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, const struct cpumask *mask,
1922 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1923 struct xps_map *map, *new_map;
1924 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1925 int cpu, numa_node_id = -2;
1926 bool active = false;
1928 mutex_lock(&xps_map_mutex);
1930 dev_maps = xmap_dereference(dev->xps_maps);
1932 /* allocate memory for queue storage */
1933 for_each_online_cpu(cpu) {
1934 if (!cpumask_test_cpu(cpu, mask))
1938 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1939 if (!new_dev_maps) {
1940 mutex_unlock(&xps_map_mutex);
1944 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1947 map = expand_xps_map(map, cpu, index);
1951 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1955 goto out_no_new_maps;
1957 for_each_possible_cpu(cpu) {
1958 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1959 /* add queue to CPU maps */
1962 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1963 while ((pos < map->len) && (map->queues[pos] != index))
1966 if (pos == map->len)
1967 map->queues[map->len++] = index;
1969 if (numa_node_id == -2)
1970 numa_node_id = cpu_to_node(cpu);
1971 else if (numa_node_id != cpu_to_node(cpu))
1974 } else if (dev_maps) {
1975 /* fill in the new device map from the old device map */
1976 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1982 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1984 /* Cleanup old maps */
1986 for_each_possible_cpu(cpu) {
1987 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 if (map && map != new_map)
1990 kfree_rcu(map, rcu);
1993 kfree_rcu(dev_maps, rcu);
1996 dev_maps = new_dev_maps;
2000 /* update Tx queue numa node */
2001 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2002 (numa_node_id >= 0) ? numa_node_id :
2008 /* removes queue from unused CPUs */
2009 for_each_possible_cpu(cpu) {
2010 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2013 if (remove_xps_queue(dev_maps, cpu, index))
2017 /* free map if not active */
2019 RCU_INIT_POINTER(dev->xps_maps, NULL);
2020 kfree_rcu(dev_maps, rcu);
2024 mutex_unlock(&xps_map_mutex);
2028 /* remove any maps that we added */
2029 for_each_possible_cpu(cpu) {
2030 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2031 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2033 if (new_map && new_map != map)
2037 mutex_unlock(&xps_map_mutex);
2039 kfree(new_dev_maps);
2042 EXPORT_SYMBOL(netif_set_xps_queue);
2046 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2047 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2049 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2053 if (txq < 1 || txq > dev->num_tx_queues)
2056 if (dev->reg_state == NETREG_REGISTERED ||
2057 dev->reg_state == NETREG_UNREGISTERING) {
2060 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066 netif_setup_tc(dev, txq);
2068 if (txq < dev->real_num_tx_queues) {
2069 qdisc_reset_all_tx_gt(dev, txq);
2071 netif_reset_xps_queues_gt(dev, txq);
2076 dev->real_num_tx_queues = txq;
2079 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2083 * netif_set_real_num_rx_queues - set actual number of RX queues used
2084 * @dev: Network device
2085 * @rxq: Actual number of RX queues
2087 * This must be called either with the rtnl_lock held or before
2088 * registration of the net device. Returns 0 on success, or a
2089 * negative error code. If called before registration, it always
2092 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2096 if (rxq < 1 || rxq > dev->num_rx_queues)
2099 if (dev->reg_state == NETREG_REGISTERED) {
2102 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108 dev->real_num_rx_queues = rxq;
2111 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2115 * netif_get_num_default_rss_queues - default number of RSS queues
2117 * This routine should set an upper limit on the number of RSS queues
2118 * used by default by multiqueue devices.
2120 int netif_get_num_default_rss_queues(void)
2122 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2124 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2126 static inline void __netif_reschedule(struct Qdisc *q)
2128 struct softnet_data *sd;
2129 unsigned long flags;
2131 local_irq_save(flags);
2132 sd = &__get_cpu_var(softnet_data);
2133 q->next_sched = NULL;
2134 *sd->output_queue_tailp = q;
2135 sd->output_queue_tailp = &q->next_sched;
2136 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2137 local_irq_restore(flags);
2140 void __netif_schedule(struct Qdisc *q)
2142 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2143 __netif_reschedule(q);
2145 EXPORT_SYMBOL(__netif_schedule);
2147 void dev_kfree_skb_irq(struct sk_buff *skb)
2149 if (atomic_dec_and_test(&skb->users)) {
2150 struct softnet_data *sd;
2151 unsigned long flags;
2153 local_irq_save(flags);
2154 sd = &__get_cpu_var(softnet_data);
2155 skb->next = sd->completion_queue;
2156 sd->completion_queue = skb;
2157 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2158 local_irq_restore(flags);
2161 EXPORT_SYMBOL(dev_kfree_skb_irq);
2163 void dev_kfree_skb_any(struct sk_buff *skb)
2165 if (in_irq() || irqs_disabled())
2166 dev_kfree_skb_irq(skb);
2170 EXPORT_SYMBOL(dev_kfree_skb_any);
2174 * netif_device_detach - mark device as removed
2175 * @dev: network device
2177 * Mark device as removed from system and therefore no longer available.
2179 void netif_device_detach(struct net_device *dev)
2181 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2182 netif_running(dev)) {
2183 netif_tx_stop_all_queues(dev);
2186 EXPORT_SYMBOL(netif_device_detach);
2189 * netif_device_attach - mark device as attached
2190 * @dev: network device
2192 * Mark device as attached from system and restart if needed.
2194 void netif_device_attach(struct net_device *dev)
2196 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2197 netif_running(dev)) {
2198 netif_tx_wake_all_queues(dev);
2199 __netdev_watchdog_up(dev);
2202 EXPORT_SYMBOL(netif_device_attach);
2204 static void skb_warn_bad_offload(const struct sk_buff *skb)
2206 static const netdev_features_t null_features = 0;
2207 struct net_device *dev = skb->dev;
2208 const char *driver = "";
2210 if (!net_ratelimit())
2213 if (dev && dev->dev.parent)
2214 driver = dev_driver_string(dev->dev.parent);
2216 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2217 "gso_type=%d ip_summed=%d\n",
2218 driver, dev ? &dev->features : &null_features,
2219 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2220 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2221 skb_shinfo(skb)->gso_type, skb->ip_summed);
2225 * Invalidate hardware checksum when packet is to be mangled, and
2226 * complete checksum manually on outgoing path.
2228 int skb_checksum_help(struct sk_buff *skb)
2231 int ret = 0, offset;
2233 if (skb->ip_summed == CHECKSUM_COMPLETE)
2234 goto out_set_summed;
2236 if (unlikely(skb_shinfo(skb)->gso_size)) {
2237 skb_warn_bad_offload(skb);
2241 /* Before computing a checksum, we should make sure no frag could
2242 * be modified by an external entity : checksum could be wrong.
2244 if (skb_has_shared_frag(skb)) {
2245 ret = __skb_linearize(skb);
2250 offset = skb_checksum_start_offset(skb);
2251 BUG_ON(offset >= skb_headlen(skb));
2252 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2254 offset += skb->csum_offset;
2255 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2257 if (skb_cloned(skb) &&
2258 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2259 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2264 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2266 skb->ip_summed = CHECKSUM_NONE;
2270 EXPORT_SYMBOL(skb_checksum_help);
2272 __be16 skb_network_protocol(struct sk_buff *skb)
2274 __be16 type = skb->protocol;
2275 int vlan_depth = ETH_HLEN;
2277 /* Tunnel gso handlers can set protocol to ethernet. */
2278 if (type == htons(ETH_P_TEB)) {
2281 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2284 eth = (struct ethhdr *)skb_mac_header(skb);
2285 type = eth->h_proto;
2288 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2289 struct vlan_hdr *vh;
2291 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2294 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2295 type = vh->h_vlan_encapsulated_proto;
2296 vlan_depth += VLAN_HLEN;
2303 * skb_mac_gso_segment - mac layer segmentation handler.
2304 * @skb: buffer to segment
2305 * @features: features for the output path (see dev->features)
2307 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2308 netdev_features_t features)
2310 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2311 struct packet_offload *ptype;
2312 __be16 type = skb_network_protocol(skb);
2314 if (unlikely(!type))
2315 return ERR_PTR(-EINVAL);
2317 __skb_pull(skb, skb->mac_len);
2320 list_for_each_entry_rcu(ptype, &offload_base, list) {
2321 if (ptype->type == type && ptype->callbacks.gso_segment) {
2322 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2325 err = ptype->callbacks.gso_send_check(skb);
2326 segs = ERR_PTR(err);
2327 if (err || skb_gso_ok(skb, features))
2329 __skb_push(skb, (skb->data -
2330 skb_network_header(skb)));
2332 segs = ptype->callbacks.gso_segment(skb, features);
2338 __skb_push(skb, skb->data - skb_mac_header(skb));
2342 EXPORT_SYMBOL(skb_mac_gso_segment);
2345 /* openvswitch calls this on rx path, so we need a different check.
2347 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2350 return skb->ip_summed != CHECKSUM_PARTIAL;
2352 return skb->ip_summed == CHECKSUM_NONE;
2356 * __skb_gso_segment - Perform segmentation on skb.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2359 * @tx_path: whether it is called in TX path
2361 * This function segments the given skb and returns a list of segments.
2363 * It may return NULL if the skb requires no segmentation. This is
2364 * only possible when GSO is used for verifying header integrity.
2366 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2367 netdev_features_t features, bool tx_path)
2369 if (unlikely(skb_needs_check(skb, tx_path))) {
2372 skb_warn_bad_offload(skb);
2374 if (skb_header_cloned(skb) &&
2375 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2376 return ERR_PTR(err);
2379 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2380 skb_reset_mac_header(skb);
2381 skb_reset_mac_len(skb);
2383 return skb_mac_gso_segment(skb, features);
2385 EXPORT_SYMBOL(__skb_gso_segment);
2387 /* Take action when hardware reception checksum errors are detected. */
2389 void netdev_rx_csum_fault(struct net_device *dev)
2391 if (net_ratelimit()) {
2392 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2396 EXPORT_SYMBOL(netdev_rx_csum_fault);
2399 /* Actually, we should eliminate this check as soon as we know, that:
2400 * 1. IOMMU is present and allows to map all the memory.
2401 * 2. No high memory really exists on this machine.
2404 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2406 #ifdef CONFIG_HIGHMEM
2408 if (!(dev->features & NETIF_F_HIGHDMA)) {
2409 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2410 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2411 if (PageHighMem(skb_frag_page(frag)))
2416 if (PCI_DMA_BUS_IS_PHYS) {
2417 struct device *pdev = dev->dev.parent;
2421 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2422 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2423 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2424 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2433 void (*destructor)(struct sk_buff *skb);
2436 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2438 static void dev_gso_skb_destructor(struct sk_buff *skb)
2440 struct dev_gso_cb *cb;
2443 struct sk_buff *nskb = skb->next;
2445 skb->next = nskb->next;
2448 } while (skb->next);
2450 cb = DEV_GSO_CB(skb);
2452 cb->destructor(skb);
2456 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2457 * @skb: buffer to segment
2458 * @features: device features as applicable to this skb
2460 * This function segments the given skb and stores the list of segments
2463 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2465 struct sk_buff *segs;
2467 segs = skb_gso_segment(skb, features);
2469 /* Verifying header integrity only. */
2474 return PTR_ERR(segs);
2477 DEV_GSO_CB(skb)->destructor = skb->destructor;
2478 skb->destructor = dev_gso_skb_destructor;
2483 static netdev_features_t harmonize_features(struct sk_buff *skb,
2484 netdev_features_t features)
2486 if (skb->ip_summed != CHECKSUM_NONE &&
2487 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2488 features &= ~NETIF_F_ALL_CSUM;
2489 } else if (illegal_highdma(skb->dev, skb)) {
2490 features &= ~NETIF_F_SG;
2496 netdev_features_t netif_skb_features(struct sk_buff *skb)
2498 __be16 protocol = skb->protocol;
2499 netdev_features_t features = skb->dev->features;
2501 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2502 features &= ~NETIF_F_GSO_MASK;
2504 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2505 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2506 protocol = veh->h_vlan_encapsulated_proto;
2507 } else if (!vlan_tx_tag_present(skb)) {
2508 return harmonize_features(skb, features);
2511 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2512 NETIF_F_HW_VLAN_STAG_TX);
2514 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2515 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2516 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2517 NETIF_F_HW_VLAN_STAG_TX;
2519 return harmonize_features(skb, features);
2521 EXPORT_SYMBOL(netif_skb_features);
2524 * Returns true if either:
2525 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2526 * 2. skb is fragmented and the device does not support SG.
2528 static inline int skb_needs_linearize(struct sk_buff *skb,
2529 netdev_features_t features)
2531 return skb_is_nonlinear(skb) &&
2532 ((skb_has_frag_list(skb) &&
2533 !(features & NETIF_F_FRAGLIST)) ||
2534 (skb_shinfo(skb)->nr_frags &&
2535 !(features & NETIF_F_SG)));
2538 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2539 struct netdev_queue *txq)
2541 const struct net_device_ops *ops = dev->netdev_ops;
2542 int rc = NETDEV_TX_OK;
2543 unsigned int skb_len;
2545 if (likely(!skb->next)) {
2546 netdev_features_t features;
2549 * If device doesn't need skb->dst, release it right now while
2550 * its hot in this cpu cache
2552 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2555 features = netif_skb_features(skb);
2557 if (vlan_tx_tag_present(skb) &&
2558 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2559 skb = __vlan_put_tag(skb, skb->vlan_proto,
2560 vlan_tx_tag_get(skb));
2567 /* If encapsulation offload request, verify we are testing
2568 * hardware encapsulation features instead of standard
2569 * features for the netdev
2571 if (skb->encapsulation)
2572 features &= dev->hw_enc_features;
2574 if (netif_needs_gso(skb, features)) {
2575 if (unlikely(dev_gso_segment(skb, features)))
2580 if (skb_needs_linearize(skb, features) &&
2581 __skb_linearize(skb))
2584 /* If packet is not checksummed and device does not
2585 * support checksumming for this protocol, complete
2586 * checksumming here.
