_boot_setup *s;
547 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
548 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
549 memset(s[i].name, 0, sizeof(s[i].name));
550 strlcpy(s[i].name, name, IFNAMSIZ);
551 memcpy(&s[i].map, map, sizeof(s[i].map));
556 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
560 * netdev_boot_setup_check - check boot time settings
561 * @dev: the netdevice
563 * Check boot time settings for the device.
564 * The found settings are set for the device to be used
565 * later in the device probing.
566 * Returns 0 if no settings found, 1 if they are.
568 int netdev_boot_setup_check(struct net_device *dev)
570 struct netdev_boot_setup *s = dev_boot_setup;
573 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
574 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
575 !strcmp(dev->name, s[i].name)) {
576 dev->irq = s[i].map.irq;
577 dev->base_addr = s[i].map.base_addr;
578 dev->mem_start = s[i].map.mem_start;
579 dev->mem_end = s[i].map.mem_end;
585 EXPORT_SYMBOL(netdev_boot_setup_check);
589 * netdev_boot_base - get address from boot time settings
590 * @prefix: prefix for network device
591 * @unit: id for network device
593 * Check boot time settings for the base address of device.
594 * The found settings are set for the device to be used
595 * later in the device probing.
596 * Returns 0 if no settings found.
598 unsigned long netdev_boot_base(const char *prefix, int unit)
600 const struct netdev_boot_setup *s = dev_boot_setup;
604 sprintf(name, "%s%d", prefix, unit);
607 * If device already registered then return base of 1
608 * to indicate not to probe for this interface
610 if (__dev_get_by_name(&init_net, name))
613 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
614 if (!strcmp(name, s[i].name))
615 return s[i].map.base_addr;
620 * Saves at boot time configured settings for any netdevice.
622 int __init netdev_boot_setup(char *str)
627 str = get_options(str, ARRAY_SIZE(ints), ints);
632 memset(&map, 0, sizeof(map));
636 map.base_addr = ints[2];
638 map.mem_start = ints[3];
640 map.mem_end = ints[4];
642 /* Add new entry to the list */
643 return netdev_boot_setup_add(str, &map);
646 __setup("netdev=", netdev_boot_setup);
648 /*******************************************************************************
650 Device Interface Subroutines
652 *******************************************************************************/
655 * __dev_get_by_name - find a device by its name
656 * @net: the applicable net namespace
657 * @name: name to find
659 * Find an interface by name. Must be called under RTNL semaphore
660 * or @dev_base_lock. If the name is found a pointer to the device
661 * is returned. If the name is not found then %NULL is returned. The
662 * reference counters are not incremented so the caller must be
663 * careful with locks.
666 struct net_device *__dev_get_by_name(struct net *net, const char *name)
668 struct net_device *dev;
669 struct hlist_head *head = dev_name_hash(net, name);
671 hlist_for_each_entry(dev, head, name_hlist)
672 if (!strncmp(dev->name, name, IFNAMSIZ))
677 EXPORT_SYMBOL(__dev_get_by_name);
680 * dev_get_by_name_rcu - find a device by its name
681 * @net: the applicable net namespace
682 * @name: name to find
684 * Find an interface by name.
685 * If the name is found a pointer to the device is returned.
686 * If the name is not found then %NULL is returned.
687 * The reference counters are not incremented so the caller must be
688 * careful with locks. The caller must hold RCU lock.
691 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
693 struct net_device *dev;
694 struct hlist_head *head = dev_name_hash(net, name);
696 hlist_for_each_entry_rcu(dev, head, name_hlist)
697 if (!strncmp(dev->name, name, IFNAMSIZ))
702 EXPORT_SYMBOL(dev_get_by_name_rcu);
705 * dev_get_by_name - find a device by its name
706 * @net: the applicable net namespace
707 * @name: name to find
709 * Find an interface by name. This can be called from any
710 * context and does its own locking. The returned handle has
711 * the usage count incremented and the caller must use dev_put() to
712 * release it when it is no longer needed. %NULL is returned if no
713 * matching device is found.
716 struct net_device *dev_get_by_name(struct net *net, const char *name)
718 struct net_device *dev;
721 dev = dev_get_by_name_rcu(net, name);
727 EXPORT_SYMBOL(dev_get_by_name);
730 * __dev_get_by_index - find a device by its ifindex
731 * @net: the applicable net namespace
732 * @ifindex: index of device
734 * Search for an interface by index. Returns %NULL if the device
735 * is not found or a pointer to the device. The device has not
736 * had its reference counter increased so the caller must be careful
737 * about locking. The caller must hold either the RTNL semaphore
741 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
743 struct net_device *dev;
744 struct hlist_head *head = dev_index_hash(net, ifindex);
746 hlist_for_each_entry(dev, head, index_hlist)
747 if (dev->ifindex == ifindex)
752 EXPORT_SYMBOL(__dev_get_by_index);
755 * dev_get_by_index_rcu - find a device by its ifindex
756 * @net: the applicable net namespace
757 * @ifindex: index of device
759 * Search for an interface by index. Returns %NULL if the device
760 * is not found or a pointer to the device. The device has not
761 * had its reference counter increased so the caller must be careful
762 * about locking. The caller must hold RCU lock.
765 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
767 struct net_device *dev;
768 struct hlist_head *head = dev_index_hash(net, ifindex);
770 hlist_for_each_entry_rcu(dev, head, index_hlist)
771 if (dev->ifindex == ifindex)
776 EXPORT_SYMBOL(dev_get_by_index_rcu);
780 * dev_get_by_index - find a device by its ifindex
781 * @net: the applicable net namespace
782 * @ifindex: index of device
784 * Search for an interface by index. Returns NULL if the device
785 * is not found or a pointer to the device. The device returned has
786 * had a reference added and the pointer is safe until the user calls
787 * dev_put to indicate they have finished with it.
790 struct net_device *dev_get_by_index(struct net *net, int ifindex)
792 struct net_device *dev;
795 dev = dev_get_by_index_rcu(net, ifindex);
801 EXPORT_SYMBOL(dev_get_by_index);
804 * netdev_get_name - get a netdevice name, knowing its ifindex.
805 * @net: network namespace
806 * @name: a pointer to the buffer where the name will be stored.
807 * @ifindex: the ifindex of the interface to get the name from.
809 * The use of raw_seqcount_begin() and cond_resched() before
810 * retrying is required as we want to give the writers a chance
811 * to complete when CONFIG_PREEMPT is not set.
813 int netdev_get_name(struct net *net, char *name, int ifindex)
815 struct net_device *dev;
819 seq = raw_seqcount_begin(&devnet_rename_seq);
821 dev = dev_get_by_index_rcu(net, ifindex);
827 strcpy(name, dev->name);
829 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
838 * dev_getbyhwaddr_rcu - find a device by its hardware address
839 * @net: the applicable net namespace
840 * @type: media type of device
841 * @ha: hardware address
843 * Search for an interface by MAC address. Returns NULL if the device
844 * is not found or a pointer to the device.
845 * The caller must hold RCU or RTNL.
846 * The returned device has not had its ref count increased
847 * and the caller must therefore be careful about locking
851 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
854 struct net_device *dev;
856 for_each_netdev_rcu(net, dev)
857 if (dev->type == type &&
858 !memcmp(dev->dev_addr, ha, dev->addr_len))
863 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
865 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
867 struct net_device *dev;
870 for_each_netdev(net, dev)
871 if (dev->type == type)
876 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
878 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
880 struct net_device *dev, *ret = NULL;
883 for_each_netdev_rcu(net, dev)
884 if (dev->type == type) {
892 EXPORT_SYMBOL(dev_getfirstbyhwtype);
895 * dev_get_by_flags_rcu - find any device with given flags
896 * @net: the applicable net namespace
897 * @if_flags: IFF_* values
898 * @mask: bitmask of bits in if_flags to check
900 * Search for any interface with the given flags. Returns NULL if a device
901 * is not found or a pointer to the device. Must be called inside
902 * rcu_read_lock(), and result refcount is unchanged.
905 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
908 struct net_device *dev, *ret;
911 for_each_netdev_rcu(net, dev) {
912 if (((dev->flags ^ if_flags) & mask) == 0) {
919 EXPORT_SYMBOL(dev_get_by_flags_rcu);
922 * dev_valid_name - check if name is okay for network device
925 * Network device names need to be valid file names to
926 * to allow sysfs to work. We also disallow any kind of
929 bool dev_valid_name(const char *name)
933 if (strlen(name) >= IFNAMSIZ)
935 if (!strcmp(name, ".") || !strcmp(name, ".."))
939 if (*name == '/' || isspace(*name))
945 EXPORT_SYMBOL(dev_valid_name);
948 * __dev_alloc_name - allocate a name for a device
949 * @net: network namespace to allocate the device name in
950 * @name: name format string
951 * @buf: scratch buffer and result name string
953 * Passed a format string - eg "lt%d" it will try and find a suitable
954 * id. It scans list of devices to build up a free map, then chooses
955 * the first empty slot. The caller must hold the dev_base or rtnl lock
956 * while allocating the name and adding the device in order to avoid
958 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
959 * Returns the number of the unit assigned or a negative errno code.
962 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
966 const int max_netdevices = 8*PAGE_SIZE;
967 unsigned long *inuse;
968 struct net_device *d;
970 p = strnchr(name, IFNAMSIZ-1, '%');
973 * Verify the string as this thing may have come from
974 * the user. There must be either one "%d" and no other "%"
977 if (p[1] != 'd' || strchr(p + 2, '%'))
980 /* Use one page as a bit array of possible slots */
981 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
985 for_each_netdev(net, d) {
986 if (!sscanf(d->name, name, &i))
988 if (i < 0 || i >= max_netdevices)
991 /* avoid cases where sscanf is not exact inverse of printf */
992 snprintf(buf, IFNAMSIZ, name, i);
993 if (!strncmp(buf, d->name, IFNAMSIZ))
997 i = find_first_zero_bit(inuse, max_netdevices);
998 free_page((unsigned long) inuse);
1002 snprintf(buf, IFNAMSIZ, name, i);
1003 if (!__dev_get_by_name(net, buf))
1006 /* It is possible to run out of possible slots
1007 * when the name is long and there isn't enough space left
1008 * for the digits, or if all bits are used.
1014 * dev_alloc_name - allocate a name for a device
1016 * @name: name format string
1018 * Passed a format string - eg "lt%d" it will try and find a suitable
1019 * id. It scans list of devices to build up a free map, then chooses
1020 * the first empty slot. The caller must hold the dev_base or rtnl lock
1021 * while allocating the name and adding the device in order to avoid
1023 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1024 * Returns the number of the unit assigned or a negative errno code.
1027 int dev_alloc_name(struct net_device *dev, const char *name)
1033 BUG_ON(!dev_net(dev));
1035 ret = __dev_alloc_name(net, name, buf);
1037 strlcpy(dev->name, buf, IFNAMSIZ);
1040 EXPORT_SYMBOL(dev_alloc_name);
1042 static int dev_alloc_name_ns(struct net *net,
1043 struct net_device *dev,
1049 ret = __dev_alloc_name(net, name, buf);
1051 strlcpy(dev->name, buf, IFNAMSIZ);
1055 static int dev_get_valid_name(struct net *net,
1056 struct net_device *dev,
1061 if (!dev_valid_name(name))
1064 if (strchr(name, '%'))
1065 return dev_alloc_name_ns(net, dev, name);
1066 else if (__dev_get_by_name(net, name))
1068 else if (dev->name != name)
1069 strlcpy(dev->name, name, IFNAMSIZ);
1075 * dev_change_name - change name of a device
1077 * @newname: name (or format string) must be at least IFNAMSIZ
1079 * Change name of a device, can pass format strings "eth%d".
1082 int dev_change_name(struct net_device *dev, const char *newname)
1084 char oldname[IFNAMSIZ];
1090 BUG_ON(!dev_net(dev));
1093 if (dev->flags & IFF_UP)
1096 write_seqcount_begin(&devnet_rename_seq);
1098 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1099 write_seqcount_end(&devnet_rename_seq);
1103 memcpy(oldname, dev->name, IFNAMSIZ);
1105 err = dev_get_valid_name(net, dev, newname);
1107 write_seqcount_end(&devnet_rename_seq);
1112 ret = device_rename(&dev->dev, dev->name);
1114 memcpy(dev->name, oldname, IFNAMSIZ);
1115 write_seqcount_end(&devnet_rename_seq);
1119 write_seqcount_end(&devnet_rename_seq);
1121 write_lock_bh(&dev_base_lock);
1122 hlist_del_rcu(&dev->name_hlist);
1123 write_unlock_bh(&dev_base_lock);
1127 write_lock_bh(&dev_base_lock);
1128 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1129 write_unlock_bh(&dev_base_lock);
1131 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1132 ret = notifier_to_errno(ret);
1135 /* err >= 0 after dev_alloc_name() or stores the first errno */
1138 write_seqcount_begin(&devnet_rename_seq);
1139 memcpy(dev->name, oldname, IFNAMSIZ);
1142 pr_err("%s: name change rollback failed: %d\n",
1151 * dev_set_alias - change ifalias of a device
1153 * @alias: name up to IFALIASZ
1154 * @len: limit of bytes to copy from info
1156 * Set ifalias for a device,
1158 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1164 if (len >= IFALIASZ)
1168 kfree(dev->ifalias);
1169 dev->ifalias = NULL;
1173 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1176 dev->ifalias = new_ifalias;
1178 strlcpy(dev->ifalias, alias, len+1);
1184 * netdev_features_change - device changes features
1185 * @dev: device to cause notification
1187 * Called to indicate a device has changed features.
1189 void netdev_features_change(struct net_device *dev)
1191 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1193 EXPORT_SYMBOL(netdev_features_change);
1196 * netdev_state_change - device changes state
1197 * @dev: device to cause notification
1199 * Called to indicate a device has changed state. This function calls
1200 * the notifier chains for netdev_chain and sends a NEWLINK message
1201 * to the routing socket.
1203 void netdev_state_change(struct net_device *dev)
1205 if (dev->flags & IFF_UP) {
1206 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1207 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1210 EXPORT_SYMBOL(netdev_state_change);
1213 * netdev_notify_peers - notify network peers about existence of @dev
1214 * @dev: network device
1216 * Generate traffic such that interested network peers are aware of
1217 * @dev, such as by generating a gratuitous ARP. This may be used when
1218 * a device wants to inform the rest of the network about some sort of
1219 * reconfiguration such as a failover event or virtual machine
1222 void netdev_notify_peers(struct net_device *dev)
1225 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1228 EXPORT_SYMBOL(netdev_notify_peers);
1230 static int __dev_open(struct net_device *dev)
1232 const struct net_device_ops *ops = dev->netdev_ops;
1237 if (!netif_device_present(dev))
1240 /* Block netpoll from trying to do any rx path servicing.
1241 * If we don't do this there is a chance ndo_poll_controller
1242 * or ndo_poll may be running while we open the device
1244 netpoll_rx_disable(dev);
1246 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1247 ret = notifier_to_errno(ret);
1251 set_bit(__LINK_STATE_START, &dev->state);
1253 if (ops->ndo_validate_addr)
1254 ret = ops->ndo_validate_addr(dev);
1256 if (!ret && ops->ndo_open)
1257 ret = ops->ndo_open(dev);
1259 netpoll_rx_enable(dev);
1262 clear_bit(__LINK_STATE_START, &dev->state);
1264 dev->flags |= IFF_UP;
1265 net_dmaengine_get();
1266 dev_set_rx_mode(dev);
1268 add_device_randomness(dev->dev_addr, dev->addr_len);
1275 * dev_open - prepare an interface for use.
1276 * @dev: device to open
1278 * Takes a device from down to up state. The device's private open
1279 * function is invoked and then the multicast lists are loaded. Finally
1280 * the device is moved into the up state and a %NETDEV_UP message is
1281 * sent to the netdev notifier chain.
1283 * Calling this function on an active interface is a nop. On a failure
1284 * a negative errno code is returned.
1286 int dev_open(struct net_device *dev)
1290 if (dev->flags & IFF_UP)
1293 ret = __dev_open(dev);
1297 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1298 call_netdevice_notifiers(NETDEV_UP, dev);
1302 EXPORT_SYMBOL(dev_open);
1304 static int __dev_close_many(struct list_head *head)
1306 struct net_device *dev;
1311 list_for_each_entry(dev, head, close_list) {
1312 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1314 clear_bit(__LINK_STATE_START, &dev->state);
1316 /* Synchronize to scheduled poll. We cannot touch poll list, it
1317 * can be even on different cpu. So just clear netif_running().
