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 <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 static DEFINE_SPINLOCK(ptype_lock);
148 static DEFINE_SPINLOCK(offload_lock);
149 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
150 struct list_head ptype_all __read_mostly; /* Taps */
151 static struct list_head offload_base __read_mostly;
153 static int netif_rx_internal(struct sk_buff *skb);
154 static int call_netdevice_notifiers_info(unsigned long val,
155 struct net_device *dev,
156 struct netdev_notifier_info *info);
159 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base_head list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 DEFINE_RWLOCK(dev_base_lock);
178 EXPORT_SYMBOL(dev_base_lock);
180 /* protects napi_hash addition/deletion and napi_gen_id */
181 static DEFINE_SPINLOCK(napi_hash_lock);
183 static unsigned int napi_gen_id;
184 static DEFINE_HASHTABLE(napi_hash, 8);
186 static seqcount_t devnet_rename_seq;
188 static inline void dev_base_seq_inc(struct net *net)
190 while (++net->dev_base_seq == 0);
193 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
195 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
197 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
200 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
202 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
205 static inline void rps_lock(struct softnet_data *sd)
208 spin_lock(&sd->input_pkt_queue.lock);
212 static inline void rps_unlock(struct softnet_data *sd)
215 spin_unlock(&sd->input_pkt_queue.lock);
219 /* Device list insertion */
220 static void list_netdevice(struct net_device *dev)
222 struct net *net = dev_net(dev);
226 write_lock_bh(&dev_base_lock);
227 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
228 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
229 hlist_add_head_rcu(&dev->index_hlist,
230 dev_index_hash(net, dev->ifindex));
231 write_unlock_bh(&dev_base_lock);
233 dev_base_seq_inc(net);
236 /* Device list removal
237 * caller must respect a RCU grace period before freeing/reusing dev
239 static void unlist_netdevice(struct net_device *dev)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del_rcu(&dev->dev_list);
246 hlist_del_rcu(&dev->name_hlist);
247 hlist_del_rcu(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(dev_net(dev));
257 static RAW_NOTIFIER_HEAD(netdev_chain);
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
264 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
265 EXPORT_PER_CPU_SYMBOL(softnet_data);
267 #ifdef CONFIG_LOCKDEP
269 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
270 * according to dev->type
272 static const unsigned short netdev_lock_type[] =
273 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
274 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
275 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
276 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
277 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
278 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
279 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
280 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
281 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
282 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
283 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
284 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
285 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
286 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
287 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
289 static const char *const netdev_lock_name[] =
290 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
291 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
292 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
293 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
294 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
295 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
296 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
297 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
298 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
299 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
300 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
301 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
302 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
303 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
304 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
306 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
313 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
314 if (netdev_lock_type[i] == dev_type)
316 /* the last key is used by default */
317 return ARRAY_SIZE(netdev_lock_type) - 1;
320 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
321 unsigned short dev_type)
325 i = netdev_lock_pos(dev_type);
326 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
327 netdev_lock_name[i]);
330 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
334 i = netdev_lock_pos(dev->type);
335 lockdep_set_class_and_name(&dev->addr_list_lock,
336 &netdev_addr_lock_key[i],
337 netdev_lock_name[i]);
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
349 /*******************************************************************************
351 Protocol management and registration routines
353 *******************************************************************************/
356 * Add a protocol ID to the list. Now that the input handler is
357 * smarter we can dispense with all the messy stuff that used to be
360 * BEWARE!!! Protocol handlers, mangling input packets,
361 * MUST BE last in hash buckets and checking protocol handlers
362 * MUST start from promiscuous ptype_all chain in net_bh.
363 * It is true now, do not change it.
364 * Explanation follows: if protocol handler, mangling packet, will
365 * be the first on list, it is not able to sense, that packet
366 * is cloned and should be copied-on-write, so that it will
367 * change it and subsequent readers will get broken packet.
371 static inline struct list_head *ptype_head(const struct packet_type *pt)
373 if (pt->type == htons(ETH_P_ALL))
374 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
376 return pt->dev ? &pt->dev->ptype_specific :
377 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
381 * dev_add_pack - add packet handler
382 * @pt: packet type declaration
384 * Add a protocol handler to the networking stack. The passed &packet_type
385 * is linked into kernel lists and may not be freed until it has been
386 * removed from the kernel lists.
388 * This call does not sleep therefore it can not
389 * guarantee all CPU's that are in middle of receiving packets
390 * will see the new packet type (until the next received packet).
393 void dev_add_pack(struct packet_type *pt)
395 struct list_head *head = ptype_head(pt);
397 spin_lock(&ptype_lock);
398 list_add_rcu(&pt->list, head);
399 spin_unlock(&ptype_lock);
401 EXPORT_SYMBOL(dev_add_pack);
404 * __dev_remove_pack - remove packet handler
405 * @pt: packet type declaration
407 * Remove a protocol handler that was previously added to the kernel
408 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
409 * from the kernel lists and can be freed or reused once this function
412 * The packet type might still be in use by receivers
413 * and must not be freed until after all the CPU's have gone
414 * through a quiescent state.
416 void __dev_remove_pack(struct packet_type *pt)
418 struct list_head *head = ptype_head(pt);
419 struct packet_type *pt1;
421 spin_lock(&ptype_lock);
423 list_for_each_entry(pt1, head, list) {
425 list_del_rcu(&pt->list);
430 pr_warn("dev_remove_pack: %p not found\n", pt);
432 spin_unlock(&ptype_lock);
434 EXPORT_SYMBOL(__dev_remove_pack);
437 * dev_remove_pack - remove packet handler
438 * @pt: packet type declaration
440 * Remove a protocol handler that was previously added to the kernel
441 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
442 * from the kernel lists and can be freed or reused once this function
445 * This call sleeps to guarantee that no CPU is looking at the packet
448 void dev_remove_pack(struct packet_type *pt)
450 __dev_remove_pack(pt);
454 EXPORT_SYMBOL(dev_remove_pack);
458 * dev_add_offload - register offload handlers
459 * @po: protocol offload declaration
461 * Add protocol offload handlers to the networking stack. The passed
462 * &proto_offload is linked into kernel lists and may not be freed until
463 * it has been removed from the kernel lists.
465 * This call does not sleep therefore it can not
466 * guarantee all CPU's that are in middle of receiving packets
467 * will see the new offload handlers (until the next received packet).
469 void dev_add_offload(struct packet_offload *po)
471 struct list_head *head = &offload_base;
473 spin_lock(&offload_lock);
474 list_add_rcu(&po->list, head);
475 spin_unlock(&offload_lock);
477 EXPORT_SYMBOL(dev_add_offload);
480 * __dev_remove_offload - remove offload handler
481 * @po: packet offload declaration
483 * Remove a protocol offload handler that was previously added to the
484 * kernel offload handlers by dev_add_offload(). The passed &offload_type
485 * is removed from the kernel lists and can be freed or reused once this
488 * The packet type might still be in use by receivers
489 * and must not be freed until after all the CPU's have gone
490 * through a quiescent state.
492 static void __dev_remove_offload(struct packet_offload *po)
494 struct list_head *head = &offload_base;
495 struct packet_offload *po1;
497 spin_lock(&offload_lock);
499 list_for_each_entry(po1, head, list) {
501 list_del_rcu(&po->list);
506 pr_warn("dev_remove_offload: %p not found\n", po);
508 spin_unlock(&offload_lock);
512 * dev_remove_offload - remove packet offload handler
513 * @po: packet offload declaration
515 * Remove a packet offload handler that was previously added to the kernel
516 * offload handlers by dev_add_offload(). The passed &offload_type is
517 * removed from the kernel lists and can be freed or reused once this
520 * This call sleeps to guarantee that no CPU is looking at the packet
523 void dev_remove_offload(struct packet_offload *po)
525 __dev_remove_offload(po);
529 EXPORT_SYMBOL(dev_remove_offload);
531 /******************************************************************************
533 Device Boot-time Settings Routines
535 *******************************************************************************/
537 /* Boot time configuration table */
538 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
541 * netdev_boot_setup_add - add new setup entry
542 * @name: name of the device
543 * @map: configured settings for the device
545 * Adds new setup entry to the dev_boot_setup list. The function
546 * returns 0 on error and 1 on success. This is a generic routine to
549 static int netdev_boot_setup_add(char *name, struct ifmap *map)
551 struct netdev_boot_setup *s;
555 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
556 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
557 memset(s[i].name, 0, sizeof(s[i].name));
558 strlcpy(s[i].name, name, IFNAMSIZ);
559 memcpy(&s[i].map, map, sizeof(s[i].map));
564 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
568 * netdev_boot_setup_check - check boot time settings
569 * @dev: the netdevice
571 * Check boot time settings for the device.
572 * The found settings are set for the device to be used
573 * later in the device probing.
574 * Returns 0 if no settings found, 1 if they are.
576 int netdev_boot_setup_check(struct net_device *dev)
578 struct netdev_boot_setup *s = dev_boot_setup;
581 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
582 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
583 !strcmp(dev->name, s[i].name)) {
584 dev->irq = s[i].map.irq;
585 dev->base_addr = s[i].map.base_addr;
586 dev->mem_start = s[i].map.mem_start;
587 dev->mem_end = s[i].map.mem_end;
593 EXPORT_SYMBOL(netdev_boot_setup_check);
597 * netdev_boot_base - get address from boot time settings
598 * @prefix: prefix for network device
599 * @unit: id for network device
601 * Check boot time settings for the base address of device.
602 * The found settings are set for the device to be used
603 * later in the device probing.
604 * Returns 0 if no settings found.
606 unsigned long netdev_boot_base(const char *prefix, int unit)
608 const struct netdev_boot_setup *s = dev_boot_setup;
612 sprintf(name, "%s%d", prefix, unit);
615 * If device already registered then return base of 1
616 * to indicate not to probe for this interface
618 if (__dev_get_by_name(&init_net, name))
621 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
622 if (!strcmp(name, s[i].name))
623 return s[i].map.base_addr;
628 * Saves at boot time configured settings for any netdevice.
630 int __init netdev_boot_setup(char *str)
635 str = get_options(str, ARRAY_SIZE(ints), ints);
640 memset(&map, 0, sizeof(map));
644 map.base_addr = ints[2];
646 map.mem_start = ints[3];
648 map.mem_end = ints[4];
650 /* Add new entry to the list */
651 return netdev_boot_setup_add(str, &map);
654 __setup("netdev=", netdev_boot_setup);
656 /*******************************************************************************
658 Device Interface Subroutines
660 *******************************************************************************/
663 * __dev_get_by_name - find a device by its name
664 * @net: the applicable net namespace
665 * @name: name to find
667 * Find an interface by name. Must be called under RTNL semaphore
668 * or @dev_base_lock. If the name is found a pointer to the device
669 * is returned. If the name is not found then %NULL is returned. The
670 * reference counters are not incremented so the caller must be
671 * careful with locks.
674 struct net_device *__dev_get_by_name(struct net *net, const char *name)
676 struct net_device *dev;
677 struct hlist_head *head = dev_name_hash(net, name);
679 hlist_for_each_entry(dev, head, name_hlist)
680 if (!strncmp(dev->name, name, IFNAMSIZ))
685 EXPORT_SYMBOL(__dev_get_by_name);
688 * dev_get_by_name_rcu - find a device by its name
689 * @net: the applicable net namespace
690 * @name: name to find
692 * Find an interface by name.
693 * If the name is found a pointer to the device is returned.
694 * If the name is not found then %NULL is returned.
695 * The reference counters are not incremented so the caller must be
696 * careful with locks. The caller must hold RCU lock.
699 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
701 struct net_device *dev;
702 struct hlist_head *head = dev_name_hash(net, name);
704 hlist_for_each_entry_rcu(dev, head, name_hlist)
705 if (!strncmp(dev->name, name, IFNAMSIZ))
710 EXPORT_SYMBOL(dev_get_by_name_rcu);
713 * dev_get_by_name - find a device by its name
714 * @net: the applicable net namespace
715 * @name: name to find
717 * Find an interface by name. This can be called from any
718 * context and does its own locking. The returned handle has
719 * the usage count incremented and the caller must use dev_put() to
720 * release it when it is no longer needed. %NULL is returned if no
721 * matching device is found.
724 struct net_device *dev_get_by_name(struct net *net, const char *name)
726 struct net_device *dev;
729 dev = dev_get_by_name_rcu(net, name);
735 EXPORT_SYMBOL(dev_get_by_name);
738 * __dev_get_by_index - find a device by its ifindex
739 * @net: the applicable net namespace
740 * @ifindex: index of device
742 * Search for an interface by index. Returns %NULL if the device
743 * is not found or a pointer to the device. The device has not
744 * had its reference counter increased so the caller must be careful
745 * about locking. The caller must hold either the RTNL semaphore
749 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
751 struct net_device *dev;
752 struct hlist_head *head = dev_index_hash(net, ifindex);
754 hlist_for_each_entry(dev, head, index_hlist)
755 if (dev->ifindex == ifindex)
760 EXPORT_SYMBOL(__dev_get_by_index);
763 * dev_get_by_index_rcu - find a device by its ifindex
764 * @net: the applicable net namespace
765 * @ifindex: index of device
767 * Search for an interface by index. Returns %NULL if the device
768 * is not found or a pointer to the device. The device has not
769 * had its reference counter increased so the caller must be careful
770 * about locking. The caller must hold RCU lock.
773 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
775 struct net_device *dev;
776 struct hlist_head *head = dev_index_hash(net, ifindex);
778 hlist_for_each_entry_rcu(dev, head, index_hlist)
779 if (dev->ifindex == ifindex)
784 EXPORT_SYMBOL(dev_get_by_index_rcu);
788 * dev_get_by_index - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns NULL if the device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_index(struct net *net, int ifindex)
800 struct net_device *dev;
803 dev = dev_get_by_index_rcu(net, ifindex);
809 EXPORT_SYMBOL(dev_get_by_index);
812 * netdev_get_name - get a netdevice name, knowing its ifindex.
813 * @net: network namespace
814 * @name: a pointer to the buffer where the name will be stored.
815 * @ifindex: the ifindex of the interface to get the name from.
817 * The use of raw_seqcount_begin() and cond_resched() before
818 * retrying is required as we want to give the writers a chance
819 * to complete when CONFIG_PREEMPT is not set.
821 int netdev_get_name(struct net *net, char *name, int ifindex)
823 struct net_device *dev;
827 seq = raw_seqcount_begin(&devnet_rename_seq);
829 dev = dev_get_by_index_rcu(net, ifindex);
835 strcpy(name, dev->name);
837 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
846 * dev_getbyhwaddr_rcu - find a device by its hardware address
847 * @net: the applicable net namespace
848 * @type: media type of device
849 * @ha: hardware address
851 * Search for an interface by MAC address. Returns NULL if the device
852 * is not found or a pointer to the device.
853 * The caller must hold RCU or RTNL.
854 * The returned device has not had its ref count increased
855 * and the caller must therefore be careful about locking
859 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
862 struct net_device *dev;
864 for_each_netdev_rcu(net, dev)
865 if (dev->type == type &&
866 !memcmp(dev->dev_addr, ha, dev->addr_len))
871 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
873 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
875 struct net_device *dev;
878 for_each_netdev(net, dev)
879 if (dev->type == type)
884 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
886 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
888 struct net_device *dev, *ret = NULL;
891 for_each_netdev_rcu(net, dev)
892 if (dev->type == type) {
900 EXPORT_SYMBOL(dev_getfirstbyhwtype);
903 * __dev_get_by_flags - find any device with given flags
904 * @net: the applicable net namespace
905 * @if_flags: IFF_* values
906 * @mask: bitmask of bits in if_flags to check
908 * Search for any interface with the given flags. Returns NULL if a device
909 * is not found or a pointer to the device. Must be called inside
910 * rtnl_lock(), and result refcount is unchanged.
913 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
916 struct net_device *dev, *ret;
921 for_each_netdev(net, dev) {
922 if (((dev->flags ^ if_flags) & mask) == 0) {
929 EXPORT_SYMBOL(__dev_get_by_flags);
932 * dev_valid_name - check if name is okay for network device
935 * Network device names need to be valid file names to
936 * to allow sysfs to work. We also disallow any kind of
939 bool dev_valid_name(const char *name)
943 if (strlen(name) >= IFNAMSIZ)
945 if (!strcmp(name, ".") || !strcmp(name, ".."))
949 if (*name == '/' || isspace(*name))
955 EXPORT_SYMBOL(dev_valid_name);
958 * __dev_alloc_name - allocate a name for a device
959 * @net: network namespace to allocate the device name in
960 * @name: name format string
961 * @buf: scratch buffer and result name string
963 * Passed a format string - eg "lt%d" it will try and find a suitable
964 * id. It scans list of devices to build up a free map, then chooses
965 * the first empty slot. The caller must hold the dev_base or rtnl lock
966 * while allocating the name and adding the device in order to avoid
968 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
969 * Returns the number of the unit assigned or a negative errno code.
972 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
976 const int max_netdevices = 8*PAGE_SIZE;
977 unsigned long *inuse;
978 struct net_device *d;
980 p = strnchr(name, IFNAMSIZ-1, '%');
983 * Verify the string as this thing may have come from
984 * the user. There must be either one "%d" and no other "%"
987 if (p[1] != 'd' || strchr(p + 2, '%'))
990 /* Use one page as a bit array of possible slots */
991 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
995 for_each_netdev(net, d) {
996 if (!sscanf(d->name, name, &i))
998 if (i < 0 || i >= max_netdevices)
1001 /* avoid cases where sscanf is not exact inverse of printf */
1002 snprintf(buf, IFNAMSIZ, name, i);
1003 if (!strncmp(buf, d->name, IFNAMSIZ))
1007 i = find_first_zero_bit(inuse, max_netdevices);
1008 free_page((unsigned long) inuse);
1012 snprintf(buf, IFNAMSIZ, name, i);
1013 if (!__dev_get_by_name(net, buf))
1016 /* It is possible to run out of possible slots
1017 * when the name is long and there isn't enough space left
1018 * for the digits, or if all bits are used.
1024 * dev_alloc_name - allocate a name for a device
1026 * @name: name format string
1028 * Passed a format string - eg "lt%d" it will try and find a suitable
1029 * id. It scans list of devices to build up a free map, then chooses
1030 * the first empty slot. The caller must hold the dev_base or rtnl lock
1031 * while allocating the name and adding the device in order to avoid
1033 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1034 * Returns the number of the unit assigned or a negative errno code.
1037 int dev_alloc_name(struct net_device *dev, const char *name)
1043 BUG_ON(!dev_net(dev));
1045 ret = __dev_alloc_name(net, name, buf);
1047 strlcpy(dev->name, buf, IFNAMSIZ);
1050 EXPORT_SYMBOL(dev_alloc_name);
1052 static int dev_alloc_name_ns(struct net *net,
1053 struct net_device *dev,
1059 ret = __dev_alloc_name(net, name, buf);
1061 strlcpy(dev->name, buf, IFNAMSIZ);
1065 static int dev_get_valid_name(struct net *net,
1066 struct net_device *dev,
1071 if (!dev_valid_name(name))
1074 if (strchr(name, '%'))
1075 return dev_alloc_name_ns(net, dev, name);
1076 else if (__dev_get_by_name(net, name))
1078 else if (dev->name != name)
1079 strlcpy(dev->name, name, IFNAMSIZ);
1085 * dev_change_name - change name of a device
1087 * @newname: name (or format string) must be at least IFNAMSIZ
1089 * Change name of a device, can pass format strings "eth%d".