2588 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2589 if (skb->encapsulation)
2590 skb_set_inner_transport_header(skb,
2591 skb_checksum_start_offset(skb));
2593 skb_set_transport_header(skb,
2594 skb_checksum_start_offset(skb));
2595 if (!(features & NETIF_F_ALL_CSUM) &&
2596 skb_checksum_help(skb))
2601 if (!list_empty(&ptype_all))
2602 dev_queue_xmit_nit(skb, dev);
2605 rc = ops->ndo_start_xmit(skb, dev);
2606 trace_net_dev_xmit(skb, rc, dev, skb_len);
2607 if (rc == NETDEV_TX_OK)
2608 txq_trans_update(txq);
2614 struct sk_buff *nskb = skb->next;
2616 skb->next = nskb->next;
2619 if (!list_empty(&ptype_all))
2620 dev_queue_xmit_nit(nskb, dev);
2622 skb_len = nskb->len;
2623 rc = ops->ndo_start_xmit(nskb, dev);
2624 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2625 if (unlikely(rc != NETDEV_TX_OK)) {
2626 if (rc & ~NETDEV_TX_MASK)
2627 goto out_kfree_gso_skb;
2628 nskb->next = skb->next;
2632 txq_trans_update(txq);
2633 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2634 return NETDEV_TX_BUSY;
2635 } while (skb->next);
2638 if (likely(skb->next == NULL)) {
2639 skb->destructor = DEV_GSO_CB(skb)->destructor;
2649 static void qdisc_pkt_len_init(struct sk_buff *skb)
2651 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2653 qdisc_skb_cb(skb)->pkt_len = skb->len;
2655 /* To get more precise estimation of bytes sent on wire,
2656 * we add to pkt_len the headers size of all segments
2658 if (shinfo->gso_size) {
2659 unsigned int hdr_len;
2660 u16 gso_segs = shinfo->gso_segs;
2662 /* mac layer + network layer */
2663 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2665 /* + transport layer */
2666 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2667 hdr_len += tcp_hdrlen(skb);
2669 hdr_len += sizeof(struct udphdr);
2671 if (shinfo->gso_type & SKB_GSO_DODGY)
2672 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2675 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2679 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2680 struct net_device *dev,
2681 struct netdev_queue *txq)
2683 spinlock_t *root_lock = qdisc_lock(q);
2687 qdisc_pkt_len_init(skb);
2688 qdisc_calculate_pkt_len(skb, q);
2690 * Heuristic to force contended enqueues to serialize on a
2691 * separate lock before trying to get qdisc main lock.
2692 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2693 * and dequeue packets faster.
2695 contended = qdisc_is_running(q);
2696 if (unlikely(contended))
2697 spin_lock(&q->busylock);
2699 spin_lock(root_lock);
2700 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2703 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2704 qdisc_run_begin(q)) {
2706 * This is a work-conserving queue; there are no old skbs
2707 * waiting to be sent out; and the qdisc is not running -
2708 * xmit the skb directly.
2710 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2713 qdisc_bstats_update(q, skb);
2715 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2716 if (unlikely(contended)) {
2717 spin_unlock(&q->busylock);
2724 rc = NET_XMIT_SUCCESS;
2727 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2728 if (qdisc_run_begin(q)) {
2729 if (unlikely(contended)) {
2730 spin_unlock(&q->busylock);
2736 spin_unlock(root_lock);
2737 if (unlikely(contended))
2738 spin_unlock(&q->busylock);
2742 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2743 static void skb_update_prio(struct sk_buff *skb)
2745 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2747 if (!skb->priority && skb->sk && map) {
2748 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2750 if (prioidx < map->priomap_len)
2751 skb->priority = map->priomap[prioidx];
2755 #define skb_update_prio(skb)
2758 static DEFINE_PER_CPU(int, xmit_recursion);
2759 #define RECURSION_LIMIT 10
2762 * dev_loopback_xmit - loop back @skb
2763 * @skb: buffer to transmit
2765 int dev_loopback_xmit(struct sk_buff *skb)
2767 skb_reset_mac_header(skb);
2768 __skb_pull(skb, skb_network_offset(skb));
2769 skb->pkt_type = PACKET_LOOPBACK;
2770 skb->ip_summed = CHECKSUM_UNNECESSARY;
2771 WARN_ON(!skb_dst(skb));
2776 EXPORT_SYMBOL(dev_loopback_xmit);
2779 * dev_queue_xmit - transmit a buffer
2780 * @skb: buffer to transmit
2782 * Queue a buffer for transmission to a network device. The caller must
2783 * have set the device and priority and built the buffer before calling
2784 * this function. The function can be called from an interrupt.
2786 * A negative errno code is returned on a failure. A success does not
2787 * guarantee the frame will be transmitted as it may be dropped due
2788 * to congestion or traffic shaping.
2790 * -----------------------------------------------------------------------------------
2791 * I notice this method can also return errors from the queue disciplines,
2792 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2795 * Regardless of the return value, the skb is consumed, so it is currently
2796 * difficult to retry a send to this method. (You can bump the ref count
2797 * before sending to hold a reference for retry if you are careful.)
2799 * When calling this method, interrupts MUST be enabled. This is because
2800 * the BH enable code must have IRQs enabled so that it will not deadlock.
2803 int dev_queue_xmit(struct sk_buff *skb)
2805 struct net_device *dev = skb->dev;
2806 struct netdev_queue *txq;
2810 skb_reset_mac_header(skb);
2812 /* Disable soft irqs for various locks below. Also
2813 * stops preemption for RCU.
2817 skb_update_prio(skb);
2819 txq = netdev_pick_tx(dev, skb);
2820 q = rcu_dereference_bh(txq->qdisc);
2822 #ifdef CONFIG_NET_CLS_ACT
2823 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2825 trace_net_dev_queue(skb);
2827 rc = __dev_xmit_skb(skb, q, dev, txq);
2831 /* The device has no queue. Common case for software devices:
2832 loopback, all the sorts of tunnels...
2834 Really, it is unlikely that netif_tx_lock protection is necessary
2835 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2837 However, it is possible, that they rely on protection
2840 Check this and shot the lock. It is not prone from deadlocks.
2841 Either shot noqueue qdisc, it is even simpler 8)
2843 if (dev->flags & IFF_UP) {
2844 int cpu = smp_processor_id(); /* ok because BHs are off */
2846 if (txq->xmit_lock_owner != cpu) {
2848 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2849 goto recursion_alert;
2851 HARD_TX_LOCK(dev, txq, cpu);
2853 if (!netif_xmit_stopped(txq)) {
2854 __this_cpu_inc(xmit_recursion);
2855 rc = dev_hard_start_xmit(skb, dev, txq);
2856 __this_cpu_dec(xmit_recursion);
2857 if (dev_xmit_complete(rc)) {
2858 HARD_TX_UNLOCK(dev, txq);
2862 HARD_TX_UNLOCK(dev, txq);
2863 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2866 /* Recursion is detected! It is possible,
2870 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2876 rcu_read_unlock_bh();
2881 rcu_read_unlock_bh();
2884 EXPORT_SYMBOL(dev_queue_xmit);
2887 /*=======================================================================
2889 =======================================================================*/
2891 int netdev_max_backlog __read_mostly = 1000;
2892 EXPORT_SYMBOL(netdev_max_backlog);
2894 int netdev_tstamp_prequeue __read_mostly = 1;
2895 int netdev_budget __read_mostly = 300;
2896 int weight_p __read_mostly = 64; /* old backlog weight */
2898 /* Called with irq disabled */
2899 static inline void ____napi_schedule(struct softnet_data *sd,
2900 struct napi_struct *napi)
2902 list_add_tail(&napi->poll_list, &sd->poll_list);
2903 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2908 /* One global table that all flow-based protocols share. */
2909 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2910 EXPORT_SYMBOL(rps_sock_flow_table);
2912 struct static_key rps_needed __read_mostly;
2914 static struct rps_dev_flow *
2915 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2916 struct rps_dev_flow *rflow, u16 next_cpu)
2918 if (next_cpu != RPS_NO_CPU) {
2919 #ifdef CONFIG_RFS_ACCEL
2920 struct netdev_rx_queue *rxqueue;
2921 struct rps_dev_flow_table *flow_table;
2922 struct rps_dev_flow *old_rflow;
2927 /* Should we steer this flow to a different hardware queue? */
2928 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2929 !(dev->features & NETIF_F_NTUPLE))
2931 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2932 if (rxq_index == skb_get_rx_queue(skb))
2935 rxqueue = dev->_rx + rxq_index;
2936 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2939 flow_id = skb->rxhash & flow_table->mask;
2940 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2941 rxq_index, flow_id);
2945 rflow = &flow_table->flows[flow_id];
2947 if (old_rflow->filter == rflow->filter)
2948 old_rflow->filter = RPS_NO_FILTER;
2952 per_cpu(softnet_data, next_cpu).input_queue_head;
2955 rflow->cpu = next_cpu;
2960 * get_rps_cpu is called from netif_receive_skb and returns the target
2961 * CPU from the RPS map of the receiving queue for a given skb.
2962 * rcu_read_lock must be held on entry.
2964 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2965 struct rps_dev_flow **rflowp)
2967 struct netdev_rx_queue *rxqueue;
2968 struct rps_map *map;
2969 struct rps_dev_flow_table *flow_table;
2970 struct rps_sock_flow_table *sock_flow_table;
2974 if (skb_rx_queue_recorded(skb)) {
2975 u16 index = skb_get_rx_queue(skb);
2976 if (unlikely(index >= dev->real_num_rx_queues)) {
2977 WARN_ONCE(dev->real_num_rx_queues > 1,
2978 "%s received packet on queue %u, but number "
2979 "of RX queues is %u\n",
2980 dev->name, index, dev->real_num_rx_queues);
2983 rxqueue = dev->_rx + index;
2987 map = rcu_dereference(rxqueue->rps_map);
2989 if (map->len == 1 &&
2990 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2991 tcpu = map->cpus[0];
2992 if (cpu_online(tcpu))
2996 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3000 skb_reset_network_header(skb);
3001 if (!skb_get_rxhash(skb))
3004 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3005 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3006 if (flow_table && sock_flow_table) {
3008 struct rps_dev_flow *rflow;
3010 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3013 next_cpu = sock_flow_table->ents[skb->rxhash &
3014 sock_flow_table->mask];
3017 * If the desired CPU (where last recvmsg was done) is
3018 * different from current CPU (one in the rx-queue flow
3019 * table entry), switch if one of the following holds:
3020 * - Current CPU is unset (equal to RPS_NO_CPU).
3021 * - Current CPU is offline.
3022 * - The current CPU's queue tail has advanced beyond the
3023 * last packet that was enqueued using this table entry.
3024 * This guarantees that all previous packets for the flow
3025 * have been dequeued, thus preserving in order delivery.
3027 if (unlikely(tcpu != next_cpu) &&
3028 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3029 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3030 rflow->last_qtail)) >= 0)) {
3032 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3035 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3043 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3045 if (cpu_online(tcpu)) {
3055 #ifdef CONFIG_RFS_ACCEL
3058 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3059 * @dev: Device on which the filter was set
3060 * @rxq_index: RX queue index
3061 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3062 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3064 * Drivers that implement ndo_rx_flow_steer() should periodically call
3065 * this function for each installed filter and remove the filters for
3066 * which it returns %true.
3068 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3069 u32 flow_id, u16 filter_id)
3071 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3072 struct rps_dev_flow_table *flow_table;
3073 struct rps_dev_flow *rflow;
3078 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3079 if (flow_table && flow_id <= flow_table->mask) {
3080 rflow = &flow_table->flows[flow_id];
3081 cpu = ACCESS_ONCE(rflow->cpu);
3082 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3083 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3084 rflow->last_qtail) <
3085 (int)(10 * flow_table->mask)))
3091 EXPORT_SYMBOL(rps_may_expire_flow);
3093 #endif /* CONFIG_RFS_ACCEL */
3095 /* Called from hardirq (IPI) context */
3096 static void rps_trigger_softirq(void *data)
3098 struct softnet_data *sd = data;
3100 ____napi_schedule(sd, &sd->backlog);
3104 #endif /* CONFIG_RPS */
3107 * Check if this softnet_data structure is another cpu one
3108 * If yes, queue it to our IPI list and return 1
3111 static int rps_ipi_queued(struct softnet_data *sd)
3114 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3117 sd->rps_ipi_next = mysd->rps_ipi_list;
3118 mysd->rps_ipi_list = sd;
3120 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3123 #endif /* CONFIG_RPS */
3127 #ifdef CONFIG_NET_FLOW_LIMIT
3128 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3131 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3133 #ifdef CONFIG_NET_FLOW_LIMIT
3134 struct sd_flow_limit *fl;
3135 struct softnet_data *sd;
3136 unsigned int old_flow, new_flow;
3138 if (qlen < (netdev_max_backlog >> 1))
3141 sd = &__get_cpu_var(softnet_data);
3144 fl = rcu_dereference(sd->flow_limit);
3146 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3147 old_flow = fl->history[fl->history_head];
3148 fl->history[fl->history_head] = new_flow;
3151 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3153 if (likely(fl->buckets[old_flow]))
3154 fl->buckets[old_flow]--;
3156 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3168 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3169 * queue (may be a remote CPU queue).