1319 * dev->stop() will invoke napi_disable() on all of it's
1320 * napi_struct instances on this device.
1322 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1325 dev_deactivate_many(head);
1327 list_for_each_entry(dev, head, close_list) {
1328 const struct net_device_ops *ops = dev->netdev_ops;
1331 * Call the device specific close. This cannot fail.
1332 * Only if device is UP
1334 * We allow it to be called even after a DETACH hot-plug
1340 dev->flags &= ~IFF_UP;
1341 net_dmaengine_put();
1347 static int __dev_close(struct net_device *dev)
1352 /* Temporarily disable netpoll until the interface is down */
1353 netpoll_rx_disable(dev);
1355 list_add(&dev->close_list, &single);
1356 retval = __dev_close_many(&single);
1359 netpoll_rx_enable(dev);
1363 static int dev_close_many(struct list_head *head)
1365 struct net_device *dev, *tmp;
1367 /* Remove the devices that don't need to be closed */
1368 list_for_each_entry_safe(dev, tmp, head, close_list)
1369 if (!(dev->flags & IFF_UP))
1370 list_del_init(&dev->close_list);
1372 __dev_close_many(head);
1374 list_for_each_entry_safe(dev, tmp, head, close_list) {
1375 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1376 call_netdevice_notifiers(NETDEV_DOWN, dev);
1377 list_del_init(&dev->close_list);
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->close_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 /* the same for macvlan devices */
1429 if (netif_is_macvlan(dev))
1430 dev = macvlan_dev_real_dev(dev);
1432 dev->wanted_features &= ~NETIF_F_LRO;
1433 netdev_update_features(dev);
1435 if (unlikely(dev->features & NETIF_F_LRO))
1436 netdev_WARN(dev, "failed to disable LRO!\n");
1438 EXPORT_SYMBOL(dev_disable_lro);
1440 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1441 struct net_device *dev)
1443 struct netdev_notifier_info info;
1445 netdev_notifier_info_init(&info, dev);
1446 return nb->notifier_call(nb, val, &info);
1449 static int dev_boot_phase = 1;
1452 * register_netdevice_notifier - register a network notifier block
1455 * Register a notifier to be called when network device events occur.
1456 * The notifier passed is linked into the kernel structures and must
1457 * not be reused until it has been unregistered. A negative errno code
1458 * is returned on a failure.
1460 * When registered all registration and up events are replayed
1461 * to the new notifier to allow device to have a race free
1462 * view of the network device list.
1465 int register_netdevice_notifier(struct notifier_block *nb)
1467 struct net_device *dev;
1468 struct net_device *last;
1473 err = raw_notifier_chain_register(&netdev_chain, nb);
1479 for_each_netdev(net, dev) {
1480 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1481 err = notifier_to_errno(err);
1485 if (!(dev->flags & IFF_UP))
1488 call_netdevice_notifier(nb, NETDEV_UP, dev);
1499 for_each_netdev(net, dev) {
1503 if (dev->flags & IFF_UP) {
1504 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1506 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1508 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1513 raw_notifier_chain_unregister(&netdev_chain, nb);
1516 EXPORT_SYMBOL(register_netdevice_notifier);
1519 * unregister_netdevice_notifier - unregister a network notifier block
1522 * Unregister a notifier previously registered by
1523 * register_netdevice_notifier(). The notifier is unlinked into the
1524 * kernel structures and may then be reused. A negative errno code
1525 * is returned on a failure.
1527 * After unregistering unregister and down device events are synthesized
1528 * for all devices on the device list to the removed notifier to remove
1529 * the need for special case cleanup code.
1532 int unregister_netdevice_notifier(struct notifier_block *nb)
1534 struct net_device *dev;
1539 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1544 for_each_netdev(net, dev) {
1545 if (dev->flags & IFF_UP) {
1546 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1548 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1550 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1557 EXPORT_SYMBOL(unregister_netdevice_notifier);
1560 * call_netdevice_notifiers_info - call all network notifier blocks
1561 * @val: value passed unmodified to notifier function
1562 * @dev: net_device pointer passed unmodified to notifier function
1563 * @info: notifier information data
1565 * Call all network notifier blocks. Parameters and return value
1566 * are as for raw_notifier_call_chain().
1569 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1570 struct netdev_notifier_info *info)
1573 netdev_notifier_info_init(info, dev);
1574 return raw_notifier_call_chain(&netdev_chain, val, info);
1576 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1579 * call_netdevice_notifiers - call all network notifier blocks
1580 * @val: value passed unmodified to notifier function
1581 * @dev: net_device pointer passed unmodified to notifier function
1583 * Call all network notifier blocks. Parameters and return value
1584 * are as for raw_notifier_call_chain().
1587 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1589 struct netdev_notifier_info info;
1591 return call_netdevice_notifiers_info(val, dev, &info);
1593 EXPORT_SYMBOL(call_netdevice_notifiers);
1595 static struct static_key netstamp_needed __read_mostly;
1596 #ifdef HAVE_JUMP_LABEL
1597 /* We are not allowed to call static_key_slow_dec() from irq context
1598 * If net_disable_timestamp() is called from irq context, defer the
1599 * static_key_slow_dec() calls.
1601 static atomic_t netstamp_needed_deferred;
1604 void net_enable_timestamp(void)
1606 #ifdef HAVE_JUMP_LABEL
1607 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1611 static_key_slow_dec(&netstamp_needed);
1615 static_key_slow_inc(&netstamp_needed);
1617 EXPORT_SYMBOL(net_enable_timestamp);
1619 void net_disable_timestamp(void)
1621 #ifdef HAVE_JUMP_LABEL
1622 if (in_interrupt()) {
1623 atomic_inc(&netstamp_needed_deferred);
1627 static_key_slow_dec(&netstamp_needed);
1629 EXPORT_SYMBOL(net_disable_timestamp);
1631 static inline void net_timestamp_set(struct sk_buff *skb)
1633 skb->tstamp.tv64 = 0;
1634 if (static_key_false(&netstamp_needed))
1635 __net_timestamp(skb);
1638 #define net_timestamp_check(COND, SKB) \
1639 if (static_key_false(&netstamp_needed)) { \
1640 if ((COND) && !(SKB)->tstamp.tv64) \
1641 __net_timestamp(SKB); \
1644 static inline bool is_skb_forwardable(struct net_device *dev,
1645 struct sk_buff *skb)
1649 if (!(dev->flags & IFF_UP))
1652 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1653 if (skb->len <= len)
1656 /* if TSO is enabled, we don't care about the length as the packet
1657 * could be forwarded without being segmented before
1659 if (skb_is_gso(skb))
1666 * dev_forward_skb - loopback an skb to another netif
1668 * @dev: destination network device
1669 * @skb: buffer to forward
1672 * NET_RX_SUCCESS (no congestion)
1673 * NET_RX_DROP (packet was dropped, but freed)
1675 * dev_forward_skb can be used for injecting an skb from the
1676 * start_xmit function of one device into the receive queue
1677 * of another device.
1679 * The receiving device may be in another namespace, so
1680 * we have to clear all information in the skb that could
1681 * impact namespace isolation.
1683 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1685 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1686 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1687 atomic_long_inc(&dev->rx_dropped);
1693 if (unlikely(!is_skb_forwardable(dev, skb))) {
1694 atomic_long_inc(&dev->rx_dropped);
1699 skb_scrub_packet(skb, true);
1700 skb->protocol = eth_type_trans(skb, dev);
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, const struct cpumask *mask,
1923 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1924 struct xps_map *map, *new_map;
1925 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1926 int cpu, numa_node_id = -2;
1927 bool active = false;
1929 mutex_lock(&xps_map_mutex);
1931 dev_maps = xmap_dereference(dev->xps_maps);
1933 /* allocate memory for queue storage */
1934 for_each_online_cpu(cpu) {
1935 if (!cpumask_test_cpu(cpu, mask))
1939 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1940 if (!new_dev_maps) {
1941 mutex_unlock(&xps_map_mutex);
1945 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1948 map = expand_xps_map(map, cpu, index);
1952 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1956 goto out_no_new_maps;
1958 for_each_possible_cpu(cpu) {
1959 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1960 /* add queue to CPU maps */
1963 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1964 while ((pos < map->len) && (map->queues[pos] != index))
1967 if (pos == map->len)
1968 map->queues[map->len++] = index;
1970 if (numa_node_id == -2)
1971 numa_node_id = cpu_to_node(cpu);
1972 else if (numa_node_id != cpu_to_node(cpu))
1975 } else if (dev_maps) {
1976 /* fill in the new device map from the old device map */
1977 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1978 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1983 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1985 /* Cleanup old maps */
1987 for_each_possible_cpu(cpu) {
1988 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1989 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1990 if (map && map != new_map)
1991 kfree_rcu(map, rcu);
1994 kfree_rcu(dev_maps, rcu);
1997 dev_maps = new_dev_maps;
2001 /* update Tx queue numa node */
2002 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2003 (numa_node_id >= 0) ? numa_node_id :
2009 /* removes queue from unused CPUs */
2010 for_each_possible_cpu(cpu) {
2011 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2014 if (remove_xps_queue(dev_maps, cpu, index))
2018 /* free map if not active */
2020 RCU_INIT_POINTER(dev->xps_maps, NULL);
2021 kfree_rcu(dev_maps, rcu);
2025 mutex_unlock(&xps_map_mutex);
2029 /* remove any maps that we added */
2030 for_each_possible_cpu(cpu) {
2031 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2032 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2034 if (new_map && new_map != map)
2038 mutex_unlock(&xps_map_mutex);
2040 kfree(new_dev_maps);
2043 EXPORT_SYMBOL(netif_set_xps_queue);
2047 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2048 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2050 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2054 if (txq < 1 || txq > dev->num_tx_queues)
2057 if (dev->reg_state == NETREG_REGISTERED ||
2058 dev->reg_state == NETREG_UNREGISTERING) {
2061 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2067 netif_setup_tc(dev, txq);
2069 if (txq < dev->real_num_tx_queues) {
2070 qdisc_reset_all_tx_gt(dev, txq);
2072 netif_reset_xps_queues_gt(dev, txq);
2077 dev->real_num_tx_queues = txq;
2080 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2084 * netif_set_real_num_rx_queues - set actual number of RX queues used
2085 * @dev: Network device
2086 * @rxq: Actual number of RX queues
2088 * This must be called either with the rtnl_lock held or before
2089 * registration of the net device. Returns 0 on success, or a
2090 * negative error code. If called before registration, it always
2093 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2097 if (rxq < 1 || rxq > dev->num_rx_queues)
2100 if (dev->reg_state == NETREG_REGISTERED) {
2103 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2109 dev->real_num_rx_queues = rxq;
2112 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2116 * netif_get_num_default_rss_queues - default number of RSS queues
2118 * This routine should set an upper limit on the number of RSS queues
2119 * used by default by multiqueue devices.
2121 int netif_get_num_default_rss_queues(void)
2123 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2125 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2127 static inline void __netif_reschedule(struct Qdisc *q)
2129 struct softnet_data *sd;
2130 unsigned long flags;
2132 local_irq_save(flags);
2133 sd = &__get_cpu_var(softnet_data);
2134 q->next_sched = NULL;
2135 *sd->output_queue_tailp = q;
2136 sd->output_queue_tailp = &q->next_sched;
2137 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2138 local_irq_restore(flags);
2141 void __netif_schedule(struct Qdisc *q)
2143 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2144 __netif_reschedule(q);
2146 EXPORT_SYMBOL(__netif_schedule);
2148 void dev_kfree_skb_irq(struct sk_buff *skb)
2150 if (atomic_dec_and_test(&skb->users)) {
2151 struct softnet_data *sd;
2152 unsigned long flags;
2154 local_irq_save(flags);
2155 sd = &__get_cpu_var(softnet_data);
2156 skb->next = sd->completion_queue;
2157 sd->completion_queue = skb;
2158 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2159 local_irq_restore(flags);
2162 EXPORT_SYMBOL(dev_kfree_skb_irq);
2164 void dev_kfree_skb_any(struct sk_buff *skb)
2166 if (in_irq() || irqs_disabled())
2167 dev_kfree_skb_irq(skb);
2171 EXPORT_SYMBOL(dev_kfree_skb_any);
2175 * netif_device_detach - mark device as removed
2176 * @dev: network device
2178 * Mark device as removed from system and therefore no longer available.
2180 void netif_device_detach(struct net_device *dev)
2182 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2183 netif_running(dev)) {
2184 netif_tx_stop_all_queues(dev);
2187 EXPORT_SYMBOL(netif_device_detach);
2190 * netif_device_attach - mark device as attached
2191 * @dev: network device
2193 * Mark device as attached from system and restart if needed.
2195 void netif_device_attach(struct net_device *dev)
2197 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2198 netif_running(dev)) {
2199 netif_tx_wake_all_queues(dev);
2200 __netdev_watchdog_up(dev);
2203 EXPORT_SYMBOL(netif_device_attach);
2205 static void skb_warn_bad_offload(const struct sk_buff *skb)
2207 static const netdev_features_t null_features = 0;
2208 struct net_device *dev = skb->dev;
2209 const char *driver = "";
2211 if (!net_ratelimit())
2214 if (dev && dev->dev.parent)
2215 driver = dev_driver_string(dev->dev.parent);
2217 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2218 "gso_type=%d ip_summed=%d\n",
2219 driver, dev ? &dev->features : &null_features,
2220 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2221 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2222 skb_shinfo(skb)->gso_type, skb->ip_summed);
2226 * Invalidate hardware checksum when packet is to be mangled, and
2227 * complete checksum manually on outgoing path.
2229 int skb_checksum_help(struct sk_buff *skb)
2232 int ret = 0, offset;
2234 if (skb->ip_summed == CHECKSUM_COMPLETE)
2235 goto out_set_summed;
2237 if (unlikely(skb_shinfo(skb)->gso_size)) {
2238 skb_warn_bad_offload(skb);
2242 /* Before computing a checksum, we should make sure no frag could
2243 * be modified by an external entity : checksum could be wrong.
2245 if (skb_has_shared_frag(skb)) {
2246 ret = __skb_linearize(skb);
2251 offset = skb_checksum_start_offset(skb);
2252 BUG_ON(offset >= skb_headlen(skb));
2253 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2255 offset += skb->csum_offset;
2256 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2258 if (skb_cloned(skb) &&
2259 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2260 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2265 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2267 skb->ip_summed = CHECKSUM_NONE;
2271 EXPORT_SYMBOL(skb_checksum_help);
2273 __be16 skb_network_protocol(struct sk_buff *skb)
2275 __be16 type = skb->protocol;
2276 int vlan_depth = ETH_HLEN;
2278 /* Tunnel gso handlers can set protocol to ethernet. */
2279 if (type == htons(ETH_P_TEB)) {
2282 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2285 eth = (struct ethhdr *)skb_mac_header(skb);
2286 type = eth->h_proto;
2289 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2290 struct vlan_hdr *vh;
2292 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2295 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2296 type = vh->h_vlan_encapsulated_proto;
2297 vlan_depth += VLAN_HLEN;
2304 * skb_mac_gso_segment - mac layer segmentation handler.
2305 * @skb: buffer to segment
2306 * @features: features for the output path (see dev->features)
2308 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2309 netdev_features_t features)
2311 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2312 struct packet_offload *ptype;
2313 __be16 type = skb_network_protocol(skb);
2315 if (unlikely(!type))
2316 return ERR_PTR(-EINVAL);
2318 __skb_pull(skb, skb->mac_len);
2321 list_for_each_entry_rcu(ptype, &offload_base, list) {
2322 if (ptype->type == type && ptype->callbacks.gso_segment) {
2323 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2326 err = ptype->callbacks.gso_send_check(skb);
2327 segs = ERR_PTR(err);
2328 if (err || skb_gso_ok(skb, features))
2330 __skb_push(skb, (skb->data -
2331 skb_network_header(skb)));
2333 segs = ptype->callbacks.gso_segment(skb, features);
2339 __skb_push(skb, skb->data - skb_mac_header(skb));
2343 EXPORT_SYMBOL(skb_mac_gso_segment);
2346 /* openvswitch calls this on rx path, so we need a different check.
