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 static 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);
503 * dev_remove_offload - remove packet offload handler
504 * @po: packet offload declaration
506 * Remove a packet offload handler that was previously added to the kernel
507 * offload handlers by dev_add_offload(). The passed &offload_type is
508 * removed from the kernel lists and can be freed or reused once this
511 * This call sleeps to guarantee that no CPU is looking at the packet
514 void dev_remove_offload(struct packet_offload *po)
516 __dev_remove_offload(po);
520 EXPORT_SYMBOL(dev_remove_offload);
522 /******************************************************************************
524 Device Boot-time Settings Routines
526 *******************************************************************************/
528 /* Boot time configuration table */
529 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
532 * netdev_boot_setup_add - add new setup entry
533 * @name: name of the device
534 * @map: configured settings for the device
536 * Adds new setup entry to the dev_boot_setup list. The function
537 * returns 0 on error and 1 on success. This is a generic routine to
540 static int netdev_boot_setup_add(char *name, struct ifmap *map)
542 struct netdev_boot_setup *s;
546 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
547 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
548 memset(s[i].name, 0, sizeof(s[i].name));
549 strlcpy(s[i].name, name, IFNAMSIZ);
550 memcpy(&s[i].map, map, sizeof(s[i].map));
555 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
559 * netdev_boot_setup_check - check boot time settings
560 * @dev: the netdevice
562 * Check boot time settings for the device.
563 * The found settings are set for the device to be used
564 * later in the device probing.
565 * Returns 0 if no settings found, 1 if they are.
567 int netdev_boot_setup_check(struct net_device *dev)
569 struct netdev_boot_setup *s = dev_boot_setup;
572 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
573 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
574 !strcmp(dev->name, s[i].name)) {
575 dev->irq = s[i].map.irq;
576 dev->base_addr = s[i].map.base_addr;
577 dev->mem_start = s[i].map.mem_start;
578 dev->mem_end = s[i].map.mem_end;
584 EXPORT_SYMBOL(netdev_boot_setup_check);
588 * netdev_boot_base - get address from boot time settings
589 * @prefix: prefix for network device
590 * @unit: id for network device
592 * Check boot time settings for the base address of device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found.
597 unsigned long netdev_boot_base(const char *prefix, int unit)
599 const struct netdev_boot_setup *s = dev_boot_setup;
603 sprintf(name, "%s%d", prefix, unit);
606 * If device already registered then return base of 1
607 * to indicate not to probe for this interface
609 if (__dev_get_by_name(&init_net, name))
612 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
613 if (!strcmp(name, s[i].name))
614 return s[i].map.base_addr;
619 * Saves at boot time configured settings for any netdevice.
621 int __init netdev_boot_setup(char *str)
626 str = get_options(str, ARRAY_SIZE(ints), ints);
631 memset(&map, 0, sizeof(map));
635 map.base_addr = ints[2];
637 map.mem_start = ints[3];
639 map.mem_end = ints[4];
641 /* Add new entry to the list */
642 return netdev_boot_setup_add(str, &map);
645 __setup("netdev=", netdev_boot_setup);
647 /*******************************************************************************
649 Device Interface Subroutines
651 *******************************************************************************/
654 * __dev_get_by_name - find a device by its name
655 * @net: the applicable net namespace
656 * @name: name to find
658 * Find an interface by name. Must be called under RTNL semaphore
659 * or @dev_base_lock. If the name is found a pointer to the device
660 * is returned. If the name is not found then %NULL is returned. The
661 * reference counters are not incremented so the caller must be
662 * careful with locks.
665 struct net_device *__dev_get_by_name(struct net *net, const char *name)
667 struct net_device *dev;
668 struct hlist_head *head = dev_name_hash(net, name);
670 hlist_for_each_entry(dev, head, name_hlist)
671 if (!strncmp(dev->name, name, IFNAMSIZ))
676 EXPORT_SYMBOL(__dev_get_by_name);
679 * dev_get_by_name_rcu - find a device by its name
680 * @net: the applicable net namespace
681 * @name: name to find
683 * Find an interface by name.
684 * If the name is found a pointer to the device is returned.
685 * If the name is not found then %NULL is returned.
686 * The reference counters are not incremented so the caller must be
687 * careful with locks. The caller must hold RCU lock.
690 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
692 struct net_device *dev;
693 struct hlist_head *head = dev_name_hash(net, name);
695 hlist_for_each_entry_rcu(dev, head, name_hlist)
696 if (!strncmp(dev->name, name, IFNAMSIZ))
701 EXPORT_SYMBOL(dev_get_by_name_rcu);
704 * dev_get_by_name - find a device by its name
705 * @net: the applicable net namespace
706 * @name: name to find
708 * Find an interface by name. This can be called from any
709 * context and does its own locking. The returned handle has
710 * the usage count incremented and the caller must use dev_put() to
711 * release it when it is no longer needed. %NULL is returned if no
712 * matching device is found.
715 struct net_device *dev_get_by_name(struct net *net, const char *name)
717 struct net_device *dev;
720 dev = dev_get_by_name_rcu(net, name);
726 EXPORT_SYMBOL(dev_get_by_name);
729 * __dev_get_by_index - find a device by its ifindex
730 * @net: the applicable net namespace
731 * @ifindex: index of device
733 * Search for an interface by index. Returns %NULL if the device
734 * is not found or a pointer to the device. The device has not
735 * had its reference counter increased so the caller must be careful
736 * about locking. The caller must hold either the RTNL semaphore
740 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
742 struct net_device *dev;
743 struct hlist_head *head = dev_index_hash(net, ifindex);
745 hlist_for_each_entry(dev, head, index_hlist)
746 if (dev->ifindex == ifindex)
751 EXPORT_SYMBOL(__dev_get_by_index);
754 * dev_get_by_index_rcu - find a device by its ifindex
755 * @net: the applicable net namespace
756 * @ifindex: index of device
758 * Search for an interface by index. Returns %NULL if the device
759 * is not found or a pointer to the device. The device has not
760 * had its reference counter increased so the caller must be careful
761 * about locking. The caller must hold RCU lock.
764 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
766 struct net_device *dev;
767 struct hlist_head *head = dev_index_hash(net, ifindex);
769 hlist_for_each_entry_rcu(dev, head, index_hlist)
770 if (dev->ifindex == ifindex)
775 EXPORT_SYMBOL(dev_get_by_index_rcu);
779 * dev_get_by_index - find a device by its ifindex
780 * @net: the applicable net namespace
781 * @ifindex: index of device
783 * Search for an interface by index. Returns NULL if the device
784 * is not found or a pointer to the device. The device returned has
785 * had a reference added and the pointer is safe until the user calls
786 * dev_put to indicate they have finished with it.
789 struct net_device *dev_get_by_index(struct net *net, int ifindex)
791 struct net_device *dev;
794 dev = dev_get_by_index_rcu(net, ifindex);
800 EXPORT_SYMBOL(dev_get_by_index);
803 * netdev_get_name - get a netdevice name, knowing its ifindex.
804 * @net: network namespace
805 * @name: a pointer to the buffer where the name will be stored.
806 * @ifindex: the ifindex of the interface to get the name from.
808 * The use of raw_seqcount_begin() and cond_resched() before
809 * retrying is required as we want to give the writers a chance
810 * to complete when CONFIG_PREEMPT is not set.
812 int netdev_get_name(struct net *net, char *name, int ifindex)
814 struct net_device *dev;
818 seq = raw_seqcount_begin(&devnet_rename_seq);
820 dev = dev_get_by_index_rcu(net, ifindex);
826 strcpy(name, dev->name);
828 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
837 * dev_getbyhwaddr_rcu - find a device by its hardware address
838 * @net: the applicable net namespace
839 * @type: media type of device
840 * @ha: hardware address
842 * Search for an interface by MAC address. Returns NULL if the device
843 * is not found or a pointer to the device.
844 * The caller must hold RCU or RTNL.
845 * The returned device has not had its ref count increased
846 * and the caller must therefore be careful about locking
850 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
853 struct net_device *dev;
855 for_each_netdev_rcu(net, dev)
856 if (dev->type == type &&
857 !memcmp(dev->dev_addr, ha, dev->addr_len))
862 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
864 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
866 struct net_device *dev;
869 for_each_netdev(net, dev)
870 if (dev->type == type)
875 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
877 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
879 struct net_device *dev, *ret = NULL;
882 for_each_netdev_rcu(net, dev)
883 if (dev->type == type) {
891 EXPORT_SYMBOL(dev_getfirstbyhwtype);
894 * dev_get_by_flags_rcu - find any device with given flags
895 * @net: the applicable net namespace
896 * @if_flags: IFF_* values
897 * @mask: bitmask of bits in if_flags to check
899 * Search for any interface with the given flags. Returns NULL if a device
900 * is not found or a pointer to the device. Must be called inside
901 * rcu_read_lock(), and result refcount is unchanged.
904 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
907 struct net_device *dev, *ret;
910 for_each_netdev_rcu(net, dev) {
911 if (((dev->flags ^ if_flags) & mask) == 0) {
918 EXPORT_SYMBOL(dev_get_by_flags_rcu);
921 * dev_valid_name - check if name is okay for network device
924 * Network device names need to be valid file names to
925 * to allow sysfs to work. We also disallow any kind of
928 bool dev_valid_name(const char *name)
932 if (strlen(name) >= IFNAMSIZ)
934 if (!strcmp(name, ".") || !strcmp(name, ".."))
938 if (*name == '/' || isspace(*name))
944 EXPORT_SYMBOL(dev_valid_name);
947 * __dev_alloc_name - allocate a name for a device
948 * @net: network namespace to allocate the device name in
949 * @name: name format string
950 * @buf: scratch buffer and result name string
952 * Passed a format string - eg "lt%d" it will try and find a suitable
953 * id. It scans list of devices to build up a free map, then chooses
954 * the first empty slot. The caller must hold the dev_base or rtnl lock
955 * while allocating the name and adding the device in order to avoid
957 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
958 * Returns the number of the unit assigned or a negative errno code.
961 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
965 const int max_netdevices = 8*PAGE_SIZE;
966 unsigned long *inuse;
967 struct net_device *d;
969 p = strnchr(name, IFNAMSIZ-1, '%');
972 * Verify the string as this thing may have come from
973 * the user. There must be either one "%d" and no other "%"
976 if (p[1] != 'd' || strchr(p + 2, '%'))
979 /* Use one page as a bit array of possible slots */
980 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
984 for_each_netdev(net, d) {
985 if (!sscanf(d->name, name, &i))
987 if (i < 0 || i >= max_netdevices)
990 /* avoid cases where sscanf is not exact inverse of printf */
991 snprintf(buf, IFNAMSIZ, name, i);
992 if (!strncmp(buf, d->name, IFNAMSIZ))
996 i = find_first_zero_bit(inuse, max_netdevices);
997 free_page((unsigned long) inuse);
1001 snprintf(buf, IFNAMSIZ, name, i);
1002 if (!__dev_get_by_name(net, buf))
1005 /* It is possible to run out of possible slots
1006 * when the name is long and there isn't enough space left
1007 * for the digits, or if all bits are used.
1013 * dev_alloc_name - allocate a name for a device
1015 * @name: name format string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 int dev_alloc_name(struct net_device *dev, const char *name)
1032 BUG_ON(!dev_net(dev));
1034 ret = __dev_alloc_name(net, name, buf);
1036 strlcpy(dev->name, buf, IFNAMSIZ);
1039 EXPORT_SYMBOL(dev_alloc_name);
1041 static int dev_alloc_name_ns(struct net *net,
1042 struct net_device *dev,
1048 ret = __dev_alloc_name(net, name, buf);
1050 strlcpy(dev->name, buf, IFNAMSIZ);
1054 static int dev_get_valid_name(struct net *net,
1055 struct net_device *dev,
1060 if (!dev_valid_name(name))
1063 if (strchr(name, '%'))
1064 return dev_alloc_name_ns(net, dev, name);
1065 else if (__dev_get_by_name(net, name))
1067 else if (dev->name != name)
1068 strlcpy(dev->name, name, IFNAMSIZ);
1074 * dev_change_name - change name of a device
1076 * @newname: name (or format string) must be at least IFNAMSIZ
1078 * Change name of a device, can pass format strings "eth%d".
1081 int dev_change_name(struct net_device *dev, const char *newname)
1083 char oldname[IFNAMSIZ];
1089 BUG_ON(!dev_net(dev));
1092 if (dev->flags & IFF_UP)
1095 write_seqcount_begin(&devnet_rename_seq);
1097 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1098 write_seqcount_end(&devnet_rename_seq);
1102 memcpy(oldname, dev->name, IFNAMSIZ);
1104 err = dev_get_valid_name(net, dev, newname);
1106 write_seqcount_end(&devnet_rename_seq);
1111 ret = device_rename(&dev->dev, dev->name);
1113 memcpy(dev->name, oldname, IFNAMSIZ);
1114 write_seqcount_end(&devnet_rename_seq);
1118 write_seqcount_end(&devnet_rename_seq);
1120 write_lock_bh(&dev_base_lock);
1121 hlist_del_rcu(&dev->name_hlist);
1122 write_unlock_bh(&dev_base_lock);
1126 write_lock_bh(&dev_base_lock);
1127 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1128 write_unlock_bh(&dev_base_lock);
1130 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1131 ret = notifier_to_errno(ret);
1134 /* err >= 0 after dev_alloc_name() or stores the first errno */
1137 write_seqcount_begin(&devnet_rename_seq);
1138 memcpy(dev->name, oldname, IFNAMSIZ);
1141 pr_err("%s: name change rollback failed: %d\n",
1150 * dev_set_alias - change ifalias of a device
1152 * @alias: name up to IFALIASZ
1153 * @len: limit of bytes to copy from info
1155 * Set ifalias for a device,
1157 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1163 if (len >= IFALIASZ)
1167 kfree(dev->ifalias);
1168 dev->ifalias = NULL;
1172 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1175 dev->ifalias = new_ifalias;
1177 strlcpy(dev->ifalias, alias, len+1);
1183 * netdev_features_change - device changes features
1184 * @dev: device to cause notification
1186 * Called to indicate a device has changed features.
1188 void netdev_features_change(struct net_device *dev)
1190 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1192 EXPORT_SYMBOL(netdev_features_change);
1195 * netdev_state_change - device changes state
1196 * @dev: device to cause notification
1198 * Called to indicate a device has changed state. This function calls
1199 * the notifier chains for netdev_chain and sends a NEWLINK message
1200 * to the routing socket.
1202 void netdev_state_change(struct net_device *dev)
1204 if (dev->flags & IFF_UP) {
1205 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1206 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1209 EXPORT_SYMBOL(netdev_state_change);
1212 * netdev_notify_peers - notify network peers about existence of @dev
1213 * @dev: network device
1215 * Generate traffic such that interested network peers are aware of
1216 * @dev, such as by generating a gratuitous ARP. This may be used when
1217 * a device wants to inform the rest of the network about some sort of
1218 * reconfiguration such as a failover event or virtual machine
1221 void netdev_notify_peers(struct net_device *dev)
1224 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1227 EXPORT_SYMBOL(netdev_notify_peers);
1229 static int __dev_open(struct net_device *dev)
1231 const struct net_device_ops *ops = dev->netdev_ops;
1236 if (!netif_device_present(dev))
1239 /* Block netpoll from trying to do any rx path servicing.
1240 * If we don't do this there is a chance ndo_poll_controller
1241 * or ndo_poll may be running while we open the device
1243 netpoll_rx_disable(dev);
1245 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1246 ret = notifier_to_errno(ret);
1250 set_bit(__LINK_STATE_START, &dev->state);
1252 if (ops->ndo_validate_addr)
1253 ret = ops->ndo_validate_addr(dev);
1255 if (!ret && ops->ndo_open)
1256 ret = ops->ndo_open(dev);
1258 netpoll_rx_enable(dev);
1261 clear_bit(__LINK_STATE_START, &dev->state);
1263 dev->flags |= IFF_UP;
1264 net_dmaengine_get();
1265 dev_set_rx_mode(dev);
1267 add_device_randomness(dev->dev_addr, dev->addr_len);
1274 * dev_open - prepare an interface for use.
1275 * @dev: device to open
1277 * Takes a device from down to up state. The device's private open
1278 * function is invoked and then the multicast lists are loaded. Finally
1279 * the device is moved into the up state and a %NETDEV_UP message is
1280 * sent to the netdev notifier chain.
