2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
140 #include "net-sysfs.h"
142 /* Instead of increasing this, you should create a hash table. */
143 #define MAX_GRO_SKBS 8
145 /* This should be increased if a protocol with a bigger head is added. */
146 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 static DEFINE_SPINLOCK(ptype_lock);
149 static DEFINE_SPINLOCK(offload_lock);
150 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
151 struct list_head ptype_all __read_mostly; /* Taps */
152 static struct list_head offload_base __read_mostly;
154 static int netif_rx_internal(struct sk_buff *skb);
155 static int call_netdevice_notifiers_info(unsigned long val,
156 struct net_device *dev,
157 struct netdev_notifier_info *info);
160 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
163 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base_head list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
178 DEFINE_RWLOCK(dev_base_lock);
179 EXPORT_SYMBOL(dev_base_lock);
181 /* protects napi_hash addition/deletion and napi_gen_id */
182 static DEFINE_SPINLOCK(napi_hash_lock);
184 static unsigned int napi_gen_id;
185 static DEFINE_HASHTABLE(napi_hash, 8);
187 static seqcount_t devnet_rename_seq;
189 static inline void dev_base_seq_inc(struct net *net)
191 while (++net->dev_base_seq == 0);
194 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
196 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
201 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
206 static inline void rps_lock(struct softnet_data *sd)
209 spin_lock(&sd->input_pkt_queue.lock);
213 static inline void rps_unlock(struct softnet_data *sd)
216 spin_unlock(&sd->input_pkt_queue.lock);
220 /* Device list insertion */
221 static void list_netdevice(struct net_device *dev)
223 struct net *net = dev_net(dev);
227 write_lock_bh(&dev_base_lock);
228 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
230 hlist_add_head_rcu(&dev->index_hlist,
231 dev_index_hash(net, dev->ifindex));
232 write_unlock_bh(&dev_base_lock);
234 dev_base_seq_inc(net);
237 /* Device list removal
238 * caller must respect a RCU grace period before freeing/reusing dev
240 static void unlist_netdevice(struct net_device *dev)
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del_rcu(&dev->dev_list);
247 hlist_del_rcu(&dev->name_hlist);
248 hlist_del_rcu(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(dev_net(dev));
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
287 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
288 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
290 static const char *const netdev_lock_name[] =
291 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
292 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
293 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
294 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
295 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
296 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
297 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
298 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
299 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
300 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
301 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
302 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
303 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
304 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
305 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
307 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
308 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
314 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
315 if (netdev_lock_type[i] == dev_type)
317 /* the last key is used by default */
318 return ARRAY_SIZE(netdev_lock_type) - 1;
321 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
322 unsigned short dev_type)
326 i = netdev_lock_pos(dev_type);
327 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
328 netdev_lock_name[i]);
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
335 i = netdev_lock_pos(dev->type);
336 lockdep_set_class_and_name(&dev->addr_list_lock,
337 &netdev_addr_lock_key[i],
338 netdev_lock_name[i]);
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
345 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
350 /*******************************************************************************
352 Protocol management and registration routines
354 *******************************************************************************/
357 * Add a protocol ID to the list. Now that the input handler is
358 * smarter we can dispense with all the messy stuff that used to be
361 * BEWARE!!! Protocol handlers, mangling input packets,
362 * MUST BE last in hash buckets and checking protocol handlers
363 * MUST start from promiscuous ptype_all chain in net_bh.
364 * It is true now, do not change it.
365 * Explanation follows: if protocol handler, mangling packet, will
366 * be the first on list, it is not able to sense, that packet
367 * is cloned and should be copied-on-write, so that it will
368 * change it and subsequent readers will get broken packet.
372 static inline struct list_head *ptype_head(const struct packet_type *pt)
374 if (pt->type == htons(ETH_P_ALL))
375 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
377 return pt->dev ? &pt->dev->ptype_specific :
378 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
382 * dev_add_pack - add packet handler
383 * @pt: packet type declaration
385 * Add a protocol handler to the networking stack. The passed &packet_type
386 * is linked into kernel lists and may not be freed until it has been
387 * removed from the kernel lists.
389 * This call does not sleep therefore it can not
390 * guarantee all CPU's that are in middle of receiving packets
391 * will see the new packet type (until the next received packet).
394 void dev_add_pack(struct packet_type *pt)
396 struct list_head *head = ptype_head(pt);
398 spin_lock(&ptype_lock);
399 list_add_rcu(&pt->list, head);
400 spin_unlock(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head = ptype_head(pt);
420 struct packet_type *pt1;
422 spin_lock(&ptype_lock);
424 list_for_each_entry(pt1, head, list) {
426 list_del_rcu(&pt->list);
431 pr_warn("dev_remove_pack: %p not found\n", pt);
433 spin_unlock(&ptype_lock);
435 EXPORT_SYMBOL(__dev_remove_pack);
438 * dev_remove_pack - remove packet handler
439 * @pt: packet type declaration
441 * Remove a protocol handler that was previously added to the kernel
442 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
443 * from the kernel lists and can be freed or reused once this function
446 * This call sleeps to guarantee that no CPU is looking at the packet
449 void dev_remove_pack(struct packet_type *pt)
451 __dev_remove_pack(pt);
455 EXPORT_SYMBOL(dev_remove_pack);
459 * dev_add_offload - register offload handlers
460 * @po: protocol offload declaration
462 * Add protocol offload handlers to the networking stack. The passed
463 * &proto_offload is linked into kernel lists and may not be freed until
464 * it has been removed from the kernel lists.
466 * This call does not sleep therefore it can not
467 * guarantee all CPU's that are in middle of receiving packets
468 * will see the new offload handlers (until the next received packet).
470 void dev_add_offload(struct packet_offload *po)
472 struct packet_offload *elem;
474 spin_lock(&offload_lock);
475 list_for_each_entry(elem, &offload_base, list) {
476 if (po->priority < elem->priority)
479 list_add_rcu(&po->list, elem->list.prev);
480 spin_unlock(&offload_lock);
482 EXPORT_SYMBOL(dev_add_offload);
485 * __dev_remove_offload - remove offload handler
486 * @po: packet offload declaration
488 * Remove a protocol offload handler that was previously added to the
489 * kernel offload handlers by dev_add_offload(). The passed &offload_type
490 * is removed from the kernel lists and can be freed or reused once this
493 * The packet type might still be in use by receivers
494 * and must not be freed until after all the CPU's have gone
495 * through a quiescent state.
497 static void __dev_remove_offload(struct packet_offload *po)
499 struct list_head *head = &offload_base;
500 struct packet_offload *po1;
502 spin_lock(&offload_lock);
504 list_for_each_entry(po1, head, list) {
506 list_del_rcu(&po->list);
511 pr_warn("dev_remove_offload: %p not found\n", po);
513 spin_unlock(&offload_lock);
517 * dev_remove_offload - remove packet offload handler
518 * @po: packet offload declaration
520 * Remove a packet offload handler that was previously added to the kernel
521 * offload handlers by dev_add_offload(). The passed &offload_type is
522 * removed from the kernel lists and can be freed or reused once this
525 * This call sleeps to guarantee that no CPU is looking at the packet
528 void dev_remove_offload(struct packet_offload *po)
530 __dev_remove_offload(po);
534 EXPORT_SYMBOL(dev_remove_offload);
536 /******************************************************************************
538 Device Boot-time Settings Routines
540 *******************************************************************************/
542 /* Boot time configuration table */
543 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
546 * netdev_boot_setup_add - add new setup entry
547 * @name: name of the device
548 * @map: configured settings for the device
550 * Adds new setup entry to the dev_boot_setup list. The function
551 * returns 0 on error and 1 on success. This is a generic routine to
554 static int netdev_boot_setup_add(char *name, struct ifmap *map)
556 struct netdev_boot_setup *s;
560 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
561 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
562 memset(s[i].name, 0, sizeof(s[i].name));
563 strlcpy(s[i].name, name, IFNAMSIZ);
564 memcpy(&s[i].map, map, sizeof(s[i].map));
569 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
573 * netdev_boot_setup_check - check boot time settings
574 * @dev: the netdevice
576 * Check boot time settings for the device.
577 * The found settings are set for the device to be used
578 * later in the device probing.
579 * Returns 0 if no settings found, 1 if they are.
581 int netdev_boot_setup_check(struct net_device *dev)
583 struct netdev_boot_setup *s = dev_boot_setup;
586 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
587 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
588 !strcmp(dev->name, s[i].name)) {
589 dev->irq = s[i].map.irq;
590 dev->base_addr = s[i].map.base_addr;
591 dev->mem_start = s[i].map.mem_start;
592 dev->mem_end = s[i].map.mem_end;
598 EXPORT_SYMBOL(netdev_boot_setup_check);
602 * netdev_boot_base - get address from boot time settings
603 * @prefix: prefix for network device
604 * @unit: id for network device
606 * Check boot time settings for the base address of device.
607 * The found settings are set for the device to be used
608 * later in the device probing.
609 * Returns 0 if no settings found.
611 unsigned long netdev_boot_base(const char *prefix, int unit)
613 const struct netdev_boot_setup *s = dev_boot_setup;
617 sprintf(name, "%s%d", prefix, unit);
620 * If device already registered then return base of 1
621 * to indicate not to probe for this interface
623 if (__dev_get_by_name(&init_net, name))
626 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
627 if (!strcmp(name, s[i].name))
628 return s[i].map.base_addr;
633 * Saves at boot time configured settings for any netdevice.
635 int __init netdev_boot_setup(char *str)
640 str = get_options(str, ARRAY_SIZE(ints), ints);
645 memset(&map, 0, sizeof(map));
649 map.base_addr = ints[2];
651 map.mem_start = ints[3];
653 map.mem_end = ints[4];
655 /* Add new entry to the list */
656 return netdev_boot_setup_add(str, &map);
659 __setup("netdev=", netdev_boot_setup);
661 /*******************************************************************************
663 Device Interface Subroutines
665 *******************************************************************************/
668 * dev_get_iflink - get 'iflink' value of a interface
669 * @dev: targeted interface
671 * Indicates the ifindex the interface is linked to.
672 * Physical interfaces have the same 'ifindex' and 'iflink' values.