2348 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2351 return skb->ip_summed != CHECKSUM_PARTIAL;
2353 return skb->ip_summed == CHECKSUM_NONE;
2357 * __skb_gso_segment - Perform segmentation on skb.
2358 * @skb: buffer to segment
2359 * @features: features for the output path (see dev->features)
2360 * @tx_path: whether it is called in TX path
2362 * This function segments the given skb and returns a list of segments.
2364 * It may return NULL if the skb requires no segmentation. This is
2365 * only possible when GSO is used for verifying header integrity.
2367 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2368 netdev_features_t features, bool tx_path)
2370 if (unlikely(skb_needs_check(skb, tx_path))) {
2373 skb_warn_bad_offload(skb);
2375 if (skb_header_cloned(skb) &&
2376 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2377 return ERR_PTR(err);
2380 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2381 SKB_GSO_CB(skb)->encap_level = 0;
2383 skb_reset_mac_header(skb);
2384 skb_reset_mac_len(skb);
2386 return skb_mac_gso_segment(skb, features);
2388 EXPORT_SYMBOL(__skb_gso_segment);
2390 /* Take action when hardware reception checksum errors are detected. */
2392 void netdev_rx_csum_fault(struct net_device *dev)
2394 if (net_ratelimit()) {
2395 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2399 EXPORT_SYMBOL(netdev_rx_csum_fault);
2402 /* Actually, we should eliminate this check as soon as we know, that:
2403 * 1. IOMMU is present and allows to map all the memory.
2404 * 2. No high memory really exists on this machine.
2407 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2409 #ifdef CONFIG_HIGHMEM
2411 if (!(dev->features & NETIF_F_HIGHDMA)) {
2412 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2413 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2414 if (PageHighMem(skb_frag_page(frag)))
2419 if (PCI_DMA_BUS_IS_PHYS) {
2420 struct device *pdev = dev->dev.parent;
2424 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2425 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2426 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2427 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2436 void (*destructor)(struct sk_buff *skb);
2439 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2441 static void dev_gso_skb_destructor(struct sk_buff *skb)
2443 struct dev_gso_cb *cb;
2446 struct sk_buff *nskb = skb->next;
2448 skb->next = nskb->next;
2451 } while (skb->next);
2453 cb = DEV_GSO_CB(skb);
2455 cb->destructor(skb);
2459 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2460 * @skb: buffer to segment
2461 * @features: device features as applicable to this skb
2463 * This function segments the given skb and stores the list of segments
2466 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2468 struct sk_buff *segs;
2470 segs = skb_gso_segment(skb, features);
2472 /* Verifying header integrity only. */
2477 return PTR_ERR(segs);
2480 DEV_GSO_CB(skb)->destructor = skb->destructor;
2481 skb->destructor = dev_gso_skb_destructor;
2486 static netdev_features_t harmonize_features(struct sk_buff *skb,
2487 netdev_features_t features)
2489 if (skb->ip_summed != CHECKSUM_NONE &&
2490 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2491 features &= ~NETIF_F_ALL_CSUM;
2492 } else if (illegal_highdma(skb->dev, skb)) {
2493 features &= ~NETIF_F_SG;
2499 netdev_features_t netif_skb_features(struct sk_buff *skb)
2501 __be16 protocol = skb->protocol;
2502 netdev_features_t features = skb->dev->features;
2504 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2505 features &= ~NETIF_F_GSO_MASK;
2507 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2508 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2509 protocol = veh->h_vlan_encapsulated_proto;
2510 } else if (!vlan_tx_tag_present(skb)) {
2511 return harmonize_features(skb, features);
2514 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2515 NETIF_F_HW_VLAN_STAG_TX);
2517 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2518 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2519 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2520 NETIF_F_HW_VLAN_STAG_TX;
2522 return harmonize_features(skb, features);
2524 EXPORT_SYMBOL(netif_skb_features);
2527 * Returns true if either:
2528 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2529 * 2. skb is fragmented and the device does not support SG.
2531 static inline int skb_needs_linearize(struct sk_buff *skb,
2532 netdev_features_t features)
2534 return skb_is_nonlinear(skb) &&
2535 ((skb_has_frag_list(skb) &&
2536 !(features & NETIF_F_FRAGLIST)) ||
2537 (skb_shinfo(skb)->nr_frags &&
2538 !(features & NETIF_F_SG)));
2541 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2542 struct netdev_queue *txq)
2544 const struct net_device_ops *ops = dev->netdev_ops;
2545 int rc = NETDEV_TX_OK;
2546 unsigned int skb_len;
2548 if (likely(!skb->next)) {
2549 netdev_features_t features;
2552 * If device doesn't need skb->dst, release it right now while
2553 * its hot in this cpu cache
2555 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2558 features = netif_skb_features(skb);
2560 if (vlan_tx_tag_present(skb) &&
2561 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2562 skb = __vlan_put_tag(skb, skb->vlan_proto,
2563 vlan_tx_tag_get(skb));
2570 /* If encapsulation offload request, verify we are testing
2571 * hardware encapsulation features instead of standard
2572 * features for the netdev
2574 if (skb->encapsulation)
2575 features &= dev->hw_enc_features;
2577 if (netif_needs_gso(skb, features)) {
2578 if (unlikely(dev_gso_segment(skb, features)))
2583 if (skb_needs_linearize(skb, features) &&
2584 __skb_linearize(skb))
2587 /* If packet is not checksummed and device does not
2588 * support checksumming for this protocol, complete
2589 * checksumming here.
2591 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2592 if (skb->encapsulation)
2593 skb_set_inner_transport_header(skb,
2594 skb_checksum_start_offset(skb));
2596 skb_set_transport_header(skb,
2597 skb_checksum_start_offset(skb));
2598 if (!(features & NETIF_F_ALL_CSUM) &&
2599 skb_checksum_help(skb))
2604 if (!list_empty(&ptype_all))
2605 dev_queue_xmit_nit(skb, dev);
2608 rc = ops->ndo_start_xmit(skb, dev);
2610 trace_net_dev_xmit(skb, rc, dev, skb_len);
2611 if (rc == NETDEV_TX_OK)
2612 txq_trans_update(txq);
2618 struct sk_buff *nskb = skb->next;
2620 skb->next = nskb->next;
2623 if (!list_empty(&ptype_all))
2624 dev_queue_xmit_nit(nskb, dev);
2626 skb_len = nskb->len;
2627 rc = ops->ndo_start_xmit(nskb, dev);
2628 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2629 if (unlikely(rc != NETDEV_TX_OK)) {
2630 if (rc & ~NETDEV_TX_MASK)
2631 goto out_kfree_gso_skb;
2632 nskb->next = skb->next;
2636 txq_trans_update(txq);
2637 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2638 return NETDEV_TX_BUSY;
2639 } while (skb->next);
2642 if (likely(skb->next == NULL)) {
2643 skb->destructor = DEV_GSO_CB(skb)->destructor;
2652 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2654 static void qdisc_pkt_len_init(struct sk_buff *skb)
2656 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2658 qdisc_skb_cb(skb)->pkt_len = skb->len;
2660 /* To get more precise estimation of bytes sent on wire,
2661 * we add to pkt_len the headers size of all segments
2663 if (shinfo->gso_size) {
2664 unsigned int hdr_len;
2665 u16 gso_segs = shinfo->gso_segs;
2667 /* mac layer + network layer */
2668 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2670 /* + transport layer */
2671 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2672 hdr_len += tcp_hdrlen(skb);
2674 hdr_len += sizeof(struct udphdr);
2676 if (shinfo->gso_type & SKB_GSO_DODGY)
2677 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2680 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2684 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2685 struct net_device *dev,
2686 struct netdev_queue *txq)
2688 spinlock_t *root_lock = qdisc_lock(q);
2692 qdisc_pkt_len_init(skb);
2693 qdisc_calculate_pkt_len(skb, q);
2695 * Heuristic to force contended enqueues to serialize on a
2696 * separate lock before trying to get qdisc main lock.
2697 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2698 * and dequeue packets faster.
2700 contended = qdisc_is_running(q);
2701 if (unlikely(contended))
2702 spin_lock(&q->busylock);
2704 spin_lock(root_lock);
2705 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2708 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2709 qdisc_run_begin(q)) {
2711 * This is a work-conserving queue; there are no old skbs
2712 * waiting to be sent out; and the qdisc is not running -
2713 * xmit the skb directly.
2715 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2718 qdisc_bstats_update(q, skb);
2720 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2721 if (unlikely(contended)) {
2722 spin_unlock(&q->busylock);
2729 rc = NET_XMIT_SUCCESS;
2732 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2733 if (qdisc_run_begin(q)) {
2734 if (unlikely(contended)) {
2735 spin_unlock(&q->busylock);
2741 spin_unlock(root_lock);
2742 if (unlikely(contended))
2743 spin_unlock(&q->busylock);
2747 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2748 static void skb_update_prio(struct sk_buff *skb)
2750 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2752 if (!skb->priority && skb->sk && map) {
2753 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2755 if (prioidx < map->priomap_len)
2756 skb->priority = map->priomap[prioidx];
2760 #define skb_update_prio(skb)
2763 static DEFINE_PER_CPU(int, xmit_recursion);
2764 #define RECURSION_LIMIT 10
2767 * dev_loopback_xmit - loop back @skb
2768 * @skb: buffer to transmit
2770 int dev_loopback_xmit(struct sk_buff *skb)
2772 skb_reset_mac_header(skb);
2773 __skb_pull(skb, skb_network_offset(skb));
2774 skb->pkt_type = PACKET_LOOPBACK;
2775 skb->ip_summed = CHECKSUM_UNNECESSARY;
2776 WARN_ON(!skb_dst(skb));
2781 EXPORT_SYMBOL(dev_loopback_xmit);
2784 * dev_queue_xmit - transmit a buffer
2785 * @skb: buffer to transmit
2787 * Queue a buffer for transmission to a network device. The caller must
2788 * have set the device and priority and built the buffer before calling
2789 * this function. The function can be called from an interrupt.
2791 * A negative errno code is returned on a failure. A success does not
2792 * guarantee the frame will be transmitted as it may be dropped due
2793 * to congestion or traffic shaping.
2795 * -----------------------------------------------------------------------------------
2796 * I notice this method can also return errors from the queue disciplines,
2797 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2800 * Regardless of the return value, the skb is consumed, so it is currently
2801 * difficult to retry a send to this method. (You can bump the ref count
2802 * before sending to hold a reference for retry if you are careful.)
2804 * When calling this method, interrupts MUST be enabled. This is because
2805 * the BH enable code must have IRQs enabled so that it will not deadlock.
2808 int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2810 struct net_device *dev = skb->dev;
2811 struct netdev_queue *txq;
2815 skb_reset_mac_header(skb);
2817 /* Disable soft irqs for various locks below. Also
2818 * stops preemption for RCU.
2822 skb_update_prio(skb);
2824 txq = netdev_pick_tx(dev, skb, accel_priv);
2825 q = rcu_dereference_bh(txq->qdisc);
2827 #ifdef CONFIG_NET_CLS_ACT
2828 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2830 trace_net_dev_queue(skb);
2832 rc = __dev_xmit_skb(skb, q, dev, txq);
2836 /* The device has no queue. Common case for software devices:
2837 loopback, all the sorts of tunnels...
2839 Really, it is unlikely that netif_tx_lock protection is necessary
2840 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2842 However, it is possible, that they rely on protection
2845 Check this and shot the lock. It is not prone from deadlocks.
2846 Either shot noqueue qdisc, it is even simpler 8)
2848 if (dev->flags & IFF_UP) {
2849 int cpu = smp_processor_id(); /* ok because BHs are off */
2851 if (txq->xmit_lock_owner != cpu) {
2853 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2854 goto recursion_alert;
2856 HARD_TX_LOCK(dev, txq, cpu);
2858 if (!netif_xmit_stopped(txq)) {
2859 __this_cpu_inc(xmit_recursion);
2860 rc = dev_hard_start_xmit(skb, dev, txq);
2861 __this_cpu_dec(xmit_recursion);
2862 if (dev_xmit_complete(rc)) {
2863 HARD_TX_UNLOCK(dev, txq);
2867 HARD_TX_UNLOCK(dev, txq);
2868 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2871 /* Recursion is detected! It is possible,
2875 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2881 rcu_read_unlock_bh();
2886 rcu_read_unlock_bh();
2890 int dev_queue_xmit(struct sk_buff *skb)
2892 return __dev_queue_xmit(skb, NULL);
2894 EXPORT_SYMBOL(dev_queue_xmit);
2896 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2898 return __dev_queue_xmit(skb, accel_priv);
2900 EXPORT_SYMBOL(dev_queue_xmit_accel);
2903 /*=======================================================================
2905 =======================================================================*/
2907 int netdev_max_backlog __read_mostly = 1000;
2908 EXPORT_SYMBOL(netdev_max_backlog);
2910 int netdev_tstamp_prequeue __read_mostly = 1;
2911 int netdev_budget __read_mostly = 300;
2912 int weight_p __read_mostly = 64; /* old backlog weight */
2914 /* Called with irq disabled */
2915 static inline void ____napi_schedule(struct softnet_data *sd,
2916 struct napi_struct *napi)
2918 list_add_tail(&napi->poll_list, &sd->poll_list);
2919 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2924 /* One global table that all flow-based protocols share. */
2925 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2926 EXPORT_SYMBOL(rps_sock_flow_table);
2928 struct static_key rps_needed __read_mostly;
2930 static struct rps_dev_flow *
2931 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2932 struct rps_dev_flow *rflow, u16 next_cpu)
2934 if (next_cpu != RPS_NO_CPU) {
2935 #ifdef CONFIG_RFS_ACCEL
2936 struct netdev_rx_queue *rxqueue;
2937 struct rps_dev_flow_table *flow_table;
2938 struct rps_dev_flow *old_rflow;
2943 /* Should we steer this flow to a different hardware queue? */
2944 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2945 !(dev->features & NETIF_F_NTUPLE))
2947 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2948 if (rxq_index == skb_get_rx_queue(skb))
2951 rxqueue = dev->_rx + rxq_index;
2952 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2955 flow_id = skb->rxhash & flow_table->mask;
2956 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2957 rxq_index, flow_id);
2961 rflow = &flow_table->flows[flow_id];
2963 if (old_rflow->filter == rflow->filter)
2964 old_rflow->filter = RPS_NO_FILTER;
2968 per_cpu(softnet_data, next_cpu).input_queue_head;
2971 rflow->cpu = next_cpu;
2976 * get_rps_cpu is called from netif_receive_skb and returns the target
2977 * CPU from the RPS map of the receiving queue for a given skb.
2978 * rcu_read_lock must be held on entry.
2980 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2981 struct rps_dev_flow **rflowp)
2983 struct netdev_rx_queue *rxqueue;
2984 struct rps_map *map;
2985 struct rps_dev_flow_table *flow_table;
2986 struct rps_sock_flow_table *sock_flow_table;
2990 if (skb_rx_queue_recorded(skb)) {
2991 u16 index = skb_get_rx_queue(skb);
2992 if (unlikely(index >= dev->real_num_rx_queues)) {
2993 WARN_ONCE(dev->real_num_rx_queues > 1,
2994 "%s received packet on queue %u, but number "
2995 "of RX queues is %u\n",
2996 dev->name, index, dev->real_num_rx_queues);
2999 rxqueue = dev->_rx + index;
3003 map = rcu_dereference(rxqueue->rps_map);
3005 if (map->len == 1 &&
3006 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3007 tcpu = map->cpus[0];
3008 if (cpu_online(tcpu))
3012 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3016 skb_reset_network_header(skb);
3017 if (!skb_get_rxhash(skb))
3020 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3021 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3022 if (flow_table && sock_flow_table) {
3024 struct rps_dev_flow *rflow;
3026 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3029 next_cpu = sock_flow_table->ents[skb->rxhash &
3030 sock_flow_table->mask];
3033 * If the desired CPU (where last recvmsg was done) is
3034 * different from current CPU (one in the rx-queue flow
3035 * table entry), switch if one of the following holds:
3036 * - Current CPU is unset (equal to RPS_NO_CPU).
3037 * - Current CPU is offline.
3038 * - The current CPU's queue tail has advanced beyond the
3039 * last packet that was enqueued using this table entry.
3040 * This guarantees that all previous packets for the flow
3041 * have been dequeued, thus preserving in order delivery.