1282 * Calling this function on an active interface is a nop. On a failure
1283 * a negative errno code is returned.
1285 int dev_open(struct net_device *dev)
1289 if (dev->flags & IFF_UP)
1292 ret = __dev_open(dev);
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1297 call_netdevice_notifiers(NETDEV_UP, dev);
1301 EXPORT_SYMBOL(dev_open);
1303 static int __dev_close_many(struct list_head *head)
1305 struct net_device *dev;
1310 list_for_each_entry(dev, head, close_list) {
1311 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1313 clear_bit(__LINK_STATE_START, &dev->state);
1315 /* Synchronize to scheduled poll. We cannot touch poll list, it
1316 * can be even on different cpu. So just clear netif_running().
1318 * dev->stop() will invoke napi_disable() on all of it's
1319 * napi_struct instances on this device.
1321 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1324 dev_deactivate_many(head);
1326 list_for_each_entry(dev, head, close_list) {
1327 const struct net_device_ops *ops = dev->netdev_ops;
1330 * Call the device specific close. This cannot fail.
1331 * Only if device is UP
1333 * We allow it to be called even after a DETACH hot-plug
1339 dev->flags &= ~IFF_UP;
1340 net_dmaengine_put();
1346 static int __dev_close(struct net_device *dev)
1351 /* Temporarily disable netpoll until the interface is down */
1352 netpoll_rx_disable(dev);
1354 list_add(&dev->close_list, &single);
1355 retval = __dev_close_many(&single);
1358 netpoll_rx_enable(dev);
1362 static int dev_close_many(struct list_head *head)
1364 struct net_device *dev, *tmp;
1366 /* Remove the devices that don't need to be closed */
1367 list_for_each_entry_safe(dev, tmp, head, close_list)
1368 if (!(dev->flags & IFF_UP))
1369 list_del_init(&dev->close_list);
1371 __dev_close_many(head);
1373 list_for_each_entry_safe(dev, tmp, head, close_list) {
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1375 call_netdevice_notifiers(NETDEV_DOWN, dev);
1376 list_del_init(&dev->close_list);
1383 * dev_close - shutdown an interface.
1384 * @dev: device to shutdown
1386 * This function moves an active device into down state. A
1387 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1388 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1391 int dev_close(struct net_device *dev)
1393 if (dev->flags & IFF_UP) {
1396 /* Block netpoll rx while the interface is going down */
1397 netpoll_rx_disable(dev);
1399 list_add(&dev->close_list, &single);
1400 dev_close_many(&single);
1403 netpoll_rx_enable(dev);
1407 EXPORT_SYMBOL(dev_close);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device *dev)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev))
1425 dev = vlan_dev_real_dev(dev);
1427 /* the same for macvlan devices */
1428 if (netif_is_macvlan(dev))
1429 dev = macvlan_dev_real_dev(dev);
1431 dev->wanted_features &= ~NETIF_F_LRO;
1432 netdev_update_features(dev);
1434 if (unlikely(dev->features & NETIF_F_LRO))
1435 netdev_WARN(dev, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro);
1439 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1440 struct net_device *dev)
1442 struct netdev_notifier_info info;
1444 netdev_notifier_info_init(&info, dev);
1445 return nb->notifier_call(nb, val, &info);
1448 static int dev_boot_phase = 1;
1451 * register_netdevice_notifier - register a network notifier block
1454 * Register a notifier to be called when network device events occur.
1455 * The notifier passed is linked into the kernel structures and must
1456 * not be reused until it has been unregistered. A negative errno code
1457 * is returned on a failure.
1459 * When registered all registration and up events are replayed
1460 * to the new notifier to allow device to have a race free
1461 * view of the network device list.
1464 int register_netdevice_notifier(struct notifier_block *nb)
1466 struct net_device *dev;
1467 struct net_device *last;
1472 err = raw_notifier_chain_register(&netdev_chain, nb);
1478 for_each_netdev(net, dev) {
1479 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1480 err = notifier_to_errno(err);
1484 if (!(dev->flags & IFF_UP))
1487 call_netdevice_notifier(nb, NETDEV_UP, dev);
1498 for_each_netdev(net, dev) {
1502 if (dev->flags & IFF_UP) {
1503 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1505 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1507 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1512 raw_notifier_chain_unregister(&netdev_chain, nb);
1515 EXPORT_SYMBOL(register_netdevice_notifier);
1518 * unregister_netdevice_notifier - unregister a network notifier block
1521 * Unregister a notifier previously registered by
1522 * register_netdevice_notifier(). The notifier is unlinked into the
1523 * kernel structures and may then be reused. A negative errno code
1524 * is returned on a failure.
1526 * After unregistering unregister and down device events are synthesized
1527 * for all devices on the device list to the removed notifier to remove
1528 * the need for special case cleanup code.
1531 int unregister_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1538 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1543 for_each_netdev(net, dev) {
1544 if (dev->flags & IFF_UP) {
1545 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1547 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1549 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier);
1559 * call_netdevice_notifiers_info - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1562 * @info: notifier information data
1564 * Call all network notifier blocks. Parameters and return value
1565 * are as for raw_notifier_call_chain().
1568 static int call_netdevice_notifiers_info(unsigned long val,
1569 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);
1578 * call_netdevice_notifiers - call all network notifier blocks
1579 * @val: value passed unmodified to notifier function
1580 * @dev: net_device pointer passed unmodified to notifier function
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1588 struct netdev_notifier_info info;
1590 return call_netdevice_notifiers_info(val, dev, &info);
1592 EXPORT_SYMBOL(call_netdevice_notifiers);
1594 static struct static_key netstamp_needed __read_mostly;
1595 #ifdef HAVE_JUMP_LABEL
1596 /* We are not allowed to call static_key_slow_dec() from irq context
1597 * If net_disable_timestamp() is called from irq context, defer the
1598 * static_key_slow_dec() calls.
1600 static atomic_t netstamp_needed_deferred;
1603 void net_enable_timestamp(void)
1605 #ifdef HAVE_JUMP_LABEL
1606 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1610 static_key_slow_dec(&netstamp_needed);
1614 static_key_slow_inc(&netstamp_needed);
1616 EXPORT_SYMBOL(net_enable_timestamp);
1618 void net_disable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 if (in_interrupt()) {
1622 atomic_inc(&netstamp_needed_deferred);
1626 static_key_slow_dec(&netstamp_needed);
1628 EXPORT_SYMBOL(net_disable_timestamp);
1630 static inline void net_timestamp_set(struct sk_buff *skb)
1632 skb->tstamp.tv64 = 0;
1633 if (static_key_false(&netstamp_needed))
1634 __net_timestamp(skb);
1637 #define net_timestamp_check(COND, SKB) \
1638 if (static_key_false(&netstamp_needed)) { \
1639 if ((COND) && !(SKB)->tstamp.tv64) \
1640 __net_timestamp(SKB); \
1643 static inline bool is_skb_forwardable(struct net_device *dev,
1644 struct sk_buff *skb)
1648 if (!(dev->flags & IFF_UP))
1651 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1652 if (skb->len <= len)
1655 /* if TSO is enabled, we don't care about the length as the packet
1656 * could be forwarded without being segmented before
1658 if (skb_is_gso(skb))
1665 * dev_forward_skb - loopback an skb to another netif
1667 * @dev: destination network device
1668 * @skb: buffer to forward
1671 * NET_RX_SUCCESS (no congestion)
1672 * NET_RX_DROP (packet was dropped, but freed)
1674 * dev_forward_skb can be used for injecting an skb from the
1675 * start_xmit function of one device into the receive queue
1676 * of another device.
1678 * The receiving device may be in another namespace, so
1679 * we have to clear all information in the skb that could
1680 * impact namespace isolation.
1682 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1686 atomic_long_inc(&dev->rx_dropped);
1692 if (unlikely(!is_skb_forwardable(dev, skb))) {
1693 atomic_long_inc(&dev->rx_dropped);
1698 skb_scrub_packet(skb, true);
1699 skb->protocol = eth_type_trans(skb, dev);
1701 return netif_rx(skb);
1703 EXPORT_SYMBOL_GPL(dev_forward_skb);
1705 static inline int deliver_skb(struct sk_buff *skb,
1706 struct packet_type *pt_prev,
1707 struct net_device *orig_dev)
1709 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1711 atomic_inc(&skb->users);
1712 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1715 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1717 if (!ptype->af_packet_priv || !skb->sk)
1720 if (ptype->id_match)
1721 return ptype->id_match(ptype, skb->sk);
1722 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1729 * Support routine. Sends outgoing frames to any network
1730 * taps currently in use.
1733 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1735 struct packet_type *ptype;
1736 struct sk_buff *skb2 = NULL;
1737 struct packet_type *pt_prev = NULL;
1740 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1741 /* Never send packets back to the socket
1742 * they originated from - MvS (miquels@drinkel.ow.org)
1744 if ((ptype->dev == dev || !ptype->dev) &&
1745 (!skb_loop_sk(ptype, skb))) {
1747 deliver_skb(skb2, pt_prev, skb->dev);
1752 skb2 = skb_clone(skb, GFP_ATOMIC);
1756 net_timestamp_set(skb2);
1758 /* skb->nh should be correctly
1759 set by sender, so that the second statement is
1760 just protection against buggy protocols.
1762 skb_reset_mac_header(skb2);
1764 if (skb_network_header(skb2) < skb2->data ||
1765 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1766 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1767 ntohs(skb2->protocol),
1769 skb_reset_network_header(skb2);
1772 skb2->transport_header = skb2->network_header;
1773 skb2->pkt_type = PACKET_OUTGOING;
1778 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1783 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1784 * @dev: Network device
1785 * @txq: number of queues available
1787 * If real_num_tx_queues is changed the tc mappings may no longer be
1788 * valid. To resolve this verify the tc mapping remains valid and if
1789 * not NULL the mapping. With no priorities mapping to this
1790 * offset/count pair it will no longer be used. In the worst case TC0
1791 * is invalid nothing can be done so disable priority mappings. If is
1792 * expected that drivers will fix this mapping if they can before
1793 * calling netif_set_real_num_tx_queues.
1795 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1798 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1800 /* If TC0 is invalidated disable TC mapping */
1801 if (tc->offset + tc->count > txq) {
1802 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1807 /* Invalidated prio to tc mappings set to TC0 */
1808 for (i = 1; i < TC_BITMASK + 1; i++) {
1809 int q = netdev_get_prio_tc_map(dev, i);
1811 tc = &dev->tc_to_txq[q];
1812 if (tc->offset + tc->count > txq) {
1813 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1815 netdev_set_prio_tc_map(dev, i, 0);
1821 static DEFINE_MUTEX(xps_map_mutex);
1822 #define xmap_dereference(P) \
1823 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1825 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1828 struct xps_map *map = NULL;
1832 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1834 for (pos = 0; map && pos < map->len; pos++) {
1835 if (map->queues[pos] == index) {
1837 map->queues[pos] = map->queues[--map->len];
1839 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1840 kfree_rcu(map, rcu);
1850 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1852 struct xps_dev_maps *dev_maps;
1854 bool active = false;
1856 mutex_lock(&xps_map_mutex);
1857 dev_maps = xmap_dereference(dev->xps_maps);
1862 for_each_possible_cpu(cpu) {
1863 for (i = index; i < dev->num_tx_queues; i++) {
1864 if (!remove_xps_queue(dev_maps, cpu, i))
1867 if (i == dev->num_tx_queues)
1872 RCU_INIT_POINTER(dev->xps_maps, NULL);
1873 kfree_rcu(dev_maps, rcu);
1876 for (i = index; i < dev->num_tx_queues; i++)
1877 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1881 mutex_unlock(&xps_map_mutex);
1884 static struct xps_map *expand_xps_map(struct xps_map *map,
1887 struct xps_map *new_map;
1888 int alloc_len = XPS_MIN_MAP_ALLOC;
1891 for (pos = 0; map && pos < map->len; pos++) {
1892 if (map->queues[pos] != index)
1897 /* Need to add queue to this CPU's existing map */
1899 if (pos < map->alloc_len)
1902 alloc_len = map->alloc_len * 2;
1905 /* Need to allocate new map to store queue on this CPU's map */
1906 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1911 for (i = 0; i < pos; i++)
1912 new_map->queues[i] = map->queues[i];
1913 new_map->alloc_len = alloc_len;
1919 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1922 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1923 struct xps_map *map, *new_map;
1924 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1925 int cpu, numa_node_id = -2;
1926 bool active = false;
1928 mutex_lock(&xps_map_mutex);
1930 dev_maps = xmap_dereference(dev->xps_maps);
1932 /* allocate memory for queue storage */
1933 for_each_online_cpu(cpu) {
1934 if (!cpumask_test_cpu(cpu, mask))
1938 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1939 if (!new_dev_maps) {
1940 mutex_unlock(&xps_map_mutex);
1944 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1947 map = expand_xps_map(map, cpu, index);
1951 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1955 goto out_no_new_maps;
1957 for_each_possible_cpu(cpu) {
1958 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1959 /* add queue to CPU maps */
1962 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1963 while ((pos < map->len) && (map->queues[pos] != index))
1966 if (pos == map->len)
1967 map->queues[map->len++] = index;
1969 if (numa_node_id == -2)
1970 numa_node_id = cpu_to_node(cpu);
1971 else if (numa_node_id != cpu_to_node(cpu))
1974 } else if (dev_maps) {
1975 /* fill in the new device map from the old device map */
1976 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1982 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1984 /* Cleanup old maps */
1986 for_each_possible_cpu(cpu) {
1987 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 if (map && map != new_map)
1990 kfree_rcu(map, rcu);
1993 kfree_rcu(dev_maps, rcu);
1996 dev_maps = new_dev_maps;
2000 /* update Tx queue numa node */
2001 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2002 (numa_node_id >= 0) ? numa_node_id :
2008 /* removes queue from unused CPUs */
2009 for_each_possible_cpu(cpu) {
2010 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2013 if (remove_xps_queue(dev_maps, cpu, index))
2017 /* free map if not active */
2019 RCU_INIT_POINTER(dev->xps_maps, NULL);
2020 kfree_rcu(dev_maps, rcu);
2024 mutex_unlock(&xps_map_mutex);
2028 /* remove any maps that we added */
2029 for_each_possible_cpu(cpu) {
2030 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2031 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2033 if (new_map && new_map != map)
2037 mutex_unlock(&xps_map_mutex);
2039 kfree(new_dev_maps);
2042 EXPORT_SYMBOL(netif_set_xps_queue);
2046 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2047 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2049 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2053 if (txq < 1 || txq > dev->num_tx_queues)
2056 if (dev->reg_state == NETREG_REGISTERED ||
2057 dev->reg_state == NETREG_UNREGISTERING) {
2060 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066 netif_setup_tc(dev, txq);
2068 if (txq < dev->real_num_tx_queues) {
2069 qdisc_reset_all_tx_gt(dev, txq);
2071 netif_reset_xps_queues_gt(dev, txq);
2076 dev->real_num_tx_queues = txq;
2079 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2083 * netif_set_real_num_rx_queues - set actual number of RX queues used
2084 * @dev: Network device
2085 * @rxq: Actual number of RX queues
2087 * This must be called either with the rtnl_lock held or before
2088 * registration of the net device. Returns 0 on success, or a
2089 * negative error code. If called before registration, it always
2092 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2096 if (rxq < 1 || rxq > dev->num_rx_queues)
2099 if (dev->reg_state == NETREG_REGISTERED) {
2102 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108 dev->real_num_rx_queues = rxq;
2111 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2115 * netif_get_num_default_rss_queues - default number of RSS queues
2117 * This routine should set an upper limit on the number of RSS queues
2118 * used by default by multiqueue devices.
2120 int netif_get_num_default_rss_queues(void)
2122 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2124 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2126 static inline void __netif_reschedule(struct Qdisc *q)
2128 struct softnet_data *sd;
2129 unsigned long flags;
2131 local_irq_save(flags);
2132 sd = &__get_cpu_var(softnet_data);
2133 q->next_sched = NULL;
2134 *sd->output_queue_tailp = q;
2135 sd->output_queue_tailp = &q->next_sched;
2136 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2137 local_irq_restore(flags);
2140 void __netif_schedule(struct Qdisc *q)
2142 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2143 __netif_reschedule(q);
2145 EXPORT_SYMBOL(__netif_schedule);
2147 struct dev_kfree_skb_cb {
2148 enum skb_free_reason reason;
2151 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2153 return (struct dev_kfree_skb_cb *)skb->cb;
2156 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2158 unsigned long flags;
2160 if (likely(atomic_read(&skb->users) == 1)) {
2162 atomic_set(&skb->users, 0);
2163 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2166 get_kfree_skb_cb(skb)->reason = reason;
2167 local_irq_save(flags);
2168 skb->next = __this_cpu_read(softnet_data.completion_queue);
2169 __this_cpu_write(softnet_data.completion_queue, skb);
2170 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2171 local_irq_restore(flags);
2173 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2175 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2177 if (in_irq() || irqs_disabled())
2178 __dev_kfree_skb_irq(skb, reason);
2182 EXPORT_SYMBOL(__dev_kfree_skb_any);
2186 * netif_device_detach - mark device as removed
2187 * @dev: network device
2189 * Mark device as removed from system and therefore no longer available.