3171 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3172 unsigned int *qtail)
3174 struct softnet_data *sd;
3175 unsigned long flags;
3178 sd = &per_cpu(softnet_data, cpu);
3180 local_irq_save(flags);
3183 qlen = skb_queue_len(&sd->input_pkt_queue);
3184 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3185 if (skb_queue_len(&sd->input_pkt_queue)) {
3187 __skb_queue_tail(&sd->input_pkt_queue, skb);
3188 input_queue_tail_incr_save(sd, qtail);
3190 local_irq_restore(flags);
3191 return NET_RX_SUCCESS;
3194 /* Schedule NAPI for backlog device
3195 * We can use non atomic operation since we own the queue lock
3197 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3198 if (!rps_ipi_queued(sd))
3199 ____napi_schedule(sd, &sd->backlog);
3207 local_irq_restore(flags);
3209 atomic_long_inc(&skb->dev->rx_dropped);
3215 * netif_rx - post buffer to the network code
3216 * @skb: buffer to post
3218 * This function receives a packet from a device driver and queues it for
3219 * the upper (protocol) levels to process. It always succeeds. The buffer
3220 * may be dropped during processing for congestion control or by the
3224 * NET_RX_SUCCESS (no congestion)
3225 * NET_RX_DROP (packet was dropped)
3229 int netif_rx(struct sk_buff *skb)
3233 /* if netpoll wants it, pretend we never saw it */
3234 if (netpoll_rx(skb))
3237 net_timestamp_check(netdev_tstamp_prequeue, skb);
3239 trace_netif_rx(skb);
3241 if (static_key_false(&rps_needed)) {
3242 struct rps_dev_flow voidflow, *rflow = &voidflow;
3248 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3250 cpu = smp_processor_id();
3252 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3260 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3265 EXPORT_SYMBOL(netif_rx);
3267 int netif_rx_ni(struct sk_buff *skb)
3272 err = netif_rx(skb);
3273 if (local_softirq_pending())
3279 EXPORT_SYMBOL(netif_rx_ni);
3281 static void net_tx_action(struct softirq_action *h)
3283 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3285 if (sd->completion_queue) {
3286 struct sk_buff *clist;
3288 local_irq_disable();
3289 clist = sd->completion_queue;
3290 sd->completion_queue = NULL;
3294 struct sk_buff *skb = clist;
3295 clist = clist->next;
3297 WARN_ON(atomic_read(&skb->users));
3298 trace_kfree_skb(skb, net_tx_action);
3303 if (sd->output_queue) {
3306 local_irq_disable();
3307 head = sd->output_queue;
3308 sd->output_queue = NULL;
3309 sd->output_queue_tailp = &sd->output_queue;
3313 struct Qdisc *q = head;
3314 spinlock_t *root_lock;
3316 head = head->next_sched;
3318 root_lock = qdisc_lock(q);
3319 if (spin_trylock(root_lock)) {
3320 smp_mb__before_clear_bit();
3321 clear_bit(__QDISC_STATE_SCHED,
3324 spin_unlock(root_lock);
3326 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3328 __netif_reschedule(q);
3330 smp_mb__before_clear_bit();
3331 clear_bit(__QDISC_STATE_SCHED,
3339 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3340 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3341 /* This hook is defined here for ATM LANE */
3342 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3343 unsigned char *addr) __read_mostly;
3344 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3347 #ifdef CONFIG_NET_CLS_ACT
3348 /* TODO: Maybe we should just force sch_ingress to be compiled in
3349 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3350 * a compare and 2 stores extra right now if we dont have it on
3351 * but have CONFIG_NET_CLS_ACT
3352 * NOTE: This doesn't stop any functionality; if you dont have
3353 * the ingress scheduler, you just can't add policies on ingress.
3356 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3358 struct net_device *dev = skb->dev;
3359 u32 ttl = G_TC_RTTL(skb->tc_verd);
3360 int result = TC_ACT_OK;
3363 if (unlikely(MAX_RED_LOOP < ttl++)) {
3364 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3365 skb->skb_iif, dev->ifindex);
3369 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3370 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3373 if (q != &noop_qdisc) {
3374 spin_lock(qdisc_lock(q));
3375 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3376 result = qdisc_enqueue_root(skb, q);
3377 spin_unlock(qdisc_lock(q));
3383 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3384 struct packet_type **pt_prev,
3385 int *ret, struct net_device *orig_dev)
3387 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3389 if (!rxq || rxq->qdisc == &noop_qdisc)
3393 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3397 switch (ing_filter(skb, rxq)) {
3411 * netdev_rx_handler_register - register receive handler
3412 * @dev: device to register a handler for
3413 * @rx_handler: receive handler to register
3414 * @rx_handler_data: data pointer that is used by rx handler
3416 * Register a receive hander for a device. This handler will then be
3417 * called from __netif_receive_skb. A negative errno code is returned
3420 * The caller must hold the rtnl_mutex.
3422 * For a general description of rx_handler, see enum rx_handler_result.
3424 int netdev_rx_handler_register(struct net_device *dev,
3425 rx_handler_func_t *rx_handler,
3426 void *rx_handler_data)
3430 if (dev->rx_handler)
3433 /* Note: rx_handler_data must be set before rx_handler */
3434 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3435 rcu_assign_pointer(dev->rx_handler, rx_handler);
3439 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3442 * netdev_rx_handler_unregister - unregister receive handler
3443 * @dev: device to unregister a handler from
3445 * Unregister a receive handler from a device.
3447 * The caller must hold the rtnl_mutex.
3449 void netdev_rx_handler_unregister(struct net_device *dev)
3453 RCU_INIT_POINTER(dev->rx_handler, NULL);
3454 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3455 * section has a guarantee to see a non NULL rx_handler_data
3459 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3461 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3464 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3465 * the special handling of PFMEMALLOC skbs.
3467 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3469 switch (skb->protocol) {
3470 case __constant_htons(ETH_P_ARP):
3471 case __constant_htons(ETH_P_IP):
3472 case __constant_htons(ETH_P_IPV6):
3473 case __constant_htons(ETH_P_8021Q):
3474 case __constant_htons(ETH_P_8021AD):
3481 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3483 struct packet_type *ptype, *pt_prev;
3484 rx_handler_func_t *rx_handler;
3485 struct net_device *orig_dev;
3486 struct net_device *null_or_dev;
3487 bool deliver_exact = false;
3488 int ret = NET_RX_DROP;
3491 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3493 trace_netif_receive_skb(skb);
3495 /* if we've gotten here through NAPI, check netpoll */
3496 if (netpoll_receive_skb(skb))
3499 orig_dev = skb->dev;
3501 skb_reset_network_header(skb);
3502 if (!skb_transport_header_was_set(skb))
3503 skb_reset_transport_header(skb);
3504 skb_reset_mac_len(skb);
3511 skb->skb_iif = skb->dev->ifindex;
3513 __this_cpu_inc(softnet_data.processed);
3515 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3516 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3517 skb = vlan_untag(skb);
3522 #ifdef CONFIG_NET_CLS_ACT
3523 if (skb->tc_verd & TC_NCLS) {
3524 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3532 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3533 if (!ptype->dev || ptype->dev == skb->dev) {
3535 ret = deliver_skb(skb, pt_prev, orig_dev);
3541 #ifdef CONFIG_NET_CLS_ACT
3542 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3548 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3551 if (vlan_tx_tag_present(skb)) {
3553 ret = deliver_skb(skb, pt_prev, orig_dev);
3556 if (vlan_do_receive(&skb))
3558 else if (unlikely(!skb))
3562 rx_handler = rcu_dereference(skb->dev->rx_handler);
3565 ret = deliver_skb(skb, pt_prev, orig_dev);
3568 switch (rx_handler(&skb)) {
3569 case RX_HANDLER_CONSUMED:
3570 ret = NET_RX_SUCCESS;
3572 case RX_HANDLER_ANOTHER:
3574 case RX_HANDLER_EXACT:
3575 deliver_exact = true;
3576 case RX_HANDLER_PASS:
3583 if (unlikely(vlan_tx_tag_present(skb))) {
3584 if (vlan_tx_tag_get_id(skb))
3585 skb->pkt_type = PACKET_OTHERHOST;
3586 /* Note: we might in the future use prio bits
3587 * and set skb->priority like in vlan_do_receive()
3588 * For the time being, just ignore Priority Code Point
3593 /* deliver only exact match when indicated */
3594 null_or_dev = deliver_exact ? skb->dev : NULL;
3596 type = skb->protocol;
3597 list_for_each_entry_rcu(ptype,
3598 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3599 if (ptype->type == type &&
3600 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3601 ptype->dev == orig_dev)) {
3603 ret = deliver_skb(skb, pt_prev, orig_dev);
3609 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3612 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3615 atomic_long_inc(&skb->dev->rx_dropped);
3617 /* Jamal, now you will not able to escape explaining
3618 * me how you were going to use this. :-)
3629 static int __netif_receive_skb(struct sk_buff *skb)
3633 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3634 unsigned long pflags = current->flags;
3637 * PFMEMALLOC skbs are special, they should
3638 * - be delivered to SOCK_MEMALLOC sockets only
3639 * - stay away from userspace
3640 * - have bounded memory usage
3642 * Use PF_MEMALLOC as this saves us from propagating the allocation
3643 * context down to all allocation sites.
3645 current->flags |= PF_MEMALLOC;
3646 ret = __netif_receive_skb_core(skb, true);
3647 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3649 ret = __netif_receive_skb_core(skb, false);
3655 * netif_receive_skb - process receive buffer from network
3656 * @skb: buffer to process
3658 * netif_receive_skb() is the main receive data processing function.
3659 * It always succeeds. The buffer may be dropped during processing
3660 * for congestion control or by the protocol layers.
3662 * This function may only be called from softirq context and interrupts
3663 * should be enabled.
3665 * Return values (usually ignored):
3666 * NET_RX_SUCCESS: no congestion
3667 * NET_RX_DROP: packet was dropped
3669 int netif_receive_skb(struct sk_buff *skb)
3671 net_timestamp_check(netdev_tstamp_prequeue, skb);
3673 if (skb_defer_rx_timestamp(skb))
3674 return NET_RX_SUCCESS;
3677 if (static_key_false(&rps_needed)) {
3678 struct rps_dev_flow voidflow, *rflow = &voidflow;
3683 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3686 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3693 return __netif_receive_skb(skb);
3695 EXPORT_SYMBOL(netif_receive_skb);
3697 /* Network device is going away, flush any packets still pending
3698 * Called with irqs disabled.
3700 static void flush_backlog(void *arg)
3702 struct net_device *dev = arg;
3703 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3704 struct sk_buff *skb, *tmp;
3707 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3708 if (skb->dev == dev) {
3709 __skb_unlink(skb, &sd->input_pkt_queue);
3711 input_queue_head_incr(sd);
3716 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3717 if (skb->dev == dev) {
3718 __skb_unlink(skb, &sd->process_queue);
3720 input_queue_head_incr(sd);
3725 static int napi_gro_complete(struct sk_buff *skb)
3727 struct packet_offload *ptype;
3728 __be16 type = skb->protocol;
3729 struct list_head *head = &offload_base;
3732 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3734 if (NAPI_GRO_CB(skb)->count == 1) {
3735 skb_shinfo(skb)->gso_size = 0;
3740 list_for_each_entry_rcu(ptype, head, list) {
3741 if (ptype->type != type || !ptype->callbacks.gro_complete)
3744 err = ptype->callbacks.gro_complete(skb);
3750 WARN_ON(&ptype->list == head);
3752 return NET_RX_SUCCESS;
3756 return netif_receive_skb(skb);
3759 /* napi->gro_list contains packets ordered by age.
3760 * youngest packets at the head of it.
3761 * Complete skbs in reverse order to reduce latencies.
