2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 static DEFINE_SPINLOCK(ptype_lock);
144 static DEFINE_SPINLOCK(offload_lock);
145 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
146 struct list_head ptype_all __read_mostly; /* Taps */
147 static struct list_head offload_base __read_mostly;
150 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
153 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
155 * Writers must hold the rtnl semaphore while they loop through the
156 * dev_base_head list, and hold dev_base_lock for writing when they do the
157 * actual updates. This allows pure readers to access the list even
158 * while a writer is preparing to update it.
160 * To put it another way, dev_base_lock is held for writing only to
161 * protect against pure readers; the rtnl semaphore provides the
162 * protection against other writers.
164 * See, for example usages, register_netdevice() and
165 * unregister_netdevice(), which must be called with the rtnl
168 DEFINE_RWLOCK(dev_base_lock);
169 EXPORT_SYMBOL(dev_base_lock);
171 /* protects napi_hash addition/deletion and napi_gen_id */
172 static DEFINE_SPINLOCK(napi_hash_lock);
174 static unsigned int napi_gen_id;
175 static DEFINE_HASHTABLE(napi_hash, 8);
177 static seqcount_t devnet_rename_seq;
179 static inline void dev_base_seq_inc(struct net *net)
181 while (++net->dev_base_seq == 0);
184 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
186 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
188 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
191 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
193 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
196 static inline void rps_lock(struct softnet_data *sd)
199 spin_lock(&sd->input_pkt_queue.lock);
203 static inline void rps_unlock(struct softnet_data *sd)
206 spin_unlock(&sd->input_pkt_queue.lock);
210 /* Device list insertion */
211 static void list_netdevice(struct net_device *dev)
213 struct net *net = dev_net(dev);
217 write_lock_bh(&dev_base_lock);
218 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head_rcu(&dev->index_hlist,
221 dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
224 dev_base_seq_inc(net);
227 /* Device list removal
228 * caller must respect a RCU grace period before freeing/reusing dev
230 static void unlist_netdevice(struct net_device *dev)
234 /* Unlink dev from the device chain */
235 write_lock_bh(&dev_base_lock);
236 list_del_rcu(&dev->dev_list);
237 hlist_del_rcu(&dev->name_hlist);
238 hlist_del_rcu(&dev->index_hlist);
239 write_unlock_bh(&dev_base_lock);
241 dev_base_seq_inc(dev_net(dev));
248 static RAW_NOTIFIER_HEAD(netdev_chain);
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
255 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
256 EXPORT_PER_CPU_SYMBOL(softnet_data);
258 #ifdef CONFIG_LOCKDEP
260 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
261 * according to dev->type
263 static const unsigned short netdev_lock_type[] =
264 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
265 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
266 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
267 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
268 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
269 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
270 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
271 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
272 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
273 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
274 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
275 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
276 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
277 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
278 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
280 static const char *const netdev_lock_name[] =
281 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
282 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
283 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
284 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
285 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
286 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
287 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
288 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
289 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
290 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
291 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
292 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
293 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
294 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
295 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
297 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
300 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
304 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
305 if (netdev_lock_type[i] == dev_type)
307 /* the last key is used by default */
308 return ARRAY_SIZE(netdev_lock_type) - 1;
311 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
312 unsigned short dev_type)
316 i = netdev_lock_pos(dev_type);
317 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
318 netdev_lock_name[i]);
321 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
325 i = netdev_lock_pos(dev->type);
326 lockdep_set_class_and_name(&dev->addr_list_lock,
327 &netdev_addr_lock_key[i],
328 netdev_lock_name[i]);
331 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
332 unsigned short dev_type)
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 /*******************************************************************************
342 Protocol management and registration routines
344 *******************************************************************************/
347 * Add a protocol ID to the list. Now that the input handler is
348 * smarter we can dispense with all the messy stuff that used to be
351 * BEWARE!!! Protocol handlers, mangling input packets,
352 * MUST BE last in hash buckets and checking protocol handlers
353 * MUST start from promiscuous ptype_all chain in net_bh.
354 * It is true now, do not change it.
355 * Explanation follows: if protocol handler, mangling packet, will
356 * be the first on list, it is not able to sense, that packet
357 * is cloned and should be copied-on-write, so that it will
358 * change it and subsequent readers will get broken packet.
362 static inline struct list_head *ptype_head(const struct packet_type *pt)
364 if (pt->type == htons(ETH_P_ALL))
367 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
371 * dev_add_pack - add packet handler
372 * @pt: packet type declaration
374 * Add a protocol handler to the networking stack. The passed &packet_type
375 * is linked into kernel lists and may not be freed until it has been
376 * removed from the kernel lists.
378 * This call does not sleep therefore it can not
379 * guarantee all CPU's that are in middle of receiving packets
380 * will see the new packet type (until the next received packet).
383 void dev_add_pack(struct packet_type *pt)
385 struct list_head *head = ptype_head(pt);
387 spin_lock(&ptype_lock);
388 list_add_rcu(&pt->list, head);
389 spin_unlock(&ptype_lock);
391 EXPORT_SYMBOL(dev_add_pack);
394 * __dev_remove_pack - remove packet handler
395 * @pt: packet type declaration
397 * Remove a protocol handler that was previously added to the kernel
398 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
399 * from the kernel lists and can be freed or reused once this function
402 * The packet type might still be in use by receivers
403 * and must not be freed until after all the CPU's have gone
404 * through a quiescent state.
406 void __dev_remove_pack(struct packet_type *pt)
408 struct list_head *head = ptype_head(pt);
409 struct packet_type *pt1;
411 spin_lock(&ptype_lock);
413 list_for_each_entry(pt1, head, list) {
415 list_del_rcu(&pt->list);
420 pr_warn("dev_remove_pack: %p not found\n", pt);
422 spin_unlock(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
444 EXPORT_SYMBOL(dev_remove_pack);
448 * dev_add_offload - register offload handlers
449 * @po: protocol offload declaration
451 * Add protocol offload handlers to the networking stack. The passed
452 * &proto_offload is linked into kernel lists and may not be freed until
453 * it has been removed from the kernel lists.
455 * This call does not sleep therefore it can not
456 * guarantee all CPU's that are in middle of receiving packets
457 * will see the new offload handlers (until the next received packet).
459 void dev_add_offload(struct packet_offload *po)
461 struct list_head *head = &offload_base;
463 spin_lock(&offload_lock);
464 list_add_rcu(&po->list, head);
465 spin_unlock(&offload_lock);
467 EXPORT_SYMBOL(dev_add_offload);
470 * __dev_remove_offload - remove offload handler
471 * @po: packet offload declaration
473 * Remove a protocol offload handler that was previously added to the
474 * kernel offload handlers by dev_add_offload(). The passed &offload_type
475 * is removed from the kernel lists and can be freed or reused once this
478 * The packet type might still be in use by receivers
479 * and must not be freed until after all the CPU's have gone
480 * through a quiescent state.
482 void __dev_remove_offload(struct packet_offload *po)
484 struct list_head *head = &offload_base;
485 struct packet_offload *po1;
487 spin_lock(&offload_lock);
489 list_for_each_entry(po1, head, list) {
491 list_del_rcu(&po->list);
496 pr_warn("dev_remove_offload: %p not found\n", po);
498 spin_unlock(&offload_lock);
500 EXPORT_SYMBOL(__dev_remove_offload);
503 * dev_remove_offload - remove packet offload handler
504 * @po: packet offload declaration
506 * Remove a packet offload handler that was previously added to the kernel
507 * offload handlers by dev_add_offload(). The passed &offload_type is
508 * removed from the kernel lists and can be freed or reused once this
511 * This call sleeps to guarantee that no CPU is looking at the packet
514 void dev_remove_offload(struct packet_offload *po)
516 __dev_remove_offload(po);
520 EXPORT_SYMBOL(dev_remove_offload);
522 /******************************************************************************
524 Device Boot-time Settings Routines
526 *******************************************************************************/
528 /* Boot time configuration table */
529 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
532 * netdev_boot_setup_add - add new setup entry
533 * @name: name of the device
534 * @map: configured settings for the device
536 * Adds new setup entry to the dev_boot_setup list. The function
537 * returns 0 on error and 1 on success. This is a generic routine to
540 static int netdev_boot_setup_add(char *name, struct ifmap *map)
542 struct netdev_boot_setup *s;
546 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
547 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
548 memset(s[i].name, 0, sizeof(s[i].name));
549 strlcpy(s[i].name, name, IFNAMSIZ);
550 memcpy(&s[i].map, map, sizeof(s[i].map));
555 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
559 * netdev_boot_setup_check - check boot time settings
560 * @dev: the netdevice
562 * Check boot time settings for the device.
563 * The found settings are set for the device to be used
564 * later in the device probing.
565 * Returns 0 if no settings found, 1 if they are.
567 int netdev_boot_setup_check(struct net_device *dev)
569 struct netdev_boot_setup *s = dev_boot_setup;
572 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
573 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
574 !strcmp(dev->name, s[i].name)) {
575 dev->irq = s[i].map.irq;
576 dev->base_addr = s[i].map.base_addr;
577 dev->mem_start = s[i].map.mem_start;
578 dev->mem_end = s[i].map.mem_end;
584 EXPORT_SYMBOL(netdev_boot_setup_check);
588 * netdev_boot_base - get address from boot time settings
589 * @prefix: prefix for network device
590 * @unit: id for network device
592 * Check boot time settings for the base address of device.
593 * The found settings are set for the device to be used
594 * later in the device probing.
595 * Returns 0 if no settings found.
597 unsigned long netdev_boot_base(const char *prefix, int unit)
599 const struct netdev_boot_setup *s = dev_boot_setup;
603 sprintf(name, "%s%d", prefix, unit);
606 * If device already registered then return base of 1
607 * to indicate not to probe for this interface
609 if (__dev_get_by_name(&init_net, name))
612 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
613 if (!strcmp(name, s[i].name))
614 return s[i].map.base_addr;
619 * Saves at boot time configured settings for any netdevice.
621 int __init netdev_boot_setup(char *str)
626 str = get_options(str, ARRAY_SIZE(ints), ints);
631 memset(&map, 0, sizeof(map));
635 map.base_addr = ints[2];
637 map.mem_start = ints[3];
639 map.mem_end = ints[4];
641 /* Add new entry to the list */
642 return netdev_boot_setup_add(str, &map);
645 __setup("netdev=", netdev_boot_setup);
647 /*******************************************************************************
649 Device Interface Subroutines
651 *******************************************************************************/
654 * __dev_get_by_name - find a device by its name
655 * @net: the applicable net namespace
656 * @name: name to find
658 * Find an interface by name. Must be called under RTNL semaphore
659 * or @dev_base_lock. If the name is found a pointer to the device
660 * is returned. If the name is not found then %NULL is returned. The
661 * reference counters are not incremented so the caller must be
662 * careful with locks.
665 struct net_device *__dev_get_by_name(struct net *net, const char *name)
667 struct net_device *dev;
668 struct hlist_head *head = dev_name_hash(net, name);
670 hlist_for_each_entry(dev, head, name_hlist)
671 if (!strncmp(dev->name, name, IFNAMSIZ))
676 EXPORT_SYMBOL(__dev_get_by_name);
679 * dev_get_by_name_rcu - find a device by its name
680 * @net: the applicable net namespace
681 * @name: name to find
683 * Find an interface by name.
684 * If the name is found a pointer to the device is returned.
685 * If the name is not found then %NULL is returned.
686 * The reference counters are not incremented so the caller must be
687 * careful with locks. The caller must hold RCU lock.
690 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
692 struct net_device *dev;
693 struct hlist_head *head = dev_name_hash(net, name);
695 hlist_for_each_entry_rcu(dev, head, name_hlist)
696 if (!strncmp(dev->name, name, IFNAMSIZ))
701 EXPORT_SYMBOL(dev_get_by_name_rcu);
704 * dev_get_by_name - find a device by its name
705 * @net: the applicable net namespace
706 * @name: name to find
708 * Find an interface by name. This can be called from any
709 * context and does its own locking. The returned handle has
710 * the usage count incremented and the caller must use dev_put() to
711 * release it when it is no longer needed. %NULL is returned if no
712 * matching device is found.
715 struct net_device *dev_get_by_name(struct net *net, const char *name)
717 struct net_device *dev;
720 dev = dev_get_by_name_rcu(net, name);
726 EXPORT_SYMBOL(dev_get_by_name);
729 * __dev_get_by_index - find a device by its ifindex
730 * @net: the applicable net namespace
731 * @ifindex: index of device
733 * Search for an interface by index. Returns %NULL if the device
734 * is not found or a pointer to the device. The device has not
735 * had its reference counter increased so the caller must be careful
736 * about locking. The caller must hold either the RTNL semaphore
740 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
742 struct net_device *dev;
743 struct hlist_head *head = dev_index_hash(net, ifindex);
745 hlist_for_each_entry(dev, head, index_hlist)
746 if (dev->ifindex == ifindex)
751 EXPORT_SYMBOL(__dev_get_by_index);
754 * dev_get_by_index_rcu - find a device by its ifindex
755 * @net: the applicable net namespace
756 * @ifindex: index of device
758 * Search for an interface by index. Returns %NULL if the device
759 * is not found or a pointer to the device. The device has not
760 * had its reference counter increased so the caller must be careful
761 * about locking. The caller must hold RCU lock.
764 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
766 struct net_device *dev;
767 struct hlist_head *head = dev_index_hash(net, ifindex);
769 hlist_for_each_entry_rcu(dev, head, index_hlist)
770 if (dev->ifindex == ifindex)
775 EXPORT_SYMBOL(dev_get_by_index_rcu);
779 * dev_get_by_index - find a device by its ifindex
780 * @net: the applicable net namespace
781 * @ifindex: index of device
783 * Search for an interface by index. Returns NULL if the device
784 * is not found or a pointer to the device. The device returned has
785 * had a reference added and the pointer is safe until the user calls
786 * dev_put to indicate they have finished with it.
789 struct net_device *dev_get_by_index(struct net *net, int ifindex)
791 struct net_device *dev;
794 dev = dev_get_by_index_rcu(net, ifindex);
800 EXPORT_SYMBOL(dev_get_by_index);
803 * netdev_get_name - get a netdevice name, knowing its ifindex.
804 * @net: network namespace
805 * @name: a pointer to the buffer where the name will be stored.
806 * @ifindex: the ifindex of the interface to get the name from.
808 * The use of raw_seqcount_begin() and cond_resched() before
809 * retrying is required as we want to give the writers a chance
810 * to complete when CONFIG_PREEMPT is not set.
812 int netdev_get_name(struct net *net, char *name, int ifindex)
814 struct net_device *dev;
818 seq = raw_seqcount_begin(&devnet_rename_seq);
820 dev = dev_get_by_index_rcu(net, ifindex);
826 strcpy(name, dev->name);
828 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
837 * dev_getbyhwaddr_rcu - find a device by its hardware address
838 * @net: the applicable net namespace
839 * @type: media type of device
840 * @ha: hardware address
842 * Search for an interface by MAC address. Returns NULL if the device
843 * is not found or a pointer to the device.
844 * The caller must hold RCU or RTNL.
845 * The returned device has not had its ref count increased
846 * and the caller must therefore be careful about locking
850 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
853 struct net_device *dev;
855 for_each_netdev_rcu(net, dev)
856 if (dev->type == type &&
857 !memcmp(dev->dev_addr, ha, dev->addr_len))
862 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
864 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
866 struct net_device *dev;
869 for_each_netdev(net, dev)
870 if (dev->type == type)
875 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
877 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
879 struct net_device *dev, *ret = NULL;
882 for_each_netdev_rcu(net, dev)
883 if (dev->type == type) {
891 EXPORT_SYMBOL(dev_getfirstbyhwtype);
894 * dev_get_by_flags_rcu - find any device with given flags
895 * @net: the applicable net namespace
896 * @if_flags: IFF_* values
897 * @mask: bitmask of bits in if_flags to check
899 * Search for any interface with the given flags. Returns NULL if a device
900 * is not found or a pointer to the device. Must be called inside
901 * rcu_read_lock(), and result refcount is unchanged.
904 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
907 struct net_device *dev, *ret;
910 for_each_netdev_rcu(net, dev) {
911 if (((dev->flags ^ if_flags) & mask) == 0) {
918 EXPORT_SYMBOL(dev_get_by_flags_rcu);
921 * dev_valid_name - check if name is okay for network device
924 * Network device names need to be valid file names to
925 * to allow sysfs to work. We also disallow any kind of
928 bool dev_valid_name(const char *name)
932 if (strlen(name) >= IFNAMSIZ)
934 if (!strcmp(name, ".") || !strcmp(name, ".."))
938 if (*name == '/' || isspace(*name))
944 EXPORT_SYMBOL(dev_valid_name);
947 * __dev_alloc_name - allocate a name for a device
948 * @net: network namespace to allocate the device name in
949 * @name: name format string
950 * @buf: scratch buffer and result name string
952 * Passed a format string - eg "lt%d" it will try and find a suitable
953 * id. It scans list of devices to build up a free map, then chooses
954 * the first empty slot. The caller must hold the dev_base or rtnl lock
955 * while allocating the name and adding the device in order to avoid
957 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
958 * Returns the number of the unit assigned or a negative errno code.
961 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
965 const int max_netdevices = 8*PAGE_SIZE;
966 unsigned long *inuse;
967 struct net_device *d;
969 p = strnchr(name, IFNAMSIZ-1, '%');
972 * Verify the string as this thing may have come from
973 * the user. There must be either one "%d" and no other "%"
976 if (p[1] != 'd' || strchr(p + 2, '%'))
979 /* Use one page as a bit array of possible slots */
980 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
984 for_each_netdev(net, d) {
985 if (!sscanf(d->name, name, &i))
987 if (i < 0 || i >= max_netdevices)
990 /* avoid cases where sscanf is not exact inverse of printf */
991 snprintf(buf, IFNAMSIZ, name, i);
992 if (!strncmp(buf, d->name, IFNAMSIZ))
996 i = find_first_zero_bit(inuse, max_netdevices);
997 free_page((unsigned long) inuse);
1001 snprintf(buf, IFNAMSIZ, name, i);
1002 if (!__dev_get_by_name(net, buf))
1005 /* It is possible to run out of possible slots
1006 * when the name is long and there isn't enough space left
1007 * for the digits, or if all bits are used.