2191 void netif_device_detach(struct net_device *dev)
2193 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2194 netif_running(dev)) {
2195 netif_tx_stop_all_queues(dev);
2198 EXPORT_SYMBOL(netif_device_detach);
2201 * netif_device_attach - mark device as attached
2202 * @dev: network device
2204 * Mark device as attached from system and restart if needed.
2206 void netif_device_attach(struct net_device *dev)
2208 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2209 netif_running(dev)) {
2210 netif_tx_wake_all_queues(dev);
2211 __netdev_watchdog_up(dev);
2214 EXPORT_SYMBOL(netif_device_attach);
2216 static void skb_warn_bad_offload(const struct sk_buff *skb)
2218 static const netdev_features_t null_features = 0;
2219 struct net_device *dev = skb->dev;
2220 const char *driver = "";
2222 if (!net_ratelimit())
2225 if (dev && dev->dev.parent)
2226 driver = dev_driver_string(dev->dev.parent);
2228 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2229 "gso_type=%d ip_summed=%d\n",
2230 driver, dev ? &dev->features : &null_features,
2231 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2232 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2233 skb_shinfo(skb)->gso_type, skb->ip_summed);
2237 * Invalidate hardware checksum when packet is to be mangled, and
2238 * complete checksum manually on outgoing path.
2240 int skb_checksum_help(struct sk_buff *skb)
2243 int ret = 0, offset;
2245 if (skb->ip_summed == CHECKSUM_COMPLETE)
2246 goto out_set_summed;
2248 if (unlikely(skb_shinfo(skb)->gso_size)) {
2249 skb_warn_bad_offload(skb);
2253 /* Before computing a checksum, we should make sure no frag could
2254 * be modified by an external entity : checksum could be wrong.
2256 if (skb_has_shared_frag(skb)) {
2257 ret = __skb_linearize(skb);
2262 offset = skb_checksum_start_offset(skb);
2263 BUG_ON(offset >= skb_headlen(skb));
2264 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2266 offset += skb->csum_offset;
2267 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2269 if (skb_cloned(skb) &&
2270 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2271 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2276 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278 skb->ip_summed = CHECKSUM_NONE;
2282 EXPORT_SYMBOL(skb_checksum_help);
2284 __be16 skb_network_protocol(struct sk_buff *skb)
2286 __be16 type = skb->protocol;
2287 int vlan_depth = ETH_HLEN;
2289 /* Tunnel gso handlers can set protocol to ethernet. */
2290 if (type == htons(ETH_P_TEB)) {
2293 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2296 eth = (struct ethhdr *)skb_mac_header(skb);
2297 type = eth->h_proto;
2300 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2301 struct vlan_hdr *vh;
2303 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2306 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2307 type = vh->h_vlan_encapsulated_proto;
2308 vlan_depth += VLAN_HLEN;
2315 * skb_mac_gso_segment - mac layer segmentation handler.
2316 * @skb: buffer to segment
2317 * @features: features for the output path (see dev->features)
2319 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2320 netdev_features_t features)
2322 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2323 struct packet_offload *ptype;
2324 __be16 type = skb_network_protocol(skb);
2326 if (unlikely(!type))
2327 return ERR_PTR(-EINVAL);
2329 __skb_pull(skb, skb->mac_len);
2332 list_for_each_entry_rcu(ptype, &offload_base, list) {
2333 if (ptype->type == type && ptype->callbacks.gso_segment) {
2334 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2337 err = ptype->callbacks.gso_send_check(skb);
2338 segs = ERR_PTR(err);
2339 if (err || skb_gso_ok(skb, features))
2341 __skb_push(skb, (skb->data -
2342 skb_network_header(skb)));
2344 segs = ptype->callbacks.gso_segment(skb, features);
2350 __skb_push(skb, skb->data - skb_mac_header(skb));
2354 EXPORT_SYMBOL(skb_mac_gso_segment);
2357 /* openvswitch calls this on rx path, so we need a different check.
2359 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2362 return skb->ip_summed != CHECKSUM_PARTIAL;
2364 return skb->ip_summed == CHECKSUM_NONE;
2368 * __skb_gso_segment - Perform segmentation on skb.
2369 * @skb: buffer to segment
2370 * @features: features for the output path (see dev->features)
2371 * @tx_path: whether it is called in TX path
2373 * This function segments the given skb and returns a list of segments.
2375 * It may return NULL if the skb requires no segmentation. This is
2376 * only possible when GSO is used for verifying header integrity.
2378 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2379 netdev_features_t features, bool tx_path)
2381 if (unlikely(skb_needs_check(skb, tx_path))) {
2384 skb_warn_bad_offload(skb);
2386 if (skb_header_cloned(skb) &&
2387 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2388 return ERR_PTR(err);
2391 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2392 SKB_GSO_CB(skb)->encap_level = 0;
2394 skb_reset_mac_header(skb);
2395 skb_reset_mac_len(skb);
2397 return skb_mac_gso_segment(skb, features);
2399 EXPORT_SYMBOL(__skb_gso_segment);
2401 /* Take action when hardware reception checksum errors are detected. */
2403 void netdev_rx_csum_fault(struct net_device *dev)
2405 if (net_ratelimit()) {
2406 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2410 EXPORT_SYMBOL(netdev_rx_csum_fault);
2413 /* Actually, we should eliminate this check as soon as we know, that:
2414 * 1. IOMMU is present and allows to map all the memory.
2415 * 2. No high memory really exists on this machine.
2418 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2420 #ifdef CONFIG_HIGHMEM
2422 if (!(dev->features & NETIF_F_HIGHDMA)) {
2423 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2424 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2425 if (PageHighMem(skb_frag_page(frag)))
2430 if (PCI_DMA_BUS_IS_PHYS) {
2431 struct device *pdev = dev->dev.parent;
2435 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2436 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2437 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2438 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2447 void (*destructor)(struct sk_buff *skb);
2450 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2452 static void dev_gso_skb_destructor(struct sk_buff *skb)
2454 struct dev_gso_cb *cb;
2456 kfree_skb_list(skb->next);
2459 cb = DEV_GSO_CB(skb);
2461 cb->destructor(skb);
2465 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2466 * @skb: buffer to segment
2467 * @features: device features as applicable to this skb
2469 * This function segments the given skb and stores the list of segments
2472 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2474 struct sk_buff *segs;
2476 segs = skb_gso_segment(skb, features);
2478 /* Verifying header integrity only. */
2483 return PTR_ERR(segs);
2486 DEV_GSO_CB(skb)->destructor = skb->destructor;
2487 skb->destructor = dev_gso_skb_destructor;
2492 static netdev_features_t harmonize_features(struct sk_buff *skb,
2493 netdev_features_t features)
2495 if (skb->ip_summed != CHECKSUM_NONE &&
2496 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2497 features &= ~NETIF_F_ALL_CSUM;
2498 } else if (illegal_highdma(skb->dev, skb)) {
2499 features &= ~NETIF_F_SG;
2505 netdev_features_t netif_skb_features(struct sk_buff *skb)
2507 __be16 protocol = skb->protocol;
2508 netdev_features_t features = skb->dev->features;
2510 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2511 features &= ~NETIF_F_GSO_MASK;
2513 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2514 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2515 protocol = veh->h_vlan_encapsulated_proto;
2516 } else if (!vlan_tx_tag_present(skb)) {
2517 return harmonize_features(skb, features);
2520 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2521 NETIF_F_HW_VLAN_STAG_TX);
2523 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2524 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2525 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2526 NETIF_F_HW_VLAN_STAG_TX;
2528 return harmonize_features(skb, features);
2530 EXPORT_SYMBOL(netif_skb_features);
2532 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2533 struct netdev_queue *txq, void *accel_priv)
2535 const struct net_device_ops *ops = dev->netdev_ops;
2536 int rc = NETDEV_TX_OK;
2537 unsigned int skb_len;
2539 if (likely(!skb->next)) {
2540 netdev_features_t features;
2543 * If device doesn't need skb->dst, release it right now while
2544 * its hot in this cpu cache
2546 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2549 features = netif_skb_features(skb);
2551 if (vlan_tx_tag_present(skb) &&
2552 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2553 skb = __vlan_put_tag(skb, skb->vlan_proto,
2554 vlan_tx_tag_get(skb));
2561 /* If encapsulation offload request, verify we are testing
2562 * hardware encapsulation features instead of standard
2563 * features for the netdev
2565 if (skb->encapsulation)
2566 features &= dev->hw_enc_features;
2568 if (netif_needs_gso(skb, features)) {
2569 if (unlikely(dev_gso_segment(skb, features)))
2574 if (skb_needs_linearize(skb, features) &&
2575 __skb_linearize(skb))
2578 /* If packet is not checksummed and device does not
2579 * support checksumming for this protocol, complete
2580 * checksumming here.
2582 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2583 if (skb->encapsulation)
2584 skb_set_inner_transport_header(skb,
2585 skb_checksum_start_offset(skb));
2587 skb_set_transport_header(skb,
2588 skb_checksum_start_offset(skb));
2589 if (!(features & NETIF_F_ALL_CSUM) &&
2590 skb_checksum_help(skb))
2595 if (!list_empty(&ptype_all))
2596 dev_queue_xmit_nit(skb, dev);
2600 rc = ops->ndo_dfwd_start_xmit(skb, dev, accel_priv);
2602 rc = ops->ndo_start_xmit(skb, dev);
2604 trace_net_dev_xmit(skb, rc, dev, skb_len);
2605 if (rc == NETDEV_TX_OK && txq)
2606 txq_trans_update(txq);
2612 struct sk_buff *nskb = skb->next;
2614 skb->next = nskb->next;
2617 if (!list_empty(&ptype_all))
2618 dev_queue_xmit_nit(nskb, dev);
2620 skb_len = nskb->len;
2622 rc = ops->ndo_dfwd_start_xmit(nskb, dev, accel_priv);
2624 rc = ops->ndo_start_xmit(nskb, dev);
2625 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2626 if (unlikely(rc != NETDEV_TX_OK)) {
2627 if (rc & ~NETDEV_TX_MASK)
2628 goto out_kfree_gso_skb;
2629 nskb->next = skb->next;
2633 txq_trans_update(txq);
2634 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2635 return NETDEV_TX_BUSY;
2636 } while (skb->next);
2639 if (likely(skb->next == NULL)) {
2640 skb->destructor = DEV_GSO_CB(skb)->destructor;
2649 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2784 * Queue a buffer for transmission to a network device. The caller must
2785 * have set the device and priority and built the buffer before calling
2786 * this function. The function can be called from an interrupt.
2788 * A negative errno code is returned on a failure. A success does not
2789 * guarantee the frame will be transmitted as it may be dropped due
2790 * to congestion or traffic shaping.
2792 * -----------------------------------------------------------------------------------
2793 * I notice this method can also return errors from the queue disciplines,
2794 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2797 * Regardless of the return value, the skb is consumed, so it is currently
2798 * difficult to retry a send to this method. (You can bump the ref count
2799 * before sending to hold a reference for retry if you are careful.)
2801 * When calling this method, interrupts MUST be enabled. This is because
2802 * the BH enable code must have IRQs enabled so that it will not deadlock.
2805 int dev_queue_xmit(struct sk_buff *skb)
2807 struct net_device *dev = skb->dev;
2808 struct netdev_queue *txq;
2812 skb_reset_mac_header(skb);
2814 /* Disable soft irqs for various locks below. Also
2815 * stops preemption for RCU.
2819 skb_update_prio(skb);
2821 txq = netdev_pick_tx(dev, skb);
2822 q = rcu_dereference_bh(txq->qdisc);
2824 #ifdef CONFIG_NET_CLS_ACT
2825 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2827 trace_net_dev_queue(skb);
2829 rc = __dev_xmit_skb(skb, q, dev, txq);
2833 /* The device has no queue. Common case for software devices:
2834 loopback, all the sorts of tunnels...
2836 Really, it is unlikely that netif_tx_lock protection is necessary
2837 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2839 However, it is possible, that they rely on protection
2842 Check this and shot the lock. It is not prone from deadlocks.
2843 Either shot noqueue qdisc, it is even simpler 8)
2845 if (dev->flags & IFF_UP) {
2846 int cpu = smp_processor_id(); /* ok because BHs are off */
2848 if (txq->xmit_lock_owner != cpu) {
2850 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2851 goto recursion_alert;
2853 HARD_TX_LOCK(dev, txq, cpu);
2855 if (!netif_xmit_stopped(txq)) {
2856 __this_cpu_inc(xmit_recursion);
2857 rc = dev_hard_start_xmit(skb, dev, txq, NULL);
2858 __this_cpu_dec(xmit_recursion);
2859 if (dev_xmit_complete(rc)) {
2860 HARD_TX_UNLOCK(dev, txq);
2864 HARD_TX_UNLOCK(dev, txq);
2865 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2868 /* Recursion is detected! It is possible,
2872 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2878 rcu_read_unlock_bh();
2883 rcu_read_unlock_bh();
2886 EXPORT_SYMBOL(dev_queue_xmit);
2889 /*=======================================================================
2891 =======================================================================*/
2893 int netdev_max_backlog __read_mostly = 1000;
2894 EXPORT_SYMBOL(netdev_max_backlog);
2896 int netdev_tstamp_prequeue __read_mostly = 1;
2897 int netdev_budget __read_mostly = 300;
2898 int weight_p __read_mostly = 64; /* old backlog weight */
2900 /* Called with irq disabled */
2901 static inline void ____napi_schedule(struct softnet_data *sd,
2902 struct napi_struct *napi)
2904 list_add_tail(&napi->poll_list, &sd->poll_list);
2905 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2910 /* One global table that all flow-based protocols share. */
2911 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2912 EXPORT_SYMBOL(rps_sock_flow_table);
2914 struct static_key rps_needed __read_mostly;
2916 static struct rps_dev_flow *
2917 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2918 struct rps_dev_flow *rflow, u16 next_cpu)
2920 if (next_cpu != RPS_NO_CPU) {
2921 #ifdef CONFIG_RFS_ACCEL
2922 struct netdev_rx_queue *rxqueue;
2923 struct rps_dev_flow_table *flow_table;
2924 struct rps_dev_flow *old_rflow;
2929 /* Should we steer this flow to a different hardware queue? */
2930 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2931 !(dev->features & NETIF_F_NTUPLE))
2933 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2934 if (rxq_index == skb_get_rx_queue(skb))
2937 rxqueue = dev->_rx + rxq_index;
2938 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2941 flow_id = skb->rxhash & flow_table->mask;
2942 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2943 rxq_index, flow_id);
2947 rflow = &flow_table->flows[flow_id];
2949 if (old_rflow->filter == rflow->filter)
2950 old_rflow->filter = RPS_NO_FILTER;
2954 per_cpu(softnet_data, next_cpu).input_queue_head;
2957 rflow->cpu = next_cpu;
2962 * get_rps_cpu is called from netif_receive_skb and returns the target
2963 * CPU from the RPS map of the receiving queue for a given skb.
2964 * rcu_read_lock must be held on entry.