3763 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3765 struct sk_buff *skb, *prev = NULL;
3767 /* scan list and build reverse chain */
3768 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3773 for (skb = prev; skb; skb = prev) {
3776 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3780 napi_gro_complete(skb);
3784 napi->gro_list = NULL;
3786 EXPORT_SYMBOL(napi_gro_flush);
3788 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3791 unsigned int maclen = skb->dev->hard_header_len;
3793 for (p = napi->gro_list; p; p = p->next) {
3794 unsigned long diffs;
3796 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3797 diffs |= p->vlan_tci ^ skb->vlan_tci;
3798 if (maclen == ETH_HLEN)
3799 diffs |= compare_ether_header(skb_mac_header(p),
3800 skb_gro_mac_header(skb));
3802 diffs = memcmp(skb_mac_header(p),
3803 skb_gro_mac_header(skb),
3805 NAPI_GRO_CB(p)->same_flow = !diffs;
3806 NAPI_GRO_CB(p)->flush = 0;
3810 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3812 struct sk_buff **pp = NULL;
3813 struct packet_offload *ptype;
3814 __be16 type = skb->protocol;
3815 struct list_head *head = &offload_base;
3817 enum gro_result ret;
3819 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3822 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3825 gro_list_prepare(napi, skb);
3828 list_for_each_entry_rcu(ptype, head, list) {
3829 if (ptype->type != type || !ptype->callbacks.gro_receive)
3832 skb_set_network_header(skb, skb_gro_offset(skb));
3833 skb_reset_mac_len(skb);
3834 NAPI_GRO_CB(skb)->same_flow = 0;
3835 NAPI_GRO_CB(skb)->flush = 0;
3836 NAPI_GRO_CB(skb)->free = 0;
3838 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3843 if (&ptype->list == head)
3846 same_flow = NAPI_GRO_CB(skb)->same_flow;
3847 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3850 struct sk_buff *nskb = *pp;
3854 napi_gro_complete(nskb);
3861 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3865 NAPI_GRO_CB(skb)->count = 1;
3866 NAPI_GRO_CB(skb)->age = jiffies;
3867 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3868 skb->next = napi->gro_list;
3869 napi->gro_list = skb;
3873 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3874 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3876 BUG_ON(skb->end - skb->tail < grow);
3878 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3881 skb->data_len -= grow;
3883 skb_shinfo(skb)->frags[0].page_offset += grow;
3884 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3886 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3887 skb_frag_unref(skb, 0);
3888 memmove(skb_shinfo(skb)->frags,
3889 skb_shinfo(skb)->frags + 1,
3890 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3903 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3907 if (netif_receive_skb(skb))
3915 case GRO_MERGED_FREE:
3916 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3917 kmem_cache_free(skbuff_head_cache, skb);
3930 static void skb_gro_reset_offset(struct sk_buff *skb)
3932 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3933 const skb_frag_t *frag0 = &pinfo->frags[0];
3935 NAPI_GRO_CB(skb)->data_offset = 0;
3936 NAPI_GRO_CB(skb)->frag0 = NULL;
3937 NAPI_GRO_CB(skb)->frag0_len = 0;
3939 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3941 !PageHighMem(skb_frag_page(frag0))) {
3942 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3943 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3947 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3949 skb_gro_reset_offset(skb);
3951 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3953 EXPORT_SYMBOL(napi_gro_receive);
3955 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3957 __skb_pull(skb, skb_headlen(skb));
3958 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3959 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3961 skb->dev = napi->dev;
3967 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3969 struct sk_buff *skb = napi->skb;
3972 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3978 EXPORT_SYMBOL(napi_get_frags);
3980 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3986 skb->protocol = eth_type_trans(skb, skb->dev);
3988 if (ret == GRO_HELD)
3989 skb_gro_pull(skb, -ETH_HLEN);
3990 else if (netif_receive_skb(skb))
3995 case GRO_MERGED_FREE:
3996 napi_reuse_skb(napi, skb);
4006 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4008 struct sk_buff *skb = napi->skb;
4015 skb_reset_mac_header(skb);
4016 skb_gro_reset_offset(skb);
4018 off = skb_gro_offset(skb);
4019 hlen = off + sizeof(*eth);
4020 eth = skb_gro_header_fast(skb, off);
4021 if (skb_gro_header_hard(skb, hlen)) {
4022 eth = skb_gro_header_slow(skb, hlen, off);
4023 if (unlikely(!eth)) {
4024 napi_reuse_skb(napi, skb);
4030 skb_gro_pull(skb, sizeof(*eth));
4033 * This works because the only protocols we care about don't require
4034 * special handling. We'll fix it up properly at the end.
4036 skb->protocol = eth->h_proto;
4042 gro_result_t napi_gro_frags(struct napi_struct *napi)
4044 struct sk_buff *skb = napi_frags_skb(napi);
4049 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4051 EXPORT_SYMBOL(napi_gro_frags);
4054 * net_rps_action sends any pending IPI's for rps.
4055 * Note: called with local irq disabled, but exits with local irq enabled.
4057 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4060 struct softnet_data *remsd = sd->rps_ipi_list;
4063 sd->rps_ipi_list = NULL;
4067 /* Send pending IPI's to kick RPS processing on remote cpus. */
4069 struct softnet_data *next = remsd->rps_ipi_next;
4071 if (cpu_online(remsd->cpu))
4072 __smp_call_function_single(remsd->cpu,
4081 static int process_backlog(struct napi_struct *napi, int quota)
4084 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4087 /* Check if we have pending ipi, its better to send them now,
4088 * not waiting net_rx_action() end.
4090 if (sd->rps_ipi_list) {
4091 local_irq_disable();
4092 net_rps_action_and_irq_enable(sd);
4095 napi->weight = weight_p;
4096 local_irq_disable();
4097 while (work < quota) {
4098 struct sk_buff *skb;
4101 while ((skb = __skb_dequeue(&sd->process_queue))) {
4103 __netif_receive_skb(skb);
4104 local_irq_disable();
4105 input_queue_head_incr(sd);
4106 if (++work >= quota) {
4113 qlen = skb_queue_len(&sd->input_pkt_queue);
4115 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4116 &sd->process_queue);
4118 if (qlen < quota - work) {
4120 * Inline a custom version of __napi_complete().
4121 * only current cpu owns and manipulates this napi,
4122 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4123 * we can use a plain write instead of clear_bit(),
4124 * and we dont need an smp_mb() memory barrier.
4126 list_del(&napi->poll_list);
4129 quota = work + qlen;
4139 * __napi_schedule - schedule for receive
4140 * @n: entry to schedule
4142 * The entry's receive function will be scheduled to run
4144 void __napi_schedule(struct napi_struct *n)
4146 unsigned long flags;
4148 local_irq_save(flags);
4149 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4150 local_irq_restore(flags);
4152 EXPORT_SYMBOL(__napi_schedule);
4154 void __napi_complete(struct napi_struct *n)
4156 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4157 BUG_ON(n->gro_list);
4159 list_del(&n->poll_list);
4160 smp_mb__before_clear_bit();
4161 clear_bit(NAPI_STATE_SCHED, &n->state);
4163 EXPORT_SYMBOL(__napi_complete);
4165 void napi_complete(struct napi_struct *n)
4167 unsigned long flags;
4170 * don't let napi dequeue from the cpu poll list
4171 * just in case its running on a different cpu
4173 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4176 napi_gro_flush(n, false);
4177 local_irq_save(flags);
4179 local_irq_restore(flags);
4181 EXPORT_SYMBOL(napi_complete);
4183 /* must be called under rcu_read_lock(), as we dont take a reference */
4184 struct napi_struct *napi_by_id(unsigned int napi_id)
4186 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4187 struct napi_struct *napi;
4189 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4190 if (napi->napi_id == napi_id)
4195 EXPORT_SYMBOL_GPL(napi_by_id);
4197 void napi_hash_add(struct napi_struct *napi)
4199 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4201 spin_lock(&napi_hash_lock);
4203 /* 0 is not a valid id, we also skip an id that is taken
4204 * we expect both events to be extremely rare
4207 while (!napi->napi_id) {
4208 napi->napi_id = ++napi_gen_id;
4209 if (napi_by_id(napi->napi_id))
4213 hlist_add_head_rcu(&napi->napi_hash_node,
4214 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4216 spin_unlock(&napi_hash_lock);
4219 EXPORT_SYMBOL_GPL(napi_hash_add);
4221 /* Warning : caller is responsible to make sure rcu grace period
4222 * is respected before freeing memory containing @napi
4224 void napi_hash_del(struct napi_struct *napi)
4226 spin_lock(&napi_hash_lock);
4228 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4229 hlist_del_rcu(&napi->napi_hash_node);
4231 spin_unlock(&napi_hash_lock);
4233 EXPORT_SYMBOL_GPL(napi_hash_del);
4235 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4236 int (*poll)(struct napi_struct *, int), int weight)
4238 INIT_LIST_HEAD(&napi->poll_list);
4239 napi->gro_count = 0;
4240 napi->gro_list = NULL;
4243 if (weight > NAPI_POLL_WEIGHT)
4244 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4246 napi->weight = weight;
4247 list_add(&napi->dev_list, &dev->napi_list);
4249 #ifdef CONFIG_NETPOLL
4250 spin_lock_init(&napi->poll_lock);
4251 napi->poll_owner = -1;
4253 set_bit(NAPI_STATE_SCHED, &napi->state);
4255 EXPORT_SYMBOL(netif_napi_add);
4257 void netif_napi_del(struct napi_struct *napi)
4259 struct sk_buff *skb, *next;
4261 list_del_init(&napi->dev_list);
4262 napi_free_frags(napi);
4264 for (skb = napi->gro_list; skb; skb = next) {
4270 napi->gro_list = NULL;
4271 napi->gro_count = 0;
4273 EXPORT_SYMBOL(netif_napi_del);
4275 static void net_rx_action(struct softirq_action *h)
4277 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4278 unsigned long time_limit = jiffies + 2;
4279 int budget = netdev_budget;
4282 local_irq_disable();
4284 while (!list_empty(&sd->poll_list)) {
4285 struct napi_struct *n;
4288 /* If softirq window is exhuasted then punt.
4289 * Allow this to run for 2 jiffies since which will allow
4290 * an average latency of 1.5/HZ.
4292 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4297 /* Even though interrupts have been re-enabled, this
4298 * access is safe because interrupts can only add new
4299 * entries to the tail of this list, and only ->poll()
4300 * calls can remove this head entry from the list.
4302 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4304 have = netpoll_poll_lock(n);
4308 /* This NAPI_STATE_SCHED test is for avoiding a race
4309 * with netpoll's poll_napi(). Only the entity which
4310 * obtains the lock and sees NAPI_STATE_SCHED set will
4311 * actually make the ->poll() call. Therefore we avoid
4312 * accidentally calling ->poll() when NAPI is not scheduled.
4315 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4316 work = n->poll(n, weight);
4320 WARN_ON_ONCE(work > weight);
4324 local_irq_disable();
4326 /* Drivers must not modify the NAPI state if they
4327 * consume the entire weight. In such cases this code
4328 * still "owns" the NAPI instance and therefore can
4329 * move the instance around on the list at-will.
4331 if (unlikely(work == weight)) {
4332 if (unlikely(napi_disable_pending(n))) {
4335 local_irq_disable();
4338 /* flush too old packets
4339 * If HZ < 1000, flush all packets.
4342 napi_gro_flush(n, HZ >= 1000);
4343 local_irq_disable();
4345 list_move_tail(&n->poll_list, &sd->poll_list);
4349 netpoll_poll_unlock(have);
4352 net_rps_action_and_irq_enable(sd);
4354 #ifdef CONFIG_NET_DMA
4356 * There may not be any more sk_buffs coming right now, so push
4357 * any pending DMA copies to hardware
4359 dma_issue_pending_all();
4366 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4370 struct netdev_adjacent {
4371 struct net_device *dev;
4373 /* upper master flag, there can only be one master device per list */
4376 /* counter for the number of times this device was added to us */
4379 /* private field for the users */
4382 struct list_head list;
4383 struct rcu_head rcu;
4386 static struct netdev_adjacent *__netdev_find_adj_rcu(struct net_device *dev,
4387 struct net_device *adj_dev,
4388 struct list_head *adj_list)
4390 struct netdev_adjacent *adj;
4392 list_for_each_entry_rcu(adj, adj_list, list) {
4393 if (adj->dev == adj_dev)
4399 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4400 struct net_device *adj_dev,
4401 struct list_head *adj_list)
4403 struct netdev_adjacent *adj;
4405 list_for_each_entry(adj, adj_list, list) {
4406 if (adj->dev == adj_dev)
4413 * netdev_has_upper_dev - Check if device is linked to an upper device
4415 * @upper_dev: upper device to check
4417 * Find out if a device is linked to specified upper device and return true
4418 * in case it is. Note that this checks only immediate upper device,
4419 * not through a complete stack of devices. The caller must hold the RTNL lock.
4421 bool netdev_has_upper_dev(struct net_device *dev,
4422 struct net_device *upper_dev)
4426 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4428 EXPORT_SYMBOL(netdev_has_upper_dev);
4431 * netdev_has_any_upper_dev - Check if device is linked to some device
4434 * Find out if a device is linked to an upper device and return true in case
4435 * it is. The caller must hold the RTNL lock.
4437 bool netdev_has_any_upper_dev(struct net_device *dev)
4441 return !list_empty(&dev->all_adj_list.upper);
4443 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4446 * netdev_master_upper_dev_get - Get master upper device
4449 * Find a master upper device and return pointer to it or NULL in case
4450 * it's not there. The caller must hold the RTNL lock.
4452 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4454 struct netdev_adjacent *upper;
4458 if (list_empty(&dev->adj_list.upper))
4461 upper = list_first_entry(&dev->adj_list.upper,
4462 struct netdev_adjacent, list);
4463 if (likely(upper->master))
4467 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4469 void *netdev_adjacent_get_private(struct list_head *adj_list)
4471 struct netdev_adjacent *adj;
4473 adj = list_entry(adj_list, struct netdev_adjacent, list);
4475 return adj->private;
4477 EXPORT_SYMBOL(netdev_adjacent_get_private);
4480 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4482 * @iter: list_head ** of the current position
4484 * Gets the next device from the dev's upper list, starting from iter
4485 * position. The caller must hold RCU read lock.
