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_GSO_CB(skb)->encap_level = 0;
2382 skb_reset_mac_header(skb);
2383 skb_reset_mac_len(skb);
2385 return skb_mac_gso_segment(skb, features);
2387 EXPORT_SYMBOL(__skb_gso_segment);
2389 /* Take action when hardware reception checksum errors are detected. */
2391 void netdev_rx_csum_fault(struct net_device *dev)
2393 if (net_ratelimit()) {
2394 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2398 EXPORT_SYMBOL(netdev_rx_csum_fault);
2401 /* Actually, we should eliminate this check as soon as we know, that:
2402 * 1. IOMMU is present and allows to map all the memory.
2403 * 2. No high memory really exists on this machine.
2406 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2408 #ifdef CONFIG_HIGHMEM
2410 if (!(dev->features & NETIF_F_HIGHDMA)) {
2411 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2412 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2413 if (PageHighMem(skb_frag_page(frag)))
2418 if (PCI_DMA_BUS_IS_PHYS) {
2419 struct device *pdev = dev->dev.parent;
2423 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2424 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2425 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2426 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2435 void (*destructor)(struct sk_buff *skb);
2438 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2440 static void dev_gso_skb_destructor(struct sk_buff *skb)
2442 struct dev_gso_cb *cb;
2445 struct sk_buff *nskb = skb->next;
2447 skb->next = nskb->next;
2450 } while (skb->next);
2452 cb = DEV_GSO_CB(skb);
2454 cb->destructor(skb);
2458 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2459 * @skb: buffer to segment
2460 * @features: device features as applicable to this skb
2462 * This function segments the given skb and stores the list of segments
2465 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2467 struct sk_buff *segs;
2469 segs = skb_gso_segment(skb, features);
2471 /* Verifying header integrity only. */
2476 return PTR_ERR(segs);
2479 DEV_GSO_CB(skb)->destructor = skb->destructor;
2480 skb->destructor = dev_gso_skb_destructor;
2485 static netdev_features_t harmonize_features(struct sk_buff *skb,
2486 netdev_features_t features)
2488 if (skb->ip_summed != CHECKSUM_NONE &&
2489 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2490 features &= ~NETIF_F_ALL_CSUM;
2491 } else if (illegal_highdma(skb->dev, skb)) {
2492 features &= ~NETIF_F_SG;
2498 netdev_features_t netif_skb_features(struct sk_buff *skb)
2500 __be16 protocol = skb->protocol;
2501 netdev_features_t features = skb->dev->features;
2503 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2504 features &= ~NETIF_F_GSO_MASK;
2506 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2507 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2508 protocol = veh->h_vlan_encapsulated_proto;
2509 } else if (!vlan_tx_tag_present(skb)) {
2510 return harmonize_features(skb, features);
2513 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2514 NETIF_F_HW_VLAN_STAG_TX);
2516 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2517 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2518 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2519 NETIF_F_HW_VLAN_STAG_TX;
2521 return harmonize_features(skb, features);
2523 EXPORT_SYMBOL(netif_skb_features);
2526 * Returns true if either:
2527 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2528 * 2. skb is fragmented and the device does not support SG.
2530 static inline int skb_needs_linearize(struct sk_buff *skb,
2531 netdev_features_t features)
2533 return skb_is_nonlinear(skb) &&
2534 ((skb_has_frag_list(skb) &&
2535 !(features & NETIF_F_FRAGLIST)) ||
2536 (skb_shinfo(skb)->nr_frags &&
2537 !(features & NETIF_F_SG)));
2540 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2541 struct netdev_queue *txq)
2543 const struct net_device_ops *ops = dev->netdev_ops;
2544 int rc = NETDEV_TX_OK;
2545 unsigned int skb_len;
2547 if (likely(!skb->next)) {
2548 netdev_features_t features;
2551 * If device doesn't need skb->dst, release it right now while
2552 * its hot in this cpu cache
2554 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2557 features = netif_skb_features(skb);
2559 if (vlan_tx_tag_present(skb) &&
2560 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2561 skb = __vlan_put_tag(skb, skb->vlan_proto,
2562 vlan_tx_tag_get(skb));
2569 /* If encapsulation offload request, verify we are testing
2570 * hardware encapsulation features instead of standard
2571 * features for the netdev
2573 if (skb->encapsulation)
2574 features &= dev->hw_enc_features;
2576 if (netif_needs_gso(skb, features)) {
2577 if (unlikely(dev_gso_segment(skb, features)))
2582 if (skb_needs_linearize(skb, features) &&
2583 __skb_linearize(skb))
2586 /* If packet is not checksummed and device does not
2587 * support checksumming for this protocol, complete
2588 * checksumming here.
2590 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2591 if (skb->encapsulation)
2592 skb_set_inner_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2595 skb_set_transport_header(skb,
2596 skb_checksum_start_offset(skb));
2597 if (!(features & NETIF_F_ALL_CSUM) &&
2598 skb_checksum_help(skb))
2603 if (!list_empty(&ptype_all))
2604 dev_queue_xmit_nit(skb, dev);
2607 rc = ops->ndo_start_xmit(skb, dev);
2608 trace_net_dev_xmit(skb, rc, dev, skb_len);
2609 if (rc == NETDEV_TX_OK)
2610 txq_trans_update(txq);
2616 struct sk_buff *nskb = skb->next;
2618 skb->next = nskb->next;
2621 if (!list_empty(&ptype_all))
2622 dev_queue_xmit_nit(nskb, dev);
2624 skb_len = nskb->len;
2625 rc = ops->ndo_start_xmit(nskb, dev);
2626 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2627 if (unlikely(rc != NETDEV_TX_OK)) {
2628 if (rc & ~NETDEV_TX_MASK)
2629 goto out_kfree_gso_skb;
2630 nskb->next = skb->next;
2634 txq_trans_update(txq);
2635 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2636 return NETDEV_TX_BUSY;
2637 } while (skb->next);
2640 if (likely(skb->next == NULL)) {
2641 skb->destructor = DEV_GSO_CB(skb)->destructor;
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2784 * Queue a buffer for transmission to a network device. The caller must
2785 * have set the device and priority and built the buffer before calling
2786 * this function. The function can be called from an interrupt.
2788 * A negative errno code is returned on a failure. A success does not
2789 * guarantee the frame will be transmitted as it may be dropped due
2790 * to congestion or traffic shaping.
2792 * -----------------------------------------------------------------------------------
2793 * I notice this method can also return errors from the queue disciplines,
2794 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2797 * Regardless of the return value, the skb is consumed, so it is currently
2798 * difficult to retry a send to this method. (You can bump the ref count
2799 * before sending to hold a reference for retry if you are careful.)
2801 * When calling this method, interrupts MUST be enabled. This is because
2802 * the BH enable code must have IRQs enabled so that it will not deadlock.
2805 int dev_queue_xmit(struct sk_buff *skb)
2807 struct net_device *dev = skb->dev;
2808 struct netdev_queue *txq;
2812 skb_reset_mac_header(skb);
2814 /* Disable soft irqs for various locks below. Also
2815 * stops preemption for RCU.
2819 skb_update_prio(skb);
2821 txq = netdev_pick_tx(dev, skb);
2822 q = rcu_dereference_bh(txq->qdisc);
2824 #ifdef CONFIG_NET_CLS_ACT
2825 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2827 trace_net_dev_queue(skb);
2829 rc = __dev_xmit_skb(skb, q, dev, txq);
2833 /* The device has no queue. Common case for software devices:
2834 loopback, all the sorts of tunnels...
2836 Really, it is unlikely that netif_tx_lock protection is necessary
2837 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2839 However, it is possible, that they rely on protection
2842 Check this and shot the lock. It is not prone from deadlocks.
2843 Either shot noqueue qdisc, it is even simpler 8)
2845 if (dev->flags & IFF_UP) {
2846 int cpu = smp_processor_id(); /* ok because BHs are off */
2848 if (txq->xmit_lock_owner != cpu) {
2850 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2851 goto recursion_alert;
2853 HARD_TX_LOCK(dev, txq, cpu);
2855 if (!netif_xmit_stopped(txq)) {
2856 __this_cpu_inc(xmit_recursion);
2857 rc = dev_hard_start_xmit(skb, dev, txq);
2858 __this_cpu_dec(xmit_recursion);
2859 if (dev_xmit_complete(rc)) {
2860 HARD_TX_UNLOCK(dev, txq);
2864 HARD_TX_UNLOCK(dev, txq);
2865 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2868 /* Recursion is detected! It is possible,
2872 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2878 rcu_read_unlock_bh();
2883 rcu_read_unlock_bh();
2886 EXPORT_SYMBOL(dev_queue_xmit);
2889 /*=======================================================================
2891 =======================================================================*/
2893 int netdev_max_backlog __read_mostly = 1000;
2894 EXPORT_SYMBOL(netdev_max_backlog);
2896 int netdev_tstamp_prequeue __read_mostly = 1;
2897 int netdev_budget __read_mostly = 300;
2898 int weight_p __read_mostly = 64; /* old backlog weight */
2900 /* Called with irq disabled */
2901 static inline void ____napi_schedule(struct softnet_data *sd,
2902 struct napi_struct *napi)
2904 list_add_tail(&napi->poll_list, &sd->poll_list);
2905 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2910 /* One global table that all flow-based protocols share. */
2911 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2912 EXPORT_SYMBOL(rps_sock_flow_table);
2914 struct static_key rps_needed __read_mostly;
2916 static struct rps_dev_flow *
2917 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2918 struct rps_dev_flow *rflow, u16 next_cpu)
2920 if (next_cpu != RPS_NO_CPU) {
2921 #ifdef CONFIG_RFS_ACCEL
2922 struct netdev_rx_queue *rxqueue;
2923 struct rps_dev_flow_table *flow_table;
2924 struct rps_dev_flow *old_rflow;
2929 /* Should we steer this flow to a different hardware queue? */
2930 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2931 !(dev->features & NETIF_F_NTUPLE))
2933 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2934 if (rxq_index == skb_get_rx_queue(skb))
2937 rxqueue = dev->_rx + rxq_index;
2938 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2941 flow_id = skb->rxhash & flow_table->mask;
2942 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2943 rxq_index, flow_id);
2947 rflow = &flow_table->flows[flow_id];
2949 if (old_rflow->filter == rflow->filter)
2950 old_rflow->filter = RPS_NO_FILTER;
2954 per_cpu(softnet_data, next_cpu).input_queue_head;
2957 rflow->cpu = next_cpu;
2962 * get_rps_cpu is called from netif_receive_skb and returns the target
2963 * CPU from the RPS map of the receiving queue for a given skb.
2964 * rcu_read_lock must be held on entry.
2966 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2967 struct rps_dev_flow **rflowp)
2969 struct netdev_rx_queue *rxqueue;
2970 struct rps_map *map;
2971 struct rps_dev_flow_table *flow_table;
2972 struct rps_sock_flow_table *sock_flow_table;
2976 if (skb_rx_queue_recorded(skb)) {
2977 u16 index = skb_get_rx_queue(skb);
2978 if (unlikely(index >= dev->real_num_rx_queues)) {
2979 WARN_ONCE(dev->real_num_rx_queues > 1,
2980 "%s received packet on queue %u, but number "
2981 "of RX queues is %u\n",
2982 dev->name, index, dev->real_num_rx_queues);
2985 rxqueue = dev->_rx + index;
2989 map = rcu_dereference(rxqueue->rps_map);
2991 if (map->len == 1 &&
2992 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2993 tcpu = map->cpus[0];
2994 if (cpu_online(tcpu))
2998 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3002 skb_reset_network_header(skb);
3003 if (!skb_get_rxhash(skb))
3006 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3007 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3008 if (flow_table && sock_flow_table) {
3010 struct rps_dev_flow *rflow;
3012 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3015 next_cpu = sock_flow_table->ents[skb->rxhash &
3016 sock_flow_table->mask];
3019 * If the desired CPU (where last recvmsg was done) is
3020 * different from current CPU (one in the rx-queue flow
3021 * table entry), switch if one of the following holds:
3022 * - Current CPU is unset (equal to RPS_NO_CPU).