1092 int dev_change_name(struct net_device *dev, const char *newname)
1094 unsigned char old_assign_type;
1095 char oldname[IFNAMSIZ];
1101 BUG_ON(!dev_net(dev));
1104 if (dev->flags & IFF_UP)
1107 write_seqcount_begin(&devnet_rename_seq);
1109 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1110 write_seqcount_end(&devnet_rename_seq);
1114 memcpy(oldname, dev->name, IFNAMSIZ);
1116 err = dev_get_valid_name(net, dev, newname);
1118 write_seqcount_end(&devnet_rename_seq);
1122 if (oldname[0] && !strchr(oldname, '%'))
1123 netdev_info(dev, "renamed from %s\n", oldname);
1125 old_assign_type = dev->name_assign_type;
1126 dev->name_assign_type = NET_NAME_RENAMED;
1129 ret = device_rename(&dev->dev, dev->name);
1131 memcpy(dev->name, oldname, IFNAMSIZ);
1132 dev->name_assign_type = old_assign_type;
1133 write_seqcount_end(&devnet_rename_seq);
1137 write_seqcount_end(&devnet_rename_seq);
1139 netdev_adjacent_rename_links(dev, oldname);
1141 write_lock_bh(&dev_base_lock);
1142 hlist_del_rcu(&dev->name_hlist);
1143 write_unlock_bh(&dev_base_lock);
1147 write_lock_bh(&dev_base_lock);
1148 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1149 write_unlock_bh(&dev_base_lock);
1151 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1152 ret = notifier_to_errno(ret);
1155 /* err >= 0 after dev_alloc_name() or stores the first errno */
1158 write_seqcount_begin(&devnet_rename_seq);
1159 memcpy(dev->name, oldname, IFNAMSIZ);
1160 memcpy(oldname, newname, IFNAMSIZ);
1161 dev->name_assign_type = old_assign_type;
1162 old_assign_type = NET_NAME_RENAMED;
1165 pr_err("%s: name change rollback failed: %d\n",
1174 * dev_set_alias - change ifalias of a device
1176 * @alias: name up to IFALIASZ
1177 * @len: limit of bytes to copy from info
1179 * Set ifalias for a device,
1181 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1187 if (len >= IFALIASZ)
1191 kfree(dev->ifalias);
1192 dev->ifalias = NULL;
1196 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1199 dev->ifalias = new_ifalias;
1201 strlcpy(dev->ifalias, alias, len+1);
1207 * netdev_features_change - device changes features
1208 * @dev: device to cause notification
1210 * Called to indicate a device has changed features.
1212 void netdev_features_change(struct net_device *dev)
1214 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1216 EXPORT_SYMBOL(netdev_features_change);
1219 * netdev_state_change - device changes state
1220 * @dev: device to cause notification
1222 * Called to indicate a device has changed state. This function calls
1223 * the notifier chains for netdev_chain and sends a NEWLINK message
1224 * to the routing socket.
1226 void netdev_state_change(struct net_device *dev)
1228 if (dev->flags & IFF_UP) {
1229 struct netdev_notifier_change_info change_info;
1231 change_info.flags_changed = 0;
1232 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1234 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1237 EXPORT_SYMBOL(netdev_state_change);
1240 * netdev_notify_peers - notify network peers about existence of @dev
1241 * @dev: network device
1243 * Generate traffic such that interested network peers are aware of
1244 * @dev, such as by generating a gratuitous ARP. This may be used when
1245 * a device wants to inform the rest of the network about some sort of
1246 * reconfiguration such as a failover event or virtual machine
1249 void netdev_notify_peers(struct net_device *dev)
1252 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1255 EXPORT_SYMBOL(netdev_notify_peers);
1257 static int __dev_open(struct net_device *dev)
1259 const struct net_device_ops *ops = dev->netdev_ops;
1264 if (!netif_device_present(dev))
1267 /* Block netpoll from trying to do any rx path servicing.
1268 * If we don't do this there is a chance ndo_poll_controller
1269 * or ndo_poll may be running while we open the device
1271 netpoll_poll_disable(dev);
1273 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1274 ret = notifier_to_errno(ret);
1278 set_bit(__LINK_STATE_START, &dev->state);
1280 if (ops->ndo_validate_addr)
1281 ret = ops->ndo_validate_addr(dev);
1283 if (!ret && ops->ndo_open)
1284 ret = ops->ndo_open(dev);
1286 netpoll_poll_enable(dev);
1289 clear_bit(__LINK_STATE_START, &dev->state);
1291 dev->flags |= IFF_UP;
1292 dev_set_rx_mode(dev);
1294 add_device_randomness(dev->dev_addr, dev->addr_len);
1301 * dev_open - prepare an interface for use.
1302 * @dev: device to open
1304 * Takes a device from down to up state. The device's private open
1305 * function is invoked and then the multicast lists are loaded. Finally
1306 * the device is moved into the up state and a %NETDEV_UP message is
1307 * sent to the netdev notifier chain.
1309 * Calling this function on an active interface is a nop. On a failure
1310 * a negative errno code is returned.
1312 int dev_open(struct net_device *dev)
1316 if (dev->flags & IFF_UP)
1319 ret = __dev_open(dev);
1323 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1324 call_netdevice_notifiers(NETDEV_UP, dev);
1328 EXPORT_SYMBOL(dev_open);
1330 static int __dev_close_many(struct list_head *head)
1332 struct net_device *dev;
1337 list_for_each_entry(dev, head, close_list) {
1338 /* Temporarily disable netpoll until the interface is down */
1339 netpoll_poll_disable(dev);
1341 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1343 clear_bit(__LINK_STATE_START, &dev->state);
1345 /* Synchronize to scheduled poll. We cannot touch poll list, it
1346 * can be even on different cpu. So just clear netif_running().
1348 * dev->stop() will invoke napi_disable() on all of it's
1349 * napi_struct instances on this device.
1351 smp_mb__after_atomic(); /* Commit netif_running(). */
1354 dev_deactivate_many(head);
1356 list_for_each_entry(dev, head, close_list) {
1357 const struct net_device_ops *ops = dev->netdev_ops;
1360 * Call the device specific close. This cannot fail.
1361 * Only if device is UP
1363 * We allow it to be called even after a DETACH hot-plug
1369 dev->flags &= ~IFF_UP;
1370 netpoll_poll_enable(dev);
1376 static int __dev_close(struct net_device *dev)
1381 list_add(&dev->close_list, &single);
1382 retval = __dev_close_many(&single);
1388 static int dev_close_many(struct list_head *head)
1390 struct net_device *dev, *tmp;
1392 /* Remove the devices that don't need to be closed */
1393 list_for_each_entry_safe(dev, tmp, head, close_list)
1394 if (!(dev->flags & IFF_UP))
1395 list_del_init(&dev->close_list);
1397 __dev_close_many(head);
1399 list_for_each_entry_safe(dev, tmp, head, close_list) {
1400 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1401 call_netdevice_notifiers(NETDEV_DOWN, dev);
1402 list_del_init(&dev->close_list);
1409 * dev_close - shutdown an interface.
1410 * @dev: device to shutdown
1412 * This function moves an active device into down state. A
1413 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1414 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1417 int dev_close(struct net_device *dev)
1419 if (dev->flags & IFF_UP) {
1422 list_add(&dev->close_list, &single);
1423 dev_close_many(&single);
1428 EXPORT_SYMBOL(dev_close);
1432 * dev_disable_lro - disable Large Receive Offload on a device
1435 * Disable Large Receive Offload (LRO) on a net device. Must be
1436 * called under RTNL. This is needed if received packets may be
1437 * forwarded to another interface.
1439 void dev_disable_lro(struct net_device *dev)
1441 struct net_device *lower_dev;
1442 struct list_head *iter;
1444 dev->wanted_features &= ~NETIF_F_LRO;
1445 netdev_update_features(dev);
1447 if (unlikely(dev->features & NETIF_F_LRO))
1448 netdev_WARN(dev, "failed to disable LRO!\n");
1450 netdev_for_each_lower_dev(dev, lower_dev, iter)
1451 dev_disable_lro(lower_dev);
1453 EXPORT_SYMBOL(dev_disable_lro);
1455 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1456 struct net_device *dev)
1458 struct netdev_notifier_info info;
1460 netdev_notifier_info_init(&info, dev);
1461 return nb->notifier_call(nb, val, &info);
1464 static int dev_boot_phase = 1;
1467 * register_netdevice_notifier - register a network notifier block
1470 * Register a notifier to be called when network device events occur.
1471 * The notifier passed is linked into the kernel structures and must
1472 * not be reused until it has been unregistered. A negative errno code
1473 * is returned on a failure.
1475 * When registered all registration and up events are replayed
1476 * to the new notifier to allow device to have a race free
1477 * view of the network device list.
1480 int register_netdevice_notifier(struct notifier_block *nb)
1482 struct net_device *dev;
1483 struct net_device *last;
1488 err = raw_notifier_chain_register(&netdev_chain, nb);
1494 for_each_netdev(net, dev) {
1495 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1496 err = notifier_to_errno(err);
1500 if (!(dev->flags & IFF_UP))
1503 call_netdevice_notifier(nb, NETDEV_UP, dev);
1514 for_each_netdev(net, dev) {
1518 if (dev->flags & IFF_UP) {
1519 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1521 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1523 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1528 raw_notifier_chain_unregister(&netdev_chain, nb);
1531 EXPORT_SYMBOL(register_netdevice_notifier);
1534 * unregister_netdevice_notifier - unregister a network notifier block
1537 * Unregister a notifier previously registered by
1538 * register_netdevice_notifier(). The notifier is unlinked into the
1539 * kernel structures and may then be reused. A negative errno code
1540 * is returned on a failure.
1542 * After unregistering unregister and down device events are synthesized
1543 * for all devices on the device list to the removed notifier to remove
1544 * the need for special case cleanup code.
1547 int unregister_netdevice_notifier(struct notifier_block *nb)
1549 struct net_device *dev;
1554 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1559 for_each_netdev(net, dev) {
1560 if (dev->flags & IFF_UP) {
1561 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1563 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1565 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1572 EXPORT_SYMBOL(unregister_netdevice_notifier);
1575 * call_netdevice_notifiers_info - call all network notifier blocks
1576 * @val: value passed unmodified to notifier function
1577 * @dev: net_device pointer passed unmodified to notifier function
1578 * @info: notifier information data
1580 * Call all network notifier blocks. Parameters and return value
1581 * are as for raw_notifier_call_chain().
1584 static int call_netdevice_notifiers_info(unsigned long val,
1585 struct net_device *dev,
1586 struct netdev_notifier_info *info)
1589 netdev_notifier_info_init(info, dev);
1590 return raw_notifier_call_chain(&netdev_chain, val, info);
1594 * call_netdevice_notifiers - call all network notifier blocks
1595 * @val: value passed unmodified to notifier function
1596 * @dev: net_device pointer passed unmodified to notifier function
1598 * Call all network notifier blocks. Parameters and return value
1599 * are as for raw_notifier_call_chain().
1602 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1604 struct netdev_notifier_info info;
1606 return call_netdevice_notifiers_info(val, dev, &info);
1608 EXPORT_SYMBOL(call_netdevice_notifiers);
1610 static struct static_key netstamp_needed __read_mostly;
1611 #ifdef HAVE_JUMP_LABEL
1612 /* We are not allowed to call static_key_slow_dec() from irq context
1613 * If net_disable_timestamp() is called from irq context, defer the
1614 * static_key_slow_dec() calls.
1616 static atomic_t netstamp_needed_deferred;
1619 void net_enable_timestamp(void)
1621 #ifdef HAVE_JUMP_LABEL
1622 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1626 static_key_slow_dec(&netstamp_needed);
1630 static_key_slow_inc(&netstamp_needed);
1632 EXPORT_SYMBOL(net_enable_timestamp);
1634 void net_disable_timestamp(void)
1636 #ifdef HAVE_JUMP_LABEL
1637 if (in_interrupt()) {
1638 atomic_inc(&netstamp_needed_deferred);
1642 static_key_slow_dec(&netstamp_needed);
1644 EXPORT_SYMBOL(net_disable_timestamp);
1646 static inline void net_timestamp_set(struct sk_buff *skb)
1648 skb->tstamp.tv64 = 0;
1649 if (static_key_false(&netstamp_needed))
1650 __net_timestamp(skb);
1653 #define net_timestamp_check(COND, SKB) \
1654 if (static_key_false(&netstamp_needed)) { \
1655 if ((COND) && !(SKB)->tstamp.tv64) \
1656 __net_timestamp(SKB); \
1659 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1663 if (!(dev->flags & IFF_UP))
1666 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1667 if (skb->len <= len)
1670 /* if TSO is enabled, we don't care about the length as the packet
1671 * could be forwarded without being segmented before
1673 if (skb_is_gso(skb))
1678 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1680 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1682 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1683 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1684 atomic_long_inc(&dev->rx_dropped);
1690 if (unlikely(!is_skb_forwardable(dev, skb))) {
1691 atomic_long_inc(&dev->rx_dropped);
1696 skb_scrub_packet(skb, true);
1697 skb->protocol = eth_type_trans(skb, dev);
1698 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1702 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1705 * dev_forward_skb - loopback an skb to another netif
1707 * @dev: destination network device
1708 * @skb: buffer to forward
1711 * NET_RX_SUCCESS (no congestion)
1712 * NET_RX_DROP (packet was dropped, but freed)
1714 * dev_forward_skb can be used for injecting an skb from the
1715 * start_xmit function of one device into the receive queue
1716 * of another device.
1718 * The receiving device may be in another namespace, so
1719 * we have to clear all information in the skb that could
1720 * impact namespace isolation.
1722 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1724 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1726 EXPORT_SYMBOL_GPL(dev_forward_skb);
1728 static inline int deliver_skb(struct sk_buff *skb,
1729 struct packet_type *pt_prev,
1730 struct net_device *orig_dev)
1732 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1734 atomic_inc(&skb->users);
1735 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1738 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1739 struct packet_type **pt,
1740 struct net_device *dev, __be16 type,
1741 struct list_head *ptype_list)
1743 struct packet_type *ptype, *pt_prev = *pt;
1745 list_for_each_entry_rcu(ptype, ptype_list, list) {
1746 if (ptype->type != type)
1749 deliver_skb(skb, pt_prev, dev);
1755 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1757 if (!ptype->af_packet_priv || !skb->sk)
1760 if (ptype->id_match)
1761 return ptype->id_match(ptype, skb->sk);
1762 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1769 * Support routine. Sends outgoing frames to any network
1770 * taps currently in use.
1773 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1775 struct packet_type *ptype;
1776 struct sk_buff *skb2 = NULL;
1777 struct packet_type *pt_prev = NULL;
1778 struct list_head *ptype_list = &ptype_all;
1782 list_for_each_entry_rcu(ptype, ptype_list, list) {
1783 /* Never send packets back to the socket
1784 * they originated from - MvS (miquels@drinkel.ow.org)
1786 if (skb_loop_sk(ptype, skb))
1790 deliver_skb(skb2, pt_prev, skb->dev);
1795 /* need to clone skb, done only once */
1796 skb2 = skb_clone(skb, GFP_ATOMIC);
1800 net_timestamp_set(skb2);
1802 /* skb->nh should be correctly
1803 * set by sender, so that the second statement is
1804 * just protection against buggy protocols.
1806 skb_reset_mac_header(skb2);
1808 if (skb_network_header(skb2) < skb2->data ||
1809 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1810 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1811 ntohs(skb2->protocol),
1813 skb_reset_network_header(skb2);
1816 skb2->transport_header = skb2->network_header;
1817 skb2->pkt_type = PACKET_OUTGOING;
1821 if (ptype_list == &ptype_all) {
1822 ptype_list = &dev->ptype_all;
1827 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1832 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1833 * @dev: Network device
1834 * @txq: number of queues available
1836 * If real_num_tx_queues is changed the tc mappings may no longer be
1837 * valid. To resolve this verify the tc mapping remains valid and if
1838 * not NULL the mapping. With no priorities mapping to this
1839 * offset/count pair it will no longer be used. In the worst case TC0
1840 * is invalid nothing can be done so disable priority mappings. If is
1841 * expected that drivers will fix this mapping if they can before
1842 * calling netif_set_real_num_tx_queues.