4487 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4488 struct list_head **iter)
4490 struct netdev_adjacent *upper;
4492 WARN_ON_ONCE(!rcu_read_lock_held());
4494 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4496 if (&upper->list == &dev->all_adj_list.upper)
4499 *iter = &upper->list;
4503 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4506 * netdev_lower_get_next_private - Get the next ->private from the
4507 * lower neighbour list
4509 * @iter: list_head ** of the current position
4511 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4512 * list, starting from iter position. The caller must hold either hold the
4513 * RTNL lock or its own locking that guarantees that the neighbour lower
4514 * list will remain unchainged.
4516 void *netdev_lower_get_next_private(struct net_device *dev,
4517 struct list_head **iter)
4519 struct netdev_adjacent *lower;
4521 lower = list_entry(*iter, struct netdev_adjacent, list);
4523 if (&lower->list == &dev->adj_list.lower)
4527 *iter = lower->list.next;
4529 return lower->private;
4531 EXPORT_SYMBOL(netdev_lower_get_next_private);
4534 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4535 * lower neighbour list, RCU
4538 * @iter: list_head ** of the current position
4540 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4541 * list, starting from iter position. The caller must hold RCU read lock.
4543 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4544 struct list_head **iter)
4546 struct netdev_adjacent *lower;
4548 WARN_ON_ONCE(!rcu_read_lock_held());
4550 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4552 if (&lower->list == &dev->adj_list.lower)
4556 *iter = &lower->list;
4558 return lower->private;
4560 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4563 * netdev_master_upper_dev_get_rcu - Get master upper device
4566 * Find a master upper device and return pointer to it or NULL in case
4567 * it's not there. The caller must hold the RCU read lock.
4569 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4571 struct netdev_adjacent *upper;
4573 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4574 struct netdev_adjacent, list);
4575 if (upper && likely(upper->master))
4579 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4581 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4582 struct net_device *adj_dev,
4583 struct list_head *dev_list,
4584 void *private, bool master)
4586 struct netdev_adjacent *adj;
4587 char linkname[IFNAMSIZ+7];
4590 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4597 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4602 adj->master = master;
4604 adj->private = private;
4607 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4608 adj_dev->name, dev->name, adj_dev->name);
4610 if (dev_list == &dev->adj_list.lower) {
4611 sprintf(linkname, "lower_%s", adj_dev->name);
4612 ret = sysfs_create_link(&(dev->dev.kobj),
4613 &(adj_dev->dev.kobj), linkname);
4616 } else if (dev_list == &dev->adj_list.upper) {
4617 sprintf(linkname, "upper_%s", adj_dev->name);
4618 ret = sysfs_create_link(&(dev->dev.kobj),
4619 &(adj_dev->dev.kobj), linkname);
4624 /* Ensure that master link is always the first item in list. */
4626 ret = sysfs_create_link(&(dev->dev.kobj),
4627 &(adj_dev->dev.kobj), "master");
4629 goto remove_symlinks;
4631 list_add_rcu(&adj->list, dev_list);
4633 list_add_tail_rcu(&adj->list, dev_list);
4639 if (dev_list == &dev->adj_list.lower) {
4640 sprintf(linkname, "lower_%s", adj_dev->name);
4641 sysfs_remove_link(&(dev->dev.kobj), linkname);
4642 } else if (dev_list == &dev->adj_list.upper) {
4643 sprintf(linkname, "upper_%s", adj_dev->name);
4644 sysfs_remove_link(&(dev->dev.kobj), linkname);
4653 void __netdev_adjacent_dev_remove(struct net_device *dev,
4654 struct net_device *adj_dev,
4655 struct list_head *dev_list)
4657 struct netdev_adjacent *adj;
4658 char linkname[IFNAMSIZ+7];
4660 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4663 pr_err("tried to remove device %s from %s\n",
4664 dev->name, adj_dev->name);
4668 if (adj->ref_nr > 1) {
4669 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4676 sysfs_remove_link(&(dev->dev.kobj), "master");
4678 if (dev_list == &dev->adj_list.lower) {
4679 sprintf(linkname, "lower_%s", adj_dev->name);
4680 sysfs_remove_link(&(dev->dev.kobj), linkname);
4681 } else if (dev_list == &dev->adj_list.upper) {
4682 sprintf(linkname, "upper_%s", adj_dev->name);
4683 sysfs_remove_link(&(dev->dev.kobj), linkname);
4686 list_del_rcu(&adj->list);
4687 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4688 adj_dev->name, dev->name, adj_dev->name);
4690 kfree_rcu(adj, rcu);
4693 int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4694 struct net_device *upper_dev,
4695 struct list_head *up_list,
4696 struct list_head *down_list,
4697 void *private, bool master)
4701 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4706 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4709 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4716 int __netdev_adjacent_dev_link(struct net_device *dev,
4717 struct net_device *upper_dev)
4719 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4720 &dev->all_adj_list.upper,
4721 &upper_dev->all_adj_list.lower,
4725 void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4726 struct net_device *upper_dev,
4727 struct list_head *up_list,
4728 struct list_head *down_list)
4730 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4731 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4734 void __netdev_adjacent_dev_unlink(struct net_device *dev,
4735 struct net_device *upper_dev)
4737 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4738 &dev->all_adj_list.upper,
4739 &upper_dev->all_adj_list.lower);
4742 int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4743 struct net_device *upper_dev,
4744 void *private, bool master)
4746 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4751 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4752 &dev->adj_list.upper,
4753 &upper_dev->adj_list.lower,
4756 __netdev_adjacent_dev_unlink(dev, upper_dev);
4763 void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4764 struct net_device *upper_dev)
4766 __netdev_adjacent_dev_unlink(dev, upper_dev);
4767 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4768 &dev->adj_list.upper,
4769 &upper_dev->adj_list.lower);
4772 static int __netdev_upper_dev_link(struct net_device *dev,
4773 struct net_device *upper_dev, bool master,
4776 struct netdev_adjacent *i, *j, *to_i, *to_j;
4781 if (dev == upper_dev)
4784 /* To prevent loops, check if dev is not upper device to upper_dev. */
4785 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4788 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4791 if (master && netdev_master_upper_dev_get(dev))
4794 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4799 /* Now that we linked these devs, make all the upper_dev's
4800 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4801 * versa, and don't forget the devices itself. All of these
4802 * links are non-neighbours.
4804 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4805 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4806 pr_debug("Interlinking %s with %s, non-neighbour\n",
4807 i->dev->name, j->dev->name);
4808 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4814 /* add dev to every upper_dev's upper device */
4815 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4816 pr_debug("linking %s's upper device %s with %s\n",
4817 upper_dev->name, i->dev->name, dev->name);
4818 ret = __netdev_adjacent_dev_link(dev, i->dev);
4820 goto rollback_upper_mesh;
4823 /* add upper_dev to every dev's lower device */
4824 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4825 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4826 i->dev->name, upper_dev->name);
4827 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4829 goto rollback_lower_mesh;
4832 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4835 rollback_lower_mesh:
4837 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4840 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4845 rollback_upper_mesh:
4847 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4850 __netdev_adjacent_dev_unlink(dev, i->dev);
4858 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4859 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4860 if (i == to_i && j == to_j)
4862 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4868 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4874 * netdev_upper_dev_link - Add a link to the upper device
4876 * @upper_dev: new upper device
4878 * Adds a link to device which is upper to this one. The caller must hold
4879 * the RTNL lock. On a failure a negative errno code is returned.
4880 * On success the reference counts are adjusted and the function
4883 int netdev_upper_dev_link(struct net_device *dev,
4884 struct net_device *upper_dev)
4886 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4888 EXPORT_SYMBOL(netdev_upper_dev_link);
4891 * netdev_master_upper_dev_link - Add a master link to the upper device
4893 * @upper_dev: new upper device
4895 * Adds a link to device which is upper to this one. In this case, only
4896 * one master upper device can be linked, although other non-master devices
4897 * might be linked as well. The caller must hold the RTNL lock.
4898 * On a failure a negative errno code is returned. On success the reference
4899 * counts are adjusted and the function returns zero.
4901 int netdev_master_upper_dev_link(struct net_device *dev,
4902 struct net_device *upper_dev)
4904 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4906 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4908 int netdev_master_upper_dev_link_private(struct net_device *dev,
4909 struct net_device *upper_dev,
4912 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4914 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4917 * netdev_upper_dev_unlink - Removes a link to upper device
4919 * @upper_dev: new upper device
4921 * Removes a link to device which is upper to this one. The caller must hold
4924 void netdev_upper_dev_unlink(struct net_device *dev,
4925 struct net_device *upper_dev)
4927 struct netdev_adjacent *i, *j;
4930 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4932 /* Here is the tricky part. We must remove all dev's lower
4933 * devices from all upper_dev's upper devices and vice
4934 * versa, to maintain the graph relationship.
4936 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4937 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4938 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4940 /* remove also the devices itself from lower/upper device
4943 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4944 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4946 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4947 __netdev_adjacent_dev_unlink(dev, i->dev);
4949 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4951 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4953 void *netdev_lower_dev_get_private_rcu(struct net_device *dev,
4954 struct net_device *lower_dev)
4956 struct netdev_adjacent *lower;
4960 lower = __netdev_find_adj_rcu(dev, lower_dev, &dev->adj_list.lower);
4964 return lower->private;
4966 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu);
4968 void *netdev_lower_dev_get_private(struct net_device *dev,
4969 struct net_device *lower_dev)
4971 struct netdev_adjacent *lower;
4975 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
4979 return lower->private;
4981 EXPORT_SYMBOL(netdev_lower_dev_get_private);
4983 static void dev_change_rx_flags(struct net_device *dev, int flags)
4985 const struct net_device_ops *ops = dev->netdev_ops;
4987 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4988 ops->ndo_change_rx_flags(dev, flags);
4991 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
4993 unsigned int old_flags = dev->flags;
4999 dev->flags |= IFF_PROMISC;
5000 dev->promiscuity += inc;
5001 if (dev->promiscuity == 0) {
5004 * If inc causes overflow, untouch promisc and return error.
5007 dev->flags &= ~IFF_PROMISC;
5009 dev->promiscuity -= inc;
5010 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5015 if (dev->flags != old_flags) {
5016 pr_info("device %s %s promiscuous mode\n",
5018 dev->flags & IFF_PROMISC ? "entered" : "left");
5019 if (audit_enabled) {
5020 current_uid_gid(&uid, &gid);
5021 audit_log(current->audit_context, GFP_ATOMIC,
5022 AUDIT_ANOM_PROMISCUOUS,
5023 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5024 dev->name, (dev->flags & IFF_PROMISC),
5025 (old_flags & IFF_PROMISC),
5026 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5027 from_kuid(&init_user_ns, uid),
5028 from_kgid(&init_user_ns, gid),
5029 audit_get_sessionid(current));
5032 dev_change_rx_flags(dev, IFF_PROMISC);
5035 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5040 * dev_set_promiscuity - update promiscuity count on a device
5044 * Add or remove promiscuity from a device. While the count in the device
5045 * remains above zero the interface remains promiscuous. Once it hits zero
5046 * the device reverts back to normal filtering operation. A negative inc
5047 * value is used to drop promiscuity on the device.
5048 * Return 0 if successful or a negative errno code on error.
5050 int dev_set_promiscuity(struct net_device *dev, int inc)
5052 unsigned int old_flags = dev->flags;
5055 err = __dev_set_promiscuity(dev, inc, true);
5058 if (dev->flags != old_flags)
5059 dev_set_rx_mode(dev);
5062 EXPORT_SYMBOL(dev_set_promiscuity);
5064 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5066 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5070 dev->flags |= IFF_ALLMULTI;
5071 dev->allmulti += inc;
5072 if (dev->allmulti == 0) {
5075 * If inc causes overflow, untouch allmulti and return error.
5078 dev->flags &= ~IFF_ALLMULTI;
5080 dev->allmulti -= inc;
5081 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5086 if (dev->flags ^ old_flags) {
5087 dev_change_rx_flags(dev, IFF_ALLMULTI);
5088 dev_set_rx_mode(dev);
5090 __dev_notify_flags(dev, old_flags,
5091 dev->gflags ^ old_gflags);
5097 * dev_set_allmulti - update allmulti count on a device
5101 * Add or remove reception of all multicast frames to a device. While the
5102 * count in the device remains above zero the interface remains listening
5103 * to all interfaces. Once it hits zero the device reverts back to normal
5104 * filtering operation. A negative @inc value is used to drop the counter
5105 * when releasing a resource needing all multicasts.
5106 * Return 0 if successful or a negative errno code on error.
5109 int dev_set_allmulti(struct net_device *dev, int inc)
5111 return __dev_set_allmulti(dev, inc, true);
5113 EXPORT_SYMBOL(dev_set_allmulti);
5116 * Upload unicast and multicast address lists to device and
5117 * configure RX filtering. When the device doesn't support unicast
5118 * filtering it is put in promiscuous mode while unicast addresses
5121 void __dev_set_rx_mode(struct net_device *dev)
5123 const struct net_device_ops *ops = dev->netdev_ops;
5125 /* dev_open will call this function so the list will stay sane. */
5126 if (!(dev->flags&IFF_UP))
5129 if (!netif_device_present(dev))
5132 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5133 /* Unicast addresses changes may only happen under the rtnl,
5134 * therefore calling __dev_set_promiscuity here is safe.