2966 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2967 struct rps_dev_flow **rflowp)
2969 struct netdev_rx_queue *rxqueue;
2970 struct rps_map *map;
2971 struct rps_dev_flow_table *flow_table;
2972 struct rps_sock_flow_table *sock_flow_table;
2976 if (skb_rx_queue_recorded(skb)) {
2977 u16 index = skb_get_rx_queue(skb);
2978 if (unlikely(index >= dev->real_num_rx_queues)) {
2979 WARN_ONCE(dev->real_num_rx_queues > 1,
2980 "%s received packet on queue %u, but number "
2981 "of RX queues is %u\n",
2982 dev->name, index, dev->real_num_rx_queues);
2985 rxqueue = dev->_rx + index;
2989 map = rcu_dereference(rxqueue->rps_map);
2991 if (map->len == 1 &&
2992 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2993 tcpu = map->cpus[0];
2994 if (cpu_online(tcpu))
2998 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3002 skb_reset_network_header(skb);
3003 if (!skb_get_hash(skb))
3006 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3007 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3008 if (flow_table && sock_flow_table) {
3010 struct rps_dev_flow *rflow;
3012 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3015 next_cpu = sock_flow_table->ents[skb->rxhash &
3016 sock_flow_table->mask];
3019 * If the desired CPU (where last recvmsg was done) is
3020 * different from current CPU (one in the rx-queue flow
3021 * table entry), switch if one of the following holds:
3022 * - Current CPU is unset (equal to RPS_NO_CPU).
3023 * - Current CPU is offline.
3024 * - The current CPU's queue tail has advanced beyond the
3025 * last packet that was enqueued using this table entry.
3026 * This guarantees that all previous packets for the flow
3027 * have been dequeued, thus preserving in order delivery.
3029 if (unlikely(tcpu != next_cpu) &&
3030 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3031 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3032 rflow->last_qtail)) >= 0)) {
3034 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3037 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3045 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3047 if (cpu_online(tcpu)) {
3057 #ifdef CONFIG_RFS_ACCEL
3060 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3061 * @dev: Device on which the filter was set
3062 * @rxq_index: RX queue index
3063 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3064 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3066 * Drivers that implement ndo_rx_flow_steer() should periodically call
3067 * this function for each installed filter and remove the filters for
3068 * which it returns %true.
3070 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3071 u32 flow_id, u16 filter_id)
3073 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3074 struct rps_dev_flow_table *flow_table;
3075 struct rps_dev_flow *rflow;
3080 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3081 if (flow_table && flow_id <= flow_table->mask) {
3082 rflow = &flow_table->flows[flow_id];
3083 cpu = ACCESS_ONCE(rflow->cpu);
3084 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3085 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3086 rflow->last_qtail) <
3087 (int)(10 * flow_table->mask)))
3093 EXPORT_SYMBOL(rps_may_expire_flow);
3095 #endif /* CONFIG_RFS_ACCEL */
3097 /* Called from hardirq (IPI) context */
3098 static void rps_trigger_softirq(void *data)
3100 struct softnet_data *sd = data;
3102 ____napi_schedule(sd, &sd->backlog);
3106 #endif /* CONFIG_RPS */
3109 * Check if this softnet_data structure is another cpu one
3110 * If yes, queue it to our IPI list and return 1
3113 static int rps_ipi_queued(struct softnet_data *sd)
3116 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3119 sd->rps_ipi_next = mysd->rps_ipi_list;
3120 mysd->rps_ipi_list = sd;
3122 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3125 #endif /* CONFIG_RPS */
3129 #ifdef CONFIG_NET_FLOW_LIMIT
3130 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3133 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3135 #ifdef CONFIG_NET_FLOW_LIMIT
3136 struct sd_flow_limit *fl;
3137 struct softnet_data *sd;
3138 unsigned int old_flow, new_flow;
3140 if (qlen < (netdev_max_backlog >> 1))
3143 sd = &__get_cpu_var(softnet_data);
3146 fl = rcu_dereference(sd->flow_limit);
3148 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3149 old_flow = fl->history[fl->history_head];
3150 fl->history[fl->history_head] = new_flow;
3153 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3155 if (likely(fl->buckets[old_flow]))
3156 fl->buckets[old_flow]--;
3158 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3170 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3171 * queue (may be a remote CPU queue).
3173 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3174 unsigned int *qtail)
3176 struct softnet_data *sd;
3177 unsigned long flags;
3180 sd = &per_cpu(softnet_data, cpu);
3182 local_irq_save(flags);
3185 qlen = skb_queue_len(&sd->input_pkt_queue);
3186 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3187 if (skb_queue_len(&sd->input_pkt_queue)) {
3189 __skb_queue_tail(&sd->input_pkt_queue, skb);
3190 input_queue_tail_incr_save(sd, qtail);
3192 local_irq_restore(flags);
3193 return NET_RX_SUCCESS;
3196 /* Schedule NAPI for backlog device
3197 * We can use non atomic operation since we own the queue lock
3199 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3200 if (!rps_ipi_queued(sd))
3201 ____napi_schedule(sd, &sd->backlog);
3209 local_irq_restore(flags);
3211 atomic_long_inc(&skb->dev->rx_dropped);
3217 * netif_rx - post buffer to the network code
3218 * @skb: buffer to post
3220 * This function receives a packet from a device driver and queues it for
3221 * the upper (protocol) levels to process. It always succeeds. The buffer
3222 * may be dropped during processing for congestion control or by the
3226 * NET_RX_SUCCESS (no congestion)
3227 * NET_RX_DROP (packet was dropped)
3231 int netif_rx(struct sk_buff *skb)
3235 /* if netpoll wants it, pretend we never saw it */
3236 if (netpoll_rx(skb))
3239 net_timestamp_check(netdev_tstamp_prequeue, skb);
3241 trace_netif_rx(skb);
3243 if (static_key_false(&rps_needed)) {
3244 struct rps_dev_flow voidflow, *rflow = &voidflow;
3250 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3252 cpu = smp_processor_id();
3254 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3262 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3267 EXPORT_SYMBOL(netif_rx);
3269 int netif_rx_ni(struct sk_buff *skb)
3274 err = netif_rx(skb);
3275 if (local_softirq_pending())
3281 EXPORT_SYMBOL(netif_rx_ni);
3283 static void net_tx_action(struct softirq_action *h)
3285 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3287 if (sd->completion_queue) {
3288 struct sk_buff *clist;
3290 local_irq_disable();
3291 clist = sd->completion_queue;
3292 sd->completion_queue = NULL;
3296 struct sk_buff *skb = clist;
3297 clist = clist->next;
3299 WARN_ON(atomic_read(&skb->users));
3300 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3301 trace_consume_skb(skb);
3303 trace_kfree_skb(skb, net_tx_action);
3308 if (sd->output_queue) {
3311 local_irq_disable();
3312 head = sd->output_queue;
3313 sd->output_queue = NULL;
3314 sd->output_queue_tailp = &sd->output_queue;
3318 struct Qdisc *q = head;
3319 spinlock_t *root_lock;
3321 head = head->next_sched;
3323 root_lock = qdisc_lock(q);
3324 if (spin_trylock(root_lock)) {
3325 smp_mb__before_clear_bit();
3326 clear_bit(__QDISC_STATE_SCHED,
3329 spin_unlock(root_lock);
3331 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3333 __netif_reschedule(q);
3335 smp_mb__before_clear_bit();
3336 clear_bit(__QDISC_STATE_SCHED,
3344 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3345 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3346 /* This hook is defined here for ATM LANE */
3347 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3348 unsigned char *addr) __read_mostly;
3349 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3352 #ifdef CONFIG_NET_CLS_ACT
3353 /* TODO: Maybe we should just force sch_ingress to be compiled in
3354 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3355 * a compare and 2 stores extra right now if we dont have it on
3356 * but have CONFIG_NET_CLS_ACT
3357 * NOTE: This doesn't stop any functionality; if you dont have
3358 * the ingress scheduler, you just can't add policies on ingress.
3361 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3363 struct net_device *dev = skb->dev;
3364 u32 ttl = G_TC_RTTL(skb->tc_verd);
3365 int result = TC_ACT_OK;
3368 if (unlikely(MAX_RED_LOOP < ttl++)) {
3369 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3370 skb->skb_iif, dev->ifindex);
3374 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3375 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3378 if (q != &noop_qdisc) {
3379 spin_lock(qdisc_lock(q));
3380 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3381 result = qdisc_enqueue_root(skb, q);
3382 spin_unlock(qdisc_lock(q));
3388 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3389 struct packet_type **pt_prev,
3390 int *ret, struct net_device *orig_dev)
3392 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3394 if (!rxq || rxq->qdisc == &noop_qdisc)
3398 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3402 switch (ing_filter(skb, rxq)) {
3416 * netdev_rx_handler_register - register receive handler
3417 * @dev: device to register a handler for
3418 * @rx_handler: receive handler to register
3419 * @rx_handler_data: data pointer that is used by rx handler
3421 * Register a receive hander for a device. This handler will then be
3422 * called from __netif_receive_skb. A negative errno code is returned
3425 * The caller must hold the rtnl_mutex.
3427 * For a general description of rx_handler, see enum rx_handler_result.
3429 int netdev_rx_handler_register(struct net_device *dev,
3430 rx_handler_func_t *rx_handler,
3431 void *rx_handler_data)
3435 if (dev->rx_handler)
3438 /* Note: rx_handler_data must be set before rx_handler */
3439 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3440 rcu_assign_pointer(dev->rx_handler, rx_handler);
3444 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3447 * netdev_rx_handler_unregister - unregister receive handler
3448 * @dev: device to unregister a handler from
3450 * Unregister a receive handler from a device.
3452 * The caller must hold the rtnl_mutex.
3454 void netdev_rx_handler_unregister(struct net_device *dev)
3458 RCU_INIT_POINTER(dev->rx_handler, NULL);
3459 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3460 * section has a guarantee to see a non NULL rx_handler_data
3464 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3466 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3469 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3470 * the special handling of PFMEMALLOC skbs.
3472 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3474 switch (skb->protocol) {
3475 case __constant_htons(ETH_P_ARP):
3476 case __constant_htons(ETH_P_IP):
3477 case __constant_htons(ETH_P_IPV6):
3478 case __constant_htons(ETH_P_8021Q):
3479 case __constant_htons(ETH_P_8021AD):
3486 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3488 struct packet_type *ptype, *pt_prev;
3489 rx_handler_func_t *rx_handler;
3490 struct net_device *orig_dev;
3491 struct net_device *null_or_dev;
3492 bool deliver_exact = false;
3493 int ret = NET_RX_DROP;
3496 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3498 trace_netif_receive_skb(skb);
3500 /* if we've gotten here through NAPI, check netpoll */
3501 if (netpoll_receive_skb(skb))
3504 orig_dev = skb->dev;
3506 skb_reset_network_header(skb);
3507 if (!skb_transport_header_was_set(skb))
3508 skb_reset_transport_header(skb);
3509 skb_reset_mac_len(skb);
3516 skb->skb_iif = skb->dev->ifindex;
3518 __this_cpu_inc(softnet_data.processed);
3520 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3521 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3522 skb = vlan_untag(skb);
3527 #ifdef CONFIG_NET_CLS_ACT
3528 if (skb->tc_verd & TC_NCLS) {
3529 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3537 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3538 if (!ptype->dev || ptype->dev == skb->dev) {
3540 ret = deliver_skb(skb, pt_prev, orig_dev);
3546 #ifdef CONFIG_NET_CLS_ACT
3547 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3553 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3556 if (vlan_tx_tag_present(skb)) {
3558 ret = deliver_skb(skb, pt_prev, orig_dev);
3561 if (vlan_do_receive(&skb))
3563 else if (unlikely(!skb))
3567 rx_handler = rcu_dereference(skb->dev->rx_handler);
3570 ret = deliver_skb(skb, pt_prev, orig_dev);
3573 switch (rx_handler(&skb)) {
3574 case RX_HANDLER_CONSUMED:
3575 ret = NET_RX_SUCCESS;
3577 case RX_HANDLER_ANOTHER:
3579 case RX_HANDLER_EXACT:
3580 deliver_exact = true;
3581 case RX_HANDLER_PASS:
3588 if (unlikely(vlan_tx_tag_present(skb))) {
3589 if (vlan_tx_tag_get_id(skb))
3590 skb->pkt_type = PACKET_OTHERHOST;
3591 /* Note: we might in the future use prio bits
3592 * and set skb->priority like in vlan_do_receive()
3593 * For the time being, just ignore Priority Code Point
3598 /* deliver only exact match when indicated */
3599 null_or_dev = deliver_exact ? skb->dev : NULL;
3601 type = skb->protocol;
3602 list_for_each_entry_rcu(ptype,
3603 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3604 if (ptype->type == type &&
3605 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3606 ptype->dev == orig_dev)) {
3608 ret = deliver_skb(skb, pt_prev, orig_dev);
3614 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3617 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3620 atomic_long_inc(&skb->dev->rx_dropped);
3622 /* Jamal, now you will not able to escape explaining
3623 * me how you were going to use this. :-)
3634 static int __netif_receive_skb(struct sk_buff *skb)
3638 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3639 unsigned long pflags = current->flags;
3642 * PFMEMALLOC skbs are special, they should
3643 * - be delivered to SOCK_MEMALLOC sockets only
3644 * - stay away from userspace
3645 * - have bounded memory usage
3647 * Use PF_MEMALLOC as this saves us from propagating the allocation
3648 * context down to all allocation sites.
3650 current->flags |= PF_MEMALLOC;
3651 ret = __netif_receive_skb_core(skb, true);
3652 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3654 ret = __netif_receive_skb_core(skb, false);
3660 * netif_receive_skb - process receive buffer from network
3661 * @skb: buffer to process
3663 * netif_receive_skb() is the main receive data processing function.
3664 * It always succeeds. The buffer may be dropped during processing
3665 * for congestion control or by the protocol layers.
3667 * This function may only be called from softirq context and interrupts
3668 * should be enabled.
3670 * Return values (usually ignored):
3671 * NET_RX_SUCCESS: no congestion
3672 * NET_RX_DROP: packet was dropped
3674 int netif_receive_skb(struct sk_buff *skb)
3676 net_timestamp_check(netdev_tstamp_prequeue, skb);
3678 if (skb_defer_rx_timestamp(skb))
3679 return NET_RX_SUCCESS;
3682 if (static_key_false(&rps_needed)) {
3683 struct rps_dev_flow voidflow, *rflow = &voidflow;
3688 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3691 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3698 return __netif_receive_skb(skb);
3700 EXPORT_SYMBOL(netif_receive_skb);
3702 /* Network device is going away, flush any packets still pending
3703 * Called with irqs disabled.
3705 static void flush_backlog(void *arg)
3707 struct net_device *dev = arg;
3708 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3709 struct sk_buff *skb, *tmp;
3712 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3713 if (skb->dev == dev) {
3714 __skb_unlink(skb, &sd->input_pkt_queue);
3716 input_queue_head_incr(sd);
3721 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3722 if (skb->dev == dev) {
3723 __skb_unlink(skb, &sd->process_queue);
3725 input_queue_head_incr(sd);
3730 static int napi_gro_complete(struct sk_buff *skb)
3732 struct packet_offload *ptype;
3733 __be16 type = skb->protocol;
3734 struct list_head *head = &offload_base;
3737 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3739 if (NAPI_GRO_CB(skb)->count == 1) {
3740 skb_shinfo(skb)->gso_size = 0;
3745 list_for_each_entry_rcu(ptype, head, list) {
3746 if (ptype->type != type || !ptype->callbacks.gro_complete)
3749 err = ptype->callbacks.gro_complete(skb, 0);
3755 WARN_ON(&ptype->list == head);
3757 return NET_RX_SUCCESS;
3761 return netif_receive_skb(skb);
3764 /* napi->gro_list contains packets ordered by age.
3765 * youngest packets at the head of it.
3766 * Complete skbs in reverse order to reduce latencies.