1013 * dev_alloc_name - allocate a name for a device
1015 * @name: name format string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 int dev_alloc_name(struct net_device *dev, const char *name)
1032 BUG_ON(!dev_net(dev));
1034 ret = __dev_alloc_name(net, name, buf);
1036 strlcpy(dev->name, buf, IFNAMSIZ);
1039 EXPORT_SYMBOL(dev_alloc_name);
1041 static int dev_alloc_name_ns(struct net *net,
1042 struct net_device *dev,
1048 ret = __dev_alloc_name(net, name, buf);
1050 strlcpy(dev->name, buf, IFNAMSIZ);
1054 static int dev_get_valid_name(struct net *net,
1055 struct net_device *dev,
1060 if (!dev_valid_name(name))
1063 if (strchr(name, '%'))
1064 return dev_alloc_name_ns(net, dev, name);
1065 else if (__dev_get_by_name(net, name))
1067 else if (dev->name != name)
1068 strlcpy(dev->name, name, IFNAMSIZ);
1074 * dev_change_name - change name of a device
1076 * @newname: name (or format string) must be at least IFNAMSIZ
1078 * Change name of a device, can pass format strings "eth%d".
1081 int dev_change_name(struct net_device *dev, const char *newname)
1083 char oldname[IFNAMSIZ];
1089 BUG_ON(!dev_net(dev));
1092 if (dev->flags & IFF_UP)
1095 write_seqcount_begin(&devnet_rename_seq);
1097 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1098 write_seqcount_end(&devnet_rename_seq);
1102 memcpy(oldname, dev->name, IFNAMSIZ);
1104 err = dev_get_valid_name(net, dev, newname);
1106 write_seqcount_end(&devnet_rename_seq);
1111 ret = device_rename(&dev->dev, dev->name);
1113 memcpy(dev->name, oldname, IFNAMSIZ);
1114 write_seqcount_end(&devnet_rename_seq);
1118 write_seqcount_end(&devnet_rename_seq);
1120 write_lock_bh(&dev_base_lock);
1121 hlist_del_rcu(&dev->name_hlist);
1122 write_unlock_bh(&dev_base_lock);
1126 write_lock_bh(&dev_base_lock);
1127 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1128 write_unlock_bh(&dev_base_lock);
1130 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1131 ret = notifier_to_errno(ret);
1134 /* err >= 0 after dev_alloc_name() or stores the first errno */
1137 write_seqcount_begin(&devnet_rename_seq);
1138 memcpy(dev->name, oldname, IFNAMSIZ);
1141 pr_err("%s: name change rollback failed: %d\n",
1150 * dev_set_alias - change ifalias of a device
1152 * @alias: name up to IFALIASZ
1153 * @len: limit of bytes to copy from info
1155 * Set ifalias for a device,
1157 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1163 if (len >= IFALIASZ)
1167 kfree(dev->ifalias);
1168 dev->ifalias = NULL;
1172 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1175 dev->ifalias = new_ifalias;
1177 strlcpy(dev->ifalias, alias, len+1);
1183 * netdev_features_change - device changes features
1184 * @dev: device to cause notification
1186 * Called to indicate a device has changed features.
1188 void netdev_features_change(struct net_device *dev)
1190 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1192 EXPORT_SYMBOL(netdev_features_change);
1195 * netdev_state_change - device changes state
1196 * @dev: device to cause notification
1198 * Called to indicate a device has changed state. This function calls
1199 * the notifier chains for netdev_chain and sends a NEWLINK message
1200 * to the routing socket.
1202 void netdev_state_change(struct net_device *dev)
1204 if (dev->flags & IFF_UP) {
1205 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1206 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1209 EXPORT_SYMBOL(netdev_state_change);
1212 * netdev_notify_peers - notify network peers about existence of @dev
1213 * @dev: network device
1215 * Generate traffic such that interested network peers are aware of
1216 * @dev, such as by generating a gratuitous ARP. This may be used when
1217 * a device wants to inform the rest of the network about some sort of
1218 * reconfiguration such as a failover event or virtual machine
1221 void netdev_notify_peers(struct net_device *dev)
1224 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1227 EXPORT_SYMBOL(netdev_notify_peers);
1229 static int __dev_open(struct net_device *dev)
1231 const struct net_device_ops *ops = dev->netdev_ops;
1236 if (!netif_device_present(dev))
1239 /* Block netpoll from trying to do any rx path servicing.
1240 * If we don't do this there is a chance ndo_poll_controller
1241 * or ndo_poll may be running while we open the device
1243 netpoll_rx_disable(dev);
1245 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1246 ret = notifier_to_errno(ret);
1250 set_bit(__LINK_STATE_START, &dev->state);
1252 if (ops->ndo_validate_addr)
1253 ret = ops->ndo_validate_addr(dev);
1255 if (!ret && ops->ndo_open)
1256 ret = ops->ndo_open(dev);
1258 netpoll_rx_enable(dev);
1261 clear_bit(__LINK_STATE_START, &dev->state);
1263 dev->flags |= IFF_UP;
1264 net_dmaengine_get();
1265 dev_set_rx_mode(dev);
1267 add_device_randomness(dev->dev_addr, dev->addr_len);
1274 * dev_open - prepare an interface for use.
1275 * @dev: device to open
1277 * Takes a device from down to up state. The device's private open
1278 * function is invoked and then the multicast lists are loaded. Finally
1279 * the device is moved into the up state and a %NETDEV_UP message is
1280 * sent to the netdev notifier chain.
1282 * Calling this function on an active interface is a nop. On a failure
1283 * a negative errno code is returned.
1285 int dev_open(struct net_device *dev)
1289 if (dev->flags & IFF_UP)
1292 ret = __dev_open(dev);
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1297 call_netdevice_notifiers(NETDEV_UP, dev);
1301 EXPORT_SYMBOL(dev_open);
1303 static int __dev_close_many(struct list_head *head)
1305 struct net_device *dev;
1310 list_for_each_entry(dev, head, close_list) {
1311 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1313 clear_bit(__LINK_STATE_START, &dev->state);
1315 /* Synchronize to scheduled poll. We cannot touch poll list, it
1316 * can be even on different cpu. So just clear netif_running().
1318 * dev->stop() will invoke napi_disable() on all of it's
1319 * napi_struct instances on this device.
1321 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1324 dev_deactivate_many(head);
1326 list_for_each_entry(dev, head, close_list) {
1327 const struct net_device_ops *ops = dev->netdev_ops;
1330 * Call the device specific close. This cannot fail.
1331 * Only if device is UP
1333 * We allow it to be called even after a DETACH hot-plug
1339 dev->flags &= ~IFF_UP;
1340 net_dmaengine_put();
1346 static int __dev_close(struct net_device *dev)
1351 /* Temporarily disable netpoll until the interface is down */
1352 netpoll_rx_disable(dev);
1354 list_add(&dev->close_list, &single);
1355 retval = __dev_close_many(&single);
1358 netpoll_rx_enable(dev);
1362 static int dev_close_many(struct list_head *head)
1364 struct net_device *dev, *tmp;
1366 /* Remove the devices that don't need to be closed */
1367 list_for_each_entry_safe(dev, tmp, head, close_list)
1368 if (!(dev->flags & IFF_UP))
1369 list_del_init(&dev->close_list);
1371 __dev_close_many(head);
1373 list_for_each_entry_safe(dev, tmp, head, close_list) {
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1375 call_netdevice_notifiers(NETDEV_DOWN, dev);
1376 list_del_init(&dev->close_list);
1383 * dev_close - shutdown an interface.
1384 * @dev: device to shutdown
1386 * This function moves an active device into down state. A
1387 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1388 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1391 int dev_close(struct net_device *dev)
1393 if (dev->flags & IFF_UP) {
1396 /* Block netpoll rx while the interface is going down */
1397 netpoll_rx_disable(dev);
1399 list_add(&dev->close_list, &single);
1400 dev_close_many(&single);
1403 netpoll_rx_enable(dev);
1407 EXPORT_SYMBOL(dev_close);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device *dev)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev))
1425 dev = vlan_dev_real_dev(dev);
1427 dev->wanted_features &= ~NETIF_F_LRO;
1428 netdev_update_features(dev);
1430 if (unlikely(dev->features & NETIF_F_LRO))
1431 netdev_WARN(dev, "failed to disable LRO!\n");
1433 EXPORT_SYMBOL(dev_disable_lro);
1435 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1436 struct net_device *dev)
1438 struct netdev_notifier_info info;
1440 netdev_notifier_info_init(&info, dev);
1441 return nb->notifier_call(nb, val, &info);
1444 static int dev_boot_phase = 1;
1447 * register_netdevice_notifier - register a network notifier block
1450 * Register a notifier to be called when network device events occur.
1451 * The notifier passed is linked into the kernel structures and must
1452 * not be reused until it has been unregistered. A negative errno code
1453 * is returned on a failure.
1455 * When registered all registration and up events are replayed
1456 * to the new notifier to allow device to have a race free
1457 * view of the network device list.
1460 int register_netdevice_notifier(struct notifier_block *nb)
1462 struct net_device *dev;
1463 struct net_device *last;
1468 err = raw_notifier_chain_register(&netdev_chain, nb);
1474 for_each_netdev(net, dev) {
1475 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1476 err = notifier_to_errno(err);
1480 if (!(dev->flags & IFF_UP))
1483 call_netdevice_notifier(nb, NETDEV_UP, dev);
1494 for_each_netdev(net, dev) {
1498 if (dev->flags & IFF_UP) {
1499 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1501 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1503 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1508 raw_notifier_chain_unregister(&netdev_chain, nb);
1511 EXPORT_SYMBOL(register_netdevice_notifier);
1514 * unregister_netdevice_notifier - unregister a network notifier block
1517 * Unregister a notifier previously registered by
1518 * register_netdevice_notifier(). The notifier is unlinked into the
1519 * kernel structures and may then be reused. A negative errno code
1520 * is returned on a failure.
1522 * After unregistering unregister and down device events are synthesized
1523 * for all devices on the device list to the removed notifier to remove
1524 * the need for special case cleanup code.
1527 int unregister_netdevice_notifier(struct notifier_block *nb)
1529 struct net_device *dev;
1534 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1539 for_each_netdev(net, dev) {
1540 if (dev->flags & IFF_UP) {
1541 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1543 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1545 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1552 EXPORT_SYMBOL(unregister_netdevice_notifier);
1555 * call_netdevice_notifiers_info - call all network notifier blocks
1556 * @val: value passed unmodified to notifier function
1557 * @dev: net_device pointer passed unmodified to notifier function
1558 * @info: notifier information data
1560 * Call all network notifier blocks. Parameters and return value
1561 * are as for raw_notifier_call_chain().
1564 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1565 struct netdev_notifier_info *info)
1568 netdev_notifier_info_init(info, dev);
1569 return raw_notifier_call_chain(&netdev_chain, val, info);
1571 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1574 * call_netdevice_notifiers - call all network notifier blocks
1575 * @val: value passed unmodified to notifier function
1576 * @dev: net_device pointer passed unmodified to notifier function
1578 * Call all network notifier blocks. Parameters and return value
1579 * are as for raw_notifier_call_chain().
1582 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1584 struct netdev_notifier_info info;
1586 return call_netdevice_notifiers_info(val, dev, &info);
1588 EXPORT_SYMBOL(call_netdevice_notifiers);
1590 static struct static_key netstamp_needed __read_mostly;
1591 #ifdef HAVE_JUMP_LABEL
1592 /* We are not allowed to call static_key_slow_dec() from irq context
1593 * If net_disable_timestamp() is called from irq context, defer the
1594 * static_key_slow_dec() calls.
1596 static atomic_t netstamp_needed_deferred;
1599 void net_enable_timestamp(void)
1601 #ifdef HAVE_JUMP_LABEL
1602 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1606 static_key_slow_dec(&netstamp_needed);
1610 static_key_slow_inc(&netstamp_needed);
1612 EXPORT_SYMBOL(net_enable_timestamp);
1614 void net_disable_timestamp(void)
1616 #ifdef HAVE_JUMP_LABEL
1617 if (in_interrupt()) {
1618 atomic_inc(&netstamp_needed_deferred);
1622 static_key_slow_dec(&netstamp_needed);
1624 EXPORT_SYMBOL(net_disable_timestamp);
1626 static inline void net_timestamp_set(struct sk_buff *skb)
1628 skb->tstamp.tv64 = 0;
1629 if (static_key_false(&netstamp_needed))
1630 __net_timestamp(skb);
1633 #define net_timestamp_check(COND, SKB) \
1634 if (static_key_false(&netstamp_needed)) { \
1635 if ((COND) && !(SKB)->tstamp.tv64) \
1636 __net_timestamp(SKB); \
1639 static inline bool is_skb_forwardable(struct net_device *dev,
1640 struct sk_buff *skb)
1644 if (!(dev->flags & IFF_UP))
1647 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1648 if (skb->len <= len)
1651 /* if TSO is enabled, we don't care about the length as the packet
1652 * could be forwarded without being segmented before
1654 if (skb_is_gso(skb))
1661 * dev_forward_skb - loopback an skb to another netif
1663 * @dev: destination network device
1664 * @skb: buffer to forward
1667 * NET_RX_SUCCESS (no congestion)
1668 * NET_RX_DROP (packet was dropped, but freed)
1670 * dev_forward_skb can be used for injecting an skb from the
1671 * start_xmit function of one device into the receive queue
1672 * of another device.
1674 * The receiving device may be in another namespace, so
1675 * we have to clear all information in the skb that could
1676 * impact namespace isolation.
1678 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1680 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1681 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1682 atomic_long_inc(&dev->rx_dropped);
1688 if (unlikely(!is_skb_forwardable(dev, skb))) {
1689 atomic_long_inc(&dev->rx_dropped);
1693 skb->protocol = eth_type_trans(skb, dev);
1695 /* eth_type_trans() can set pkt_type.
1696 * call skb_scrub_packet() after it to clear pkt_type _after_ calling
1699 skb_scrub_packet(skb, true);
1701 return netif_rx(skb);
1703 EXPORT_SYMBOL_GPL(dev_forward_skb);
1705 static inline int deliver_skb(struct sk_buff *skb,
1706 struct packet_type *pt_prev,
1707 struct net_device *orig_dev)
1709 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1711 atomic_inc(&skb->users);
1712 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1715 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1717 if (!ptype->af_packet_priv || !skb->sk)
1720 if (ptype->id_match)
1721 return ptype->id_match(ptype, skb->sk);
1722 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1729 * Support routine. Sends outgoing frames to any network
1730 * taps currently in use.
1733 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1735 struct packet_type *ptype;
1736 struct sk_buff *skb2 = NULL;
1737 struct packet_type *pt_prev = NULL;
1740 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1741 /* Never send packets back to the socket
1742 * they originated from - MvS (miquels@drinkel.ow.org)
1744 if ((ptype->dev == dev || !ptype->dev) &&
1745 (!skb_loop_sk(ptype, skb))) {
1747 deliver_skb(skb2, pt_prev, skb->dev);
1752 skb2 = skb_clone(skb, GFP_ATOMIC);
1756 net_timestamp_set(skb2);
1758 /* skb->nh should be correctly
1759 set by sender, so that the second statement is
1760 just protection against buggy protocols.
1762 skb_reset_mac_header(skb2);
1764 if (skb_network_header(skb2) < skb2->data ||
1765 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1766 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1767 ntohs(skb2->protocol),
1769 skb_reset_network_header(skb2);
1772 skb2->transport_header = skb2->network_header;
1773 skb2->pkt_type = PACKET_OUTGOING;
1778 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1783 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1784 * @dev: Network device
1785 * @txq: number of queues available
1787 * If real_num_tx_queues is changed the tc mappings may no longer be
1788 * valid. To resolve this verify the tc mapping remains valid and if
1789 * not NULL the mapping. With no priorities mapping to this
1790 * offset/count pair it will no longer be used. In the worst case TC0
1791 * is invalid nothing can be done so disable priority mappings. If is
1792 * expected that drivers will fix this mapping if they can before
1793 * calling netif_set_real_num_tx_queues.