3043 if (unlikely(tcpu != next_cpu) &&
3044 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3045 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3046 rflow->last_qtail)) >= 0)) {
3048 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3051 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3059 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3061 if (cpu_online(tcpu)) {
3071 #ifdef CONFIG_RFS_ACCEL
3074 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3075 * @dev: Device on which the filter was set
3076 * @rxq_index: RX queue index
3077 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3078 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3080 * Drivers that implement ndo_rx_flow_steer() should periodically call
3081 * this function for each installed filter and remove the filters for
3082 * which it returns %true.
3084 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3085 u32 flow_id, u16 filter_id)
3087 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3088 struct rps_dev_flow_table *flow_table;
3089 struct rps_dev_flow *rflow;
3094 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3095 if (flow_table && flow_id <= flow_table->mask) {
3096 rflow = &flow_table->flows[flow_id];
3097 cpu = ACCESS_ONCE(rflow->cpu);
3098 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3099 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3100 rflow->last_qtail) <
3101 (int)(10 * flow_table->mask)))
3107 EXPORT_SYMBOL(rps_may_expire_flow);
3109 #endif /* CONFIG_RFS_ACCEL */
3111 /* Called from hardirq (IPI) context */
3112 static void rps_trigger_softirq(void *data)
3114 struct softnet_data *sd = data;
3116 ____napi_schedule(sd, &sd->backlog);
3120 #endif /* CONFIG_RPS */
3123 * Check if this softnet_data structure is another cpu one
3124 * If yes, queue it to our IPI list and return 1
3127 static int rps_ipi_queued(struct softnet_data *sd)
3130 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3133 sd->rps_ipi_next = mysd->rps_ipi_list;
3134 mysd->rps_ipi_list = sd;
3136 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3139 #endif /* CONFIG_RPS */
3143 #ifdef CONFIG_NET_FLOW_LIMIT
3144 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3147 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3149 #ifdef CONFIG_NET_FLOW_LIMIT
3150 struct sd_flow_limit *fl;
3151 struct softnet_data *sd;
3152 unsigned int old_flow, new_flow;
3154 if (qlen < (netdev_max_backlog >> 1))
3157 sd = &__get_cpu_var(softnet_data);
3160 fl = rcu_dereference(sd->flow_limit);
3162 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3163 old_flow = fl->history[fl->history_head];
3164 fl->history[fl->history_head] = new_flow;
3167 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3169 if (likely(fl->buckets[old_flow]))
3170 fl->buckets[old_flow]--;
3172 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3184 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3185 * queue (may be a remote CPU queue).
3187 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3188 unsigned int *qtail)
3190 struct softnet_data *sd;
3191 unsigned long flags;
3194 sd = &per_cpu(softnet_data, cpu);
3196 local_irq_save(flags);
3199 qlen = skb_queue_len(&sd->input_pkt_queue);
3200 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3201 if (skb_queue_len(&sd->input_pkt_queue)) {
3203 __skb_queue_tail(&sd->input_pkt_queue, skb);
3204 input_queue_tail_incr_save(sd, qtail);
3206 local_irq_restore(flags);
3207 return NET_RX_SUCCESS;
3210 /* Schedule NAPI for backlog device
3211 * We can use non atomic operation since we own the queue lock
3213 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3214 if (!rps_ipi_queued(sd))
3215 ____napi_schedule(sd, &sd->backlog);
3223 local_irq_restore(flags);
3225 atomic_long_inc(&skb->dev->rx_dropped);
3231 * netif_rx - post buffer to the network code
3232 * @skb: buffer to post
3234 * This function receives a packet from a device driver and queues it for
3235 * the upper (protocol) levels to process. It always succeeds. The buffer
3236 * may be dropped during processing for congestion control or by the
3240 * NET_RX_SUCCESS (no congestion)
3241 * NET_RX_DROP (packet was dropped)
3245 int netif_rx(struct sk_buff *skb)
3249 /* if netpoll wants it, pretend we never saw it */
3250 if (netpoll_rx(skb))
3253 net_timestamp_check(netdev_tstamp_prequeue, skb);
3255 trace_netif_rx(skb);
3257 if (static_key_false(&rps_needed)) {
3258 struct rps_dev_flow voidflow, *rflow = &voidflow;
3264 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3266 cpu = smp_processor_id();
3268 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3276 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3281 EXPORT_SYMBOL(netif_rx);
3283 int netif_rx_ni(struct sk_buff *skb)
3288 err = netif_rx(skb);
3289 if (local_softirq_pending())
3295 EXPORT_SYMBOL(netif_rx_ni);
3297 static void net_tx_action(struct softirq_action *h)
3299 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3301 if (sd->completion_queue) {
3302 struct sk_buff *clist;
3304 local_irq_disable();
3305 clist = sd->completion_queue;
3306 sd->completion_queue = NULL;
3310 struct sk_buff *skb = clist;
3311 clist = clist->next;
3313 WARN_ON(atomic_read(&skb->users));
3314 trace_kfree_skb(skb, net_tx_action);
3319 if (sd->output_queue) {
3322 local_irq_disable();
3323 head = sd->output_queue;
3324 sd->output_queue = NULL;
3325 sd->output_queue_tailp = &sd->output_queue;
3329 struct Qdisc *q = head;
3330 spinlock_t *root_lock;
3332 head = head->next_sched;
3334 root_lock = qdisc_lock(q);
3335 if (spin_trylock(root_lock)) {
3336 smp_mb__before_clear_bit();
3337 clear_bit(__QDISC_STATE_SCHED,
3340 spin_unlock(root_lock);
3342 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3344 __netif_reschedule(q);
3346 smp_mb__before_clear_bit();
3347 clear_bit(__QDISC_STATE_SCHED,
3355 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3356 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3357 /* This hook is defined here for ATM LANE */
3358 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3359 unsigned char *addr) __read_mostly;
3360 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3363 #ifdef CONFIG_NET_CLS_ACT
3364 /* TODO: Maybe we should just force sch_ingress to be compiled in
3365 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3366 * a compare and 2 stores extra right now if we dont have it on
3367 * but have CONFIG_NET_CLS_ACT
3368 * NOTE: This doesn't stop any functionality; if you dont have
3369 * the ingress scheduler, you just can't add policies on ingress.
3372 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3374 struct net_device *dev = skb->dev;
3375 u32 ttl = G_TC_RTTL(skb->tc_verd);
3376 int result = TC_ACT_OK;
3379 if (unlikely(MAX_RED_LOOP < ttl++)) {
3380 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3381 skb->skb_iif, dev->ifindex);
3385 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3386 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3389 if (q != &noop_qdisc) {
3390 spin_lock(qdisc_lock(q));
3391 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3392 result = qdisc_enqueue_root(skb, q);
3393 spin_unlock(qdisc_lock(q));
3399 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3400 struct packet_type **pt_prev,
3401 int *ret, struct net_device *orig_dev)
3403 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3405 if (!rxq || rxq->qdisc == &noop_qdisc)
3409 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3413 switch (ing_filter(skb, rxq)) {
3427 * netdev_rx_handler_register - register receive handler
3428 * @dev: device to register a handler for
3429 * @rx_handler: receive handler to register
3430 * @rx_handler_data: data pointer that is used by rx handler
3432 * Register a receive hander for a device. This handler will then be
3433 * called from __netif_receive_skb. A negative errno code is returned
3436 * The caller must hold the rtnl_mutex.
3438 * For a general description of rx_handler, see enum rx_handler_result.
3440 int netdev_rx_handler_register(struct net_device *dev,
3441 rx_handler_func_t *rx_handler,
3442 void *rx_handler_data)
3446 if (dev->rx_handler)
3449 /* Note: rx_handler_data must be set before rx_handler */
3450 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3451 rcu_assign_pointer(dev->rx_handler, rx_handler);
3455 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3458 * netdev_rx_handler_unregister - unregister receive handler
3459 * @dev: device to unregister a handler from
3461 * Unregister a receive handler from a device.
3463 * The caller must hold the rtnl_mutex.
3465 void netdev_rx_handler_unregister(struct net_device *dev)
3469 RCU_INIT_POINTER(dev->rx_handler, NULL);
3470 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3471 * section has a guarantee to see a non NULL rx_handler_data
3475 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3477 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3480 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3481 * the special handling of PFMEMALLOC skbs.
3483 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3485 switch (skb->protocol) {
3486 case __constant_htons(ETH_P_ARP):
3487 case __constant_htons(ETH_P_IP):
3488 case __constant_htons(ETH_P_IPV6):
3489 case __constant_htons(ETH_P_8021Q):
3490 case __constant_htons(ETH_P_8021AD):
3497 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3499 struct packet_type *ptype, *pt_prev;
3500 rx_handler_func_t *rx_handler;
3501 struct net_device *orig_dev;
3502 struct net_device *null_or_dev;
3503 bool deliver_exact = false;
3504 int ret = NET_RX_DROP;
3507 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3509 trace_netif_receive_skb(skb);
3511 /* if we've gotten here through NAPI, check netpoll */
3512 if (netpoll_receive_skb(skb))
3515 orig_dev = skb->dev;
3517 skb_reset_network_header(skb);
3518 if (!skb_transport_header_was_set(skb))
3519 skb_reset_transport_header(skb);
3520 skb_reset_mac_len(skb);
3527 skb->skb_iif = skb->dev->ifindex;
3529 __this_cpu_inc(softnet_data.processed);
3531 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3532 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3533 skb = vlan_untag(skb);
3538 #ifdef CONFIG_NET_CLS_ACT
3539 if (skb->tc_verd & TC_NCLS) {
3540 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3548 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3549 if (!ptype->dev || ptype->dev == skb->dev) {
3551 ret = deliver_skb(skb, pt_prev, orig_dev);
3557 #ifdef CONFIG_NET_CLS_ACT
3558 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3564 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3567 if (vlan_tx_tag_present(skb)) {
3569 ret = deliver_skb(skb, pt_prev, orig_dev);
3572 if (vlan_do_receive(&skb))
3574 else if (unlikely(!skb))
3578 rx_handler = rcu_dereference(skb->dev->rx_handler);
3581 ret = deliver_skb(skb, pt_prev, orig_dev);
3584 switch (rx_handler(&skb)) {
3585 case RX_HANDLER_CONSUMED:
3586 ret = NET_RX_SUCCESS;
3588 case RX_HANDLER_ANOTHER:
3590 case RX_HANDLER_EXACT:
3591 deliver_exact = true;
3592 case RX_HANDLER_PASS:
3599 if (unlikely(vlan_tx_tag_present(skb))) {
3600 if (vlan_tx_tag_get_id(skb))
3601 skb->pkt_type = PACKET_OTHERHOST;
3602 /* Note: we might in the future use prio bits
3603 * and set skb->priority like in vlan_do_receive()
3604 * For the time being, just ignore Priority Code Point
3609 /* deliver only exact match when indicated */
3610 null_or_dev = deliver_exact ? skb->dev : NULL;
3612 type = skb->protocol;
3613 list_for_each_entry_rcu(ptype,
3614 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3615 if (ptype->type == type &&
3616 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3617 ptype->dev == orig_dev)) {
3619 ret = deliver_skb(skb, pt_prev, orig_dev);
3625 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3628 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3631 atomic_long_inc(&skb->dev->rx_dropped);
3633 /* Jamal, now you will not able to escape explaining
3634 * me how you were going to use this. :-)
3645 static int __netif_receive_skb(struct sk_buff *skb)
3649 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3650 unsigned long pflags = current->flags;
3653 * PFMEMALLOC skbs are special, they should
3654 * - be delivered to SOCK_MEMALLOC sockets only
3655 * - stay away from userspace
3656 * - have bounded memory usage
3658 * Use PF_MEMALLOC as this saves us from propagating the allocation
3659 * context down to all allocation sites.
3661 current->flags |= PF_MEMALLOC;
3662 ret = __netif_receive_skb_core(skb, true);
3663 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3665 ret = __netif_receive_skb_core(skb, false);
3671 * netif_receive_skb - process receive buffer from network
3672 * @skb: buffer to process
3674 * netif_receive_skb() is the main receive data processing function.
3675 * It always succeeds. The buffer may be dropped during processing
3676 * for congestion control or by the protocol layers.
3678 * This function may only be called from softirq context and interrupts
3679 * should be enabled.
3681 * Return values (usually ignored):
3682 * NET_RX_SUCCESS: no congestion
3683 * NET_RX_DROP: packet was dropped
3685 int netif_receive_skb(struct sk_buff *skb)
3687 net_timestamp_check(netdev_tstamp_prequeue, skb);
3689 if (skb_defer_rx_timestamp(skb))
3690 return NET_RX_SUCCESS;
3693 if (static_key_false(&rps_needed)) {
3694 struct rps_dev_flow voidflow, *rflow = &voidflow;
3699 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3702 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3709 return __netif_receive_skb(skb);
3711 EXPORT_SYMBOL(netif_receive_skb);
3713 /* Network device is going away, flush any packets still pending
3714 * Called with irqs disabled.
3716 static void flush_backlog(void *arg)
3718 struct net_device *dev = arg;
3719 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3720 struct sk_buff *skb, *tmp;
3723 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3724 if (skb->dev == dev) {
3725 __skb_unlink(skb, &sd->input_pkt_queue);
3727 input_queue_head_incr(sd);
3732 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3733 if (skb->dev == dev) {
3734 __skb_unlink(skb, &sd->process_queue);
3736 input_queue_head_incr(sd);
3741 static int napi_gro_complete(struct sk_buff *skb)
3743 struct packet_offload *ptype;
3744 __be16 type = skb->protocol;
3745 struct list_head *head = &offload_base;
3748 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3750 if (NAPI_GRO_CB(skb)->count == 1) {
3751 skb_shinfo(skb)->gso_size = 0;
3756 list_for_each_entry_rcu(ptype, head, list) {
3757 if (ptype->type != type || !ptype->callbacks.gro_complete)
3760 err = ptype->callbacks.gro_complete(skb);
3766 WARN_ON(&ptype->list == head);
3768 return NET_RX_SUCCESS;
3772 return netif_receive_skb(skb);
3775 /* napi->gro_list contains packets ordered by age.
3776 * youngest packets at the head of it.
3777 * Complete skbs in reverse order to reduce latencies.
3779 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3781 struct sk_buff *skb, *prev = NULL;
3783 /* scan list and build reverse chain */
3784 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3789 for (skb = prev; skb; skb = prev) {
3792 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3796 napi_gro_complete(skb);
3800 napi->gro_list = NULL;
3802 EXPORT_SYMBOL(napi_gro_flush);
3804 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3807 unsigned int maclen = skb->dev->hard_header_len;
3809 for (p = napi->gro_list; p; p = p->next) {
3810 unsigned long diffs;
3812 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3813 diffs |= p->vlan_tci ^ skb->vlan_tci;
3814 if (maclen == ETH_HLEN)
3815 diffs |= compare_ether_header(skb_mac_header(p),
3816 skb_gro_mac_header(skb));
3818 diffs = memcmp(skb_mac_header(p),
3819 skb_gro_mac_header(skb),
3821 NAPI_GRO_CB(p)->same_flow = !diffs;
3822 NAPI_GRO_CB(p)->flush = 0;
3826 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3828 struct sk_buff **pp = NULL;
3829 struct packet_offload *ptype;
3830 __be16 type = skb->protocol;
3831 struct list_head *head = &offload_base;
3833 enum gro_result ret;
3835 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3838 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3841 gro_list_prepare(napi, skb);
3844 list_for_each_entry_rcu(ptype, head, list) {
3845 if (ptype->type != type || !ptype->callbacks.gro_receive)
3848 skb_set_network_header(skb, skb_gro_offset(skb));
3849 skb_reset_mac_len(skb);
3850 NAPI_GRO_CB(skb)->same_flow = 0;
3851 NAPI_GRO_CB(skb)->flush = 0;
3852 NAPI_GRO_CB(skb)->free = 0;
3854 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3859 if (&ptype->list == head)
3862 same_flow = NAPI_GRO_CB(skb)->same_flow;
3863 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3866 struct sk_buff *nskb = *pp;
3870 napi_gro_complete(nskb);
3877 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3881 NAPI_GRO_CB(skb)->count = 1;
3882 NAPI_GRO_CB(skb)->age = jiffies;
3883 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3884 skb->next = napi->gro_list;
3885 napi->gro_list = skb;
3889 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3890 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3892 BUG_ON(skb->end - skb->tail < grow);
3894 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3897 skb->data_len -= grow;
3899 skb_shinfo(skb)->frags[0].page_offset += grow;
3900 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3902 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3903 skb_frag_unref(skb, 0);
3904 memmove(skb_shinfo(skb)->frags,
3905 skb_shinfo(skb)->frags + 1,
3906 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3919 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3923 if (netif_receive_skb(skb))
3931 case GRO_MERGED_FREE:
3932 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3933 kmem_cache_free(skbuff_head_cache, skb);
3946 static void skb_gro_reset_offset(struct sk_buff *skb)
3948 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3949 const skb_frag_t *frag0 = &pinfo->frags[0];
3951 NAPI_GRO_CB(skb)->data_offset = 0;
3952 NAPI_GRO_CB(skb)->frag0 = NULL;
3953 NAPI_GRO_CB(skb)->frag0_len = 0;
3955 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3957 !PageHighMem(skb_frag_page(frag0))) {
3958 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3959 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3963 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3965 skb_gro_reset_offset(skb);
3967 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3969 EXPORT_SYMBOL(napi_gro_receive);
3971 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3973 __skb_pull(skb, skb_headlen(skb));
3974 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3975 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3977 skb->dev = napi->dev;
3983 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3985 struct sk_buff *skb = napi->skb;
3988 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3994 EXPORT_SYMBOL(napi_get_frags);
3996 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4002 skb->protocol = eth_type_trans(skb, skb->dev);
4004 if (ret == GRO_HELD)
4005 skb_gro_pull(skb, -ETH_HLEN);
4006 else if (netif_receive_skb(skb))
4011 case GRO_MERGED_FREE:
4012 napi_reuse_skb(napi, skb);
4022 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4024 struct sk_buff *skb = napi->skb;
4031 skb_reset_mac_header(skb);
4032 skb_gro_reset_offset(skb);
4034 off = skb_gro_offset(skb);
4035 hlen = off + sizeof(*eth);
4036 eth = skb_gro_header_fast(skb, off);
4037 if (skb_gro_header_hard(skb, hlen)) {
4038 eth = skb_gro_header_slow(skb, hlen, off);
4039 if (unlikely(!eth)) {
4040 napi_reuse_skb(napi, skb);
4046 skb_gro_pull(skb, sizeof(*eth));
4049 * This works because the only protocols we care about don't require
4050 * special handling. We'll fix it up properly at the end.