3768 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3770 struct sk_buff *skb, *prev = NULL;
3772 /* scan list and build reverse chain */
3773 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3778 for (skb = prev; skb; skb = prev) {
3781 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3785 napi_gro_complete(skb);
3789 napi->gro_list = NULL;
3791 EXPORT_SYMBOL(napi_gro_flush);
3793 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3796 unsigned int maclen = skb->dev->hard_header_len;
3798 for (p = napi->gro_list; p; p = p->next) {
3799 unsigned long diffs;
3801 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3802 diffs |= p->vlan_tci ^ skb->vlan_tci;
3803 if (maclen == ETH_HLEN)
3804 diffs |= compare_ether_header(skb_mac_header(p),
3805 skb_gro_mac_header(skb));
3807 diffs = memcmp(skb_mac_header(p),
3808 skb_gro_mac_header(skb),
3810 NAPI_GRO_CB(p)->same_flow = !diffs;
3811 NAPI_GRO_CB(p)->flush = 0;
3815 static void skb_gro_reset_offset(struct sk_buff *skb)
3817 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3818 const skb_frag_t *frag0 = &pinfo->frags[0];
3820 NAPI_GRO_CB(skb)->data_offset = 0;
3821 NAPI_GRO_CB(skb)->frag0 = NULL;
3822 NAPI_GRO_CB(skb)->frag0_len = 0;
3824 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3826 !PageHighMem(skb_frag_page(frag0))) {
3827 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3828 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3832 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3834 struct sk_buff **pp = NULL;
3835 struct packet_offload *ptype;
3836 __be16 type = skb->protocol;
3837 struct list_head *head = &offload_base;
3839 enum gro_result ret;
3841 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3844 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3847 skb_gro_reset_offset(skb);
3848 gro_list_prepare(napi, skb);
3849 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3852 list_for_each_entry_rcu(ptype, head, list) {
3853 if (ptype->type != type || !ptype->callbacks.gro_receive)
3856 skb_set_network_header(skb, skb_gro_offset(skb));
3857 skb_reset_mac_len(skb);
3858 NAPI_GRO_CB(skb)->same_flow = 0;
3859 NAPI_GRO_CB(skb)->flush = 0;
3860 NAPI_GRO_CB(skb)->free = 0;
3862 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3867 if (&ptype->list == head)
3870 same_flow = NAPI_GRO_CB(skb)->same_flow;
3871 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3874 struct sk_buff *nskb = *pp;
3878 napi_gro_complete(nskb);
3885 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3889 NAPI_GRO_CB(skb)->count = 1;
3890 NAPI_GRO_CB(skb)->age = jiffies;
3891 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3892 skb->next = napi->gro_list;
3893 napi->gro_list = skb;
3897 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3898 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3900 BUG_ON(skb->end - skb->tail < grow);
3902 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3905 skb->data_len -= grow;
3907 skb_shinfo(skb)->frags[0].page_offset += grow;
3908 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3910 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3911 skb_frag_unref(skb, 0);
3912 memmove(skb_shinfo(skb)->frags,
3913 skb_shinfo(skb)->frags + 1,
3914 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3926 struct packet_offload *gro_find_receive_by_type(__be16 type)
3928 struct list_head *offload_head = &offload_base;
3929 struct packet_offload *ptype;
3931 list_for_each_entry_rcu(ptype, offload_head, list) {
3932 if (ptype->type != type || !ptype->callbacks.gro_receive)
3939 struct packet_offload *gro_find_complete_by_type(__be16 type)
3941 struct list_head *offload_head = &offload_base;
3942 struct packet_offload *ptype;
3944 list_for_each_entry_rcu(ptype, offload_head, list) {
3945 if (ptype->type != type || !ptype->callbacks.gro_complete)
3952 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3956 if (netif_receive_skb(skb))
3964 case GRO_MERGED_FREE:
3965 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3966 kmem_cache_free(skbuff_head_cache, skb);
3979 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3981 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3983 EXPORT_SYMBOL(napi_gro_receive);
3985 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3987 __skb_pull(skb, skb_headlen(skb));
3988 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3989 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3991 skb->dev = napi->dev;
3997 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3999 struct sk_buff *skb = napi->skb;
4002 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4007 EXPORT_SYMBOL(napi_get_frags);
4009 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4014 if (netif_receive_skb(skb))
4019 case GRO_MERGED_FREE:
4020 napi_reuse_skb(napi, skb);
4031 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4033 struct sk_buff *skb = napi->skb;
4037 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4038 napi_reuse_skb(napi, skb);
4041 skb->protocol = eth_type_trans(skb, skb->dev);
4046 gro_result_t napi_gro_frags(struct napi_struct *napi)
4048 struct sk_buff *skb = napi_frags_skb(napi);
4053 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4055 EXPORT_SYMBOL(napi_gro_frags);
4058 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4059 * Note: called with local irq disabled, but exits with local irq enabled.
4061 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4064 struct softnet_data *remsd = sd->rps_ipi_list;
4067 sd->rps_ipi_list = NULL;
4071 /* Send pending IPI's to kick RPS processing on remote cpus. */
4073 struct softnet_data *next = remsd->rps_ipi_next;
4075 if (cpu_online(remsd->cpu))
4076 __smp_call_function_single(remsd->cpu,
4085 static int process_backlog(struct napi_struct *napi, int quota)
4088 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4091 /* Check if we have pending ipi, its better to send them now,
4092 * not waiting net_rx_action() end.
4094 if (sd->rps_ipi_list) {
4095 local_irq_disable();
4096 net_rps_action_and_irq_enable(sd);
4099 napi->weight = weight_p;
4100 local_irq_disable();
4101 while (work < quota) {
4102 struct sk_buff *skb;
4105 while ((skb = __skb_dequeue(&sd->process_queue))) {
4107 __netif_receive_skb(skb);
4108 local_irq_disable();
4109 input_queue_head_incr(sd);
4110 if (++work >= quota) {
4117 qlen = skb_queue_len(&sd->input_pkt_queue);
4119 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4120 &sd->process_queue);
4122 if (qlen < quota - work) {
4124 * Inline a custom version of __napi_complete().
4125 * only current cpu owns and manipulates this napi,
4126 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4127 * we can use a plain write instead of clear_bit(),
4128 * and we dont need an smp_mb() memory barrier.
4130 list_del(&napi->poll_list);
4133 quota = work + qlen;
4143 * __napi_schedule - schedule for receive
4144 * @n: entry to schedule
4146 * The entry's receive function will be scheduled to run
4148 void __napi_schedule(struct napi_struct *n)
4150 unsigned long flags;
4152 local_irq_save(flags);
4153 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4154 local_irq_restore(flags);
4156 EXPORT_SYMBOL(__napi_schedule);
4158 void __napi_complete(struct napi_struct *n)
4160 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4161 BUG_ON(n->gro_list);
4163 list_del(&n->poll_list);
4164 smp_mb__before_clear_bit();
4165 clear_bit(NAPI_STATE_SCHED, &n->state);
4167 EXPORT_SYMBOL(__napi_complete);
4169 void napi_complete(struct napi_struct *n)
4171 unsigned long flags;
4174 * don't let napi dequeue from the cpu poll list
4175 * just in case its running on a different cpu
4177 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4180 napi_gro_flush(n, false);
4181 local_irq_save(flags);
4183 local_irq_restore(flags);
4185 EXPORT_SYMBOL(napi_complete);
4187 /* must be called under rcu_read_lock(), as we dont take a reference */
4188 struct napi_struct *napi_by_id(unsigned int napi_id)
4190 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4191 struct napi_struct *napi;
4193 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4194 if (napi->napi_id == napi_id)
4199 EXPORT_SYMBOL_GPL(napi_by_id);
4201 void napi_hash_add(struct napi_struct *napi)
4203 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4205 spin_lock(&napi_hash_lock);
4207 /* 0 is not a valid id, we also skip an id that is taken
4208 * we expect both events to be extremely rare
4211 while (!napi->napi_id) {
4212 napi->napi_id = ++napi_gen_id;
4213 if (napi_by_id(napi->napi_id))
4217 hlist_add_head_rcu(&napi->napi_hash_node,
4218 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4220 spin_unlock(&napi_hash_lock);
4223 EXPORT_SYMBOL_GPL(napi_hash_add);
4225 /* Warning : caller is responsible to make sure rcu grace period
4226 * is respected before freeing memory containing @napi
4228 void napi_hash_del(struct napi_struct *napi)
4230 spin_lock(&napi_hash_lock);
4232 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4233 hlist_del_rcu(&napi->napi_hash_node);
4235 spin_unlock(&napi_hash_lock);
4237 EXPORT_SYMBOL_GPL(napi_hash_del);
4239 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4240 int (*poll)(struct napi_struct *, int), int weight)
4242 INIT_LIST_HEAD(&napi->poll_list);
4243 napi->gro_count = 0;
4244 napi->gro_list = NULL;
4247 if (weight > NAPI_POLL_WEIGHT)
4248 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4250 napi->weight = weight;
4251 list_add(&napi->dev_list, &dev->napi_list);
4253 #ifdef CONFIG_NETPOLL
4254 spin_lock_init(&napi->poll_lock);
4255 napi->poll_owner = -1;
4257 set_bit(NAPI_STATE_SCHED, &napi->state);
4259 EXPORT_SYMBOL(netif_napi_add);
4261 void netif_napi_del(struct napi_struct *napi)
4263 list_del_init(&napi->dev_list);
4264 napi_free_frags(napi);
4266 kfree_skb_list(napi->gro_list);
4267 napi->gro_list = NULL;
4268 napi->gro_count = 0;
4270 EXPORT_SYMBOL(netif_napi_del);
4272 static void net_rx_action(struct softirq_action *h)
4274 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4275 unsigned long time_limit = jiffies + 2;
4276 int budget = netdev_budget;
4279 local_irq_disable();
4281 while (!list_empty(&sd->poll_list)) {
4282 struct napi_struct *n;
4285 /* If softirq window is exhuasted then punt.
4286 * Allow this to run for 2 jiffies since which will allow
4287 * an average latency of 1.5/HZ.
4289 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4294 /* Even though interrupts have been re-enabled, this
4295 * access is safe because interrupts can only add new
4296 * entries to the tail of this list, and only ->poll()
4297 * calls can remove this head entry from the list.
4299 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4301 have = netpoll_poll_lock(n);
4305 /* This NAPI_STATE_SCHED test is for avoiding a race
4306 * with netpoll's poll_napi(). Only the entity which
4307 * obtains the lock and sees NAPI_STATE_SCHED set will
4308 * actually make the ->poll() call. Therefore we avoid
4309 * accidentally calling ->poll() when NAPI is not scheduled.
4312 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4313 work = n->poll(n, weight);
4317 WARN_ON_ONCE(work > weight);
4321 local_irq_disable();
4323 /* Drivers must not modify the NAPI state if they
4324 * consume the entire weight. In such cases this code
4325 * still "owns" the NAPI instance and therefore can
4326 * move the instance around on the list at-will.
4328 if (unlikely(work == weight)) {
4329 if (unlikely(napi_disable_pending(n))) {
4332 local_irq_disable();
4335 /* flush too old packets
4336 * If HZ < 1000, flush all packets.
4339 napi_gro_flush(n, HZ >= 1000);
4340 local_irq_disable();
4342 list_move_tail(&n->poll_list, &sd->poll_list);
4346 netpoll_poll_unlock(have);
4349 net_rps_action_and_irq_enable(sd);
4351 #ifdef CONFIG_NET_DMA
4353 * There may not be any more sk_buffs coming right now, so push
4354 * any pending DMA copies to hardware
4356 dma_issue_pending_all();
4363 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4367 struct netdev_adjacent {
4368 struct net_device *dev;
4370 /* upper master flag, there can only be one master device per list */
4373 /* counter for the number of times this device was added to us */
4376 /* private field for the users */
4379 struct list_head list;
4380 struct rcu_head rcu;
4383 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4384 struct net_device *adj_dev,
4385 struct list_head *adj_list)
4387 struct netdev_adjacent *adj;
4389 list_for_each_entry(adj, adj_list, list) {
4390 if (adj->dev == adj_dev)
4397 * netdev_has_upper_dev - Check if device is linked to an upper device
4399 * @upper_dev: upper device to check
4401 * Find out if a device is linked to specified upper device and return true
4402 * in case it is. Note that this checks only immediate upper device,
4403 * not through a complete stack of devices. The caller must hold the RTNL lock.
4405 bool netdev_has_upper_dev(struct net_device *dev,
4406 struct net_device *upper_dev)
4410 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4412 EXPORT_SYMBOL(netdev_has_upper_dev);
4415 * netdev_has_any_upper_dev - Check if device is linked to some device
4418 * Find out if a device is linked to an upper device and return true in case
4419 * it is. The caller must hold the RTNL lock.
4421 static bool netdev_has_any_upper_dev(struct net_device *dev)
4425 return !list_empty(&dev->all_adj_list.upper);
4429 * netdev_master_upper_dev_get - Get master upper device
4432 * Find a master upper device and return pointer to it or NULL in case
4433 * it's not there. The caller must hold the RTNL lock.
4435 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4437 struct netdev_adjacent *upper;
4441 if (list_empty(&dev->adj_list.upper))
4444 upper = list_first_entry(&dev->adj_list.upper,
4445 struct netdev_adjacent, list);
4446 if (likely(upper->master))
4450 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4452 void *netdev_adjacent_get_private(struct list_head *adj_list)
4454 struct netdev_adjacent *adj;
4456 adj = list_entry(adj_list, struct netdev_adjacent, list);
4458 return adj->private;
4460 EXPORT_SYMBOL(netdev_adjacent_get_private);
4463 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4465 * @iter: list_head ** of the current position
4467 * Gets the next device from the dev's upper list, starting from iter
4468 * position. The caller must hold RCU read lock.
4470 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4471 struct list_head **iter)
4473 struct netdev_adjacent *upper;
4475 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4477 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4479 if (&upper->list == &dev->all_adj_list.upper)
4482 *iter = &upper->list;
4486 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4489 * netdev_lower_get_next_private - Get the next ->private from the
4490 * lower neighbour list
4492 * @iter: list_head ** of the current position
4494 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4495 * list, starting from iter position. The caller must hold either hold the
4496 * RTNL lock or its own locking that guarantees that the neighbour lower
4497 * list will remain unchainged.
4499 void *netdev_lower_get_next_private(struct net_device *dev,
4500 struct list_head **iter)
4502 struct netdev_adjacent *lower;
4504 lower = list_entry(*iter, struct netdev_adjacent, list);
4506 if (&lower->list == &dev->adj_list.lower)
4510 *iter = lower->list.next;
4512 return lower->private;
4514 EXPORT_SYMBOL(netdev_lower_get_next_private);
4517 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4518 * lower neighbour list, RCU
4521 * @iter: list_head ** of the current position
4523 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4524 * list, starting from iter position. The caller must hold RCU read lock.
4526 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4527 struct list_head **iter)
4529 struct netdev_adjacent *lower;
4531 WARN_ON_ONCE(!rcu_read_lock_held());
4533 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4535 if (&lower->list == &dev->adj_list.lower)
4539 *iter = &lower->list;
4541 return lower->private;
4543 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4546 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4547 * lower neighbour list, RCU
4551 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4552 * list. The caller must hold RCU read lock.
4554 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4556 struct netdev_adjacent *lower;
4558 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4559 struct netdev_adjacent, list);
4561 return lower->private;
4564 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4567 * netdev_master_upper_dev_get_rcu - Get master upper device
4570 * Find a master upper device and return pointer to it or NULL in case
4571 * it's not there. The caller must hold the RCU read lock.