1795 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1798 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1800 /* If TC0 is invalidated disable TC mapping */
1801 if (tc->offset + tc->count > txq) {
1802 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1807 /* Invalidated prio to tc mappings set to TC0 */
1808 for (i = 1; i < TC_BITMASK + 1; i++) {
1809 int q = netdev_get_prio_tc_map(dev, i);
1811 tc = &dev->tc_to_txq[q];
1812 if (tc->offset + tc->count > txq) {
1813 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1815 netdev_set_prio_tc_map(dev, i, 0);
1821 static DEFINE_MUTEX(xps_map_mutex);
1822 #define xmap_dereference(P) \
1823 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1825 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1828 struct xps_map *map = NULL;
1832 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1834 for (pos = 0; map && pos < map->len; pos++) {
1835 if (map->queues[pos] == index) {
1837 map->queues[pos] = map->queues[--map->len];
1839 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1840 kfree_rcu(map, rcu);
1850 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1852 struct xps_dev_maps *dev_maps;
1854 bool active = false;
1856 mutex_lock(&xps_map_mutex);
1857 dev_maps = xmap_dereference(dev->xps_maps);
1862 for_each_possible_cpu(cpu) {
1863 for (i = index; i < dev->num_tx_queues; i++) {
1864 if (!remove_xps_queue(dev_maps, cpu, i))
1867 if (i == dev->num_tx_queues)
1872 RCU_INIT_POINTER(dev->xps_maps, NULL);
1873 kfree_rcu(dev_maps, rcu);
1876 for (i = index; i < dev->num_tx_queues; i++)
1877 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1881 mutex_unlock(&xps_map_mutex);
1884 static struct xps_map *expand_xps_map(struct xps_map *map,
1887 struct xps_map *new_map;
1888 int alloc_len = XPS_MIN_MAP_ALLOC;
1891 for (pos = 0; map && pos < map->len; pos++) {
1892 if (map->queues[pos] != index)
1897 /* Need to add queue to this CPU's existing map */
1899 if (pos < map->alloc_len)
1902 alloc_len = map->alloc_len * 2;
1905 /* Need to allocate new map to store queue on this CPU's map */
1906 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1911 for (i = 0; i < pos; i++)
1912 new_map->queues[i] = map->queues[i];
1913 new_map->alloc_len = alloc_len;
1919 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1921 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1922 struct xps_map *map, *new_map;
1923 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1924 int cpu, numa_node_id = -2;
1925 bool active = false;
1927 mutex_lock(&xps_map_mutex);
1929 dev_maps = xmap_dereference(dev->xps_maps);
1931 /* allocate memory for queue storage */
1932 for_each_online_cpu(cpu) {
1933 if (!cpumask_test_cpu(cpu, mask))
1937 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1938 if (!new_dev_maps) {
1939 mutex_unlock(&xps_map_mutex);
1943 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1946 map = expand_xps_map(map, cpu, index);
1950 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1954 goto out_no_new_maps;
1956 for_each_possible_cpu(cpu) {
1957 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1958 /* add queue to CPU maps */
1961 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1962 while ((pos < map->len) && (map->queues[pos] != index))
1965 if (pos == map->len)
1966 map->queues[map->len++] = index;
1968 if (numa_node_id == -2)
1969 numa_node_id = cpu_to_node(cpu);
1970 else if (numa_node_id != cpu_to_node(cpu))
1973 } else if (dev_maps) {
1974 /* fill in the new device map from the old device map */
1975 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1976 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1981 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1983 /* Cleanup old maps */
1985 for_each_possible_cpu(cpu) {
1986 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1987 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1988 if (map && map != new_map)
1989 kfree_rcu(map, rcu);
1992 kfree_rcu(dev_maps, rcu);
1995 dev_maps = new_dev_maps;
1999 /* update Tx queue numa node */
2000 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2001 (numa_node_id >= 0) ? numa_node_id :
2007 /* removes queue from unused CPUs */
2008 for_each_possible_cpu(cpu) {
2009 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2012 if (remove_xps_queue(dev_maps, cpu, index))
2016 /* free map if not active */
2018 RCU_INIT_POINTER(dev->xps_maps, NULL);
2019 kfree_rcu(dev_maps, rcu);
2023 mutex_unlock(&xps_map_mutex);
2027 /* remove any maps that we added */
2028 for_each_possible_cpu(cpu) {
2029 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2030 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2032 if (new_map && new_map != map)
2036 mutex_unlock(&xps_map_mutex);
2038 kfree(new_dev_maps);
2041 EXPORT_SYMBOL(netif_set_xps_queue);
2045 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2046 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2048 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2052 if (txq < 1 || txq > dev->num_tx_queues)
2055 if (dev->reg_state == NETREG_REGISTERED ||
2056 dev->reg_state == NETREG_UNREGISTERING) {
2059 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2065 netif_setup_tc(dev, txq);
2067 if (txq < dev->real_num_tx_queues) {
2068 qdisc_reset_all_tx_gt(dev, txq);
2070 netif_reset_xps_queues_gt(dev, txq);
2075 dev->real_num_tx_queues = txq;
2078 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2082 * netif_set_real_num_rx_queues - set actual number of RX queues used
2083 * @dev: Network device
2084 * @rxq: Actual number of RX queues
2086 * This must be called either with the rtnl_lock held or before
2087 * registration of the net device. Returns 0 on success, or a
2088 * negative error code. If called before registration, it always
2091 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2095 if (rxq < 1 || rxq > dev->num_rx_queues)
2098 if (dev->reg_state == NETREG_REGISTERED) {
2101 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2107 dev->real_num_rx_queues = rxq;
2110 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2114 * netif_get_num_default_rss_queues - default number of RSS queues
2116 * This routine should set an upper limit on the number of RSS queues
2117 * used by default by multiqueue devices.
2119 int netif_get_num_default_rss_queues(void)
2121 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2123 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2125 static inline void __netif_reschedule(struct Qdisc *q)
2127 struct softnet_data *sd;
2128 unsigned long flags;
2130 local_irq_save(flags);
2131 sd = &__get_cpu_var(softnet_data);
2132 q->next_sched = NULL;
2133 *sd->output_queue_tailp = q;
2134 sd->output_queue_tailp = &q->next_sched;
2135 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2136 local_irq_restore(flags);
2139 void __netif_schedule(struct Qdisc *q)
2141 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2142 __netif_reschedule(q);
2144 EXPORT_SYMBOL(__netif_schedule);
2146 void dev_kfree_skb_irq(struct sk_buff *skb)
2148 if (atomic_dec_and_test(&skb->users)) {
2149 struct softnet_data *sd;
2150 unsigned long flags;
2152 local_irq_save(flags);
2153 sd = &__get_cpu_var(softnet_data);
2154 skb->next = sd->completion_queue;
2155 sd->completion_queue = skb;
2156 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2157 local_irq_restore(flags);
2160 EXPORT_SYMBOL(dev_kfree_skb_irq);
2162 void dev_kfree_skb_any(struct sk_buff *skb)
2164 if (in_irq() || irqs_disabled())
2165 dev_kfree_skb_irq(skb);
2169 EXPORT_SYMBOL(dev_kfree_skb_any);
2173 * netif_device_detach - mark device as removed
2174 * @dev: network device
2176 * Mark device as removed from system and therefore no longer available.
2178 void netif_device_detach(struct net_device *dev)
2180 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2181 netif_running(dev)) {
2182 netif_tx_stop_all_queues(dev);
2185 EXPORT_SYMBOL(netif_device_detach);
2188 * netif_device_attach - mark device as attached
2189 * @dev: network device
2191 * Mark device as attached from system and restart if needed.
2193 void netif_device_attach(struct net_device *dev)
2195 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2196 netif_running(dev)) {
2197 netif_tx_wake_all_queues(dev);
2198 __netdev_watchdog_up(dev);
2201 EXPORT_SYMBOL(netif_device_attach);
2203 static void skb_warn_bad_offload(const struct sk_buff *skb)
2205 static const netdev_features_t null_features = 0;
2206 struct net_device *dev = skb->dev;
2207 const char *driver = "";
2209 if (!net_ratelimit())
2212 if (dev && dev->dev.parent)
2213 driver = dev_driver_string(dev->dev.parent);
2215 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2216 "gso_type=%d ip_summed=%d\n",
2217 driver, dev ? &dev->features : &null_features,
2218 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2219 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2220 skb_shinfo(skb)->gso_type, skb->ip_summed);
2224 * Invalidate hardware checksum when packet is to be mangled, and
2225 * complete checksum manually on outgoing path.
2227 int skb_checksum_help(struct sk_buff *skb)
2230 int ret = 0, offset;
2232 if (skb->ip_summed == CHECKSUM_COMPLETE)
2233 goto out_set_summed;
2235 if (unlikely(skb_shinfo(skb)->gso_size)) {
2236 skb_warn_bad_offload(skb);
2240 /* Before computing a checksum, we should make sure no frag could
2241 * be modified by an external entity : checksum could be wrong.
2243 if (skb_has_shared_frag(skb)) {
2244 ret = __skb_linearize(skb);
2249 offset = skb_checksum_start_offset(skb);
2250 BUG_ON(offset >= skb_headlen(skb));
2251 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2253 offset += skb->csum_offset;
2254 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2256 if (skb_cloned(skb) &&
2257 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2258 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2263 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2265 skb->ip_summed = CHECKSUM_NONE;
2269 EXPORT_SYMBOL(skb_checksum_help);
2271 __be16 skb_network_protocol(struct sk_buff *skb)
2273 __be16 type = skb->protocol;
2274 int vlan_depth = ETH_HLEN;
2276 /* Tunnel gso handlers can set protocol to ethernet. */
2277 if (type == htons(ETH_P_TEB)) {
2280 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2283 eth = (struct ethhdr *)skb_mac_header(skb);
2284 type = eth->h_proto;
2287 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2288 struct vlan_hdr *vh;
2290 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2293 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2294 type = vh->h_vlan_encapsulated_proto;
2295 vlan_depth += VLAN_HLEN;
2302 * skb_mac_gso_segment - mac layer segmentation handler.
2303 * @skb: buffer to segment
2304 * @features: features for the output path (see dev->features)
2306 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2307 netdev_features_t features)
2309 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2310 struct packet_offload *ptype;
2311 __be16 type = skb_network_protocol(skb);
2313 if (unlikely(!type))
2314 return ERR_PTR(-EINVAL);
2316 __skb_pull(skb, skb->mac_len);
2319 list_for_each_entry_rcu(ptype, &offload_base, list) {
2320 if (ptype->type == type && ptype->callbacks.gso_segment) {
2321 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2324 err = ptype->callbacks.gso_send_check(skb);
2325 segs = ERR_PTR(err);
2326 if (err || skb_gso_ok(skb, features))
2328 __skb_push(skb, (skb->data -
2329 skb_network_header(skb)));
2331 segs = ptype->callbacks.gso_segment(skb, features);
2337 __skb_push(skb, skb->data - skb_mac_header(skb));
2341 EXPORT_SYMBOL(skb_mac_gso_segment);
2344 /* openvswitch calls this on rx path, so we need a different check.
2346 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2349 return skb->ip_summed != CHECKSUM_PARTIAL;
2351 return skb->ip_summed == CHECKSUM_NONE;
2355 * __skb_gso_segment - Perform segmentation on skb.
2356 * @skb: buffer to segment
2357 * @features: features for the output path (see dev->features)
2358 * @tx_path: whether it is called in TX path
2360 * This function segments the given skb and returns a list of segments.
2362 * It may return NULL if the skb requires no segmentation. This is
2363 * only possible when GSO is used for verifying header integrity.
2365 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2366 netdev_features_t features, bool tx_path)
2368 if (unlikely(skb_needs_check(skb, tx_path))) {
2371 skb_warn_bad_offload(skb);
2373 if (skb_header_cloned(skb) &&
2374 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2375 return ERR_PTR(err);
2378 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2379 skb_reset_mac_header(skb);
2380 skb_reset_mac_len(skb);
2382 return skb_mac_gso_segment(skb, features);
2384 EXPORT_SYMBOL(__skb_gso_segment);
2386 /* Take action when hardware reception checksum errors are detected. */
2388 void netdev_rx_csum_fault(struct net_device *dev)
2390 if (net_ratelimit()) {
2391 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2395 EXPORT_SYMBOL(netdev_rx_csum_fault);
2398 /* Actually, we should eliminate this check as soon as we know, that:
2399 * 1. IOMMU is present and allows to map all the memory.
2400 * 2. No high memory really exists on this machine.
2403 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2405 #ifdef CONFIG_HIGHMEM
2407 if (!(dev->features & NETIF_F_HIGHDMA)) {
2408 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2409 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2410 if (PageHighMem(skb_frag_page(frag)))
2415 if (PCI_DMA_BUS_IS_PHYS) {
2416 struct device *pdev = dev->dev.parent;
2420 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2421 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2422 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2423 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2432 void (*destructor)(struct sk_buff *skb);
2435 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2437 static void dev_gso_skb_destructor(struct sk_buff *skb)
2439 struct dev_gso_cb *cb;
2442 struct sk_buff *nskb = skb->next;
2444 skb->next = nskb->next;
2447 } while (skb->next);
2449 cb = DEV_GSO_CB(skb);
2451 cb->destructor(skb);
2455 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2456 * @skb: buffer to segment
2457 * @features: device features as applicable to this skb
2459 * This function segments the given skb and stores the list of segments
2462 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2464 struct sk_buff *segs;
2466 segs = skb_gso_segment(skb, features);
2468 /* Verifying header integrity only. */
2473 return PTR_ERR(segs);
2476 DEV_GSO_CB(skb)->destructor = skb->destructor;
2477 skb->destructor = dev_gso_skb_destructor;
2482 static netdev_features_t harmonize_features(struct sk_buff *skb,
2483 netdev_features_t features)
2485 if (skb->ip_summed != CHECKSUM_NONE &&
2486 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2487 features &= ~NETIF_F_ALL_CSUM;
2488 } else if (illegal_highdma(skb->dev, skb)) {
2489 features &= ~NETIF_F_SG;
2495 netdev_features_t netif_skb_features(struct sk_buff *skb)
2497 __be16 protocol = skb->protocol;
2498 netdev_features_t features = skb->dev->features;
2500 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2501 features &= ~NETIF_F_GSO_MASK;
2503 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2504 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2505 protocol = veh->h_vlan_encapsulated_proto;
2506 } else if (!vlan_tx_tag_present(skb)) {
2507 return harmonize_features(skb, features);
2510 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2511 NETIF_F_HW_VLAN_STAG_TX);
2513 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2514 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2515 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2516 NETIF_F_HW_VLAN_STAG_TX;
2518 return harmonize_features(skb, features);
2520 EXPORT_SYMBOL(netif_skb_features);
2523 * Returns true if either:
2524 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2525 * 2. skb is fragmented and the device does not support SG.
2527 static inline int skb_needs_linearize(struct sk_buff *skb,
2528 netdev_features_t features)
2530 return skb_is_nonlinear(skb) &&
2531 ((skb_has_frag_list(skb) &&
2532 !(features & NETIF_F_FRAGLIST)) ||
2533 (skb_shinfo(skb)->nr_frags &&
2534 !(features & NETIF_F_SG)));
2537 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2538 struct netdev_queue *txq)
2540 const struct net_device_ops *ops = dev->netdev_ops;
2541 int rc = NETDEV_TX_OK;
2542 unsigned int skb_len;
2544 if (likely(!skb->next)) {
2545 netdev_features_t features;
2548 * If device doesn't need skb->dst, release it right now while
2549 * its hot in this cpu cache
2551 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2554 features = netif_skb_features(skb);
2556 if (vlan_tx_tag_present(skb) &&
2557 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2558 skb = __vlan_put_tag(skb, skb->vlan_proto,
2559 vlan_tx_tag_get(skb));
2566 /* If encapsulation offload request, verify we are testing
2567 * hardware encapsulation features instead of standard
2568 * features for the netdev
2570 if (skb->encapsulation)
2571 features &= dev->hw_enc_features;
2573 if (netif_needs_gso(skb, features)) {
2574 if (unlikely(dev_gso_segment(skb, features)))
2579 if (skb_needs_linearize(skb, features) &&
2580 __skb_linearize(skb))
2583 /* If packet is not checksummed and device does not
2584 * support checksumming for this protocol, complete
2585 * checksumming here.
2587 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2588 if (skb->encapsulation)
2589 skb_set_inner_transport_header(skb,
2590 skb_checksum_start_offset(skb));
2592 skb_set_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2594 if (!(features & NETIF_F_ALL_CSUM) &&
2595 skb_checksum_help(skb))
2600 if (!list_empty(&ptype_all))
2601 dev_queue_xmit_nit(skb, dev);
2604 rc = ops->ndo_start_xmit(skb, dev);
2605 trace_net_dev_xmit(skb, rc, dev, skb_len);
2606 if (rc == NETDEV_TX_OK)
2607 txq_trans_update(txq);
2613 struct sk_buff *nskb = skb->next;
2615 skb->next = nskb->next;
2618 if (!list_empty(&ptype_all))
2619 dev_queue_xmit_nit(nskb, dev);
2621 skb_len = nskb->len;
2622 rc = ops->ndo_start_xmit(nskb, dev);
2623 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2624 if (unlikely(rc != NETDEV_TX_OK)) {
2625 if (rc & ~NETDEV_TX_MASK)
2626 goto out_kfree_gso_skb;
2627 nskb->next = skb->next;
2631 txq_trans_update(txq);
2632 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2633 return NETDEV_TX_BUSY;
2634 } while (skb->next);
2637 if (likely(skb->next == NULL)) {
2638 skb->destructor = DEV_GSO_CB(skb)->destructor;
2648 static void qdisc_pkt_len_init(struct sk_buff *skb)
2650 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2652 qdisc_skb_cb(skb)->pkt_len = skb->len;
2654 /* To get more precise estimation of bytes sent on wire,
2655 * we add to pkt_len the headers size of all segments
2657 if (shinfo->gso_size) {
2658 unsigned int hdr_len;
2659 u16 gso_segs = shinfo->gso_segs;
2661 /* mac layer + network layer */
2662 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2664 /* + transport layer */
2665 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2666 hdr_len += tcp_hdrlen(skb);
2668 hdr_len += sizeof(struct udphdr);
2670 if (shinfo->gso_type & SKB_GSO_DODGY)
2671 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2674 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2678 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2679 struct net_device *dev,
2680 struct netdev_queue *txq)
2682 spinlock_t *root_lock = qdisc_lock(q);
2686 qdisc_pkt_len_init(skb);
2687 qdisc_calculate_pkt_len(skb, q);
2689 * Heuristic to force contended enqueues to serialize on a
2690 * separate lock before trying to get qdisc main lock.
2691 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2692 * and dequeue packets faster.
2694 contended = qdisc_is_running(q);
2695 if (unlikely(contended))
2696 spin_lock(&q->busylock);
2698 spin_lock(root_lock);
2699 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2702 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2703 qdisc_run_begin(q)) {
2705 * This is a work-conserving queue; there are no old skbs
2706 * waiting to be sent out; and the qdisc is not running -
2707 * xmit the skb directly.