1844 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1847 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1849 /* If TC0 is invalidated disable TC mapping */
1850 if (tc->offset + tc->count > txq) {
1851 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1856 /* Invalidated prio to tc mappings set to TC0 */
1857 for (i = 1; i < TC_BITMASK + 1; i++) {
1858 int q = netdev_get_prio_tc_map(dev, i);
1860 tc = &dev->tc_to_txq[q];
1861 if (tc->offset + tc->count > txq) {
1862 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1864 netdev_set_prio_tc_map(dev, i, 0);
1870 static DEFINE_MUTEX(xps_map_mutex);
1871 #define xmap_dereference(P) \
1872 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1874 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1877 struct xps_map *map = NULL;
1881 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1883 for (pos = 0; map && pos < map->len; pos++) {
1884 if (map->queues[pos] == index) {
1886 map->queues[pos] = map->queues[--map->len];
1888 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1889 kfree_rcu(map, rcu);
1899 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1901 struct xps_dev_maps *dev_maps;
1903 bool active = false;
1905 mutex_lock(&xps_map_mutex);
1906 dev_maps = xmap_dereference(dev->xps_maps);
1911 for_each_possible_cpu(cpu) {
1912 for (i = index; i < dev->num_tx_queues; i++) {
1913 if (!remove_xps_queue(dev_maps, cpu, i))
1916 if (i == dev->num_tx_queues)
1921 RCU_INIT_POINTER(dev->xps_maps, NULL);
1922 kfree_rcu(dev_maps, rcu);
1925 for (i = index; i < dev->num_tx_queues; i++)
1926 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1930 mutex_unlock(&xps_map_mutex);
1933 static struct xps_map *expand_xps_map(struct xps_map *map,
1936 struct xps_map *new_map;
1937 int alloc_len = XPS_MIN_MAP_ALLOC;
1940 for (pos = 0; map && pos < map->len; pos++) {
1941 if (map->queues[pos] != index)
1946 /* Need to add queue to this CPU's existing map */
1948 if (pos < map->alloc_len)
1951 alloc_len = map->alloc_len * 2;
1954 /* Need to allocate new map to store queue on this CPU's map */
1955 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1960 for (i = 0; i < pos; i++)
1961 new_map->queues[i] = map->queues[i];
1962 new_map->alloc_len = alloc_len;
1968 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1971 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1972 struct xps_map *map, *new_map;
1973 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1974 int cpu, numa_node_id = -2;
1975 bool active = false;
1977 mutex_lock(&xps_map_mutex);
1979 dev_maps = xmap_dereference(dev->xps_maps);
1981 /* allocate memory for queue storage */
1982 for_each_online_cpu(cpu) {
1983 if (!cpumask_test_cpu(cpu, mask))
1987 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1988 if (!new_dev_maps) {
1989 mutex_unlock(&xps_map_mutex);
1993 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1996 map = expand_xps_map(map, cpu, index);
2000 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2004 goto out_no_new_maps;
2006 for_each_possible_cpu(cpu) {
2007 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2008 /* add queue to CPU maps */
2011 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2012 while ((pos < map->len) && (map->queues[pos] != index))
2015 if (pos == map->len)
2016 map->queues[map->len++] = index;
2018 if (numa_node_id == -2)
2019 numa_node_id = cpu_to_node(cpu);
2020 else if (numa_node_id != cpu_to_node(cpu))
2023 } else if (dev_maps) {
2024 /* fill in the new device map from the old device map */
2025 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2026 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2031 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2033 /* Cleanup old maps */
2035 for_each_possible_cpu(cpu) {
2036 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2037 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2038 if (map && map != new_map)
2039 kfree_rcu(map, rcu);
2042 kfree_rcu(dev_maps, rcu);
2045 dev_maps = new_dev_maps;
2049 /* update Tx queue numa node */
2050 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2051 (numa_node_id >= 0) ? numa_node_id :
2057 /* removes queue from unused CPUs */
2058 for_each_possible_cpu(cpu) {
2059 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2062 if (remove_xps_queue(dev_maps, cpu, index))
2066 /* free map if not active */
2068 RCU_INIT_POINTER(dev->xps_maps, NULL);
2069 kfree_rcu(dev_maps, rcu);
2073 mutex_unlock(&xps_map_mutex);
2077 /* remove any maps that we added */
2078 for_each_possible_cpu(cpu) {
2079 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2080 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2082 if (new_map && new_map != map)
2086 mutex_unlock(&xps_map_mutex);
2088 kfree(new_dev_maps);
2091 EXPORT_SYMBOL(netif_set_xps_queue);
2095 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2096 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2098 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2102 if (txq < 1 || txq > dev->num_tx_queues)
2105 if (dev->reg_state == NETREG_REGISTERED ||
2106 dev->reg_state == NETREG_UNREGISTERING) {
2109 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2115 netif_setup_tc(dev, txq);
2117 if (txq < dev->real_num_tx_queues) {
2118 qdisc_reset_all_tx_gt(dev, txq);
2120 netif_reset_xps_queues_gt(dev, txq);
2125 dev->real_num_tx_queues = txq;
2128 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2132 * netif_set_real_num_rx_queues - set actual number of RX queues used
2133 * @dev: Network device
2134 * @rxq: Actual number of RX queues
2136 * This must be called either with the rtnl_lock held or before
2137 * registration of the net device. Returns 0 on success, or a
2138 * negative error code. If called before registration, it always
2141 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2145 if (rxq < 1 || rxq > dev->num_rx_queues)
2148 if (dev->reg_state == NETREG_REGISTERED) {
2151 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2157 dev->real_num_rx_queues = rxq;
2160 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2164 * netif_get_num_default_rss_queues - default number of RSS queues
2166 * This routine should set an upper limit on the number of RSS queues
2167 * used by default by multiqueue devices.
2169 int netif_get_num_default_rss_queues(void)
2171 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2173 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2175 static inline void __netif_reschedule(struct Qdisc *q)
2177 struct softnet_data *sd;
2178 unsigned long flags;
2180 local_irq_save(flags);
2181 sd = this_cpu_ptr(&softnet_data);
2182 q->next_sched = NULL;
2183 *sd->output_queue_tailp = q;
2184 sd->output_queue_tailp = &q->next_sched;
2185 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2186 local_irq_restore(flags);
2189 void __netif_schedule(struct Qdisc *q)
2191 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2192 __netif_reschedule(q);
2194 EXPORT_SYMBOL(__netif_schedule);
2196 struct dev_kfree_skb_cb {
2197 enum skb_free_reason reason;
2200 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2202 return (struct dev_kfree_skb_cb *)skb->cb;
2205 void netif_schedule_queue(struct netdev_queue *txq)
2208 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2209 struct Qdisc *q = rcu_dereference(txq->qdisc);
2211 __netif_schedule(q);
2215 EXPORT_SYMBOL(netif_schedule_queue);
2218 * netif_wake_subqueue - allow sending packets on subqueue
2219 * @dev: network device
2220 * @queue_index: sub queue index
2222 * Resume individual transmit queue of a device with multiple transmit queues.
2224 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2226 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2228 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2232 q = rcu_dereference(txq->qdisc);
2233 __netif_schedule(q);
2237 EXPORT_SYMBOL(netif_wake_subqueue);
2239 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2241 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2245 q = rcu_dereference(dev_queue->qdisc);
2246 __netif_schedule(q);
2250 EXPORT_SYMBOL(netif_tx_wake_queue);
2252 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2254 unsigned long flags;
2256 if (likely(atomic_read(&skb->users) == 1)) {
2258 atomic_set(&skb->users, 0);
2259 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2262 get_kfree_skb_cb(skb)->reason = reason;
2263 local_irq_save(flags);
2264 skb->next = __this_cpu_read(softnet_data.completion_queue);
2265 __this_cpu_write(softnet_data.completion_queue, skb);
2266 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2267 local_irq_restore(flags);
2269 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2271 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2273 if (in_irq() || irqs_disabled())
2274 __dev_kfree_skb_irq(skb, reason);
2278 EXPORT_SYMBOL(__dev_kfree_skb_any);
2282 * netif_device_detach - mark device as removed
2283 * @dev: network device
2285 * Mark device as removed from system and therefore no longer available.
2287 void netif_device_detach(struct net_device *dev)
2289 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2290 netif_running(dev)) {
2291 netif_tx_stop_all_queues(dev);
2294 EXPORT_SYMBOL(netif_device_detach);
2297 * netif_device_attach - mark device as attached
2298 * @dev: network device
2300 * Mark device as attached from system and restart if needed.
2302 void netif_device_attach(struct net_device *dev)
2304 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2305 netif_running(dev)) {
2306 netif_tx_wake_all_queues(dev);
2307 __netdev_watchdog_up(dev);
2310 EXPORT_SYMBOL(netif_device_attach);
2312 static void skb_warn_bad_offload(const struct sk_buff *skb)
2314 static const netdev_features_t null_features = 0;
2315 struct net_device *dev = skb->dev;
2316 const char *driver = "";
2318 if (!net_ratelimit())
2321 if (dev && dev->dev.parent)
2322 driver = dev_driver_string(dev->dev.parent);
2324 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2325 "gso_type=%d ip_summed=%d\n",
2326 driver, dev ? &dev->features : &null_features,
2327 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2328 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2329 skb_shinfo(skb)->gso_type, skb->ip_summed);
2333 * Invalidate hardware checksum when packet is to be mangled, and
2334 * complete checksum manually on outgoing path.
2336 int skb_checksum_help(struct sk_buff *skb)
2339 int ret = 0, offset;
2341 if (skb->ip_summed == CHECKSUM_COMPLETE)
2342 goto out_set_summed;
2344 if (unlikely(skb_shinfo(skb)->gso_size)) {
2345 skb_warn_bad_offload(skb);
2349 /* Before computing a checksum, we should make sure no frag could
2350 * be modified by an external entity : checksum could be wrong.
2352 if (skb_has_shared_frag(skb)) {
2353 ret = __skb_linearize(skb);
2358 offset = skb_checksum_start_offset(skb);
2359 BUG_ON(offset >= skb_headlen(skb));
2360 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2362 offset += skb->csum_offset;
2363 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2365 if (skb_cloned(skb) &&
2366 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2367 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2372 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2374 skb->ip_summed = CHECKSUM_NONE;
2378 EXPORT_SYMBOL(skb_checksum_help);
2380 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2382 unsigned int vlan_depth = skb->mac_len;
2383 __be16 type = skb->protocol;
2385 /* Tunnel gso handlers can set protocol to ethernet. */
2386 if (type == htons(ETH_P_TEB)) {
2389 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2392 eth = (struct ethhdr *)skb_mac_header(skb);
2393 type = eth->h_proto;
2396 /* if skb->protocol is 802.1Q/AD then the header should already be
2397 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2398 * ETH_HLEN otherwise
2400 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2402 if (WARN_ON(vlan_depth < VLAN_HLEN))
2404 vlan_depth -= VLAN_HLEN;
2406 vlan_depth = ETH_HLEN;
2409 struct vlan_hdr *vh;
2411 if (unlikely(!pskb_may_pull(skb,
2412 vlan_depth + VLAN_HLEN)))
2415 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2416 type = vh->h_vlan_encapsulated_proto;
2417 vlan_depth += VLAN_HLEN;
2418 } while (type == htons(ETH_P_8021Q) ||
2419 type == htons(ETH_P_8021AD));
2422 *depth = vlan_depth;
2428 * skb_mac_gso_segment - mac layer segmentation handler.
2429 * @skb: buffer to segment
2430 * @features: features for the output path (see dev->features)
2432 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2433 netdev_features_t features)
2435 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2436 struct packet_offload *ptype;
2437 int vlan_depth = skb->mac_len;
2438 __be16 type = skb_network_protocol(skb, &vlan_depth);
2440 if (unlikely(!type))
2441 return ERR_PTR(-EINVAL);
2443 __skb_pull(skb, vlan_depth);
2446 list_for_each_entry_rcu(ptype, &offload_base, list) {
2447 if (ptype->type == type && ptype->callbacks.gso_segment) {
2448 segs = ptype->callbacks.gso_segment(skb, features);
2454 __skb_push(skb, skb->data - skb_mac_header(skb));
2458 EXPORT_SYMBOL(skb_mac_gso_segment);
2461 /* openvswitch calls this on rx path, so we need a different check.
2463 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2466 return skb->ip_summed != CHECKSUM_PARTIAL;
2468 return skb->ip_summed == CHECKSUM_NONE;
2472 * __skb_gso_segment - Perform segmentation on skb.
2473 * @skb: buffer to segment
2474 * @features: features for the output path (see dev->features)
2475 * @tx_path: whether it is called in TX path
2477 * This function segments the given skb and returns a list of segments.
2479 * It may return NULL if the skb requires no segmentation. This is
2480 * only possible when GSO is used for verifying header integrity.
2482 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2483 netdev_features_t features, bool tx_path)
2485 if (unlikely(skb_needs_check(skb, tx_path))) {
2488 skb_warn_bad_offload(skb);
2490 err = skb_cow_head(skb, 0);
2492 return ERR_PTR(err);
2495 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2496 SKB_GSO_CB(skb)->encap_level = 0;
2498 skb_reset_mac_header(skb);
2499 skb_reset_mac_len(skb);
2501 return skb_mac_gso_segment(skb, features);
2503 EXPORT_SYMBOL(__skb_gso_segment);
2505 /* Take action when hardware reception checksum errors are detected. */
2507 void netdev_rx_csum_fault(struct net_device *dev)
2509 if (net_ratelimit()) {
2510 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2514 EXPORT_SYMBOL(netdev_rx_csum_fault);
2517 /* Actually, we should eliminate this check as soon as we know, that:
2518 * 1. IOMMU is present and allows to map all the memory.
2519 * 2. No high memory really exists on this machine.
2522 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2524 #ifdef CONFIG_HIGHMEM
2526 if (!(dev->features & NETIF_F_HIGHDMA)) {
2527 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2528 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2529 if (PageHighMem(skb_frag_page(frag)))
2534 if (PCI_DMA_BUS_IS_PHYS) {
2535 struct device *pdev = dev->dev.parent;
2539 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2540 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2541 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2542 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2550 /* If MPLS offload request, verify we are testing hardware MPLS features
2551 * instead of standard features for the netdev.
2553 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2554 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2555 netdev_features_t features,
2558 if (eth_p_mpls(type))
2559 features &= skb->dev->mpls_features;
2564 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2565 netdev_features_t features,
2572 static netdev_features_t harmonize_features(struct sk_buff *skb,
2573 netdev_features_t features)
2578 type = skb_network_protocol(skb, &tmp);
2579 features = net_mpls_features(skb, features, type);
2581 if (skb->ip_summed != CHECKSUM_NONE &&
2582 !can_checksum_protocol(features, type)) {
2583 features &= ~NETIF_F_ALL_CSUM;
2584 } else if (illegal_highdma(skb->dev, skb)) {
2585 features &= ~NETIF_F_SG;
2591 netdev_features_t netif_skb_features(struct sk_buff *skb)
2593 struct net_device *dev = skb->dev;
2594 netdev_features_t features = dev->features;
2595 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2596 __be16 protocol = skb->protocol;
2598 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2599 features &= ~NETIF_F_GSO_MASK;
2601 /* If encapsulation offload request, verify we are testing
2602 * hardware encapsulation features instead of standard
2603 * features for the netdev
2605 if (skb->encapsulation)
2606 features &= dev->hw_enc_features;
2608 if (!skb_vlan_tag_present(skb)) {
2609 if (unlikely(protocol == htons(ETH_P_8021Q) ||
2610 protocol == htons(ETH_P_8021AD))) {
2611 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2612 protocol = veh->h_vlan_encapsulated_proto;
2618 features = netdev_intersect_features(features,
2619 dev->vlan_features |
2620 NETIF_F_HW_VLAN_CTAG_TX |
2621 NETIF_F_HW_VLAN_STAG_TX);
2623 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2624 features = netdev_intersect_features(features,
2629 NETIF_F_HW_VLAN_CTAG_TX |
2630 NETIF_F_HW_VLAN_STAG_TX);
2633 if (dev->netdev_ops->ndo_features_check)
2634 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2637 return harmonize_features(skb, features);
2639 EXPORT_SYMBOL(netif_skb_features);
2641 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2642 struct netdev_queue *txq, bool more)
2647 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2648 dev_queue_xmit_nit(skb, dev);
2651 trace_net_dev_start_xmit(skb, dev);
2652 rc = netdev_start_xmit(skb, dev, txq, more);
2653 trace_net_dev_xmit(skb, rc, dev, len);
2658 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2659 struct netdev_queue *txq, int *ret)
2661 struct sk_buff *skb = first;
2662 int rc = NETDEV_TX_OK;
2665 struct sk_buff *next = skb->next;
2668 rc = xmit_one(skb, dev, txq, next != NULL);
2669 if (unlikely(!dev_xmit_complete(rc))) {
2675 if (netif_xmit_stopped(txq) && skb) {
2676 rc = NETDEV_TX_BUSY;
2686 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2687 netdev_features_t features)
2689 if (skb_vlan_tag_present(skb) &&
2690 !vlan_hw_offload_capable(features, skb->vlan_proto))
2691 skb = __vlan_hwaccel_push_inside(skb);
2695 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2697 netdev_features_t features;
2702 features = netif_skb_features(skb);
2703 skb = validate_xmit_vlan(skb, features);
2707 if (netif_needs_gso(dev, skb, features)) {
2708 struct sk_buff *segs;
2710 segs = skb_gso_segment(skb, features);
2718 if (skb_needs_linearize(skb, features) &&
2719 __skb_linearize(skb))
2722 /* If packet is not checksummed and device does not
2723 * support checksumming for this protocol, complete
2724 * checksumming here.
2726 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2727 if (skb->encapsulation)
2728 skb_set_inner_transport_header(skb,
2729 skb_checksum_start_offset(skb));
2731 skb_set_transport_header(skb,
2732 skb_checksum_start_offset(skb));
2733 if (!(features & NETIF_F_ALL_CSUM) &&
2734 skb_checksum_help(skb))
2747 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2749 struct sk_buff *next, *head = NULL, *tail;
2751 for (; skb != NULL; skb = next) {
2755 /* in case skb wont be segmented, point to itself */
2758 skb = validate_xmit_skb(skb, dev);
2766 /* If skb was segmented, skb->prev points to
2767 * the last segment. If not, it still contains skb.
2774 static void qdisc_pkt_len_init(struct sk_buff *skb)
2776 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2778 qdisc_skb_cb(skb)->pkt_len = skb->len;
2780 /* To get more precise estimation of bytes sent on wire,
2781 * we add to pkt_len the headers size of all segments
2783 if (shinfo->gso_size) {
2784 unsigned int hdr_len;
2785 u16 gso_segs = shinfo->gso_segs;
2787 /* mac layer + network layer */
2788 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2790 /* + transport layer */
2791 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2792 hdr_len += tcp_hdrlen(skb);
2794 hdr_len += sizeof(struct udphdr);
2796 if (shinfo->gso_type & SKB_GSO_DODGY)
2797 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2800 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2804 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2805 struct net_device *dev,
2806 struct netdev_queue *txq)
2808 spinlock_t *root_lock = qdisc_lock(q);
2812 qdisc_pkt_len_init(skb);
2813 qdisc_calculate_pkt_len(skb, q);
2815 * Heuristic to force contended enqueues to serialize on a
2816 * separate lock before trying to get qdisc main lock.
2817 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2818 * often and dequeue packets faster.
2820 contended = qdisc_is_running(q);
2821 if (unlikely(contended))
2822 spin_lock(&q->busylock);
2824 spin_lock(root_lock);
2825 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2828 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2829 qdisc_run_begin(q)) {
2831 * This is a work-conserving queue; there are no old skbs
2832 * waiting to be sent out; and the qdisc is not running -
2833 * xmit the skb directly.
2836 qdisc_bstats_update(q, skb);
2838 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2839 if (unlikely(contended)) {
2840 spin_unlock(&q->busylock);
2847 rc = NET_XMIT_SUCCESS;
2849 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2850 if (qdisc_run_begin(q)) {
2851 if (unlikely(contended)) {
2852 spin_unlock(&q->busylock);
2858 spin_unlock(root_lock);
2859 if (unlikely(contended))
2860 spin_unlock(&q->busylock);
2864 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2865 static void skb_update_prio(struct sk_buff *skb)
2867 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2869 if (!skb->priority && skb->sk && map) {
2870 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2872 if (prioidx < map->priomap_len)
2873 skb->priority = map->priomap[prioidx];
2877 #define skb_update_prio(skb)
2880 static DEFINE_PER_CPU(int, xmit_recursion);
2881 #define RECURSION_LIMIT 10
2884 * dev_loopback_xmit - loop back @skb
2885 * @skb: buffer to transmit
2887 int dev_loopback_xmit(struct sk_buff *skb)
2889 skb_reset_mac_header(skb);
2890 __skb_pull(skb, skb_network_offset(skb));
2891 skb->pkt_type = PACKET_LOOPBACK;
2892 skb->ip_summed = CHECKSUM_UNNECESSARY;
2893 WARN_ON(!skb_dst(skb));
2898 EXPORT_SYMBOL(dev_loopback_xmit);
2901 * __dev_queue_xmit - transmit a buffer
2902 * @skb: buffer to transmit
2903 * @accel_priv: private data used for L2 forwarding offload
2905 * Queue a buffer for transmission to a network device. The caller must
2906 * have set the device and priority and built the buffer before calling
2907 * this function. The function can be called from an interrupt.
2909 * A negative errno code is returned on a failure. A success does not
2910 * guarantee the frame will be transmitted as it may be dropped due
2911 * to congestion or traffic shaping.
2913 * -----------------------------------------------------------------------------------
2914 * I notice this method can also return errors from the queue disciplines,
2915 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2918 * Regardless of the return value, the skb is consumed, so it is currently
2919 * difficult to retry a send to this method. (You can bump the ref count
2920 * before sending to hold a reference for retry if you are careful.)