675 int dev_get_iflink(const struct net_device *dev)
677 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
678 return dev->netdev_ops->ndo_get_iflink(dev);
680 /* If dev->rtnl_link_ops is set, it's a virtual interface. */
681 if (dev->rtnl_link_ops)
686 EXPORT_SYMBOL(dev_get_iflink);
689 * __dev_get_by_name - find a device by its name
690 * @net: the applicable net namespace
691 * @name: name to find
693 * Find an interface by name. Must be called under RTNL semaphore
694 * or @dev_base_lock. If the name is found a pointer to the device
695 * is returned. If the name is not found then %NULL is returned. The
696 * reference counters are not incremented so the caller must be
697 * careful with locks.
700 struct net_device *__dev_get_by_name(struct net *net, const char *name)
702 struct net_device *dev;
703 struct hlist_head *head = dev_name_hash(net, name);
705 hlist_for_each_entry(dev, head, name_hlist)
706 if (!strncmp(dev->name, name, IFNAMSIZ))
711 EXPORT_SYMBOL(__dev_get_by_name);
714 * dev_get_by_name_rcu - find a device by its name
715 * @net: the applicable net namespace
716 * @name: name to find
718 * Find an interface by name.
719 * If the name is found a pointer to the device is returned.
720 * If the name is not found then %NULL is returned.
721 * The reference counters are not incremented so the caller must be
722 * careful with locks. The caller must hold RCU lock.
725 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
727 struct net_device *dev;
728 struct hlist_head *head = dev_name_hash(net, name);
730 hlist_for_each_entry_rcu(dev, head, name_hlist)
731 if (!strncmp(dev->name, name, IFNAMSIZ))
736 EXPORT_SYMBOL(dev_get_by_name_rcu);
739 * dev_get_by_name - find a device by its name
740 * @net: the applicable net namespace
741 * @name: name to find
743 * Find an interface by name. This can be called from any
744 * context and does its own locking. The returned handle has
745 * the usage count incremented and the caller must use dev_put() to
746 * release it when it is no longer needed. %NULL is returned if no
747 * matching device is found.
750 struct net_device *dev_get_by_name(struct net *net, const char *name)
752 struct net_device *dev;
755 dev = dev_get_by_name_rcu(net, name);
761 EXPORT_SYMBOL(dev_get_by_name);
764 * __dev_get_by_index - find a device by its ifindex
765 * @net: the applicable net namespace
766 * @ifindex: index of device
768 * Search for an interface by index. Returns %NULL if the device
769 * is not found or a pointer to the device. The device has not
770 * had its reference counter increased so the caller must be careful
771 * about locking. The caller must hold either the RTNL semaphore
775 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
777 struct net_device *dev;
778 struct hlist_head *head = dev_index_hash(net, ifindex);
780 hlist_for_each_entry(dev, head, index_hlist)
781 if (dev->ifindex == ifindex)
786 EXPORT_SYMBOL(__dev_get_by_index);
789 * dev_get_by_index_rcu - find a device by its ifindex
790 * @net: the applicable net namespace
791 * @ifindex: index of device
793 * Search for an interface by index. Returns %NULL if the device
794 * is not found or a pointer to the device. The device has not
795 * had its reference counter increased so the caller must be careful
796 * about locking. The caller must hold RCU lock.
799 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
801 struct net_device *dev;
802 struct hlist_head *head = dev_index_hash(net, ifindex);
804 hlist_for_each_entry_rcu(dev, head, index_hlist)
805 if (dev->ifindex == ifindex)
810 EXPORT_SYMBOL(dev_get_by_index_rcu);
814 * dev_get_by_index - find a device by its ifindex
815 * @net: the applicable net namespace
816 * @ifindex: index of device
818 * Search for an interface by index. Returns NULL if the device
819 * is not found or a pointer to the device. The device returned has
820 * had a reference added and the pointer is safe until the user calls
821 * dev_put to indicate they have finished with it.
824 struct net_device *dev_get_by_index(struct net *net, int ifindex)
826 struct net_device *dev;
829 dev = dev_get_by_index_rcu(net, ifindex);
835 EXPORT_SYMBOL(dev_get_by_index);
838 * netdev_get_name - get a netdevice name, knowing its ifindex.
839 * @net: network namespace
840 * @name: a pointer to the buffer where the name will be stored.
841 * @ifindex: the ifindex of the interface to get the name from.
843 * The use of raw_seqcount_begin() and cond_resched() before
844 * retrying is required as we want to give the writers a chance
845 * to complete when CONFIG_PREEMPT is not set.
847 int netdev_get_name(struct net *net, char *name, int ifindex)
849 struct net_device *dev;
853 seq = raw_seqcount_begin(&devnet_rename_seq);
855 dev = dev_get_by_index_rcu(net, ifindex);
861 strcpy(name, dev->name);
863 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
872 * dev_getbyhwaddr_rcu - find a device by its hardware address
873 * @net: the applicable net namespace
874 * @type: media type of device
875 * @ha: hardware address
877 * Search for an interface by MAC address. Returns NULL if the device
878 * is not found or a pointer to the device.
879 * The caller must hold RCU or RTNL.
880 * The returned device has not had its ref count increased
881 * and the caller must therefore be careful about locking
885 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
888 struct net_device *dev;
890 for_each_netdev_rcu(net, dev)
891 if (dev->type == type &&
892 !memcmp(dev->dev_addr, ha, dev->addr_len))
897 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
899 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
901 struct net_device *dev;
904 for_each_netdev(net, dev)
905 if (dev->type == type)
910 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
912 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
914 struct net_device *dev, *ret = NULL;
917 for_each_netdev_rcu(net, dev)
918 if (dev->type == type) {
926 EXPORT_SYMBOL(dev_getfirstbyhwtype);
929 * __dev_get_by_flags - find any device with given flags
930 * @net: the applicable net namespace
931 * @if_flags: IFF_* values
932 * @mask: bitmask of bits in if_flags to check
934 * Search for any interface with the given flags. Returns NULL if a device
935 * is not found or a pointer to the device. Must be called inside
936 * rtnl_lock(), and result refcount is unchanged.
939 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
942 struct net_device *dev, *ret;
947 for_each_netdev(net, dev) {
948 if (((dev->flags ^ if_flags) & mask) == 0) {
955 EXPORT_SYMBOL(__dev_get_by_flags);
958 * dev_valid_name - check if name is okay for network device
961 * Network device names need to be valid file names to
962 * to allow sysfs to work. We also disallow any kind of
965 bool dev_valid_name(const char *name)
969 if (strlen(name) >= IFNAMSIZ)
971 if (!strcmp(name, ".") || !strcmp(name, ".."))
975 if (*name == '/' || *name == ':' || isspace(*name))
981 EXPORT_SYMBOL(dev_valid_name);
984 * __dev_alloc_name - allocate a name for a device
985 * @net: network namespace to allocate the device name in
986 * @name: name format string
987 * @buf: scratch buffer and result name string
989 * Passed a format string - eg "lt%d" it will try and find a suitable
990 * id. It scans list of devices to build up a free map, then chooses
991 * the first empty slot. The caller must hold the dev_base or rtnl lock
992 * while allocating the name and adding the device in order to avoid
994 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
995 * Returns the number of the unit assigned or a negative errno code.
998 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1002 const int max_netdevices = 8*PAGE_SIZE;
1003 unsigned long *inuse;
1004 struct net_device *d;
1006 p = strnchr(name, IFNAMSIZ-1, '%');
1009 * Verify the string as this thing may have come from
1010 * the user. There must be either one "%d" and no other "%"
1013 if (p[1] != 'd' || strchr(p + 2, '%'))
1016 /* Use one page as a bit array of possible slots */
1017 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1021 for_each_netdev(net, d) {
1022 if (!sscanf(d->name, name, &i))
1024 if (i < 0 || i >= max_netdevices)
1027 /* avoid cases where sscanf is not exact inverse of printf */
1028 snprintf(buf, IFNAMSIZ, name, i);
1029 if (!strncmp(buf, d->name, IFNAMSIZ))
1033 i = find_first_zero_bit(inuse, max_netdevices);
1034 free_page((unsigned long) inuse);
1038 snprintf(buf, IFNAMSIZ, name, i);
1039 if (!__dev_get_by_name(net, buf))
1042 /* It is possible to run out of possible slots
1043 * when the name is long and there isn't enough space left
1044 * for the digits, or if all bits are used.
1050 * dev_alloc_name - allocate a name for a device
1052 * @name: name format string
1054 * Passed a format string - eg "lt%d" it will try and find a suitable
1055 * id. It scans list of devices to build up a free map, then chooses
1056 * the first empty slot. The caller must hold the dev_base or rtnl lock
1057 * while allocating the name and adding the device in order to avoid
1059 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1060 * Returns the number of the unit assigned or a negative errno code.
1063 int dev_alloc_name(struct net_device *dev, const char *name)
1069 BUG_ON(!dev_net(dev));
1071 ret = __dev_alloc_name(net, name, buf);
1073 strlcpy(dev->name, buf, IFNAMSIZ);
1076 EXPORT_SYMBOL(dev_alloc_name);
1078 static int dev_alloc_name_ns(struct net *net,
1079 struct net_device *dev,
1085 ret = __dev_alloc_name(net, name, buf);
1087 strlcpy(dev->name, buf, IFNAMSIZ);
1091 static int dev_get_valid_name(struct net *net,
1092 struct net_device *dev,
1097 if (!dev_valid_name(name))
1100 if (strchr(name, '%'))
1101 return dev_alloc_name_ns(net, dev, name);
1102 else if (__dev_get_by_name(net, name))
1104 else if (dev->name != name)
1105 strlcpy(dev->name, name, IFNAMSIZ);
1111 * dev_change_name - change name of a device
1113 * @newname: name (or format string) must be at least IFNAMSIZ
1115 * Change name of a device, can pass format strings "eth%d".