5136 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5137 __dev_set_promiscuity(dev, 1, false);
5138 dev->uc_promisc = true;
5139 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5140 __dev_set_promiscuity(dev, -1, false);
5141 dev->uc_promisc = false;
5145 if (ops->ndo_set_rx_mode)
5146 ops->ndo_set_rx_mode(dev);
5149 void dev_set_rx_mode(struct net_device *dev)
5151 netif_addr_lock_bh(dev);
5152 __dev_set_rx_mode(dev);
5153 netif_addr_unlock_bh(dev);
5157 * dev_get_flags - get flags reported to userspace
5160 * Get the combination of flag bits exported through APIs to userspace.
5162 unsigned int dev_get_flags(const struct net_device *dev)
5166 flags = (dev->flags & ~(IFF_PROMISC |
5171 (dev->gflags & (IFF_PROMISC |
5174 if (netif_running(dev)) {
5175 if (netif_oper_up(dev))
5176 flags |= IFF_RUNNING;
5177 if (netif_carrier_ok(dev))
5178 flags |= IFF_LOWER_UP;
5179 if (netif_dormant(dev))
5180 flags |= IFF_DORMANT;
5185 EXPORT_SYMBOL(dev_get_flags);
5187 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5189 unsigned int old_flags = dev->flags;
5195 * Set the flags on our device.
5198 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5199 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5201 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5205 * Load in the correct multicast list now the flags have changed.
5208 if ((old_flags ^ flags) & IFF_MULTICAST)
5209 dev_change_rx_flags(dev, IFF_MULTICAST);
5211 dev_set_rx_mode(dev);
5214 * Have we downed the interface. We handle IFF_UP ourselves
5215 * according to user attempts to set it, rather than blindly
5220 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5221 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5224 dev_set_rx_mode(dev);
5227 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5228 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5229 unsigned int old_flags = dev->flags;
5231 dev->gflags ^= IFF_PROMISC;
5233 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5234 if (dev->flags != old_flags)
5235 dev_set_rx_mode(dev);
5238 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5239 is important. Some (broken) drivers set IFF_PROMISC, when
5240 IFF_ALLMULTI is requested not asking us and not reporting.
5242 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5243 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5245 dev->gflags ^= IFF_ALLMULTI;
5246 __dev_set_allmulti(dev, inc, false);
5252 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5253 unsigned int gchanges)
5255 unsigned int changes = dev->flags ^ old_flags;
5258 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges);
5260 if (changes & IFF_UP) {
5261 if (dev->flags & IFF_UP)
5262 call_netdevice_notifiers(NETDEV_UP, dev);
5264 call_netdevice_notifiers(NETDEV_DOWN, dev);
5267 if (dev->flags & IFF_UP &&
5268 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5269 struct netdev_notifier_change_info change_info;
5271 change_info.flags_changed = changes;
5272 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5278 * dev_change_flags - change device settings
5280 * @flags: device state flags
5282 * Change settings on device based state flags. The flags are
5283 * in the userspace exported format.
5285 int dev_change_flags(struct net_device *dev, unsigned int flags)
5288 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5290 ret = __dev_change_flags(dev, flags);
5294 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5295 __dev_notify_flags(dev, old_flags, changes);
5298 EXPORT_SYMBOL(dev_change_flags);
5301 * dev_set_mtu - Change maximum transfer unit
5303 * @new_mtu: new transfer unit
5305 * Change the maximum transfer size of the network device.
5307 int dev_set_mtu(struct net_device *dev, int new_mtu)
5309 const struct net_device_ops *ops = dev->netdev_ops;
5312 if (new_mtu == dev->mtu)
5315 /* MTU must be positive. */
5319 if (!netif_device_present(dev))
5323 if (ops->ndo_change_mtu)
5324 err = ops->ndo_change_mtu(dev, new_mtu);
5329 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5332 EXPORT_SYMBOL(dev_set_mtu);
5335 * dev_set_group - Change group this device belongs to
5337 * @new_group: group this device should belong to
5339 void dev_set_group(struct net_device *dev, int new_group)
5341 dev->group = new_group;
5343 EXPORT_SYMBOL(dev_set_group);
5346 * dev_set_mac_address - Change Media Access Control Address
5350 * Change the hardware (MAC) address of the device
5352 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5354 const struct net_device_ops *ops = dev->netdev_ops;
5357 if (!ops->ndo_set_mac_address)
5359 if (sa->sa_family != dev->type)
5361 if (!netif_device_present(dev))
5363 err = ops->ndo_set_mac_address(dev, sa);
5366 dev->addr_assign_type = NET_ADDR_SET;
5367 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5368 add_device_randomness(dev->dev_addr, dev->addr_len);
5371 EXPORT_SYMBOL(dev_set_mac_address);
5374 * dev_change_carrier - Change device carrier
5376 * @new_carrier: new value
5378 * Change device carrier
5380 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5382 const struct net_device_ops *ops = dev->netdev_ops;
5384 if (!ops->ndo_change_carrier)
5386 if (!netif_device_present(dev))
5388 return ops->ndo_change_carrier(dev, new_carrier);
5390 EXPORT_SYMBOL(dev_change_carrier);
5393 * dev_get_phys_port_id - Get device physical port ID
5397 * Get device physical port ID
5399 int dev_get_phys_port_id(struct net_device *dev,
5400 struct netdev_phys_port_id *ppid)
5402 const struct net_device_ops *ops = dev->netdev_ops;
5404 if (!ops->ndo_get_phys_port_id)
5406 return ops->ndo_get_phys_port_id(dev, ppid);
5408 EXPORT_SYMBOL(dev_get_phys_port_id);
5411 * dev_new_index - allocate an ifindex
5412 * @net: the applicable net namespace
5414 * Returns a suitable unique value for a new device interface
5415 * number. The caller must hold the rtnl semaphore or the
5416 * dev_base_lock to be sure it remains unique.
5418 static int dev_new_index(struct net *net)
5420 int ifindex = net->ifindex;
5424 if (!__dev_get_by_index(net, ifindex))
5425 return net->ifindex = ifindex;
5429 /* Delayed registration/unregisteration */
5430 static LIST_HEAD(net_todo_list);
5431 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5433 static void net_set_todo(struct net_device *dev)
5435 list_add_tail(&dev->todo_list, &net_todo_list);
5436 dev_net(dev)->dev_unreg_count++;
5439 static void rollback_registered_many(struct list_head *head)
5441 struct net_device *dev, *tmp;
5442 LIST_HEAD(close_head);
5444 BUG_ON(dev_boot_phase);
5447 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5448 /* Some devices call without registering
5449 * for initialization unwind. Remove those
5450 * devices and proceed with the remaining.
5452 if (dev->reg_state == NETREG_UNINITIALIZED) {
5453 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5457 list_del(&dev->unreg_list);
5460 dev->dismantle = true;
5461 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5464 /* If device is running, close it first. */
5465 list_for_each_entry(dev, head, unreg_list)
5466 list_add_tail(&dev->close_list, &close_head);
5467 dev_close_many(&close_head);
5469 list_for_each_entry(dev, head, unreg_list) {
5470 /* And unlink it from device chain. */
5471 unlist_netdevice(dev);
5473 dev->reg_state = NETREG_UNREGISTERING;
5478 list_for_each_entry(dev, head, unreg_list) {
5479 /* Shutdown queueing discipline. */
5483 /* Notify protocols, that we are about to destroy
5484 this device. They should clean all the things.
5486 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5488 if (!dev->rtnl_link_ops ||
5489 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5490 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5493 * Flush the unicast and multicast chains
5498 if (dev->netdev_ops->ndo_uninit)
5499 dev->netdev_ops->ndo_uninit(dev);
5501 /* Notifier chain MUST detach us all upper devices. */
5502 WARN_ON(netdev_has_any_upper_dev(dev));
5504 /* Remove entries from kobject tree */
5505 netdev_unregister_kobject(dev);
5507 /* Remove XPS queueing entries */
5508 netif_reset_xps_queues_gt(dev, 0);
5514 list_for_each_entry(dev, head, unreg_list)
5518 static void rollback_registered(struct net_device *dev)
5522 list_add(&dev->unreg_list, &single);
5523 rollback_registered_many(&single);
5527 static netdev_features_t netdev_fix_features(struct net_device *dev,
5528 netdev_features_t features)
5530 /* Fix illegal checksum combinations */
5531 if ((features & NETIF_F_HW_CSUM) &&
5532 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5533 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5534 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5537 /* TSO requires that SG is present as well. */
5538 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5539 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5540 features &= ~NETIF_F_ALL_TSO;
5543 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5544 !(features & NETIF_F_IP_CSUM)) {
5545 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5546 features &= ~NETIF_F_TSO;
5547 features &= ~NETIF_F_TSO_ECN;
5550 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5551 !(features & NETIF_F_IPV6_CSUM)) {
5552 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5553 features &= ~NETIF_F_TSO6;
5556 /* TSO ECN requires that TSO is present as well. */
5557 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5558 features &= ~NETIF_F_TSO_ECN;
5560 /* Software GSO depends on SG. */
5561 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5562 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5563 features &= ~NETIF_F_GSO;
5566 /* UFO needs SG and checksumming */
5567 if (features & NETIF_F_UFO) {
5568 /* maybe split UFO into V4 and V6? */
5569 if (!((features & NETIF_F_GEN_CSUM) ||
5570 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5571 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5573 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5574 features &= ~NETIF_F_UFO;
5577 if (!(features & NETIF_F_SG)) {
5579 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5580 features &= ~NETIF_F_UFO;
5587 int __netdev_update_features(struct net_device *dev)
5589 netdev_features_t features;
5594 features = netdev_get_wanted_features(dev);
5596 if (dev->netdev_ops->ndo_fix_features)
5597 features = dev->netdev_ops->ndo_fix_features(dev, features);
5599 /* driver might be less strict about feature dependencies */
5600 features = netdev_fix_features(dev, features);
5602 if (dev->features == features)
5605 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5606 &dev->features, &features);
5608 if (dev->netdev_ops->ndo_set_features)
5609 err = dev->netdev_ops->ndo_set_features(dev, features);
5611 if (unlikely(err < 0)) {
5613 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5614 err, &features, &dev->features);
5619 dev->features = features;
5625 * netdev_update_features - recalculate device features
5626 * @dev: the device to check
5628 * Recalculate dev->features set and send notifications if it
5629 * has changed. Should be called after driver or hardware dependent
5630 * conditions might have changed that influence the features.
5632 void netdev_update_features(struct net_device *dev)
5634 if (__netdev_update_features(dev))
5635 netdev_features_change(dev);
5637 EXPORT_SYMBOL(netdev_update_features);
5640 * netdev_change_features - recalculate device features
5641 * @dev: the device to check
5643 * Recalculate dev->features set and send notifications even
5644 * if they have not changed. Should be called instead of
5645 * netdev_update_features() if also dev->vlan_features might
5646 * have changed to allow the changes to be propagated to stacked
5649 void netdev_change_features(struct net_device *dev)
5651 __netdev_update_features(dev);
5652 netdev_features_change(dev);
5654 EXPORT_SYMBOL(netdev_change_features);
5657 * netif_stacked_transfer_operstate - transfer operstate
5658 * @rootdev: the root or lower level device to transfer state from
5659 * @dev: the device to transfer operstate to
5661 * Transfer operational state from root to device. This is normally
5662 * called when a stacking relationship exists between the root
5663 * device and the device(a leaf device).
5665 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5666 struct net_device *dev)
5668 if (rootdev->operstate == IF_OPER_DORMANT)
5669 netif_dormant_on(dev);
5671 netif_dormant_off(dev);
5673 if (netif_carrier_ok(rootdev)) {
5674 if (!netif_carrier_ok(dev))
5675 netif_carrier_on(dev);
5677 if (netif_carrier_ok(dev))
5678 netif_carrier_off(dev);
5681 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5684 static int netif_alloc_rx_queues(struct net_device *dev)
5686 unsigned int i, count = dev->num_rx_queues;
5687 struct netdev_rx_queue *rx;
5691 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5697 for (i = 0; i < count; i++)
5703 static void netdev_init_one_queue(struct net_device *dev,
5704 struct netdev_queue *queue, void *_unused)
5706 /* Initialize queue lock */
5707 spin_lock_init(&queue->_xmit_lock);
5708 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5709 queue->xmit_lock_owner = -1;
5710 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5713 dql_init(&queue->dql, HZ);
5717 static void netif_free_tx_queues(struct net_device *dev)
5719 if (is_vmalloc_addr(dev->_tx))
5725 static int netif_alloc_netdev_queues(struct net_device *dev)
5727 unsigned int count = dev->num_tx_queues;
5728 struct netdev_queue *tx;
5729 size_t sz = count * sizeof(*tx);
5731 BUG_ON(count < 1 || count > 0xffff);
5733 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5741 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5742 spin_lock_init(&dev->tx_global_lock);
5748 * register_netdevice - register a network device
5749 * @dev: device to register
5751 * Take a completed network device structure and add it to the kernel
5752 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5753 * chain. 0 is returned on success. A negative errno code is returned
5754 * on a failure to set up the device, or if the name is a duplicate.
5756 * Callers must hold the rtnl semaphore. You may want
5757 * register_netdev() instead of this.
5760 * The locking appears insufficient to guarantee two parallel registers
5761 * will not get the same name.