4052 skb->protocol = eth->h_proto;
4058 gro_result_t napi_gro_frags(struct napi_struct *napi)
4060 struct sk_buff *skb = napi_frags_skb(napi);
4065 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4067 EXPORT_SYMBOL(napi_gro_frags);
4070 * net_rps_action sends any pending IPI's for rps.
4071 * Note: called with local irq disabled, but exits with local irq enabled.
4073 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4076 struct softnet_data *remsd = sd->rps_ipi_list;
4079 sd->rps_ipi_list = NULL;
4083 /* Send pending IPI's to kick RPS processing on remote cpus. */
4085 struct softnet_data *next = remsd->rps_ipi_next;
4087 if (cpu_online(remsd->cpu))
4088 __smp_call_function_single(remsd->cpu,
4097 static int process_backlog(struct napi_struct *napi, int quota)
4100 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4103 /* Check if we have pending ipi, its better to send them now,
4104 * not waiting net_rx_action() end.
4106 if (sd->rps_ipi_list) {
4107 local_irq_disable();
4108 net_rps_action_and_irq_enable(sd);
4111 napi->weight = weight_p;
4112 local_irq_disable();
4113 while (work < quota) {
4114 struct sk_buff *skb;
4117 while ((skb = __skb_dequeue(&sd->process_queue))) {
4119 __netif_receive_skb(skb);
4120 local_irq_disable();
4121 input_queue_head_incr(sd);
4122 if (++work >= quota) {
4129 qlen = skb_queue_len(&sd->input_pkt_queue);
4131 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4132 &sd->process_queue);
4134 if (qlen < quota - work) {
4136 * Inline a custom version of __napi_complete().
4137 * only current cpu owns and manipulates this napi,
4138 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4139 * we can use a plain write instead of clear_bit(),
4140 * and we dont need an smp_mb() memory barrier.
4142 list_del(&napi->poll_list);
4145 quota = work + qlen;
4155 * __napi_schedule - schedule for receive
4156 * @n: entry to schedule
4158 * The entry's receive function will be scheduled to run
4160 void __napi_schedule(struct napi_struct *n)
4162 unsigned long flags;
4164 local_irq_save(flags);
4165 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4166 local_irq_restore(flags);
4168 EXPORT_SYMBOL(__napi_schedule);
4170 void __napi_complete(struct napi_struct *n)
4172 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4173 BUG_ON(n->gro_list);
4175 list_del(&n->poll_list);
4176 smp_mb__before_clear_bit();
4177 clear_bit(NAPI_STATE_SCHED, &n->state);
4179 EXPORT_SYMBOL(__napi_complete);
4181 void napi_complete(struct napi_struct *n)
4183 unsigned long flags;
4186 * don't let napi dequeue from the cpu poll list
4187 * just in case its running on a different cpu
4189 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4192 napi_gro_flush(n, false);
4193 local_irq_save(flags);
4195 local_irq_restore(flags);
4197 EXPORT_SYMBOL(napi_complete);
4199 /* must be called under rcu_read_lock(), as we dont take a reference */
4200 struct napi_struct *napi_by_id(unsigned int napi_id)
4202 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4203 struct napi_struct *napi;
4205 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4206 if (napi->napi_id == napi_id)
4211 EXPORT_SYMBOL_GPL(napi_by_id);
4213 void napi_hash_add(struct napi_struct *napi)
4215 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4217 spin_lock(&napi_hash_lock);
4219 /* 0 is not a valid id, we also skip an id that is taken
4220 * we expect both events to be extremely rare
4223 while (!napi->napi_id) {
4224 napi->napi_id = ++napi_gen_id;
4225 if (napi_by_id(napi->napi_id))
4229 hlist_add_head_rcu(&napi->napi_hash_node,
4230 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4232 spin_unlock(&napi_hash_lock);
4235 EXPORT_SYMBOL_GPL(napi_hash_add);
4237 /* Warning : caller is responsible to make sure rcu grace period
4238 * is respected before freeing memory containing @napi
4240 void napi_hash_del(struct napi_struct *napi)
4242 spin_lock(&napi_hash_lock);
4244 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4245 hlist_del_rcu(&napi->napi_hash_node);
4247 spin_unlock(&napi_hash_lock);
4249 EXPORT_SYMBOL_GPL(napi_hash_del);
4251 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4252 int (*poll)(struct napi_struct *, int), int weight)
4254 INIT_LIST_HEAD(&napi->poll_list);
4255 napi->gro_count = 0;
4256 napi->gro_list = NULL;
4259 if (weight > NAPI_POLL_WEIGHT)
4260 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4262 napi->weight = weight;
4263 list_add(&napi->dev_list, &dev->napi_list);
4265 #ifdef CONFIG_NETPOLL
4266 spin_lock_init(&napi->poll_lock);
4267 napi->poll_owner = -1;
4269 set_bit(NAPI_STATE_SCHED, &napi->state);
4271 EXPORT_SYMBOL(netif_napi_add);
4273 void netif_napi_del(struct napi_struct *napi)
4275 struct sk_buff *skb, *next;
4277 list_del_init(&napi->dev_list);
4278 napi_free_frags(napi);
4280 for (skb = napi->gro_list; skb; skb = next) {
4286 napi->gro_list = NULL;
4287 napi->gro_count = 0;
4289 EXPORT_SYMBOL(netif_napi_del);
4291 static void net_rx_action(struct softirq_action *h)
4293 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4294 unsigned long time_limit = jiffies + 2;
4295 int budget = netdev_budget;
4298 local_irq_disable();
4300 while (!list_empty(&sd->poll_list)) {
4301 struct napi_struct *n;
4304 /* If softirq window is exhuasted then punt.
4305 * Allow this to run for 2 jiffies since which will allow
4306 * an average latency of 1.5/HZ.
4308 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4313 /* Even though interrupts have been re-enabled, this
4314 * access is safe because interrupts can only add new
4315 * entries to the tail of this list, and only ->poll()
4316 * calls can remove this head entry from the list.
4318 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4320 have = netpoll_poll_lock(n);
4324 /* This NAPI_STATE_SCHED test is for avoiding a race
4325 * with netpoll's poll_napi(). Only the entity which
4326 * obtains the lock and sees NAPI_STATE_SCHED set will
4327 * actually make the ->poll() call. Therefore we avoid
4328 * accidentally calling ->poll() when NAPI is not scheduled.
4331 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4332 work = n->poll(n, weight);
4336 WARN_ON_ONCE(work > weight);
4340 local_irq_disable();
4342 /* Drivers must not modify the NAPI state if they
4343 * consume the entire weight. In such cases this code
4344 * still "owns" the NAPI instance and therefore can
4345 * move the instance around on the list at-will.
4347 if (unlikely(work == weight)) {
4348 if (unlikely(napi_disable_pending(n))) {
4351 local_irq_disable();
4354 /* flush too old packets
4355 * If HZ < 1000, flush all packets.
4358 napi_gro_flush(n, HZ >= 1000);
4359 local_irq_disable();
4361 list_move_tail(&n->poll_list, &sd->poll_list);
4365 netpoll_poll_unlock(have);
4368 net_rps_action_and_irq_enable(sd);
4370 #ifdef CONFIG_NET_DMA
4372 * There may not be any more sk_buffs coming right now, so push
4373 * any pending DMA copies to hardware
4375 dma_issue_pending_all();
4382 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4386 struct netdev_adjacent {
4387 struct net_device *dev;
4389 /* upper master flag, there can only be one master device per list */
4392 /* counter for the number of times this device was added to us */
4395 /* private field for the users */
4398 struct list_head list;
4399 struct rcu_head rcu;
4402 static struct netdev_adjacent *__netdev_find_adj_rcu(struct net_device *dev,
4403 struct net_device *adj_dev,
4404 struct list_head *adj_list)
4406 struct netdev_adjacent *adj;
4408 list_for_each_entry_rcu(adj, adj_list, list) {
4409 if (adj->dev == adj_dev)
4415 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4416 struct net_device *adj_dev,
4417 struct list_head *adj_list)
4419 struct netdev_adjacent *adj;
4421 list_for_each_entry(adj, adj_list, list) {
4422 if (adj->dev == adj_dev)
4429 * netdev_has_upper_dev - Check if device is linked to an upper device
4431 * @upper_dev: upper device to check
4433 * Find out if a device is linked to specified upper device and return true
4434 * in case it is. Note that this checks only immediate upper device,
4435 * not through a complete stack of devices. The caller must hold the RTNL lock.
4437 bool netdev_has_upper_dev(struct net_device *dev,
4438 struct net_device *upper_dev)
4442 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4444 EXPORT_SYMBOL(netdev_has_upper_dev);
4447 * netdev_has_any_upper_dev - Check if device is linked to some device
4450 * Find out if a device is linked to an upper device and return true in case
4451 * it is. The caller must hold the RTNL lock.
4453 bool netdev_has_any_upper_dev(struct net_device *dev)
4457 return !list_empty(&dev->all_adj_list.upper);
4459 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4462 * netdev_master_upper_dev_get - Get master upper device
4465 * Find a master upper device and return pointer to it or NULL in case
4466 * it's not there. The caller must hold the RTNL lock.
4468 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4470 struct netdev_adjacent *upper;
4474 if (list_empty(&dev->adj_list.upper))
4477 upper = list_first_entry(&dev->adj_list.upper,
4478 struct netdev_adjacent, list);
4479 if (likely(upper->master))
4483 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4485 void *netdev_adjacent_get_private(struct list_head *adj_list)
4487 struct netdev_adjacent *adj;
4489 adj = list_entry(adj_list, struct netdev_adjacent, list);
4491 return adj->private;
4493 EXPORT_SYMBOL(netdev_adjacent_get_private);
4496 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4498 * @iter: list_head ** of the current position
4500 * Gets the next device from the dev's upper list, starting from iter
4501 * position. The caller must hold RCU read lock.
4503 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4504 struct list_head **iter)
4506 struct netdev_adjacent *upper;
4508 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4510 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4512 if (&upper->list == &dev->all_adj_list.upper)
4515 *iter = &upper->list;
4519 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4522 * netdev_lower_get_next_private - Get the next ->private from the
4523 * lower neighbour list
4525 * @iter: list_head ** of the current position
4527 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4528 * list, starting from iter position. The caller must hold either hold the
4529 * RTNL lock or its own locking that guarantees that the neighbour lower
4530 * list will remain unchainged.
4532 void *netdev_lower_get_next_private(struct net_device *dev,
4533 struct list_head **iter)
4535 struct netdev_adjacent *lower;
4537 lower = list_entry(*iter, struct netdev_adjacent, list);
4539 if (&lower->list == &dev->adj_list.lower)
4543 *iter = lower->list.next;
4545 return lower->private;
4547 EXPORT_SYMBOL(netdev_lower_get_next_private);
4550 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4551 * lower neighbour list, RCU
4554 * @iter: list_head ** of the current position
4556 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4557 * list, starting from iter position. The caller must hold RCU read lock.
4559 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4560 struct list_head **iter)
4562 struct netdev_adjacent *lower;
4564 WARN_ON_ONCE(!rcu_read_lock_held());
4566 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4568 if (&lower->list == &dev->adj_list.lower)
4572 *iter = &lower->list;
4574 return lower->private;
4576 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4579 * netdev_master_upper_dev_get_rcu - Get master upper device
4582 * Find a master upper device and return pointer to it or NULL in case
4583 * it's not there. The caller must hold the RCU read lock.
4585 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4587 struct netdev_adjacent *upper;
4589 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4590 struct netdev_adjacent, list);
4591 if (upper && likely(upper->master))
4595 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4597 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4598 struct net_device *adj_dev,
4599 struct list_head *dev_list,
4600 void *private, bool master)
4602 struct netdev_adjacent *adj;
4603 char linkname[IFNAMSIZ+7];
4606 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4613 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4618 adj->master = master;
4620 adj->private = private;
4623 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4624 adj_dev->name, dev->name, adj_dev->name);
4626 if (dev_list == &dev->adj_list.lower) {
4627 sprintf(linkname, "lower_%s", adj_dev->name);
4628 ret = sysfs_create_link(&(dev->dev.kobj),
4629 &(adj_dev->dev.kobj), linkname);
4632 } else if (dev_list == &dev->adj_list.upper) {
4633 sprintf(linkname, "upper_%s", adj_dev->name);
4634 ret = sysfs_create_link(&(dev->dev.kobj),
4635 &(adj_dev->dev.kobj), linkname);
4640 /* Ensure that master link is always the first item in list. */
4642 ret = sysfs_create_link(&(dev->dev.kobj),
4643 &(adj_dev->dev.kobj), "master");
4645 goto remove_symlinks;
4647 list_add_rcu(&adj->list, dev_list);
4649 list_add_tail_rcu(&adj->list, dev_list);
4655 if (dev_list == &dev->adj_list.lower) {
4656 sprintf(linkname, "lower_%s", adj_dev->name);
4657 sysfs_remove_link(&(dev->dev.kobj), linkname);
4658 } else if (dev_list == &dev->adj_list.upper) {
4659 sprintf(linkname, "upper_%s", adj_dev->name);
4660 sysfs_remove_link(&(dev->dev.kobj), linkname);
4670 void __netdev_adjacent_dev_remove(struct net_device *dev,
4671 struct net_device *adj_dev,
4672 struct list_head *dev_list)
4674 struct netdev_adjacent *adj;
4675 char linkname[IFNAMSIZ+7];
4677 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4680 pr_err("tried to remove device %s from %s\n",
4681 dev->name, adj_dev->name);
4685 if (adj->ref_nr > 1) {
4686 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4693 sysfs_remove_link(&(dev->dev.kobj), "master");
4695 if (dev_list == &dev->adj_list.lower) {
4696 sprintf(linkname, "lower_%s", adj_dev->name);
4697 sysfs_remove_link(&(dev->dev.kobj), linkname);
4698 } else if (dev_list == &dev->adj_list.upper) {
4699 sprintf(linkname, "upper_%s", adj_dev->name);
4700 sysfs_remove_link(&(dev->dev.kobj), linkname);
4703 list_del_rcu(&adj->list);
4704 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4705 adj_dev->name, dev->name, adj_dev->name);
4707 kfree_rcu(adj, rcu);
4710 int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4711 struct net_device *upper_dev,
4712 struct list_head *up_list,
4713 struct list_head *down_list,
4714 void *private, bool master)
4718 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4723 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4726 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4733 int __netdev_adjacent_dev_link(struct net_device *dev,
4734 struct net_device *upper_dev)
4736 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4737 &dev->all_adj_list.upper,
4738 &upper_dev->all_adj_list.lower,
4742 void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4743 struct net_device *upper_dev,
4744 struct list_head *up_list,
4745 struct list_head *down_list)
4747 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4748 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4751 void __netdev_adjacent_dev_unlink(struct net_device *dev,
4752 struct net_device *upper_dev)
4754 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4755 &dev->all_adj_list.upper,
4756 &upper_dev->all_adj_list.lower);
4759 int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4760 struct net_device *upper_dev,
4761 void *private, bool master)
4763 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4768 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4769 &dev->adj_list.upper,
4770 &upper_dev->adj_list.lower,
4773 __netdev_adjacent_dev_unlink(dev, upper_dev);
4780 void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4781 struct net_device *upper_dev)
4783 __netdev_adjacent_dev_unlink(dev, upper_dev);
4784 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4785 &dev->adj_list.upper,
4786 &upper_dev->adj_list.lower);
4789 static int __netdev_upper_dev_link(struct net_device *dev,
4790 struct net_device *upper_dev, bool master,
4793 struct netdev_adjacent *i, *j, *to_i, *to_j;
4798 if (dev == upper_dev)
4801 /* To prevent loops, check if dev is not upper device to upper_dev. */
4802 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4805 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4808 if (master && netdev_master_upper_dev_get(dev))
4811 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4816 /* Now that we linked these devs, make all the upper_dev's
4817 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4818 * versa, and don't forget the devices itself. All of these
4819 * links are non-neighbours.