3023 * - Current CPU is offline.
3024 * - The current CPU's queue tail has advanced beyond the
3025 * last packet that was enqueued using this table entry.
3026 * This guarantees that all previous packets for the flow
3027 * have been dequeued, thus preserving in order delivery.
3029 if (unlikely(tcpu != next_cpu) &&
3030 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3031 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3032 rflow->last_qtail)) >= 0)) {
3034 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3037 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3045 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3047 if (cpu_online(tcpu)) {
3057 #ifdef CONFIG_RFS_ACCEL
3060 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3061 * @dev: Device on which the filter was set
3062 * @rxq_index: RX queue index
3063 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3064 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3066 * Drivers that implement ndo_rx_flow_steer() should periodically call
3067 * this function for each installed filter and remove the filters for
3068 * which it returns %true.
3070 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3071 u32 flow_id, u16 filter_id)
3073 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3074 struct rps_dev_flow_table *flow_table;
3075 struct rps_dev_flow *rflow;
3080 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3081 if (flow_table && flow_id <= flow_table->mask) {
3082 rflow = &flow_table->flows[flow_id];
3083 cpu = ACCESS_ONCE(rflow->cpu);
3084 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3085 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3086 rflow->last_qtail) <
3087 (int)(10 * flow_table->mask)))
3093 EXPORT_SYMBOL(rps_may_expire_flow);
3095 #endif /* CONFIG_RFS_ACCEL */
3097 /* Called from hardirq (IPI) context */
3098 static void rps_trigger_softirq(void *data)
3100 struct softnet_data *sd = data;
3102 ____napi_schedule(sd, &sd->backlog);
3106 #endif /* CONFIG_RPS */
3109 * Check if this softnet_data structure is another cpu one
3110 * If yes, queue it to our IPI list and return 1
3113 static int rps_ipi_queued(struct softnet_data *sd)
3116 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3119 sd->rps_ipi_next = mysd->rps_ipi_list;
3120 mysd->rps_ipi_list = sd;
3122 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3125 #endif /* CONFIG_RPS */
3129 #ifdef CONFIG_NET_FLOW_LIMIT
3130 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3133 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3135 #ifdef CONFIG_NET_FLOW_LIMIT
3136 struct sd_flow_limit *fl;
3137 struct softnet_data *sd;
3138 unsigned int old_flow, new_flow;
3140 if (qlen < (netdev_max_backlog >> 1))
3143 sd = &__get_cpu_var(softnet_data);
3146 fl = rcu_dereference(sd->flow_limit);
3148 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3149 old_flow = fl->history[fl->history_head];
3150 fl->history[fl->history_head] = new_flow;
3153 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3155 if (likely(fl->buckets[old_flow]))
3156 fl->buckets[old_flow]--;
3158 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3170 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3171 * queue (may be a remote CPU queue).
3173 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3174 unsigned int *qtail)
3176 struct softnet_data *sd;
3177 unsigned long flags;
3180 sd = &per_cpu(softnet_data, cpu);
3182 local_irq_save(flags);
3185 qlen = skb_queue_len(&sd->input_pkt_queue);
3186 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3187 if (skb_queue_len(&sd->input_pkt_queue)) {
3189 __skb_queue_tail(&sd->input_pkt_queue, skb);
3190 input_queue_tail_incr_save(sd, qtail);
3192 local_irq_restore(flags);
3193 return NET_RX_SUCCESS;
3196 /* Schedule NAPI for backlog device
3197 * We can use non atomic operation since we own the queue lock
3199 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3200 if (!rps_ipi_queued(sd))
3201 ____napi_schedule(sd, &sd->backlog);
3209 local_irq_restore(flags);
3211 atomic_long_inc(&skb->dev->rx_dropped);
3217 * netif_rx - post buffer to the network code
3218 * @skb: buffer to post
3220 * This function receives a packet from a device driver and queues it for
3221 * the upper (protocol) levels to process. It always succeeds. The buffer
3222 * may be dropped during processing for congestion control or by the
3226 * NET_RX_SUCCESS (no congestion)
3227 * NET_RX_DROP (packet was dropped)
3231 int netif_rx(struct sk_buff *skb)
3235 /* if netpoll wants it, pretend we never saw it */
3236 if (netpoll_rx(skb))
3239 net_timestamp_check(netdev_tstamp_prequeue, skb);
3241 trace_netif_rx(skb);
3243 if (static_key_false(&rps_needed)) {
3244 struct rps_dev_flow voidflow, *rflow = &voidflow;
3250 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3252 cpu = smp_processor_id();
3254 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3262 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3267 EXPORT_SYMBOL(netif_rx);
3269 int netif_rx_ni(struct sk_buff *skb)
3274 err = netif_rx(skb);
3275 if (local_softirq_pending())
3281 EXPORT_SYMBOL(netif_rx_ni);
3283 static void net_tx_action(struct softirq_action *h)
3285 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3287 if (sd->completion_queue) {
3288 struct sk_buff *clist;
3290 local_irq_disable();
3291 clist = sd->completion_queue;
3292 sd->completion_queue = NULL;
3296 struct sk_buff *skb = clist;
3297 clist = clist->next;
3299 WARN_ON(atomic_read(&skb->users));
3300 trace_kfree_skb(skb, net_tx_action);
3305 if (sd->output_queue) {
3308 local_irq_disable();
3309 head = sd->output_queue;
3310 sd->output_queue = NULL;
3311 sd->output_queue_tailp = &sd->output_queue;
3315 struct Qdisc *q = head;
3316 spinlock_t *root_lock;
3318 head = head->next_sched;
3320 root_lock = qdisc_lock(q);
3321 if (spin_trylock(root_lock)) {
3322 smp_mb__before_clear_bit();
3323 clear_bit(__QDISC_STATE_SCHED,
3326 spin_unlock(root_lock);
3328 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3330 __netif_reschedule(q);
3332 smp_mb__before_clear_bit();
3333 clear_bit(__QDISC_STATE_SCHED,
3341 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3342 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3343 /* This hook is defined here for ATM LANE */
3344 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3345 unsigned char *addr) __read_mostly;
3346 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3349 #ifdef CONFIG_NET_CLS_ACT
3350 /* TODO: Maybe we should just force sch_ingress to be compiled in
3351 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3352 * a compare and 2 stores extra right now if we dont have it on
3353 * but have CONFIG_NET_CLS_ACT
3354 * NOTE: This doesn't stop any functionality; if you dont have
3355 * the ingress scheduler, you just can't add policies on ingress.
3358 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3360 struct net_device *dev = skb->dev;
3361 u32 ttl = G_TC_RTTL(skb->tc_verd);
3362 int result = TC_ACT_OK;
3365 if (unlikely(MAX_RED_LOOP < ttl++)) {
3366 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3367 skb->skb_iif, dev->ifindex);
3371 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3372 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3375 if (q != &noop_qdisc) {
3376 spin_lock(qdisc_lock(q));
3377 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3378 result = qdisc_enqueue_root(skb, q);
3379 spin_unlock(qdisc_lock(q));
3385 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3386 struct packet_type **pt_prev,
3387 int *ret, struct net_device *orig_dev)
3389 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3391 if (!rxq || rxq->qdisc == &noop_qdisc)
3395 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3399 switch (ing_filter(skb, rxq)) {
3413 * netdev_rx_handler_register - register receive handler
3414 * @dev: device to register a handler for
3415 * @rx_handler: receive handler to register
3416 * @rx_handler_data: data pointer that is used by rx handler
3418 * Register a receive hander for a device. This handler will then be
3419 * called from __netif_receive_skb. A negative errno code is returned
3422 * The caller must hold the rtnl_mutex.
3424 * For a general description of rx_handler, see enum rx_handler_result.
3426 int netdev_rx_handler_register(struct net_device *dev,
3427 rx_handler_func_t *rx_handler,
3428 void *rx_handler_data)
3432 if (dev->rx_handler)
3435 /* Note: rx_handler_data must be set before rx_handler */
3436 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3437 rcu_assign_pointer(dev->rx_handler, rx_handler);
3441 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3444 * netdev_rx_handler_unregister - unregister receive handler
3445 * @dev: device to unregister a handler from
3447 * Unregister a receive handler from a device.
3449 * The caller must hold the rtnl_mutex.
3451 void netdev_rx_handler_unregister(struct net_device *dev)
3455 RCU_INIT_POINTER(dev->rx_handler, NULL);
3456 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3457 * section has a guarantee to see a non NULL rx_handler_data
3461 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3463 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3466 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3467 * the special handling of PFMEMALLOC skbs.
3469 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3471 switch (skb->protocol) {
3472 case __constant_htons(ETH_P_ARP):
3473 case __constant_htons(ETH_P_IP):
3474 case __constant_htons(ETH_P_IPV6):
3475 case __constant_htons(ETH_P_8021Q):
3476 case __constant_htons(ETH_P_8021AD):
3483 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3485 struct packet_type *ptype, *pt_prev;
3486 rx_handler_func_t *rx_handler;
3487 struct net_device *orig_dev;
3488 struct net_device *null_or_dev;
3489 bool deliver_exact = false;
3490 int ret = NET_RX_DROP;
3493 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3495 trace_netif_receive_skb(skb);
3497 /* if we've gotten here through NAPI, check netpoll */
3498 if (netpoll_receive_skb(skb))
3501 orig_dev = skb->dev;
3503 skb_reset_network_header(skb);
3504 if (!skb_transport_header_was_set(skb))
3505 skb_reset_transport_header(skb);
3506 skb_reset_mac_len(skb);
3513 skb->skb_iif = skb->dev->ifindex;
3515 __this_cpu_inc(softnet_data.processed);
3517 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3518 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3519 skb = vlan_untag(skb);
3524 #ifdef CONFIG_NET_CLS_ACT
3525 if (skb->tc_verd & TC_NCLS) {
3526 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3534 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3535 if (!ptype->dev || ptype->dev == skb->dev) {
3537 ret = deliver_skb(skb, pt_prev, orig_dev);
3543 #ifdef CONFIG_NET_CLS_ACT
3544 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3550 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3553 if (vlan_tx_tag_present(skb)) {
3555 ret = deliver_skb(skb, pt_prev, orig_dev);
3558 if (vlan_do_receive(&skb))
3560 else if (unlikely(!skb))
3564 rx_handler = rcu_dereference(skb->dev->rx_handler);
3567 ret = deliver_skb(skb, pt_prev, orig_dev);
3570 switch (rx_handler(&skb)) {
3571 case RX_HANDLER_CONSUMED:
3572 ret = NET_RX_SUCCESS;
3574 case RX_HANDLER_ANOTHER:
3576 case RX_HANDLER_EXACT:
3577 deliver_exact = true;
3578 case RX_HANDLER_PASS:
3585 if (unlikely(vlan_tx_tag_present(skb))) {
3586 if (vlan_tx_tag_get_id(skb))
3587 skb->pkt_type = PACKET_OTHERHOST;
3588 /* Note: we might in the future use prio bits
3589 * and set skb->priority like in vlan_do_receive()
3590 * For the time being, just ignore Priority Code Point
3595 /* deliver only exact match when indicated */
3596 null_or_dev = deliver_exact ? skb->dev : NULL;
3598 type = skb->protocol;
3599 list_for_each_entry_rcu(ptype,
3600 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3601 if (ptype->type == type &&
3602 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3603 ptype->dev == orig_dev)) {
3605 ret = deliver_skb(skb, pt_prev, orig_dev);
3611 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3614 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3617 atomic_long_inc(&skb->dev->rx_dropped);
3619 /* Jamal, now you will not able to escape explaining
3620 * me how you were going to use this. :-)
3631 static int __netif_receive_skb(struct sk_buff *skb)
3635 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3636 unsigned long pflags = current->flags;
3639 * PFMEMALLOC skbs are special, they should
3640 * - be delivered to SOCK_MEMALLOC sockets only
3641 * - stay away from userspace
3642 * - have bounded memory usage
3644 * Use PF_MEMALLOC as this saves us from propagating the allocation
3645 * context down to all allocation sites.