5764 int register_netdevice(struct net_device *dev)
5767 struct net *net = dev_net(dev);
5769 BUG_ON(dev_boot_phase);
5774 /* When net_device's are persistent, this will be fatal. */
5775 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5778 spin_lock_init(&dev->addr_list_lock);
5779 netdev_set_addr_lockdep_class(dev);
5783 ret = dev_get_valid_name(net, dev, dev->name);
5787 /* Init, if this function is available */
5788 if (dev->netdev_ops->ndo_init) {
5789 ret = dev->netdev_ops->ndo_init(dev);
5797 if (((dev->hw_features | dev->features) &
5798 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5799 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5800 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5801 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5808 dev->ifindex = dev_new_index(net);
5809 else if (__dev_get_by_index(net, dev->ifindex))
5812 if (dev->iflink == -1)
5813 dev->iflink = dev->ifindex;
5815 /* Transfer changeable features to wanted_features and enable
5816 * software offloads (GSO and GRO).
5818 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5819 dev->features |= NETIF_F_SOFT_FEATURES;
5820 dev->wanted_features = dev->features & dev->hw_features;
5822 /* Turn on no cache copy if HW is doing checksum */
5823 if (!(dev->flags & IFF_LOOPBACK)) {
5824 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5825 if (dev->features & NETIF_F_ALL_CSUM) {
5826 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5827 dev->features |= NETIF_F_NOCACHE_COPY;
5831 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5833 dev->vlan_features |= NETIF_F_HIGHDMA;
5835 /* Make NETIF_F_SG inheritable to tunnel devices.
5837 dev->hw_enc_features |= NETIF_F_SG;
5839 /* Make NETIF_F_SG inheritable to MPLS.
5841 dev->mpls_features |= NETIF_F_SG;
5843 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5844 ret = notifier_to_errno(ret);
5848 ret = netdev_register_kobject(dev);
5851 dev->reg_state = NETREG_REGISTERED;
5853 __netdev_update_features(dev);
5856 * Default initial state at registry is that the
5857 * device is present.
5860 set_bit(__LINK_STATE_PRESENT, &dev->state);
5862 linkwatch_init_dev(dev);
5864 dev_init_scheduler(dev);
5866 list_netdevice(dev);
5867 add_device_randomness(dev->dev_addr, dev->addr_len);
5869 /* If the device has permanent device address, driver should
5870 * set dev_addr and also addr_assign_type should be set to
5871 * NET_ADDR_PERM (default value).
5873 if (dev->addr_assign_type == NET_ADDR_PERM)
5874 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5876 /* Notify protocols, that a new device appeared. */
5877 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5878 ret = notifier_to_errno(ret);
5880 rollback_registered(dev);
5881 dev->reg_state = NETREG_UNREGISTERED;
5884 * Prevent userspace races by waiting until the network
5885 * device is fully setup before sending notifications.
5887 if (!dev->rtnl_link_ops ||
5888 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5889 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5895 if (dev->netdev_ops->ndo_uninit)
5896 dev->netdev_ops->ndo_uninit(dev);
5899 EXPORT_SYMBOL(register_netdevice);
5902 * init_dummy_netdev - init a dummy network device for NAPI
5903 * @dev: device to init
5905 * This takes a network device structure and initialize the minimum
5906 * amount of fields so it can be used to schedule NAPI polls without
5907 * registering a full blown interface. This is to be used by drivers
5908 * that need to tie several hardware interfaces to a single NAPI
5909 * poll scheduler due to HW limitations.
5911 int init_dummy_netdev(struct net_device *dev)
5913 /* Clear everything. Note we don't initialize spinlocks
5914 * are they aren't supposed to be taken by any of the
5915 * NAPI code and this dummy netdev is supposed to be
5916 * only ever used for NAPI polls
5918 memset(dev, 0, sizeof(struct net_device));
5920 /* make sure we BUG if trying to hit standard
5921 * register/unregister code path
5923 dev->reg_state = NETREG_DUMMY;
5925 /* NAPI wants this */
5926 INIT_LIST_HEAD(&dev->napi_list);
5928 /* a dummy interface is started by default */
5929 set_bit(__LINK_STATE_PRESENT, &dev->state);
5930 set_bit(__LINK_STATE_START, &dev->state);
5932 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5933 * because users of this 'device' dont need to change
5939 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5943 * register_netdev - register a network device
5944 * @dev: device to register
5946 * Take a completed network device structure and add it to the kernel
5947 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5948 * chain. 0 is returned on success. A negative errno code is returned
5949 * on a failure to set up the device, or if the name is a duplicate.
5951 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5952 * and expands the device name if you passed a format string to
5955 int register_netdev(struct net_device *dev)
5960 err = register_netdevice(dev);
5964 EXPORT_SYMBOL(register_netdev);
5966 int netdev_refcnt_read(const struct net_device *dev)
5970 for_each_possible_cpu(i)
5971 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5974 EXPORT_SYMBOL(netdev_refcnt_read);
5977 * netdev_wait_allrefs - wait until all references are gone.
5978 * @dev: target net_device
5980 * This is called when unregistering network devices.
5982 * Any protocol or device that holds a reference should register
5983 * for netdevice notification, and cleanup and put back the
5984 * reference if they receive an UNREGISTER event.
5985 * We can get stuck here if buggy protocols don't correctly
5988 static void netdev_wait_allrefs(struct net_device *dev)
5990 unsigned long rebroadcast_time, warning_time;
5993 linkwatch_forget_dev(dev);
5995 rebroadcast_time = warning_time = jiffies;
5996 refcnt = netdev_refcnt_read(dev);
5998 while (refcnt != 0) {
5999 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6002 /* Rebroadcast unregister notification */
6003 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6009 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6010 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6012 /* We must not have linkwatch events
6013 * pending on unregister. If this
6014 * happens, we simply run the queue
6015 * unscheduled, resulting in a noop
6018 linkwatch_run_queue();
6023 rebroadcast_time = jiffies;
6028 refcnt = netdev_refcnt_read(dev);
6030 if (time_after(jiffies, warning_time + 10 * HZ)) {
6031 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6033 warning_time = jiffies;
6042 * register_netdevice(x1);
6043 * register_netdevice(x2);
6045 * unregister_netdevice(y1);
6046 * unregister_netdevice(y2);
6052 * We are invoked by rtnl_unlock().
6053 * This allows us to deal with problems:
6054 * 1) We can delete sysfs objects which invoke hotplug
6055 * without deadlocking with linkwatch via keventd.
6056 * 2) Since we run with the RTNL semaphore not held, we can sleep
6057 * safely in order to wait for the netdev refcnt to drop to zero.
6059 * We must not return until all unregister events added during
6060 * the interval the lock was held have been completed.
6062 void netdev_run_todo(void)
6064 struct list_head list;
6066 /* Snapshot list, allow later requests */
6067 list_replace_init(&net_todo_list, &list);
6072 /* Wait for rcu callbacks to finish before next phase */
6073 if (!list_empty(&list))
6076 while (!list_empty(&list)) {
6077 struct net_device *dev
6078 = list_first_entry(&list, struct net_device, todo_list);
6079 list_del(&dev->todo_list);
6082 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6085 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6086 pr_err("network todo '%s' but state %d\n",
6087 dev->name, dev->reg_state);
6092 dev->reg_state = NETREG_UNREGISTERED;
6094 on_each_cpu(flush_backlog, dev, 1);
6096 netdev_wait_allrefs(dev);
6099 BUG_ON(netdev_refcnt_read(dev));
6100 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6101 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6102 WARN_ON(dev->dn_ptr);
6104 if (dev->destructor)
6105 dev->destructor(dev);
6107 /* Report a network device has been unregistered */
6109 dev_net(dev)->dev_unreg_count--;
6111 wake_up(&netdev_unregistering_wq);
6113 /* Free network device */
6114 kobject_put(&dev->dev.kobj);
6118 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6119 * fields in the same order, with only the type differing.
6121 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6122 const struct net_device_stats *netdev_stats)
6124 #if BITS_PER_LONG == 64
6125 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6126 memcpy(stats64, netdev_stats, sizeof(*stats64));
6128 size_t i, n = sizeof(*stats64) / sizeof(u64);
6129 const unsigned long *src = (const unsigned long *)netdev_stats;
6130 u64 *dst = (u64 *)stats64;
6132 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6133 sizeof(*stats64) / sizeof(u64));
6134 for (i = 0; i < n; i++)
6138 EXPORT_SYMBOL(netdev_stats_to_stats64);
6141 * dev_get_stats - get network device statistics
6142 * @dev: device to get statistics from
6143 * @storage: place to store stats
6145 * Get network statistics from device. Return @storage.
6146 * The device driver may provide its own method by setting
6147 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6148 * otherwise the internal statistics structure is used.
6150 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6151 struct rtnl_link_stats64 *storage)
6153 const struct net_device_ops *ops = dev->netdev_ops;
6155 if (ops->ndo_get_stats64) {
6156 memset(storage, 0, sizeof(*storage));
6157 ops->ndo_get_stats64(dev, storage);
6158 } else if (ops->ndo_get_stats) {
6159 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6161 netdev_stats_to_stats64(storage, &dev->stats);
6163 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6166 EXPORT_SYMBOL(dev_get_stats);
6168 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6170 struct netdev_queue *queue = dev_ingress_queue(dev);
6172 #ifdef CONFIG_NET_CLS_ACT
6175 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6178 netdev_init_one_queue(dev, queue, NULL);
6179 queue->qdisc = &noop_qdisc;
6180 queue->qdisc_sleeping = &noop_qdisc;
6181 rcu_assign_pointer(dev->ingress_queue, queue);
6186 static const struct ethtool_ops default_ethtool_ops;
6188 void netdev_set_default_ethtool_ops(struct net_device *dev,
6189 const struct ethtool_ops *ops)
6191 if (dev->ethtool_ops == &default_ethtool_ops)
6192 dev->ethtool_ops = ops;
6194 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6197 * alloc_netdev_mqs - allocate network device
6198 * @sizeof_priv: size of private data to allocate space for
6199 * @name: device name format string
6200 * @setup: callback to initialize device
6201 * @txqs: the number of TX subqueues to allocate
6202 * @rxqs: the number of RX subqueues to allocate
6204 * Allocates a struct net_device with private data area for driver use
6205 * and performs basic initialization. Also allocates subquue structs
6206 * for each queue on the device.
6208 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6209 void (*setup)(struct net_device *),
6210 unsigned int txqs, unsigned int rxqs)
6212 struct net_device *dev;
6214 struct net_device *p;
6216 BUG_ON(strlen(name) >= sizeof(dev->name));
6219 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6225 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6230 alloc_size = sizeof(struct net_device);
6232 /* ensure 32-byte alignment of private area */
6233 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6234 alloc_size += sizeof_priv;
6236 /* ensure 32-byte alignment of whole construct */
6237 alloc_size += NETDEV_ALIGN - 1;
6239 p = kzalloc(alloc_size, GFP_KERNEL);
6243 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6244 dev->padded = (char *)dev - (char *)p;
6246 dev->pcpu_refcnt = alloc_percpu(int);
6247 if (!dev->pcpu_refcnt)
6250 if (dev_addr_init(dev))
6256 dev_net_set(dev, &init_net);
6258 dev->gso_max_size = GSO_MAX_SIZE;
6259 dev->gso_max_segs = GSO_MAX_SEGS;
6261 INIT_LIST_HEAD(&dev->napi_list);
6262 INIT_LIST_HEAD(&dev->unreg_list);
6263 INIT_LIST_HEAD(&dev->close_list);
6264 INIT_LIST_HEAD(&dev->link_watch_list);
6265 INIT_LIST_HEAD(&dev->adj_list.upper);
6266 INIT_LIST_HEAD(&dev->adj_list.lower);
6267 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6268 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6269 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6272 dev->num_tx_queues = txqs;
6273 dev->real_num_tx_queues = txqs;
6274 if (netif_alloc_netdev_queues(dev))
6278 dev->num_rx_queues = rxqs;
6279 dev->real_num_rx_queues = rxqs;
6280 if (netif_alloc_rx_queues(dev))
6284 strcpy(dev->name, name);
6285 dev->group = INIT_NETDEV_GROUP;
6286 if (!dev->ethtool_ops)
6287 dev->ethtool_ops = &default_ethtool_ops;
6295 free_percpu(dev->pcpu_refcnt);
6296 netif_free_tx_queues(dev);
6305 EXPORT_SYMBOL(alloc_netdev_mqs);
6308 * free_netdev - free network device
6311 * This function does the last stage of destroying an allocated device
6312 * interface. The reference to the device object is released.
6313 * If this is the last reference then it will be freed.
6315 void free_netdev(struct net_device *dev)
6317 struct napi_struct *p, *n;
6319 release_net(dev_net(dev));
6321 netif_free_tx_queues(dev);
6326 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6328 /* Flush device addresses */
6329 dev_addr_flush(dev);
6331 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6334 free_percpu(dev->pcpu_refcnt);
6335 dev->pcpu_refcnt = NULL;
6337 /* Compatibility with error handling in drivers */
6338 if (dev->reg_state == NETREG_UNINITIALIZED) {
6339 kfree((char *)dev - dev->padded);
6343 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6344 dev->reg_state = NETREG_RELEASED;
6346 /* will free via device release */
6347 put_device(&dev->dev);
6349 EXPORT_SYMBOL(free_netdev);
6352 * synchronize_net - Synchronize with packet receive processing
6354 * Wait for packets currently being received to be done.
6355 * Does not block later packets from starting.