4821 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4822 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4823 pr_debug("Interlinking %s with %s, non-neighbour\n",
4824 i->dev->name, j->dev->name);
4825 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4831 /* add dev to every upper_dev's upper device */
4832 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4833 pr_debug("linking %s's upper device %s with %s\n",
4834 upper_dev->name, i->dev->name, dev->name);
4835 ret = __netdev_adjacent_dev_link(dev, i->dev);
4837 goto rollback_upper_mesh;
4840 /* add upper_dev to every dev's lower device */
4841 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4842 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4843 i->dev->name, upper_dev->name);
4844 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4846 goto rollback_lower_mesh;
4849 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4852 rollback_lower_mesh:
4854 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4857 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4862 rollback_upper_mesh:
4864 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4867 __netdev_adjacent_dev_unlink(dev, i->dev);
4875 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4876 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4877 if (i == to_i && j == to_j)
4879 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4885 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4891 * netdev_upper_dev_link - Add a link to the upper device
4893 * @upper_dev: new upper device
4895 * Adds a link to device which is upper to this one. The caller must hold
4896 * the RTNL lock. On a failure a negative errno code is returned.
4897 * On success the reference counts are adjusted and the function
4900 int netdev_upper_dev_link(struct net_device *dev,
4901 struct net_device *upper_dev)
4903 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4905 EXPORT_SYMBOL(netdev_upper_dev_link);
4908 * netdev_master_upper_dev_link - Add a master link to the upper device
4910 * @upper_dev: new upper device
4912 * Adds a link to device which is upper to this one. In this case, only
4913 * one master upper device can be linked, although other non-master devices
4914 * might be linked as well. The caller must hold the RTNL lock.
4915 * On a failure a negative errno code is returned. On success the reference
4916 * counts are adjusted and the function returns zero.
4918 int netdev_master_upper_dev_link(struct net_device *dev,
4919 struct net_device *upper_dev)
4921 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4923 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4925 int netdev_master_upper_dev_link_private(struct net_device *dev,
4926 struct net_device *upper_dev,
4929 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4931 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4934 * netdev_upper_dev_unlink - Removes a link to upper device
4936 * @upper_dev: new upper device
4938 * Removes a link to device which is upper to this one. The caller must hold
4941 void netdev_upper_dev_unlink(struct net_device *dev,
4942 struct net_device *upper_dev)
4944 struct netdev_adjacent *i, *j;
4947 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4949 /* Here is the tricky part. We must remove all dev's lower
4950 * devices from all upper_dev's upper devices and vice
4951 * versa, to maintain the graph relationship.
4953 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4954 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4955 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4957 /* remove also the devices itself from lower/upper device
4960 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4961 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4963 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4964 __netdev_adjacent_dev_unlink(dev, i->dev);
4966 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4968 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4970 void *netdev_lower_dev_get_private_rcu(struct net_device *dev,
4971 struct net_device *lower_dev)
4973 struct netdev_adjacent *lower;
4977 lower = __netdev_find_adj_rcu(dev, lower_dev, &dev->adj_list.lower);
4981 return lower->private;
4983 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu);
4985 void *netdev_lower_dev_get_private(struct net_device *dev,
4986 struct net_device *lower_dev)
4988 struct netdev_adjacent *lower;
4992 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
4996 return lower->private;
4998 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5000 static void dev_change_rx_flags(struct net_device *dev, int flags)
5002 const struct net_device_ops *ops = dev->netdev_ops;
5004 if (ops->ndo_change_rx_flags)
5005 ops->ndo_change_rx_flags(dev, flags);
5008 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5010 unsigned int old_flags = dev->flags;
5016 dev->flags |= IFF_PROMISC;
5017 dev->promiscuity += inc;
5018 if (dev->promiscuity == 0) {
5021 * If inc causes overflow, untouch promisc and return error.
5024 dev->flags &= ~IFF_PROMISC;
5026 dev->promiscuity -= inc;
5027 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5032 if (dev->flags != old_flags) {
5033 pr_info("device %s %s promiscuous mode\n",
5035 dev->flags & IFF_PROMISC ? "entered" : "left");
5036 if (audit_enabled) {
5037 current_uid_gid(&uid, &gid);
5038 audit_log(current->audit_context, GFP_ATOMIC,
5039 AUDIT_ANOM_PROMISCUOUS,
5040 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5041 dev->name, (dev->flags & IFF_PROMISC),
5042 (old_flags & IFF_PROMISC),
5043 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5044 from_kuid(&init_user_ns, uid),
5045 from_kgid(&init_user_ns, gid),
5046 audit_get_sessionid(current));
5049 dev_change_rx_flags(dev, IFF_PROMISC);
5052 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5057 * dev_set_promiscuity - update promiscuity count on a device
5061 * Add or remove promiscuity from a device. While the count in the device
5062 * remains above zero the interface remains promiscuous. Once it hits zero
5063 * the device reverts back to normal filtering operation. A negative inc
5064 * value is used to drop promiscuity on the device.
5065 * Return 0 if successful or a negative errno code on error.
5067 int dev_set_promiscuity(struct net_device *dev, int inc)
5069 unsigned int old_flags = dev->flags;
5072 err = __dev_set_promiscuity(dev, inc, true);
5075 if (dev->flags != old_flags)
5076 dev_set_rx_mode(dev);
5079 EXPORT_SYMBOL(dev_set_promiscuity);
5081 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5083 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5087 dev->flags |= IFF_ALLMULTI;
5088 dev->allmulti += inc;
5089 if (dev->allmulti == 0) {
5092 * If inc causes overflow, untouch allmulti and return error.
5095 dev->flags &= ~IFF_ALLMULTI;
5097 dev->allmulti -= inc;
5098 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5103 if (dev->flags ^ old_flags) {
5104 dev_change_rx_flags(dev, IFF_ALLMULTI);
5105 dev_set_rx_mode(dev);
5107 __dev_notify_flags(dev, old_flags,
5108 dev->gflags ^ old_gflags);
5114 * dev_set_allmulti - update allmulti count on a device
5118 * Add or remove reception of all multicast frames to a device. While the
5119 * count in the device remains above zero the interface remains listening
5120 * to all interfaces. Once it hits zero the device reverts back to normal
5121 * filtering operation. A negative @inc value is used to drop the counter
5122 * when releasing a resource needing all multicasts.
5123 * Return 0 if successful or a negative errno code on error.
5126 int dev_set_allmulti(struct net_device *dev, int inc)
5128 return __dev_set_allmulti(dev, inc, true);
5130 EXPORT_SYMBOL(dev_set_allmulti);
5133 * Upload unicast and multicast address lists to device and
5134 * configure RX filtering. When the device doesn't support unicast
5135 * filtering it is put in promiscuous mode while unicast addresses
5138 void __dev_set_rx_mode(struct net_device *dev)
5140 const struct net_device_ops *ops = dev->netdev_ops;
5142 /* dev_open will call this function so the list will stay sane. */
5143 if (!(dev->flags&IFF_UP))
5146 if (!netif_device_present(dev))
5149 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5150 /* Unicast addresses changes may only happen under the rtnl,
5151 * therefore calling __dev_set_promiscuity here is safe.
5153 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5154 __dev_set_promiscuity(dev, 1, false);
5155 dev->uc_promisc = true;
5156 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5157 __dev_set_promiscuity(dev, -1, false);
5158 dev->uc_promisc = false;
5162 if (ops->ndo_set_rx_mode)
5163 ops->ndo_set_rx_mode(dev);
5166 void dev_set_rx_mode(struct net_device *dev)
5168 netif_addr_lock_bh(dev);
5169 __dev_set_rx_mode(dev);
5170 netif_addr_unlock_bh(dev);
5174 * dev_get_flags - get flags reported to userspace
5177 * Get the combination of flag bits exported through APIs to userspace.
5179 unsigned int dev_get_flags(const struct net_device *dev)
5183 flags = (dev->flags & ~(IFF_PROMISC |
5188 (dev->gflags & (IFF_PROMISC |
5191 if (netif_running(dev)) {
5192 if (netif_oper_up(dev))
5193 flags |= IFF_RUNNING;
5194 if (netif_carrier_ok(dev))
5195 flags |= IFF_LOWER_UP;
5196 if (netif_dormant(dev))
5197 flags |= IFF_DORMANT;
5202 EXPORT_SYMBOL(dev_get_flags);
5204 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5206 unsigned int old_flags = dev->flags;
5212 * Set the flags on our device.
5215 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5216 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5218 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5222 * Load in the correct multicast list now the flags have changed.
5225 if ((old_flags ^ flags) & IFF_MULTICAST)
5226 dev_change_rx_flags(dev, IFF_MULTICAST);
5228 dev_set_rx_mode(dev);
5231 * Have we downed the interface. We handle IFF_UP ourselves
5232 * according to user attempts to set it, rather than blindly
5237 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5238 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5241 dev_set_rx_mode(dev);
5244 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5245 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5246 unsigned int old_flags = dev->flags;
5248 dev->gflags ^= IFF_PROMISC;
5250 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5251 if (dev->flags != old_flags)
5252 dev_set_rx_mode(dev);
5255 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5256 is important. Some (broken) drivers set IFF_PROMISC, when
5257 IFF_ALLMULTI is requested not asking us and not reporting.
5259 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5260 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5262 dev->gflags ^= IFF_ALLMULTI;
5263 __dev_set_allmulti(dev, inc, false);
5269 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5270 unsigned int gchanges)
5272 unsigned int changes = dev->flags ^ old_flags;
5275 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5277 if (changes & IFF_UP) {
5278 if (dev->flags & IFF_UP)
5279 call_netdevice_notifiers(NETDEV_UP, dev);
5281 call_netdevice_notifiers(NETDEV_DOWN, dev);
5284 if (dev->flags & IFF_UP &&
5285 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5286 struct netdev_notifier_change_info change_info;
5288 change_info.flags_changed = changes;
5289 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5295 * dev_change_flags - change device settings
5297 * @flags: device state flags
5299 * Change settings on device based state flags. The flags are
5300 * in the userspace exported format.
5302 int dev_change_flags(struct net_device *dev, unsigned int flags)
5305 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5307 ret = __dev_change_flags(dev, flags);
5311 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5312 __dev_notify_flags(dev, old_flags, changes);
5315 EXPORT_SYMBOL(dev_change_flags);
5318 * dev_set_mtu - Change maximum transfer unit
5320 * @new_mtu: new transfer unit
5322 * Change the maximum transfer size of the network device.
5324 int dev_set_mtu(struct net_device *dev, int new_mtu)
5326 const struct net_device_ops *ops = dev->netdev_ops;
5329 if (new_mtu == dev->mtu)
5332 /* MTU must be positive. */
5336 if (!netif_device_present(dev))
5340 if (ops->ndo_change_mtu)
5341 err = ops->ndo_change_mtu(dev, new_mtu);
5346 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5349 EXPORT_SYMBOL(dev_set_mtu);
5352 * dev_set_group - Change group this device belongs to
5354 * @new_group: group this device should belong to
5356 void dev_set_group(struct net_device *dev, int new_group)
5358 dev->group = new_group;
5360 EXPORT_SYMBOL(dev_set_group);
5363 * dev_set_mac_address - Change Media Access Control Address
5367 * Change the hardware (MAC) address of the device
5369 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5371 const struct net_device_ops *ops = dev->netdev_ops;
5374 if (!ops->ndo_set_mac_address)
5376 if (sa->sa_family != dev->type)
5378 if (!netif_device_present(dev))
5380 err = ops->ndo_set_mac_address(dev, sa);
5383 dev->addr_assign_type = NET_ADDR_SET;
5384 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5385 add_device_randomness(dev->dev_addr, dev->addr_len);
5388 EXPORT_SYMBOL(dev_set_mac_address);
5391 * dev_change_carrier - Change device carrier
5393 * @new_carrier: new value
5395 * Change device carrier
5397 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5399 const struct net_device_ops *ops = dev->netdev_ops;
5401 if (!ops->ndo_change_carrier)
5403 if (!netif_device_present(dev))
5405 return ops->ndo_change_carrier(dev, new_carrier);
5407 EXPORT_SYMBOL(dev_change_carrier);
5410 * dev_get_phys_port_id - Get device physical port ID
5414 * Get device physical port ID
5416 int dev_get_phys_port_id(struct net_device *dev,
5417 struct netdev_phys_port_id *ppid)
5419 const struct net_device_ops *ops = dev->netdev_ops;
5421 if (!ops->ndo_get_phys_port_id)
5423 return ops->ndo_get_phys_port_id(dev, ppid);
5425 EXPORT_SYMBOL(dev_get_phys_port_id);
5428 * dev_new_index - allocate an ifindex
5429 * @net: the applicable net namespace
5431 * Returns a suitable unique value for a new device interface
5432 * number. The caller must hold the rtnl semaphore or the
5433 * dev_base_lock to be sure it remains unique.
5435 static int dev_new_index(struct net *net)
5437 int ifindex = net->ifindex;
5441 if (!__dev_get_by_index(net, ifindex))
5442 return net->ifindex = ifindex;
5446 /* Delayed registration/unregisteration */
5447 static LIST_HEAD(net_todo_list);
5448 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5450 static void net_set_todo(struct net_device *dev)
5452 list_add_tail(&dev->todo_list, &net_todo_list);
5453 dev_net(dev)->dev_unreg_count++;
5456 static void rollback_registered_many(struct list_head *head)
5458 struct net_device *dev, *tmp;
5459 LIST_HEAD(close_head);
5461 BUG_ON(dev_boot_phase);
5464 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5465 /* Some devices call without registering
5466 * for initialization unwind. Remove those
5467 * devices and proceed with the remaining.
5469 if (dev->reg_state == NETREG_UNINITIALIZED) {
5470 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5474 list_del(&dev->unreg_list);
5477 dev->dismantle = true;
5478 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5481 /* If device is running, close it first. */
5482 list_for_each_entry(dev, head, unreg_list)
5483 list_add_tail(&dev->close_list, &close_head);
5484 dev_close_many(&close_head);
5486 list_for_each_entry(dev, head, unreg_list) {
5487 /* And unlink it from device chain. */
5488 unlist_netdevice(dev);
5490 dev->reg_state = NETREG_UNREGISTERING;
5495 list_for_each_entry(dev, head, unreg_list) {
5496 /* Shutdown queueing discipline. */
5500 /* Notify protocols, that we are about to destroy
5501 this device. They should clean all the things.