2922 * When calling this method, interrupts MUST be enabled. This is because
2923 * the BH enable code must have IRQs enabled so that it will not deadlock.
2926 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2928 struct net_device *dev = skb->dev;
2929 struct netdev_queue *txq;
2933 skb_reset_mac_header(skb);
2935 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2936 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2938 /* Disable soft irqs for various locks below. Also
2939 * stops preemption for RCU.
2943 skb_update_prio(skb);
2945 /* If device/qdisc don't need skb->dst, release it right now while
2946 * its hot in this cpu cache.
2948 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2953 txq = netdev_pick_tx(dev, skb, accel_priv);
2954 q = rcu_dereference_bh(txq->qdisc);
2956 #ifdef CONFIG_NET_CLS_ACT
2957 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2959 trace_net_dev_queue(skb);
2961 rc = __dev_xmit_skb(skb, q, dev, txq);
2965 /* The device has no queue. Common case for software devices:
2966 loopback, all the sorts of tunnels...
2968 Really, it is unlikely that netif_tx_lock protection is necessary
2969 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2971 However, it is possible, that they rely on protection
2974 Check this and shot the lock. It is not prone from deadlocks.
2975 Either shot noqueue qdisc, it is even simpler 8)
2977 if (dev->flags & IFF_UP) {
2978 int cpu = smp_processor_id(); /* ok because BHs are off */
2980 if (txq->xmit_lock_owner != cpu) {
2982 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2983 goto recursion_alert;
2985 skb = validate_xmit_skb(skb, dev);
2989 HARD_TX_LOCK(dev, txq, cpu);
2991 if (!netif_xmit_stopped(txq)) {
2992 __this_cpu_inc(xmit_recursion);
2993 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2994 __this_cpu_dec(xmit_recursion);
2995 if (dev_xmit_complete(rc)) {
2996 HARD_TX_UNLOCK(dev, txq);
3000 HARD_TX_UNLOCK(dev, txq);
3001 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3004 /* Recursion is detected! It is possible,
3008 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3015 rcu_read_unlock_bh();
3017 atomic_long_inc(&dev->tx_dropped);
3018 kfree_skb_list(skb);
3021 rcu_read_unlock_bh();
3025 int dev_queue_xmit(struct sk_buff *skb)
3027 return __dev_queue_xmit(skb, NULL);
3029 EXPORT_SYMBOL(dev_queue_xmit);
3031 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3033 return __dev_queue_xmit(skb, accel_priv);
3035 EXPORT_SYMBOL(dev_queue_xmit_accel);
3038 /*=======================================================================
3040 =======================================================================*/
3042 int netdev_max_backlog __read_mostly = 1000;
3043 EXPORT_SYMBOL(netdev_max_backlog);
3045 int netdev_tstamp_prequeue __read_mostly = 1;
3046 int netdev_budget __read_mostly = 300;
3047 int weight_p __read_mostly = 64; /* old backlog weight */
3049 /* Called with irq disabled */
3050 static inline void ____napi_schedule(struct softnet_data *sd,
3051 struct napi_struct *napi)
3053 list_add_tail(&napi->poll_list, &sd->poll_list);
3054 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3059 /* One global table that all flow-based protocols share. */
3060 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3061 EXPORT_SYMBOL(rps_sock_flow_table);
3063 struct static_key rps_needed __read_mostly;
3065 static struct rps_dev_flow *
3066 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3067 struct rps_dev_flow *rflow, u16 next_cpu)
3069 if (next_cpu != RPS_NO_CPU) {
3070 #ifdef CONFIG_RFS_ACCEL
3071 struct netdev_rx_queue *rxqueue;
3072 struct rps_dev_flow_table *flow_table;
3073 struct rps_dev_flow *old_rflow;
3078 /* Should we steer this flow to a different hardware queue? */
3079 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3080 !(dev->features & NETIF_F_NTUPLE))
3082 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3083 if (rxq_index == skb_get_rx_queue(skb))
3086 rxqueue = dev->_rx + rxq_index;
3087 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3090 flow_id = skb_get_hash(skb) & flow_table->mask;
3091 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3092 rxq_index, flow_id);
3096 rflow = &flow_table->flows[flow_id];
3098 if (old_rflow->filter == rflow->filter)
3099 old_rflow->filter = RPS_NO_FILTER;
3103 per_cpu(softnet_data, next_cpu).input_queue_head;
3106 rflow->cpu = next_cpu;
3111 * get_rps_cpu is called from netif_receive_skb and returns the target
3112 * CPU from the RPS map of the receiving queue for a given skb.
3113 * rcu_read_lock must be held on entry.
3115 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3116 struct rps_dev_flow **rflowp)
3118 struct netdev_rx_queue *rxqueue;
3119 struct rps_map *map;
3120 struct rps_dev_flow_table *flow_table;
3121 struct rps_sock_flow_table *sock_flow_table;
3126 if (skb_rx_queue_recorded(skb)) {
3127 u16 index = skb_get_rx_queue(skb);
3128 if (unlikely(index >= dev->real_num_rx_queues)) {
3129 WARN_ONCE(dev->real_num_rx_queues > 1,
3130 "%s received packet on queue %u, but number "
3131 "of RX queues is %u\n",
3132 dev->name, index, dev->real_num_rx_queues);
3135 rxqueue = dev->_rx + index;
3139 map = rcu_dereference(rxqueue->rps_map);
3141 if (map->len == 1 &&
3142 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3143 tcpu = map->cpus[0];
3144 if (cpu_online(tcpu))
3148 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3152 skb_reset_network_header(skb);
3153 hash = skb_get_hash(skb);
3157 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3158 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3159 if (flow_table && sock_flow_table) {
3161 struct rps_dev_flow *rflow;
3163 rflow = &flow_table->flows[hash & flow_table->mask];
3166 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3169 * If the desired CPU (where last recvmsg was done) is
3170 * different from current CPU (one in the rx-queue flow
3171 * table entry), switch if one of the following holds:
3172 * - Current CPU is unset (equal to RPS_NO_CPU).
3173 * - Current CPU is offline.
3174 * - The current CPU's queue tail has advanced beyond the
3175 * last packet that was enqueued using this table entry.
3176 * This guarantees that all previous packets for the flow
3177 * have been dequeued, thus preserving in order delivery.
3179 if (unlikely(tcpu != next_cpu) &&
3180 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3181 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3182 rflow->last_qtail)) >= 0)) {
3184 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3187 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3195 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3196 if (cpu_online(tcpu)) {
3206 #ifdef CONFIG_RFS_ACCEL
3209 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3210 * @dev: Device on which the filter was set
3211 * @rxq_index: RX queue index
3212 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3213 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3215 * Drivers that implement ndo_rx_flow_steer() should periodically call
3216 * this function for each installed filter and remove the filters for
3217 * which it returns %true.
3219 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3220 u32 flow_id, u16 filter_id)
3222 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3223 struct rps_dev_flow_table *flow_table;
3224 struct rps_dev_flow *rflow;
3229 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3230 if (flow_table && flow_id <= flow_table->mask) {
3231 rflow = &flow_table->flows[flow_id];
3232 cpu = ACCESS_ONCE(rflow->cpu);
3233 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3234 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3235 rflow->last_qtail) <
3236 (int)(10 * flow_table->mask)))
3242 EXPORT_SYMBOL(rps_may_expire_flow);
3244 #endif /* CONFIG_RFS_ACCEL */
3246 /* Called from hardirq (IPI) context */
3247 static void rps_trigger_softirq(void *data)
3249 struct softnet_data *sd = data;
3251 ____napi_schedule(sd, &sd->backlog);
3255 #endif /* CONFIG_RPS */
3258 * Check if this softnet_data structure is another cpu one
3259 * If yes, queue it to our IPI list and return 1
3262 static int rps_ipi_queued(struct softnet_data *sd)
3265 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3268 sd->rps_ipi_next = mysd->rps_ipi_list;
3269 mysd->rps_ipi_list = sd;
3271 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3274 #endif /* CONFIG_RPS */
3278 #ifdef CONFIG_NET_FLOW_LIMIT
3279 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3282 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3284 #ifdef CONFIG_NET_FLOW_LIMIT
3285 struct sd_flow_limit *fl;
3286 struct softnet_data *sd;
3287 unsigned int old_flow, new_flow;
3289 if (qlen < (netdev_max_backlog >> 1))
3292 sd = this_cpu_ptr(&softnet_data);
3295 fl = rcu_dereference(sd->flow_limit);
3297 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3298 old_flow = fl->history[fl->history_head];
3299 fl->history[fl->history_head] = new_flow;
3302 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3304 if (likely(fl->buckets[old_flow]))
3305 fl->buckets[old_flow]--;
3307 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3319 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3320 * queue (may be a remote CPU queue).
3322 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3323 unsigned int *qtail)
3325 struct softnet_data *sd;
3326 unsigned long flags;
3329 sd = &per_cpu(softnet_data, cpu);
3331 local_irq_save(flags);
3334 qlen = skb_queue_len(&sd->input_pkt_queue);
3335 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3338 __skb_queue_tail(&sd->input_pkt_queue, skb);
3339 input_queue_tail_incr_save(sd, qtail);
3341 local_irq_restore(flags);
3342 return NET_RX_SUCCESS;
3345 /* Schedule NAPI for backlog device
3346 * We can use non atomic operation since we own the queue lock
3348 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3349 if (!rps_ipi_queued(sd))
3350 ____napi_schedule(sd, &sd->backlog);
3358 local_irq_restore(flags);
3360 atomic_long_inc(&skb->dev->rx_dropped);
3365 static int netif_rx_internal(struct sk_buff *skb)
3369 net_timestamp_check(netdev_tstamp_prequeue, skb);
3371 trace_netif_rx(skb);
3373 if (static_key_false(&rps_needed)) {
3374 struct rps_dev_flow voidflow, *rflow = &voidflow;
3380 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3382 cpu = smp_processor_id();
3384 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3392 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3399 * netif_rx - post buffer to the network code
3400 * @skb: buffer to post
3402 * This function receives a packet from a device driver and queues it for
3403 * the upper (protocol) levels to process. It always succeeds. The buffer
3404 * may be dropped during processing for congestion control or by the
3408 * NET_RX_SUCCESS (no congestion)
3409 * NET_RX_DROP (packet was dropped)
3413 int netif_rx(struct sk_buff *skb)
3415 trace_netif_rx_entry(skb);
3417 return netif_rx_internal(skb);
3419 EXPORT_SYMBOL(netif_rx);
3421 int netif_rx_ni(struct sk_buff *skb)
3425 trace_netif_rx_ni_entry(skb);
3428 err = netif_rx_internal(skb);
3429 if (local_softirq_pending())
3435 EXPORT_SYMBOL(netif_rx_ni);
3437 static void net_tx_action(struct softirq_action *h)
3439 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3441 if (sd->completion_queue) {
3442 struct sk_buff *clist;
3444 local_irq_disable();
3445 clist = sd->completion_queue;
3446 sd->completion_queue = NULL;
3450 struct sk_buff *skb = clist;
3451 clist = clist->next;
3453 WARN_ON(atomic_read(&skb->users));
3454 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3455 trace_consume_skb(skb);
3457 trace_kfree_skb(skb, net_tx_action);
3462 if (sd->output_queue) {
3465 local_irq_disable();
3466 head = sd->output_queue;
3467 sd->output_queue = NULL;
3468 sd->output_queue_tailp = &sd->output_queue;
3472 struct Qdisc *q = head;
3473 spinlock_t *root_lock;
3475 head = head->next_sched;
3477 root_lock = qdisc_lock(q);
3478 if (spin_trylock(root_lock)) {
3479 smp_mb__before_atomic();
3480 clear_bit(__QDISC_STATE_SCHED,
3483 spin_unlock(root_lock);
3485 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3487 __netif_reschedule(q);
3489 smp_mb__before_atomic();
3490 clear_bit(__QDISC_STATE_SCHED,
3498 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3499 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3500 /* This hook is defined here for ATM LANE */
3501 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3502 unsigned char *addr) __read_mostly;
3503 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3506 #ifdef CONFIG_NET_CLS_ACT
3507 /* TODO: Maybe we should just force sch_ingress to be compiled in
3508 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3509 * a compare and 2 stores extra right now if we dont have it on
3510 * but have CONFIG_NET_CLS_ACT
3511 * NOTE: This doesn't stop any functionality; if you dont have
3512 * the ingress scheduler, you just can't add policies on ingress.
3515 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3517 struct net_device *dev = skb->dev;
3518 u32 ttl = G_TC_RTTL(skb->tc_verd);
3519 int result = TC_ACT_OK;
3522 if (unlikely(MAX_RED_LOOP < ttl++)) {
3523 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3524 skb->skb_iif, dev->ifindex);
3528 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3529 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3531 q = rcu_dereference(rxq->qdisc);
3532 if (q != &noop_qdisc) {
3533 spin_lock(qdisc_lock(q));
3534 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3535 result = qdisc_enqueue_root(skb, q);
3536 spin_unlock(qdisc_lock(q));
3542 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3543 struct packet_type **pt_prev,
3544 int *ret, struct net_device *orig_dev)
3546 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3548 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3552 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3556 switch (ing_filter(skb, rxq)) {
3570 * netdev_rx_handler_register - register receive handler
3571 * @dev: device to register a handler for
3572 * @rx_handler: receive handler to register
3573 * @rx_handler_data: data pointer that is used by rx handler
3575 * Register a receive handler for a device. This handler will then be
3576 * called from __netif_receive_skb. A negative errno code is returned
3579 * The caller must hold the rtnl_mutex.
3581 * For a general description of rx_handler, see enum rx_handler_result.
3583 int netdev_rx_handler_register(struct net_device *dev,
3584 rx_handler_func_t *rx_handler,
3585 void *rx_handler_data)
3589 if (dev->rx_handler)
3592 /* Note: rx_handler_data must be set before rx_handler */
3593 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3594 rcu_assign_pointer(dev->rx_handler, rx_handler);
3598 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3601 * netdev_rx_handler_unregister - unregister receive handler
3602 * @dev: device to unregister a handler from
3604 * Unregister a receive handler from a device.
3606 * The caller must hold the rtnl_mutex.
3608 void netdev_rx_handler_unregister(struct net_device *dev)
3612 RCU_INIT_POINTER(dev->rx_handler, NULL);
3613 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3614 * section has a guarantee to see a non NULL rx_handler_data
3618 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3620 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3623 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3624 * the special handling of PFMEMALLOC skbs.
3626 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3628 switch (skb->protocol) {
3629 case htons(ETH_P_ARP):
3630 case htons(ETH_P_IP):
3631 case htons(ETH_P_IPV6):
3632 case htons(ETH_P_8021Q):
3633 case htons(ETH_P_8021AD):
3640 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3642 struct packet_type *ptype, *pt_prev;
3643 rx_handler_func_t *rx_handler;
3644 struct net_device *orig_dev;
3645 bool deliver_exact = false;
3646 int ret = NET_RX_DROP;
3649 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3651 trace_netif_receive_skb(skb);
3653 orig_dev = skb->dev;
3655 skb_reset_network_header(skb);
3656 if (!skb_transport_header_was_set(skb))
3657 skb_reset_transport_header(skb);
3658 skb_reset_mac_len(skb);
3665 skb->skb_iif = skb->dev->ifindex;
3667 __this_cpu_inc(softnet_data.processed);
3669 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3670 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3671 skb = skb_vlan_untag(skb);
3676 #ifdef CONFIG_NET_CLS_ACT
3677 if (skb->tc_verd & TC_NCLS) {
3678 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3686 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3688 ret = deliver_skb(skb, pt_prev, orig_dev);
3692 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3694 ret = deliver_skb(skb, pt_prev, orig_dev);
3699 #ifdef CONFIG_NET_CLS_ACT
3700 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3706 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3709 if (skb_vlan_tag_present(skb)) {
3711 ret = deliver_skb(skb, pt_prev, orig_dev);
3714 if (vlan_do_receive(&skb))
3716 else if (unlikely(!skb))
3720 rx_handler = rcu_dereference(skb->dev->rx_handler);
3723 ret = deliver_skb(skb, pt_prev, orig_dev);
3726 switch (rx_handler(&skb)) {
3727 case RX_HANDLER_CONSUMED:
3728 ret = NET_RX_SUCCESS;
3730 case RX_HANDLER_ANOTHER:
3732 case RX_HANDLER_EXACT:
3733 deliver_exact = true;
3734 case RX_HANDLER_PASS:
3741 if (unlikely(skb_vlan_tag_present(skb))) {
3742 if (skb_vlan_tag_get_id(skb))
3743 skb->pkt_type = PACKET_OTHERHOST;
3744 /* Note: we might in the future use prio bits
3745 * and set skb->priority like in vlan_do_receive()
3746 * For the time being, just ignore Priority Code Point
3751 type = skb->protocol;
3753 /* deliver only exact match when indicated */
3754 if (likely(!deliver_exact)) {
3755 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3756 &ptype_base[ntohs(type) &
3760 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3761 &orig_dev->ptype_specific);
3763 if (unlikely(skb->dev != orig_dev)) {
3764 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3765 &skb->dev->ptype_specific);
3769 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3772 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3775 atomic_long_inc(&skb->dev->rx_dropped);
3777 /* Jamal, now you will not able to escape explaining
3778 * me how you were going to use this. :-)
3788 static int __netif_receive_skb(struct sk_buff *skb)
3792 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3793 unsigned long pflags = current->flags;
3796 * PFMEMALLOC skbs are special, they should
3797 * - be delivered to SOCK_MEMALLOC sockets only
3798 * - stay away from userspace
3799 * - have bounded memory usage
3801 * Use PF_MEMALLOC as this saves us from propagating the allocation
3802 * context down to all allocation sites.
3804 current->flags |= PF_MEMALLOC;
3805 ret = __netif_receive_skb_core(skb, true);
3806 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3808 ret = __netif_receive_skb_core(skb, false);
3813 static int netif_receive_skb_internal(struct sk_buff *skb)
3815 net_timestamp_check(netdev_tstamp_prequeue, skb);
3817 if (skb_defer_rx_timestamp(skb))
3818 return NET_RX_SUCCESS;
3821 if (static_key_false(&rps_needed)) {
3822 struct rps_dev_flow voidflow, *rflow = &voidflow;
3827 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3830 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3837 return __netif_receive_skb(skb);
3841 * netif_receive_skb - process receive buffer from network
3842 * @skb: buffer to process
3844 * netif_receive_skb() is the main receive data processing function.
3845 * It always succeeds. The buffer may be dropped during processing
3846 * for congestion control or by the protocol layers.
3848 * This function may only be called from softirq context and interrupts
3849 * should be enabled.
3851 * Return values (usually ignored):
3852 * NET_RX_SUCCESS: no congestion
3853 * NET_RX_DROP: packet was dropped
3855 int netif_receive_skb(struct sk_buff *skb)
3857 trace_netif_receive_skb_entry(skb);
3859 return netif_receive_skb_internal(skb);
3861 EXPORT_SYMBOL(netif_receive_skb);
3863 /* Network device is going away, flush any packets still pending
3864 * Called with irqs disabled.