3647 current->flags |= PF_MEMALLOC;
3648 ret = __netif_receive_skb_core(skb, true);
3649 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3651 ret = __netif_receive_skb_core(skb, false);
3657 * netif_receive_skb - process receive buffer from network
3658 * @skb: buffer to process
3660 * netif_receive_skb() is the main receive data processing function.
3661 * It always succeeds. The buffer may be dropped during processing
3662 * for congestion control or by the protocol layers.
3664 * This function may only be called from softirq context and interrupts
3665 * should be enabled.
3667 * Return values (usually ignored):
3668 * NET_RX_SUCCESS: no congestion
3669 * NET_RX_DROP: packet was dropped
3671 int netif_receive_skb(struct sk_buff *skb)
3673 net_timestamp_check(netdev_tstamp_prequeue, skb);
3675 if (skb_defer_rx_timestamp(skb))
3676 return NET_RX_SUCCESS;
3679 if (static_key_false(&rps_needed)) {
3680 struct rps_dev_flow voidflow, *rflow = &voidflow;
3685 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3688 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3695 return __netif_receive_skb(skb);
3697 EXPORT_SYMBOL(netif_receive_skb);
3699 /* Network device is going away, flush any packets still pending
3700 * Called with irqs disabled.
3702 static void flush_backlog(void *arg)
3704 struct net_device *dev = arg;
3705 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3706 struct sk_buff *skb, *tmp;
3709 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3710 if (skb->dev == dev) {
3711 __skb_unlink(skb, &sd->input_pkt_queue);
3713 input_queue_head_incr(sd);
3718 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3719 if (skb->dev == dev) {
3720 __skb_unlink(skb, &sd->process_queue);
3722 input_queue_head_incr(sd);
3727 static int napi_gro_complete(struct sk_buff *skb)
3729 struct packet_offload *ptype;
3730 __be16 type = skb->protocol;
3731 struct list_head *head = &offload_base;
3734 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3736 if (NAPI_GRO_CB(skb)->count == 1) {
3737 skb_shinfo(skb)->gso_size = 0;
3742 list_for_each_entry_rcu(ptype, head, list) {
3743 if (ptype->type != type || !ptype->callbacks.gro_complete)
3746 err = ptype->callbacks.gro_complete(skb);
3752 WARN_ON(&ptype->list == head);
3754 return NET_RX_SUCCESS;
3758 return netif_receive_skb(skb);
3761 /* napi->gro_list contains packets ordered by age.
3762 * youngest packets at the head of it.
3763 * Complete skbs in reverse order to reduce latencies.
3765 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3767 struct sk_buff *skb, *prev = NULL;
3769 /* scan list and build reverse chain */
3770 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3775 for (skb = prev; skb; skb = prev) {
3778 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3782 napi_gro_complete(skb);
3786 napi->gro_list = NULL;
3788 EXPORT_SYMBOL(napi_gro_flush);
3790 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3793 unsigned int maclen = skb->dev->hard_header_len;
3795 for (p = napi->gro_list; p; p = p->next) {
3796 unsigned long diffs;
3798 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3799 diffs |= p->vlan_tci ^ skb->vlan_tci;
3800 if (maclen == ETH_HLEN)
3801 diffs |= compare_ether_header(skb_mac_header(p),
3802 skb_gro_mac_header(skb));
3804 diffs = memcmp(skb_mac_header(p),
3805 skb_gro_mac_header(skb),
3807 NAPI_GRO_CB(p)->same_flow = !diffs;
3808 NAPI_GRO_CB(p)->flush = 0;
3812 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3814 struct sk_buff **pp = NULL;
3815 struct packet_offload *ptype;
3816 __be16 type = skb->protocol;
3817 struct list_head *head = &offload_base;
3819 enum gro_result ret;
3821 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3824 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3827 gro_list_prepare(napi, skb);
3830 list_for_each_entry_rcu(ptype, head, list) {
3831 if (ptype->type != type || !ptype->callbacks.gro_receive)
3834 skb_set_network_header(skb, skb_gro_offset(skb));
3835 skb_reset_mac_len(skb);
3836 NAPI_GRO_CB(skb)->same_flow = 0;
3837 NAPI_GRO_CB(skb)->flush = 0;
3838 NAPI_GRO_CB(skb)->free = 0;
3840 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3845 if (&ptype->list == head)
3848 same_flow = NAPI_GRO_CB(skb)->same_flow;
3849 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3852 struct sk_buff *nskb = *pp;
3856 napi_gro_complete(nskb);
3863 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3867 NAPI_GRO_CB(skb)->count = 1;
3868 NAPI_GRO_CB(skb)->age = jiffies;
3869 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3870 skb->next = napi->gro_list;
3871 napi->gro_list = skb;
3875 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3876 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3878 BUG_ON(skb->end - skb->tail < grow);
3880 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3883 skb->data_len -= grow;
3885 skb_shinfo(skb)->frags[0].page_offset += grow;
3886 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3888 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3889 skb_frag_unref(skb, 0);
3890 memmove(skb_shinfo(skb)->frags,
3891 skb_shinfo(skb)->frags + 1,
3892 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3905 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3909 if (netif_receive_skb(skb))
3917 case GRO_MERGED_FREE:
3918 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3919 kmem_cache_free(skbuff_head_cache, skb);
3932 static void skb_gro_reset_offset(struct sk_buff *skb)
3934 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3935 const skb_frag_t *frag0 = &pinfo->frags[0];
3937 NAPI_GRO_CB(skb)->data_offset = 0;
3938 NAPI_GRO_CB(skb)->frag0 = NULL;
3939 NAPI_GRO_CB(skb)->frag0_len = 0;
3941 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3943 !PageHighMem(skb_frag_page(frag0))) {
3944 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3945 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3949 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3951 skb_gro_reset_offset(skb);
3953 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3955 EXPORT_SYMBOL(napi_gro_receive);
3957 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3959 __skb_pull(skb, skb_headlen(skb));
3960 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3961 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3963 skb->dev = napi->dev;
3969 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3971 struct sk_buff *skb = napi->skb;
3974 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3980 EXPORT_SYMBOL(napi_get_frags);
3982 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3988 skb->protocol = eth_type_trans(skb, skb->dev);
3990 if (ret == GRO_HELD)
3991 skb_gro_pull(skb, -ETH_HLEN);
3992 else if (netif_receive_skb(skb))
3997 case GRO_MERGED_FREE:
3998 napi_reuse_skb(napi, skb);
4008 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4010 struct sk_buff *skb = napi->skb;
4017 skb_reset_mac_header(skb);
4018 skb_gro_reset_offset(skb);
4020 off = skb_gro_offset(skb);
4021 hlen = off + sizeof(*eth);
4022 eth = skb_gro_header_fast(skb, off);
4023 if (skb_gro_header_hard(skb, hlen)) {
4024 eth = skb_gro_header_slow(skb, hlen, off);
4025 if (unlikely(!eth)) {
4026 napi_reuse_skb(napi, skb);
4032 skb_gro_pull(skb, sizeof(*eth));
4035 * This works because the only protocols we care about don't require
4036 * special handling. We'll fix it up properly at the end.
4038 skb->protocol = eth->h_proto;
4044 gro_result_t napi_gro_frags(struct napi_struct *napi)
4046 struct sk_buff *skb = napi_frags_skb(napi);
4051 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4053 EXPORT_SYMBOL(napi_gro_frags);
4056 * net_rps_action sends any pending IPI's for rps.
4057 * Note: called with local irq disabled, but exits with local irq enabled.
4059 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4062 struct softnet_data *remsd = sd->rps_ipi_list;
4065 sd->rps_ipi_list = NULL;
4069 /* Send pending IPI's to kick RPS processing on remote cpus. */
4071 struct softnet_data *next = remsd->rps_ipi_next;
4073 if (cpu_online(remsd->cpu))
4074 __smp_call_function_single(remsd->cpu,
4083 static int process_backlog(struct napi_struct *napi, int quota)
4086 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4089 /* Check if we have pending ipi, its better to send them now,
4090 * not waiting net_rx_action() end.
4092 if (sd->rps_ipi_list) {
4093 local_irq_disable();
4094 net_rps_action_and_irq_enable(sd);
4097 napi->weight = weight_p;
4098 local_irq_disable();
4099 while (work < quota) {
4100 struct sk_buff *skb;
4103 while ((skb = __skb_dequeue(&sd->process_queue))) {
4105 __netif_receive_skb(skb);
4106 local_irq_disable();
4107 input_queue_head_incr(sd);
4108 if (++work >= quota) {
4115 qlen = skb_queue_len(&sd->input_pkt_queue);
4117 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4118 &sd->process_queue);
4120 if (qlen < quota - work) {
4122 * Inline a custom version of __napi_complete().
4123 * only current cpu owns and manipulates this napi,
4124 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4125 * we can use a plain write instead of clear_bit(),
4126 * and we dont need an smp_mb() memory barrier.
4128 list_del(&napi->poll_list);
4131 quota = work + qlen;
4141 * __napi_schedule - schedule for receive
4142 * @n: entry to schedule
4144 * The entry's receive function will be scheduled to run
4146 void __napi_schedule(struct napi_struct *n)
4148 unsigned long flags;
4150 local_irq_save(flags);
4151 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4152 local_irq_restore(flags);
4154 EXPORT_SYMBOL(__napi_schedule);
4156 void __napi_complete(struct napi_struct *n)
4158 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4159 BUG_ON(n->gro_list);
4161 list_del(&n->poll_list);
4162 smp_mb__before_clear_bit();
4163 clear_bit(NAPI_STATE_SCHED, &n->state);
4165 EXPORT_SYMBOL(__napi_complete);
4167 void napi_complete(struct napi_struct *n)
4169 unsigned long flags;
4172 * don't let napi dequeue from the cpu poll list
4173 * just in case its running on a different cpu
4175 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4178 napi_gro_flush(n, false);
4179 local_irq_save(flags);
4181 local_irq_restore(flags);
4183 EXPORT_SYMBOL(napi_complete);
4185 /* must be called under rcu_read_lock(), as we dont take a reference */
4186 struct napi_struct *napi_by_id(unsigned int napi_id)
4188 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4189 struct napi_struct *napi;
4191 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4192 if (napi->napi_id == napi_id)
4197 EXPORT_SYMBOL_GPL(napi_by_id);
4199 void napi_hash_add(struct napi_struct *napi)
4201 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4203 spin_lock(&napi_hash_lock);
4205 /* 0 is not a valid id, we also skip an id that is taken
4206 * we expect both events to be extremely rare
4209 while (!napi->napi_id) {
4210 napi->napi_id = ++napi_gen_id;
4211 if (napi_by_id(napi->napi_id))
4215 hlist_add_head_rcu(&napi->napi_hash_node,
4216 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4218 spin_unlock(&napi_hash_lock);
4221 EXPORT_SYMBOL_GPL(napi_hash_add);
4223 /* Warning : caller is responsible to make sure rcu grace period
4224 * is respected before freeing memory containing @napi
4226 void napi_hash_del(struct napi_struct *napi)
4228 spin_lock(&napi_hash_lock);
4230 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4231 hlist_del_rcu(&napi->napi_hash_node);
4233 spin_unlock(&napi_hash_lock);
4235 EXPORT_SYMBOL_GPL(napi_hash_del);
4237 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4238 int (*poll)(struct napi_struct *, int), int weight)
4240 INIT_LIST_HEAD(&napi->poll_list);
4241 napi->gro_count = 0;
4242 napi->gro_list = NULL;
4245 if (weight > NAPI_POLL_WEIGHT)
4246 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4248 napi->weight = weight;
4249 list_add(&napi->dev_list, &dev->napi_list);
4251 #ifdef CONFIG_NETPOLL
4252 spin_lock_init(&napi->poll_lock);
4253 napi->poll_owner = -1;
4255 set_bit(NAPI_STATE_SCHED, &napi->state);
4257 EXPORT_SYMBOL(netif_napi_add);
4259 void netif_napi_del(struct napi_struct *napi)
4261 struct sk_buff *skb, *next;
4263 list_del_init(&napi->dev_list);
4264 napi_free_frags(napi);
4266 for (skb = napi->gro_list; skb; skb = next) {
4272 napi->gro_list = NULL;
4273 napi->gro_count = 0;
4275 EXPORT_SYMBOL(netif_napi_del);
4277 static void net_rx_action(struct softirq_action *h)
4279 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4280 unsigned long time_limit = jiffies + 2;
4281 int budget = netdev_budget;
4284 local_irq_disable();
4286 while (!list_empty(&sd->poll_list)) {
4287 struct napi_struct *n;
4290 /* If softirq window is exhuasted then punt.