6357 void synchronize_net(void)
6360 if (rtnl_is_locked())
6361 synchronize_rcu_expedited();
6365 EXPORT_SYMBOL(synchronize_net);
6368 * unregister_netdevice_queue - remove device from the kernel
6372 * This function shuts down a device interface and removes it
6373 * from the kernel tables.
6374 * If head not NULL, device is queued to be unregistered later.
6376 * Callers must hold the rtnl semaphore. You may want
6377 * unregister_netdev() instead of this.
6380 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6385 list_move_tail(&dev->unreg_list, head);
6387 rollback_registered(dev);
6388 /* Finish processing unregister after unlock */
6392 EXPORT_SYMBOL(unregister_netdevice_queue);
6395 * unregister_netdevice_many - unregister many devices
6396 * @head: list of devices
6398 void unregister_netdevice_many(struct list_head *head)
6400 struct net_device *dev;
6402 if (!list_empty(head)) {
6403 rollback_registered_many(head);
6404 list_for_each_entry(dev, head, unreg_list)
6408 EXPORT_SYMBOL(unregister_netdevice_many);
6411 * unregister_netdev - remove device from the kernel
6414 * This function shuts down a device interface and removes it
6415 * from the kernel tables.
6417 * This is just a wrapper for unregister_netdevice that takes
6418 * the rtnl semaphore. In general you want to use this and not
6419 * unregister_netdevice.
6421 void unregister_netdev(struct net_device *dev)
6424 unregister_netdevice(dev);
6427 EXPORT_SYMBOL(unregister_netdev);
6430 * dev_change_net_namespace - move device to different nethost namespace
6432 * @net: network namespace
6433 * @pat: If not NULL name pattern to try if the current device name
6434 * is already taken in the destination network namespace.
6436 * This function shuts down a device interface and moves it
6437 * to a new network namespace. On success 0 is returned, on
6438 * a failure a netagive errno code is returned.
6440 * Callers must hold the rtnl semaphore.
6443 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6449 /* Don't allow namespace local devices to be moved. */
6451 if (dev->features & NETIF_F_NETNS_LOCAL)
6454 /* Ensure the device has been registrered */
6455 if (dev->reg_state != NETREG_REGISTERED)
6458 /* Get out if there is nothing todo */
6460 if (net_eq(dev_net(dev), net))
6463 /* Pick the destination device name, and ensure
6464 * we can use it in the destination network namespace.
6467 if (__dev_get_by_name(net, dev->name)) {
6468 /* We get here if we can't use the current device name */
6471 if (dev_get_valid_name(net, dev, pat) < 0)
6476 * And now a mini version of register_netdevice unregister_netdevice.
6479 /* If device is running close it first. */
6482 /* And unlink it from device chain */
6484 unlist_netdevice(dev);
6488 /* Shutdown queueing discipline. */
6491 /* Notify protocols, that we are about to destroy
6492 this device. They should clean all the things.
6494 Note that dev->reg_state stays at NETREG_REGISTERED.
6495 This is wanted because this way 8021q and macvlan know
6496 the device is just moving and can keep their slaves up.
6498 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6500 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6501 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6504 * Flush the unicast and multicast chains
6509 /* Send a netdev-removed uevent to the old namespace */
6510 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6512 /* Actually switch the network namespace */
6513 dev_net_set(dev, net);
6515 /* If there is an ifindex conflict assign a new one */
6516 if (__dev_get_by_index(net, dev->ifindex)) {
6517 int iflink = (dev->iflink == dev->ifindex);
6518 dev->ifindex = dev_new_index(net);
6520 dev->iflink = dev->ifindex;
6523 /* Send a netdev-add uevent to the new namespace */
6524 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6526 /* Fixup kobjects */
6527 err = device_rename(&dev->dev, dev->name);
6530 /* Add the device back in the hashes */
6531 list_netdevice(dev);
6533 /* Notify protocols, that a new device appeared. */
6534 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6537 * Prevent userspace races by waiting until the network
6538 * device is fully setup before sending notifications.
6540 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6547 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6549 static int dev_cpu_callback(struct notifier_block *nfb,
6550 unsigned long action,
6553 struct sk_buff **list_skb;
6554 struct sk_buff *skb;
6555 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6556 struct softnet_data *sd, *oldsd;
6558 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6561 local_irq_disable();
6562 cpu = smp_processor_id();
6563 sd = &per_cpu(softnet_data, cpu);
6564 oldsd = &per_cpu(softnet_data, oldcpu);
6566 /* Find end of our completion_queue. */
6567 list_skb = &sd->completion_queue;
6569 list_skb = &(*list_skb)->next;
6570 /* Append completion queue from offline CPU. */
6571 *list_skb = oldsd->completion_queue;
6572 oldsd->completion_queue = NULL;
6574 /* Append output queue from offline CPU. */
6575 if (oldsd->output_queue) {
6576 *sd->output_queue_tailp = oldsd->output_queue;
6577 sd->output_queue_tailp = oldsd->output_queue_tailp;
6578 oldsd->output_queue = NULL;
6579 oldsd->output_queue_tailp = &oldsd->output_queue;
6581 /* Append NAPI poll list from offline CPU. */
6582 if (!list_empty(&oldsd->poll_list)) {
6583 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6584 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6587 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6590 /* Process offline CPU's input_pkt_queue */
6591 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6593 input_queue_head_incr(oldsd);
6595 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6597 input_queue_head_incr(oldsd);
6605 * netdev_increment_features - increment feature set by one
6606 * @all: current feature set
6607 * @one: new feature set
6608 * @mask: mask feature set
6610 * Computes a new feature set after adding a device with feature set
6611 * @one to the master device with current feature set @all. Will not
6612 * enable anything that is off in @mask. Returns the new feature set.
6614 netdev_features_t netdev_increment_features(netdev_features_t all,
6615 netdev_features_t one, netdev_features_t mask)
6617 if (mask & NETIF_F_GEN_CSUM)
6618 mask |= NETIF_F_ALL_CSUM;
6619 mask |= NETIF_F_VLAN_CHALLENGED;
6621 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6622 all &= one | ~NETIF_F_ALL_FOR_ALL;
6624 /* If one device supports hw checksumming, set for all. */
6625 if (all & NETIF_F_GEN_CSUM)
6626 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6630 EXPORT_SYMBOL(netdev_increment_features);
6632 static struct hlist_head * __net_init netdev_create_hash(void)
6635 struct hlist_head *hash;
6637 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6639 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6640 INIT_HLIST_HEAD(&hash[i]);
6645 /* Initialize per network namespace state */
6646 static int __net_init netdev_init(struct net *net)
6648 if (net != &init_net)
6649 INIT_LIST_HEAD(&net->dev_base_head);
6651 net->dev_name_head = netdev_create_hash();
6652 if (net->dev_name_head == NULL)
6655 net->dev_index_head = netdev_create_hash();
6656 if (net->dev_index_head == NULL)
6662 kfree(net->dev_name_head);
6668 * netdev_drivername - network driver for the device
6669 * @dev: network device
6671 * Determine network driver for device.
6673 const char *netdev_drivername(const struct net_device *dev)
6675 const struct device_driver *driver;
6676 const struct device *parent;
6677 const char *empty = "";
6679 parent = dev->dev.parent;
6683 driver = parent->driver;
6684 if (driver && driver->name)
6685 return driver->name;
6689 static int __netdev_printk(const char *level, const struct net_device *dev,
6690 struct va_format *vaf)
6694 if (dev && dev->dev.parent) {
6695 r = dev_printk_emit(level[1] - '0',
6698 dev_driver_string(dev->dev.parent),
6699 dev_name(dev->dev.parent),
6700 netdev_name(dev), vaf);
6702 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6704 r = printk("%s(NULL net_device): %pV", level, vaf);
6710 int netdev_printk(const char *level, const struct net_device *dev,
6711 const char *format, ...)
6713 struct va_format vaf;
6717 va_start(args, format);
6722 r = __netdev_printk(level, dev, &vaf);
6728 EXPORT_SYMBOL(netdev_printk);
6730 #define define_netdev_printk_level(func, level) \
6731 int func(const struct net_device *dev, const char *fmt, ...) \
6734 struct va_format vaf; \
6737 va_start(args, fmt); \
6742 r = __netdev_printk(level, dev, &vaf); \
6748 EXPORT_SYMBOL(func);
6750 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6751 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6752 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6753 define_netdev_printk_level(netdev_err, KERN_ERR);
6754 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6755 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6756 define_netdev_printk_level(netdev_info, KERN_INFO);
6758 static void __net_exit netdev_exit(struct net *net)
6760 kfree(net->dev_name_head);
6761 kfree(net->dev_index_head);
6764 static struct pernet_operations __net_initdata netdev_net_ops = {
6765 .init = netdev_init,
6766 .exit = netdev_exit,
6769 static void __net_exit default_device_exit(struct net *net)
6771 struct net_device *dev, *aux;
6773 * Push all migratable network devices back to the
6774 * initial network namespace
6777 for_each_netdev_safe(net, dev, aux) {
6779 char fb_name[IFNAMSIZ];
6781 /* Ignore unmoveable devices (i.e. loopback) */
6782 if (dev->features & NETIF_F_NETNS_LOCAL)
6785 /* Leave virtual devices for the generic cleanup */
6786 if (dev->rtnl_link_ops)
6789 /* Push remaining network devices to init_net */
6790 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6791 err = dev_change_net_namespace(dev, &init_net, fb_name);
6793 pr_emerg("%s: failed to move %s to init_net: %d\n",
6794 __func__, dev->name, err);
6801 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6803 /* Return with the rtnl_lock held when there are no network
6804 * devices unregistering in any network namespace in net_list.
6811 prepare_to_wait(&netdev_unregistering_wq, &wait,
6812 TASK_UNINTERRUPTIBLE);
6813 unregistering = false;
6815 list_for_each_entry(net, net_list, exit_list) {
6816 if (net->dev_unreg_count > 0) {
6817 unregistering = true;
6826 finish_wait(&netdev_unregistering_wq, &wait);
6829 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6831 /* At exit all network devices most be removed from a network
6832 * namespace. Do this in the reverse order of registration.
6833 * Do this across as many network namespaces as possible to
6834 * improve batching efficiency.
6836 struct net_device *dev;
6838 LIST_HEAD(dev_kill_list);
6840 /* To prevent network device cleanup code from dereferencing
6841 * loopback devices or network devices that have been freed
6842 * wait here for all pending unregistrations to complete,
6843 * before unregistring the loopback device and allowing the
6844 * network namespace be freed.
6846 * The netdev todo list containing all network devices
6847 * unregistrations that happen in default_device_exit_batch
6848 * will run in the rtnl_unlock() at the end of
6849 * default_device_exit_batch.
6851 rtnl_lock_unregistering(net_list);
6852 list_for_each_entry(net, net_list, exit_list) {
6853 for_each_netdev_reverse(net, dev) {
6854 if (dev->rtnl_link_ops)
6855 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6857 unregister_netdevice_queue(dev, &dev_kill_list);
6860 unregister_netdevice_many(&dev_kill_list);
6861 list_del(&dev_kill_list);
6865 static struct pernet_operations __net_initdata default_device_ops = {
6866 .exit = default_device_exit,
6867 .exit_batch = default_device_exit_batch,
6871 * Initialize the DEV module. At boot time this walks the device list and
6872 * unhooks any devices that fail to initialise (normally hardware not
6873 * present) and leaves us with a valid list of present and active devices.
6878 * This is called single threaded during boot, so no need
6879 * to take the rtnl semaphore.
6881 static int __init net_dev_init(void)
6883 int i, rc = -ENOMEM;
6885 BUG_ON(!dev_boot_phase);
6887 if (dev_proc_init())
6890 if (netdev_kobject_init())
6893 INIT_LIST_HEAD(&ptype_all);
6894 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6895 INIT_LIST_HEAD(&ptype_base[i]);
6897 INIT_LIST_HEAD(&offload_base);
6899 if (register_pernet_subsys(&netdev_net_ops))
6903 * Initialise the packet receive queues.
6906 for_each_possible_cpu(i) {
6907 struct softnet_data *sd = &per_cpu(softnet_data, i);
6909 memset(sd, 0, sizeof(*sd));
6910 skb_queue_head_init(&sd->input_pkt_queue);
6911 skb_queue_head_init(&sd->process_queue);
6912 sd->completion_queue = NULL;
6913 INIT_LIST_HEAD(&sd->poll_list);
6914 sd->output_queue = NULL;
6915 sd->output_queue_tailp = &sd->output_queue;
6917 sd->csd.func = rps_trigger_softirq;
6923 sd->backlog.poll = process_backlog;
6924 sd->backlog.weight = weight_p;
6925 sd->backlog.gro_list = NULL;
6926 sd->backlog.gro_count = 0;
6928 #ifdef CONFIG_NET_FLOW_LIMIT
6929 sd->flow_limit = NULL;
6935 /* The loopback device is special if any other network devices
6936 * is present in a network namespace the loopback device must
6937 * be present. Since we now dynamically allocate and free the
6938 * loopback device ensure this invariant is maintained by
6939 * keeping the loopback device as the first device on the
6940 * list of network devices. Ensuring the loopback devices
6941 * is the first device that appears and the last network device
6944 if (register_pernet_device(&loopback_net_ops))
6947 if (register_pernet_device(&default_device_ops))
6950 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6951 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6953 hotcpu_notifier(dev_cpu_callback, 0);
6960 subsys_initcall(net_dev_init);