2709 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2712 qdisc_bstats_update(q, skb);
2714 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2715 if (unlikely(contended)) {
2716 spin_unlock(&q->busylock);
2723 rc = NET_XMIT_SUCCESS;
2726 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2727 if (qdisc_run_begin(q)) {
2728 if (unlikely(contended)) {
2729 spin_unlock(&q->busylock);
2735 spin_unlock(root_lock);
2736 if (unlikely(contended))
2737 spin_unlock(&q->busylock);
2741 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2742 static void skb_update_prio(struct sk_buff *skb)
2744 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2746 if (!skb->priority && skb->sk && map) {
2747 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2749 if (prioidx < map->priomap_len)
2750 skb->priority = map->priomap[prioidx];
2754 #define skb_update_prio(skb)
2757 static DEFINE_PER_CPU(int, xmit_recursion);
2758 #define RECURSION_LIMIT 10
2761 * dev_loopback_xmit - loop back @skb
2762 * @skb: buffer to transmit
2764 int dev_loopback_xmit(struct sk_buff *skb)
2766 skb_reset_mac_header(skb);
2767 __skb_pull(skb, skb_network_offset(skb));
2768 skb->pkt_type = PACKET_LOOPBACK;
2769 skb->ip_summed = CHECKSUM_UNNECESSARY;
2770 WARN_ON(!skb_dst(skb));
2775 EXPORT_SYMBOL(dev_loopback_xmit);
2778 * dev_queue_xmit - transmit a buffer
2779 * @skb: buffer to transmit
2781 * Queue a buffer for transmission to a network device. The caller must
2782 * have set the device and priority and built the buffer before calling
2783 * this function. The function can be called from an interrupt.
2785 * A negative errno code is returned on a failure. A success does not
2786 * guarantee the frame will be transmitted as it may be dropped due
2787 * to congestion or traffic shaping.
2789 * -----------------------------------------------------------------------------------
2790 * I notice this method can also return errors from the queue disciplines,
2791 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2794 * Regardless of the return value, the skb is consumed, so it is currently
2795 * difficult to retry a send to this method. (You can bump the ref count
2796 * before sending to hold a reference for retry if you are careful.)
2798 * When calling this method, interrupts MUST be enabled. This is because
2799 * the BH enable code must have IRQs enabled so that it will not deadlock.
2802 int dev_queue_xmit(struct sk_buff *skb)
2804 struct net_device *dev = skb->dev;
2805 struct netdev_queue *txq;
2809 skb_reset_mac_header(skb);
2811 /* Disable soft irqs for various locks below. Also
2812 * stops preemption for RCU.
2816 skb_update_prio(skb);
2818 txq = netdev_pick_tx(dev, skb);
2819 q = rcu_dereference_bh(txq->qdisc);
2821 #ifdef CONFIG_NET_CLS_ACT
2822 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2824 trace_net_dev_queue(skb);
2826 rc = __dev_xmit_skb(skb, q, dev, txq);
2830 /* The device has no queue. Common case for software devices:
2831 loopback, all the sorts of tunnels...
2833 Really, it is unlikely that netif_tx_lock protection is necessary
2834 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2836 However, it is possible, that they rely on protection
2839 Check this and shot the lock. It is not prone from deadlocks.
2840 Either shot noqueue qdisc, it is even simpler 8)
2842 if (dev->flags & IFF_UP) {
2843 int cpu = smp_processor_id(); /* ok because BHs are off */
2845 if (txq->xmit_lock_owner != cpu) {
2847 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2848 goto recursion_alert;
2850 HARD_TX_LOCK(dev, txq, cpu);
2852 if (!netif_xmit_stopped(txq)) {
2853 __this_cpu_inc(xmit_recursion);
2854 rc = dev_hard_start_xmit(skb, dev, txq);
2855 __this_cpu_dec(xmit_recursion);
2856 if (dev_xmit_complete(rc)) {
2857 HARD_TX_UNLOCK(dev, txq);
2861 HARD_TX_UNLOCK(dev, txq);
2862 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2865 /* Recursion is detected! It is possible,
2869 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2875 rcu_read_unlock_bh();
2880 rcu_read_unlock_bh();
2883 EXPORT_SYMBOL(dev_queue_xmit);
2886 /*=======================================================================
2888 =======================================================================*/
2890 int netdev_max_backlog __read_mostly = 1000;
2891 EXPORT_SYMBOL(netdev_max_backlog);
2893 int netdev_tstamp_prequeue __read_mostly = 1;
2894 int netdev_budget __read_mostly = 300;
2895 int weight_p __read_mostly = 64; /* old backlog weight */
2897 /* Called with irq disabled */
2898 static inline void ____napi_schedule(struct softnet_data *sd,
2899 struct napi_struct *napi)
2901 list_add_tail(&napi->poll_list, &sd->poll_list);
2902 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2907 /* One global table that all flow-based protocols share. */
2908 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2909 EXPORT_SYMBOL(rps_sock_flow_table);
2911 struct static_key rps_needed __read_mostly;
2913 static struct rps_dev_flow *
2914 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2915 struct rps_dev_flow *rflow, u16 next_cpu)
2917 if (next_cpu != RPS_NO_CPU) {
2918 #ifdef CONFIG_RFS_ACCEL
2919 struct netdev_rx_queue *rxqueue;
2920 struct rps_dev_flow_table *flow_table;
2921 struct rps_dev_flow *old_rflow;
2926 /* Should we steer this flow to a different hardware queue? */
2927 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2928 !(dev->features & NETIF_F_NTUPLE))
2930 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2931 if (rxq_index == skb_get_rx_queue(skb))
2934 rxqueue = dev->_rx + rxq_index;
2935 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2938 flow_id = skb->rxhash & flow_table->mask;
2939 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2940 rxq_index, flow_id);
2944 rflow = &flow_table->flows[flow_id];
2946 if (old_rflow->filter == rflow->filter)
2947 old_rflow->filter = RPS_NO_FILTER;
2951 per_cpu(softnet_data, next_cpu).input_queue_head;
2954 rflow->cpu = next_cpu;
2959 * get_rps_cpu is called from netif_receive_skb and returns the target
2960 * CPU from the RPS map of the receiving queue for a given skb.
2961 * rcu_read_lock must be held on entry.
2963 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2964 struct rps_dev_flow **rflowp)
2966 struct netdev_rx_queue *rxqueue;
2967 struct rps_map *map;
2968 struct rps_dev_flow_table *flow_table;
2969 struct rps_sock_flow_table *sock_flow_table;
2973 if (skb_rx_queue_recorded(skb)) {
2974 u16 index = skb_get_rx_queue(skb);
2975 if (unlikely(index >= dev->real_num_rx_queues)) {
2976 WARN_ONCE(dev->real_num_rx_queues > 1,
2977 "%s received packet on queue %u, but number "
2978 "of RX queues is %u\n",
2979 dev->name, index, dev->real_num_rx_queues);
2982 rxqueue = dev->_rx + index;
2986 map = rcu_dereference(rxqueue->rps_map);
2988 if (map->len == 1 &&
2989 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2990 tcpu = map->cpus[0];
2991 if (cpu_online(tcpu))
2995 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2999 skb_reset_network_header(skb);
3000 if (!skb_get_rxhash(skb))
3003 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3004 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3005 if (flow_table && sock_flow_table) {
3007 struct rps_dev_flow *rflow;
3009 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3012 next_cpu = sock_flow_table->ents[skb->rxhash &
3013 sock_flow_table->mask];
3016 * If the desired CPU (where last recvmsg was done) is
3017 * different from current CPU (one in the rx-queue flow
3018 * table entry), switch if one of the following holds:
3019 * - Current CPU is unset (equal to RPS_NO_CPU).
3020 * - Current CPU is offline.
3021 * - The current CPU's queue tail has advanced beyond the
3022 * last packet that was enqueued using this table entry.
3023 * This guarantees that all previous packets for the flow
3024 * have been dequeued, thus preserving in order delivery.
3026 if (unlikely(tcpu != next_cpu) &&
3027 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3028 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3029 rflow->last_qtail)) >= 0)) {
3031 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3034 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3042 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3044 if (cpu_online(tcpu)) {
3054 #ifdef CONFIG_RFS_ACCEL
3057 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3058 * @dev: Device on which the filter was set
3059 * @rxq_index: RX queue index
3060 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3061 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3063 * Drivers that implement ndo_rx_flow_steer() should periodically call
3064 * this function for each installed filter and remove the filters for
3065 * which it returns %true.
3067 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3068 u32 flow_id, u16 filter_id)
3070 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3071 struct rps_dev_flow_table *flow_table;
3072 struct rps_dev_flow *rflow;
3077 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3078 if (flow_table && flow_id <= flow_table->mask) {
3079 rflow = &flow_table->flows[flow_id];
3080 cpu = ACCESS_ONCE(rflow->cpu);
3081 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3082 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3083 rflow->last_qtail) <
3084 (int)(10 * flow_table->mask)))
3090 EXPORT_SYMBOL(rps_may_expire_flow);
3092 #endif /* CONFIG_RFS_ACCEL */
3094 /* Called from hardirq (IPI) context */
3095 static void rps_trigger_softirq(void *data)
3097 struct softnet_data *sd = data;
3099 ____napi_schedule(sd, &sd->backlog);
3103 #endif /* CONFIG_RPS */
3106 * Check if this softnet_data structure is another cpu one
3107 * If yes, queue it to our IPI list and return 1
3110 static int rps_ipi_queued(struct softnet_data *sd)
3113 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3116 sd->rps_ipi_next = mysd->rps_ipi_list;
3117 mysd->rps_ipi_list = sd;
3119 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3122 #endif /* CONFIG_RPS */
3126 #ifdef CONFIG_NET_FLOW_LIMIT
3127 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3130 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3132 #ifdef CONFIG_NET_FLOW_LIMIT
3133 struct sd_flow_limit *fl;
3134 struct softnet_data *sd;
3135 unsigned int old_flow, new_flow;
3137 if (qlen < (netdev_max_backlog >> 1))
3140 sd = &__get_cpu_var(softnet_data);
3143 fl = rcu_dereference(sd->flow_limit);
3145 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3146 old_flow = fl->history[fl->history_head];
3147 fl->history[fl->history_head] = new_flow;
3150 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3152 if (likely(fl->buckets[old_flow]))
3153 fl->buckets[old_flow]--;
3155 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3167 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3168 * queue (may be a remote CPU queue).
3170 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3171 unsigned int *qtail)
3173 struct softnet_data *sd;
3174 unsigned long flags;
3177 sd = &per_cpu(softnet_data, cpu);
3179 local_irq_save(flags);
3182 qlen = skb_queue_len(&sd->input_pkt_queue);
3183 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3184 if (skb_queue_len(&sd->input_pkt_queue)) {
3186 __skb_queue_tail(&sd->input_pkt_queue, skb);
3187 input_queue_tail_incr_save(sd, qtail);
3189 local_irq_restore(flags);
3190 return NET_RX_SUCCESS;
3193 /* Schedule NAPI for backlog device
3194 * We can use non atomic operation since we own the queue lock
3196 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3197 if (!rps_ipi_queued(sd))
3198 ____napi_schedule(sd, &sd->backlog);
3206 local_irq_restore(flags);
3208 atomic_long_inc(&skb->dev->rx_dropped);
3214 * netif_rx - post buffer to the network code
3215 * @skb: buffer to post
3217 * This function receives a packet from a device driver and queues it for
3218 * the upper (protocol) levels to process. It always succeeds. The buffer
3219 * may be dropped during processing for congestion control or by the
3223 * NET_RX_SUCCESS (no congestion)
3224 * NET_RX_DROP (packet was dropped)
3228 int netif_rx(struct sk_buff *skb)
3232 /* if netpoll wants it, pretend we never saw it */
3233 if (netpoll_rx(skb))
3236 net_timestamp_check(netdev_tstamp_prequeue, skb);
3238 trace_netif_rx(skb);
3240 if (static_key_false(&rps_needed)) {
3241 struct rps_dev_flow voidflow, *rflow = &voidflow;
3247 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3249 cpu = smp_processor_id();
3251 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3259 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3264 EXPORT_SYMBOL(netif_rx);
3266 int netif_rx_ni(struct sk_buff *skb)
3271 err = netif_rx(skb);
3272 if (local_softirq_pending())
3278 EXPORT_SYMBOL(netif_rx_ni);
3280 static void net_tx_action(struct softirq_action *h)
3282 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3284 if (sd->completion_queue) {
3285 struct sk_buff *clist;
3287 local_irq_disable();
3288 clist = sd->completion_queue;
3289 sd->completion_queue = NULL;
3293 struct sk_buff *skb = clist;
3294 clist = clist->next;
3296 WARN_ON(atomic_read(&skb->users));
3297 trace_kfree_skb(skb, net_tx_action);
3302 if (sd->output_queue) {
3305 local_irq_disable();
3306 head = sd->output_queue;
3307 sd->output_queue = NULL;
3308 sd->output_queue_tailp = &sd->output_queue;
3312 struct Qdisc *q = head;
3313 spinlock_t *root_lock;
3315 head = head->next_sched;
3317 root_lock = qdisc_lock(q);
3318 if (spin_trylock(root_lock)) {
3319 smp_mb__before_clear_bit();
3320 clear_bit(__QDISC_STATE_SCHED,
3323 spin_unlock(root_lock);
3325 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3327 __netif_reschedule(q);
3329 smp_mb__before_clear_bit();
3330 clear_bit(__QDISC_STATE_SCHED,
3338 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3339 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3340 /* This hook is defined here for ATM LANE */
3341 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3342 unsigned char *addr) __read_mostly;
3343 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3346 #ifdef CONFIG_NET_CLS_ACT
3347 /* TODO: Maybe we should just force sch_ingress to be compiled in
3348 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3349 * a compare and 2 stores extra right now if we dont have it on
3350 * but have CONFIG_NET_CLS_ACT
3351 * NOTE: This doesn't stop any functionality; if you dont have
3352 * the ingress scheduler, you just can't add policies on ingress.
3355 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3357 struct net_device *dev = skb->dev;
3358 u32 ttl = G_TC_RTTL(skb->tc_verd);
3359 int result = TC_ACT_OK;
3362 if (unlikely(MAX_RED_LOOP < ttl++)) {
3363 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3364 skb->skb_iif, dev->ifindex);
3368 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3369 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3372 if (q != &noop_qdisc) {
3373 spin_lock(qdisc_lock(q));
3374 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3375 result = qdisc_enqueue_root(skb, q);
3376 spin_unlock(qdisc_lock(q));
3382 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3383 struct packet_type **pt_prev,
3384 int *ret, struct net_device *orig_dev)
3386 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3388 if (!rxq || rxq->qdisc == &noop_qdisc)
3392 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3396 switch (ing_filter(skb, rxq)) {
3410 * netdev_rx_handler_register - register receive handler
3411 * @dev: device to register a handler for
3412 * @rx_handler: receive handler to register
3413 * @rx_handler_data: data pointer that is used by rx handler
3415 * Register a receive hander for a device. This handler will then be
3416 * called from __netif_receive_skb. A negative errno code is returned
3419 * The caller must hold the rtnl_mutex.
3421 * For a general description of rx_handler, see enum rx_handler_result.
3423 int netdev_rx_handler_register(struct net_device *dev,
3424 rx_handler_func_t *rx_handler,
3425 void *rx_handler_data)
3429 if (dev->rx_handler)
3432 /* Note: rx_handler_data must be set before rx_handler */
3433 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3434 rcu_assign_pointer(dev->rx_handler, rx_handler);
3438 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3441 * netdev_rx_handler_unregister - unregister receive handler
3442 * @dev: device to unregister a handler from
3444 * Unregister a receive handler from a device.
3446 * The caller must hold the rtnl_mutex.
3448 void netdev_rx_handler_unregister(struct net_device *dev)
3452 RCU_INIT_POINTER(dev->rx_handler, NULL);
3453 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3454 * section has a guarantee to see a non NULL rx_handler_data
3458 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3460 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3463 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3464 * the special handling of PFMEMALLOC skbs.
3466 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3468 switch (skb->protocol) {
3469 case __constant_htons(ETH_P_ARP):
3470 case __constant_htons(ETH_P_IP):
3471 case __constant_htons(ETH_P_IPV6):
3472 case __constant_htons(ETH_P_8021Q):
3473 case __constant_htons(ETH_P_8021AD):
3480 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3482 struct packet_type *ptype, *pt_prev;
3483 rx_handler_func_t *rx_handler;
3484 struct net_device *orig_dev;
3485 struct net_device *null_or_dev;
3486 bool deliver_exact = false;
3487 int ret = NET_RX_DROP;
3490 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3492 trace_netif_receive_skb(skb);
3494 /* if we've gotten here through NAPI, check netpoll */
3495 if (netpoll_receive_skb(skb))
3498 orig_dev = skb->dev;
3500 skb_reset_network_header(skb);
3501 if (!skb_transport_header_was_set(skb))
3502 skb_reset_transport_header(skb);
3503 skb_reset_mac_len(skb);
3510 skb->skb_iif = skb->dev->ifindex;
3512 __this_cpu_inc(softnet_data.processed);
3514 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3515 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3516 skb = vlan_untag(skb);
3521 #ifdef CONFIG_NET_CLS_ACT
3522 if (skb->tc_verd & TC_NCLS) {
3523 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3531 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3532 if (!ptype->dev || ptype->dev == skb->dev) {
3534 ret = deliver_skb(skb, pt_prev, orig_dev);
3540 #ifdef CONFIG_NET_CLS_ACT
3541 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3547 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3550 if (vlan_tx_tag_present(skb)) {
3552 ret = deliver_skb(skb, pt_prev, orig_dev);
3555 if (vlan_do_receive(&skb))
3557 else if (unlikely(!skb))
3561 rx_handler = rcu_dereference(skb->dev->rx_handler);
3564 ret = deliver_skb(skb, pt_prev, orig_dev);
3567 switch (rx_handler(&skb)) {
3568 case RX_HANDLER_CONSUMED:
3569 ret = NET_RX_SUCCESS;
3571 case RX_HANDLER_ANOTHER:
3573 case RX_HANDLER_EXACT:
3574 deliver_exact = true;
3575 case RX_HANDLER_PASS:
3582 if (unlikely(vlan_tx_tag_present(skb))) {
3583 if (vlan_tx_tag_get_id(skb))
3584 skb->pkt_type = PACKET_OTHERHOST;
3585 /* Note: we might in the future use prio bits
3586 * and set skb->priority like in vlan_do_receive()
3587 * For the time being, just ignore Priority Code Point
3592 /* deliver only exact match when indicated */
3593 null_or_dev = deliver_exact ? skb->dev : NULL;
3595 type = skb->protocol;
3596 list_for_each_entry_rcu(ptype,
3597 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3598 if (ptype->type == type &&
3599 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3600 ptype->dev == orig_dev)) {
3602 ret = deliver_skb(skb, pt_prev, orig_dev);
3608 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3611 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3614 atomic_long_inc(&skb->dev->rx_dropped);
3616 /* Jamal, now you will not able to escape explaining
3617 * me how you were going to use this. :-)
3628 static int __netif_receive_skb(struct sk_buff *skb)
3632 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3633 unsigned long pflags = current->flags;
3636 * PFMEMALLOC skbs are special, they should
3637 * - be delivered to SOCK_MEMALLOC sockets only
3638 * - stay away from userspace
3639 * - have bounded memory usage
3641 * Use PF_MEMALLOC as this saves us from propagating the allocation
3642 * context down to all allocation sites.