1118 int dev_change_name(struct net_device *dev, const char *newname)
1120 unsigned char old_assign_type;
1121 char oldname[IFNAMSIZ];
1127 BUG_ON(!dev_net(dev));
1130 if (dev->flags & IFF_UP)
1133 write_seqcount_begin(&devnet_rename_seq);
1135 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1136 write_seqcount_end(&devnet_rename_seq);
1140 memcpy(oldname, dev->name, IFNAMSIZ);
1142 err = dev_get_valid_name(net, dev, newname);
1144 write_seqcount_end(&devnet_rename_seq);
1148 if (oldname[0] && !strchr(oldname, '%'))
1149 netdev_info(dev, "renamed from %s\n", oldname);
1151 old_assign_type = dev->name_assign_type;
1152 dev->name_assign_type = NET_NAME_RENAMED;
1155 ret = device_rename(&dev->dev, dev->name);
1157 memcpy(dev->name, oldname, IFNAMSIZ);
1158 dev->name_assign_type = old_assign_type;
1159 write_seqcount_end(&devnet_rename_seq);
1163 write_seqcount_end(&devnet_rename_seq);
1165 netdev_adjacent_rename_links(dev, oldname);
1167 write_lock_bh(&dev_base_lock);
1168 hlist_del_rcu(&dev->name_hlist);
1169 write_unlock_bh(&dev_base_lock);
1173 write_lock_bh(&dev_base_lock);
1174 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1175 write_unlock_bh(&dev_base_lock);
1177 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1178 ret = notifier_to_errno(ret);
1181 /* err >= 0 after dev_alloc_name() or stores the first errno */
1184 write_seqcount_begin(&devnet_rename_seq);
1185 memcpy(dev->name, oldname, IFNAMSIZ);
1186 memcpy(oldname, newname, IFNAMSIZ);
1187 dev->name_assign_type = old_assign_type;
1188 old_assign_type = NET_NAME_RENAMED;
1191 pr_err("%s: name change rollback failed: %d\n",
1200 * dev_set_alias - change ifalias of a device
1202 * @alias: name up to IFALIASZ
1203 * @len: limit of bytes to copy from info
1205 * Set ifalias for a device,
1207 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1213 if (len >= IFALIASZ)
1217 kfree(dev->ifalias);
1218 dev->ifalias = NULL;
1222 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1225 dev->ifalias = new_ifalias;
1227 strlcpy(dev->ifalias, alias, len+1);
1233 * netdev_features_change - device changes features
1234 * @dev: device to cause notification
1236 * Called to indicate a device has changed features.
1238 void netdev_features_change(struct net_device *dev)
1240 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1242 EXPORT_SYMBOL(netdev_features_change);
1245 * netdev_state_change - device changes state
1246 * @dev: device to cause notification
1248 * Called to indicate a device has changed state. This function calls
1249 * the notifier chains for netdev_chain and sends a NEWLINK message
1250 * to the routing socket.
1252 void netdev_state_change(struct net_device *dev)
1254 if (dev->flags & IFF_UP) {
1255 struct netdev_notifier_change_info change_info;
1257 change_info.flags_changed = 0;
1258 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1260 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1263 EXPORT_SYMBOL(netdev_state_change);
1266 * netdev_notify_peers - notify network peers about existence of @dev
1267 * @dev: network device
1269 * Generate traffic such that interested network peers are aware of
1270 * @dev, such as by generating a gratuitous ARP. This may be used when
1271 * a device wants to inform the rest of the network about some sort of
1272 * reconfiguration such as a failover event or virtual machine
1275 void netdev_notify_peers(struct net_device *dev)
1278 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1281 EXPORT_SYMBOL(netdev_notify_peers);
1283 static int __dev_open(struct net_device *dev)
1285 const struct net_device_ops *ops = dev->netdev_ops;
1290 if (!netif_device_present(dev))
1293 /* Block netpoll from trying to do any rx path servicing.
1294 * If we don't do this there is a chance ndo_poll_controller
1295 * or ndo_poll may be running while we open the device
1297 netpoll_poll_disable(dev);
1299 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1300 ret = notifier_to_errno(ret);
1304 set_bit(__LINK_STATE_START, &dev->state);
1306 if (ops->ndo_validate_addr)
1307 ret = ops->ndo_validate_addr(dev);
1309 if (!ret && ops->ndo_open)
1310 ret = ops->ndo_open(dev);
1312 netpoll_poll_enable(dev);
1315 clear_bit(__LINK_STATE_START, &dev->state);
1317 dev->flags |= IFF_UP;
1318 dev_set_rx_mode(dev);
1320 add_device_randomness(dev->dev_addr, dev->addr_len);
1327 * dev_open - prepare an interface for use.
1328 * @dev: device to open
1330 * Takes a device from down to up state. The device's private open
1331 * function is invoked and then the multicast lists are loaded. Finally
1332 * the device is moved into the up state and a %NETDEV_UP message is
1333 * sent to the netdev notifier chain.
1335 * Calling this function on an active interface is a nop. On a failure
1336 * a negative errno code is returned.
1338 int dev_open(struct net_device *dev)
1342 if (dev->flags & IFF_UP)
1345 ret = __dev_open(dev);
1349 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1350 call_netdevice_notifiers(NETDEV_UP, dev);
1354 EXPORT_SYMBOL(dev_open);
1356 static int __dev_close_many(struct list_head *head)
1358 struct net_device *dev;
1363 list_for_each_entry(dev, head, close_list) {
1364 /* Temporarily disable netpoll until the interface is down */
1365 netpoll_poll_disable(dev);
1367 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1369 clear_bit(__LINK_STATE_START, &dev->state);
1371 /* Synchronize to scheduled poll. We cannot touch poll list, it
1372 * can be even on different cpu. So just clear netif_running().
1374 * dev->stop() will invoke napi_disable() on all of it's
1375 * napi_struct instances on this device.
1377 smp_mb__after_atomic(); /* Commit netif_running(). */
1380 dev_deactivate_many(head);
1382 list_for_each_entry(dev, head, close_list) {
1383 const struct net_device_ops *ops = dev->netdev_ops;
1386 * Call the device specific close. This cannot fail.
1387 * Only if device is UP
1389 * We allow it to be called even after a DETACH hot-plug
1395 dev->flags &= ~IFF_UP;
1396 netpoll_poll_enable(dev);
1402 static int __dev_close(struct net_device *dev)
1407 list_add(&dev->close_list, &single);
1408 retval = __dev_close_many(&single);
1414 int dev_close_many(struct list_head *head, bool unlink)
1416 struct net_device *dev, *tmp;
1418 /* Remove the devices that don't need to be closed */
1419 list_for_each_entry_safe(dev, tmp, head, close_list)
1420 if (!(dev->flags & IFF_UP))
1421 list_del_init(&dev->close_list);
1423 __dev_close_many(head);
1425 list_for_each_entry_safe(dev, tmp, head, close_list) {
1426 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1427 call_netdevice_notifiers(NETDEV_DOWN, dev);
1429 list_del_init(&dev->close_list);
1434 EXPORT_SYMBOL(dev_close_many);
1437 * dev_close - shutdown an interface.
1438 * @dev: device to shutdown
1440 * This function moves an active device into down state. A
1441 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1442 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1445 int dev_close(struct net_device *dev)
1447 if (dev->flags & IFF_UP) {
1450 list_add(&dev->close_list, &single);
1451 dev_close_many(&single, true);
1456 EXPORT_SYMBOL(dev_close);
1460 * dev_disable_lro - disable Large Receive Offload on a device
1463 * Disable Large Receive Offload (LRO) on a net device. Must be
1464 * called under RTNL. This is needed if received packets may be
1465 * forwarded to another interface.
1467 void dev_disable_lro(struct net_device *dev)
1469 struct net_device *lower_dev;
1470 struct list_head *iter;
1472 dev->wanted_features &= ~NETIF_F_LRO;
1473 netdev_update_features(dev);
1475 if (unlikely(dev->features & NETIF_F_LRO))
1476 netdev_WARN(dev, "failed to disable LRO!\n");
1478 netdev_for_each_lower_dev(dev, lower_dev, iter)
1479 dev_disable_lro(lower_dev);
1481 EXPORT_SYMBOL(dev_disable_lro);
1483 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1484 struct net_device *dev)
1486 struct netdev_notifier_info info;
1488 netdev_notifier_info_init(&info, dev);
1489 return nb->notifier_call(nb, val, &info);
1492 static int dev_boot_phase = 1;
1495 * register_netdevice_notifier - register a network notifier block
1498 * Register a notifier to be called when network device events occur.
1499 * The notifier passed is linked into the kernel structures and must
1500 * not be reused until it has been unregistered. A negative errno code
1501 * is returned on a failure.
1503 * When registered all registration and up events are replayed
1504 * to the new notifier to allow device to have a race free
1505 * view of the network device list.
1508 int register_netdevice_notifier(struct notifier_block *nb)
1510 struct net_device *dev;
1511 struct net_device *last;
1516 err = raw_notifier_chain_register(&netdev_chain, nb);
1522 for_each_netdev(net, dev) {
1523 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1524 err = notifier_to_errno(err);
1528 if (!(dev->flags & IFF_UP))
1531 call_netdevice_notifier(nb, NETDEV_UP, dev);
1542 for_each_netdev(net, dev) {
1546 if (dev->flags & IFF_UP) {
1547 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1549 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1551 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 raw_notifier_chain_unregister(&netdev_chain, nb);
1559 EXPORT_SYMBOL(register_netdevice_notifier);
1562 * unregister_netdevice_notifier - unregister a network notifier block
1565 * Unregister a notifier previously registered by
1566 * register_netdevice_notifier(). The notifier is unlinked into the
1567 * kernel structures and may then be reused. A negative errno code
1568 * is returned on a failure.
1570 * After unregistering unregister and down device events are synthesized
1571 * for all devices on the device list to the removed notifier to remove
1572 * the need for special case cleanup code.
1575 int unregister_netdevice_notifier(struct notifier_block *nb)
1577 struct net_device *dev;
1582 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1587 for_each_netdev(net, dev) {
1588 if (dev->flags & IFF_UP) {
1589 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1591 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1593 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1600 EXPORT_SYMBOL(unregister_netdevice_notifier);
1603 * call_netdevice_notifiers_info - call all network notifier blocks
1604 * @val: value passed unmodified to notifier function
1605 * @dev: net_device pointer passed unmodified to notifier function
1606 * @info: notifier information data
1608 * Call all network notifier blocks. Parameters and return value
1609 * are as for raw_notifier_call_chain().
1612 static int call_netdevice_notifiers_info(unsigned long val,
1613 struct net_device *dev,
1614 struct netdev_notifier_info *info)
1617 netdev_notifier_info_init(info, dev);
1618 return raw_notifier_call_chain(&netdev_chain, val, info);
1622 * call_netdevice_notifiers - call all network notifier blocks
1623 * @val: value passed unmodified to notifier function
1624 * @dev: net_device pointer passed unmodified to notifier function
1626 * Call all network notifier blocks. Parameters and return value
1627 * are as for raw_notifier_call_chain().