4291 * Allow this to run for 2 jiffies since which will allow
4292 * an average latency of 1.5/HZ.
4294 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4299 /* Even though interrupts have been re-enabled, this
4300 * access is safe because interrupts can only add new
4301 * entries to the tail of this list, and only ->poll()
4302 * calls can remove this head entry from the list.
4304 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4306 have = netpoll_poll_lock(n);
4310 /* This NAPI_STATE_SCHED test is for avoiding a race
4311 * with netpoll's poll_napi(). Only the entity which
4312 * obtains the lock and sees NAPI_STATE_SCHED set will
4313 * actually make the ->poll() call. Therefore we avoid
4314 * accidentally calling ->poll() when NAPI is not scheduled.
4317 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4318 work = n->poll(n, weight);
4322 WARN_ON_ONCE(work > weight);
4326 local_irq_disable();
4328 /* Drivers must not modify the NAPI state if they
4329 * consume the entire weight. In such cases this code
4330 * still "owns" the NAPI instance and therefore can
4331 * move the instance around on the list at-will.
4333 if (unlikely(work == weight)) {
4334 if (unlikely(napi_disable_pending(n))) {
4337 local_irq_disable();
4340 /* flush too old packets
4341 * If HZ < 1000, flush all packets.
4344 napi_gro_flush(n, HZ >= 1000);
4345 local_irq_disable();
4347 list_move_tail(&n->poll_list, &sd->poll_list);
4351 netpoll_poll_unlock(have);
4354 net_rps_action_and_irq_enable(sd);
4356 #ifdef CONFIG_NET_DMA
4358 * There may not be any more sk_buffs coming right now, so push
4359 * any pending DMA copies to hardware
4361 dma_issue_pending_all();
4368 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4372 struct netdev_adjacent {
4373 struct net_device *dev;
4375 /* upper master flag, there can only be one master device per list */
4378 /* counter for the number of times this device was added to us */
4381 /* private field for the users */
4384 struct list_head list;
4385 struct rcu_head rcu;
4388 static struct netdev_adjacent *__netdev_find_adj_rcu(struct net_device *dev,
4389 struct net_device *adj_dev,
4390 struct list_head *adj_list)
4392 struct netdev_adjacent *adj;
4394 list_for_each_entry_rcu(adj, adj_list, list) {
4395 if (adj->dev == adj_dev)
4401 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4402 struct net_device *adj_dev,
4403 struct list_head *adj_list)
4405 struct netdev_adjacent *adj;
4407 list_for_each_entry(adj, adj_list, list) {
4408 if (adj->dev == adj_dev)
4415 * netdev_has_upper_dev - Check if device is linked to an upper device
4417 * @upper_dev: upper device to check
4419 * Find out if a device is linked to specified upper device and return true
4420 * in case it is. Note that this checks only immediate upper device,
4421 * not through a complete stack of devices. The caller must hold the RTNL lock.
4423 bool netdev_has_upper_dev(struct net_device *dev,
4424 struct net_device *upper_dev)
4428 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4430 EXPORT_SYMBOL(netdev_has_upper_dev);
4433 * netdev_has_any_upper_dev - Check if device is linked to some device
4436 * Find out if a device is linked to an upper device and return true in case
4437 * it is. The caller must hold the RTNL lock.
4439 bool netdev_has_any_upper_dev(struct net_device *dev)
4443 return !list_empty(&dev->all_adj_list.upper);
4445 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4448 * netdev_master_upper_dev_get - Get master upper device
4451 * Find a master upper device and return pointer to it or NULL in case
4452 * it's not there. The caller must hold the RTNL lock.
4454 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4456 struct netdev_adjacent *upper;
4460 if (list_empty(&dev->adj_list.upper))
4463 upper = list_first_entry(&dev->adj_list.upper,
4464 struct netdev_adjacent, list);
4465 if (likely(upper->master))
4469 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4471 void *netdev_adjacent_get_private(struct list_head *adj_list)
4473 struct netdev_adjacent *adj;
4475 adj = list_entry(adj_list, struct netdev_adjacent, list);
4477 return adj->private;
4479 EXPORT_SYMBOL(netdev_adjacent_get_private);
4482 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4484 * @iter: list_head ** of the current position
4486 * Gets the next device from the dev's upper list, starting from iter
4487 * position. The caller must hold RCU read lock.
4489 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4490 struct list_head **iter)
4492 struct netdev_adjacent *upper;
4494 WARN_ON_ONCE(!rcu_read_lock_held());
4496 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4498 if (&upper->list == &dev->all_adj_list.upper)
4501 *iter = &upper->list;
4505 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4508 * netdev_lower_get_next_private - Get the next ->private from the
4509 * lower neighbour list
4511 * @iter: list_head ** of the current position
4513 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4514 * list, starting from iter position. The caller must hold either hold the
4515 * RTNL lock or its own locking that guarantees that the neighbour lower
4516 * list will remain unchainged.
4518 void *netdev_lower_get_next_private(struct net_device *dev,
4519 struct list_head **iter)
4521 struct netdev_adjacent *lower;
4523 lower = list_entry(*iter, struct netdev_adjacent, list);
4525 if (&lower->list == &dev->adj_list.lower)
4529 *iter = lower->list.next;
4531 return lower->private;
4533 EXPORT_SYMBOL(netdev_lower_get_next_private);
4536 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4537 * lower neighbour list, RCU
4540 * @iter: list_head ** of the current position
4542 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4543 * list, starting from iter position. The caller must hold RCU read lock.
4545 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4546 struct list_head **iter)
4548 struct netdev_adjacent *lower;
4550 WARN_ON_ONCE(!rcu_read_lock_held());
4552 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4554 if (&lower->list == &dev->adj_list.lower)
4558 *iter = &lower->list;
4560 return lower->private;
4562 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4565 * netdev_master_upper_dev_get_rcu - Get master upper device
4568 * Find a master upper device and return pointer to it or NULL in case
4569 * it's not there. The caller must hold the RCU read lock.
4571 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4573 struct netdev_adjacent *upper;
4575 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4576 struct netdev_adjacent, list);
4577 if (upper && likely(upper->master))
4581 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4583 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4584 struct net_device *adj_dev,
4585 struct list_head *dev_list,
4586 void *private, bool master)
4588 struct netdev_adjacent *adj;
4589 char linkname[IFNAMSIZ+7];
4592 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4599 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4604 adj->master = master;
4606 adj->private = private;
4609 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4610 adj_dev->name, dev->name, adj_dev->name);
4612 if (dev_list == &dev->adj_list.lower) {
4613 sprintf(linkname, "lower_%s", adj_dev->name);
4614 ret = sysfs_create_link(&(dev->dev.kobj),
4615 &(adj_dev->dev.kobj), linkname);
4618 } else if (dev_list == &dev->adj_list.upper) {
4619 sprintf(linkname, "upper_%s", adj_dev->name);
4620 ret = sysfs_create_link(&(dev->dev.kobj),
4621 &(adj_dev->dev.kobj), linkname);
4626 /* Ensure that master link is always the first item in list. */
4628 ret = sysfs_create_link(&(dev->dev.kobj),
4629 &(adj_dev->dev.kobj), "master");
4631 goto remove_symlinks;
4633 list_add_rcu(&adj->list, dev_list);
4635 list_add_tail_rcu(&adj->list, dev_list);
4641 if (dev_list == &dev->adj_list.lower) {
4642 sprintf(linkname, "lower_%s", adj_dev->name);
4643 sysfs_remove_link(&(dev->dev.kobj), linkname);
4644 } else if (dev_list == &dev->adj_list.upper) {
4645 sprintf(linkname, "upper_%s", adj_dev->name);
4646 sysfs_remove_link(&(dev->dev.kobj), linkname);
4655 void __netdev_adjacent_dev_remove(struct net_device *dev,
4656 struct net_device *adj_dev,
4657 struct list_head *dev_list)
4659 struct netdev_adjacent *adj;
4660 char linkname[IFNAMSIZ+7];
4662 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4665 pr_err("tried to remove device %s from %s\n",
4666 dev->name, adj_dev->name);
4670 if (adj->ref_nr > 1) {
4671 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4678 sysfs_remove_link(&(dev->dev.kobj), "master");
4680 if (dev_list == &dev->adj_list.lower) {
4681 sprintf(linkname, "lower_%s", adj_dev->name);
4682 sysfs_remove_link(&(dev->dev.kobj), linkname);
4683 } else if (dev_list == &dev->adj_list.upper) {
4684 sprintf(linkname, "upper_%s", adj_dev->name);
4685 sysfs_remove_link(&(dev->dev.kobj), linkname);
4688 list_del_rcu(&adj->list);
4689 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4690 adj_dev->name, dev->name, adj_dev->name);
4692 kfree_rcu(adj, rcu);
4695 int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4696 struct net_device *upper_dev,
4697 struct list_head *up_list,
4698 struct list_head *down_list,
4699 void *private, bool master)
4703 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4708 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4711 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4718 int __netdev_adjacent_dev_link(struct net_device *dev,
4719 struct net_device *upper_dev)
4721 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4722 &dev->all_adj_list.upper,
4723 &upper_dev->all_adj_list.lower,
4727 void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4728 struct net_device *upper_dev,
4729 struct list_head *up_list,
4730 struct list_head *down_list)
4732 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4733 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4736 void __netdev_adjacent_dev_unlink(struct net_device *dev,
4737 struct net_device *upper_dev)
4739 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4740 &dev->all_adj_list.upper,
4741 &upper_dev->all_adj_list.lower);
4744 int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4745 struct net_device *upper_dev,
4746 void *private, bool master)
4748 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4753 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4754 &dev->adj_list.upper,
4755 &upper_dev->adj_list.lower,
4758 __netdev_adjacent_dev_unlink(dev, upper_dev);
4765 void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4766 struct net_device *upper_dev)
4768 __netdev_adjacent_dev_unlink(dev, upper_dev);
4769 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4770 &dev->adj_list.upper,
4771 &upper_dev->adj_list.lower);
4774 static int __netdev_upper_dev_link(struct net_device *dev,
4775 struct net_device *upper_dev, bool master,
4778 struct netdev_adjacent *i, *j, *to_i, *to_j;
4783 if (dev == upper_dev)
4786 /* To prevent loops, check if dev is not upper device to upper_dev. */
4787 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4790 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4793 if (master && netdev_master_upper_dev_get(dev))
4796 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4801 /* Now that we linked these devs, make all the upper_dev's
4802 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4803 * versa, and don't forget the devices itself. All of these
4804 * links are non-neighbours.