3866 static void flush_backlog(void *arg)
3868 struct net_device *dev = arg;
3869 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3870 struct sk_buff *skb, *tmp;
3873 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3874 if (skb->dev == dev) {
3875 __skb_unlink(skb, &sd->input_pkt_queue);
3877 input_queue_head_incr(sd);
3882 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3883 if (skb->dev == dev) {
3884 __skb_unlink(skb, &sd->process_queue);
3886 input_queue_head_incr(sd);
3891 static int napi_gro_complete(struct sk_buff *skb)
3893 struct packet_offload *ptype;
3894 __be16 type = skb->protocol;
3895 struct list_head *head = &offload_base;
3898 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3900 if (NAPI_GRO_CB(skb)->count == 1) {
3901 skb_shinfo(skb)->gso_size = 0;
3906 list_for_each_entry_rcu(ptype, head, list) {
3907 if (ptype->type != type || !ptype->callbacks.gro_complete)
3910 err = ptype->callbacks.gro_complete(skb, 0);
3916 WARN_ON(&ptype->list == head);
3918 return NET_RX_SUCCESS;
3922 return netif_receive_skb_internal(skb);
3925 /* napi->gro_list contains packets ordered by age.
3926 * youngest packets at the head of it.
3927 * Complete skbs in reverse order to reduce latencies.
3929 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3931 struct sk_buff *skb, *prev = NULL;
3933 /* scan list and build reverse chain */
3934 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3939 for (skb = prev; skb; skb = prev) {
3942 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3946 napi_gro_complete(skb);
3950 napi->gro_list = NULL;
3952 EXPORT_SYMBOL(napi_gro_flush);
3954 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3957 unsigned int maclen = skb->dev->hard_header_len;
3958 u32 hash = skb_get_hash_raw(skb);
3960 for (p = napi->gro_list; p; p = p->next) {
3961 unsigned long diffs;
3963 NAPI_GRO_CB(p)->flush = 0;
3965 if (hash != skb_get_hash_raw(p)) {
3966 NAPI_GRO_CB(p)->same_flow = 0;
3970 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3971 diffs |= p->vlan_tci ^ skb->vlan_tci;
3972 if (maclen == ETH_HLEN)
3973 diffs |= compare_ether_header(skb_mac_header(p),
3974 skb_mac_header(skb));
3976 diffs = memcmp(skb_mac_header(p),
3977 skb_mac_header(skb),
3979 NAPI_GRO_CB(p)->same_flow = !diffs;
3983 static void skb_gro_reset_offset(struct sk_buff *skb)
3985 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3986 const skb_frag_t *frag0 = &pinfo->frags[0];
3988 NAPI_GRO_CB(skb)->data_offset = 0;
3989 NAPI_GRO_CB(skb)->frag0 = NULL;
3990 NAPI_GRO_CB(skb)->frag0_len = 0;
3992 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3994 !PageHighMem(skb_frag_page(frag0))) {
3995 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3996 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4000 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4002 struct skb_shared_info *pinfo = skb_shinfo(skb);
4004 BUG_ON(skb->end - skb->tail < grow);
4006 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4008 skb->data_len -= grow;
4011 pinfo->frags[0].page_offset += grow;
4012 skb_frag_size_sub(&pinfo->frags[0], grow);
4014 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4015 skb_frag_unref(skb, 0);
4016 memmove(pinfo->frags, pinfo->frags + 1,
4017 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4021 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4023 struct sk_buff **pp = NULL;
4024 struct packet_offload *ptype;
4025 __be16 type = skb->protocol;
4026 struct list_head *head = &offload_base;
4028 enum gro_result ret;
4031 if (!(skb->dev->features & NETIF_F_GRO))
4034 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4037 gro_list_prepare(napi, skb);
4040 list_for_each_entry_rcu(ptype, head, list) {
4041 if (ptype->type != type || !ptype->callbacks.gro_receive)
4044 skb_set_network_header(skb, skb_gro_offset(skb));
4045 skb_reset_mac_len(skb);
4046 NAPI_GRO_CB(skb)->same_flow = 0;
4047 NAPI_GRO_CB(skb)->flush = 0;
4048 NAPI_GRO_CB(skb)->free = 0;
4049 NAPI_GRO_CB(skb)->udp_mark = 0;
4051 /* Setup for GRO checksum validation */
4052 switch (skb->ip_summed) {
4053 case CHECKSUM_COMPLETE:
4054 NAPI_GRO_CB(skb)->csum = skb->csum;
4055 NAPI_GRO_CB(skb)->csum_valid = 1;
4056 NAPI_GRO_CB(skb)->csum_cnt = 0;
4058 case CHECKSUM_UNNECESSARY:
4059 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4060 NAPI_GRO_CB(skb)->csum_valid = 0;
4063 NAPI_GRO_CB(skb)->csum_cnt = 0;
4064 NAPI_GRO_CB(skb)->csum_valid = 0;
4067 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4072 if (&ptype->list == head)
4075 same_flow = NAPI_GRO_CB(skb)->same_flow;
4076 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4079 struct sk_buff *nskb = *pp;
4083 napi_gro_complete(nskb);
4090 if (NAPI_GRO_CB(skb)->flush)
4093 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4094 struct sk_buff *nskb = napi->gro_list;
4096 /* locate the end of the list to select the 'oldest' flow */
4097 while (nskb->next) {
4103 napi_gro_complete(nskb);
4107 NAPI_GRO_CB(skb)->count = 1;
4108 NAPI_GRO_CB(skb)->age = jiffies;
4109 NAPI_GRO_CB(skb)->last = skb;
4110 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4111 skb->next = napi->gro_list;
4112 napi->gro_list = skb;
4116 grow = skb_gro_offset(skb) - skb_headlen(skb);
4118 gro_pull_from_frag0(skb, grow);
4127 struct packet_offload *gro_find_receive_by_type(__be16 type)
4129 struct list_head *offload_head = &offload_base;
4130 struct packet_offload *ptype;
4132 list_for_each_entry_rcu(ptype, offload_head, list) {
4133 if (ptype->type != type || !ptype->callbacks.gro_receive)
4139 EXPORT_SYMBOL(gro_find_receive_by_type);
4141 struct packet_offload *gro_find_complete_by_type(__be16 type)
4143 struct list_head *offload_head = &offload_base;
4144 struct packet_offload *ptype;
4146 list_for_each_entry_rcu(ptype, offload_head, list) {
4147 if (ptype->type != type || !ptype->callbacks.gro_complete)
4153 EXPORT_SYMBOL(gro_find_complete_by_type);
4155 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4159 if (netif_receive_skb_internal(skb))
4167 case GRO_MERGED_FREE:
4168 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4169 kmem_cache_free(skbuff_head_cache, skb);
4182 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4184 trace_napi_gro_receive_entry(skb);
4186 skb_gro_reset_offset(skb);
4188 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4190 EXPORT_SYMBOL(napi_gro_receive);
4192 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4194 if (unlikely(skb->pfmemalloc)) {
4198 __skb_pull(skb, skb_headlen(skb));
4199 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4200 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4202 skb->dev = napi->dev;
4204 skb->encapsulation = 0;
4205 skb_shinfo(skb)->gso_type = 0;
4206 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4211 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4213 struct sk_buff *skb = napi->skb;
4216 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4221 EXPORT_SYMBOL(napi_get_frags);
4223 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4224 struct sk_buff *skb,
4230 __skb_push(skb, ETH_HLEN);
4231 skb->protocol = eth_type_trans(skb, skb->dev);
4232 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4237 case GRO_MERGED_FREE:
4238 napi_reuse_skb(napi, skb);
4248 /* Upper GRO stack assumes network header starts at gro_offset=0
4249 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4250 * We copy ethernet header into skb->data to have a common layout.
4252 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4254 struct sk_buff *skb = napi->skb;
4255 const struct ethhdr *eth;
4256 unsigned int hlen = sizeof(*eth);
4260 skb_reset_mac_header(skb);
4261 skb_gro_reset_offset(skb);
4263 eth = skb_gro_header_fast(skb, 0);
4264 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4265 eth = skb_gro_header_slow(skb, hlen, 0);
4266 if (unlikely(!eth)) {
4267 napi_reuse_skb(napi, skb);
4271 gro_pull_from_frag0(skb, hlen);
4272 NAPI_GRO_CB(skb)->frag0 += hlen;
4273 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4275 __skb_pull(skb, hlen);
4278 * This works because the only protocols we care about don't require
4280 * We'll fix it up properly in napi_frags_finish()
4282 skb->protocol = eth->h_proto;
4287 gro_result_t napi_gro_frags(struct napi_struct *napi)
4289 struct sk_buff *skb = napi_frags_skb(napi);
4294 trace_napi_gro_frags_entry(skb);
4296 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4298 EXPORT_SYMBOL(napi_gro_frags);
4300 /* Compute the checksum from gro_offset and return the folded value
4301 * after adding in any pseudo checksum.
4303 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4308 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4310 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4311 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4313 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4314 !skb->csum_complete_sw)
4315 netdev_rx_csum_fault(skb->dev);
4318 NAPI_GRO_CB(skb)->csum = wsum;
4319 NAPI_GRO_CB(skb)->csum_valid = 1;
4323 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4326 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4327 * Note: called with local irq disabled, but exits with local irq enabled.
4329 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4332 struct softnet_data *remsd = sd->rps_ipi_list;
4335 sd->rps_ipi_list = NULL;
4339 /* Send pending IPI's to kick RPS processing on remote cpus. */
4341 struct softnet_data *next = remsd->rps_ipi_next;
4343 if (cpu_online(remsd->cpu))
4344 smp_call_function_single_async(remsd->cpu,
4353 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4356 return sd->rps_ipi_list != NULL;
4362 static int process_backlog(struct napi_struct *napi, int quota)
4365 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4367 /* Check if we have pending ipi, its better to send them now,
4368 * not waiting net_rx_action() end.
4370 if (sd_has_rps_ipi_waiting(sd)) {
4371 local_irq_disable();
4372 net_rps_action_and_irq_enable(sd);
4375 napi->weight = weight_p;
4376 local_irq_disable();
4378 struct sk_buff *skb;
4380 while ((skb = __skb_dequeue(&sd->process_queue))) {
4382 __netif_receive_skb(skb);
4383 local_irq_disable();
4384 input_queue_head_incr(sd);
4385 if (++work >= quota) {
4392 if (skb_queue_empty(&sd->input_pkt_queue)) {
4394 * Inline a custom version of __napi_complete().
4395 * only current cpu owns and manipulates this napi,
4396 * and NAPI_STATE_SCHED is the only possible flag set
4398 * We can use a plain write instead of clear_bit(),
4399 * and we dont need an smp_mb() memory barrier.
4407 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4408 &sd->process_queue);
4417 * __napi_schedule - schedule for receive
4418 * @n: entry to schedule
4420 * The entry's receive function will be scheduled to run.
4421 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4423 void __napi_schedule(struct napi_struct *n)
4425 unsigned long flags;
4427 local_irq_save(flags);
4428 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4429 local_irq_restore(flags);
4431 EXPORT_SYMBOL(__napi_schedule);
4434 * __napi_schedule_irqoff - schedule for receive
4435 * @n: entry to schedule
4437 * Variant of __napi_schedule() assuming hard irqs are masked
4439 void __napi_schedule_irqoff(struct napi_struct *n)
4441 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4443 EXPORT_SYMBOL(__napi_schedule_irqoff);
4445 void __napi_complete(struct napi_struct *n)
4447 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4449 list_del_init(&n->poll_list);
4450 smp_mb__before_atomic();
4451 clear_bit(NAPI_STATE_SCHED, &n->state);
4453 EXPORT_SYMBOL(__napi_complete);
4455 void napi_complete_done(struct napi_struct *n, int work_done)
4457 unsigned long flags;
4460 * don't let napi dequeue from the cpu poll list
4461 * just in case its running on a different cpu
4463 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4467 unsigned long timeout = 0;
4470 timeout = n->dev->gro_flush_timeout;
4473 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4474 HRTIMER_MODE_REL_PINNED);
4476 napi_gro_flush(n, false);
4478 if (likely(list_empty(&n->poll_list))) {
4479 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4481 /* If n->poll_list is not empty, we need to mask irqs */
4482 local_irq_save(flags);
4484 local_irq_restore(flags);
4487 EXPORT_SYMBOL(napi_complete_done);
4489 /* must be called under rcu_read_lock(), as we dont take a reference */
4490 struct napi_struct *napi_by_id(unsigned int napi_id)
4492 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4493 struct napi_struct *napi;
4495 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4496 if (napi->napi_id == napi_id)
4501 EXPORT_SYMBOL_GPL(napi_by_id);
4503 void napi_hash_add(struct napi_struct *napi)
4505 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4507 spin_lock(&napi_hash_lock);
4509 /* 0 is not a valid id, we also skip an id that is taken
4510 * we expect both events to be extremely rare
4513 while (!napi->napi_id) {
4514 napi->napi_id = ++napi_gen_id;
4515 if (napi_by_id(napi->napi_id))
4519 hlist_add_head_rcu(&napi->napi_hash_node,
4520 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4522 spin_unlock(&napi_hash_lock);
4525 EXPORT_SYMBOL_GPL(napi_hash_add);
4527 /* Warning : caller is responsible to make sure rcu grace period
4528 * is respected before freeing memory containing @napi
4530 void napi_hash_del(struct napi_struct *napi)
4532 spin_lock(&napi_hash_lock);
4534 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4535 hlist_del_rcu(&napi->napi_hash_node);
4537 spin_unlock(&napi_hash_lock);
4539 EXPORT_SYMBOL_GPL(napi_hash_del);
4541 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4543 struct napi_struct *napi;
4545 napi = container_of(timer, struct napi_struct, timer);
4547 napi_schedule(napi);
4549 return HRTIMER_NORESTART;
4552 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4553 int (*poll)(struct napi_struct *, int), int weight)
4555 INIT_LIST_HEAD(&napi->poll_list);
4556 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4557 napi->timer.function = napi_watchdog;
4558 napi->gro_count = 0;
4559 napi->gro_list = NULL;
4562 if (weight > NAPI_POLL_WEIGHT)
4563 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4565 napi->weight = weight;
4566 list_add(&napi->dev_list, &dev->napi_list);
4568 #ifdef CONFIG_NETPOLL
4569 spin_lock_init(&napi->poll_lock);
4570 napi->poll_owner = -1;
4572 set_bit(NAPI_STATE_SCHED, &napi->state);
4574 EXPORT_SYMBOL(netif_napi_add);
4576 void napi_disable(struct napi_struct *n)
4579 set_bit(NAPI_STATE_DISABLE, &n->state);
4581 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4584 hrtimer_cancel(&n->timer);
4586 clear_bit(NAPI_STATE_DISABLE, &n->state);
4588 EXPORT_SYMBOL(napi_disable);
4590 void netif_napi_del(struct napi_struct *napi)
4592 list_del_init(&napi->dev_list);
4593 napi_free_frags(napi);
4595 kfree_skb_list(napi->gro_list);
4596 napi->gro_list = NULL;
4597 napi->gro_count = 0;
4599 EXPORT_SYMBOL(netif_napi_del);
4601 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4606 list_del_init(&n->poll_list);
4608 have = netpoll_poll_lock(n);
4612 /* This NAPI_STATE_SCHED test is for avoiding a race
4613 * with netpoll's poll_napi(). Only the entity which
4614 * obtains the lock and sees NAPI_STATE_SCHED set will
4615 * actually make the ->poll() call. Therefore we avoid
4616 * accidentally calling ->poll() when NAPI is not scheduled.
4619 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4620 work = n->poll(n, weight);
4624 WARN_ON_ONCE(work > weight);
4626 if (likely(work < weight))
4629 /* Drivers must not modify the NAPI state if they
4630 * consume the entire weight. In such cases this code
4631 * still "owns" the NAPI instance and therefore can
4632 * move the instance around on the list at-will.
4634 if (unlikely(napi_disable_pending(n))) {
4640 /* flush too old packets
4641 * If HZ < 1000, flush all packets.
4643 napi_gro_flush(n, HZ >= 1000);
4646 /* Some drivers may have called napi_schedule
4647 * prior to exhausting their budget.
4649 if (unlikely(!list_empty(&n->poll_list))) {
4650 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4651 n->dev ? n->dev->name : "backlog");
4655 list_add_tail(&n->poll_list, repoll);
4658 netpoll_poll_unlock(have);
4663 static void net_rx_action(struct softirq_action *h)
4665 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4666 unsigned long time_limit = jiffies + 2;
4667 int budget = netdev_budget;
4671 local_irq_disable();
4672 list_splice_init(&sd->poll_list, &list);
4676 struct napi_struct *n;
4678 if (list_empty(&list)) {
4679 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4684 n = list_first_entry(&list, struct napi_struct, poll_list);
4685 budget -= napi_poll(n, &repoll);
4687 /* If softirq window is exhausted then punt.
4688 * Allow this to run for 2 jiffies since which will allow
4689 * an average latency of 1.5/HZ.
4691 if (unlikely(budget <= 0 ||
4692 time_after_eq(jiffies, time_limit))) {
4698 local_irq_disable();
4700 list_splice_tail_init(&sd->poll_list, &list);
4701 list_splice_tail(&repoll, &list);
4702 list_splice(&list, &sd->poll_list);
4703 if (!list_empty(&sd->poll_list))
4704 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4706 net_rps_action_and_irq_enable(sd);
4709 struct netdev_adjacent {
4710 struct net_device *dev;
4712 /* upper master flag, there can only be one master device per list */
4715 /* counter for the number of times this device was added to us */
4718 /* private field for the users */
4721 struct list_head list;
4722 struct rcu_head rcu;
4725 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4726 struct net_device *adj_dev,
4727 struct list_head *adj_list)
4729 struct netdev_adjacent *adj;
4731 list_for_each_entry(adj, adj_list, list) {
4732 if (adj->dev == adj_dev)
4739 * netdev_has_upper_dev - Check if device is linked to an upper device
4741 * @upper_dev: upper device to check
4743 * Find out if a device is linked to specified upper device and return true
4744 * in case it is. Note that this checks only immediate upper device,
4745 * not through a complete stack of devices. The caller must hold the RTNL lock.
4747 bool netdev_has_upper_dev(struct net_device *dev,
4748 struct net_device *upper_dev)
4752 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4754 EXPORT_SYMBOL(netdev_has_upper_dev);
4757 * netdev_has_any_upper_dev - Check if device is linked to some device
4760 * Find out if a device is linked to an upper device and return true in case
4761 * it is. The caller must hold the RTNL lock.
4763 static bool netdev_has_any_upper_dev(struct net_device *dev)
4767 return !list_empty(&dev->all_adj_list.upper);
4771 * netdev_master_upper_dev_get - Get master upper device
4774 * Find a master upper device and return pointer to it or NULL in case
4775 * it's not there. The caller must hold the RTNL lock.
4777 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4779 struct netdev_adjacent *upper;
4783 if (list_empty(&dev->adj_list.upper))
4786 upper = list_first_entry(&dev->adj_list.upper,
4787 struct netdev_adjacent, list);
4788 if (likely(upper->master))
4792 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4794 void *netdev_adjacent_get_private(struct list_head *adj_list)
4796 struct netdev_adjacent *adj;
4798 adj = list_entry(adj_list, struct netdev_adjacent, list);
4800 return adj->private;
4802 EXPORT_SYMBOL(netdev_adjacent_get_private);
4805 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4807 * @iter: list_head ** of the current position
4809 * Gets the next device from the dev's upper list, starting from iter
4810 * position. The caller must hold RCU read lock.