1630 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1632 struct netdev_notifier_info info;
1634 return call_netdevice_notifiers_info(val, dev, &info);
1636 EXPORT_SYMBOL(call_netdevice_notifiers);
1638 #ifdef CONFIG_NET_INGRESS
1639 static struct static_key ingress_needed __read_mostly;
1641 void net_inc_ingress_queue(void)
1643 static_key_slow_inc(&ingress_needed);
1645 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1647 void net_dec_ingress_queue(void)
1649 static_key_slow_dec(&ingress_needed);
1651 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1654 static struct static_key netstamp_needed __read_mostly;
1655 #ifdef HAVE_JUMP_LABEL
1656 /* We are not allowed to call static_key_slow_dec() from irq context
1657 * If net_disable_timestamp() is called from irq context, defer the
1658 * static_key_slow_dec() calls.
1660 static atomic_t netstamp_needed_deferred;
1663 void net_enable_timestamp(void)
1665 #ifdef HAVE_JUMP_LABEL
1666 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1670 static_key_slow_dec(&netstamp_needed);
1674 static_key_slow_inc(&netstamp_needed);
1676 EXPORT_SYMBOL(net_enable_timestamp);
1678 void net_disable_timestamp(void)
1680 #ifdef HAVE_JUMP_LABEL
1681 if (in_interrupt()) {
1682 atomic_inc(&netstamp_needed_deferred);
1686 static_key_slow_dec(&netstamp_needed);
1688 EXPORT_SYMBOL(net_disable_timestamp);
1690 static inline void net_timestamp_set(struct sk_buff *skb)
1692 skb->tstamp.tv64 = 0;
1693 if (static_key_false(&netstamp_needed))
1694 __net_timestamp(skb);
1697 #define net_timestamp_check(COND, SKB) \
1698 if (static_key_false(&netstamp_needed)) { \
1699 if ((COND) && !(SKB)->tstamp.tv64) \
1700 __net_timestamp(SKB); \
1703 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1707 if (!(dev->flags & IFF_UP))
1710 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1711 if (skb->len <= len)
1714 /* if TSO is enabled, we don't care about the length as the packet
1715 * could be forwarded without being segmented before
1717 if (skb_is_gso(skb))
1722 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1724 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1726 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1727 unlikely(!is_skb_forwardable(dev, skb))) {
1728 atomic_long_inc(&dev->rx_dropped);
1733 skb_scrub_packet(skb, true);
1735 skb->protocol = eth_type_trans(skb, dev);
1736 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1740 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1743 * dev_forward_skb - loopback an skb to another netif
1745 * @dev: destination network device
1746 * @skb: buffer to forward
1749 * NET_RX_SUCCESS (no congestion)
1750 * NET_RX_DROP (packet was dropped, but freed)
1752 * dev_forward_skb can be used for injecting an skb from the
1753 * start_xmit function of one device into the receive queue
1754 * of another device.
1756 * The receiving device may be in another namespace, so
1757 * we have to clear all information in the skb that could
1758 * impact namespace isolation.
1760 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1762 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1764 EXPORT_SYMBOL_GPL(dev_forward_skb);
1766 static inline int deliver_skb(struct sk_buff *skb,
1767 struct packet_type *pt_prev,
1768 struct net_device *orig_dev)
1770 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1772 atomic_inc(&skb->users);
1773 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1776 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1777 struct packet_type **pt,
1778 struct net_device *orig_dev,
1780 struct list_head *ptype_list)
1782 struct packet_type *ptype, *pt_prev = *pt;
1784 list_for_each_entry_rcu(ptype, ptype_list, list) {
1785 if (ptype->type != type)
1788 deliver_skb(skb, pt_prev, orig_dev);
1794 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1796 if (!ptype->af_packet_priv || !skb->sk)
1799 if (ptype->id_match)
1800 return ptype->id_match(ptype, skb->sk);
1801 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1808 * Support routine. Sends outgoing frames to any network
1809 * taps currently in use.
1812 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1814 struct packet_type *ptype;
1815 struct sk_buff *skb2 = NULL;
1816 struct packet_type *pt_prev = NULL;
1817 struct list_head *ptype_list = &ptype_all;
1821 list_for_each_entry_rcu(ptype, ptype_list, list) {
1822 /* Never send packets back to the socket
1823 * they originated from - MvS (miquels@drinkel.ow.org)
1825 if (skb_loop_sk(ptype, skb))
1829 deliver_skb(skb2, pt_prev, skb->dev);
1834 /* need to clone skb, done only once */
1835 skb2 = skb_clone(skb, GFP_ATOMIC);
1839 net_timestamp_set(skb2);
1841 /* skb->nh should be correctly
1842 * set by sender, so that the second statement is
1843 * just protection against buggy protocols.
1845 skb_reset_mac_header(skb2);
1847 if (skb_network_header(skb2) < skb2->data ||
1848 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1849 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1850 ntohs(skb2->protocol),
1852 skb_reset_network_header(skb2);
1855 skb2->transport_header = skb2->network_header;
1856 skb2->pkt_type = PACKET_OUTGOING;
1860 if (ptype_list == &ptype_all) {
1861 ptype_list = &dev->ptype_all;
1866 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1871 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1872 * @dev: Network device
1873 * @txq: number of queues available
1875 * If real_num_tx_queues is changed the tc mappings may no longer be
1876 * valid. To resolve this verify the tc mapping remains valid and if
1877 * not NULL the mapping. With no priorities mapping to this
1878 * offset/count pair it will no longer be used. In the worst case TC0
1879 * is invalid nothing can be done so disable priority mappings. If is
1880 * expected that drivers will fix this mapping if they can before
1881 * calling netif_set_real_num_tx_queues.
1883 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1886 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1888 /* If TC0 is invalidated disable TC mapping */
1889 if (tc->offset + tc->count > txq) {
1890 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1895 /* Invalidated prio to tc mappings set to TC0 */
1896 for (i = 1; i < TC_BITMASK + 1; i++) {
1897 int q = netdev_get_prio_tc_map(dev, i);
1899 tc = &dev->tc_to_txq[q];
1900 if (tc->offset + tc->count > txq) {
1901 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1903 netdev_set_prio_tc_map(dev, i, 0);
1909 static DEFINE_MUTEX(xps_map_mutex);
1910 #define xmap_dereference(P) \
1911 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1913 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1916 struct xps_map *map = NULL;
1920 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1922 for (pos = 0; map && pos < map->len; pos++) {
1923 if (map->queues[pos] == index) {
1925 map->queues[pos] = map->queues[--map->len];
1927 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1928 kfree_rcu(map, rcu);
1938 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1940 struct xps_dev_maps *dev_maps;
1942 bool active = false;
1944 mutex_lock(&xps_map_mutex);
1945 dev_maps = xmap_dereference(dev->xps_maps);
1950 for_each_possible_cpu(cpu) {
1951 for (i = index; i < dev->num_tx_queues; i++) {
1952 if (!remove_xps_queue(dev_maps, cpu, i))
1955 if (i == dev->num_tx_queues)
1960 RCU_INIT_POINTER(dev->xps_maps, NULL);
1961 kfree_rcu(dev_maps, rcu);
1964 for (i = index; i < dev->num_tx_queues; i++)
1965 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1969 mutex_unlock(&xps_map_mutex);
1972 static struct xps_map *expand_xps_map(struct xps_map *map,
1975 struct xps_map *new_map;
1976 int alloc_len = XPS_MIN_MAP_ALLOC;
1979 for (pos = 0; map && pos < map->len; pos++) {
1980 if (map->queues[pos] != index)
1985 /* Need to add queue to this CPU's existing map */
1987 if (pos < map->alloc_len)
1990 alloc_len = map->alloc_len * 2;
1993 /* Need to allocate new map to store queue on this CPU's map */
1994 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1999 for (i = 0; i < pos; i++)
2000 new_map->queues[i] = map->queues[i];
2001 new_map->alloc_len = alloc_len;
2007 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2010 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2011 struct xps_map *map, *new_map;
2012 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2013 int cpu, numa_node_id = -2;
2014 bool active = false;
2016 mutex_lock(&xps_map_mutex);
2018 dev_maps = xmap_dereference(dev->xps_maps);
2020 /* allocate memory for queue storage */
2021 for_each_online_cpu(cpu) {
2022 if (!cpumask_test_cpu(cpu, mask))
2026 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2027 if (!new_dev_maps) {
2028 mutex_unlock(&xps_map_mutex);
2032 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2035 map = expand_xps_map(map, cpu, index);
2039 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2043 goto out_no_new_maps;
2045 for_each_possible_cpu(cpu) {
2046 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2047 /* add queue to CPU maps */
2050 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2051 while ((pos < map->len) && (map->queues[pos] != index))
2054 if (pos == map->len)
2055 map->queues[map->len++] = index;
2057 if (numa_node_id == -2)
2058 numa_node_id = cpu_to_node(cpu);
2059 else if (numa_node_id != cpu_to_node(cpu))
2062 } else if (dev_maps) {
2063 /* fill in the new device map from the old device map */
2064 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2065 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2070 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2072 /* Cleanup old maps */
2074 for_each_possible_cpu(cpu) {
2075 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2076 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2077 if (map && map != new_map)
2078 kfree_rcu(map, rcu);
2081 kfree_rcu(dev_maps, rcu);
2084 dev_maps = new_dev_maps;
2088 /* update Tx queue numa node */
2089 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2090 (numa_node_id >= 0) ? numa_node_id :
2096 /* removes queue from unused CPUs */
2097 for_each_possible_cpu(cpu) {
2098 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2101 if (remove_xps_queue(dev_maps, cpu, index))
2105 /* free map if not active */
2107 RCU_INIT_POINTER(dev->xps_maps, NULL);
2108 kfree_rcu(dev_maps, rcu);
2112 mutex_unlock(&xps_map_mutex);
2116 /* remove any maps that we added */
2117 for_each_possible_cpu(cpu) {
2118 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2119 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2121 if (new_map && new_map != map)
2125 mutex_unlock(&xps_map_mutex);
2127 kfree(new_dev_maps);
2130 EXPORT_SYMBOL(netif_set_xps_queue);
2134 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2135 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2137 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2141 if (txq < 1 || txq > dev->num_tx_queues)
2144 if (dev->reg_state == NETREG_REGISTERED ||
2145 dev->reg_state == NETREG_UNREGISTERING) {
2148 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2154 netif_setup_tc(dev, txq);
2156 if (txq < dev->real_num_tx_queues) {
2157 qdisc_reset_all_tx_gt(dev, txq);
2159 netif_reset_xps_queues_gt(dev, txq);
2164 dev->real_num_tx_queues = txq;
2167 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2171 * netif_set_real_num_rx_queues - set actual number of RX queues used
2172 * @dev: Network device
2173 * @rxq: Actual number of RX queues
2175 * This must be called either with the rtnl_lock held or before
2176 * registration of the net device. Returns 0 on success, or a
2177 * negative error code. If called before registration, it always
2180 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2184 if (rxq < 1 || rxq > dev->num_rx_queues)
2187 if (dev->reg_state == NETREG_REGISTERED) {
2190 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2196 dev->real_num_rx_queues = rxq;
2199 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2203 * netif_get_num_default_rss_queues - default number of RSS queues
2205 * This routine should set an upper limit on the number of RSS queues
2206 * used by default by multiqueue devices.