5503 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5505 if (!dev->rtnl_link_ops ||
5506 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5507 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5510 * Flush the unicast and multicast chains
5515 if (dev->netdev_ops->ndo_uninit)
5516 dev->netdev_ops->ndo_uninit(dev);
5518 /* Notifier chain MUST detach us all upper devices. */
5519 WARN_ON(netdev_has_any_upper_dev(dev));
5521 /* Remove entries from kobject tree */
5522 netdev_unregister_kobject(dev);
5524 /* Remove XPS queueing entries */
5525 netif_reset_xps_queues_gt(dev, 0);
5531 list_for_each_entry(dev, head, unreg_list)
5535 static void rollback_registered(struct net_device *dev)
5539 list_add(&dev->unreg_list, &single);
5540 rollback_registered_many(&single);
5544 static netdev_features_t netdev_fix_features(struct net_device *dev,
5545 netdev_features_t features)
5547 /* Fix illegal checksum combinations */
5548 if ((features & NETIF_F_HW_CSUM) &&
5549 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5550 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5551 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5554 /* TSO requires that SG is present as well. */
5555 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5556 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5557 features &= ~NETIF_F_ALL_TSO;
5560 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5561 !(features & NETIF_F_IP_CSUM)) {
5562 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5563 features &= ~NETIF_F_TSO;
5564 features &= ~NETIF_F_TSO_ECN;
5567 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5568 !(features & NETIF_F_IPV6_CSUM)) {
5569 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5570 features &= ~NETIF_F_TSO6;
5573 /* TSO ECN requires that TSO is present as well. */
5574 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5575 features &= ~NETIF_F_TSO_ECN;
5577 /* Software GSO depends on SG. */
5578 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5579 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5580 features &= ~NETIF_F_GSO;
5583 /* UFO needs SG and checksumming */
5584 if (features & NETIF_F_UFO) {
5585 /* maybe split UFO into V4 and V6? */
5586 if (!((features & NETIF_F_GEN_CSUM) ||
5587 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5588 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5590 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5591 features &= ~NETIF_F_UFO;
5594 if (!(features & NETIF_F_SG)) {
5596 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5597 features &= ~NETIF_F_UFO;
5604 int __netdev_update_features(struct net_device *dev)
5606 netdev_features_t features;
5611 features = netdev_get_wanted_features(dev);
5613 if (dev->netdev_ops->ndo_fix_features)
5614 features = dev->netdev_ops->ndo_fix_features(dev, features);
5616 /* driver might be less strict about feature dependencies */
5617 features = netdev_fix_features(dev, features);
5619 if (dev->features == features)
5622 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5623 &dev->features, &features);
5625 if (dev->netdev_ops->ndo_set_features)
5626 err = dev->netdev_ops->ndo_set_features(dev, features);
5628 if (unlikely(err < 0)) {
5630 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5631 err, &features, &dev->features);
5636 dev->features = features;
5642 * netdev_update_features - recalculate device features
5643 * @dev: the device to check
5645 * Recalculate dev->features set and send notifications if it
5646 * has changed. Should be called after driver or hardware dependent
5647 * conditions might have changed that influence the features.
5649 void netdev_update_features(struct net_device *dev)
5651 if (__netdev_update_features(dev))
5652 netdev_features_change(dev);
5654 EXPORT_SYMBOL(netdev_update_features);
5657 * netdev_change_features - recalculate device features
5658 * @dev: the device to check
5660 * Recalculate dev->features set and send notifications even
5661 * if they have not changed. Should be called instead of
5662 * netdev_update_features() if also dev->vlan_features might
5663 * have changed to allow the changes to be propagated to stacked
5666 void netdev_change_features(struct net_device *dev)
5668 __netdev_update_features(dev);
5669 netdev_features_change(dev);
5671 EXPORT_SYMBOL(netdev_change_features);
5674 * netif_stacked_transfer_operstate - transfer operstate
5675 * @rootdev: the root or lower level device to transfer state from
5676 * @dev: the device to transfer operstate to
5678 * Transfer operational state from root to device. This is normally
5679 * called when a stacking relationship exists between the root
5680 * device and the device(a leaf device).
5682 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5683 struct net_device *dev)
5685 if (rootdev->operstate == IF_OPER_DORMANT)
5686 netif_dormant_on(dev);
5688 netif_dormant_off(dev);
5690 if (netif_carrier_ok(rootdev)) {
5691 if (!netif_carrier_ok(dev))
5692 netif_carrier_on(dev);
5694 if (netif_carrier_ok(dev))
5695 netif_carrier_off(dev);
5698 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5701 static int netif_alloc_rx_queues(struct net_device *dev)
5703 unsigned int i, count = dev->num_rx_queues;
5704 struct netdev_rx_queue *rx;
5708 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5714 for (i = 0; i < count; i++)
5720 static void netdev_init_one_queue(struct net_device *dev,
5721 struct netdev_queue *queue, void *_unused)
5723 /* Initialize queue lock */
5724 spin_lock_init(&queue->_xmit_lock);
5725 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5726 queue->xmit_lock_owner = -1;
5727 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5730 dql_init(&queue->dql, HZ);
5734 static void netif_free_tx_queues(struct net_device *dev)
5736 if (is_vmalloc_addr(dev->_tx))
5742 static int netif_alloc_netdev_queues(struct net_device *dev)
5744 unsigned int count = dev->num_tx_queues;
5745 struct netdev_queue *tx;
5746 size_t sz = count * sizeof(*tx);
5748 BUG_ON(count < 1 || count > 0xffff);
5750 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5758 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5759 spin_lock_init(&dev->tx_global_lock);
5765 * register_netdevice - register a network device
5766 * @dev: device to register
5768 * Take a completed network device structure and add it to the kernel
5769 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5770 * chain. 0 is returned on success. A negative errno code is returned
5771 * on a failure to set up the device, or if the name is a duplicate.
5773 * Callers must hold the rtnl semaphore. You may want
5774 * register_netdev() instead of this.
5777 * The locking appears insufficient to guarantee two parallel registers
5778 * will not get the same name.
5781 int register_netdevice(struct net_device *dev)
5784 struct net *net = dev_net(dev);
5786 BUG_ON(dev_boot_phase);
5791 /* When net_device's are persistent, this will be fatal. */
5792 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5795 spin_lock_init(&dev->addr_list_lock);
5796 netdev_set_addr_lockdep_class(dev);
5800 ret = dev_get_valid_name(net, dev, dev->name);
5804 /* Init, if this function is available */
5805 if (dev->netdev_ops->ndo_init) {
5806 ret = dev->netdev_ops->ndo_init(dev);
5814 if (((dev->hw_features | dev->features) &
5815 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5816 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5817 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5818 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5825 dev->ifindex = dev_new_index(net);
5826 else if (__dev_get_by_index(net, dev->ifindex))
5829 if (dev->iflink == -1)
5830 dev->iflink = dev->ifindex;
5832 /* Transfer changeable features to wanted_features and enable
5833 * software offloads (GSO and GRO).
5835 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5836 dev->features |= NETIF_F_SOFT_FEATURES;
5837 dev->wanted_features = dev->features & dev->hw_features;
5839 /* Turn on no cache copy if HW is doing checksum */
5840 if (!(dev->flags & IFF_LOOPBACK)) {
5841 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5842 if (dev->features & NETIF_F_ALL_CSUM) {
5843 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5844 dev->features |= NETIF_F_NOCACHE_COPY;
5848 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5850 dev->vlan_features |= NETIF_F_HIGHDMA;
5852 /* Make NETIF_F_SG inheritable to tunnel devices.
5854 dev->hw_enc_features |= NETIF_F_SG;
5856 /* Make NETIF_F_SG inheritable to MPLS.
5858 dev->mpls_features |= NETIF_F_SG;
5860 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5861 ret = notifier_to_errno(ret);
5865 ret = netdev_register_kobject(dev);
5868 dev->reg_state = NETREG_REGISTERED;
5870 __netdev_update_features(dev);
5873 * Default initial state at registry is that the
5874 * device is present.
5877 set_bit(__LINK_STATE_PRESENT, &dev->state);
5879 linkwatch_init_dev(dev);
5881 dev_init_scheduler(dev);
5883 list_netdevice(dev);
5884 add_device_randomness(dev->dev_addr, dev->addr_len);
5886 /* If the device has permanent device address, driver should
5887 * set dev_addr and also addr_assign_type should be set to
5888 * NET_ADDR_PERM (default value).
5890 if (dev->addr_assign_type == NET_ADDR_PERM)
5891 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5893 /* Notify protocols, that a new device appeared. */
5894 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5895 ret = notifier_to_errno(ret);
5897 rollback_registered(dev);
5898 dev->reg_state = NETREG_UNREGISTERED;
5901 * Prevent userspace races by waiting until the network
5902 * device is fully setup before sending notifications.
5904 if (!dev->rtnl_link_ops ||
5905 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5906 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5912 if (dev->netdev_ops->ndo_uninit)
5913 dev->netdev_ops->ndo_uninit(dev);
5916 EXPORT_SYMBOL(register_netdevice);
5919 * init_dummy_netdev - init a dummy network device for NAPI
5920 * @dev: device to init
5922 * This takes a network device structure and initialize the minimum
5923 * amount of fields so it can be used to schedule NAPI polls without
5924 * registering a full blown interface. This is to be used by drivers
5925 * that need to tie several hardware interfaces to a single NAPI
5926 * poll scheduler due to HW limitations.
5928 int init_dummy_netdev(struct net_device *dev)
5930 /* Clear everything. Note we don't initialize spinlocks
5931 * are they aren't supposed to be taken by any of the
5932 * NAPI code and this dummy netdev is supposed to be
5933 * only ever used for NAPI polls
5935 memset(dev, 0, sizeof(struct net_device));
5937 /* make sure we BUG if trying to hit standard
5938 * register/unregister code path
5940 dev->reg_state = NETREG_DUMMY;
5942 /* NAPI wants this */
5943 INIT_LIST_HEAD(&dev->napi_list);
5945 /* a dummy interface is started by default */
5946 set_bit(__LINK_STATE_PRESENT, &dev->state);
5947 set_bit(__LINK_STATE_START, &dev->state);
5949 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5950 * because users of this 'device' dont need to change
5956 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5960 * register_netdev - register a network device
5961 * @dev: device to register
5963 * Take a completed network device structure and add it to the kernel
5964 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5965 * chain. 0 is returned on success. A negative errno code is returned
5966 * on a failure to set up the device, or if the name is a duplicate.
5968 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5969 * and expands the device name if you passed a format string to
5972 int register_netdev(struct net_device *dev)
5977 err = register_netdevice(dev);
5981 EXPORT_SYMBOL(register_netdev);
5983 int netdev_refcnt_read(const struct net_device *dev)
5987 for_each_possible_cpu(i)
5988 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5991 EXPORT_SYMBOL(netdev_refcnt_read);
5994 * netdev_wait_allrefs - wait until all references are gone.
5995 * @dev: target net_device
5997 * This is called when unregistering network devices.
5999 * Any protocol or device that holds a reference should register
6000 * for netdevice notification, and cleanup and put back the
6001 * reference if they receive an UNREGISTER event.
6002 * We can get stuck here if buggy protocols don't correctly
6005 static void netdev_wait_allrefs(struct net_device *dev)
6007 unsigned long rebroadcast_time, warning_time;
6010 linkwatch_forget_dev(dev);
6012 rebroadcast_time = warning_time = jiffies;
6013 refcnt = netdev_refcnt_read(dev);
6015 while (refcnt != 0) {
6016 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6019 /* Rebroadcast unregister notification */
6020 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6026 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6027 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6029 /* We must not have linkwatch events
6030 * pending on unregister. If this
6031 * happens, we simply run the queue
6032 * unscheduled, resulting in a noop
6035 linkwatch_run_queue();
6040 rebroadcast_time = jiffies;
6045 refcnt = netdev_refcnt_read(dev);
6047 if (time_after(jiffies, warning_time + 10 * HZ)) {
6048 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6050 warning_time = jiffies;
6059 * register_netdevice(x1);
6060 * register_netdevice(x2);
6062 * unregister_netdevice(y1);
6063 * unregister_netdevice(y2);
6069 * We are invoked by rtnl_unlock().
6070 * This allows us to deal with problems:
6071 * 1) We can delete sysfs objects which invoke hotplug
6072 * without deadlocking with linkwatch via keventd.
6073 * 2) Since we run with the RTNL semaphore not held, we can sleep
6074 * safely in order to wait for the netdev refcnt to drop to zero.
6076 * We must not return until all unregister events added during
6077 * the interval the lock was held have been completed.
6079 void netdev_run_todo(void)
6081 struct list_head list;
6083 /* Snapshot list, allow later requests */
6084 list_replace_init(&net_todo_list, &list);
6089 /* Wait for rcu callbacks to finish before next phase */
6090 if (!list_empty(&list))
6093 while (!list_empty(&list)) {
6094 struct net_device *dev
6095 = list_first_entry(&list, struct net_device, todo_list);
6096 list_del(&dev->todo_list);
6099 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6102 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6103 pr_err("network todo '%s' but state %d\n",
6104 dev->name, dev->reg_state);
6109 dev->reg_state = NETREG_UNREGISTERED;
6111 on_each_cpu(flush_backlog, dev, 1);
6113 netdev_wait_allrefs(dev);
6116 BUG_ON(netdev_refcnt_read(dev));
6117 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6118 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6119 WARN_ON(dev->dn_ptr);
6121 if (dev->destructor)
6122 dev->destructor(dev);
6124 /* Report a network device has been unregistered */
6126 dev_net(dev)->dev_unreg_count--;
6128 wake_up(&netdev_unregistering_wq);
6130 /* Free network device */
6131 kobject_put(&dev->dev.kobj);
6135 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6136 * fields in the same order, with only the type differing.
6138 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6139 const struct net_device_stats *netdev_stats)
6141 #if BITS_PER_LONG == 64
6142 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6143 memcpy(stats64, netdev_stats, sizeof(*stats64));
6145 size_t i, n = sizeof(*stats64) / sizeof(u64);
6146 const unsigned long *src = (const unsigned long *)netdev_stats;
6147 u64 *dst = (u64 *)stats64;
6149 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6150 sizeof(*stats64) / sizeof(u64));
6151 for (i = 0; i < n; i++)
6155 EXPORT_SYMBOL(netdev_stats_to_stats64);
6158 * dev_get_stats - get network device statistics
6159 * @dev: device to get statistics from
6160 * @storage: place to store stats
6162 * Get network statistics from device. Return @storage.
6163 * The device driver may provide its own method by setting
6164 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6165 * otherwise the internal statistics structure is used.
6167 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6168 struct rtnl_link_stats64 *storage)
6170 const struct net_device_ops *ops = dev->netdev_ops;
6172 if (ops->ndo_get_stats64) {
6173 memset(storage, 0, sizeof(*storage));
6174 ops->ndo_get_stats64(dev, storage);
6175 } else if (ops->ndo_get_stats) {
6176 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6178 netdev_stats_to_stats64(storage, &dev->stats);
6180 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6183 EXPORT_SYMBOL(dev_get_stats);
6185 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6187 struct netdev_queue *queue = dev_ingress_queue(dev);
6189 #ifdef CONFIG_NET_CLS_ACT
6192 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6195 netdev_init_one_queue(dev, queue, NULL);
6196 queue->qdisc = &noop_qdisc;
6197 queue->qdisc_sleeping = &noop_qdisc;
6198 rcu_assign_pointer(dev->ingress_queue, queue);
6203 static const struct ethtool_ops default_ethtool_ops;
6205 void netdev_set_default_ethtool_ops(struct net_device *dev,
6206 const struct ethtool_ops *ops)
6208 if (dev->ethtool_ops == &default_ethtool_ops)
6209 dev->ethtool_ops = ops;
6211 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6213 void netdev_freemem(struct net_device *dev)
6215 char *addr = (char *)dev - dev->padded;
6217 if (is_vmalloc_addr(addr))
6224 * alloc_netdev_mqs - allocate network device
6225 * @sizeof_priv: size of private data to allocate space for
6226 * @name: device name format string
6227 * @setup: callback to initialize device
6228 * @txqs: the number of TX subqueues to allocate
6229 * @rxqs: the number of RX subqueues to allocate
6231 * Allocates a struct net_device with private data area for driver use
6232 * and performs basic initialization. Also allocates subqueue structs
6233 * for each queue on the device.