4806 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4807 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4808 pr_debug("Interlinking %s with %s, non-neighbour\n",
4809 i->dev->name, j->dev->name);
4810 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4816 /* add dev to every upper_dev's upper device */
4817 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4818 pr_debug("linking %s's upper device %s with %s\n",
4819 upper_dev->name, i->dev->name, dev->name);
4820 ret = __netdev_adjacent_dev_link(dev, i->dev);
4822 goto rollback_upper_mesh;
4825 /* add upper_dev to every dev's lower device */
4826 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4827 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4828 i->dev->name, upper_dev->name);
4829 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4831 goto rollback_lower_mesh;
4834 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4837 rollback_lower_mesh:
4839 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4842 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4847 rollback_upper_mesh:
4849 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4852 __netdev_adjacent_dev_unlink(dev, i->dev);
4860 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4861 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4862 if (i == to_i && j == to_j)
4864 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4870 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4876 * netdev_upper_dev_link - Add a link to the upper device
4878 * @upper_dev: new upper device
4880 * Adds a link to device which is upper to this one. The caller must hold
4881 * the RTNL lock. On a failure a negative errno code is returned.
4882 * On success the reference counts are adjusted and the function
4885 int netdev_upper_dev_link(struct net_device *dev,
4886 struct net_device *upper_dev)
4888 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4890 EXPORT_SYMBOL(netdev_upper_dev_link);
4893 * netdev_master_upper_dev_link - Add a master link to the upper device
4895 * @upper_dev: new upper device
4897 * Adds a link to device which is upper to this one. In this case, only
4898 * one master upper device can be linked, although other non-master devices
4899 * might be linked as well. The caller must hold the RTNL lock.
4900 * On a failure a negative errno code is returned. On success the reference
4901 * counts are adjusted and the function returns zero.
4903 int netdev_master_upper_dev_link(struct net_device *dev,
4904 struct net_device *upper_dev)
4906 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4908 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4910 int netdev_master_upper_dev_link_private(struct net_device *dev,
4911 struct net_device *upper_dev,
4914 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4916 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4919 * netdev_upper_dev_unlink - Removes a link to upper device
4921 * @upper_dev: new upper device
4923 * Removes a link to device which is upper to this one. The caller must hold
4926 void netdev_upper_dev_unlink(struct net_device *dev,
4927 struct net_device *upper_dev)
4929 struct netdev_adjacent *i, *j;
4932 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4934 /* Here is the tricky part. We must remove all dev's lower
4935 * devices from all upper_dev's upper devices and vice
4936 * versa, to maintain the graph relationship.
4938 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4939 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4940 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4942 /* remove also the devices itself from lower/upper device
4945 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4946 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4948 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4949 __netdev_adjacent_dev_unlink(dev, i->dev);
4951 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4953 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4955 void *netdev_lower_dev_get_private_rcu(struct net_device *dev,
4956 struct net_device *lower_dev)
4958 struct netdev_adjacent *lower;
4962 lower = __netdev_find_adj_rcu(dev, lower_dev, &dev->adj_list.lower);
4966 return lower->private;
4968 EXPORT_SYMBOL(netdev_lower_dev_get_private_rcu);
4970 void *netdev_lower_dev_get_private(struct net_device *dev,
4971 struct net_device *lower_dev)
4973 struct netdev_adjacent *lower;
4977 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
4981 return lower->private;
4983 EXPORT_SYMBOL(netdev_lower_dev_get_private);
4985 static void dev_change_rx_flags(struct net_device *dev, int flags)
4987 const struct net_device_ops *ops = dev->netdev_ops;
4989 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4990 ops->ndo_change_rx_flags(dev, flags);
4993 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
4995 unsigned int old_flags = dev->flags;
5001 dev->flags |= IFF_PROMISC;
5002 dev->promiscuity += inc;
5003 if (dev->promiscuity == 0) {
5006 * If inc causes overflow, untouch promisc and return error.
5009 dev->flags &= ~IFF_PROMISC;
5011 dev->promiscuity -= inc;
5012 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5017 if (dev->flags != old_flags) {
5018 pr_info("device %s %s promiscuous mode\n",
5020 dev->flags & IFF_PROMISC ? "entered" : "left");
5021 if (audit_enabled) {
5022 current_uid_gid(&uid, &gid);
5023 audit_log(current->audit_context, GFP_ATOMIC,
5024 AUDIT_ANOM_PROMISCUOUS,
5025 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5026 dev->name, (dev->flags & IFF_PROMISC),
5027 (old_flags & IFF_PROMISC),
5028 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5029 from_kuid(&init_user_ns, uid),
5030 from_kgid(&init_user_ns, gid),
5031 audit_get_sessionid(current));
5034 dev_change_rx_flags(dev, IFF_PROMISC);
5037 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5042 * dev_set_promiscuity - update promiscuity count on a device
5046 * Add or remove promiscuity from a device. While the count in the device
5047 * remains above zero the interface remains promiscuous. Once it hits zero
5048 * the device reverts back to normal filtering operation. A negative inc
5049 * value is used to drop promiscuity on the device.
5050 * Return 0 if successful or a negative errno code on error.
5052 int dev_set_promiscuity(struct net_device *dev, int inc)
5054 unsigned int old_flags = dev->flags;
5057 err = __dev_set_promiscuity(dev, inc, true);
5060 if (dev->flags != old_flags)
5061 dev_set_rx_mode(dev);
5064 EXPORT_SYMBOL(dev_set_promiscuity);
5066 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5068 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5072 dev->flags |= IFF_ALLMULTI;
5073 dev->allmulti += inc;
5074 if (dev->allmulti == 0) {
5077 * If inc causes overflow, untouch allmulti and return error.
5080 dev->flags &= ~IFF_ALLMULTI;
5082 dev->allmulti -= inc;
5083 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5088 if (dev->flags ^ old_flags) {
5089 dev_change_rx_flags(dev, IFF_ALLMULTI);
5090 dev_set_rx_mode(dev);
5092 __dev_notify_flags(dev, old_flags,
5093 dev->gflags ^ old_gflags);
5099 * dev_set_allmulti - update allmulti count on a device
5103 * Add or remove reception of all multicast frames to a device. While the
5104 * count in the device remains above zero the interface remains listening
5105 * to all interfaces. Once it hits zero the device reverts back to normal
5106 * filtering operation. A negative @inc value is used to drop the counter
5107 * when releasing a resource needing all multicasts.
5108 * Return 0 if successful or a negative errno code on error.
5111 int dev_set_allmulti(struct net_device *dev, int inc)
5113 return __dev_set_allmulti(dev, inc, true);
5115 EXPORT_SYMBOL(dev_set_allmulti);
5118 * Upload unicast and multicast address lists to device and
5119 * configure RX filtering. When the device doesn't support unicast
5120 * filtering it is put in promiscuous mode while unicast addresses
5123 void __dev_set_rx_mode(struct net_device *dev)
5125 const struct net_device_ops *ops = dev->netdev_ops;
5127 /* dev_open will call this function so the list will stay sane. */
5128 if (!(dev->flags&IFF_UP))
5131 if (!netif_device_present(dev))
5134 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5135 /* Unicast addresses changes may only happen under the rtnl,
5136 * therefore calling __dev_set_promiscuity here is safe.
5138 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5139 __dev_set_promiscuity(dev, 1, false);
5140 dev->uc_promisc = true;
5141 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5142 __dev_set_promiscuity(dev, -1, false);
5143 dev->uc_promisc = false;
5147 if (ops->ndo_set_rx_mode)
5148 ops->ndo_set_rx_mode(dev);
5151 void dev_set_rx_mode(struct net_device *dev)
5153 netif_addr_lock_bh(dev);
5154 __dev_set_rx_mode(dev);
5155 netif_addr_unlock_bh(dev);
5159 * dev_get_flags - get flags reported to userspace
5162 * Get the combination of flag bits exported through APIs to userspace.
5164 unsigned int dev_get_flags(const struct net_device *dev)
5168 flags = (dev->flags & ~(IFF_PROMISC |
5173 (dev->gflags & (IFF_PROMISC |
5176 if (netif_running(dev)) {
5177 if (netif_oper_up(dev))
5178 flags |= IFF_RUNNING;
5179 if (netif_carrier_ok(dev))
5180 flags |= IFF_LOWER_UP;
5181 if (netif_dormant(dev))
5182 flags |= IFF_DORMANT;
5187 EXPORT_SYMBOL(dev_get_flags);
5189 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5191 unsigned int old_flags = dev->flags;
5197 * Set the flags on our device.
5200 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5201 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5203 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5207 * Load in the correct multicast list now the flags have changed.
5210 if ((old_flags ^ flags) & IFF_MULTICAST)
5211 dev_change_rx_flags(dev, IFF_MULTICAST);
5213 dev_set_rx_mode(dev);
5216 * Have we downed the interface. We handle IFF_UP ourselves
5217 * according to user attempts to set it, rather than blindly
5222 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5223 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5226 dev_set_rx_mode(dev);
5229 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5230 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5231 unsigned int old_flags = dev->flags;
5233 dev->gflags ^= IFF_PROMISC;
5235 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5236 if (dev->flags != old_flags)
5237 dev_set_rx_mode(dev);
5240 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5241 is important. Some (broken) drivers set IFF_PROMISC, when
5242 IFF_ALLMULTI is requested not asking us and not reporting.
5244 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5245 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5247 dev->gflags ^= IFF_ALLMULTI;
5248 __dev_set_allmulti(dev, inc, false);
5254 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5255 unsigned int gchanges)
5257 unsigned int changes = dev->flags ^ old_flags;
5260 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges);
5262 if (changes & IFF_UP) {
5263 if (dev->flags & IFF_UP)
5264 call_netdevice_notifiers(NETDEV_UP, dev);
5266 call_netdevice_notifiers(NETDEV_DOWN, dev);
5269 if (dev->flags & IFF_UP &&
5270 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5271 struct netdev_notifier_change_info change_info;
5273 change_info.flags_changed = changes;
5274 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5280 * dev_change_flags - change device settings
5282 * @flags: device state flags
5284 * Change settings on device based state flags. The flags are
5285 * in the userspace exported format.
5287 int dev_change_flags(struct net_device *dev, unsigned int flags)
5290 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5292 ret = __dev_change_flags(dev, flags);
5296 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5297 __dev_notify_flags(dev, old_flags, changes);
5300 EXPORT_SYMBOL(dev_change_flags);
5303 * dev_set_mtu - Change maximum transfer unit
5305 * @new_mtu: new transfer unit
5307 * Change the maximum transfer size of the network device.
5309 int dev_set_mtu(struct net_device *dev, int new_mtu)
5311 const struct net_device_ops *ops = dev->netdev_ops;
5314 if (new_mtu == dev->mtu)
5317 /* MTU must be positive. */
5321 if (!netif_device_present(dev))
5325 if (ops->ndo_change_mtu)
5326 err = ops->ndo_change_mtu(dev, new_mtu);
5331 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5334 EXPORT_SYMBOL(dev_set_mtu);
5337 * dev_set_group - Change group this device belongs to
5339 * @new_group: group this device should belong to
5341 void dev_set_group(struct net_device *dev, int new_group)
5343 dev->group = new_group;
5345 EXPORT_SYMBOL(dev_set_group);
5348 * dev_set_mac_address - Change Media Access Control Address
5352 * Change the hardware (MAC) address of the device
5354 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5356 const struct net_device_ops *ops = dev->netdev_ops;
5359 if (!ops->ndo_set_mac_address)
5361 if (sa->sa_family != dev->type)
5363 if (!netif_device_present(dev))
5365 err = ops->ndo_set_mac_address(dev, sa);
5368 dev->addr_assign_type = NET_ADDR_SET;
5369 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5370 add_device_randomness(dev->dev_addr, dev->addr_len);
5373 EXPORT_SYMBOL(dev_set_mac_address);
5376 * dev_change_carrier - Change device carrier
5378 * @new_carrier: new value
5380 * Change device carrier
5382 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5384 const struct net_device_ops *ops = dev->netdev_ops;
5386 if (!ops->ndo_change_carrier)
5388 if (!netif_device_present(dev))
5390 return ops->ndo_change_carrier(dev, new_carrier);
5392 EXPORT_SYMBOL(dev_change_carrier);
5395 * dev_get_phys_port_id - Get device physical port ID
5399 * Get device physical port ID
5401 int dev_get_phys_port_id(struct net_device *dev,
5402 struct netdev_phys_port_id *ppid)
5404 const struct net_device_ops *ops = dev->netdev_ops;
5406 if (!ops->ndo_get_phys_port_id)
5408 return ops->ndo_get_phys_port_id(dev, ppid);
5410 EXPORT_SYMBOL(dev_get_phys_port_id);
5413 * dev_new_index - allocate an ifindex
5414 * @net: the applicable net namespace
5416 * Returns a suitable unique value for a new device interface
5417 * number. The caller must hold the rtnl semaphore or the
5418 * dev_base_lock to be sure it remains unique.