3644 current->flags |= PF_MEMALLOC;
3645 ret = __netif_receive_skb_core(skb, true);
3646 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3648 ret = __netif_receive_skb_core(skb, false);
3654 * netif_receive_skb - process receive buffer from network
3655 * @skb: buffer to process
3657 * netif_receive_skb() is the main receive data processing function.
3658 * It always succeeds. The buffer may be dropped during processing
3659 * for congestion control or by the protocol layers.
3661 * This function may only be called from softirq context and interrupts
3662 * should be enabled.
3664 * Return values (usually ignored):
3665 * NET_RX_SUCCESS: no congestion
3666 * NET_RX_DROP: packet was dropped
3668 int netif_receive_skb(struct sk_buff *skb)
3670 net_timestamp_check(netdev_tstamp_prequeue, skb);
3672 if (skb_defer_rx_timestamp(skb))
3673 return NET_RX_SUCCESS;
3676 if (static_key_false(&rps_needed)) {
3677 struct rps_dev_flow voidflow, *rflow = &voidflow;
3682 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3685 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3692 return __netif_receive_skb(skb);
3694 EXPORT_SYMBOL(netif_receive_skb);
3696 /* Network device is going away, flush any packets still pending
3697 * Called with irqs disabled.
3699 static void flush_backlog(void *arg)
3701 struct net_device *dev = arg;
3702 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3703 struct sk_buff *skb, *tmp;
3706 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3707 if (skb->dev == dev) {
3708 __skb_unlink(skb, &sd->input_pkt_queue);
3710 input_queue_head_incr(sd);
3715 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3716 if (skb->dev == dev) {
3717 __skb_unlink(skb, &sd->process_queue);
3719 input_queue_head_incr(sd);
3724 static int napi_gro_complete(struct sk_buff *skb)
3726 struct packet_offload *ptype;
3727 __be16 type = skb->protocol;
3728 struct list_head *head = &offload_base;
3731 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3733 if (NAPI_GRO_CB(skb)->count == 1) {
3734 skb_shinfo(skb)->gso_size = 0;
3739 list_for_each_entry_rcu(ptype, head, list) {
3740 if (ptype->type != type || !ptype->callbacks.gro_complete)
3743 err = ptype->callbacks.gro_complete(skb);
3749 WARN_ON(&ptype->list == head);
3751 return NET_RX_SUCCESS;
3755 return netif_receive_skb(skb);
3758 /* napi->gro_list contains packets ordered by age.
3759 * youngest packets at the head of it.
3760 * Complete skbs in reverse order to reduce latencies.
3762 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3764 struct sk_buff *skb, *prev = NULL;
3766 /* scan list and build reverse chain */
3767 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3772 for (skb = prev; skb; skb = prev) {
3775 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3779 napi_gro_complete(skb);
3783 napi->gro_list = NULL;
3785 EXPORT_SYMBOL(napi_gro_flush);
3787 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3790 unsigned int maclen = skb->dev->hard_header_len;
3792 for (p = napi->gro_list; p; p = p->next) {
3793 unsigned long diffs;
3795 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3796 diffs |= p->vlan_tci ^ skb->vlan_tci;
3797 if (maclen == ETH_HLEN)
3798 diffs |= compare_ether_header(skb_mac_header(p),
3799 skb_gro_mac_header(skb));
3801 diffs = memcmp(skb_mac_header(p),
3802 skb_gro_mac_header(skb),
3804 NAPI_GRO_CB(p)->same_flow = !diffs;
3805 NAPI_GRO_CB(p)->flush = 0;
3809 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3811 struct sk_buff **pp = NULL;
3812 struct packet_offload *ptype;
3813 __be16 type = skb->protocol;
3814 struct list_head *head = &offload_base;
3816 enum gro_result ret;
3818 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3821 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3824 gro_list_prepare(napi, skb);
3827 list_for_each_entry_rcu(ptype, head, list) {
3828 if (ptype->type != type || !ptype->callbacks.gro_receive)
3831 skb_set_network_header(skb, skb_gro_offset(skb));
3832 skb_reset_mac_len(skb);
3833 NAPI_GRO_CB(skb)->same_flow = 0;
3834 NAPI_GRO_CB(skb)->flush = 0;
3835 NAPI_GRO_CB(skb)->free = 0;
3837 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3842 if (&ptype->list == head)
3845 same_flow = NAPI_GRO_CB(skb)->same_flow;
3846 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3849 struct sk_buff *nskb = *pp;
3853 napi_gro_complete(nskb);
3860 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3864 NAPI_GRO_CB(skb)->count = 1;
3865 NAPI_GRO_CB(skb)->age = jiffies;
3866 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3867 skb->next = napi->gro_list;
3868 napi->gro_list = skb;
3872 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3873 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3875 BUG_ON(skb->end - skb->tail < grow);
3877 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3880 skb->data_len -= grow;
3882 skb_shinfo(skb)->frags[0].page_offset += grow;
3883 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3885 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3886 skb_frag_unref(skb, 0);
3887 memmove(skb_shinfo(skb)->frags,
3888 skb_shinfo(skb)->frags + 1,
3889 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3902 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3906 if (netif_receive_skb(skb))
3914 case GRO_MERGED_FREE:
3915 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3916 kmem_cache_free(skbuff_head_cache, skb);
3929 static void skb_gro_reset_offset(struct sk_buff *skb)
3931 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3932 const skb_frag_t *frag0 = &pinfo->frags[0];
3934 NAPI_GRO_CB(skb)->data_offset = 0;
3935 NAPI_GRO_CB(skb)->frag0 = NULL;
3936 NAPI_GRO_CB(skb)->frag0_len = 0;
3938 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3940 !PageHighMem(skb_frag_page(frag0))) {
3941 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3942 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3946 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3948 skb_gro_reset_offset(skb);
3950 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3952 EXPORT_SYMBOL(napi_gro_receive);
3954 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3956 __skb_pull(skb, skb_headlen(skb));
3957 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3958 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3960 skb->dev = napi->dev;
3966 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3968 struct sk_buff *skb = napi->skb;
3971 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3977 EXPORT_SYMBOL(napi_get_frags);
3979 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3985 skb->protocol = eth_type_trans(skb, skb->dev);
3987 if (ret == GRO_HELD)
3988 skb_gro_pull(skb, -ETH_HLEN);
3989 else if (netif_receive_skb(skb))
3994 case GRO_MERGED_FREE:
3995 napi_reuse_skb(napi, skb);
4005 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4007 struct sk_buff *skb = napi->skb;
4014 skb_reset_mac_header(skb);
4015 skb_gro_reset_offset(skb);
4017 off = skb_gro_offset(skb);
4018 hlen = off + sizeof(*eth);
4019 eth = skb_gro_header_fast(skb, off);
4020 if (skb_gro_header_hard(skb, hlen)) {
4021 eth = skb_gro_header_slow(skb, hlen, off);
4022 if (unlikely(!eth)) {
4023 napi_reuse_skb(napi, skb);
4029 skb_gro_pull(skb, sizeof(*eth));
4032 * This works because the only protocols we care about don't require
4033 * special handling. We'll fix it up properly at the end.
4035 skb->protocol = eth->h_proto;
4041 gro_result_t napi_gro_frags(struct napi_struct *napi)
4043 struct sk_buff *skb = napi_frags_skb(napi);
4048 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4050 EXPORT_SYMBOL(napi_gro_frags);
4053 * net_rps_action sends any pending IPI's for rps.
4054 * Note: called with local irq disabled, but exits with local irq enabled.
4056 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4059 struct softnet_data *remsd = sd->rps_ipi_list;
4062 sd->rps_ipi_list = NULL;
4066 /* Send pending IPI's to kick RPS processing on remote cpus. */
4068 struct softnet_data *next = remsd->rps_ipi_next;
4070 if (cpu_online(remsd->cpu))
4071 __smp_call_function_single(remsd->cpu,
4080 static int process_backlog(struct napi_struct *napi, int quota)
4083 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4086 /* Check if we have pending ipi, its better to send them now,
4087 * not waiting net_rx_action() end.
4089 if (sd->rps_ipi_list) {
4090 local_irq_disable();
4091 net_rps_action_and_irq_enable(sd);
4094 napi->weight = weight_p;
4095 local_irq_disable();
4096 while (work < quota) {
4097 struct sk_buff *skb;
4100 while ((skb = __skb_dequeue(&sd->process_queue))) {
4102 __netif_receive_skb(skb);
4103 local_irq_disable();
4104 input_queue_head_incr(sd);
4105 if (++work >= quota) {
4112 qlen = skb_queue_len(&sd->input_pkt_queue);
4114 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4115 &sd->process_queue);
4117 if (qlen < quota - work) {
4119 * Inline a custom version of __napi_complete().
4120 * only current cpu owns and manipulates this napi,
4121 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4122 * we can use a plain write instead of clear_bit(),
4123 * and we dont need an smp_mb() memory barrier.
4125 list_del(&napi->poll_list);
4128 quota = work + qlen;
4138 * __napi_schedule - schedule for receive
4139 * @n: entry to schedule
4141 * The entry's receive function will be scheduled to run
4143 void __napi_schedule(struct napi_struct *n)
4145 unsigned long flags;
4147 local_irq_save(flags);
4148 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4149 local_irq_restore(flags);
4151 EXPORT_SYMBOL(__napi_schedule);
4153 void __napi_complete(struct napi_struct *n)
4155 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4156 BUG_ON(n->gro_list);
4158 list_del(&n->poll_list);
4159 smp_mb__before_clear_bit();
4160 clear_bit(NAPI_STATE_SCHED, &n->state);
4162 EXPORT_SYMBOL(__napi_complete);
4164 void napi_complete(struct napi_struct *n)
4166 unsigned long flags;
4169 * don't let napi dequeue from the cpu poll list
4170 * just in case its running on a different cpu
4172 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4175 napi_gro_flush(n, false);
4176 local_irq_save(flags);
4178 local_irq_restore(flags);
4180 EXPORT_SYMBOL(napi_complete);
4182 /* must be called under rcu_read_lock(), as we dont take a reference */
4183 struct napi_struct *napi_by_id(unsigned int napi_id)
4185 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4186 struct napi_struct *napi;
4188 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4189 if (napi->napi_id == napi_id)
4194 EXPORT_SYMBOL_GPL(napi_by_id);
4196 void napi_hash_add(struct napi_struct *napi)
4198 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4200 spin_lock(&napi_hash_lock);
4202 /* 0 is not a valid id, we also skip an id that is taken
4203 * we expect both events to be extremely rare
4206 while (!napi->napi_id) {
4207 napi->napi_id = ++napi_gen_id;
4208 if (napi_by_id(napi->napi_id))
4212 hlist_add_head_rcu(&napi->napi_hash_node,
4213 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4215 spin_unlock(&napi_hash_lock);
4218 EXPORT_SYMBOL_GPL(napi_hash_add);
4220 /* Warning : caller is responsible to make sure rcu grace period
4221 * is respected before freeing memory containing @napi
4223 void napi_hash_del(struct napi_struct *napi)
4225 spin_lock(&napi_hash_lock);
4227 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4228 hlist_del_rcu(&napi->napi_hash_node);
4230 spin_unlock(&napi_hash_lock);
4232 EXPORT_SYMBOL_GPL(napi_hash_del);
4234 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4235 int (*poll)(struct napi_struct *, int), int weight)
4237 INIT_LIST_HEAD(&napi->poll_list);
4238 napi->gro_count = 0;
4239 napi->gro_list = NULL;
4242 if (weight > NAPI_POLL_WEIGHT)
4243 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4245 napi->weight = weight;
4246 list_add(&napi->dev_list, &dev->napi_list);
4248 #ifdef CONFIG_NETPOLL
4249 spin_lock_init(&napi->poll_lock);
4250 napi->poll_owner = -1;
4252 set_bit(NAPI_STATE_SCHED, &napi->state);
4254 EXPORT_SYMBOL(netif_napi_add);
4256 void netif_napi_del(struct napi_struct *napi)
4258 struct sk_buff *skb, *next;
4260 list_del_init(&napi->dev_list);
4261 napi_free_frags(napi);
4263 for (skb = napi->gro_list; skb; skb = next) {
4269 napi->gro_list = NULL;
4270 napi->gro_count = 0;
4272 EXPORT_SYMBOL(netif_napi_del);
4274 static void net_rx_action(struct softirq_action *h)
4276 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4277 unsigned long time_limit = jiffies + 2;
4278 int budget = netdev_budget;
4281 local_irq_disable();
4283 while (!list_empty(&sd->poll_list)) {
4284 struct napi_struct *n;
4287 /* If softirq window is exhuasted then punt.
4288 * Allow this to run for 2 jiffies since which will allow
4289 * an average latency of 1.5/HZ.
4291 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4296 /* Even though interrupts have been re-enabled, this
4297 * access is safe because interrupts can only add new
4298 * entries to the tail of this list, and only ->poll()
4299 * calls can remove this head entry from the list.
4301 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4303 have = netpoll_poll_lock(n);
4307 /* This NAPI_STATE_SCHED test is for avoiding a race
4308 * with netpoll's poll_napi(). Only the entity which
4309 * obtains the lock and sees NAPI_STATE_SCHED set will
4310 * actually make the ->poll() call. Therefore we avoid
4311 * accidentally calling ->poll() when NAPI is not scheduled.
4314 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4315 work = n->poll(n, weight);
4319 WARN_ON_ONCE(work > weight);
4323 local_irq_disable();
4325 /* Drivers must not modify the NAPI state if they
4326 * consume the entire weight. In such cases this code
4327 * still "owns" the NAPI instance and therefore can
4328 * move the instance around on the list at-will.
4330 if (unlikely(work == weight)) {
4331 if (unlikely(napi_disable_pending(n))) {
4334 local_irq_disable();
4337 /* flush too old packets
4338 * If HZ < 1000, flush all packets.
4341 napi_gro_flush(n, HZ >= 1000);
4342 local_irq_disable();
4344 list_move_tail(&n->poll_list, &sd->poll_list);
4348 netpoll_poll_unlock(have);
4351 net_rps_action_and_irq_enable(sd);
4353 #ifdef CONFIG_NET_DMA
4355 * There may not be any more sk_buffs coming right now, so push
4356 * any pending DMA copies to hardware
4358 dma_issue_pending_all();
4365 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4369 struct netdev_adjacent {
4370 struct net_device *dev;
4372 /* upper master flag, there can only be one master device per list */
4375 /* counter for the number of times this device was added to us */
4378 /* private field for the users */
4381 struct list_head list;
4382 struct rcu_head rcu;
4385 static struct netdev_adjacent *__netdev_find_adj_rcu(struct net_device *dev,
4386 struct net_device *adj_dev,
4387 struct list_head *adj_list)
4389 struct netdev_adjacent *adj;
4391 list_for_each_entry_rcu(adj, adj_list, list) {
4392 if (adj->dev == adj_dev)
4398 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4399 struct net_device *adj_dev,
4400 struct list_head *adj_list)
4402 struct netdev_adjacent *adj;
4404 list_for_each_entry(adj, adj_list, list) {
4405 if (adj->dev == adj_dev)
4412 * netdev_has_upper_dev - Check if device is linked to an upper device
4414 * @upper_dev: upper device to check
4416 * Find out if a device is linked to specified upper device and return true
4417 * in case it is. Note that this checks only immediate upper device,
4418 * not through a complete stack of devices. The caller must hold the RTNL lock.
4420 bool netdev_has_upper_dev(struct net_device *dev,
4421 struct net_device *upper_dev)
4425 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4427 EXPORT_SYMBOL(netdev_has_upper_dev);
4430 * netdev_has_any_upper_dev - Check if device is linked to some device
4433 * Find out if a device is linked to an upper device and return true in case
4434 * it is. The caller must hold the RTNL lock.
4436 bool netdev_has_any_upper_dev(struct net_device *dev)
4440 return !list_empty(&dev->all_adj_list.upper);
4442 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4445 * netdev_master_upper_dev_get - Get master upper device
4448 * Find a master upper device and return pointer to it or NULL in case
4449 * it's not there. The caller must hold the RTNL lock.
4451 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4453 struct netdev_adjacent *upper;
4457 if (list_empty(&dev->adj_list.upper))
4460 upper = list_first_entry(&dev->adj_list.upper,
4461 struct netdev_adjacent, list);
4462 if (likely(upper->master))
4466 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4468 void *netdev_adjacent_get_private(struct list_head *adj_list)
4470 struct netdev_adjacent *adj;
4472 adj = list_entry(adj_list, struct netdev_adjacent, list);
4474 return adj->private;
4476 EXPORT_SYMBOL(netdev_adjacent_get_private);
4479 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4481 * @iter: list_head ** of the current position
4483 * Gets the next device from the dev's upper list, starting from iter
4484 * position. The caller must hold RCU read lock.