6235 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6236 void (*setup)(struct net_device *),
6237 unsigned int txqs, unsigned int rxqs)
6239 struct net_device *dev;
6241 struct net_device *p;
6243 BUG_ON(strlen(name) >= sizeof(dev->name));
6246 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6252 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6257 alloc_size = sizeof(struct net_device);
6259 /* ensure 32-byte alignment of private area */
6260 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6261 alloc_size += sizeof_priv;
6263 /* ensure 32-byte alignment of whole construct */
6264 alloc_size += NETDEV_ALIGN - 1;
6266 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6268 p = vzalloc(alloc_size);
6272 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6273 dev->padded = (char *)dev - (char *)p;
6275 dev->pcpu_refcnt = alloc_percpu(int);
6276 if (!dev->pcpu_refcnt)
6279 if (dev_addr_init(dev))
6285 dev_net_set(dev, &init_net);
6287 dev->gso_max_size = GSO_MAX_SIZE;
6288 dev->gso_max_segs = GSO_MAX_SEGS;
6290 INIT_LIST_HEAD(&dev->napi_list);
6291 INIT_LIST_HEAD(&dev->unreg_list);
6292 INIT_LIST_HEAD(&dev->close_list);
6293 INIT_LIST_HEAD(&dev->link_watch_list);
6294 INIT_LIST_HEAD(&dev->adj_list.upper);
6295 INIT_LIST_HEAD(&dev->adj_list.lower);
6296 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6297 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6298 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6301 dev->num_tx_queues = txqs;
6302 dev->real_num_tx_queues = txqs;
6303 if (netif_alloc_netdev_queues(dev))
6307 dev->num_rx_queues = rxqs;
6308 dev->real_num_rx_queues = rxqs;
6309 if (netif_alloc_rx_queues(dev))
6313 strcpy(dev->name, name);
6314 dev->group = INIT_NETDEV_GROUP;
6315 if (!dev->ethtool_ops)
6316 dev->ethtool_ops = &default_ethtool_ops;
6324 free_percpu(dev->pcpu_refcnt);
6325 netif_free_tx_queues(dev);
6331 netdev_freemem(dev);
6334 EXPORT_SYMBOL(alloc_netdev_mqs);
6337 * free_netdev - free network device
6340 * This function does the last stage of destroying an allocated device
6341 * interface. The reference to the device object is released.
6342 * If this is the last reference then it will be freed.
6344 void free_netdev(struct net_device *dev)
6346 struct napi_struct *p, *n;
6348 release_net(dev_net(dev));
6350 netif_free_tx_queues(dev);
6355 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6357 /* Flush device addresses */
6358 dev_addr_flush(dev);
6360 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6363 free_percpu(dev->pcpu_refcnt);
6364 dev->pcpu_refcnt = NULL;
6366 /* Compatibility with error handling in drivers */
6367 if (dev->reg_state == NETREG_UNINITIALIZED) {
6368 netdev_freemem(dev);
6372 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6373 dev->reg_state = NETREG_RELEASED;
6375 /* will free via device release */
6376 put_device(&dev->dev);
6378 EXPORT_SYMBOL(free_netdev);
6381 * synchronize_net - Synchronize with packet receive processing
6383 * Wait for packets currently being received to be done.
6384 * Does not block later packets from starting.
6386 void synchronize_net(void)
6389 if (rtnl_is_locked())
6390 synchronize_rcu_expedited();
6394 EXPORT_SYMBOL(synchronize_net);
6397 * unregister_netdevice_queue - remove device from the kernel
6401 * This function shuts down a device interface and removes it
6402 * from the kernel tables.
6403 * If head not NULL, device is queued to be unregistered later.
6405 * Callers must hold the rtnl semaphore. You may want
6406 * unregister_netdev() instead of this.
6409 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6414 list_move_tail(&dev->unreg_list, head);
6416 rollback_registered(dev);
6417 /* Finish processing unregister after unlock */
6421 EXPORT_SYMBOL(unregister_netdevice_queue);
6424 * unregister_netdevice_many - unregister many devices
6425 * @head: list of devices
6427 void unregister_netdevice_many(struct list_head *head)
6429 struct net_device *dev;
6431 if (!list_empty(head)) {
6432 rollback_registered_many(head);
6433 list_for_each_entry(dev, head, unreg_list)
6437 EXPORT_SYMBOL(unregister_netdevice_many);
6440 * unregister_netdev - remove device from the kernel
6443 * This function shuts down a device interface and removes it
6444 * from the kernel tables.
6446 * This is just a wrapper for unregister_netdevice that takes
6447 * the rtnl semaphore. In general you want to use this and not
6448 * unregister_netdevice.
6450 void unregister_netdev(struct net_device *dev)
6453 unregister_netdevice(dev);
6456 EXPORT_SYMBOL(unregister_netdev);
6459 * dev_change_net_namespace - move device to different nethost namespace
6461 * @net: network namespace
6462 * @pat: If not NULL name pattern to try if the current device name
6463 * is already taken in the destination network namespace.
6465 * This function shuts down a device interface and moves it
6466 * to a new network namespace. On success 0 is returned, on
6467 * a failure a netagive errno code is returned.
6469 * Callers must hold the rtnl semaphore.
6472 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6478 /* Don't allow namespace local devices to be moved. */
6480 if (dev->features & NETIF_F_NETNS_LOCAL)
6483 /* Ensure the device has been registrered */
6484 if (dev->reg_state != NETREG_REGISTERED)
6487 /* Get out if there is nothing todo */
6489 if (net_eq(dev_net(dev), net))
6492 /* Pick the destination device name, and ensure
6493 * we can use it in the destination network namespace.
6496 if (__dev_get_by_name(net, dev->name)) {
6497 /* We get here if we can't use the current device name */
6500 if (dev_get_valid_name(net, dev, pat) < 0)
6505 * And now a mini version of register_netdevice unregister_netdevice.
6508 /* If device is running close it first. */
6511 /* And unlink it from device chain */
6513 unlist_netdevice(dev);
6517 /* Shutdown queueing discipline. */
6520 /* Notify protocols, that we are about to destroy
6521 this device. They should clean all the things.
6523 Note that dev->reg_state stays at NETREG_REGISTERED.
6524 This is wanted because this way 8021q and macvlan know
6525 the device is just moving and can keep their slaves up.
6527 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6529 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6530 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6533 * Flush the unicast and multicast chains
6538 /* Send a netdev-removed uevent to the old namespace */
6539 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6541 /* Actually switch the network namespace */
6542 dev_net_set(dev, net);
6544 /* If there is an ifindex conflict assign a new one */
6545 if (__dev_get_by_index(net, dev->ifindex)) {
6546 int iflink = (dev->iflink == dev->ifindex);
6547 dev->ifindex = dev_new_index(net);
6549 dev->iflink = dev->ifindex;
6552 /* Send a netdev-add uevent to the new namespace */
6553 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6555 /* Fixup kobjects */
6556 err = device_rename(&dev->dev, dev->name);
6559 /* Add the device back in the hashes */
6560 list_netdevice(dev);
6562 /* Notify protocols, that a new device appeared. */
6563 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6566 * Prevent userspace races by waiting until the network
6567 * device is fully setup before sending notifications.
6569 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6576 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6578 static int dev_cpu_callback(struct notifier_block *nfb,
6579 unsigned long action,
6582 struct sk_buff **list_skb;
6583 struct sk_buff *skb;
6584 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6585 struct softnet_data *sd, *oldsd;
6587 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6590 local_irq_disable();
6591 cpu = smp_processor_id();
6592 sd = &per_cpu(softnet_data, cpu);
6593 oldsd = &per_cpu(softnet_data, oldcpu);
6595 /* Find end of our completion_queue. */
6596 list_skb = &sd->completion_queue;
6598 list_skb = &(*list_skb)->next;
6599 /* Append completion queue from offline CPU. */
6600 *list_skb = oldsd->completion_queue;
6601 oldsd->completion_queue = NULL;
6603 /* Append output queue from offline CPU. */
6604 if (oldsd->output_queue) {
6605 *sd->output_queue_tailp = oldsd->output_queue;
6606 sd->output_queue_tailp = oldsd->output_queue_tailp;
6607 oldsd->output_queue = NULL;
6608 oldsd->output_queue_tailp = &oldsd->output_queue;
6610 /* Append NAPI poll list from offline CPU. */
6611 if (!list_empty(&oldsd->poll_list)) {
6612 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6613 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6616 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6619 /* Process offline CPU's input_pkt_queue */
6620 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6622 input_queue_head_incr(oldsd);
6624 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6626 input_queue_head_incr(oldsd);
6634 * netdev_increment_features - increment feature set by one
6635 * @all: current feature set
6636 * @one: new feature set
6637 * @mask: mask feature set
6639 * Computes a new feature set after adding a device with feature set
6640 * @one to the master device with current feature set @all. Will not
6641 * enable anything that is off in @mask. Returns the new feature set.
6643 netdev_features_t netdev_increment_features(netdev_features_t all,
6644 netdev_features_t one, netdev_features_t mask)
6646 if (mask & NETIF_F_GEN_CSUM)
6647 mask |= NETIF_F_ALL_CSUM;
6648 mask |= NETIF_F_VLAN_CHALLENGED;
6650 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6651 all &= one | ~NETIF_F_ALL_FOR_ALL;
6653 /* If one device supports hw checksumming, set for all. */
6654 if (all & NETIF_F_GEN_CSUM)
6655 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6659 EXPORT_SYMBOL(netdev_increment_features);
6661 static struct hlist_head * __net_init netdev_create_hash(void)
6664 struct hlist_head *hash;
6666 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6668 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6669 INIT_HLIST_HEAD(&hash[i]);
6674 /* Initialize per network namespace state */
6675 static int __net_init netdev_init(struct net *net)
6677 if (net != &init_net)
6678 INIT_LIST_HEAD(&net->dev_base_head);
6680 net->dev_name_head = netdev_create_hash();
6681 if (net->dev_name_head == NULL)
6684 net->dev_index_head = netdev_create_hash();
6685 if (net->dev_index_head == NULL)
6691 kfree(net->dev_name_head);
6697 * netdev_drivername - network driver for the device
6698 * @dev: network device
6700 * Determine network driver for device.
6702 const char *netdev_drivername(const struct net_device *dev)
6704 const struct device_driver *driver;
6705 const struct device *parent;
6706 const char *empty = "";
6708 parent = dev->dev.parent;
6712 driver = parent->driver;
6713 if (driver && driver->name)
6714 return driver->name;
6718 static int __netdev_printk(const char *level, const struct net_device *dev,
6719 struct va_format *vaf)
6723 if (dev && dev->dev.parent) {
6724 r = dev_printk_emit(level[1] - '0',
6727 dev_driver_string(dev->dev.parent),
6728 dev_name(dev->dev.parent),
6729 netdev_name(dev), vaf);
6731 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6733 r = printk("%s(NULL net_device): %pV", level, vaf);
6739 int netdev_printk(const char *level, const struct net_device *dev,
6740 const char *format, ...)
6742 struct va_format vaf;
6746 va_start(args, format);
6751 r = __netdev_printk(level, dev, &vaf);
6757 EXPORT_SYMBOL(netdev_printk);
6759 #define define_netdev_printk_level(func, level) \
6760 int func(const struct net_device *dev, const char *fmt, ...) \
6763 struct va_format vaf; \
6766 va_start(args, fmt); \
6771 r = __netdev_printk(level, dev, &vaf); \
6777 EXPORT_SYMBOL(func);
6779 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6780 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6781 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6782 define_netdev_printk_level(netdev_err, KERN_ERR);
6783 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6784 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6785 define_netdev_printk_level(netdev_info, KERN_INFO);
6787 static void __net_exit netdev_exit(struct net *net)
6789 kfree(net->dev_name_head);
6790 kfree(net->dev_index_head);
6793 static struct pernet_operations __net_initdata netdev_net_ops = {
6794 .init = netdev_init,
6795 .exit = netdev_exit,
6798 static void __net_exit default_device_exit(struct net *net)
6800 struct net_device *dev, *aux;
6802 * Push all migratable network devices back to the
6803 * initial network namespace
6806 for_each_netdev_safe(net, dev, aux) {
6808 char fb_name[IFNAMSIZ];
6810 /* Ignore unmoveable devices (i.e. loopback) */
6811 if (dev->features & NETIF_F_NETNS_LOCAL)
6814 /* Leave virtual devices for the generic cleanup */
6815 if (dev->rtnl_link_ops)
6818 /* Push remaining network devices to init_net */
6819 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6820 err = dev_change_net_namespace(dev, &init_net, fb_name);
6822 pr_emerg("%s: failed to move %s to init_net: %d\n",
6823 __func__, dev->name, err);
6830 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6832 /* Return with the rtnl_lock held when there are no network
6833 * devices unregistering in any network namespace in net_list.
6840 prepare_to_wait(&netdev_unregistering_wq, &wait,
6841 TASK_UNINTERRUPTIBLE);
6842 unregistering = false;
6844 list_for_each_entry(net, net_list, exit_list) {
6845 if (net->dev_unreg_count > 0) {
6846 unregistering = true;
6855 finish_wait(&netdev_unregistering_wq, &wait);
6858 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6860 /* At exit all network devices most be removed from a network
6861 * namespace. Do this in the reverse order of registration.
6862 * Do this across as many network namespaces as possible to
6863 * improve batching efficiency.
6865 struct net_device *dev;
6867 LIST_HEAD(dev_kill_list);
6869 /* To prevent network device cleanup code from dereferencing
6870 * loopback devices or network devices that have been freed
6871 * wait here for all pending unregistrations to complete,
6872 * before unregistring the loopback device and allowing the
6873 * network namespace be freed.
6875 * The netdev todo list containing all network devices
6876 * unregistrations that happen in default_device_exit_batch
6877 * will run in the rtnl_unlock() at the end of
6878 * default_device_exit_batch.
6880 rtnl_lock_unregistering(net_list);
6881 list_for_each_entry(net, net_list, exit_list) {
6882 for_each_netdev_reverse(net, dev) {
6883 if (dev->rtnl_link_ops)
6884 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6886 unregister_netdevice_queue(dev, &dev_kill_list);
6889 unregister_netdevice_many(&dev_kill_list);
6890 list_del(&dev_kill_list);
6894 static struct pernet_operations __net_initdata default_device_ops = {
6895 .exit = default_device_exit,
6896 .exit_batch = default_device_exit_batch,
6900 * Initialize the DEV module. At boot time this walks the device list and
6901 * unhooks any devices that fail to initialise (normally hardware not
6902 * present) and leaves us with a valid list of present and active devices.
6907 * This is called single threaded during boot, so no need
6908 * to take the rtnl semaphore.
6910 static int __init net_dev_init(void)
6912 int i, rc = -ENOMEM;
6914 BUG_ON(!dev_boot_phase);
6916 if (dev_proc_init())
6919 if (netdev_kobject_init())
6922 INIT_LIST_HEAD(&ptype_all);
6923 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6924 INIT_LIST_HEAD(&ptype_base[i]);
6926 INIT_LIST_HEAD(&offload_base);
6928 if (register_pernet_subsys(&netdev_net_ops))
6932 * Initialise the packet receive queues.
6935 for_each_possible_cpu(i) {
6936 struct softnet_data *sd = &per_cpu(softnet_data, i);
6938 memset(sd, 0, sizeof(*sd));
6939 skb_queue_head_init(&sd->input_pkt_queue);
6940 skb_queue_head_init(&sd->process_queue);
6941 sd->completion_queue = NULL;
6942 INIT_LIST_HEAD(&sd->poll_list);
6943 sd->output_queue = NULL;
6944 sd->output_queue_tailp = &sd->output_queue;
6946 sd->csd.func = rps_trigger_softirq;
6952 sd->backlog.poll = process_backlog;
6953 sd->backlog.weight = weight_p;
6954 sd->backlog.gro_list = NULL;
6955 sd->backlog.gro_count = 0;
6957 #ifdef CONFIG_NET_FLOW_LIMIT
6958 sd->flow_limit = NULL;
6964 /* The loopback device is special if any other network devices
6965 * is present in a network namespace the loopback device must
6966 * be present. Since we now dynamically allocate and free the
6967 * loopback device ensure this invariant is maintained by
6968 * keeping the loopback device as the first device on the
6969 * list of network devices. Ensuring the loopback devices
6970 * is the first device that appears and the last network device
6973 if (register_pernet_device(&loopback_net_ops))
6976 if (register_pernet_device(&default_device_ops))
6979 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6980 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6982 hotcpu_notifier(dev_cpu_callback, 0);
6989 subsys_initcall(net_dev_init);