5420 static int dev_new_index(struct net *net)
5422 int ifindex = net->ifindex;
5426 if (!__dev_get_by_index(net, ifindex))
5427 return net->ifindex = ifindex;
5431 /* Delayed registration/unregisteration */
5432 static LIST_HEAD(net_todo_list);
5433 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5435 static void net_set_todo(struct net_device *dev)
5437 list_add_tail(&dev->todo_list, &net_todo_list);
5438 dev_net(dev)->dev_unreg_count++;
5441 static void rollback_registered_many(struct list_head *head)
5443 struct net_device *dev, *tmp;
5444 LIST_HEAD(close_head);
5446 BUG_ON(dev_boot_phase);
5449 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5450 /* Some devices call without registering
5451 * for initialization unwind. Remove those
5452 * devices and proceed with the remaining.
5454 if (dev->reg_state == NETREG_UNINITIALIZED) {
5455 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5459 list_del(&dev->unreg_list);
5462 dev->dismantle = true;
5463 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5466 /* If device is running, close it first. */
5467 list_for_each_entry(dev, head, unreg_list)
5468 list_add_tail(&dev->close_list, &close_head);
5469 dev_close_many(&close_head);
5471 list_for_each_entry(dev, head, unreg_list) {
5472 /* And unlink it from device chain. */
5473 unlist_netdevice(dev);
5475 dev->reg_state = NETREG_UNREGISTERING;
5480 list_for_each_entry(dev, head, unreg_list) {
5481 /* Shutdown queueing discipline. */
5485 /* Notify protocols, that we are about to destroy
5486 this device. They should clean all the things.
5488 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5490 if (!dev->rtnl_link_ops ||
5491 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5492 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5495 * Flush the unicast and multicast chains
5500 if (dev->netdev_ops->ndo_uninit)
5501 dev->netdev_ops->ndo_uninit(dev);
5503 /* Notifier chain MUST detach us all upper devices. */
5504 WARN_ON(netdev_has_any_upper_dev(dev));
5506 /* Remove entries from kobject tree */
5507 netdev_unregister_kobject(dev);
5509 /* Remove XPS queueing entries */
5510 netif_reset_xps_queues_gt(dev, 0);
5516 list_for_each_entry(dev, head, unreg_list)
5520 static void rollback_registered(struct net_device *dev)
5524 list_add(&dev->unreg_list, &single);
5525 rollback_registered_many(&single);
5529 static netdev_features_t netdev_fix_features(struct net_device *dev,
5530 netdev_features_t features)
5532 /* Fix illegal checksum combinations */
5533 if ((features & NETIF_F_HW_CSUM) &&
5534 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5535 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5536 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5539 /* TSO requires that SG is present as well. */
5540 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5541 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5542 features &= ~NETIF_F_ALL_TSO;
5545 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5546 !(features & NETIF_F_IP_CSUM)) {
5547 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5548 features &= ~NETIF_F_TSO;
5549 features &= ~NETIF_F_TSO_ECN;
5552 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5553 !(features & NETIF_F_IPV6_CSUM)) {
5554 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5555 features &= ~NETIF_F_TSO6;
5558 /* TSO ECN requires that TSO is present as well. */
5559 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5560 features &= ~NETIF_F_TSO_ECN;
5562 /* Software GSO depends on SG. */
5563 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5564 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5565 features &= ~NETIF_F_GSO;
5568 /* UFO needs SG and checksumming */
5569 if (features & NETIF_F_UFO) {
5570 /* maybe split UFO into V4 and V6? */
5571 if (!((features & NETIF_F_GEN_CSUM) ||
5572 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5573 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5575 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5576 features &= ~NETIF_F_UFO;
5579 if (!(features & NETIF_F_SG)) {
5581 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5582 features &= ~NETIF_F_UFO;
5589 int __netdev_update_features(struct net_device *dev)
5591 netdev_features_t features;
5596 features = netdev_get_wanted_features(dev);
5598 if (dev->netdev_ops->ndo_fix_features)
5599 features = dev->netdev_ops->ndo_fix_features(dev, features);
5601 /* driver might be less strict about feature dependencies */
5602 features = netdev_fix_features(dev, features);
5604 if (dev->features == features)
5607 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5608 &dev->features, &features);
5610 if (dev->netdev_ops->ndo_set_features)
5611 err = dev->netdev_ops->ndo_set_features(dev, features);
5613 if (unlikely(err < 0)) {
5615 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5616 err, &features, &dev->features);
5621 dev->features = features;
5627 * netdev_update_features - recalculate device features
5628 * @dev: the device to check
5630 * Recalculate dev->features set and send notifications if it
5631 * has changed. Should be called after driver or hardware dependent
5632 * conditions might have changed that influence the features.
5634 void netdev_update_features(struct net_device *dev)
5636 if (__netdev_update_features(dev))
5637 netdev_features_change(dev);
5639 EXPORT_SYMBOL(netdev_update_features);
5642 * netdev_change_features - recalculate device features
5643 * @dev: the device to check
5645 * Recalculate dev->features set and send notifications even
5646 * if they have not changed. Should be called instead of
5647 * netdev_update_features() if also dev->vlan_features might
5648 * have changed to allow the changes to be propagated to stacked
5651 void netdev_change_features(struct net_device *dev)
5653 __netdev_update_features(dev);
5654 netdev_features_change(dev);
5656 EXPORT_SYMBOL(netdev_change_features);
5659 * netif_stacked_transfer_operstate - transfer operstate
5660 * @rootdev: the root or lower level device to transfer state from
5661 * @dev: the device to transfer operstate to
5663 * Transfer operational state from root to device. This is normally
5664 * called when a stacking relationship exists between the root
5665 * device and the device(a leaf device).
5667 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5668 struct net_device *dev)
5670 if (rootdev->operstate == IF_OPER_DORMANT)
5671 netif_dormant_on(dev);
5673 netif_dormant_off(dev);
5675 if (netif_carrier_ok(rootdev)) {
5676 if (!netif_carrier_ok(dev))
5677 netif_carrier_on(dev);
5679 if (netif_carrier_ok(dev))
5680 netif_carrier_off(dev);
5683 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5686 static int netif_alloc_rx_queues(struct net_device *dev)
5688 unsigned int i, count = dev->num_rx_queues;
5689 struct netdev_rx_queue *rx;
5693 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5699 for (i = 0; i < count; i++)
5705 static void netdev_init_one_queue(struct net_device *dev,
5706 struct netdev_queue *queue, void *_unused)
5708 /* Initialize queue lock */
5709 spin_lock_init(&queue->_xmit_lock);
5710 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5711 queue->xmit_lock_owner = -1;
5712 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5715 dql_init(&queue->dql, HZ);
5719 static void netif_free_tx_queues(struct net_device *dev)
5721 if (is_vmalloc_addr(dev->_tx))
5727 static int netif_alloc_netdev_queues(struct net_device *dev)
5729 unsigned int count = dev->num_tx_queues;
5730 struct netdev_queue *tx;
5731 size_t sz = count * sizeof(*tx);
5733 BUG_ON(count < 1 || count > 0xffff);
5735 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5743 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5744 spin_lock_init(&dev->tx_global_lock);
5750 * register_netdevice - register a network device
5751 * @dev: device to register
5753 * Take a completed network device structure and add it to the kernel
5754 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5755 * chain. 0 is returned on success. A negative errno code is returned
5756 * on a failure to set up the device, or if the name is a duplicate.
5758 * Callers must hold the rtnl semaphore. You may want
5759 * register_netdev() instead of this.
5762 * The locking appears insufficient to guarantee two parallel registers
5763 * will not get the same name.
5766 int register_netdevice(struct net_device *dev)
5769 struct net *net = dev_net(dev);
5771 BUG_ON(dev_boot_phase);
5776 /* When net_device's are persistent, this will be fatal. */
5777 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5780 spin_lock_init(&dev->addr_list_lock);
5781 netdev_set_addr_lockdep_class(dev);
5785 ret = dev_get_valid_name(net, dev, dev->name);
5789 /* Init, if this function is available */
5790 if (dev->netdev_ops->ndo_init) {
5791 ret = dev->netdev_ops->ndo_init(dev);
5799 if (((dev->hw_features | dev->features) &
5800 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5801 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5802 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5803 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5810 dev->ifindex = dev_new_index(net);
5811 else if (__dev_get_by_index(net, dev->ifindex))
5814 if (dev->iflink == -1)
5815 dev->iflink = dev->ifindex;
5817 /* Transfer changeable features to wanted_features and enable
5818 * software offloads (GSO and GRO).
5820 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5821 dev->features |= NETIF_F_SOFT_FEATURES;
5822 dev->wanted_features = dev->features & dev->hw_features;
5824 /* Turn on no cache copy if HW is doing checksum */
5825 if (!(dev->flags & IFF_LOOPBACK)) {
5826 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5827 if (dev->features & NETIF_F_ALL_CSUM) {
5828 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5829 dev->features |= NETIF_F_NOCACHE_COPY;
5833 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5835 dev->vlan_features |= NETIF_F_HIGHDMA;
5837 /* Make NETIF_F_SG inheritable to tunnel devices.
5839 dev->hw_enc_features |= NETIF_F_SG;
5841 /* Make NETIF_F_SG inheritable to MPLS.
5843 dev->mpls_features |= NETIF_F_SG;
5845 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5846 ret = notifier_to_errno(ret);
5850 ret = netdev_register_kobject(dev);
5853 dev->reg_state = NETREG_REGISTERED;
5855 __netdev_update_features(dev);
5858 * Default initial state at registry is that the
5859 * device is present.
5862 set_bit(__LINK_STATE_PRESENT, &dev->state);
5864 linkwatch_init_dev(dev);
5866 dev_init_scheduler(dev);
5868 list_netdevice(dev);
5869 add_device_randomness(dev->dev_addr, dev->addr_len);
5871 /* If the device has permanent device address, driver should
5872 * set dev_addr and also addr_assign_type should be set to
5873 * NET_ADDR_PERM (default value).
5875 if (dev->addr_assign_type == NET_ADDR_PERM)
5876 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5878 /* Notify protocols, that a new device appeared. */
5879 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5880 ret = notifier_to_errno(ret);
5882 rollback_registered(dev);
5883 dev->reg_state = NETREG_UNREGISTERED;
5886 * Prevent userspace races by waiting until the network
5887 * device is fully setup before sending notifications.
5889 if (!dev->rtnl_link_ops ||
5890 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5891 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5897 if (dev->netdev_ops->ndo_uninit)
5898 dev->netdev_ops->ndo_uninit(dev);
5901 EXPORT_SYMBOL(register_netdevice);
5904 * init_dummy_netdev - init a dummy network device for NAPI
5905 * @dev: device to init
5907 * This takes a network device structure and initialize the minimum
5908 * amount of fields so it can be used to schedule NAPI polls without
5909 * registering a full blown interface. This is to be used by drivers
5910 * that need to tie several hardware interfaces to a single NAPI
5911 * poll scheduler due to HW limitations.