4812 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4813 struct list_head **iter)
4815 struct netdev_adjacent *upper;
4817 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4819 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4821 if (&upper->list == &dev->adj_list.upper)
4824 *iter = &upper->list;
4828 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4831 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4833 * @iter: list_head ** of the current position
4835 * Gets the next device from the dev's upper list, starting from iter
4836 * position. The caller must hold RCU read lock.
4838 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4839 struct list_head **iter)
4841 struct netdev_adjacent *upper;
4843 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4845 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4847 if (&upper->list == &dev->all_adj_list.upper)
4850 *iter = &upper->list;
4854 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4857 * netdev_lower_get_next_private - Get the next ->private from the
4858 * lower neighbour list
4860 * @iter: list_head ** of the current position
4862 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4863 * list, starting from iter position. The caller must hold either hold the
4864 * RTNL lock or its own locking that guarantees that the neighbour lower
4865 * list will remain unchainged.
4867 void *netdev_lower_get_next_private(struct net_device *dev,
4868 struct list_head **iter)
4870 struct netdev_adjacent *lower;
4872 lower = list_entry(*iter, struct netdev_adjacent, list);
4874 if (&lower->list == &dev->adj_list.lower)
4877 *iter = lower->list.next;
4879 return lower->private;
4881 EXPORT_SYMBOL(netdev_lower_get_next_private);
4884 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4885 * lower neighbour list, RCU
4888 * @iter: list_head ** of the current position
4890 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4891 * list, starting from iter position. The caller must hold RCU read lock.
4893 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4894 struct list_head **iter)
4896 struct netdev_adjacent *lower;
4898 WARN_ON_ONCE(!rcu_read_lock_held());
4900 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4902 if (&lower->list == &dev->adj_list.lower)
4905 *iter = &lower->list;
4907 return lower->private;
4909 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4912 * netdev_lower_get_next - Get the next device from the lower neighbour
4915 * @iter: list_head ** of the current position
4917 * Gets the next netdev_adjacent from the dev's lower neighbour
4918 * list, starting from iter position. The caller must hold RTNL lock or
4919 * its own locking that guarantees that the neighbour lower
4920 * list will remain unchainged.
4922 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4924 struct netdev_adjacent *lower;
4926 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4928 if (&lower->list == &dev->adj_list.lower)
4931 *iter = &lower->list;
4935 EXPORT_SYMBOL(netdev_lower_get_next);
4938 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4939 * lower neighbour list, RCU
4943 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4944 * list. The caller must hold RCU read lock.
4946 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4948 struct netdev_adjacent *lower;
4950 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4951 struct netdev_adjacent, list);
4953 return lower->private;
4956 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4959 * netdev_master_upper_dev_get_rcu - Get master upper device
4962 * Find a master upper device and return pointer to it or NULL in case
4963 * it's not there. The caller must hold the RCU read lock.
4965 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4967 struct netdev_adjacent *upper;
4969 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4970 struct netdev_adjacent, list);
4971 if (upper && likely(upper->master))
4975 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4977 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4978 struct net_device *adj_dev,
4979 struct list_head *dev_list)
4981 char linkname[IFNAMSIZ+7];
4982 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4983 "upper_%s" : "lower_%s", adj_dev->name);
4984 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4987 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4989 struct list_head *dev_list)
4991 char linkname[IFNAMSIZ+7];
4992 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4993 "upper_%s" : "lower_%s", name);
4994 sysfs_remove_link(&(dev->dev.kobj), linkname);
4997 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4998 struct net_device *adj_dev,
4999 struct list_head *dev_list)
5001 return (dev_list == &dev->adj_list.upper ||
5002 dev_list == &dev->adj_list.lower) &&
5003 net_eq(dev_net(dev), dev_net(adj_dev));
5006 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5007 struct net_device *adj_dev,
5008 struct list_head *dev_list,
5009 void *private, bool master)
5011 struct netdev_adjacent *adj;
5014 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5021 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5026 adj->master = master;
5028 adj->private = private;
5031 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5032 adj_dev->name, dev->name, adj_dev->name);
5034 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5035 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5040 /* Ensure that master link is always the first item in list. */
5042 ret = sysfs_create_link(&(dev->dev.kobj),
5043 &(adj_dev->dev.kobj), "master");
5045 goto remove_symlinks;
5047 list_add_rcu(&adj->list, dev_list);
5049 list_add_tail_rcu(&adj->list, dev_list);
5055 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5056 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5064 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5065 struct net_device *adj_dev,
5066 struct list_head *dev_list)
5068 struct netdev_adjacent *adj;
5070 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5073 pr_err("tried to remove device %s from %s\n",
5074 dev->name, adj_dev->name);
5078 if (adj->ref_nr > 1) {
5079 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5086 sysfs_remove_link(&(dev->dev.kobj), "master");
5088 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5089 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5091 list_del_rcu(&adj->list);
5092 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5093 adj_dev->name, dev->name, adj_dev->name);
5095 kfree_rcu(adj, rcu);
5098 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5099 struct net_device *upper_dev,
5100 struct list_head *up_list,
5101 struct list_head *down_list,
5102 void *private, bool master)
5106 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5111 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5114 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5121 static int __netdev_adjacent_dev_link(struct net_device *dev,
5122 struct net_device *upper_dev)
5124 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5125 &dev->all_adj_list.upper,
5126 &upper_dev->all_adj_list.lower,
5130 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5131 struct net_device *upper_dev,
5132 struct list_head *up_list,
5133 struct list_head *down_list)
5135 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5136 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5139 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5140 struct net_device *upper_dev)
5142 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5143 &dev->all_adj_list.upper,
5144 &upper_dev->all_adj_list.lower);
5147 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5148 struct net_device *upper_dev,
5149 void *private, bool master)
5151 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5156 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5157 &dev->adj_list.upper,
5158 &upper_dev->adj_list.lower,
5161 __netdev_adjacent_dev_unlink(dev, upper_dev);
5168 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5169 struct net_device *upper_dev)
5171 __netdev_adjacent_dev_unlink(dev, upper_dev);
5172 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5173 &dev->adj_list.upper,
5174 &upper_dev->adj_list.lower);
5177 static int __netdev_upper_dev_link(struct net_device *dev,
5178 struct net_device *upper_dev, bool master,
5181 struct netdev_adjacent *i, *j, *to_i, *to_j;
5186 if (dev == upper_dev)
5189 /* To prevent loops, check if dev is not upper device to upper_dev. */
5190 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5193 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5196 if (master && netdev_master_upper_dev_get(dev))
5199 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5204 /* Now that we linked these devs, make all the upper_dev's
5205 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5206 * versa, and don't forget the devices itself. All of these
5207 * links are non-neighbours.
5209 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5210 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5211 pr_debug("Interlinking %s with %s, non-neighbour\n",
5212 i->dev->name, j->dev->name);
5213 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5219 /* add dev to every upper_dev's upper device */
5220 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5221 pr_debug("linking %s's upper device %s with %s\n",
5222 upper_dev->name, i->dev->name, dev->name);
5223 ret = __netdev_adjacent_dev_link(dev, i->dev);
5225 goto rollback_upper_mesh;
5228 /* add upper_dev to every dev's lower device */
5229 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5230 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5231 i->dev->name, upper_dev->name);
5232 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5234 goto rollback_lower_mesh;
5237 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5240 rollback_lower_mesh:
5242 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5245 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5250 rollback_upper_mesh:
5252 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5255 __netdev_adjacent_dev_unlink(dev, i->dev);
5263 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5264 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5265 if (i == to_i && j == to_j)
5267 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5273 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5279 * netdev_upper_dev_link - Add a link to the upper device
5281 * @upper_dev: new upper device
5283 * Adds a link to device which is upper to this one. The caller must hold
5284 * the RTNL lock. On a failure a negative errno code is returned.
5285 * On success the reference counts are adjusted and the function
5288 int netdev_upper_dev_link(struct net_device *dev,
5289 struct net_device *upper_dev)
5291 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5293 EXPORT_SYMBOL(netdev_upper_dev_link);
5296 * netdev_master_upper_dev_link - Add a master link to the upper device
5298 * @upper_dev: new upper device
5300 * Adds a link to device which is upper to this one. In this case, only
5301 * one master upper device can be linked, although other non-master devices
5302 * might be linked as well. The caller must hold the RTNL lock.
5303 * On a failure a negative errno code is returned. On success the reference
5304 * counts are adjusted and the function returns zero.
5306 int netdev_master_upper_dev_link(struct net_device *dev,
5307 struct net_device *upper_dev)
5309 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5311 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5313 int netdev_master_upper_dev_link_private(struct net_device *dev,
5314 struct net_device *upper_dev,
5317 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5319 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5322 * netdev_upper_dev_unlink - Removes a link to upper device
5324 * @upper_dev: new upper device
5326 * Removes a link to device which is upper to this one. The caller must hold
5329 void netdev_upper_dev_unlink(struct net_device *dev,
5330 struct net_device *upper_dev)
5332 struct netdev_adjacent *i, *j;
5335 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5337 /* Here is the tricky part. We must remove all dev's lower
5338 * devices from all upper_dev's upper devices and vice
5339 * versa, to maintain the graph relationship.
5341 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5342 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5343 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5345 /* remove also the devices itself from lower/upper device
5348 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5349 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5351 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5352 __netdev_adjacent_dev_unlink(dev, i->dev);
5354 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5356 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5359 * netdev_bonding_info_change - Dispatch event about slave change
5361 * @netdev_bonding_info: info to dispatch
5363 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5364 * The caller must hold the RTNL lock.
5366 void netdev_bonding_info_change(struct net_device *dev,
5367 struct netdev_bonding_info *bonding_info)
5369 struct netdev_notifier_bonding_info info;
5371 memcpy(&info.bonding_info, bonding_info,
5372 sizeof(struct netdev_bonding_info));
5373 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5376 EXPORT_SYMBOL(netdev_bonding_info_change);
5378 void netdev_adjacent_add_links(struct net_device *dev)
5380 struct netdev_adjacent *iter;
5382 struct net *net = dev_net(dev);
5384 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5385 if (!net_eq(net,dev_net(iter->dev)))
5387 netdev_adjacent_sysfs_add(iter->dev, dev,
5388 &iter->dev->adj_list.lower);
5389 netdev_adjacent_sysfs_add(dev, iter->dev,
5390 &dev->adj_list.upper);
5393 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5394 if (!net_eq(net,dev_net(iter->dev)))
5396 netdev_adjacent_sysfs_add(iter->dev, dev,
5397 &iter->dev->adj_list.upper);
5398 netdev_adjacent_sysfs_add(dev, iter->dev,
5399 &dev->adj_list.lower);
5403 void netdev_adjacent_del_links(struct net_device *dev)
5405 struct netdev_adjacent *iter;
5407 struct net *net = dev_net(dev);
5409 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5410 if (!net_eq(net,dev_net(iter->dev)))
5412 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5413 &iter->dev->adj_list.lower);
5414 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5415 &dev->adj_list.upper);
5418 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5419 if (!net_eq(net,dev_net(iter->dev)))
5421 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5422 &iter->dev->adj_list.upper);
5423 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5424 &dev->adj_list.lower);
5428 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5430 struct netdev_adjacent *iter;
5432 struct net *net = dev_net(dev);
5434 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5435 if (!net_eq(net,dev_net(iter->dev)))
5437 netdev_adjacent_sysfs_del(iter->dev, oldname,
5438 &iter->dev->adj_list.lower);
5439 netdev_adjacent_sysfs_add(iter->dev, dev,
5440 &iter->dev->adj_list.lower);
5443 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5444 if (!net_eq(net,dev_net(iter->dev)))
5446 netdev_adjacent_sysfs_del(iter->dev, oldname,
5447 &iter->dev->adj_list.upper);
5448 netdev_adjacent_sysfs_add(iter->dev, dev,
5449 &iter->dev->adj_list.upper);
5453 void *netdev_lower_dev_get_private(struct net_device *dev,
5454 struct net_device *lower_dev)
5456 struct netdev_adjacent *lower;
5460 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5464 return lower->private;
5466 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5469 int dev_get_nest_level(struct net_device *dev,
5470 bool (*type_check)(struct net_device *dev))
5472 struct net_device *lower = NULL;
5473 struct list_head *iter;
5479 netdev_for_each_lower_dev(dev, lower, iter) {
5480 nest = dev_get_nest_level(lower, type_check);
5481 if (max_nest < nest)
5485 if (type_check(dev))
5490 EXPORT_SYMBOL(dev_get_nest_level);
5492 static void dev_change_rx_flags(struct net_device *dev, int flags)
5494 const struct net_device_ops *ops = dev->netdev_ops;
5496 if (ops->ndo_change_rx_flags)
5497 ops->ndo_change_rx_flags(dev, flags);
5500 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5502 unsigned int old_flags = dev->flags;
5508 dev->flags |= IFF_PROMISC;
5509 dev->promiscuity += inc;
5510 if (dev->promiscuity == 0) {
5513 * If inc causes overflow, untouch promisc and return error.
5516 dev->flags &= ~IFF_PROMISC;
5518 dev->promiscuity -= inc;
5519 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5524 if (dev->flags != old_flags) {
5525 pr_info("device %s %s promiscuous mode\n",
5527 dev->flags & IFF_PROMISC ? "entered" : "left");
5528 if (audit_enabled) {
5529 current_uid_gid(&uid, &gid);
5530 audit_log(current->audit_context, GFP_ATOMIC,
5531 AUDIT_ANOM_PROMISCUOUS,
5532 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5533 dev->name, (dev->flags & IFF_PROMISC),
5534 (old_flags & IFF_PROMISC),
5535 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5536 from_kuid(&init_user_ns, uid),
5537 from_kgid(&init_user_ns, gid),
5538 audit_get_sessionid(current));
5541 dev_change_rx_flags(dev, IFF_PROMISC);
5544 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5549 * dev_set_promiscuity - update promiscuity count on a device
5553 * Add or remove promiscuity from a device. While the count in the device
5554 * remains above zero the interface remains promiscuous. Once it hits zero
5555 * the device reverts back to normal filtering operation. A negative inc
5556 * value is used to drop promiscuity on the device.
5557 * Return 0 if successful or a negative errno code on error.
5559 int dev_set_promiscuity(struct net_device *dev, int inc)
5561 unsigned int old_flags = dev->flags;
5564 err = __dev_set_promiscuity(dev, inc, true);
5567 if (dev->flags != old_flags)
5568 dev_set_rx_mode(dev);
5571 EXPORT_SYMBOL(dev_set_promiscuity);
5573 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5575 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5579 dev->flags |= IFF_ALLMULTI;
5580 dev->allmulti += inc;
5581 if (dev->allmulti == 0) {
5584 * If inc causes overflow, untouch allmulti and return error.
5587 dev->flags &= ~IFF_ALLMULTI;
5589 dev->allmulti -= inc;
5590 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5595 if (dev->flags ^ old_flags) {
5596 dev_change_rx_flags(dev, IFF_ALLMULTI);
5597 dev_set_rx_mode(dev);
5599 __dev_notify_flags(dev, old_flags,
5600 dev->gflags ^ old_gflags);
5606 * dev_set_allmulti - update allmulti count on a device
5610 * Add or remove reception of all multicast frames to a device. While the
5611 * count in the device remains above zero the interface remains listening
5612 * to all interfaces. Once it hits zero the device reverts back to normal
5613 * filtering operation. A negative @inc value is used to drop the counter
5614 * when releasing a resource needing all multicasts.
5615 * Return 0 if successful or a negative errno code on error.
5618 int dev_set_allmulti(struct net_device *dev, int inc)
5620 return __dev_set_allmulti(dev, inc, true);
5622 EXPORT_SYMBOL(dev_set_allmulti);
5625 * Upload unicast and multicast address lists to device and
5626 * configure RX filtering. When the device doesn't support unicast
5627 * filtering it is put in promiscuous mode while unicast addresses
5630 void __dev_set_rx_mode(struct net_device *dev)
5632 const struct net_device_ops *ops = dev->netdev_ops;
5634 /* dev_open will call this function so the list will stay sane. */
5635 if (!(dev->flags&IFF_UP))
5638 if (!netif_device_present(dev))
5641 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5642 /* Unicast addresses changes may only happen under the rtnl,
5643 * therefore calling __dev_set_promiscuity here is safe.
5645 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5646 __dev_set_promiscuity(dev, 1, false);
5647 dev->uc_promisc = true;
5648 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5649 __dev_set_promiscuity(dev, -1, false);
5650 dev->uc_promisc = false;
5654 if (ops->ndo_set_rx_mode)
5655 ops->ndo_set_rx_mode(dev);
5658 void dev_set_rx_mode(struct net_device *dev)
5660 netif_addr_lock_bh(dev);
5661 __dev_set_rx_mode(dev);
5662 netif_addr_unlock_bh(dev);
5666 * dev_get_flags - get flags reported to userspace
5669 * Get the combination of flag bits exported through APIs to userspace.
5671 unsigned int dev_get_flags(const struct net_device *dev)
5675 flags = (dev->flags & ~(IFF_PROMISC |
5680 (dev->gflags & (IFF_PROMISC |
5683 if (netif_running(dev)) {
5684 if (netif_oper_up(dev))
5685 flags |= IFF_RUNNING;
5686 if (netif_carrier_ok(dev))
5687 flags |= IFF_LOWER_UP;
5688 if (netif_dormant(dev))
5689 flags |= IFF_DORMANT;
5694 EXPORT_SYMBOL(dev_get_flags);
5696 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5698 unsigned int old_flags = dev->flags;
5704 * Set the flags on our device.
5707 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5708 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5710 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5714 * Load in the correct multicast list now the flags have changed.
5717 if ((old_flags ^ flags) & IFF_MULTICAST)
5718 dev_change_rx_flags(dev, IFF_MULTICAST);
5720 dev_set_rx_mode(dev);
5723 * Have we downed the interface. We handle IFF_UP ourselves
5724 * according to user attempts to set it, rather than blindly
5729 if ((old_flags ^ flags) & IFF_UP)
5730 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5732 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5733 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5734 unsigned int old_flags = dev->flags;
5736 dev->gflags ^= IFF_PROMISC;
5738 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5739 if (dev->flags != old_flags)
5740 dev_set_rx_mode(dev);
5743 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5744 is important. Some (broken) drivers set IFF_PROMISC, when
5745 IFF_ALLMULTI is requested not asking us and not reporting.
5747 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5748 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5750 dev->gflags ^= IFF_ALLMULTI;
5751 __dev_set_allmulti(dev, inc, false);
5757 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5758 unsigned int gchanges)
5760 unsigned int changes = dev->flags ^ old_flags;
5763 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5765 if (changes & IFF_UP) {
5766 if (dev->flags & IFF_UP)
5767 call_netdevice_notifiers(NETDEV_UP, dev);
5769 call_netdevice_notifiers(NETDEV_DOWN, dev);
5772 if (dev->flags & IFF_UP &&
5773 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5774 struct netdev_notifier_change_info change_info;
5776 change_info.flags_changed = changes;
5777 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5783 * dev_change_flags - change device settings
5785 * @flags: device state flags
5787 * Change settings on device based state flags. The flags are
5788 * in the userspace exported format.