4573 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4575 struct netdev_adjacent *upper;
4577 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4578 struct netdev_adjacent, list);
4579 if (upper && likely(upper->master))
4583 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4585 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4586 struct net_device *adj_dev,
4587 struct list_head *dev_list,
4588 void *private, bool master)
4590 struct netdev_adjacent *adj;
4591 char linkname[IFNAMSIZ+7];
4594 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4601 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4606 adj->master = master;
4608 adj->private = private;
4611 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4612 adj_dev->name, dev->name, adj_dev->name);
4614 if (dev_list == &dev->adj_list.lower) {
4615 sprintf(linkname, "lower_%s", adj_dev->name);
4616 ret = sysfs_create_link(&(dev->dev.kobj),
4617 &(adj_dev->dev.kobj), linkname);
4620 } else if (dev_list == &dev->adj_list.upper) {
4621 sprintf(linkname, "upper_%s", adj_dev->name);
4622 ret = sysfs_create_link(&(dev->dev.kobj),
4623 &(adj_dev->dev.kobj), linkname);
4628 /* Ensure that master link is always the first item in list. */
4630 ret = sysfs_create_link(&(dev->dev.kobj),
4631 &(adj_dev->dev.kobj), "master");
4633 goto remove_symlinks;
4635 list_add_rcu(&adj->list, dev_list);
4637 list_add_tail_rcu(&adj->list, dev_list);
4643 if (dev_list == &dev->adj_list.lower) {
4644 sprintf(linkname, "lower_%s", adj_dev->name);
4645 sysfs_remove_link(&(dev->dev.kobj), linkname);
4646 } else if (dev_list == &dev->adj_list.upper) {
4647 sprintf(linkname, "upper_%s", adj_dev->name);
4648 sysfs_remove_link(&(dev->dev.kobj), linkname);
4658 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4659 struct net_device *adj_dev,
4660 struct list_head *dev_list)
4662 struct netdev_adjacent *adj;
4663 char linkname[IFNAMSIZ+7];
4665 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4668 pr_err("tried to remove device %s from %s\n",
4669 dev->name, adj_dev->name);
4673 if (adj->ref_nr > 1) {
4674 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4681 sysfs_remove_link(&(dev->dev.kobj), "master");
4683 if (dev_list == &dev->adj_list.lower) {
4684 sprintf(linkname, "lower_%s", adj_dev->name);
4685 sysfs_remove_link(&(dev->dev.kobj), linkname);
4686 } else if (dev_list == &dev->adj_list.upper) {
4687 sprintf(linkname, "upper_%s", adj_dev->name);
4688 sysfs_remove_link(&(dev->dev.kobj), linkname);
4691 list_del_rcu(&adj->list);
4692 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4693 adj_dev->name, dev->name, adj_dev->name);
4695 kfree_rcu(adj, rcu);
4698 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4699 struct net_device *upper_dev,
4700 struct list_head *up_list,
4701 struct list_head *down_list,
4702 void *private, bool master)
4706 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4711 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4714 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4721 static int __netdev_adjacent_dev_link(struct net_device *dev,
4722 struct net_device *upper_dev)
4724 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4725 &dev->all_adj_list.upper,
4726 &upper_dev->all_adj_list.lower,
4730 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4731 struct net_device *upper_dev,
4732 struct list_head *up_list,
4733 struct list_head *down_list)
4735 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4736 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4739 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4740 struct net_device *upper_dev)
4742 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4743 &dev->all_adj_list.upper,
4744 &upper_dev->all_adj_list.lower);
4747 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4748 struct net_device *upper_dev,
4749 void *private, bool master)
4751 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4756 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4757 &dev->adj_list.upper,
4758 &upper_dev->adj_list.lower,
4761 __netdev_adjacent_dev_unlink(dev, upper_dev);
4768 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4769 struct net_device *upper_dev)
4771 __netdev_adjacent_dev_unlink(dev, upper_dev);
4772 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4773 &dev->adj_list.upper,
4774 &upper_dev->adj_list.lower);
4777 static int __netdev_upper_dev_link(struct net_device *dev,
4778 struct net_device *upper_dev, bool master,
4781 struct netdev_adjacent *i, *j, *to_i, *to_j;
4786 if (dev == upper_dev)
4789 /* To prevent loops, check if dev is not upper device to upper_dev. */
4790 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4793 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4796 if (master && netdev_master_upper_dev_get(dev))
4799 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4804 /* Now that we linked these devs, make all the upper_dev's
4805 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4806 * versa, and don't forget the devices itself. All of these
4807 * links are non-neighbours.
4809 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4810 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4811 pr_debug("Interlinking %s with %s, non-neighbour\n",
4812 i->dev->name, j->dev->name);
4813 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4819 /* add dev to every upper_dev's upper device */
4820 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4821 pr_debug("linking %s's upper device %s with %s\n",
4822 upper_dev->name, i->dev->name, dev->name);
4823 ret = __netdev_adjacent_dev_link(dev, i->dev);
4825 goto rollback_upper_mesh;
4828 /* add upper_dev to every dev's lower device */
4829 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4830 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4831 i->dev->name, upper_dev->name);
4832 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4834 goto rollback_lower_mesh;
4837 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4840 rollback_lower_mesh:
4842 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4845 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4850 rollback_upper_mesh:
4852 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4855 __netdev_adjacent_dev_unlink(dev, i->dev);
4863 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4864 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4865 if (i == to_i && j == to_j)
4867 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4873 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4879 * netdev_upper_dev_link - Add a link to the upper device
4881 * @upper_dev: new upper device
4883 * Adds a link to device which is upper to this one. The caller must hold
4884 * the RTNL lock. On a failure a negative errno code is returned.
4885 * On success the reference counts are adjusted and the function
4888 int netdev_upper_dev_link(struct net_device *dev,
4889 struct net_device *upper_dev)
4891 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4893 EXPORT_SYMBOL(netdev_upper_dev_link);
4896 * netdev_master_upper_dev_link - Add a master link to the upper device
4898 * @upper_dev: new upper device
4900 * Adds a link to device which is upper to this one. In this case, only
4901 * one master upper device can be linked, although other non-master devices
4902 * might be linked as well. The caller must hold the RTNL lock.
4903 * On a failure a negative errno code is returned. On success the reference
4904 * counts are adjusted and the function returns zero.
4906 int netdev_master_upper_dev_link(struct net_device *dev,
4907 struct net_device *upper_dev)
4909 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4911 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4913 int netdev_master_upper_dev_link_private(struct net_device *dev,
4914 struct net_device *upper_dev,
4917 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4919 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4922 * netdev_upper_dev_unlink - Removes a link to upper device
4924 * @upper_dev: new upper device
4926 * Removes a link to device which is upper to this one. The caller must hold
4929 void netdev_upper_dev_unlink(struct net_device *dev,
4930 struct net_device *upper_dev)
4932 struct netdev_adjacent *i, *j;
4935 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4937 /* Here is the tricky part. We must remove all dev's lower
4938 * devices from all upper_dev's upper devices and vice
4939 * versa, to maintain the graph relationship.
4941 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4942 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4943 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4945 /* remove also the devices itself from lower/upper device
4948 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4949 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4951 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4952 __netdev_adjacent_dev_unlink(dev, i->dev);
4954 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4956 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4958 void *netdev_lower_dev_get_private(struct net_device *dev,
4959 struct net_device *lower_dev)
4961 struct netdev_adjacent *lower;
4965 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
4969 return lower->private;
4971 EXPORT_SYMBOL(netdev_lower_dev_get_private);
4973 static void dev_change_rx_flags(struct net_device *dev, int flags)
4975 const struct net_device_ops *ops = dev->netdev_ops;
4977 if (ops->ndo_change_rx_flags)
4978 ops->ndo_change_rx_flags(dev, flags);
4981 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
4983 unsigned int old_flags = dev->flags;
4989 dev->flags |= IFF_PROMISC;
4990 dev->promiscuity += inc;
4991 if (dev->promiscuity == 0) {
4994 * If inc causes overflow, untouch promisc and return error.
4997 dev->flags &= ~IFF_PROMISC;
4999 dev->promiscuity -= inc;
5000 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5005 if (dev->flags != old_flags) {
5006 pr_info("device %s %s promiscuous mode\n",
5008 dev->flags & IFF_PROMISC ? "entered" : "left");
5009 if (audit_enabled) {
5010 current_uid_gid(&uid, &gid);
5011 audit_log(current->audit_context, GFP_ATOMIC,
5012 AUDIT_ANOM_PROMISCUOUS,
5013 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5014 dev->name, (dev->flags & IFF_PROMISC),
5015 (old_flags & IFF_PROMISC),
5016 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5017 from_kuid(&init_user_ns, uid),
5018 from_kgid(&init_user_ns, gid),
5019 audit_get_sessionid(current));
5022 dev_change_rx_flags(dev, IFF_PROMISC);
5025 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5030 * dev_set_promiscuity - update promiscuity count on a device
5034 * Add or remove promiscuity from a device. While the count in the device
5035 * remains above zero the interface remains promiscuous. Once it hits zero
5036 * the device reverts back to normal filtering operation. A negative inc
5037 * value is used to drop promiscuity on the device.
5038 * Return 0 if successful or a negative errno code on error.
5040 int dev_set_promiscuity(struct net_device *dev, int inc)
5042 unsigned int old_flags = dev->flags;
5045 err = __dev_set_promiscuity(dev, inc, true);
5048 if (dev->flags != old_flags)
5049 dev_set_rx_mode(dev);
5052 EXPORT_SYMBOL(dev_set_promiscuity);
5054 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5056 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5060 dev->flags |= IFF_ALLMULTI;
5061 dev->allmulti += inc;
5062 if (dev->allmulti == 0) {
5065 * If inc causes overflow, untouch allmulti and return error.
5068 dev->flags &= ~IFF_ALLMULTI;
5070 dev->allmulti -= inc;
5071 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5076 if (dev->flags ^ old_flags) {
5077 dev_change_rx_flags(dev, IFF_ALLMULTI);
5078 dev_set_rx_mode(dev);
5080 __dev_notify_flags(dev, old_flags,
5081 dev->gflags ^ old_gflags);
5087 * dev_set_allmulti - update allmulti count on a device
5091 * Add or remove reception of all multicast frames to a device. While the
5092 * count in the device remains above zero the interface remains listening
5093 * to all interfaces. Once it hits zero the device reverts back to normal
5094 * filtering operation. A negative @inc value is used to drop the counter
5095 * when releasing a resource needing all multicasts.
5096 * Return 0 if successful or a negative errno code on error.
5099 int dev_set_allmulti(struct net_device *dev, int inc)
5101 return __dev_set_allmulti(dev, inc, true);
5103 EXPORT_SYMBOL(dev_set_allmulti);
5106 * Upload unicast and multicast address lists to device and
5107 * configure RX filtering. When the device doesn't support unicast
5108 * filtering it is put in promiscuous mode while unicast addresses
5111 void __dev_set_rx_mode(struct net_device *dev)
5113 const struct net_device_ops *ops = dev->netdev_ops;
5115 /* dev_open will call this function so the list will stay sane. */
5116 if (!(dev->flags&IFF_UP))
5119 if (!netif_device_present(dev))
5122 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5123 /* Unicast addresses changes may only happen under the rtnl,
5124 * therefore calling __dev_set_promiscuity here is safe.
5126 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5127 __dev_set_promiscuity(dev, 1, false);
5128 dev->uc_promisc = true;
5129 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5130 __dev_set_promiscuity(dev, -1, false);
5131 dev->uc_promisc = false;
5135 if (ops->ndo_set_rx_mode)
5136 ops->ndo_set_rx_mode(dev);
5139 void dev_set_rx_mode(struct net_device *dev)
5141 netif_addr_lock_bh(dev);
5142 __dev_set_rx_mode(dev);
5143 netif_addr_unlock_bh(dev);
5147 * dev_get_flags - get flags reported to userspace
5150 * Get the combination of flag bits exported through APIs to userspace.
5152 unsigned int dev_get_flags(const struct net_device *dev)
5156 flags = (dev->flags & ~(IFF_PROMISC |
5161 (dev->gflags & (IFF_PROMISC |
5164 if (netif_running(dev)) {
5165 if (netif_oper_up(dev))
5166 flags |= IFF_RUNNING;
5167 if (netif_carrier_ok(dev))
5168 flags |= IFF_LOWER_UP;
5169 if (netif_dormant(dev))
5170 flags |= IFF_DORMANT;
5175 EXPORT_SYMBOL(dev_get_flags);
5177 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5179 unsigned int old_flags = dev->flags;
5185 * Set the flags on our device.
5188 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5189 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5191 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5195 * Load in the correct multicast list now the flags have changed.
5198 if ((old_flags ^ flags) & IFF_MULTICAST)
5199 dev_change_rx_flags(dev, IFF_MULTICAST);
5201 dev_set_rx_mode(dev);
5204 * Have we downed the interface. We handle IFF_UP ourselves
5205 * according to user attempts to set it, rather than blindly
5210 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5211 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5214 dev_set_rx_mode(dev);
5217 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5218 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5219 unsigned int old_flags = dev->flags;
5221 dev->gflags ^= IFF_PROMISC;
5223 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5224 if (dev->flags != old_flags)
5225 dev_set_rx_mode(dev);
5228 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5229 is important. Some (broken) drivers set IFF_PROMISC, when
5230 IFF_ALLMULTI is requested not asking us and not reporting.
5232 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5233 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5235 dev->gflags ^= IFF_ALLMULTI;
5236 __dev_set_allmulti(dev, inc, false);
5242 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5243 unsigned int gchanges)
5245 unsigned int changes = dev->flags ^ old_flags;
5248 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5250 if (changes & IFF_UP) {
5251 if (dev->flags & IFF_UP)
5252 call_netdevice_notifiers(NETDEV_UP, dev);
5254 call_netdevice_notifiers(NETDEV_DOWN, dev);
5257 if (dev->flags & IFF_UP &&
5258 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5259 struct netdev_notifier_change_info change_info;
5261 change_info.flags_changed = changes;
5262 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5268 * dev_change_flags - change device settings
5270 * @flags: device state flags
5272 * Change settings on device based state flags. The flags are
5273 * in the userspace exported format.
5275 int dev_change_flags(struct net_device *dev, unsigned int flags)
5278 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5280 ret = __dev_change_flags(dev, flags);
5284 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5285 __dev_notify_flags(dev, old_flags, changes);
5288 EXPORT_SYMBOL(dev_change_flags);
5291 * dev_set_mtu - Change maximum transfer unit
5293 * @new_mtu: new transfer unit
5295 * Change the maximum transfer size of the network device.
5297 int dev_set_mtu(struct net_device *dev, int new_mtu)
5299 const struct net_device_ops *ops = dev->netdev_ops;
5302 if (new_mtu == dev->mtu)
5305 /* MTU must be positive. */
5309 if (!netif_device_present(dev))
5313 if (ops->ndo_change_mtu)
5314 err = ops->ndo_change_mtu(dev, new_mtu);
5319 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5322 EXPORT_SYMBOL(dev_set_mtu);
5325 * dev_set_group - Change group this device belongs to
5327 * @new_group: group this device should belong to
5329 void dev_set_group(struct net_device *dev, int new_group)
5331 dev->group = new_group;
5333 EXPORT_SYMBOL(dev_set_group);
5336 * dev_set_mac_address - Change Media Access Control Address
5340 * Change the hardware (MAC) address of the device
5342 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5344 const struct net_device_ops *ops = dev->netdev_ops;
5347 if (!ops->ndo_set_mac_address)
5349 if (sa->sa_family != dev->type)
5351 if (!netif_device_present(dev))
5353 err = ops->ndo_set_mac_address(dev, sa);
5356 dev->addr_assign_type = NET_ADDR_SET;
5357 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5358 add_device_randomness(dev->dev_addr, dev->addr_len);
5361 EXPORT_SYMBOL(dev_set_mac_address);
5364 * dev_change_carrier - Change device carrier
5366 * @new_carrier: new value
5368 * Change device carrier
5370 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5372 const struct net_device_ops *ops = dev->netdev_ops;
5374 if (!ops->ndo_change_carrier)
5376 if (!netif_device_present(dev))
5378 return ops->ndo_change_carrier(dev, new_carrier);
5380 EXPORT_SYMBOL(dev_change_carrier);
5383 * dev_get_phys_port_id - Get device physical port ID
5387 * Get device physical port ID
5389 int dev_get_phys_port_id(struct net_device *dev,
5390 struct netdev_phys_port_id *ppid)
5392 const struct net_device_ops *ops = dev->netdev_ops;
5394 if (!ops->ndo_get_phys_port_id)
5396 return ops->ndo_get_phys_port_id(dev, ppid);
5398 EXPORT_SYMBOL(dev_get_phys_port_id);
5401 * dev_new_index - allocate an ifindex
5402 * @net: the applicable net namespace
5404 * Returns a suitable unique value for a new device interface
5405 * number. The caller must hold the rtnl semaphore or the
5406 * dev_base_lock to be sure it remains unique.
5408 static int dev_new_index(struct net *net)
5410 int ifindex = net->ifindex;
5414 if (!__dev_get_by_index(net, ifindex))
5415 return net->ifindex = ifindex;
5419 /* Delayed registration/unregisteration */
5420 static LIST_HEAD(net_todo_list);
5421 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5423 static void net_set_todo(struct net_device *dev)
5425 list_add_tail(&dev->todo_list, &net_todo_list);
5426 dev_net(dev)->dev_unreg_count++;
5429 static void rollback_registered_many(struct list_head *head)
5431 struct net_device *dev, *tmp;
5432 LIST_HEAD(close_head);
5434 BUG_ON(dev_boot_phase);
5437 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5438 /* Some devices call without registering
5439 * for initialization unwind. Remove those
5440 * devices and proceed with the remaining.
5442 if (dev->reg_state == NETREG_UNINITIALIZED) {
5443 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5447 list_del(&dev->unreg_list);
5450 dev->dismantle = true;
5451 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5454 /* If device is running, close it first. */
5455 list_for_each_entry(dev, head, unreg_list)
5456 list_add_tail(&dev->close_list, &close_head);
5457 dev_close_many(&close_head);
5459 list_for_each_entry(dev, head, unreg_list) {
5460 /* And unlink it from device chain. */
5461 unlist_netdevice(dev);
5463 dev->reg_state = NETREG_UNREGISTERING;
5468 list_for_each_entry(dev, head, unreg_list) {
5469 /* Shutdown queueing discipline. */
5473 /* Notify protocols, that we are about to destroy
5474 this device. They should clean all the things.