5913 int init_dummy_netdev(struct net_device *dev)
5915 /* Clear everything. Note we don't initialize spinlocks
5916 * are they aren't supposed to be taken by any of the
5917 * NAPI code and this dummy netdev is supposed to be
5918 * only ever used for NAPI polls
5920 memset(dev, 0, sizeof(struct net_device));
5922 /* make sure we BUG if trying to hit standard
5923 * register/unregister code path
5925 dev->reg_state = NETREG_DUMMY;
5927 /* NAPI wants this */
5928 INIT_LIST_HEAD(&dev->napi_list);
5930 /* a dummy interface is started by default */
5931 set_bit(__LINK_STATE_PRESENT, &dev->state);
5932 set_bit(__LINK_STATE_START, &dev->state);
5934 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5935 * because users of this 'device' dont need to change
5941 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5945 * register_netdev - register a network device
5946 * @dev: device to register
5948 * Take a completed network device structure and add it to the kernel
5949 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5950 * chain. 0 is returned on success. A negative errno code is returned
5951 * on a failure to set up the device, or if the name is a duplicate.
5953 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5954 * and expands the device name if you passed a format string to
5957 int register_netdev(struct net_device *dev)
5962 err = register_netdevice(dev);
5966 EXPORT_SYMBOL(register_netdev);
5968 int netdev_refcnt_read(const struct net_device *dev)
5972 for_each_possible_cpu(i)
5973 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5976 EXPORT_SYMBOL(netdev_refcnt_read);
5979 * netdev_wait_allrefs - wait until all references are gone.
5980 * @dev: target net_device
5982 * This is called when unregistering network devices.
5984 * Any protocol or device that holds a reference should register
5985 * for netdevice notification, and cleanup and put back the
5986 * reference if they receive an UNREGISTER event.
5987 * We can get stuck here if buggy protocols don't correctly
5990 static void netdev_wait_allrefs(struct net_device *dev)
5992 unsigned long rebroadcast_time, warning_time;
5995 linkwatch_forget_dev(dev);
5997 rebroadcast_time = warning_time = jiffies;
5998 refcnt = netdev_refcnt_read(dev);
6000 while (refcnt != 0) {
6001 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6004 /* Rebroadcast unregister notification */
6005 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6011 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6012 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6014 /* We must not have linkwatch events
6015 * pending on unregister. If this
6016 * happens, we simply run the queue
6017 * unscheduled, resulting in a noop
6020 linkwatch_run_queue();
6025 rebroadcast_time = jiffies;
6030 refcnt = netdev_refcnt_read(dev);
6032 if (time_after(jiffies, warning_time + 10 * HZ)) {
6033 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6035 warning_time = jiffies;
6044 * register_netdevice(x1);
6045 * register_netdevice(x2);
6047 * unregister_netdevice(y1);
6048 * unregister_netdevice(y2);
6054 * We are invoked by rtnl_unlock().
6055 * This allows us to deal with problems:
6056 * 1) We can delete sysfs objects which invoke hotplug
6057 * without deadlocking with linkwatch via keventd.
6058 * 2) Since we run with the RTNL semaphore not held, we can sleep
6059 * safely in order to wait for the netdev refcnt to drop to zero.
6061 * We must not return until all unregister events added during
6062 * the interval the lock was held have been completed.
6064 void netdev_run_todo(void)
6066 struct list_head list;
6068 /* Snapshot list, allow later requests */
6069 list_replace_init(&net_todo_list, &list);
6074 /* Wait for rcu callbacks to finish before next phase */
6075 if (!list_empty(&list))
6078 while (!list_empty(&list)) {
6079 struct net_device *dev
6080 = list_first_entry(&list, struct net_device, todo_list);
6081 list_del(&dev->todo_list);
6084 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6087 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6088 pr_err("network todo '%s' but state %d\n",
6089 dev->name, dev->reg_state);
6094 dev->reg_state = NETREG_UNREGISTERED;
6096 on_each_cpu(flush_backlog, dev, 1);
6098 netdev_wait_allrefs(dev);
6101 BUG_ON(netdev_refcnt_read(dev));
6102 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6103 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6104 WARN_ON(dev->dn_ptr);
6106 if (dev->destructor)
6107 dev->destructor(dev);
6109 /* Report a network device has been unregistered */
6111 dev_net(dev)->dev_unreg_count--;
6113 wake_up(&netdev_unregistering_wq);
6115 /* Free network device */
6116 kobject_put(&dev->dev.kobj);
6120 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6121 * fields in the same order, with only the type differing.
6123 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6124 const struct net_device_stats *netdev_stats)
6126 #if BITS_PER_LONG == 64
6127 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6128 memcpy(stats64, netdev_stats, sizeof(*stats64));
6130 size_t i, n = sizeof(*stats64) / sizeof(u64);
6131 const unsigned long *src = (const unsigned long *)netdev_stats;
6132 u64 *dst = (u64 *)stats64;
6134 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6135 sizeof(*stats64) / sizeof(u64));
6136 for (i = 0; i < n; i++)
6140 EXPORT_SYMBOL(netdev_stats_to_stats64);
6143 * dev_get_stats - get network device statistics
6144 * @dev: device to get statistics from
6145 * @storage: place to store stats
6147 * Get network statistics from device. Return @storage.
6148 * The device driver may provide its own method by setting
6149 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6150 * otherwise the internal statistics structure is used.
6152 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6153 struct rtnl_link_stats64 *storage)
6155 const struct net_device_ops *ops = dev->netdev_ops;
6157 if (ops->ndo_get_stats64) {
6158 memset(storage, 0, sizeof(*storage));
6159 ops->ndo_get_stats64(dev, storage);
6160 } else if (ops->ndo_get_stats) {
6161 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6163 netdev_stats_to_stats64(storage, &dev->stats);
6165 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6168 EXPORT_SYMBOL(dev_get_stats);
6170 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6172 struct netdev_queue *queue = dev_ingress_queue(dev);
6174 #ifdef CONFIG_NET_CLS_ACT
6177 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6180 netdev_init_one_queue(dev, queue, NULL);
6181 queue->qdisc = &noop_qdisc;
6182 queue->qdisc_sleeping = &noop_qdisc;
6183 rcu_assign_pointer(dev->ingress_queue, queue);
6188 static const struct ethtool_ops default_ethtool_ops;
6190 void netdev_set_default_ethtool_ops(struct net_device *dev,
6191 const struct ethtool_ops *ops)
6193 if (dev->ethtool_ops == &default_ethtool_ops)
6194 dev->ethtool_ops = ops;
6196 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6199 * alloc_netdev_mqs - allocate network device
6200 * @sizeof_priv: size of private data to allocate space for
6201 * @name: device name format string
6202 * @setup: callback to initialize device
6203 * @txqs: the number of TX subqueues to allocate
6204 * @rxqs: the number of RX subqueues to allocate
6206 * Allocates a struct net_device with private data area for driver use
6207 * and performs basic initialization. Also allocates subquue structs
6208 * for each queue on the device.
6210 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6211 void (*setup)(struct net_device *),
6212 unsigned int txqs, unsigned int rxqs)
6214 struct net_device *dev;
6216 struct net_device *p;
6218 BUG_ON(strlen(name) >= sizeof(dev->name));
6221 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6227 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6232 alloc_size = sizeof(struct net_device);
6234 /* ensure 32-byte alignment of private area */
6235 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6236 alloc_size += sizeof_priv;
6238 /* ensure 32-byte alignment of whole construct */
6239 alloc_size += NETDEV_ALIGN - 1;
6241 p = kzalloc(alloc_size, GFP_KERNEL);
6245 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6246 dev->padded = (char *)dev - (char *)p;
6248 dev->pcpu_refcnt = alloc_percpu(int);
6249 if (!dev->pcpu_refcnt)
6252 if (dev_addr_init(dev))
6258 dev_net_set(dev, &init_net);
6260 dev->gso_max_size = GSO_MAX_SIZE;
6261 dev->gso_max_segs = GSO_MAX_SEGS;
6263 INIT_LIST_HEAD(&dev->napi_list);
6264 INIT_LIST_HEAD(&dev->unreg_list);
6265 INIT_LIST_HEAD(&dev->close_list);
6266 INIT_LIST_HEAD(&dev->link_watch_list);
6267 INIT_LIST_HEAD(&dev->adj_list.upper);
6268 INIT_LIST_HEAD(&dev->adj_list.lower);
6269 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6270 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6271 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6274 dev->num_tx_queues = txqs;
6275 dev->real_num_tx_queues = txqs;
6276 if (netif_alloc_netdev_queues(dev))
6280 dev->num_rx_queues = rxqs;
6281 dev->real_num_rx_queues = rxqs;
6282 if (netif_alloc_rx_queues(dev))
6286 strcpy(dev->name, name);
6287 dev->group = INIT_NETDEV_GROUP;
6288 if (!dev->ethtool_ops)
6289 dev->ethtool_ops = &default_ethtool_ops;
6297 free_percpu(dev->pcpu_refcnt);
6298 netif_free_tx_queues(dev);
6307 EXPORT_SYMBOL(alloc_netdev_mqs);
6310 * free_netdev - free network device
6313 * This function does the last stage of destroying an allocated device
6314 * interface. The reference to the device object is released.
6315 * If this is the last reference then it will be freed.
6317 void free_netdev(struct net_device *dev)
6319 struct napi_struct *p, *n;
6321 release_net(dev_net(dev));
6323 netif_free_tx_queues(dev);
6328 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6330 /* Flush device addresses */
6331 dev_addr_flush(dev);
6333 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6336 free_percpu(dev->pcpu_refcnt);
6337 dev->pcpu_refcnt = NULL;
6339 /* Compatibility with error handling in drivers */
6340 if (dev->reg_state == NETREG_UNINITIALIZED) {
6341 kfree((char *)dev - dev->padded);
6345 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6346 dev->reg_state = NETREG_RELEASED;
6348 /* will free via device release */
6349 put_device(&dev->dev);
6351 EXPORT_SYMBOL(free_netdev);
6354 * synchronize_net - Synchronize with packet receive processing
6356 * Wait for packets currently being received to be done.
6357 * Does not block later packets from starting.
6359 void synchronize_net(void)
6362 if (rtnl_is_locked())
6363 synchronize_rcu_expedited();
6367 EXPORT_SYMBOL(synchronize_net);
6370 * unregister_netdevice_queue - remove device from the kernel
6374 * This function shuts down a device interface and removes it
6375 * from the kernel tables.
6376 * If head not NULL, device is queued to be unregistered later.
6378 * Callers must hold the rtnl semaphore. You may want
6379 * unregister_netdev() instead of this.
6382 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6387 list_move_tail(&dev->unreg_list, head);
6389 rollback_registered(dev);
6390 /* Finish processing unregister after unlock */
6394 EXPORT_SYMBOL(unregister_netdevice_queue);
6397 * unregister_netdevice_many - unregister many devices
6398 * @head: list of devices
6400 void unregister_netdevice_many(struct list_head *head)
6402 struct net_device *dev;
6404 if (!list_empty(head)) {
6405 rollback_registered_many(head);
6406 list_for_each_entry(dev, head, unreg_list)
6410 EXPORT_SYMBOL(unregister_netdevice_many);
6413 * unregister_netdev - remove device from the kernel
6416 * This function shuts down a device interface and removes it
6417 * from the kernel tables.
6419 * This is just a wrapper for unregister_netdevice that takes
6420 * the rtnl semaphore. In general you want to use this and not
6421 * unregister_netdevice.
6423 void unregister_netdev(struct net_device *dev)
6426 unregister_netdevice(dev);
6429 EXPORT_SYMBOL(unregister_netdev);
6432 * dev_change_net_namespace - move device to different nethost namespace
6434 * @net: network namespace
6435 * @pat: If not NULL name pattern to try if the current device name
6436 * is already taken in the destination network namespace.