4486 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4487 struct list_head **iter)
4489 struct netdev_adjacent *upper;
4491 WARN_ON_ONCE(!rcu_read_lock_held());
4493 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4495 if (&upper->list == &dev->all_adj_list.upper)
4498 *iter = &upper->list;
4502 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4505 * netdev_lower_get_next_private - Get the next ->private from the
4506 * lower neighbour list
4508 * @iter: list_head ** of the current position
4510 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4511 * list, starting from iter position. The caller must hold either hold the
4512 * RTNL lock or its own locking that guarantees that the neighbour lower
4513 * list will remain unchainged.
4515 void *netdev_lower_get_next_private(struct net_device *dev,
4516 struct list_head **iter)
4518 struct netdev_adjacent *lower;
4520 lower = list_entry(*iter, struct netdev_adjacent, list);
4522 if (&lower->list == &dev->adj_list.lower)
4526 *iter = lower->list.next;
4528 return lower->private;
4530 EXPORT_SYMBOL(netdev_lower_get_next_private);
4533 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4534 * lower neighbour list, RCU
4537 * @iter: list_head ** of the current position
4539 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4540 * list, starting from iter position. The caller must hold RCU read lock.
4542 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4543 struct list_head **iter)
4545 struct netdev_adjacent *lower;
4547 WARN_ON_ONCE(!rcu_read_lock_held());
4549 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4551 if (&lower->list == &dev->adj_list.lower)
4555 *iter = &lower->list;
4557 return lower->private;
4559 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4562 * netdev_master_upper_dev_get_rcu - Get master upper device
4565 * Find a master upper device and return pointer to it or NULL in case
4566 * it's not there. The caller must hold the RCU read lock.
4568 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4570 struct netdev_adjacent *upper;
4572 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4573 struct netdev_adjacent, list);
4574 if (upper && likely(upper->master))
4578 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4580 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4581 struct net_device *adj_dev,
4582 struct list_head *dev_list,
4583 void *private, bool master)
4585 struct netdev_adjacent *adj;
4586 char linkname[IFNAMSIZ+7];
4589 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4596 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4601 adj->master = master;
4603 adj->private = private;
4606 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4607 adj_dev->name, dev->name, adj_dev->name);
4609 if (dev_list == &dev->adj_list.lower) {
4610 sprintf(linkname, "lower_%s", adj_dev->name);
4611 ret = sysfs_create_link(&(dev->dev.kobj),
4612 &(adj_dev->dev.kobj), linkname);
4615 } else if (dev_list == &dev->adj_list.upper) {
4616 sprintf(linkname, "upper_%s", adj_dev->name);
4617 ret = sysfs_create_link(&(dev->dev.kobj),
4618 &(adj_dev->dev.kobj), linkname);
4623 /* Ensure that master link is always the first item in list. */
4625 ret = sysfs_create_link(&(dev->dev.kobj),
4626 &(adj_dev->dev.kobj), "master");
4628 goto remove_symlinks;
4630 list_add_rcu(&adj->list, dev_list);
4632 list_add_tail_rcu(&adj->list, dev_list);
4638 if (dev_list == &dev->adj_list.lower) {
4639 sprintf(linkname, "lower_%s", adj_dev->name);
4640 sysfs_remove_link(&(dev->dev.kobj), linkname);
4641 } else if (dev_list == &dev->adj_list.upper) {
4642 sprintf(linkname, "upper_%s", adj_dev->name);
4643 sysfs_remove_link(&(dev->dev.kobj), linkname);
4652 void __netdev_adjacent_dev_remove(struct net_device *dev,
4653 struct net_device *adj_dev,
4654 struct list_head *dev_list)
4656 struct netdev_adjacent *adj;
4657 char linkname[IFNAMSIZ+7];
4659 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4662 pr_err("tried to remove device %s from %s\n",
4663 dev->name, adj_dev->name);
4667 if (adj->ref_nr > 1) {
4668 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4675 sysfs_remove_link(&(dev->dev.kobj), "master");
4677 if (dev_list == &dev->adj_list.lower) {
4678 sprintf(linkname, "lower_%s", adj_dev->name);
4679 sysfs_remove_link(&(dev->dev.kobj), linkname);
4680 } else if (dev_list == &dev->adj_list.upper) {
4681 sprintf(linkname, "upper_%s", adj_dev->name);
4682 sysfs_remove_link(&(dev->dev.kobj), linkname);
4685 list_del_rcu(&adj->list);
4686 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4687 adj_dev->name, dev->name, adj_dev->name);
4689 kfree_rcu(adj, rcu);
4692 int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4693 struct net_device *upper_dev,
4694 struct list_head *up_list,
4695 struct list_head *down_list,
4696 void *private, bool master)
4700 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4705 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4708 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4715 int __netdev_adjacent_dev_link(struct net_device *dev,
4716 struct net_device *upper_dev)
4718 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4719 &dev->all_adj_list.upper,
4720 &upper_dev->all_adj_list.lower,
4724 void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4725 struct net_device *upper_dev,
4726 struct list_head *up_list,
4727 struct list_head *down_list)
4729 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4730 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4733 void __netdev_adjacent_dev_unlink(struct net_device *dev,
4734 struct net_device *upper_dev)
4736 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4737 &dev->all_adj_list.upper,
4738 &upper_dev->all_adj_list.lower);
4741 int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4742 struct net_device *upper_dev,
4743 void *private, bool master)
4745 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4750 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4751 &dev->adj_list.upper,
4752 &upper_dev->adj_list.lower,
4755 __netdev_adjacent_dev_unlink(dev, upper_dev);
4762 void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4763 struct net_device *upper_dev)
4765 __netdev_adjacent_dev_unlink(dev, upper_dev);
4766 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4767 &dev->adj_list.upper,
4768 &upper_dev->adj_list.lower);
4771 static int __netdev_upper_dev_link(struct net_device *dev,
4772 struct net_device *upper_dev, bool master,
4775 struct netdev_adjacent *i, *j, *to_i, *to_j;
4780 if (dev == upper_dev)
4783 /* To prevent loops, check if dev is not upper device to upper_dev. */
4784 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4787 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4790 if (master && netdev_master_upper_dev_get(dev))
4793 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4798 /* Now that we linked these devs, make all the upper_dev's
4799 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4800 * versa, and don't forget the devices itself. All of these
4801 * links are non-neighbours.
4803 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4804 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4805 pr_debug("Interlinking %s with %s, non-neighbour\n",
4806 i->dev->name, j->dev->name);
4807 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4813 /* add dev to every upper_dev's upper device */
4814 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4815 pr_debug("linking %s's upper device %s with %s\n",
4816 upper_dev->name, i->dev->name, dev->name);
4817 ret = __netdev_adjacent_dev_link(dev, i->dev);
4819 goto rollback_upper_mesh;
4822 /* add upper_dev to every dev's lower device */
4823 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4824 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4825 i->dev->name, upper_dev->name);
4826 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4828 goto rollback_lower_mesh;
4831 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4834 rollback_lower_mesh:
4836 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4839 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4844 rollback_upper_mesh:
4846 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4849 __netdev_adjacent_dev_unlink(dev, i->dev);
4857 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4858 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4859 if (i == to_i && j == to_j)
4861 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4867 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4873 * netdev_upper_dev_link - Add a link to the upper device
4875 * @upper_dev: new upper device
4877 * Adds a link to device which is upper to this one. The caller must hold
4878 * the RTNL lock. On a failure a negative errno code is returned.
4879 * On success the reference counts are adjusted and the function
4882 int netdev_upper_dev_link(struct net_device *dev,
4883 struct net_device *upper_dev)
4885 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4887 EXPORT_SYMBOL(netdev_upper_dev_link);
4890 * netdev_master_upper_dev_link - Add a master link to the upper device
4892 * @upper_dev: new upper device
4894 * Adds a link to device which is upper to this one. In this case, only
4895 * one master upper device can be linked, although other non-master devices
4896 * might be linked as well. The caller must hold the RTNL lock.
4897 * On a failure a negative errno code is returned. On success the reference
4898 * counts are adjusted and the function returns zero.
4900 int netdev_master_upper_dev_link(struct net_device *dev,
4901 struct net_device *upper_dev)
4903 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4905 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4907 int netdev_master_upper_dev_link_private(struct net_device *dev,
4908 struct net_device *upper_dev,
4911 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4913 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4916 * netdev_upper_dev_unlink - Removes a link to upper device
4918 * @upper_dev: new upper device
4920 * Removes a link to device which is upper to this one. The caller must hold
4923 void netdev_upper_dev_unlink(struct net_device *dev,
4924 struct net_device *upper_dev)
4926 struct netdev_adjacent *i, *j;
4929 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4931 /* Here is the tricky part. We must remove all dev's lower
4932 * devices from all upper_dev's upper devices and vice
4933 * versa, to maintain the graph relationship.
4935 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4936 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4937 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4939 /* remove also the devices itself from lower/upper device
4942 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4943 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4945 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4946 __netdev_adjacent_dev_unlink(dev, i->dev);
4948 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4950 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4952 void *netdev_lower_dev_get_private_rcu(struct net_device *dev,
4953 struct net_device *lower_dev)
4955 struct netdev_adjacent *lower;
4959 lower = __netdev_find_adj_rcu(dev, lower_dev, &dev->adj_list.lower);
4963 return lower->private;
4965 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu);
4967 void *netdev_lower_dev_get_private(struct net_device *dev,
4968 struct net_device *lower_dev)
4970 struct netdev_adjacent *lower;
4974 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
4978 return lower->private;
4980 EXPORT_SYMBOL(netdev_lower_dev_get_private);
4982 static void dev_change_rx_flags(struct net_device *dev, int flags)
4984 const struct net_device_ops *ops = dev->netdev_ops;
4986 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4987 ops->ndo_change_rx_flags(dev, flags);
4990 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
4992 unsigned int old_flags = dev->flags;
4998 dev->flags |= IFF_PROMISC;
4999 dev->promiscuity += inc;
5000 if (dev->promiscuity == 0) {
5003 * If inc causes overflow, untouch promisc and return error.
5006 dev->flags &= ~IFF_PROMISC;
5008 dev->promiscuity -= inc;
5009 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5014 if (dev->flags != old_flags) {
5015 pr_info("device %s %s promiscuous mode\n",
5017 dev->flags & IFF_PROMISC ? "entered" : "left");
5018 if (audit_enabled) {
5019 current_uid_gid(&uid, &gid);
5020 audit_log(current->audit_context, GFP_ATOMIC,
5021 AUDIT_ANOM_PROMISCUOUS,
5022 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5023 dev->name, (dev->flags & IFF_PROMISC),
5024 (old_flags & IFF_PROMISC),
5025 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5026 from_kuid(&init_user_ns, uid),
5027 from_kgid(&init_user_ns, gid),
5028 audit_get_sessionid(current));
5031 dev_change_rx_flags(dev, IFF_PROMISC);
5034 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5039 * dev_set_promiscuity - update promiscuity count on a device
5043 * Add or remove promiscuity from a device. While the count in the device
5044 * remains above zero the interface remains promiscuous. Once it hits zero
5045 * the device reverts back to normal filtering operation. A negative inc
5046 * value is used to drop promiscuity on the device.
5047 * Return 0 if successful or a negative errno code on error.
5049 int dev_set_promiscuity(struct net_device *dev, int inc)
5051 unsigned int old_flags = dev->flags;
5054 err = __dev_set_promiscuity(dev, inc, true);
5057 if (dev->flags != old_flags)
5058 dev_set_rx_mode(dev);
5061 EXPORT_SYMBOL(dev_set_promiscuity);
5063 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5065 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5069 dev->flags |= IFF_ALLMULTI;
5070 dev->allmulti += inc;
5071 if (dev->allmulti == 0) {
5074 * If inc causes overflow, untouch allmulti and return error.
5077 dev->flags &= ~IFF_ALLMULTI;
5079 dev->allmulti -= inc;
5080 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5085 if (dev->flags ^ old_flags) {
5086 dev_change_rx_flags(dev, IFF_ALLMULTI);
5087 dev_set_rx_mode(dev);
5089 __dev_notify_flags(dev, old_flags,
5090 dev->gflags ^ old_gflags);
5096 * dev_set_allmulti - update allmulti count on a device
5100 * Add or remove reception of all multicast frames to a device. While the
5101 * count in the device remains above zero the interface remains listening
5102 * to all interfaces. Once it hits zero the device reverts back to normal
5103 * filtering operation. A negative @inc value is used to drop the counter
5104 * when releasing a resource needing all multicasts.
5105 * Return 0 if successful or a negative errno code on error.
5108 int dev_set_allmulti(struct net_device *dev, int inc)
5110 return __dev_set_allmulti(dev, inc, true);
5112 EXPORT_SYMBOL(dev_set_allmulti);
5115 * Upload unicast and multicast address lists to device and
5116 * configure RX filtering. When the device doesn't support unicast
5117 * filtering it is put in promiscuous mode while unicast addresses
5120 void __dev_set_rx_mode(struct net_device *dev)
5122 const struct net_device_ops *ops = dev->netdev_ops;
5124 /* dev_open will call this function so the list will stay sane. */
5125 if (!(dev->flags&IFF_UP))
5128 if (!netif_device_present(dev))
5131 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5132 /* Unicast addresses changes may only happen under the rtnl,
5133 * therefore calling __dev_set_promiscuity here is safe.
5135 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5136 __dev_set_promiscuity(dev, 1, false);
5137 dev->uc_promisc = true;
5138 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5139 __dev_set_promiscuity(dev, -1, false);
5140 dev->uc_promisc = false;
5144 if (ops->ndo_set_rx_mode)
5145 ops->ndo_set_rx_mode(dev);
5148 void dev_set_rx_mode(struct net_device *dev)
5150 netif_addr_lock_bh(dev);
5151 __dev_set_rx_mode(dev);
5152 netif_addr_unlock_bh(dev);
5156 * dev_get_flags - get flags reported to userspace
5159 * Get the combination of flag bits exported through APIs to userspace.
5161 unsigned int dev_get_flags(const struct net_device *dev)
5165 flags = (dev->flags & ~(IFF_PROMISC |
5170 (dev->gflags & (IFF_PROMISC |
5173 if (netif_running(dev)) {
5174 if (netif_oper_up(dev))
5175 flags |= IFF_RUNNING;
5176 if (netif_carrier_ok(dev))
5177 flags |= IFF_LOWER_UP;
5178 if (netif_dormant(dev))
5179 flags |= IFF_DORMANT;
5184 EXPORT_SYMBOL(dev_get_flags);
5186 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5188 unsigned int old_flags = dev->flags;
5194 * Set the flags on our device.
5197 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5198 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5200 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5204 * Load in the correct multicast list now the flags have changed.
5207 if ((old_flags ^ flags) & IFF_MULTICAST)
5208 dev_change_rx_flags(dev, IFF_MULTICAST);
5210 dev_set_rx_mode(dev);
5213 * Have we downed the interface. We handle IFF_UP ourselves
5214 * according to user attempts to set it, rather than blindly
5219 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5220 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5223 dev_set_rx_mode(dev);
5226 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5227 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5228 unsigned int old_flags = dev->flags;
5230 dev->gflags ^= IFF_PROMISC;
5232 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5233 if (dev->flags != old_flags)
5234 dev_set_rx_mode(dev);
5237 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5238 is important. Some (broken) drivers set IFF_PROMISC, when
5239 IFF_ALLMULTI is requested not asking us and not reporting.
5241 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5242 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5244 dev->gflags ^= IFF_ALLMULTI;
5245 __dev_set_allmulti(dev, inc, false);
5251 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5252 unsigned int gchanges)
5254 unsigned int changes = dev->flags ^ old_flags;
5257 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges);
5259 if (changes & IFF_UP) {
5260 if (dev->flags & IFF_UP)
5261 call_netdevice_notifiers(NETDEV_UP, dev);
5263 call_netdevice_notifiers(NETDEV_DOWN, dev);
5266 if (dev->flags & IFF_UP &&
5267 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5268 struct netdev_notifier_change_info change_info;
5270 change_info.flags_changed = changes;
5271 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5277 * dev_change_flags - change device settings
5279 * @flags: device state flags
5281 * Change settings on device based state flags. The flags are
5282 * in the userspace exported format.
5284 int dev_change_flags(struct net_device *dev, unsigned int flags)
5287 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5289 ret = __dev_change_flags(dev, flags);
5293 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5294 __dev_notify_flags(dev, old_flags, changes);
5297 EXPORT_SYMBOL(dev_change_flags);
5300 * dev_set_mtu - Change maximum transfer unit
5302 * @new_mtu: new transfer unit
5304 * Change the maximum transfer size of the network device.