2208 int netif_get_num_default_rss_queues(void)
2210 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2212 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2214 static inline void __netif_reschedule(struct Qdisc *q)
2216 struct softnet_data *sd;
2217 unsigned long flags;
2219 local_irq_save(flags);
2220 sd = this_cpu_ptr(&softnet_data);
2221 q->next_sched = NULL;
2222 *sd->output_queue_tailp = q;
2223 sd->output_queue_tailp = &q->next_sched;
2224 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2225 local_irq_restore(flags);
2228 void __netif_schedule(struct Qdisc *q)
2230 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2231 __netif_reschedule(q);
2233 EXPORT_SYMBOL(__netif_schedule);
2235 struct dev_kfree_skb_cb {
2236 enum skb_free_reason reason;
2239 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2241 return (struct dev_kfree_skb_cb *)skb->cb;
2244 void netif_schedule_queue(struct netdev_queue *txq)
2247 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2248 struct Qdisc *q = rcu_dereference(txq->qdisc);
2250 __netif_schedule(q);
2254 EXPORT_SYMBOL(netif_schedule_queue);
2257 * netif_wake_subqueue - allow sending packets on subqueue
2258 * @dev: network device
2259 * @queue_index: sub queue index
2261 * Resume individual transmit queue of a device with multiple transmit queues.
2263 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2265 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2267 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2271 q = rcu_dereference(txq->qdisc);
2272 __netif_schedule(q);
2276 EXPORT_SYMBOL(netif_wake_subqueue);
2278 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2280 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2284 q = rcu_dereference(dev_queue->qdisc);
2285 __netif_schedule(q);
2289 EXPORT_SYMBOL(netif_tx_wake_queue);
2291 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2293 unsigned long flags;
2295 if (likely(atomic_read(&skb->users) == 1)) {
2297 atomic_set(&skb->users, 0);
2298 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2301 get_kfree_skb_cb(skb)->reason = reason;
2302 local_irq_save(flags);
2303 skb->next = __this_cpu_read(softnet_data.completion_queue);
2304 __this_cpu_write(softnet_data.completion_queue, skb);
2305 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2306 local_irq_restore(flags);
2308 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2310 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2312 if (in_irq() || irqs_disabled())
2313 __dev_kfree_skb_irq(skb, reason);
2317 EXPORT_SYMBOL(__dev_kfree_skb_any);
2321 * netif_device_detach - mark device as removed
2322 * @dev: network device
2324 * Mark device as removed from system and therefore no longer available.
2326 void netif_device_detach(struct net_device *dev)
2328 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2329 netif_running(dev)) {
2330 netif_tx_stop_all_queues(dev);
2333 EXPORT_SYMBOL(netif_device_detach);
2336 * netif_device_attach - mark device as attached
2337 * @dev: network device
2339 * Mark device as attached from system and restart if needed.
2341 void netif_device_attach(struct net_device *dev)
2343 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2344 netif_running(dev)) {
2345 netif_tx_wake_all_queues(dev);
2346 __netdev_watchdog_up(dev);
2349 EXPORT_SYMBOL(netif_device_attach);
2352 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2353 * to be used as a distribution range.
2355 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2356 unsigned int num_tx_queues)
2360 u16 qcount = num_tx_queues;
2362 if (skb_rx_queue_recorded(skb)) {
2363 hash = skb_get_rx_queue(skb);
2364 while (unlikely(hash >= num_tx_queues))
2365 hash -= num_tx_queues;
2370 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2371 qoffset = dev->tc_to_txq[tc].offset;
2372 qcount = dev->tc_to_txq[tc].count;
2375 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2377 EXPORT_SYMBOL(__skb_tx_hash);
2379 static void skb_warn_bad_offload(const struct sk_buff *skb)
2381 static const netdev_features_t null_features = 0;
2382 struct net_device *dev = skb->dev;
2383 const char *driver = "";
2385 if (!net_ratelimit())
2388 if (dev && dev->dev.parent)
2389 driver = dev_driver_string(dev->dev.parent);
2391 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2392 "gso_type=%d ip_summed=%d\n",
2393 driver, dev ? &dev->features : &null_features,
2394 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2395 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2396 skb_shinfo(skb)->gso_type, skb->ip_summed);
2400 * Invalidate hardware checksum when packet is to be mangled, and
2401 * complete checksum manually on outgoing path.
2403 int skb_checksum_help(struct sk_buff *skb)
2406 int ret = 0, offset;
2408 if (skb->ip_summed == CHECKSUM_COMPLETE)
2409 goto out_set_summed;
2411 if (unlikely(skb_shinfo(skb)->gso_size)) {
2412 skb_warn_bad_offload(skb);
2416 /* Before computing a checksum, we should make sure no frag could
2417 * be modified by an external entity : checksum could be wrong.
2419 if (skb_has_shared_frag(skb)) {
2420 ret = __skb_linearize(skb);
2425 offset = skb_checksum_start_offset(skb);
2426 BUG_ON(offset >= skb_headlen(skb));
2427 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2429 offset += skb->csum_offset;
2430 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2432 if (skb_cloned(skb) &&
2433 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2434 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2439 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2441 skb->ip_summed = CHECKSUM_NONE;
2445 EXPORT_SYMBOL(skb_checksum_help);
2447 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2449 __be16 type = skb->protocol;
2451 /* Tunnel gso handlers can set protocol to ethernet. */
2452 if (type == htons(ETH_P_TEB)) {
2455 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2458 eth = (struct ethhdr *)skb_mac_header(skb);
2459 type = eth->h_proto;
2462 return __vlan_get_protocol(skb, type, depth);
2466 * skb_mac_gso_segment - mac layer segmentation handler.
2467 * @skb: buffer to segment
2468 * @features: features for the output path (see dev->features)
2470 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2471 netdev_features_t features)
2473 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2474 struct packet_offload *ptype;
2475 int vlan_depth = skb->mac_len;
2476 __be16 type = skb_network_protocol(skb, &vlan_depth);
2478 if (unlikely(!type))
2479 return ERR_PTR(-EINVAL);
2481 __skb_pull(skb, vlan_depth);
2484 list_for_each_entry_rcu(ptype, &offload_base, list) {
2485 if (ptype->type == type && ptype->callbacks.gso_segment) {
2486 segs = ptype->callbacks.gso_segment(skb, features);
2492 __skb_push(skb, skb->data - skb_mac_header(skb));
2496 EXPORT_SYMBOL(skb_mac_gso_segment);
2499 /* openvswitch calls this on rx path, so we need a different check.
2501 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2504 return skb->ip_summed != CHECKSUM_PARTIAL;
2506 return skb->ip_summed == CHECKSUM_NONE;
2510 * __skb_gso_segment - Perform segmentation on skb.
2511 * @skb: buffer to segment
2512 * @features: features for the output path (see dev->features)
2513 * @tx_path: whether it is called in TX path
2515 * This function segments the given skb and returns a list of segments.
2517 * It may return NULL if the skb requires no segmentation. This is
2518 * only possible when GSO is used for verifying header integrity.
2520 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2521 netdev_features_t features, bool tx_path)
2523 if (unlikely(skb_needs_check(skb, tx_path))) {
2526 skb_warn_bad_offload(skb);
2528 err = skb_cow_head(skb, 0);
2530 return ERR_PTR(err);
2533 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2534 SKB_GSO_CB(skb)->encap_level = 0;
2536 skb_reset_mac_header(skb);
2537 skb_reset_mac_len(skb);
2539 return skb_mac_gso_segment(skb, features);
2541 EXPORT_SYMBOL(__skb_gso_segment);
2543 /* Take action when hardware reception checksum errors are detected. */
2545 void netdev_rx_csum_fault(struct net_device *dev)
2547 if (net_ratelimit()) {
2548 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2552 EXPORT_SYMBOL(netdev_rx_csum_fault);
2555 /* Actually, we should eliminate this check as soon as we know, that:
2556 * 1. IOMMU is present and allows to map all the memory.
2557 * 2. No high memory really exists on this machine.
2560 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2562 #ifdef CONFIG_HIGHMEM
2564 if (!(dev->features & NETIF_F_HIGHDMA)) {
2565 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2566 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2567 if (PageHighMem(skb_frag_page(frag)))
2572 if (PCI_DMA_BUS_IS_PHYS) {
2573 struct device *pdev = dev->dev.parent;
2577 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2578 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2579 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2580 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2588 /* If MPLS offload request, verify we are testing hardware MPLS features
2589 * instead of standard features for the netdev.