6438 * This function shuts down a device interface and moves it
6439 * to a new network namespace. On success 0 is returned, on
6440 * a failure a netagive errno code is returned.
6442 * Callers must hold the rtnl semaphore.
6445 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6451 /* Don't allow namespace local devices to be moved. */
6453 if (dev->features & NETIF_F_NETNS_LOCAL)
6456 /* Ensure the device has been registrered */
6457 if (dev->reg_state != NETREG_REGISTERED)
6460 /* Get out if there is nothing todo */
6462 if (net_eq(dev_net(dev), net))
6465 /* Pick the destination device name, and ensure
6466 * we can use it in the destination network namespace.
6469 if (__dev_get_by_name(net, dev->name)) {
6470 /* We get here if we can't use the current device name */
6473 if (dev_get_valid_name(net, dev, pat) < 0)
6478 * And now a mini version of register_netdevice unregister_netdevice.
6481 /* If device is running close it first. */
6484 /* And unlink it from device chain */
6486 unlist_netdevice(dev);
6490 /* Shutdown queueing discipline. */
6493 /* Notify protocols, that we are about to destroy
6494 this device. They should clean all the things.
6496 Note that dev->reg_state stays at NETREG_REGISTERED.
6497 This is wanted because this way 8021q and macvlan know
6498 the device is just moving and can keep their slaves up.
6500 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6502 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6503 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6506 * Flush the unicast and multicast chains
6511 /* Send a netdev-removed uevent to the old namespace */
6512 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6514 /* Actually switch the network namespace */
6515 dev_net_set(dev, net);
6517 /* If there is an ifindex conflict assign a new one */
6518 if (__dev_get_by_index(net, dev->ifindex)) {
6519 int iflink = (dev->iflink == dev->ifindex);
6520 dev->ifindex = dev_new_index(net);
6522 dev->iflink = dev->ifindex;
6525 /* Send a netdev-add uevent to the new namespace */
6526 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6528 /* Fixup kobjects */
6529 err = device_rename(&dev->dev, dev->name);
6532 /* Add the device back in the hashes */
6533 list_netdevice(dev);
6535 /* Notify protocols, that a new device appeared. */
6536 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6539 * Prevent userspace races by waiting until the network
6540 * device is fully setup before sending notifications.
6542 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6549 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6551 static int dev_cpu_callback(struct notifier_block *nfb,
6552 unsigned long action,
6555 struct sk_buff **list_skb;
6556 struct sk_buff *skb;
6557 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6558 struct softnet_data *sd, *oldsd;
6560 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6563 local_irq_disable();
6564 cpu = smp_processor_id();
6565 sd = &per_cpu(softnet_data, cpu);
6566 oldsd = &per_cpu(softnet_data, oldcpu);
6568 /* Find end of our completion_queue. */
6569 list_skb = &sd->completion_queue;
6571 list_skb = &(*list_skb)->next;
6572 /* Append completion queue from offline CPU. */
6573 *list_skb = oldsd->completion_queue;
6574 oldsd->completion_queue = NULL;
6576 /* Append output queue from offline CPU. */
6577 if (oldsd->output_queue) {
6578 *sd->output_queue_tailp = oldsd->output_queue;
6579 sd->output_queue_tailp = oldsd->output_queue_tailp;
6580 oldsd->output_queue = NULL;
6581 oldsd->output_queue_tailp = &oldsd->output_queue;
6583 /* Append NAPI poll list from offline CPU. */
6584 if (!list_empty(&oldsd->poll_list)) {
6585 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6586 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6589 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6592 /* Process offline CPU's input_pkt_queue */
6593 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6595 input_queue_head_incr(oldsd);
6597 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6599 input_queue_head_incr(oldsd);
6607 * netdev_increment_features - increment feature set by one
6608 * @all: current feature set
6609 * @one: new feature set
6610 * @mask: mask feature set
6612 * Computes a new feature set after adding a device with feature set
6613 * @one to the master device with current feature set @all. Will not
6614 * enable anything that is off in @mask. Returns the new feature set.
6616 netdev_features_t netdev_increment_features(netdev_features_t all,
6617 netdev_features_t one, netdev_features_t mask)
6619 if (mask & NETIF_F_GEN_CSUM)
6620 mask |= NETIF_F_ALL_CSUM;
6621 mask |= NETIF_F_VLAN_CHALLENGED;
6623 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6624 all &= one | ~NETIF_F_ALL_FOR_ALL;
6626 /* If one device supports hw checksumming, set for all. */
6627 if (all & NETIF_F_GEN_CSUM)
6628 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6632 EXPORT_SYMBOL(netdev_increment_features);
6634 static struct hlist_head * __net_init netdev_create_hash(void)
6637 struct hlist_head *hash;
6639 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6641 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6642 INIT_HLIST_HEAD(&hash[i]);
6647 /* Initialize per network namespace state */
6648 static int __net_init netdev_init(struct net *net)
6650 if (net != &init_net)
6651 INIT_LIST_HEAD(&net->dev_base_head);
6653 net->dev_name_head = netdev_create_hash();
6654 if (net->dev_name_head == NULL)
6657 net->dev_index_head = netdev_create_hash();
6658 if (net->dev_index_head == NULL)
6664 kfree(net->dev_name_head);
6670 * netdev_drivername - network driver for the device
6671 * @dev: network device
6673 * Determine network driver for device.
6675 const char *netdev_drivername(const struct net_device *dev)
6677 const struct device_driver *driver;
6678 const struct device *parent;
6679 const char *empty = "";
6681 parent = dev->dev.parent;
6685 driver = parent->driver;
6686 if (driver && driver->name)
6687 return driver->name;
6691 static int __netdev_printk(const char *level, const struct net_device *dev,
6692 struct va_format *vaf)
6696 if (dev && dev->dev.parent) {
6697 r = dev_printk_emit(level[1] - '0',
6700 dev_driver_string(dev->dev.parent),
6701 dev_name(dev->dev.parent),
6702 netdev_name(dev), vaf);
6704 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6706 r = printk("%s(NULL net_device): %pV", level, vaf);
6712 int netdev_printk(const char *level, const struct net_device *dev,
6713 const char *format, ...)
6715 struct va_format vaf;
6719 va_start(args, format);
6724 r = __netdev_printk(level, dev, &vaf);
6730 EXPORT_SYMBOL(netdev_printk);
6732 #define define_netdev_printk_level(func, level) \
6733 int func(const struct net_device *dev, const char *fmt, ...) \
6736 struct va_format vaf; \
6739 va_start(args, fmt); \
6744 r = __netdev_printk(level, dev, &vaf); \
6750 EXPORT_SYMBOL(func);
6752 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6753 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6754 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6755 define_netdev_printk_level(netdev_err, KERN_ERR);
6756 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6757 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6758 define_netdev_printk_level(netdev_info, KERN_INFO);
6760 static void __net_exit netdev_exit(struct net *net)
6762 kfree(net->dev_name_head);
6763 kfree(net->dev_index_head);
6766 static struct pernet_operations __net_initdata netdev_net_ops = {
6767 .init = netdev_init,
6768 .exit = netdev_exit,
6771 static void __net_exit default_device_exit(struct net *net)
6773 struct net_device *dev, *aux;
6775 * Push all migratable network devices back to the
6776 * initial network namespace
6779 for_each_netdev_safe(net, dev, aux) {
6781 char fb_name[IFNAMSIZ];
6783 /* Ignore unmoveable devices (i.e. loopback) */
6784 if (dev->features & NETIF_F_NETNS_LOCAL)
6787 /* Leave virtual devices for the generic cleanup */
6788 if (dev->rtnl_link_ops)
6791 /* Push remaining network devices to init_net */
6792 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6793 err = dev_change_net_namespace(dev, &init_net, fb_name);
6795 pr_emerg("%s: failed to move %s to init_net: %d\n",
6796 __func__, dev->name, err);
6803 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6805 /* Return with the rtnl_lock held when there are no network
6806 * devices unregistering in any network namespace in net_list.
6813 prepare_to_wait(&netdev_unregistering_wq, &wait,
6814 TASK_UNINTERRUPTIBLE);
6815 unregistering = false;
6817 list_for_each_entry(net, net_list, exit_list) {
6818 if (net->dev_unreg_count > 0) {
6819 unregistering = true;
6828 finish_wait(&netdev_unregistering_wq, &wait);
6831 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6833 /* At exit all network devices most be removed from a network
6834 * namespace. Do this in the reverse order of registration.
6835 * Do this across as many network namespaces as possible to
6836 * improve batching efficiency.
6838 struct net_device *dev;
6840 LIST_HEAD(dev_kill_list);
6842 /* To prevent network device cleanup code from dereferencing
6843 * loopback devices or network devices that have been freed
6844 * wait here for all pending unregistrations to complete,
6845 * before unregistring the loopback device and allowing the
6846 * network namespace be freed.
6848 * The netdev todo list containing all network devices
6849 * unregistrations that happen in default_device_exit_batch
6850 * will run in the rtnl_unlock() at the end of
6851 * default_device_exit_batch.
6853 rtnl_lock_unregistering(net_list);
6854 list_for_each_entry(net, net_list, exit_list) {
6855 for_each_netdev_reverse(net, dev) {
6856 if (dev->rtnl_link_ops)
6857 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6859 unregister_netdevice_queue(dev, &dev_kill_list);
6862 unregister_netdevice_many(&dev_kill_list);
6863 list_del(&dev_kill_list);
6867 static struct pernet_operations __net_initdata default_device_ops = {
6868 .exit = default_device_exit,
6869 .exit_batch = default_device_exit_batch,
6873 * Initialize the DEV module. At boot time this walks the device list and
6874 * unhooks any devices that fail to initialise (normally hardware not
6875 * present) and leaves us with a valid list of present and active devices.
6880 * This is called single threaded during boot, so no need
6881 * to take the rtnl semaphore.
6883 static int __init net_dev_init(void)
6885 int i, rc = -ENOMEM;
6887 BUG_ON(!dev_boot_phase);
6889 if (dev_proc_init())
6892 if (netdev_kobject_init())
6895 INIT_LIST_HEAD(&ptype_all);
6896 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6897 INIT_LIST_HEAD(&ptype_base[i]);
6899 INIT_LIST_HEAD(&offload_base);
6901 if (register_pernet_subsys(&netdev_net_ops))
6905 * Initialise the packet receive queues.
6908 for_each_possible_cpu(i) {
6909 struct softnet_data *sd = &per_cpu(softnet_data, i);
6911 memset(sd, 0, sizeof(*sd));
6912 skb_queue_head_init(&sd->input_pkt_queue);
6913 skb_queue_head_init(&sd->process_queue);
6914 sd->completion_queue = NULL;
6915 INIT_LIST_HEAD(&sd->poll_list);
6916 sd->output_queue = NULL;
6917 sd->output_queue_tailp = &sd->output_queue;
6919 sd->csd.func = rps_trigger_softirq;
6925 sd->backlog.poll = process_backlog;
6926 sd->backlog.weight = weight_p;
6927 sd->backlog.gro_list = NULL;
6928 sd->backlog.gro_count = 0;
6930 #ifdef CONFIG_NET_FLOW_LIMIT
6931 sd->flow_limit = NULL;
6937 /* The loopback device is special if any other network devices
6938 * is present in a network namespace the loopback device must
6939 * be present. Since we now dynamically allocate and free the
6940 * loopback device ensure this invariant is maintained by
6941 * keeping the loopback device as the first device on the
6942 * list of network devices. Ensuring the loopback devices
6943 * is the first device that appears and the last network device
6946 if (register_pernet_device(&loopback_net_ops))
6949 if (register_pernet_device(&default_device_ops))
6952 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6953 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6955 hotcpu_notifier(dev_cpu_callback, 0);
6962 subsys_initcall(net_dev_init);