5306 int dev_set_mtu(struct net_device *dev, int new_mtu)
5308 const struct net_device_ops *ops = dev->netdev_ops;
5311 if (new_mtu == dev->mtu)
5314 /* MTU must be positive. */
5318 if (!netif_device_present(dev))
5322 if (ops->ndo_change_mtu)
5323 err = ops->ndo_change_mtu(dev, new_mtu);
5328 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5331 EXPORT_SYMBOL(dev_set_mtu);
5334 * dev_set_group - Change group this device belongs to
5336 * @new_group: group this device should belong to
5338 void dev_set_group(struct net_device *dev, int new_group)
5340 dev->group = new_group;
5342 EXPORT_SYMBOL(dev_set_group);
5345 * dev_set_mac_address - Change Media Access Control Address
5349 * Change the hardware (MAC) address of the device
5351 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5353 const struct net_device_ops *ops = dev->netdev_ops;
5356 if (!ops->ndo_set_mac_address)
5358 if (sa->sa_family != dev->type)
5360 if (!netif_device_present(dev))
5362 err = ops->ndo_set_mac_address(dev, sa);
5365 dev->addr_assign_type = NET_ADDR_SET;
5366 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5367 add_device_randomness(dev->dev_addr, dev->addr_len);
5370 EXPORT_SYMBOL(dev_set_mac_address);
5373 * dev_change_carrier - Change device carrier
5375 * @new_carrier: new value
5377 * Change device carrier
5379 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5381 const struct net_device_ops *ops = dev->netdev_ops;
5383 if (!ops->ndo_change_carrier)
5385 if (!netif_device_present(dev))
5387 return ops->ndo_change_carrier(dev, new_carrier);
5389 EXPORT_SYMBOL(dev_change_carrier);
5392 * dev_get_phys_port_id - Get device physical port ID
5396 * Get device physical port ID
5398 int dev_get_phys_port_id(struct net_device *dev,
5399 struct netdev_phys_port_id *ppid)
5401 const struct net_device_ops *ops = dev->netdev_ops;
5403 if (!ops->ndo_get_phys_port_id)
5405 return ops->ndo_get_phys_port_id(dev, ppid);
5407 EXPORT_SYMBOL(dev_get_phys_port_id);
5410 * dev_new_index - allocate an ifindex
5411 * @net: the applicable net namespace
5413 * Returns a suitable unique value for a new device interface
5414 * number. The caller must hold the rtnl semaphore or the
5415 * dev_base_lock to be sure it remains unique.
5417 static int dev_new_index(struct net *net)
5419 int ifindex = net->ifindex;
5423 if (!__dev_get_by_index(net, ifindex))
5424 return net->ifindex = ifindex;
5428 /* Delayed registration/unregisteration */
5429 static LIST_HEAD(net_todo_list);
5430 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5432 static void net_set_todo(struct net_device *dev)
5434 list_add_tail(&dev->todo_list, &net_todo_list);
5435 dev_net(dev)->dev_unreg_count++;
5438 static void rollback_registered_many(struct list_head *head)
5440 struct net_device *dev, *tmp;
5441 LIST_HEAD(close_head);
5443 BUG_ON(dev_boot_phase);
5446 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5447 /* Some devices call without registering
5448 * for initialization unwind. Remove those
5449 * devices and proceed with the remaining.
5451 if (dev->reg_state == NETREG_UNINITIALIZED) {
5452 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5456 list_del(&dev->unreg_list);
5459 dev->dismantle = true;
5460 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5463 /* If device is running, close it first. */
5464 list_for_each_entry(dev, head, unreg_list)
5465 list_add_tail(&dev->close_list, &close_head);
5466 dev_close_many(&close_head);
5468 list_for_each_entry(dev, head, unreg_list) {
5469 /* And unlink it from device chain. */
5470 unlist_netdevice(dev);
5472 dev->reg_state = NETREG_UNREGISTERING;
5477 list_for_each_entry(dev, head, unreg_list) {
5478 /* Shutdown queueing discipline. */
5482 /* Notify protocols, that we are about to destroy
5483 this device. They should clean all the things.
5485 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5487 if (!dev->rtnl_link_ops ||
5488 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5489 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5492 * Flush the unicast and multicast chains
5497 if (dev->netdev_ops->ndo_uninit)
5498 dev->netdev_ops->ndo_uninit(dev);
5500 /* Notifier chain MUST detach us all upper devices. */
5501 WARN_ON(netdev_has_any_upper_dev(dev));
5503 /* Remove entries from kobject tree */
5504 netdev_unregister_kobject(dev);
5506 /* Remove XPS queueing entries */
5507 netif_reset_xps_queues_gt(dev, 0);
5513 list_for_each_entry(dev, head, unreg_list)
5517 static void rollback_registered(struct net_device *dev)
5521 list_add(&dev->unreg_list, &single);
5522 rollback_registered_many(&single);
5526 static netdev_features_t netdev_fix_features(struct net_device *dev,
5527 netdev_features_t features)
5529 /* Fix illegal checksum combinations */
5530 if ((features & NETIF_F_HW_CSUM) &&
5531 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5532 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5533 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5536 /* TSO requires that SG is present as well. */
5537 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5538 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5539 features &= ~NETIF_F_ALL_TSO;
5542 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5543 !(features & NETIF_F_IP_CSUM)) {
5544 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5545 features &= ~NETIF_F_TSO;
5546 features &= ~NETIF_F_TSO_ECN;
5549 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5550 !(features & NETIF_F_IPV6_CSUM)) {
5551 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5552 features &= ~NETIF_F_TSO6;
5555 /* TSO ECN requires that TSO is present as well. */
5556 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5557 features &= ~NETIF_F_TSO_ECN;
5559 /* Software GSO depends on SG. */
5560 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5561 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5562 features &= ~NETIF_F_GSO;
5565 /* UFO needs SG and checksumming */
5566 if (features & NETIF_F_UFO) {
5567 /* maybe split UFO into V4 and V6? */
5568 if (!((features & NETIF_F_GEN_CSUM) ||
5569 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5570 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5572 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5573 features &= ~NETIF_F_UFO;
5576 if (!(features & NETIF_F_SG)) {
5578 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5579 features &= ~NETIF_F_UFO;
5586 int __netdev_update_features(struct net_device *dev)
5588 netdev_features_t features;
5593 features = netdev_get_wanted_features(dev);
5595 if (dev->netdev_ops->ndo_fix_features)
5596 features = dev->netdev_ops->ndo_fix_features(dev, features);
5598 /* driver might be less strict about feature dependencies */
5599 features = netdev_fix_features(dev, features);
5601 if (dev->features == features)
5604 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5605 &dev->features, &features);
5607 if (dev->netdev_ops->ndo_set_features)
5608 err = dev->netdev_ops->ndo_set_features(dev, features);
5610 if (unlikely(err < 0)) {
5612 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5613 err, &features, &dev->features);
5618 dev->features = features;
5624 * netdev_update_features - recalculate device features
5625 * @dev: the device to check
5627 * Recalculate dev->features set and send notifications if it
5628 * has changed. Should be called after driver or hardware dependent
5629 * conditions might have changed that influence the features.
5631 void netdev_update_features(struct net_device *dev)
5633 if (__netdev_update_features(dev))
5634 netdev_features_change(dev);
5636 EXPORT_SYMBOL(netdev_update_features);
5639 * netdev_change_features - recalculate device features
5640 * @dev: the device to check
5642 * Recalculate dev->features set and send notifications even
5643 * if they have not changed. Should be called instead of
5644 * netdev_update_features() if also dev->vlan_features might
5645 * have changed to allow the changes to be propagated to stacked
5648 void netdev_change_features(struct net_device *dev)
5650 __netdev_update_features(dev);
5651 netdev_features_change(dev);
5653 EXPORT_SYMBOL(netdev_change_features);
5656 * netif_stacked_transfer_operstate - transfer operstate
5657 * @rootdev: the root or lower level device to transfer state from
5658 * @dev: the device to transfer operstate to
5660 * Transfer operational state from root to device. This is normally
5661 * called when a stacking relationship exists between the root
5662 * device and the device(a leaf device).
5664 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5665 struct net_device *dev)
5667 if (rootdev->operstate == IF_OPER_DORMANT)
5668 netif_dormant_on(dev);
5670 netif_dormant_off(dev);
5672 if (netif_carrier_ok(rootdev)) {
5673 if (!netif_carrier_ok(dev))
5674 netif_carrier_on(dev);
5676 if (netif_carrier_ok(dev))
5677 netif_carrier_off(dev);
5680 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5683 static int netif_alloc_rx_queues(struct net_device *dev)
5685 unsigned int i, count = dev->num_rx_queues;
5686 struct netdev_rx_queue *rx;
5690 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5696 for (i = 0; i < count; i++)
5702 static void netdev_init_one_queue(struct net_device *dev,
5703 struct netdev_queue *queue, void *_unused)
5705 /* Initialize queue lock */
5706 spin_lock_init(&queue->_xmit_lock);
5707 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5708 queue->xmit_lock_owner = -1;
5709 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5712 dql_init(&queue->dql, HZ);
5716 static void netif_free_tx_queues(struct net_device *dev)
5718 if (is_vmalloc_addr(dev->_tx))
5724 static int netif_alloc_netdev_queues(struct net_device *dev)
5726 unsigned int count = dev->num_tx_queues;
5727 struct netdev_queue *tx;
5728 size_t sz = count * sizeof(*tx);
5730 BUG_ON(count < 1 || count > 0xffff);
5732 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5740 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5741 spin_lock_init(&dev->tx_global_lock);
5747 * register_netdevice - register a network device
5748 * @dev: device to register
5750 * Take a completed network device structure and add it to the kernel
5751 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5752 * chain. 0 is returned on success. A negative errno code is returned
5753 * on a failure to set up the device, or if the name is a duplicate.
5755 * Callers must hold the rtnl semaphore. You may want
5756 * register_netdev() instead of this.
5759 * The locking appears insufficient to guarantee two parallel registers
5760 * will not get the same name.
5763 int register_netdevice(struct net_device *dev)
5766 struct net *net = dev_net(dev);
5768 BUG_ON(dev_boot_phase);
5773 /* When net_device's are persistent, this will be fatal. */
5774 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5777 spin_lock_init(&dev->addr_list_lock);
5778 netdev_set_addr_lockdep_class(dev);
5782 ret = dev_get_valid_name(net, dev, dev->name);
5786 /* Init, if this function is available */
5787 if (dev->netdev_ops->ndo_init) {
5788 ret = dev->netdev_ops->ndo_init(dev);
5796 if (((dev->hw_features | dev->features) &
5797 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5798 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5799 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5800 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5807 dev->ifindex = dev_new_index(net);
5808 else if (__dev_get_by_index(net, dev->ifindex))
5811 if (dev->iflink == -1)
5812 dev->iflink = dev->ifindex;
5814 /* Transfer changeable features to wanted_features and enable
5815 * software offloads (GSO and GRO).
5817 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5818 dev->features |= NETIF_F_SOFT_FEATURES;
5819 dev->wanted_features = dev->features & dev->hw_features;
5821 /* Turn on no cache copy if HW is doing checksum */
5822 if (!(dev->flags & IFF_LOOPBACK)) {
5823 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5824 if (dev->features & NETIF_F_ALL_CSUM) {
5825 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5826 dev->features |= NETIF_F_NOCACHE_COPY;
5830 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5832 dev->vlan_features |= NETIF_F_HIGHDMA;
5834 /* Make NETIF_F_SG inheritable to tunnel devices.
5836 dev->hw_enc_features |= NETIF_F_SG;
5838 /* Make NETIF_F_SG inheritable to MPLS.
5840 dev->mpls_features |= NETIF_F_SG;
5842 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5843 ret = notifier_to_errno(ret);
5847 ret = netdev_register_kobject(dev);
5850 dev->reg_state = NETREG_REGISTERED;
5852 __netdev_update_features(dev);
5855 * Default initial state at registry is that the
5856 * device is present.
5859 set_bit(__LINK_STATE_PRESENT, &dev->state);
5861 linkwatch_init_dev(dev);
5863 dev_init_scheduler(dev);
5865 list_netdevice(dev);
5866 add_device_randomness(dev->dev_addr, dev->addr_len);
5868 /* If the device has permanent device address, driver should
5869 * set dev_addr and also addr_assign_type should be set to
5870 * NET_ADDR_PERM (default value).
5872 if (dev->addr_assign_type == NET_ADDR_PERM)
5873 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5875 /* Notify protocols, that a new device appeared. */
5876 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5877 ret = notifier_to_errno(ret);
5879 rollback_registered(dev);
5880 dev->reg_state = NETREG_UNREGISTERED;
5883 * Prevent userspace races by waiting until the network
5884 * device is fully setup before sending notifications.
5886 if (!dev->rtnl_link_ops ||
5887 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5888 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5894 if (dev->netdev_ops->ndo_uninit)
5895 dev->netdev_ops->ndo_uninit(dev);
5898 EXPORT_SYMBOL(register_netdevice);
5901 * init_dummy_netdev - init a dummy network device for NAPI
5902 * @dev: device to init
5904 * This takes a network device structure and initialize the minimum
5905 * amount of fields so it can be used to schedule NAPI polls without
5906 * registering a full blown interface. This is to be used by drivers
5907 * that need to tie several hardware interfaces to a single NAPI
5908 * poll scheduler due to HW limitations.
5910 int init_dummy_netdev(struct net_device *dev)
5912 /* Clear everything. Note we don't initialize spinlocks
5913 * are they aren't supposed to be taken by any of the
5914 * NAPI code and this dummy netdev is supposed to be
5915 * only ever used for NAPI polls
5917 memset(dev, 0, sizeof(struct net_device));
5919 /* make sure we BUG if trying to hit standard
5920 * register/unregister code path
5922 dev->reg_state = NETREG_DUMMY;
5924 /* NAPI wants this */
5925 INIT_LIST_HEAD(&dev->napi_list);
5927 /* a dummy interface is started by default */
5928 set_bit(__LINK_STATE_PRESENT, &dev->state);
5929 set_bit(__LINK_STATE_START, &dev->state);
5931 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5932 * because users of this 'device' dont need to change
5938 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5942 * register_netdev - register a network device
5943 * @dev: device to register
5945 * Take a completed network device structure and add it to the kernel
5946 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5947 * chain. 0 is returned on success. A negative errno code is returned
5948 * on a failure to set up the device, or if the name is a duplicate.
5950 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5951 * and expands the device name if you passed a format string to
5954 int register_netdev(struct net_device *dev)
5959 err = register_netdevice(dev);
5963 EXPORT_SYMBOL(register_netdev);
5965 int netdev_refcnt_read(const struct net_device *dev)
5969 for_each_possible_cpu(i)
5970 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5973 EXPORT_SYMBOL(netdev_refcnt_read);
5976 * netdev_wait_allrefs - wait until all references are gone.
5977 * @dev: target net_device
5979 * This is called when unregistering network devices.
5981 * Any protocol or device that holds a reference should register
5982 * for netdevice notification, and cleanup and put back the
5983 * reference if they receive an UNREGISTER event.
5984 * We can get stuck here if buggy protocols don't correctly
5987 static void netdev_wait_allrefs(struct net_device *dev)
5989 unsigned long rebroadcast_time, warning_time;
5992 linkwatch_forget_dev(dev);
5994 rebroadcast_time = warning_time = jiffies;
5995 refcnt = netdev_refcnt_read(dev);
5997 while (refcnt != 0) {
5998 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6001 /* Rebroadcast unregister notification */
6002 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6008 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6009 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6011 /* We must not have linkwatch events
6012 * pending on unregister. If this
6013 * happens, we simply run the queue
6014 * unscheduled, resulting in a noop
6017 linkwatch_run_queue();
6022 rebroadcast_time = jiffies;
6027 refcnt = netdev_refcnt_read(dev);
6029 if (time_after(jiffies, warning_time + 10 * HZ)) {
6030 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6032 warning_time = jiffies;
6041 * register_netdevice(x1);
6042 * register_netdevice(x2);
6044 * unregister_netdevice(y1);
6045 * unregister_netdevice(y2);
6051 * We are invoked by rtnl_unlock().
6052 * This allows us to deal with problems:
6053 * 1) We can delete sysfs objects which invoke hotplug
6054 * without deadlocking with linkwatch via keventd.
6055 * 2) Since we run with the RTNL semaphore not held, we can sleep
6056 * safely in order to wait for the netdev refcnt to drop to zero.
6058 * We must not return until all unregister events added during
6059 * the interval the lock was held have been completed.
6061 void netdev_run_todo(void)
6063 struct list_head list;
6065 /* Snapshot list, allow later requests */
6066 list_replace_init(&net_todo_list, &list);
6071 /* Wait for rcu callbacks to finish before next phase */
6072 if (!list_empty(&list))
6075 while (!list_empty(&list)) {
6076 struct net_device *dev
6077 = list_first_entry(&list, struct net_device, todo_list);
6078 list_del(&dev->todo_list);
6081 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6084 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6085 pr_err("network todo '%s' but state %d\n",
6086 dev->name, dev->reg_state);
6091 dev->reg_state = NETREG_UNREGISTERED;
6093 on_each_cpu(flush_backlog, dev, 1);
6095 netdev_wait_allrefs(dev);
6098 BUG_ON(netdev_refcnt_read(dev));
6099 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6100 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6101 WARN_ON(dev->dn_ptr);
6103 if (dev->destructor)
6104 dev->destructor(dev);
6106 /* Report a network device has been unregistered */
6108 dev_net(dev)->dev_unreg_count--;
6110 wake_up(&netdev_unregistering_wq);
6112 /* Free network device */
6113 kobject_put(&dev->dev.kobj);
6117 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6118 * fields in the same order, with only the type differing.
6120 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6121 const struct net_device_stats *netdev_stats)
6123 #if BITS_PER_LONG == 64
6124 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6125 memcpy(stats64, netdev_stats, sizeof(*stats64));
6127 size_t i, n = sizeof(*stats64) / sizeof(u64);
6128 const unsigned long *src = (const unsigned long *)netdev_stats;
6129 u64 *dst = (u64 *)stats64;
6131 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6132 sizeof(*stats64) / sizeof(u64));
6133 for (i = 0; i < n; i++)
6137 EXPORT_SYMBOL(netdev_stats_to_stats64);
6140 * dev_get_stats - get network device statistics
6141 * @dev: device to get statistics from
6142 * @storage: place to store stats
6144 * Get network statistics from device. Return @storage.
6145 * The device driver may provide its own method by setting
6146 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6147 * otherwise the internal statistics structure is used.