5476 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5478 if (!dev->rtnl_link_ops ||
5479 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5480 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5483 * Flush the unicast and multicast chains
5488 if (dev->netdev_ops->ndo_uninit)
5489 dev->netdev_ops->ndo_uninit(dev);
5491 /* Notifier chain MUST detach us all upper devices. */
5492 WARN_ON(netdev_has_any_upper_dev(dev));
5494 /* Remove entries from kobject tree */
5495 netdev_unregister_kobject(dev);
5497 /* Remove XPS queueing entries */
5498 netif_reset_xps_queues_gt(dev, 0);
5504 list_for_each_entry(dev, head, unreg_list)
5508 static void rollback_registered(struct net_device *dev)
5512 list_add(&dev->unreg_list, &single);
5513 rollback_registered_many(&single);
5517 static netdev_features_t netdev_fix_features(struct net_device *dev,
5518 netdev_features_t features)
5520 /* Fix illegal checksum combinations */
5521 if ((features & NETIF_F_HW_CSUM) &&
5522 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5523 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5524 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5527 /* TSO requires that SG is present as well. */
5528 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5529 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5530 features &= ~NETIF_F_ALL_TSO;
5533 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5534 !(features & NETIF_F_IP_CSUM)) {
5535 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5536 features &= ~NETIF_F_TSO;
5537 features &= ~NETIF_F_TSO_ECN;
5540 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5541 !(features & NETIF_F_IPV6_CSUM)) {
5542 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5543 features &= ~NETIF_F_TSO6;
5546 /* TSO ECN requires that TSO is present as well. */
5547 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5548 features &= ~NETIF_F_TSO_ECN;
5550 /* Software GSO depends on SG. */
5551 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5552 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5553 features &= ~NETIF_F_GSO;
5556 /* UFO needs SG and checksumming */
5557 if (features & NETIF_F_UFO) {
5558 /* maybe split UFO into V4 and V6? */
5559 if (!((features & NETIF_F_GEN_CSUM) ||
5560 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5561 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5563 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5564 features &= ~NETIF_F_UFO;
5567 if (!(features & NETIF_F_SG)) {
5569 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5570 features &= ~NETIF_F_UFO;
5577 int __netdev_update_features(struct net_device *dev)
5579 netdev_features_t features;
5584 features = netdev_get_wanted_features(dev);
5586 if (dev->netdev_ops->ndo_fix_features)
5587 features = dev->netdev_ops->ndo_fix_features(dev, features);
5589 /* driver might be less strict about feature dependencies */
5590 features = netdev_fix_features(dev, features);
5592 if (dev->features == features)
5595 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5596 &dev->features, &features);
5598 if (dev->netdev_ops->ndo_set_features)
5599 err = dev->netdev_ops->ndo_set_features(dev, features);
5601 if (unlikely(err < 0)) {
5603 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5604 err, &features, &dev->features);
5609 dev->features = features;
5615 * netdev_update_features - recalculate device features
5616 * @dev: the device to check
5618 * Recalculate dev->features set and send notifications if it
5619 * has changed. Should be called after driver or hardware dependent
5620 * conditions might have changed that influence the features.
5622 void netdev_update_features(struct net_device *dev)
5624 if (__netdev_update_features(dev))
5625 netdev_features_change(dev);
5627 EXPORT_SYMBOL(netdev_update_features);
5630 * netdev_change_features - recalculate device features
5631 * @dev: the device to check
5633 * Recalculate dev->features set and send notifications even
5634 * if they have not changed. Should be called instead of
5635 * netdev_update_features() if also dev->vlan_features might
5636 * have changed to allow the changes to be propagated to stacked
5639 void netdev_change_features(struct net_device *dev)
5641 __netdev_update_features(dev);
5642 netdev_features_change(dev);
5644 EXPORT_SYMBOL(netdev_change_features);
5647 * netif_stacked_transfer_operstate - transfer operstate
5648 * @rootdev: the root or lower level device to transfer state from
5649 * @dev: the device to transfer operstate to
5651 * Transfer operational state from root to device. This is normally
5652 * called when a stacking relationship exists between the root
5653 * device and the device(a leaf device).
5655 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5656 struct net_device *dev)
5658 if (rootdev->operstate == IF_OPER_DORMANT)
5659 netif_dormant_on(dev);
5661 netif_dormant_off(dev);
5663 if (netif_carrier_ok(rootdev)) {
5664 if (!netif_carrier_ok(dev))
5665 netif_carrier_on(dev);
5667 if (netif_carrier_ok(dev))
5668 netif_carrier_off(dev);
5671 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5674 static int netif_alloc_rx_queues(struct net_device *dev)
5676 unsigned int i, count = dev->num_rx_queues;
5677 struct netdev_rx_queue *rx;
5681 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5687 for (i = 0; i < count; i++)
5693 static void netdev_init_one_queue(struct net_device *dev,
5694 struct netdev_queue *queue, void *_unused)
5696 /* Initialize queue lock */
5697 spin_lock_init(&queue->_xmit_lock);
5698 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5699 queue->xmit_lock_owner = -1;
5700 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5703 dql_init(&queue->dql, HZ);
5707 static void netif_free_tx_queues(struct net_device *dev)
5709 if (is_vmalloc_addr(dev->_tx))
5715 static int netif_alloc_netdev_queues(struct net_device *dev)
5717 unsigned int count = dev->num_tx_queues;
5718 struct netdev_queue *tx;
5719 size_t sz = count * sizeof(*tx);
5721 BUG_ON(count < 1 || count > 0xffff);
5723 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5731 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5732 spin_lock_init(&dev->tx_global_lock);
5738 * register_netdevice - register a network device
5739 * @dev: device to register
5741 * Take a completed network device structure and add it to the kernel
5742 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5743 * chain. 0 is returned on success. A negative errno code is returned
5744 * on a failure to set up the device, or if the name is a duplicate.
5746 * Callers must hold the rtnl semaphore. You may want
5747 * register_netdev() instead of this.
5750 * The locking appears insufficient to guarantee two parallel registers
5751 * will not get the same name.
5754 int register_netdevice(struct net_device *dev)
5757 struct net *net = dev_net(dev);
5759 BUG_ON(dev_boot_phase);
5764 /* When net_device's are persistent, this will be fatal. */
5765 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5768 spin_lock_init(&dev->addr_list_lock);
5769 netdev_set_addr_lockdep_class(dev);
5773 ret = dev_get_valid_name(net, dev, dev->name);
5777 /* Init, if this function is available */
5778 if (dev->netdev_ops->ndo_init) {
5779 ret = dev->netdev_ops->ndo_init(dev);
5787 if (((dev->hw_features | dev->features) &
5788 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5789 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5790 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5791 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5798 dev->ifindex = dev_new_index(net);
5799 else if (__dev_get_by_index(net, dev->ifindex))
5802 if (dev->iflink == -1)
5803 dev->iflink = dev->ifindex;
5805 /* Transfer changeable features to wanted_features and enable
5806 * software offloads (GSO and GRO).
5808 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5809 dev->features |= NETIF_F_SOFT_FEATURES;
5810 dev->wanted_features = dev->features & dev->hw_features;
5812 if (!(dev->flags & IFF_LOOPBACK)) {
5813 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5816 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5818 dev->vlan_features |= NETIF_F_HIGHDMA;
5820 /* Make NETIF_F_SG inheritable to tunnel devices.
5822 dev->hw_enc_features |= NETIF_F_SG;
5824 /* Make NETIF_F_SG inheritable to MPLS.
5826 dev->mpls_features |= NETIF_F_SG;
5828 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5829 ret = notifier_to_errno(ret);
5833 ret = netdev_register_kobject(dev);
5836 dev->reg_state = NETREG_REGISTERED;
5838 __netdev_update_features(dev);
5841 * Default initial state at registry is that the
5842 * device is present.
5845 set_bit(__LINK_STATE_PRESENT, &dev->state);
5847 linkwatch_init_dev(dev);
5849 dev_init_scheduler(dev);
5851 list_netdevice(dev);
5852 add_device_randomness(dev->dev_addr, dev->addr_len);
5854 /* If the device has permanent device address, driver should
5855 * set dev_addr and also addr_assign_type should be set to
5856 * NET_ADDR_PERM (default value).
5858 if (dev->addr_assign_type == NET_ADDR_PERM)
5859 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5861 /* Notify protocols, that a new device appeared. */
5862 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5863 ret = notifier_to_errno(ret);
5865 rollback_registered(dev);
5866 dev->reg_state = NETREG_UNREGISTERED;
5869 * Prevent userspace races by waiting until the network
5870 * device is fully setup before sending notifications.
5872 if (!dev->rtnl_link_ops ||
5873 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5874 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5880 if (dev->netdev_ops->ndo_uninit)
5881 dev->netdev_ops->ndo_uninit(dev);
5884 EXPORT_SYMBOL(register_netdevice);
5887 * init_dummy_netdev - init a dummy network device for NAPI
5888 * @dev: device to init
5890 * This takes a network device structure and initialize the minimum
5891 * amount of fields so it can be used to schedule NAPI polls without
5892 * registering a full blown interface. This is to be used by drivers
5893 * that need to tie several hardware interfaces to a single NAPI
5894 * poll scheduler due to HW limitations.
5896 int init_dummy_netdev(struct net_device *dev)
5898 /* Clear everything. Note we don't initialize spinlocks
5899 * are they aren't supposed to be taken by any of the
5900 * NAPI code and this dummy netdev is supposed to be
5901 * only ever used for NAPI polls
5903 memset(dev, 0, sizeof(struct net_device));
5905 /* make sure we BUG if trying to hit standard
5906 * register/unregister code path
5908 dev->reg_state = NETREG_DUMMY;
5910 /* NAPI wants this */
5911 INIT_LIST_HEAD(&dev->napi_list);
5913 /* a dummy interface is started by default */
5914 set_bit(__LINK_STATE_PRESENT, &dev->state);
5915 set_bit(__LINK_STATE_START, &dev->state);
5917 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5918 * because users of this 'device' dont need to change
5924 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5928 * register_netdev - register a network device
5929 * @dev: device to register
5931 * Take a completed network device structure and add it to the kernel
5932 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5933 * chain. 0 is returned on success. A negative errno code is returned
5934 * on a failure to set up the device, or if the name is a duplicate.
5936 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5937 * and expands the device name if you passed a format string to
5940 int register_netdev(struct net_device *dev)
5945 err = register_netdevice(dev);
5949 EXPORT_SYMBOL(register_netdev);
5951 int netdev_refcnt_read(const struct net_device *dev)
5955 for_each_possible_cpu(i)
5956 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5959 EXPORT_SYMBOL(netdev_refcnt_read);
5962 * netdev_wait_allrefs - wait until all references are gone.
5963 * @dev: target net_device
5965 * This is called when unregistering network devices.
5967 * Any protocol or device that holds a reference should register
5968 * for netdevice notification, and cleanup and put back the
5969 * reference if they receive an UNREGISTER event.
5970 * We can get stuck here if buggy protocols don't correctly
5973 static void netdev_wait_allrefs(struct net_device *dev)
5975 unsigned long rebroadcast_time, warning_time;
5978 linkwatch_forget_dev(dev);
5980 rebroadcast_time = warning_time = jiffies;
5981 refcnt = netdev_refcnt_read(dev);
5983 while (refcnt != 0) {
5984 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5987 /* Rebroadcast unregister notification */
5988 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5994 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5995 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5997 /* We must not have linkwatch events
5998 * pending on unregister. If this
5999 * happens, we simply run the queue
6000 * unscheduled, resulting in a noop
6003 linkwatch_run_queue();
6008 rebroadcast_time = jiffies;
6013 refcnt = netdev_refcnt_read(dev);
6015 if (time_after(jiffies, warning_time + 10 * HZ)) {
6016 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6018 warning_time = jiffies;
6027 * register_netdevice(x1);
6028 * register_netdevice(x2);
6030 * unregister_netdevice(y1);
6031 * unregister_netdevice(y2);
6037 * We are invoked by rtnl_unlock().
6038 * This allows us to deal with problems:
6039 * 1) We can delete sysfs objects which invoke hotplug
6040 * without deadlocking with linkwatch via keventd.
6041 * 2) Since we run with the RTNL semaphore not held, we can sleep
6042 * safely in order to wait for the netdev refcnt to drop to zero.
6044 * We must not return until all unregister events added during
6045 * the interval the lock was held have been completed.
6047 void netdev_run_todo(void)
6049 struct list_head list;
6051 /* Snapshot list, allow later requests */
6052 list_replace_init(&net_todo_list, &list);
6057 /* Wait for rcu callbacks to finish before next phase */
6058 if (!list_empty(&list))
6061 while (!list_empty(&list)) {
6062 struct net_device *dev
6063 = list_first_entry(&list, struct net_device, todo_list);
6064 list_del(&dev->todo_list);
6067 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6070 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6071 pr_err("network todo '%s' but state %d\n",
6072 dev->name, dev->reg_state);
6077 dev->reg_state = NETREG_UNREGISTERED;
6079 on_each_cpu(flush_backlog, dev, 1);
6081 netdev_wait_allrefs(dev);
6084 BUG_ON(netdev_refcnt_read(dev));
6085 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6086 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6087 WARN_ON(dev->dn_ptr);
6089 if (dev->destructor)
6090 dev->destructor(dev);
6092 /* Report a network device has been unregistered */
6094 dev_net(dev)->dev_unreg_count--;
6096 wake_up(&netdev_unregistering_wq);
6098 /* Free network device */
6099 kobject_put(&dev->dev.kobj);
6103 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6104 * fields in the same order, with only the type differing.
6106 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6107 const struct net_device_stats *netdev_stats)
6109 #if BITS_PER_LONG == 64
6110 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6111 memcpy(stats64, netdev_stats, sizeof(*stats64));
6113 size_t i, n = sizeof(*stats64) / sizeof(u64);
6114 const unsigned long *src = (const unsigned long *)netdev_stats;
6115 u64 *dst = (u64 *)stats64;
6117 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6118 sizeof(*stats64) / sizeof(u64));
6119 for (i = 0; i < n; i++)
6123 EXPORT_SYMBOL(netdev_stats_to_stats64);
6126 * dev_get_stats - get network device statistics
6127 * @dev: device to get statistics from
6128 * @storage: place to store stats
6130 * Get network statistics from device. Return @storage.
6131 * The device driver may provide its own method by setting
6132 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6133 * otherwise the internal statistics structure is used.
6135 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6136 struct rtnl_link_stats64 *storage)
6138 const struct net_device_ops *ops = dev->netdev_ops;
6140 if (ops->ndo_get_stats64) {
6141 memset(storage, 0, sizeof(*storage));
6142 ops->ndo_get_stats64(dev, storage);
6143 } else if (ops->ndo_get_stats) {
6144 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6146 netdev_stats_to_stats64(storage, &dev->stats);
6148 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6151 EXPORT_SYMBOL(dev_get_stats);
6153 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6155 struct netdev_queue *queue = dev_ingress_queue(dev);
6157 #ifdef CONFIG_NET_CLS_ACT
6160 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6163 netdev_init_one_queue(dev, queue, NULL);
6164 queue->qdisc = &noop_qdisc;
6165 queue->qdisc_sleeping = &noop_qdisc;
6166 rcu_assign_pointer(dev->ingress_queue, queue);
6171 static const struct ethtool_ops default_ethtool_ops;
6173 void netdev_set_default_ethtool_ops(struct net_device *dev,
6174 const struct ethtool_ops *ops)
6176 if (dev->ethtool_ops == &default_ethtool_ops)
6177 dev->ethtool_ops = ops;
6179 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6181 void netdev_freemem(struct net_device *dev)
6183 char *addr = (char *)dev - dev->padded;
6185 if (is_vmalloc_addr(addr))
6192 * alloc_netdev_mqs - allocate network device
6193 * @sizeof_priv: size of private data to allocate space for
6194 * @name: device name format string
6195 * @setup: callback to initialize device
6196 * @txqs: the number of TX subqueues to allocate
6197 * @rxqs: the number of RX subqueues to allocate
6199 * Allocates a struct net_device with private data area for driver use
6200 * and performs basic initialization. Also allocates subquue structs
6201 * for each queue on the device.