2591 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2592 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2593 netdev_features_t features,
2596 if (eth_p_mpls(type))
2597 features &= skb->dev->mpls_features;
2602 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2603 netdev_features_t features,
2610 static netdev_features_t harmonize_features(struct sk_buff *skb,
2611 netdev_features_t features)
2616 type = skb_network_protocol(skb, &tmp);
2617 features = net_mpls_features(skb, features, type);
2619 if (skb->ip_summed != CHECKSUM_NONE &&
2620 !can_checksum_protocol(features, type)) {
2621 features &= ~NETIF_F_ALL_CSUM;
2622 } else if (illegal_highdma(skb->dev, skb)) {
2623 features &= ~NETIF_F_SG;
2629 netdev_features_t passthru_features_check(struct sk_buff *skb,
2630 struct net_device *dev,
2631 netdev_features_t features)
2635 EXPORT_SYMBOL(passthru_features_check);
2637 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2638 struct net_device *dev,
2639 netdev_features_t features)
2641 return vlan_features_check(skb, features);
2644 netdev_features_t netif_skb_features(struct sk_buff *skb)
2646 struct net_device *dev = skb->dev;
2647 netdev_features_t features = dev->features;
2648 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2650 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2651 features &= ~NETIF_F_GSO_MASK;
2653 /* If encapsulation offload request, verify we are testing
2654 * hardware encapsulation features instead of standard
2655 * features for the netdev
2657 if (skb->encapsulation)
2658 features &= dev->hw_enc_features;
2660 if (skb_vlan_tagged(skb))
2661 features = netdev_intersect_features(features,
2662 dev->vlan_features |
2663 NETIF_F_HW_VLAN_CTAG_TX |
2664 NETIF_F_HW_VLAN_STAG_TX);
2666 if (dev->netdev_ops->ndo_features_check)
2667 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2670 features &= dflt_features_check(skb, dev, features);
2672 return harmonize_features(skb, features);
2674 EXPORT_SYMBOL(netif_skb_features);
2676 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2677 struct netdev_queue *txq, bool more)
2682 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2683 dev_queue_xmit_nit(skb, dev);
2686 trace_net_dev_start_xmit(skb, dev);
2687 rc = netdev_start_xmit(skb, dev, txq, more);
2688 trace_net_dev_xmit(skb, rc, dev, len);
2693 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2694 struct netdev_queue *txq, int *ret)
2696 struct sk_buff *skb = first;
2697 int rc = NETDEV_TX_OK;
2700 struct sk_buff *next = skb->next;
2703 rc = xmit_one(skb, dev, txq, next != NULL);
2704 if (unlikely(!dev_xmit_complete(rc))) {
2710 if (netif_xmit_stopped(txq) && skb) {
2711 rc = NETDEV_TX_BUSY;
2721 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2722 netdev_features_t features)
2724 if (skb_vlan_tag_present(skb) &&
2725 !vlan_hw_offload_capable(features, skb->vlan_proto))
2726 skb = __vlan_hwaccel_push_inside(skb);
2730 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2732 netdev_features_t features;
2737 features = netif_skb_features(skb);
2738 skb = validate_xmit_vlan(skb, features);
2742 if (netif_needs_gso(skb, features)) {
2743 struct sk_buff *segs;
2745 segs = skb_gso_segment(skb, features);
2753 if (skb_needs_linearize(skb, features) &&
2754 __skb_linearize(skb))
2757 /* If packet is not checksummed and device does not
2758 * support checksumming for this protocol, complete
2759 * checksumming here.
2761 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2762 if (skb->encapsulation)
2763 skb_set_inner_transport_header(skb,
2764 skb_checksum_start_offset(skb));
2766 skb_set_transport_header(skb,
2767 skb_checksum_start_offset(skb));
2768 if (!(features & NETIF_F_ALL_CSUM) &&
2769 skb_checksum_help(skb))
2782 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2784 struct sk_buff *next, *head = NULL, *tail;
2786 for (; skb != NULL; skb = next) {
2790 /* in case skb wont be segmented, point to itself */
2793 skb = validate_xmit_skb(skb, dev);
2801 /* If skb was segmented, skb->prev points to
2802 * the last segment. If not, it still contains skb.
2809 static void qdisc_pkt_len_init(struct sk_buff *skb)
2811 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2813 qdisc_skb_cb(skb)->pkt_len = skb->len;
2815 /* To get more precise estimation of bytes sent on wire,
2816 * we add to pkt_len the headers size of all segments
2818 if (shinfo->gso_size) {
2819 unsigned int hdr_len;
2820 u16 gso_segs = shinfo->gso_segs;
2822 /* mac layer + network layer */
2823 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2825 /* + transport layer */
2826 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2827 hdr_len += tcp_hdrlen(skb);
2829 hdr_len += sizeof(struct udphdr);
2831 if (shinfo->gso_type & SKB_GSO_DODGY)
2832 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2835 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2839 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2840 struct net_device *dev,
2841 struct netdev_queue *txq)
2843 spinlock_t *root_lock = qdisc_lock(q);
2847 qdisc_pkt_len_init(skb);
2848 qdisc_calculate_pkt_len(skb, q);
2850 * Heuristic to force contended enqueues to serialize on a
2851 * separate lock before trying to get qdisc main lock.
2852 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2853 * often and dequeue packets faster.
2855 contended = qdisc_is_running(q);
2856 if (unlikely(contended))
2857 spin_lock(&q->busylock);
2859 spin_lock(root_lock);
2860 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2863 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2864 qdisc_run_begin(q)) {
2866 * This is a work-conserving queue; there are no old skbs
2867 * waiting to be sent out; and the qdisc is not running -
2868 * xmit the skb directly.
2871 qdisc_bstats_update(q, skb);
2873 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2874 if (unlikely(contended)) {
2875 spin_unlock(&q->busylock);
2882 rc = NET_XMIT_SUCCESS;
2884 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2885 if (qdisc_run_begin(q)) {
2886 if (unlikely(contended)) {
2887 spin_unlock(&q->busylock);
2893 spin_unlock(root_lock);
2894 if (unlikely(contended))
2895 spin_unlock(&q->busylock);
2899 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2900 static void skb_update_prio(struct sk_buff *skb)
2902 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2904 if (!skb->priority && skb->sk && map) {
2905 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2907 if (prioidx < map->priomap_len)
2908 skb->priority = map->priomap[prioidx];
2912 #define skb_update_prio(skb)
2915 DEFINE_PER_CPU(int, xmit_recursion);
2916 EXPORT_SYMBOL(xmit_recursion);
2918 #define RECURSION_LIMIT 10
2921 * dev_loopback_xmit - loop back @skb
2922 * @skb: buffer to transmit
2924 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb)
2926 skb_reset_mac_header(skb);
2927 __skb_pull(skb, skb_network_offset(skb));
2928 skb->pkt_type = PACKET_LOOPBACK;
2929 skb->ip_summed = CHECKSUM_UNNECESSARY;
2930 WARN_ON(!skb_dst(skb));
2935 EXPORT_SYMBOL(dev_loopback_xmit);
2937 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2940 struct xps_dev_maps *dev_maps;
2941 struct xps_map *map;
2942 int queue_index = -1;
2945 dev_maps = rcu_dereference(dev->xps_maps);
2947 map = rcu_dereference(
2948 dev_maps->cpu_map[skb->sender_cpu - 1]);
2951 queue_index = map->queues[0];
2953 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2955 if (unlikely(queue_index >= dev->real_num_tx_queues))
2967 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2969 struct sock *sk = skb->sk;
2970 int queue_index = sk_tx_queue_get(sk);
2972 if (queue_index < 0 || skb->ooo_okay ||
2973 queue_index >= dev->real_num_tx_queues) {
2974 int new_index = get_xps_queue(dev, skb);
2976 new_index = skb_tx_hash(dev, skb);
2978 if (queue_index != new_index && sk &&
2979 rcu_access_pointer(sk->sk_dst_cache))
2980 sk_tx_queue_set(sk, new_index);
2982 queue_index = new_index;
2988 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2989 struct sk_buff *skb,
2992 int queue_index = 0;
2995 if (skb->sender_cpu == 0)
2996 skb->sender_cpu = raw_smp_processor_id() + 1;
2999 if (dev->real_num_tx_queues != 1) {
3000 const struct net_device_ops *ops = dev->netdev_ops;
3001 if (ops->ndo_select_queue)
3002 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3005 queue_index = __netdev_pick_tx(dev, skb);
3008 queue_index = netdev_cap_txqueue(dev, queue_index);
3011 skb_set_queue_mapping(skb, queue_index);
3012 return netdev_get_tx_queue(dev, queue_index);
3016 * __dev_queue_xmit - transmit a buffer
3017 * @skb: buffer to transmit
3018 * @accel_priv: private data used for L2 forwarding offload
3020 * Queue a buffer for transmission to a network device. The caller must
3021 * have set the device and priority and built the buffer before calling
3022 * this function. The function can be called from an interrupt.
3024 * A negative errno code is returned on a failure. A success does not
3025 * guarantee the frame will be transmitted as it may be dropped due
3026 * to congestion or traffic shaping.
3028 * -----------------------------------------------------------------------------------
3029 * I notice this method can also return errors from the queue disciplines,
3030 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3033 * Regardless of the return value, the skb is consumed, so it is currently
3034 * difficult to retry a send to this method. (You can bump the ref count
3035 * before sending to hold a reference for retry if you are careful.)
3037 * When calling this method, interrupts MUST be enabled. This is because
3038 * the BH enable code must have IRQs enabled so that it will not deadlock.
3041 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3043 struct net_device *dev = skb->dev;
3044 struct netdev_queue *txq;
3048 skb_reset_mac_header(skb);
3050 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3051 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3053 /* Disable soft irqs for various locks below. Also
3054 * stops preemption for RCU.
3058 skb_update_prio(skb);
3060 /* If device/qdisc don't need skb->dst, release it right now while
3061 * its hot in this cpu cache.
3063 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3068 txq = netdev_pick_tx(dev, skb, accel_priv);
3069 q = rcu_dereference_bh(txq->qdisc);
3071 #ifdef CONFIG_NET_CLS_ACT
3072 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3074 trace_net_dev_queue(skb);
3076 rc = __dev_xmit_skb(skb, q, dev, txq);
3080 /* The device has no queue. Common case for software devices:
3081 loopback, all the sorts of tunnels...