6149 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6150 struct rtnl_link_stats64 *storage)
6152 const struct net_device_ops *ops = dev->netdev_ops;
6154 if (ops->ndo_get_stats64) {
6155 memset(storage, 0, sizeof(*storage));
6156 ops->ndo_get_stats64(dev, storage);
6157 } else if (ops->ndo_get_stats) {
6158 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6160 netdev_stats_to_stats64(storage, &dev->stats);
6162 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6165 EXPORT_SYMBOL(dev_get_stats);
6167 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6169 struct netdev_queue *queue = dev_ingress_queue(dev);
6171 #ifdef CONFIG_NET_CLS_ACT
6174 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6177 netdev_init_one_queue(dev, queue, NULL);
6178 queue->qdisc = &noop_qdisc;
6179 queue->qdisc_sleeping = &noop_qdisc;
6180 rcu_assign_pointer(dev->ingress_queue, queue);
6185 static const struct ethtool_ops default_ethtool_ops;
6187 void netdev_set_default_ethtool_ops(struct net_device *dev,
6188 const struct ethtool_ops *ops)
6190 if (dev->ethtool_ops == &default_ethtool_ops)
6191 dev->ethtool_ops = ops;
6193 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6196 * alloc_netdev_mqs - allocate network device
6197 * @sizeof_priv: size of private data to allocate space for
6198 * @name: device name format string
6199 * @setup: callback to initialize device
6200 * @txqs: the number of TX subqueues to allocate
6201 * @rxqs: the number of RX subqueues to allocate
6203 * Allocates a struct net_device with private data area for driver use
6204 * and performs basic initialization. Also allocates subquue structs
6205 * for each queue on the device.
6207 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6208 void (*setup)(struct net_device *),
6209 unsigned int txqs, unsigned int rxqs)
6211 struct net_device *dev;
6213 struct net_device *p;
6215 BUG_ON(strlen(name) >= sizeof(dev->name));
6218 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6224 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6229 alloc_size = sizeof(struct net_device);
6231 /* ensure 32-byte alignment of private area */
6232 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6233 alloc_size += sizeof_priv;
6235 /* ensure 32-byte alignment of whole construct */
6236 alloc_size += NETDEV_ALIGN - 1;
6238 p = kzalloc(alloc_size, GFP_KERNEL);
6242 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6243 dev->padded = (char *)dev - (char *)p;
6245 dev->pcpu_refcnt = alloc_percpu(int);
6246 if (!dev->pcpu_refcnt)
6249 if (dev_addr_init(dev))
6255 dev_net_set(dev, &init_net);
6257 dev->gso_max_size = GSO_MAX_SIZE;
6258 dev->gso_max_segs = GSO_MAX_SEGS;
6260 INIT_LIST_HEAD(&dev->napi_list);
6261 INIT_LIST_HEAD(&dev->unreg_list);
6262 INIT_LIST_HEAD(&dev->close_list);
6263 INIT_LIST_HEAD(&dev->link_watch_list);
6264 INIT_LIST_HEAD(&dev->adj_list.upper);
6265 INIT_LIST_HEAD(&dev->adj_list.lower);
6266 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6267 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6268 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6271 dev->num_tx_queues = txqs;
6272 dev->real_num_tx_queues = txqs;
6273 if (netif_alloc_netdev_queues(dev))
6277 dev->num_rx_queues = rxqs;
6278 dev->real_num_rx_queues = rxqs;
6279 if (netif_alloc_rx_queues(dev))
6283 strcpy(dev->name, name);
6284 dev->group = INIT_NETDEV_GROUP;
6285 if (!dev->ethtool_ops)
6286 dev->ethtool_ops = &default_ethtool_ops;
6294 free_percpu(dev->pcpu_refcnt);
6295 netif_free_tx_queues(dev);
6304 EXPORT_SYMBOL(alloc_netdev_mqs);
6307 * free_netdev - free network device
6310 * This function does the last stage of destroying an allocated device
6311 * interface. The reference to the device object is released.
6312 * If this is the last reference then it will be freed.
6314 void free_netdev(struct net_device *dev)
6316 struct napi_struct *p, *n;
6318 release_net(dev_net(dev));
6320 netif_free_tx_queues(dev);
6325 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6327 /* Flush device addresses */
6328 dev_addr_flush(dev);
6330 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6333 free_percpu(dev->pcpu_refcnt);
6334 dev->pcpu_refcnt = NULL;
6336 /* Compatibility with error handling in drivers */
6337 if (dev->reg_state == NETREG_UNINITIALIZED) {
6338 kfree((char *)dev - dev->padded);
6342 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6343 dev->reg_state = NETREG_RELEASED;
6345 /* will free via device release */
6346 put_device(&dev->dev);
6348 EXPORT_SYMBOL(free_netdev);
6351 * synchronize_net - Synchronize with packet receive processing
6353 * Wait for packets currently being received to be done.
6354 * Does not block later packets from starting.
6356 void synchronize_net(void)
6359 if (rtnl_is_locked())
6360 synchronize_rcu_expedited();
6364 EXPORT_SYMBOL(synchronize_net);
6367 * unregister_netdevice_queue - remove device from the kernel
6371 * This function shuts down a device interface and removes it
6372 * from the kernel tables.
6373 * If head not NULL, device is queued to be unregistered later.
6375 * Callers must hold the rtnl semaphore. You may want
6376 * unregister_netdev() instead of this.
6379 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6384 list_move_tail(&dev->unreg_list, head);
6386 rollback_registered(dev);
6387 /* Finish processing unregister after unlock */
6391 EXPORT_SYMBOL(unregister_netdevice_queue);
6394 * unregister_netdevice_many - unregister many devices
6395 * @head: list of devices
6397 void unregister_netdevice_many(struct list_head *head)
6399 struct net_device *dev;
6401 if (!list_empty(head)) {
6402 rollback_registered_many(head);
6403 list_for_each_entry(dev, head, unreg_list)
6407 EXPORT_SYMBOL(unregister_netdevice_many);
6410 * unregister_netdev - remove device from the kernel
6413 * This function shuts down a device interface and removes it
6414 * from the kernel tables.
6416 * This is just a wrapper for unregister_netdevice that takes
6417 * the rtnl semaphore. In general you want to use this and not
6418 * unregister_netdevice.
6420 void unregister_netdev(struct net_device *dev)
6423 unregister_netdevice(dev);
6426 EXPORT_SYMBOL(unregister_netdev);
6429 * dev_change_net_namespace - move device to different nethost namespace
6431 * @net: network namespace
6432 * @pat: If not NULL name pattern to try if the current device name
6433 * is already taken in the destination network namespace.
6435 * This function shuts down a device interface and moves it
6436 * to a new network namespace. On success 0 is returned, on
6437 * a failure a netagive errno code is returned.
6439 * Callers must hold the rtnl semaphore.
6442 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6448 /* Don't allow namespace local devices to be moved. */
6450 if (dev->features & NETIF_F_NETNS_LOCAL)
6453 /* Ensure the device has been registrered */
6454 if (dev->reg_state != NETREG_REGISTERED)
6457 /* Get out if there is nothing todo */
6459 if (net_eq(dev_net(dev), net))
6462 /* Pick the destination device name, and ensure
6463 * we can use it in the destination network namespace.
6466 if (__dev_get_by_name(net, dev->name)) {
6467 /* We get here if we can't use the current device name */
6470 if (dev_get_valid_name(net, dev, pat) < 0)
6475 * And now a mini version of register_netdevice unregister_netdevice.
6478 /* If device is running close it first. */
6481 /* And unlink it from device chain */
6483 unlist_netdevice(dev);
6487 /* Shutdown queueing discipline. */
6490 /* Notify protocols, that we are about to destroy
6491 this device. They should clean all the things.
6493 Note that dev->reg_state stays at NETREG_REGISTERED.
6494 This is wanted because this way 8021q and macvlan know
6495 the device is just moving and can keep their slaves up.
6497 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6499 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6500 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6503 * Flush the unicast and multicast chains
6508 /* Send a netdev-removed uevent to the old namespace */
6509 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6511 /* Actually switch the network namespace */
6512 dev_net_set(dev, net);
6514 /* If there is an ifindex conflict assign a new one */
6515 if (__dev_get_by_index(net, dev->ifindex)) {
6516 int iflink = (dev->iflink == dev->ifindex);
6517 dev->ifindex = dev_new_index(net);
6519 dev->iflink = dev->ifindex;
6522 /* Send a netdev-add uevent to the new namespace */
6523 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6525 /* Fixup kobjects */
6526 err = device_rename(&dev->dev, dev->name);
6529 /* Add the device back in the hashes */
6530 list_netdevice(dev);
6532 /* Notify protocols, that a new device appeared. */
6533 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6536 * Prevent userspace races by waiting until the network
6537 * device is fully setup before sending notifications.
6539 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6546 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6548 static int dev_cpu_callback(struct notifier_block *nfb,
6549 unsigned long action,
6552 struct sk_buff **list_skb;
6553 struct sk_buff *skb;
6554 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6555 struct softnet_data *sd, *oldsd;
6557 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6560 local_irq_disable();
6561 cpu = smp_processor_id();
6562 sd = &per_cpu(softnet_data, cpu);
6563 oldsd = &per_cpu(softnet_data, oldcpu);
6565 /* Find end of our completion_queue. */
6566 list_skb = &sd->completion_queue;
6568 list_skb = &(*list_skb)->next;
6569 /* Append completion queue from offline CPU. */
6570 *list_skb = oldsd->completion_queue;
6571 oldsd->completion_queue = NULL;
6573 /* Append output queue from offline CPU. */
6574 if (oldsd->output_queue) {
6575 *sd->output_queue_tailp = oldsd->output_queue;
6576 sd->output_queue_tailp = oldsd->output_queue_tailp;
6577 oldsd->output_queue = NULL;
6578 oldsd->output_queue_tailp = &oldsd->output_queue;
6580 /* Append NAPI poll list from offline CPU. */
6581 if (!list_empty(&oldsd->poll_list)) {
6582 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6583 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6586 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6589 /* Process offline CPU's input_pkt_queue */
6590 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6592 input_queue_head_incr(oldsd);
6594 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6596 input_queue_head_incr(oldsd);
6604 * netdev_increment_features - increment feature set by one
6605 * @all: current feature set
6606 * @one: new feature set
6607 * @mask: mask feature set
6609 * Computes a new feature set after adding a device with feature set
6610 * @one to the master device with current feature set @all. Will not
6611 * enable anything that is off in @mask. Returns the new feature set.
6613 netdev_features_t netdev_increment_features(netdev_features_t all,
6614 netdev_features_t one, netdev_features_t mask)
6616 if (mask & NETIF_F_GEN_CSUM)
6617 mask |= NETIF_F_ALL_CSUM;
6618 mask |= NETIF_F_VLAN_CHALLENGED;
6620 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6621 all &= one | ~NETIF_F_ALL_FOR_ALL;
6623 /* If one device supports hw checksumming, set for all. */
6624 if (all & NETIF_F_GEN_CSUM)
6625 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6629 EXPORT_SYMBOL(netdev_increment_features);
6631 static struct hlist_head * __net_init netdev_create_hash(void)
6634 struct hlist_head *hash;
6636 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6638 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6639 INIT_HLIST_HEAD(&hash[i]);
6644 /* Initialize per network namespace state */
6645 static int __net_init netdev_init(struct net *net)
6647 if (net != &init_net)
6648 INIT_LIST_HEAD(&net->dev_base_head);
6650 net->dev_name_head = netdev_create_hash();
6651 if (net->dev_name_head == NULL)
6654 net->dev_index_head = netdev_create_hash();
6655 if (net->dev_index_head == NULL)
6661 kfree(net->dev_name_head);
6667 * netdev_drivername - network driver for the device
6668 * @dev: network device
6670 * Determine network driver for device.
6672 const char *netdev_drivername(const struct net_device *dev)
6674 const struct device_driver *driver;
6675 const struct device *parent;
6676 const char *empty = "";
6678 parent = dev->dev.parent;
6682 driver = parent->driver;
6683 if (driver && driver->name)
6684 return driver->name;
6688 static int __netdev_printk(const char *level, const struct net_device *dev,
6689 struct va_format *vaf)
6693 if (dev && dev->dev.parent) {
6694 r = dev_printk_emit(level[1] - '0',
6697 dev_driver_string(dev->dev.parent),
6698 dev_name(dev->dev.parent),
6699 netdev_name(dev), vaf);
6701 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6703 r = printk("%s(NULL net_device): %pV", level, vaf);
6709 int netdev_printk(const char *level, const struct net_device *dev,
6710 const char *format, ...)
6712 struct va_format vaf;
6716 va_start(args, format);
6721 r = __netdev_printk(level, dev, &vaf);
6727 EXPORT_SYMBOL(netdev_printk);
6729 #define define_netdev_printk_level(func, level) \
6730 int func(const struct net_device *dev, const char *fmt, ...) \
6733 struct va_format vaf; \
6736 va_start(args, fmt); \
6741 r = __netdev_printk(level, dev, &vaf); \
6747 EXPORT_SYMBOL(func);
6749 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6750 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6751 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6752 define_netdev_printk_level(netdev_err, KERN_ERR);
6753 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6754 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6755 define_netdev_printk_level(netdev_info, KERN_INFO);
6757 static void __net_exit netdev_exit(struct net *net)
6759 kfree(net->dev_name_head);
6760 kfree(net->dev_index_head);
6763 static struct pernet_operations __net_initdata netdev_net_ops = {
6764 .init = netdev_init,
6765 .exit = netdev_exit,
6768 static void __net_exit default_device_exit(struct net *net)
6770 struct net_device *dev, *aux;
6772 * Push all migratable network devices back to the
6773 * initial network namespace
6776 for_each_netdev_safe(net, dev, aux) {
6778 char fb_name[IFNAMSIZ];
6780 /* Ignore unmoveable devices (i.e. loopback) */
6781 if (dev->features & NETIF_F_NETNS_LOCAL)
6784 /* Leave virtual devices for the generic cleanup */
6785 if (dev->rtnl_link_ops)
6788 /* Push remaining network devices to init_net */
6789 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6790 err = dev_change_net_namespace(dev, &init_net, fb_name);
6792 pr_emerg("%s: failed to move %s to init_net: %d\n",
6793 __func__, dev->name, err);
6800 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6802 /* Return with the rtnl_lock held when there are no network
6803 * devices unregistering in any network namespace in net_list.
6810 prepare_to_wait(&netdev_unregistering_wq, &wait,
6811 TASK_UNINTERRUPTIBLE);
6812 unregistering = false;
6814 list_for_each_entry(net, net_list, exit_list) {
6815 if (net->dev_unreg_count > 0) {
6816 unregistering = true;
6825 finish_wait(&netdev_unregistering_wq, &wait);
6828 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6830 /* At exit all network devices most be removed from a network
6831 * namespace. Do this in the reverse order of registration.
6832 * Do this across as many network namespaces as possible to
6833 * improve batching efficiency.
6835 struct net_device *dev;
6837 LIST_HEAD(dev_kill_list);
6839 /* To prevent network device cleanup code from dereferencing
6840 * loopback devices or network devices that have been freed
6841 * wait here for all pending unregistrations to complete,
6842 * before unregistring the loopback device and allowing the
6843 * network namespace be freed.
6845 * The netdev todo list containing all network devices
6846 * unregistrations that happen in default_device_exit_batch
6847 * will run in the rtnl_unlock() at the end of
6848 * default_device_exit_batch.
6850 rtnl_lock_unregistering(net_list);
6851 list_for_each_entry(net, net_list, exit_list) {
6852 for_each_netdev_reverse(net, dev) {
6853 if (dev->rtnl_link_ops)
6854 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6856 unregister_netdevice_queue(dev, &dev_kill_list);
6859 unregister_netdevice_many(&dev_kill_list);
6860 list_del(&dev_kill_list);
6864 static struct pernet_operations __net_initdata default_device_ops = {
6865 .exit = default_device_exit,
6866 .exit_batch = default_device_exit_batch,
6870 * Initialize the DEV module. At boot time this walks the device list and
6871 * unhooks any devices that fail to initialise (normally hardware not
6872 * present) and leaves us with a valid list of present and active devices.
6877 * This is called single threaded during boot, so no need
6878 * to take the rtnl semaphore.
6880 static int __init net_dev_init(void)
6882 int i, rc = -ENOMEM;
6884 BUG_ON(!dev_boot_phase);
6886 if (dev_proc_init())
6889 if (netdev_kobject_init())
6892 INIT_LIST_HEAD(&ptype_all);
6893 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6894 INIT_LIST_HEAD(&ptype_base[i]);
6896 INIT_LIST_HEAD(&offload_base);
6898 if (register_pernet_subsys(&netdev_net_ops))
6902 * Initialise the packet receive queues.
6905 for_each_possible_cpu(i) {
6906 struct softnet_data *sd = &per_cpu(softnet_data, i);
6908 memset(sd, 0, sizeof(*sd));
6909 skb_queue_head_init(&sd->input_pkt_queue);
6910 skb_queue_head_init(&sd->process_queue);
6911 sd->completion_queue = NULL;
6912 INIT_LIST_HEAD(&sd->poll_list);
6913 sd->output_queue = NULL;
6914 sd->output_queue_tailp = &sd->output_queue;
6916 sd->csd.func = rps_trigger_softirq;
6922 sd->backlog.poll = process_backlog;
6923 sd->backlog.weight = weight_p;
6924 sd->backlog.gro_list = NULL;
6925 sd->backlog.gro_count = 0;
6927 #ifdef CONFIG_NET_FLOW_LIMIT
6928 sd->flow_limit = NULL;
6934 /* The loopback device is special if any other network devices
6935 * is present in a network namespace the loopback device must
6936 * be present. Since we now dynamically allocate and free the
6937 * loopback device ensure this invariant is maintained by
6938 * keeping the loopback device as the first device on the
6939 * list of network devices. Ensuring the loopback devices
6940 * is the first device that appears and the last network device
6943 if (register_pernet_device(&loopback_net_ops))
6946 if (register_pernet_device(&default_device_ops))
6949 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6950 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6952 hotcpu_notifier(dev_cpu_callback, 0);
6959 subsys_initcall(net_dev_init);