6203 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6204 void (*setup)(struct net_device *),
6205 unsigned int txqs, unsigned int rxqs)
6207 struct net_device *dev;
6209 struct net_device *p;
6211 BUG_ON(strlen(name) >= sizeof(dev->name));
6214 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6220 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6225 alloc_size = sizeof(struct net_device);
6227 /* ensure 32-byte alignment of private area */
6228 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6229 alloc_size += sizeof_priv;
6231 /* ensure 32-byte alignment of whole construct */
6232 alloc_size += NETDEV_ALIGN - 1;
6234 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6236 p = vzalloc(alloc_size);
6240 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6241 dev->padded = (char *)dev - (char *)p;
6243 dev->pcpu_refcnt = alloc_percpu(int);
6244 if (!dev->pcpu_refcnt)
6247 if (dev_addr_init(dev))
6253 dev_net_set(dev, &init_net);
6255 dev->gso_max_size = GSO_MAX_SIZE;
6256 dev->gso_max_segs = GSO_MAX_SEGS;
6258 INIT_LIST_HEAD(&dev->napi_list);
6259 INIT_LIST_HEAD(&dev->unreg_list);
6260 INIT_LIST_HEAD(&dev->close_list);
6261 INIT_LIST_HEAD(&dev->link_watch_list);
6262 INIT_LIST_HEAD(&dev->adj_list.upper);
6263 INIT_LIST_HEAD(&dev->adj_list.lower);
6264 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6265 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6266 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6269 dev->num_tx_queues = txqs;
6270 dev->real_num_tx_queues = txqs;
6271 if (netif_alloc_netdev_queues(dev))
6275 dev->num_rx_queues = rxqs;
6276 dev->real_num_rx_queues = rxqs;
6277 if (netif_alloc_rx_queues(dev))
6281 strcpy(dev->name, name);
6282 dev->group = INIT_NETDEV_GROUP;
6283 if (!dev->ethtool_ops)
6284 dev->ethtool_ops = &default_ethtool_ops;
6292 free_percpu(dev->pcpu_refcnt);
6293 netif_free_tx_queues(dev);
6299 netdev_freemem(dev);
6302 EXPORT_SYMBOL(alloc_netdev_mqs);
6305 * free_netdev - free network device
6308 * This function does the last stage of destroying an allocated device
6309 * interface. The reference to the device object is released.
6310 * If this is the last reference then it will be freed.
6312 void free_netdev(struct net_device *dev)
6314 struct napi_struct *p, *n;
6316 release_net(dev_net(dev));
6318 netif_free_tx_queues(dev);
6323 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6325 /* Flush device addresses */
6326 dev_addr_flush(dev);
6328 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6331 free_percpu(dev->pcpu_refcnt);
6332 dev->pcpu_refcnt = NULL;
6334 /* Compatibility with error handling in drivers */
6335 if (dev->reg_state == NETREG_UNINITIALIZED) {
6336 netdev_freemem(dev);
6340 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6341 dev->reg_state = NETREG_RELEASED;
6343 /* will free via device release */
6344 put_device(&dev->dev);
6346 EXPORT_SYMBOL(free_netdev);
6349 * synchronize_net - Synchronize with packet receive processing
6351 * Wait for packets currently being received to be done.
6352 * Does not block later packets from starting.
6354 void synchronize_net(void)
6357 if (rtnl_is_locked())
6358 synchronize_rcu_expedited();
6362 EXPORT_SYMBOL(synchronize_net);
6365 * unregister_netdevice_queue - remove device from the kernel
6369 * This function shuts down a device interface and removes it
6370 * from the kernel tables.
6371 * If head not NULL, device is queued to be unregistered later.
6373 * Callers must hold the rtnl semaphore. You may want
6374 * unregister_netdev() instead of this.
6377 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6382 list_move_tail(&dev->unreg_list, head);
6384 rollback_registered(dev);
6385 /* Finish processing unregister after unlock */
6389 EXPORT_SYMBOL(unregister_netdevice_queue);
6392 * unregister_netdevice_many - unregister many devices
6393 * @head: list of devices
6395 void unregister_netdevice_many(struct list_head *head)
6397 struct net_device *dev;
6399 if (!list_empty(head)) {
6400 rollback_registered_many(head);
6401 list_for_each_entry(dev, head, unreg_list)
6405 EXPORT_SYMBOL(unregister_netdevice_many);
6408 * unregister_netdev - remove device from the kernel
6411 * This function shuts down a device interface and removes it
6412 * from the kernel tables.
6414 * This is just a wrapper for unregister_netdevice that takes
6415 * the rtnl semaphore. In general you want to use this and not
6416 * unregister_netdevice.
6418 void unregister_netdev(struct net_device *dev)
6421 unregister_netdevice(dev);
6424 EXPORT_SYMBOL(unregister_netdev);
6427 * dev_change_net_namespace - move device to different nethost namespace
6429 * @net: network namespace
6430 * @pat: If not NULL name pattern to try if the current device name
6431 * is already taken in the destination network namespace.
6433 * This function shuts down a device interface and moves it
6434 * to a new network namespace. On success 0 is returned, on
6435 * a failure a netagive errno code is returned.
6437 * Callers must hold the rtnl semaphore.
6440 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6446 /* Don't allow namespace local devices to be moved. */
6448 if (dev->features & NETIF_F_NETNS_LOCAL)
6451 /* Ensure the device has been registrered */
6452 if (dev->reg_state != NETREG_REGISTERED)
6455 /* Get out if there is nothing todo */
6457 if (net_eq(dev_net(dev), net))
6460 /* Pick the destination device name, and ensure
6461 * we can use it in the destination network namespace.
6464 if (__dev_get_by_name(net, dev->name)) {
6465 /* We get here if we can't use the current device name */
6468 if (dev_get_valid_name(net, dev, pat) < 0)
6473 * And now a mini version of register_netdevice unregister_netdevice.
6476 /* If device is running close it first. */
6479 /* And unlink it from device chain */
6481 unlist_netdevice(dev);
6485 /* Shutdown queueing discipline. */
6488 /* Notify protocols, that we are about to destroy
6489 this device. They should clean all the things.
6491 Note that dev->reg_state stays at NETREG_REGISTERED.
6492 This is wanted because this way 8021q and macvlan know
6493 the device is just moving and can keep their slaves up.
6495 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6497 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6498 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6501 * Flush the unicast and multicast chains
6506 /* Send a netdev-removed uevent to the old namespace */
6507 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6509 /* Actually switch the network namespace */
6510 dev_net_set(dev, net);
6512 /* If there is an ifindex conflict assign a new one */
6513 if (__dev_get_by_index(net, dev->ifindex)) {
6514 int iflink = (dev->iflink == dev->ifindex);
6515 dev->ifindex = dev_new_index(net);
6517 dev->iflink = dev->ifindex;
6520 /* Send a netdev-add uevent to the new namespace */
6521 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6523 /* Fixup kobjects */
6524 err = device_rename(&dev->dev, dev->name);
6527 /* Add the device back in the hashes */
6528 list_netdevice(dev);
6530 /* Notify protocols, that a new device appeared. */
6531 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6534 * Prevent userspace races by waiting until the network
6535 * device is fully setup before sending notifications.
6537 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6544 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6546 static int dev_cpu_callback(struct notifier_block *nfb,
6547 unsigned long action,
6550 struct sk_buff **list_skb;
6551 struct sk_buff *skb;
6552 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6553 struct softnet_data *sd, *oldsd;
6555 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6558 local_irq_disable();
6559 cpu = smp_processor_id();
6560 sd = &per_cpu(softnet_data, cpu);
6561 oldsd = &per_cpu(softnet_data, oldcpu);
6563 /* Find end of our completion_queue. */
6564 list_skb = &sd->completion_queue;
6566 list_skb = &(*list_skb)->next;
6567 /* Append completion queue from offline CPU. */
6568 *list_skb = oldsd->completion_queue;
6569 oldsd->completion_queue = NULL;
6571 /* Append output queue from offline CPU. */
6572 if (oldsd->output_queue) {
6573 *sd->output_queue_tailp = oldsd->output_queue;
6574 sd->output_queue_tailp = oldsd->output_queue_tailp;
6575 oldsd->output_queue = NULL;
6576 oldsd->output_queue_tailp = &oldsd->output_queue;
6578 /* Append NAPI poll list from offline CPU. */
6579 if (!list_empty(&oldsd->poll_list)) {
6580 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6581 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6584 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6587 /* Process offline CPU's input_pkt_queue */
6588 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6590 input_queue_head_incr(oldsd);
6592 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6594 input_queue_head_incr(oldsd);
6602 * netdev_increment_features - increment feature set by one
6603 * @all: current feature set
6604 * @one: new feature set
6605 * @mask: mask feature set
6607 * Computes a new feature set after adding a device with feature set
6608 * @one to the master device with current feature set @all. Will not
6609 * enable anything that is off in @mask. Returns the new feature set.
6611 netdev_features_t netdev_increment_features(netdev_features_t all,
6612 netdev_features_t one, netdev_features_t mask)
6614 if (mask & NETIF_F_GEN_CSUM)
6615 mask |= NETIF_F_ALL_CSUM;
6616 mask |= NETIF_F_VLAN_CHALLENGED;
6618 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6619 all &= one | ~NETIF_F_ALL_FOR_ALL;
6621 /* If one device supports hw checksumming, set for all. */
6622 if (all & NETIF_F_GEN_CSUM)
6623 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6627 EXPORT_SYMBOL(netdev_increment_features);
6629 static struct hlist_head * __net_init netdev_create_hash(void)
6632 struct hlist_head *hash;
6634 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6636 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6637 INIT_HLIST_HEAD(&hash[i]);
6642 /* Initialize per network namespace state */
6643 static int __net_init netdev_init(struct net *net)
6645 if (net != &init_net)
6646 INIT_LIST_HEAD(&net->dev_base_head);
6648 net->dev_name_head = netdev_create_hash();
6649 if (net->dev_name_head == NULL)
6652 net->dev_index_head = netdev_create_hash();
6653 if (net->dev_index_head == NULL)
6659 kfree(net->dev_name_head);
6665 * netdev_drivername - network driver for the device
6666 * @dev: network device
6668 * Determine network driver for device.
6670 const char *netdev_drivername(const struct net_device *dev)
6672 const struct device_driver *driver;
6673 const struct device *parent;
6674 const char *empty = "";
6676 parent = dev->dev.parent;
6680 driver = parent->driver;
6681 if (driver && driver->name)
6682 return driver->name;
6686 static int __netdev_printk(const char *level, const struct net_device *dev,
6687 struct va_format *vaf)
6691 if (dev && dev->dev.parent) {
6692 r = dev_printk_emit(level[1] - '0',
6695 dev_driver_string(dev->dev.parent),
6696 dev_name(dev->dev.parent),
6697 netdev_name(dev), vaf);
6699 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6701 r = printk("%s(NULL net_device): %pV", level, vaf);
6707 int netdev_printk(const char *level, const struct net_device *dev,
6708 const char *format, ...)
6710 struct va_format vaf;
6714 va_start(args, format);
6719 r = __netdev_printk(level, dev, &vaf);
6725 EXPORT_SYMBOL(netdev_printk);
6727 #define define_netdev_printk_level(func, level) \
6728 int func(const struct net_device *dev, const char *fmt, ...) \
6731 struct va_format vaf; \
6734 va_start(args, fmt); \
6739 r = __netdev_printk(level, dev, &vaf); \
6745 EXPORT_SYMBOL(func);
6747 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6748 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6749 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6750 define_netdev_printk_level(netdev_err, KERN_ERR);
6751 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6752 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6753 define_netdev_printk_level(netdev_info, KERN_INFO);
6755 static void __net_exit netdev_exit(struct net *net)
6757 kfree(net->dev_name_head);
6758 kfree(net->dev_index_head);
6761 static struct pernet_operations __net_initdata netdev_net_ops = {
6762 .init = netdev_init,
6763 .exit = netdev_exit,
6766 static void __net_exit default_device_exit(struct net *net)
6768 struct net_device *dev, *aux;
6770 * Push all migratable network devices back to the
6771 * initial network namespace
6774 for_each_netdev_safe(net, dev, aux) {
6776 char fb_name[IFNAMSIZ];
6778 /* Ignore unmoveable devices (i.e. loopback) */
6779 if (dev->features & NETIF_F_NETNS_LOCAL)
6782 /* Leave virtual devices for the generic cleanup */
6783 if (dev->rtnl_link_ops)
6786 /* Push remaining network devices to init_net */
6787 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6788 err = dev_change_net_namespace(dev, &init_net, fb_name);
6790 pr_emerg("%s: failed to move %s to init_net: %d\n",
6791 __func__, dev->name, err);
6798 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6800 /* Return with the rtnl_lock held when there are no network
6801 * devices unregistering in any network namespace in net_list.
6808 prepare_to_wait(&netdev_unregistering_wq, &wait,
6809 TASK_UNINTERRUPTIBLE);
6810 unregistering = false;
6812 list_for_each_entry(net, net_list, exit_list) {
6813 if (net->dev_unreg_count > 0) {
6814 unregistering = true;
6823 finish_wait(&netdev_unregistering_wq, &wait);
6826 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6828 /* At exit all network devices most be removed from a network
6829 * namespace. Do this in the reverse order of registration.
6830 * Do this across as many network namespaces as possible to
6831 * improve batching efficiency.
6833 struct net_device *dev;
6835 LIST_HEAD(dev_kill_list);
6837 /* To prevent network device cleanup code from dereferencing
6838 * loopback devices or network devices that have been freed
6839 * wait here for all pending unregistrations to complete,
6840 * before unregistring the loopback device and allowing the
6841 * network namespace be freed.
6843 * The netdev todo list containing all network devices
6844 * unregistrations that happen in default_device_exit_batch
6845 * will run in the rtnl_unlock() at the end of
6846 * default_device_exit_batch.
6848 rtnl_lock_unregistering(net_list);
6849 list_for_each_entry(net, net_list, exit_list) {
6850 for_each_netdev_reverse(net, dev) {
6851 if (dev->rtnl_link_ops)
6852 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6854 unregister_netdevice_queue(dev, &dev_kill_list);
6857 unregister_netdevice_many(&dev_kill_list);
6858 list_del(&dev_kill_list);
6862 static struct pernet_operations __net_initdata default_device_ops = {
6863 .exit = default_device_exit,
6864 .exit_batch = default_device_exit_batch,
6868 * Initialize the DEV module. At boot time this walks the device list and
6869 * unhooks any devices that fail to initialise (normally hardware not
6870 * present) and leaves us with a valid list of present and active devices.
6875 * This is called single threaded during boot, so no need
6876 * to take the rtnl semaphore.
6878 static int __init net_dev_init(void)
6880 int i, rc = -ENOMEM;
6882 BUG_ON(!dev_boot_phase);
6884 if (dev_proc_init())
6887 if (netdev_kobject_init())
6890 INIT_LIST_HEAD(&ptype_all);
6891 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6892 INIT_LIST_HEAD(&ptype_base[i]);
6894 INIT_LIST_HEAD(&offload_base);
6896 if (register_pernet_subsys(&netdev_net_ops))
6900 * Initialise the packet receive queues.
6903 for_each_possible_cpu(i) {
6904 struct softnet_data *sd = &per_cpu(softnet_data, i);
6906 memset(sd, 0, sizeof(*sd));
6907 skb_queue_head_init(&sd->input_pkt_queue);
6908 skb_queue_head_init(&sd->process_queue);
6909 sd->completion_queue = NULL;
6910 INIT_LIST_HEAD(&sd->poll_list);
6911 sd->output_queue = NULL;
6912 sd->output_queue_tailp = &sd->output_queue;
6914 sd->csd.func = rps_trigger_softirq;
6920 sd->backlog.poll = process_backlog;
6921 sd->backlog.weight = weight_p;
6922 sd->backlog.gro_list = NULL;
6923 sd->backlog.gro_count = 0;
6925 #ifdef CONFIG_NET_FLOW_LIMIT
6926 sd->flow_limit = NULL;
6932 /* The loopback device is special if any other network devices
6933 * is present in a network namespace the loopback device must
6934 * be present. Since we now dynamically allocate and free the
6935 * loopback device ensure this invariant is maintained by
6936 * keeping the loopback device as the first device on the
6937 * list of network devices. Ensuring the loopback devices
6938 * is the first device that appears and the last network device
6941 if (register_pernet_device(&loopback_net_ops))
6944 if (register_pernet_device(&default_device_ops))
6947 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6948 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6950 hotcpu_notifier(dev_cpu_callback, 0);
6957 subsys_initcall(net_dev_init);