5790 int dev_change_flags(struct net_device *dev, unsigned int flags)
5793 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5795 ret = __dev_change_flags(dev, flags);
5799 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5800 __dev_notify_flags(dev, old_flags, changes);
5803 EXPORT_SYMBOL(dev_change_flags);
5805 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5807 const struct net_device_ops *ops = dev->netdev_ops;
5809 if (ops->ndo_change_mtu)
5810 return ops->ndo_change_mtu(dev, new_mtu);
5817 * dev_set_mtu - Change maximum transfer unit
5819 * @new_mtu: new transfer unit
5821 * Change the maximum transfer size of the network device.
5823 int dev_set_mtu(struct net_device *dev, int new_mtu)
5827 if (new_mtu == dev->mtu)
5830 /* MTU must be positive. */
5834 if (!netif_device_present(dev))
5837 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5838 err = notifier_to_errno(err);
5842 orig_mtu = dev->mtu;
5843 err = __dev_set_mtu(dev, new_mtu);
5846 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5847 err = notifier_to_errno(err);
5849 /* setting mtu back and notifying everyone again,
5850 * so that they have a chance to revert changes.
5852 __dev_set_mtu(dev, orig_mtu);
5853 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5858 EXPORT_SYMBOL(dev_set_mtu);
5861 * dev_set_group - Change group this device belongs to
5863 * @new_group: group this device should belong to
5865 void dev_set_group(struct net_device *dev, int new_group)
5867 dev->group = new_group;
5869 EXPORT_SYMBOL(dev_set_group);
5872 * dev_set_mac_address - Change Media Access Control Address
5876 * Change the hardware (MAC) address of the device
5878 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5880 const struct net_device_ops *ops = dev->netdev_ops;
5883 if (!ops->ndo_set_mac_address)
5885 if (sa->sa_family != dev->type)
5887 if (!netif_device_present(dev))
5889 err = ops->ndo_set_mac_address(dev, sa);
5892 dev->addr_assign_type = NET_ADDR_SET;
5893 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5894 add_device_randomness(dev->dev_addr, dev->addr_len);
5897 EXPORT_SYMBOL(dev_set_mac_address);
5900 * dev_change_carrier - Change device carrier
5902 * @new_carrier: new value
5904 * Change device carrier
5906 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5908 const struct net_device_ops *ops = dev->netdev_ops;
5910 if (!ops->ndo_change_carrier)
5912 if (!netif_device_present(dev))
5914 return ops->ndo_change_carrier(dev, new_carrier);
5916 EXPORT_SYMBOL(dev_change_carrier);
5919 * dev_get_phys_port_id - Get device physical port ID
5923 * Get device physical port ID
5925 int dev_get_phys_port_id(struct net_device *dev,
5926 struct netdev_phys_item_id *ppid)
5928 const struct net_device_ops *ops = dev->netdev_ops;
5930 if (!ops->ndo_get_phys_port_id)
5932 return ops->ndo_get_phys_port_id(dev, ppid);
5934 EXPORT_SYMBOL(dev_get_phys_port_id);
5937 * dev_new_index - allocate an ifindex
5938 * @net: the applicable net namespace
5940 * Returns a suitable unique value for a new device interface
5941 * number. The caller must hold the rtnl semaphore or the
5942 * dev_base_lock to be sure it remains unique.
5944 static int dev_new_index(struct net *net)
5946 int ifindex = net->ifindex;
5950 if (!__dev_get_by_index(net, ifindex))
5951 return net->ifindex = ifindex;
5955 /* Delayed registration/unregisteration */
5956 static LIST_HEAD(net_todo_list);
5957 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5959 static void net_set_todo(struct net_device *dev)
5961 list_add_tail(&dev->todo_list, &net_todo_list);
5962 dev_net(dev)->dev_unreg_count++;
5965 static void rollback_registered_many(struct list_head *head)
5967 struct net_device *dev, *tmp;
5968 LIST_HEAD(close_head);
5970 BUG_ON(dev_boot_phase);
5973 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5974 /* Some devices call without registering
5975 * for initialization unwind. Remove those
5976 * devices and proceed with the remaining.
5978 if (dev->reg_state == NETREG_UNINITIALIZED) {
5979 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5983 list_del(&dev->unreg_list);
5986 dev->dismantle = true;
5987 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5990 /* If device is running, close it first. */
5991 list_for_each_entry(dev, head, unreg_list)
5992 list_add_tail(&dev->close_list, &close_head);
5993 dev_close_many(&close_head);
5995 list_for_each_entry(dev, head, unreg_list) {
5996 /* And unlink it from device chain. */
5997 unlist_netdevice(dev);
5999 dev->reg_state = NETREG_UNREGISTERING;
6004 list_for_each_entry(dev, head, unreg_list) {
6005 struct sk_buff *skb = NULL;
6007 /* Shutdown queueing discipline. */
6011 /* Notify protocols, that we are about to destroy
6012 this device. They should clean all the things.
6014 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6016 if (!dev->rtnl_link_ops ||
6017 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6018 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6022 * Flush the unicast and multicast chains
6027 if (dev->netdev_ops->ndo_uninit)
6028 dev->netdev_ops->ndo_uninit(dev);
6031 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6033 /* Notifier chain MUST detach us all upper devices. */
6034 WARN_ON(netdev_has_any_upper_dev(dev));
6036 /* Remove entries from kobject tree */
6037 netdev_unregister_kobject(dev);
6039 /* Remove XPS queueing entries */
6040 netif_reset_xps_queues_gt(dev, 0);
6046 list_for_each_entry(dev, head, unreg_list)
6050 static void rollback_registered(struct net_device *dev)
6054 list_add(&dev->unreg_list, &single);
6055 rollback_registered_many(&single);
6059 static netdev_features_t netdev_fix_features(struct net_device *dev,
6060 netdev_features_t features)
6062 /* Fix illegal checksum combinations */
6063 if ((features & NETIF_F_HW_CSUM) &&
6064 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6065 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6066 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6069 /* TSO requires that SG is present as well. */
6070 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6071 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6072 features &= ~NETIF_F_ALL_TSO;
6075 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6076 !(features & NETIF_F_IP_CSUM)) {
6077 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6078 features &= ~NETIF_F_TSO;
6079 features &= ~NETIF_F_TSO_ECN;
6082 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6083 !(features & NETIF_F_IPV6_CSUM)) {
6084 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6085 features &= ~NETIF_F_TSO6;
6088 /* TSO ECN requires that TSO is present as well. */
6089 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6090 features &= ~NETIF_F_TSO_ECN;
6092 /* Software GSO depends on SG. */
6093 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6094 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6095 features &= ~NETIF_F_GSO;
6098 /* UFO needs SG and checksumming */
6099 if (features & NETIF_F_UFO) {
6100 /* maybe split UFO into V4 and V6? */
6101 if (!((features & NETIF_F_GEN_CSUM) ||
6102 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6103 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6105 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6106 features &= ~NETIF_F_UFO;
6109 if (!(features & NETIF_F_SG)) {
6111 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6112 features &= ~NETIF_F_UFO;
6116 #ifdef CONFIG_NET_RX_BUSY_POLL
6117 if (dev->netdev_ops->ndo_busy_poll)
6118 features |= NETIF_F_BUSY_POLL;
6121 features &= ~NETIF_F_BUSY_POLL;
6126 int __netdev_update_features(struct net_device *dev)
6128 netdev_features_t features;
6133 features = netdev_get_wanted_features(dev);
6135 if (dev->netdev_ops->ndo_fix_features)
6136 features = dev->netdev_ops->ndo_fix_features(dev, features);
6138 /* driver might be less strict about feature dependencies */
6139 features = netdev_fix_features(dev, features);
6141 if (dev->features == features)
6144 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6145 &dev->features, &features);
6147 if (dev->netdev_ops->ndo_set_features)
6148 err = dev->netdev_ops->ndo_set_features(dev, features);
6150 if (unlikely(err < 0)) {
6152 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6153 err, &features, &dev->features);
6158 dev->features = features;
6164 * netdev_update_features - recalculate device features
6165 * @dev: the device to check
6167 * Recalculate dev->features set and send notifications if it
6168 * has changed. Should be called after driver or hardware dependent
6169 * conditions might have changed that influence the features.
6171 void netdev_update_features(struct net_device *dev)
6173 if (__netdev_update_features(dev))
6174 netdev_features_change(dev);
6176 EXPORT_SYMBOL(netdev_update_features);
6179 * netdev_change_features - recalculate device features
6180 * @dev: the device to check
6182 * Recalculate dev->features set and send notifications even
6183 * if they have not changed. Should be called instead of
6184 * netdev_update_features() if also dev->vlan_features might
6185 * have changed to allow the changes to be propagated to stacked
6188 void netdev_change_features(struct net_device *dev)
6190 __netdev_update_features(dev);
6191 netdev_features_change(dev);
6193 EXPORT_SYMBOL(netdev_change_features);
6196 * netif_stacked_transfer_operstate - transfer operstate
6197 * @rootdev: the root or lower level device to transfer state from
6198 * @dev: the device to transfer operstate to
6200 * Transfer operational state from root to device. This is normally
6201 * called when a stacking relationship exists between the root
6202 * device and the device(a leaf device).
6204 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6205 struct net_device *dev)
6207 if (rootdev->operstate == IF_OPER_DORMANT)
6208 netif_dormant_on(dev);
6210 netif_dormant_off(dev);
6212 if (netif_carrier_ok(rootdev)) {
6213 if (!netif_carrier_ok(dev))
6214 netif_carrier_on(dev);
6216 if (netif_carrier_ok(dev))
6217 netif_carrier_off(dev);
6220 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6223 static int netif_alloc_rx_queues(struct net_device *dev)
6225 unsigned int i, count = dev->num_rx_queues;
6226 struct netdev_rx_queue *rx;
6227 size_t sz = count * sizeof(*rx);
6231 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6239 for (i = 0; i < count; i++)
6245 static void netdev_init_one_queue(struct net_device *dev,
6246 struct netdev_queue *queue, void *_unused)
6248 /* Initialize queue lock */
6249 spin_lock_init(&queue->_xmit_lock);
6250 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6251 queue->xmit_lock_owner = -1;
6252 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6255 dql_init(&queue->dql, HZ);
6259 static void netif_free_tx_queues(struct net_device *dev)
6264 static int netif_alloc_netdev_queues(struct net_device *dev)
6266 unsigned int count = dev->num_tx_queues;
6267 struct netdev_queue *tx;
6268 size_t sz = count * sizeof(*tx);
6270 BUG_ON(count < 1 || count > 0xffff);
6272 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6280 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6281 spin_lock_init(&dev->tx_global_lock);
6287 * register_netdevice - register a network device
6288 * @dev: device to register
6290 * Take a completed network device structure and add it to the kernel
6291 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6292 * chain. 0 is returned on success. A negative errno code is returned
6293 * on a failure to set up the device, or if the name is a duplicate.
6295 * Callers must hold the rtnl semaphore. You may want
6296 * register_netdev() instead of this.
6299 * The locking appears insufficient to guarantee two parallel registers
6300 * will not get the same name.
6303 int register_netdevice(struct net_device *dev)
6306 struct net *net = dev_net(dev);
6308 BUG_ON(dev_boot_phase);
6313 /* When net_device's are persistent, this will be fatal. */
6314 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6317 spin_lock_init(&dev->addr_list_lock);
6318 netdev_set_addr_lockdep_class(dev);
6322 ret = dev_get_valid_name(net, dev, dev->name);
6326 /* Init, if this function is available */
6327 if (dev->netdev_ops->ndo_init) {
6328 ret = dev->netdev_ops->ndo_init(dev);
6336 if (((dev->hw_features | dev->features) &
6337 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6338 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6339 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6340 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6347 dev->ifindex = dev_new_index(net);
6348 else if (__dev_get_by_index(net, dev->ifindex))
6351 if (dev->iflink == -1)
6352 dev->iflink = dev->ifindex;
6354 /* Transfer changeable features to wanted_features and enable
6355 * software offloads (GSO and GRO).
6357 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6358 dev->features |= NETIF_F_SOFT_FEATURES;
6359 dev->wanted_features = dev->features & dev->hw_features;
6361 if (!(dev->flags & IFF_LOOPBACK)) {
6362 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6365 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6367 dev->vlan_features |= NETIF_F_HIGHDMA;
6369 /* Make NETIF_F_SG inheritable to tunnel devices.
6371 dev->hw_enc_features |= NETIF_F_SG;
6373 /* Make NETIF_F_SG inheritable to MPLS.
6375 dev->mpls_features |= NETIF_F_SG;
6377 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6378 ret = notifier_to_errno(ret);
6382 ret = netdev_register_kobject(dev);
6385 dev->reg_state = NETREG_REGISTERED;
6387 __netdev_update_features(dev);
6390 * Default initial state at registry is that the
6391 * device is present.
6394 set_bit(__LINK_STATE_PRESENT, &dev->state);
6396 linkwatch_init_dev(dev);
6398 dev_init_scheduler(dev);
6400 list_netdevice(dev);
6401 add_device_randomness(dev->dev_addr, dev->addr_len);
6403 /* If the device has permanent device address, driver should
6404 * set dev_addr and also addr_assign_type should be set to
6405 * NET_ADDR_PERM (default value).
6407 if (dev->addr_assign_type == NET_ADDR_PERM)
6408 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6410 /* Notify protocols, that a new device appeared. */
6411 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6412 ret = notifier_to_errno(ret);
6414 rollback_registered(dev);
6415 dev->reg_state = NETREG_UNREGISTERED;
6418 * Prevent userspace races by waiting until the network
6419 * device is fully setup before sending notifications.
6421 if (!dev->rtnl_link_ops ||
6422 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6423 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6429 if (dev->netdev_ops->ndo_uninit)
6430 dev->netdev_ops->ndo_uninit(dev);
6433 EXPORT_SYMBOL(register_netdevice);
6436 * init_dummy_netdev - init a dummy network device for NAPI
6437 * @dev: device to init
6439 * This takes a network device structure and initialize the minimum
6440 * amount of fields so it can be used to schedule NAPI polls without
6441 * registering a full blown interface. This is to be used by drivers
6442 * that need to tie several hardware interfaces to a single NAPI
6443 * poll scheduler due to HW limitations.
6445 int init_dummy_netdev(struct net_device *dev)
6447 /* Clear everything. Note we don't initialize spinlocks
6448 * are they aren't supposed to be taken by any of the
6449 * NAPI code and this dummy netdev is supposed to be
6450 * only ever used for NAPI polls
6452 memset(dev, 0, sizeof(struct net_device));
6454 /* make sure we BUG if trying to hit standard
6455 * register/unregister code path
6457 dev->reg_state = NETREG_DUMMY;
6459 /* NAPI wants this */
6460 INIT_LIST_HEAD(&dev->napi_list);
6462 /* a dummy interface is started by default */
6463 set_bit(__LINK_STATE_PRESENT, &dev->state);
6464 set_bit(__LINK_STATE_START, &dev->state);
6466 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6467 * because users of this 'device' dont need to change
6473 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6477 * register_netdev - register a network device
6478 * @dev: device to register
6480 * Take a completed network device structure and add it to the kernel
6481 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6482 * chain. 0 is returned on success. A negative errno code is returned
6483 * on a failure to set up the device, or if the name is a duplicate.
6485 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6486 * and expands the device name if you passed a format string to
6489 int register_netdev(struct net_device *dev)
6494 err = register_netdevice(dev);
6498 EXPORT_SYMBOL(register_netdev);
6500 int netdev_refcnt_read(const struct net_device *dev)
6504 for_each_possible_cpu(i)
6505 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6508 EXPORT_SYMBOL(netdev_refcnt_read);
6511 * netdev_wait_allrefs - wait until all references are gone.
6512 * @dev: target net_device
6514 * This is called when unregistering network devices.
6516 * Any protocol or device that holds a reference should register
6517 * for netdevice notification, and cleanup and put back the
6518 * reference if they receive an UNREGISTER event.
6519 * We can get stuck here if buggy protocols don't correctly
6522 static void netdev_wait_allrefs(struct net_device *dev)
6524 unsigned long rebroadcast_time, warning_time;
6527 linkwatch_forget_dev(dev);
6529 rebroadcast_time = warning_time = jiffies;
6530 refcnt = netdev_refcnt_read(dev);
6532 while (refcnt != 0) {
6533 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6536 /* Rebroadcast unregister notification */
6537 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6543 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6544 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6546 /* We must not have linkwatch events
6547 * pending on unregister. If this
6548 * happens, we simply run the queue
6549 * unscheduled, resulting in a noop
6552 linkwatch_run_queue();
6557 rebroadcast_time = jiffies;
6562 refcnt = netdev_refcnt_read(dev);
6564 if (time_after(jiffies, warning_time + 10 * HZ)) {
6565 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6567 warning_time = jiffies;
6576 * register_netdevice(x1);
6577 * register_netdevice(x2);
6579 * unregister_netdevice(y1);
6580 * unregister_netdevice(y2);
6586 * We are invoked by rtnl_unlock().
6587 * This allows us to deal with problems:
6588 * 1) We can delete sysfs objects which invoke hotplug
6589 * without deadlocking with linkwatch via keventd.
6590 * 2) Since we run with the RTNL semaphore not held, we can sleep
6591 * safely in order to wait for the netdev refcnt to drop to zero.
6593 * We must not return until all unregister events added during
6594 * the interval the lock was held have been completed.
6596 void netdev_run_todo(void)
6598 struct list_head list;
6600 /* Snapshot list, allow later requests */
6601 list_replace_init(&net_todo_list, &list);
6606 /* Wait for rcu callbacks to finish before next phase */
6607 if (!list_empty(&list))
6610 while (!list_empty(&list)) {
6611 struct net_device *dev
6612 = list_first_entry(&list, struct net_device, todo_list);
6613 list_del(&dev->todo_list);
6616 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6619 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6620 pr_err("network todo '%s' but state %d\n",
6621 dev->name, dev->reg_state);
6626 dev->reg_state = NETREG_UNREGISTERED;
6628 on_each_cpu(flush_backlog, dev, 1);
6630 netdev_wait_allrefs(dev);
6633 BUG_ON(netdev_refcnt_read(dev));
6634 BUG_ON(!list_empty(&dev->ptype_all));
6635 BUG_ON(!list_empty(&dev->ptype_specific));
6636 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6637 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6638 WARN_ON(dev->dn_ptr);
6640 if (dev->destructor)
6641 dev->destructor(dev);
6643 /* Report a network device has been unregistered */
6645 dev_net(dev)->dev_unreg_count--;
6647 wake_up(&netdev_unregistering_wq);
6649 /* Free network device */
6650 kobject_put(&dev->dev.kobj);
6654 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6655 * fields in the same order, with only the type differing.