3083 Really, it is unlikely that netif_tx_lock protection is necessary
3084 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3086 However, it is possible, that they rely on protection
3089 Check this and shot the lock. It is not prone from deadlocks.
3090 Either shot noqueue qdisc, it is even simpler 8)
3092 if (dev->flags & IFF_UP) {
3093 int cpu = smp_processor_id(); /* ok because BHs are off */
3095 if (txq->xmit_lock_owner != cpu) {
3097 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3098 goto recursion_alert;
3100 skb = validate_xmit_skb(skb, dev);
3104 HARD_TX_LOCK(dev, txq, cpu);
3106 if (!netif_xmit_stopped(txq)) {
3107 __this_cpu_inc(xmit_recursion);
3108 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3109 __this_cpu_dec(xmit_recursion);
3110 if (dev_xmit_complete(rc)) {
3111 HARD_TX_UNLOCK(dev, txq);
3115 HARD_TX_UNLOCK(dev, txq);
3116 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3119 /* Recursion is detected! It is possible,
3123 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3130 rcu_read_unlock_bh();
3132 atomic_long_inc(&dev->tx_dropped);
3133 kfree_skb_list(skb);
3136 rcu_read_unlock_bh();
3140 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb)
3142 return __dev_queue_xmit(skb, NULL);
3144 EXPORT_SYMBOL(dev_queue_xmit_sk);
3146 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3148 return __dev_queue_xmit(skb, accel_priv);
3150 EXPORT_SYMBOL(dev_queue_xmit_accel);
3153 /*=======================================================================
3155 =======================================================================*/
3157 int netdev_max_backlog __read_mostly = 1000;
3158 EXPORT_SYMBOL(netdev_max_backlog);
3160 int netdev_tstamp_prequeue __read_mostly = 1;
3161 int netdev_budget __read_mostly = 300;
3162 int weight_p __read_mostly = 64; /* old backlog weight */
3164 /* Called with irq disabled */
3165 static inline void ____napi_schedule(struct softnet_data *sd,
3166 struct napi_struct *napi)
3168 list_add_tail(&napi->poll_list, &sd->poll_list);
3169 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3174 /* One global table that all flow-based protocols share. */
3175 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3176 EXPORT_SYMBOL(rps_sock_flow_table);
3177 u32 rps_cpu_mask __read_mostly;
3178 EXPORT_SYMBOL(rps_cpu_mask);
3180 struct static_key rps_needed __read_mostly;
3182 static struct rps_dev_flow *
3183 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3184 struct rps_dev_flow *rflow, u16 next_cpu)
3186 if (next_cpu < nr_cpu_ids) {
3187 #ifdef CONFIG_RFS_ACCEL
3188 struct netdev_rx_queue *rxqueue;
3189 struct rps_dev_flow_table *flow_table;
3190 struct rps_dev_flow *old_rflow;
3195 /* Should we steer this flow to a different hardware queue? */
3196 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3197 !(dev->features & NETIF_F_NTUPLE))
3199 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3200 if (rxq_index == skb_get_rx_queue(skb))
3203 rxqueue = dev->_rx + rxq_index;
3204 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3207 flow_id = skb_get_hash(skb) & flow_table->mask;
3208 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3209 rxq_index, flow_id);
3213 rflow = &flow_table->flows[flow_id];
3215 if (old_rflow->filter == rflow->filter)
3216 old_rflow->filter = RPS_NO_FILTER;
3220 per_cpu(softnet_data, next_cpu).input_queue_head;
3223 rflow->cpu = next_cpu;
3228 * get_rps_cpu is called from netif_receive_skb and returns the target
3229 * CPU from the RPS map of the receiving queue for a given skb.
3230 * rcu_read_lock must be held on entry.
3232 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3233 struct rps_dev_flow **rflowp)
3235 const struct rps_sock_flow_table *sock_flow_table;
3236 struct netdev_rx_queue *rxqueue = dev->_rx;
3237 struct rps_dev_flow_table *flow_table;
3238 struct rps_map *map;
3243 if (skb_rx_queue_recorded(skb)) {
3244 u16 index = skb_get_rx_queue(skb);
3246 if (unlikely(index >= dev->real_num_rx_queues)) {
3247 WARN_ONCE(dev->real_num_rx_queues > 1,
3248 "%s received packet on queue %u, but number "
3249 "of RX queues is %u\n",
3250 dev->name, index, dev->real_num_rx_queues);
3256 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3258 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3259 map = rcu_dereference(rxqueue->rps_map);
3260 if (!flow_table && !map)
3263 skb_reset_network_header(skb);
3264 hash = skb_get_hash(skb);
3268 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3269 if (flow_table && sock_flow_table) {
3270 struct rps_dev_flow *rflow;
3274 /* First check into global flow table if there is a match */
3275 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3276 if ((ident ^ hash) & ~rps_cpu_mask)
3279 next_cpu = ident & rps_cpu_mask;
3281 /* OK, now we know there is a match,
3282 * we can look at the local (per receive queue) flow table
3284 rflow = &flow_table->flows[hash & flow_table->mask];
3288 * If the desired CPU (where last recvmsg was done) is
3289 * different from current CPU (one in the rx-queue flow
3290 * table entry), switch if one of the following holds:
3291 * - Current CPU is unset (>= nr_cpu_ids).
3292 * - Current CPU is offline.
3293 * - The current CPU's queue tail has advanced beyond the
3294 * last packet that was enqueued using this table entry.
3295 * This guarantees that all previous packets for the flow
3296 * have been dequeued, thus preserving in order delivery.
3298 if (unlikely(tcpu != next_cpu) &&
3299 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3300 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3301 rflow->last_qtail)) >= 0)) {
3303 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3306 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3316 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3317 if (cpu_online(tcpu)) {
3327 #ifdef CONFIG_RFS_ACCEL
3330 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3331 * @dev: Device on which the filter was set
3332 * @rxq_index: RX queue index
3333 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3334 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3336 * Drivers that implement ndo_rx_flow_steer() should periodically call
3337 * this function for each installed filter and remove the filters for
3338 * which it returns %true.
3340 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3341 u32 flow_id, u16 filter_id)
3343 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3344 struct rps_dev_flow_table *flow_table;
3345 struct rps_dev_flow *rflow;
3350 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3351 if (flow_table && flow_id <= flow_table->mask) {
3352 rflow = &flow_table->flows[flow_id];
3353 cpu = ACCESS_ONCE(rflow->cpu);
3354 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3355 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3356 rflow->last_qtail) <
3357 (int)(10 * flow_table->mask)))
3363 EXPORT_SYMBOL(rps_may_expire_flow);
3365 #endif /* CONFIG_RFS_ACCEL */
3367 /* Called from hardirq (IPI) context */
3368 static void rps_trigger_softirq(void *data)
3370 struct softnet_data *sd = data;
3372 ____napi_schedule(sd, &sd->backlog);
3376 #endif /* CONFIG_RPS */
3379 * Check if this softnet_data structure is another cpu one
3380 * If yes, queue it to our IPI list and return 1
3383 static int rps_ipi_queued(struct softnet_data *sd)
3386 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3389 sd->rps_ipi_next = mysd->rps_ipi_list;
3390 mysd->rps_ipi_list = sd;
3392 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3395 #endif /* CONFIG_RPS */
3399 #ifdef CONFIG_NET_FLOW_LIMIT
3400 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3403 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3405 #ifdef CONFIG_NET_FLOW_LIMIT
3406 struct sd_flow_limit *fl;
3407 struct softnet_data *sd;
3408 unsigned int old_flow, new_flow;
3410 if (qlen < (netdev_max_backlog >> 1))
3413 sd = this_cpu_ptr(&softnet_data);
3416 fl = rcu_dereference(sd->flow_limit);
3418 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3419 old_flow = fl->history[fl->history_head];
3420 fl->history[fl->history_head] = new_flow;
3423 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3425 if (likely(fl->buckets[old_flow]))
3426 fl->buckets[old_flow]--;
3428 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3440 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3441 * queue (may be a remote CPU queue).
3443 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3444 unsigned int *qtail)
3446 struct softnet_data *sd;
3447 unsigned long flags;
3450 sd = &per_cpu(softnet_data, cpu);
3452 local_irq_save(flags);
3455 qlen = skb_queue_len(&sd->input_pkt_queue);
3456 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3459 __skb_queue_tail(&sd->input_pkt_queue, skb);
3460 input_queue_tail_incr_save(sd, qtail);
3462 local_irq_restore(flags);
3463 return NET_RX_SUCCESS;
3466 /* Schedule NAPI for backlog device
3467 * We can use non atomic operation since we own the queue lock
3469 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3470 if (!rps_ipi_queued(sd))
3471 ____napi_schedule(sd, &sd->backlog);
3479 local_irq_restore(flags);
3481 atomic_long_inc(&skb->dev->rx_dropped);
3486 static int netif_rx_internal(struct sk_buff *skb)
3490 net_timestamp_check(netdev_tstamp_prequeue, skb);
3492 trace_netif_rx(skb);
3494 if (static_key_false(&rps_needed)) {
3495 struct rps_dev_flow voidflow, *rflow = &voidflow;
3501 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3503 cpu = smp_processor_id();
3505 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3513 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3520 * netif_rx - post buffer to the network code
3521 * @skb: buffer to post
3523 * This function receives a packet from a device driver and queues it for
3524 * the upper (protocol) levels to process. It always succeeds. The buffer
3525 * may be dropped during processing for congestion control or by the
3529 * NET_RX_SUCCESS (no congestion)
3530 * NET_RX_DROP (packet was dropped)
3534 int netif_rx(struct sk_buff *skb)
3536 trace_netif_rx_entry(skb);
3538 return netif_rx_internal(skb);
3540 EXPORT_SYMBOL(netif_rx);
3542 int netif_rx_ni(struct sk_buff *skb)
3546 trace_netif_rx_ni_entry(skb);
3549 err = netif_rx_internal(skb);
3550 if (local_softirq_pending())
3556 EXPORT_SYMBOL(netif_rx_ni);
3558 static void net_tx_action(struct softirq_action *h)
3560 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3562 if (sd->completion_queue) {
3563 struct sk_buff *clist;
3565 local_irq_disable();
3566 clist = sd->completion_queue;
3567 sd->completion_queue = NULL;
3571 struct sk_buff *skb = clist;
3572 clist = clist->next;
3574 WARN_ON(atomic_read(&skb->users));
3575 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3576 trace_consume_skb(skb);
3578 trace_kfree_skb(skb, net_tx_action);
3583 if (sd->output_queue) {
3586 local_irq_disable();
3587 head = sd->output_queue;
3588 sd->output_queue = NULL;
3589 sd->output_queue_tailp = &sd->output_queue;
3593 struct Qdisc *q = head;
3594 spinlock_t *root_lock;
3596 head = head->next_sched;
3598 root_lock = qdisc_lock(q);
3599 if (spin_trylock(root_lock)) {
3600 smp_mb__before_atomic();
3601 clear_bit(__QDISC_STATE_SCHED,
3604 spin_unlock(root_lock);
3606 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3608 __netif_reschedule(q);
3610 smp_mb__before_atomic();
3611 clear_bit(__QDISC_STATE_SCHED,
3619 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3620 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3621 /* This hook is defined here for ATM LANE */
3622 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3623 unsigned char *addr) __read_mostly;
3624 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3627 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3628 struct packet_type **pt_prev,
3629 int *ret, struct net_device *orig_dev)
3631 #ifdef CONFIG_NET_CLS_ACT
3632 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3633 struct tcf_result cl_res;
3635 /* If there's at least one ingress present somewhere (so
3636 * we get here via enabled static key), remaining devices
3637 * that are not configured with an ingress qdisc will bail
3643 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3647 qdisc_skb_cb(skb)->pkt_len = skb->len;
3648 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3649 qdisc_bstats_update_cpu(cl->q, skb);
3651 switch (tc_classify(skb, cl, &cl_res)) {
3653 case TC_ACT_RECLASSIFY:
3654 skb->tc_index = TC_H_MIN(cl_res.classid);
3657 qdisc_qstats_drop_cpu(cl->q);
3665 #endif /* CONFIG_NET_CLS_ACT */
3670 * netdev_rx_handler_register - register receive handler
3671 * @dev: device to register a handler for
3672 * @rx_handler: receive handler to register
3673 * @rx_handler_data: data pointer that is used by rx handler
3675 * Register a receive handler for a device. This handler will then be
3676 * called from __netif_receive_skb. A negative errno code is returned
3679 * The caller must hold the rtnl_mutex.