6657 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6658 const struct net_device_stats *netdev_stats)
6660 #if BITS_PER_LONG == 64
6661 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6662 memcpy(stats64, netdev_stats, sizeof(*stats64));
6664 size_t i, n = sizeof(*stats64) / sizeof(u64);
6665 const unsigned long *src = (const unsigned long *)netdev_stats;
6666 u64 *dst = (u64 *)stats64;
6668 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6669 sizeof(*stats64) / sizeof(u64));
6670 for (i = 0; i < n; i++)
6674 EXPORT_SYMBOL(netdev_stats_to_stats64);
6677 * dev_get_stats - get network device statistics
6678 * @dev: device to get statistics from
6679 * @storage: place to store stats
6681 * Get network statistics from device. Return @storage.
6682 * The device driver may provide its own method by setting
6683 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6684 * otherwise the internal statistics structure is used.
6686 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6687 struct rtnl_link_stats64 *storage)
6689 const struct net_device_ops *ops = dev->netdev_ops;
6691 if (ops->ndo_get_stats64) {
6692 memset(storage, 0, sizeof(*storage));
6693 ops->ndo_get_stats64(dev, storage);
6694 } else if (ops->ndo_get_stats) {
6695 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6697 netdev_stats_to_stats64(storage, &dev->stats);
6699 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6700 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6703 EXPORT_SYMBOL(dev_get_stats);
6705 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6707 struct netdev_queue *queue = dev_ingress_queue(dev);
6709 #ifdef CONFIG_NET_CLS_ACT
6712 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6715 netdev_init_one_queue(dev, queue, NULL);
6716 queue->qdisc = &noop_qdisc;
6717 queue->qdisc_sleeping = &noop_qdisc;
6718 rcu_assign_pointer(dev->ingress_queue, queue);
6723 static const struct ethtool_ops default_ethtool_ops;
6725 void netdev_set_default_ethtool_ops(struct net_device *dev,
6726 const struct ethtool_ops *ops)
6728 if (dev->ethtool_ops == &default_ethtool_ops)
6729 dev->ethtool_ops = ops;
6731 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6733 void netdev_freemem(struct net_device *dev)
6735 char *addr = (char *)dev - dev->padded;
6741 * alloc_netdev_mqs - allocate network device
6742 * @sizeof_priv: size of private data to allocate space for
6743 * @name: device name format string
6744 * @name_assign_type: origin of device name
6745 * @setup: callback to initialize device
6746 * @txqs: the number of TX subqueues to allocate
6747 * @rxqs: the number of RX subqueues to allocate
6749 * Allocates a struct net_device with private data area for driver use
6750 * and performs basic initialization. Also allocates subqueue structs
6751 * for each queue on the device.
6753 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6754 unsigned char name_assign_type,
6755 void (*setup)(struct net_device *),
6756 unsigned int txqs, unsigned int rxqs)
6758 struct net_device *dev;
6760 struct net_device *p;
6762 BUG_ON(strlen(name) >= sizeof(dev->name));
6765 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6771 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6776 alloc_size = sizeof(struct net_device);
6778 /* ensure 32-byte alignment of private area */
6779 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6780 alloc_size += sizeof_priv;
6782 /* ensure 32-byte alignment of whole construct */
6783 alloc_size += NETDEV_ALIGN - 1;
6785 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6787 p = vzalloc(alloc_size);
6791 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6792 dev->padded = (char *)dev - (char *)p;
6794 dev->pcpu_refcnt = alloc_percpu(int);
6795 if (!dev->pcpu_refcnt)
6798 if (dev_addr_init(dev))
6804 dev_net_set(dev, &init_net);
6806 dev->gso_max_size = GSO_MAX_SIZE;
6807 dev->gso_max_segs = GSO_MAX_SEGS;
6808 dev->gso_min_segs = 0;
6810 INIT_LIST_HEAD(&dev->napi_list);
6811 INIT_LIST_HEAD(&dev->unreg_list);
6812 INIT_LIST_HEAD(&dev->close_list);
6813 INIT_LIST_HEAD(&dev->link_watch_list);
6814 INIT_LIST_HEAD(&dev->adj_list.upper);
6815 INIT_LIST_HEAD(&dev->adj_list.lower);
6816 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6817 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6818 INIT_LIST_HEAD(&dev->ptype_all);
6819 INIT_LIST_HEAD(&dev->ptype_specific);
6820 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6823 dev->num_tx_queues = txqs;
6824 dev->real_num_tx_queues = txqs;
6825 if (netif_alloc_netdev_queues(dev))
6829 dev->num_rx_queues = rxqs;
6830 dev->real_num_rx_queues = rxqs;
6831 if (netif_alloc_rx_queues(dev))
6835 strcpy(dev->name, name);
6836 dev->name_assign_type = name_assign_type;
6837 dev->group = INIT_NETDEV_GROUP;
6838 if (!dev->ethtool_ops)
6839 dev->ethtool_ops = &default_ethtool_ops;
6847 free_percpu(dev->pcpu_refcnt);
6849 netdev_freemem(dev);
6852 EXPORT_SYMBOL(alloc_netdev_mqs);
6855 * free_netdev - free network device
6858 * This function does the last stage of destroying an allocated device
6859 * interface. The reference to the device object is released.
6860 * If this is the last reference then it will be freed.
6862 void free_netdev(struct net_device *dev)
6864 struct napi_struct *p, *n;
6866 release_net(dev_net(dev));
6868 netif_free_tx_queues(dev);
6873 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6875 /* Flush device addresses */
6876 dev_addr_flush(dev);
6878 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6881 free_percpu(dev->pcpu_refcnt);
6882 dev->pcpu_refcnt = NULL;
6884 /* Compatibility with error handling in drivers */
6885 if (dev->reg_state == NETREG_UNINITIALIZED) {
6886 netdev_freemem(dev);
6890 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6891 dev->reg_state = NETREG_RELEASED;
6893 /* will free via device release */
6894 put_device(&dev->dev);
6896 EXPORT_SYMBOL(free_netdev);
6899 * synchronize_net - Synchronize with packet receive processing
6901 * Wait for packets currently being received to be done.
6902 * Does not block later packets from starting.
6904 void synchronize_net(void)
6907 if (rtnl_is_locked())
6908 synchronize_rcu_expedited();
6912 EXPORT_SYMBOL(synchronize_net);
6915 * unregister_netdevice_queue - remove device from the kernel
6919 * This function shuts down a device interface and removes it
6920 * from the kernel tables.
6921 * If head not NULL, device is queued to be unregistered later.
6923 * Callers must hold the rtnl semaphore. You may want
6924 * unregister_netdev() instead of this.
6927 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6932 list_move_tail(&dev->unreg_list, head);
6934 rollback_registered(dev);
6935 /* Finish processing unregister after unlock */
6939 EXPORT_SYMBOL(unregister_netdevice_queue);
6942 * unregister_netdevice_many - unregister many devices
6943 * @head: list of devices
6945 * Note: As most callers use a stack allocated list_head,
6946 * we force a list_del() to make sure stack wont be corrupted later.
6948 void unregister_netdevice_many(struct list_head *head)
6950 struct net_device *dev;
6952 if (!list_empty(head)) {
6953 rollback_registered_many(head);
6954 list_for_each_entry(dev, head, unreg_list)
6959 EXPORT_SYMBOL(unregister_netdevice_many);
6962 * unregister_netdev - remove device from the kernel
6965 * This function shuts down a device interface and removes it
6966 * from the kernel tables.
6968 * This is just a wrapper for unregister_netdevice that takes
6969 * the rtnl semaphore. In general you want to use this and not
6970 * unregister_netdevice.
6972 void unregister_netdev(struct net_device *dev)
6975 unregister_netdevice(dev);
6978 EXPORT_SYMBOL(unregister_netdev);
6981 * dev_change_net_namespace - move device to different nethost namespace
6983 * @net: network namespace
6984 * @pat: If not NULL name pattern to try if the current device name
6985 * is already taken in the destination network namespace.
6987 * This function shuts down a device interface and moves it
6988 * to a new network namespace. On success 0 is returned, on
6989 * a failure a netagive errno code is returned.
6991 * Callers must hold the rtnl semaphore.
6994 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7000 /* Don't allow namespace local devices to be moved. */
7002 if (dev->features & NETIF_F_NETNS_LOCAL)
7005 /* Ensure the device has been registrered */
7006 if (dev->reg_state != NETREG_REGISTERED)
7009 /* Get out if there is nothing todo */
7011 if (net_eq(dev_net(dev), net))
7014 /* Pick the destination device name, and ensure
7015 * we can use it in the destination network namespace.
7018 if (__dev_get_by_name(net, dev->name)) {
7019 /* We get here if we can't use the current device name */
7022 if (dev_get_valid_name(net, dev, pat) < 0)
7027 * And now a mini version of register_netdevice unregister_netdevice.
7030 /* If device is running close it first. */
7033 /* And unlink it from device chain */
7035 unlist_netdevice(dev);
7039 /* Shutdown queueing discipline. */
7042 /* Notify protocols, that we are about to destroy
7043 this device. They should clean all the things.
7045 Note that dev->reg_state stays at NETREG_REGISTERED.
7046 This is wanted because this way 8021q and macvlan know
7047 the device is just moving and can keep their slaves up.
7049 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7051 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7052 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7055 * Flush the unicast and multicast chains
7060 /* Send a netdev-removed uevent to the old namespace */
7061 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7062 netdev_adjacent_del_links(dev);
7064 /* Actually switch the network namespace */
7065 dev_net_set(dev, net);
7067 /* If there is an ifindex conflict assign a new one */
7068 if (__dev_get_by_index(net, dev->ifindex)) {
7069 int iflink = (dev->iflink == dev->ifindex);
7070 dev->ifindex = dev_new_index(net);
7072 dev->iflink = dev->ifindex;
7075 /* Send a netdev-add uevent to the new namespace */
7076 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7077 netdev_adjacent_add_links(dev);
7079 /* Fixup kobjects */
7080 err = device_rename(&dev->dev, dev->name);
7083 /* Add the device back in the hashes */
7084 list_netdevice(dev);
7086 /* Notify protocols, that a new device appeared. */
7087 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7090 * Prevent userspace races by waiting until the network
7091 * device is fully setup before sending notifications.
7093 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7100 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7102 static int dev_cpu_callback(struct notifier_block *nfb,
7103 unsigned long action,
7106 struct sk_buff **list_skb;
7107 struct sk_buff *skb;
7108 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7109 struct softnet_data *sd, *oldsd;
7111 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7114 local_irq_disable();
7115 cpu = smp_processor_id();
7116 sd = &per_cpu(softnet_data, cpu);
7117 oldsd = &per_cpu(softnet_data, oldcpu);
7119 /* Find end of our completion_queue. */
7120 list_skb = &sd->completion_queue;
7122 list_skb = &(*list_skb)->next;
7123 /* Append completion queue from offline CPU. */
7124 *list_skb = oldsd->completion_queue;
7125 oldsd->completion_queue = NULL;
7127 /* Append output queue from offline CPU. */
7128 if (oldsd->output_queue) {
7129 *sd->output_queue_tailp = oldsd->output_queue;
7130 sd->output_queue_tailp = oldsd->output_queue_tailp;
7131 oldsd->output_queue = NULL;
7132 oldsd->output_queue_tailp = &oldsd->output_queue;
7134 /* Append NAPI poll list from offline CPU, with one exception :
7135 * process_backlog() must be called by cpu owning percpu backlog.
7136 * We properly handle process_queue & input_pkt_queue later.
7138 while (!list_empty(&oldsd->poll_list)) {
7139 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7143 list_del_init(&napi->poll_list);
7144 if (napi->poll == process_backlog)
7147 ____napi_schedule(sd, napi);
7150 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7153 /* Process offline CPU's input_pkt_queue */
7154 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7155 netif_rx_internal(skb);
7156 input_queue_head_incr(oldsd);
7158 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7159 netif_rx_internal(skb);
7160 input_queue_head_incr(oldsd);
7168 * netdev_increment_features - increment feature set by one
7169 * @all: current feature set
7170 * @one: new feature set
7171 * @mask: mask feature set
7173 * Computes a new feature set after adding a device with feature set
7174 * @one to the master device with current feature set @all. Will not
7175 * enable anything that is off in @mask. Returns the new feature set.
7177 netdev_features_t netdev_increment_features(netdev_features_t all,
7178 netdev_features_t one, netdev_features_t mask)
7180 if (mask & NETIF_F_GEN_CSUM)
7181 mask |= NETIF_F_ALL_CSUM;
7182 mask |= NETIF_F_VLAN_CHALLENGED;
7184 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7185 all &= one | ~NETIF_F_ALL_FOR_ALL;
7187 /* If one device supports hw checksumming, set for all. */
7188 if (all & NETIF_F_GEN_CSUM)
7189 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7193 EXPORT_SYMBOL(netdev_increment_features);
7195 static struct hlist_head * __net_init netdev_create_hash(void)
7198 struct hlist_head *hash;
7200 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7202 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7203 INIT_HLIST_HEAD(&hash[i]);
7208 /* Initialize per network namespace state */
7209 static int __net_init netdev_init(struct net *net)
7211 if (net != &init_net)
7212 INIT_LIST_HEAD(&net->dev_base_head);
7214 net->dev_name_head = netdev_create_hash();
7215 if (net->dev_name_head == NULL)
7218 net->dev_index_head = netdev_create_hash();
7219 if (net->dev_index_head == NULL)
7225 kfree(net->dev_name_head);
7231 * netdev_drivername - network driver for the device
7232 * @dev: network device
7234 * Determine network driver for device.
7236 const char *netdev_drivername(const struct net_device *dev)
7238 const struct device_driver *driver;
7239 const struct device *parent;
7240 const char *empty = "";
7242 parent = dev->dev.parent;
7246 driver = parent->driver;
7247 if (driver && driver->name)
7248 return driver->name;
7252 static void __netdev_printk(const char *level, const struct net_device *dev,
7253 struct va_format *vaf)
7255 if (dev && dev->dev.parent) {
7256 dev_printk_emit(level[1] - '0',
7259 dev_driver_string(dev->dev.parent),
7260 dev_name(dev->dev.parent),
7261 netdev_name(dev), netdev_reg_state(dev),
7264 printk("%s%s%s: %pV",
7265 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7267 printk("%s(NULL net_device): %pV", level, vaf);
7271 void netdev_printk(const char *level, const struct net_device *dev,
7272 const char *format, ...)
7274 struct va_format vaf;
7277 va_start(args, format);
7282 __netdev_printk(level, dev, &vaf);
7286 EXPORT_SYMBOL(netdev_printk);
7288 #define define_netdev_printk_level(func, level) \
7289 void func(const struct net_device *dev, const char *fmt, ...) \
7291 struct va_format vaf; \
7294 va_start(args, fmt); \
7299 __netdev_printk(level, dev, &vaf); \
7303 EXPORT_SYMBOL(func);
7305 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7306 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7307 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7308 define_netdev_printk_level(netdev_err, KERN_ERR);
7309 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7310 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7311 define_netdev_printk_level(netdev_info, KERN_INFO);
7313 static void __net_exit netdev_exit(struct net *net)
7315 kfree(net->dev_name_head);
7316 kfree(net->dev_index_head);
7319 static struct pernet_operations __net_initdata netdev_net_ops = {
7320 .init = netdev_init,
7321 .exit = netdev_exit,
7324 static void __net_exit default_device_exit(struct net *net)
7326 struct net_device *dev, *aux;
7328 * Push all migratable network devices back to the
7329 * initial network namespace
7332 for_each_netdev_safe(net, dev, aux) {
7334 char fb_name[IFNAMSIZ];
7336 /* Ignore unmoveable devices (i.e. loopback) */
7337 if (dev->features & NETIF_F_NETNS_LOCAL)
7340 /* Leave virtual devices for the generic cleanup */
7341 if (dev->rtnl_link_ops)
7344 /* Push remaining network devices to init_net */
7345 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7346 err = dev_change_net_namespace(dev, &init_net, fb_name);
7348 pr_emerg("%s: failed to move %s to init_net: %d\n",
7349 __func__, dev->name, err);
7356 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7358 /* Return with the rtnl_lock held when there are no network
7359 * devices unregistering in any network namespace in net_list.
7363 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7365 add_wait_queue(&netdev_unregistering_wq, &wait);
7367 unregistering = false;
7369 list_for_each_entry(net, net_list, exit_list) {
7370 if (net->dev_unreg_count > 0) {
7371 unregistering = true;
7379 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7381 remove_wait_queue(&netdev_unregistering_wq, &wait);
7384 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7386 /* At exit all network devices most be removed from a network
7387 * namespace. Do this in the reverse order of registration.
7388 * Do this across as many network namespaces as possible to
7389 * improve batching efficiency.
7391 struct net_device *dev;
7393 LIST_HEAD(dev_kill_list);
7395 /* To prevent network device cleanup code from dereferencing
7396 * loopback devices or network devices that have been freed
7397 * wait here for all pending unregistrations to complete,
7398 * before unregistring the loopback device and allowing the
7399 * network namespace be freed.
7401 * The netdev todo list containing all network devices
7402 * unregistrations that happen in default_device_exit_batch
7403 * will run in the rtnl_unlock() at the end of
7404 * default_device_exit_batch.
7406 rtnl_lock_unregistering(net_list);
7407 list_for_each_entry(net, net_list, exit_list) {
7408 for_each_netdev_reverse(net, dev) {
7409 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7410 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7412 unregister_netdevice_queue(dev, &dev_kill_list);
7415 unregister_netdevice_many(&dev_kill_list);
7419 static struct pernet_operations __net_initdata default_device_ops = {
7420 .exit = default_device_exit,
7421 .exit_batch = default_device_exit_batch,
7425 * Initialize the DEV module. At boot time this walks the device list and
7426 * unhooks any devices that fail to initialise (normally hardware not
7427 * present) and leaves us with a valid list of present and active devices.
7432 * This is called single threaded during boot, so no need
7433 * to take the rtnl semaphore.
7435 static int __init net_dev_init(void)
7437 int i, rc = -ENOMEM;
7439 BUG_ON(!dev_boot_phase);
7441 if (dev_proc_init())
7444 if (netdev_kobject_init())
7447 INIT_LIST_HEAD(&ptype_all);
7448 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7449 INIT_LIST_HEAD(&ptype_base[i]);
7451 INIT_LIST_HEAD(&offload_base);
7453 if (register_pernet_subsys(&netdev_net_ops))
7457 * Initialise the packet receive queues.
7460 for_each_possible_cpu(i) {
7461 struct softnet_data *sd = &per_cpu(softnet_data, i);
7463 skb_queue_head_init(&sd->input_pkt_queue);
7464 skb_queue_head_init(&sd->process_queue);
7465 INIT_LIST_HEAD(&sd->poll_list);
7466 sd->output_queue_tailp = &sd->output_queue;
7468 sd->csd.func = rps_trigger_softirq;
7473 sd->backlog.poll = process_backlog;
7474 sd->backlog.weight = weight_p;
7479 /* The loopback device is special if any other network devices
7480 * is present in a network namespace the loopback device must
7481 * be present. Since we now dynamically allocate and free the
7482 * loopback device ensure this invariant is maintained by
7483 * keeping the loopback device as the first device on the
7484 * list of network devices. Ensuring the loopback devices
7485 * is the first device that appears and the last network device
7488 if (register_pernet_device(&loopback_net_ops))
7491 if (register_pernet_device(&default_device_ops))
7494 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7495 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7497 hotcpu_notifier(dev_cpu_callback, 0);
7504 subsys_initcall(net_dev_init);