3681 * For a general description of rx_handler, see enum rx_handler_result.
3683 int netdev_rx_handler_register(struct net_device *dev,
3684 rx_handler_func_t *rx_handler,
3685 void *rx_handler_data)
3689 if (dev->rx_handler)
3692 /* Note: rx_handler_data must be set before rx_handler */
3693 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3694 rcu_assign_pointer(dev->rx_handler, rx_handler);
3698 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3701 * netdev_rx_handler_unregister - unregister receive handler
3702 * @dev: device to unregister a handler from
3704 * Unregister a receive handler from a device.
3706 * The caller must hold the rtnl_mutex.
3708 void netdev_rx_handler_unregister(struct net_device *dev)
3712 RCU_INIT_POINTER(dev->rx_handler, NULL);
3713 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3714 * section has a guarantee to see a non NULL rx_handler_data
3718 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3720 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3723 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3724 * the special handling of PFMEMALLOC skbs.
3726 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3728 switch (skb->protocol) {
3729 case htons(ETH_P_ARP):
3730 case htons(ETH_P_IP):
3731 case htons(ETH_P_IPV6):
3732 case htons(ETH_P_8021Q):
3733 case htons(ETH_P_8021AD):
3740 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3741 int *ret, struct net_device *orig_dev)
3743 #ifdef CONFIG_NETFILTER_INGRESS
3744 if (nf_hook_ingress_active(skb)) {
3746 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3750 return nf_hook_ingress(skb);
3752 #endif /* CONFIG_NETFILTER_INGRESS */
3756 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3758 struct packet_type *ptype, *pt_prev;
3759 rx_handler_func_t *rx_handler;
3760 struct net_device *orig_dev;
3761 bool deliver_exact = false;
3762 int ret = NET_RX_DROP;
3765 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3767 trace_netif_receive_skb(skb);
3769 orig_dev = skb->dev;
3771 skb_reset_network_header(skb);
3772 if (!skb_transport_header_was_set(skb))
3773 skb_reset_transport_header(skb);
3774 skb_reset_mac_len(skb);
3781 skb->skb_iif = skb->dev->ifindex;
3783 __this_cpu_inc(softnet_data.processed);
3785 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3786 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3787 skb = skb_vlan_untag(skb);
3792 #ifdef CONFIG_NET_CLS_ACT
3793 if (skb->tc_verd & TC_NCLS) {
3794 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3802 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3804 ret = deliver_skb(skb, pt_prev, orig_dev);
3808 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3810 ret = deliver_skb(skb, pt_prev, orig_dev);
3815 #ifdef CONFIG_NET_INGRESS
3816 if (static_key_false(&ingress_needed)) {
3817 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3821 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3825 #ifdef CONFIG_NET_CLS_ACT
3829 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3832 if (skb_vlan_tag_present(skb)) {
3834 ret = deliver_skb(skb, pt_prev, orig_dev);
3837 if (vlan_do_receive(&skb))
3839 else if (unlikely(!skb))
3843 rx_handler = rcu_dereference(skb->dev->rx_handler);
3846 ret = deliver_skb(skb, pt_prev, orig_dev);
3849 switch (rx_handler(&skb)) {
3850 case RX_HANDLER_CONSUMED:
3851 ret = NET_RX_SUCCESS;
3853 case RX_HANDLER_ANOTHER:
3855 case RX_HANDLER_EXACT:
3856 deliver_exact = true;
3857 case RX_HANDLER_PASS:
3864 if (unlikely(skb_vlan_tag_present(skb))) {
3865 if (skb_vlan_tag_get_id(skb))
3866 skb->pkt_type = PACKET_OTHERHOST;
3867 /* Note: we might in the future use prio bits
3868 * and set skb->priority like in vlan_do_receive()
3869 * For the time being, just ignore Priority Code Point
3874 type = skb->protocol;
3876 /* deliver only exact match when indicated */
3877 if (likely(!deliver_exact)) {
3878 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3879 &ptype_base[ntohs(type) &
3883 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3884 &orig_dev->ptype_specific);
3886 if (unlikely(skb->dev != orig_dev)) {
3887 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3888 &skb->dev->ptype_specific);
3892 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3895 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3898 atomic_long_inc(&skb->dev->rx_dropped);
3900 /* Jamal, now you will not able to escape explaining
3901 * me how you were going to use this. :-)
3911 static int __netif_receive_skb(struct sk_buff *skb)
3915 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3916 unsigned long pflags = current->flags;
3919 * PFMEMALLOC skbs are special, they should
3920 * - be delivered to SOCK_MEMALLOC sockets only
3921 * - stay away from userspace
3922 * - have bounded memory usage
3924 * Use PF_MEMALLOC as this saves us from propagating the allocation
3925 * context down to all allocation sites.
3927 current->flags |= PF_MEMALLOC;
3928 ret = __netif_receive_skb_core(skb, true);
3929 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3931 ret = __netif_receive_skb_core(skb, false);
3936 static int netif_receive_skb_internal(struct sk_buff *skb)
3938 net_timestamp_check(netdev_tstamp_prequeue, skb);
3940 if (skb_defer_rx_timestamp(skb))
3941 return NET_RX_SUCCESS;
3944 if (static_key_false(&rps_needed)) {
3945 struct rps_dev_flow voidflow, *rflow = &voidflow;
3950 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3953 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3960 return __netif_receive_skb(skb);
3964 * netif_receive_skb - process receive buffer from network
3965 * @skb: buffer to process
3967 * netif_receive_skb() is the main receive data processing function.
3968 * It always succeeds. The buffer may be dropped during processing
3969 * for congestion control or by the protocol layers.
3971 * This function may only be called from softirq context and interrupts
3972 * should be enabled.
3974 * Return values (usually ignored):
3975 * NET_RX_SUCCESS: no congestion
3976 * NET_RX_DROP: packet was dropped
3978 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb)
3980 trace_netif_receive_skb_entry(skb);
3982 return netif_receive_skb_internal(skb);
3984 EXPORT_SYMBOL(netif_receive_skb_sk);
3986 /* Network device is going away, flush any packets still pending
3987 * Called with irqs disabled.
3989 static void flush_backlog(void *arg)
3991 struct net_device *dev = arg;
3992 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3993 struct sk_buff *skb, *tmp;
3996 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3997 if (skb->dev == dev) {
3998 __skb_unlink(skb, &sd->input_pkt_queue);
4000 input_queue_head_incr(sd);
4005 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4006 if (skb->dev == dev) {
4007 __skb_unlink(skb, &sd->process_queue);
4009 input_queue_head_incr(sd);
4014 static int napi_gro_complete(struct sk_buff *skb)
4016 struct packet_offload *ptype;
4017 __be16 type = skb->protocol;
4018 struct list_head *head = &offload_base;
4021 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4023 if (NAPI_GRO_CB(skb)->count == 1) {
4024 skb_shinfo(skb)->gso_size = 0;
4029 list_for_each_entry_rcu(ptype, head, list) {
4030 if (ptype->type != type || !ptype->callbacks.gro_complete)
4033 err = ptype->callbacks.gro_complete(skb, 0);
4039 WARN_ON(&ptype->list == head);
4041 return NET_RX_SUCCESS;
4045 return netif_receive_skb_internal(skb);
4048 /* napi->gro_list contains packets ordered by age.
4049 * youngest packets at the head of it.
4050 * Complete skbs in reverse order to reduce latencies.
4052 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4054 struct sk_buff *skb, *prev = NULL;
4056 /* scan list and build reverse chain */
4057 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4062 for (skb = prev; skb; skb = prev) {
4065 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4069 napi_gro_complete(skb);
4073 napi->gro_list = NULL;
4075 EXPORT_SYMBOL(napi_gro_flush);
4077 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4080 unsigned int maclen = skb->dev->hard_header_len;
4081 u32 hash = skb_get_hash_raw(skb);
4083 for (p = napi->gro_list; p; p = p->next) {
4084 unsigned long diffs;
4086 NAPI_GRO_CB(p)->flush = 0;
4088 if (hash != skb_get_hash_raw(p)) {
4089 NAPI_GRO_CB(p)->same_flow = 0;
4093 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4094 diffs |= p->vlan_tci ^ skb->vlan_tci;
4095 if (maclen == ETH_HLEN)
4096 diffs |= compare_ether_header(skb_mac_header(p),
4097 skb_mac_header(skb));
4099 diffs = memcmp(skb_mac_header(p),
4100 skb_mac_header(skb),
4102 NAPI_GRO_CB(p)->same_flow = !diffs;
4106 static void skb_gro_reset_offset(struct sk_buff *skb)
4108 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4109 const skb_frag_t *frag0 = &pinfo->frags[0];
4111 NAPI_GRO_CB(skb)->data_offset = 0;
4112 NAPI_GRO_CB(skb)->frag0 = NULL;
4113 NAPI_GRO_CB(skb)->frag0_len = 0;
4115 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4117 !PageHighMem(skb_frag_page(frag0))) {
4118 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4119 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4123 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4125 struct skb_shared_info *pinfo = skb_shinfo(skb);
4127 BUG_ON(skb->end - skb->tail < grow);
4129 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4131 skb->data_len -= grow;
4134 pinfo->frags[0].page_offset += grow;
4135 skb_frag_size_sub(&pinfo->frags[0], grow);
4137 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4138 skb_frag_unref(skb, 0);
4139 memmove(pinfo->frags, pinfo->frags + 1,
4140 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4144 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4146 struct sk_buff **pp = NULL;
4147 struct packet_offload *ptype;
4148 __be16 type = skb->protocol;
4149 struct list_head *head = &offload_base;
4151 enum gro_result ret;
4154 if (!(skb->dev->features & NETIF_F_GRO))
4157 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4160 gro_list_prepare(napi, skb);
4163 list_for_each_entry_rcu(ptype, head, list) {
4164 if (ptype->type != type || !ptype->callbacks.gro_receive)
4167 skb_set_network_header(skb, skb_gro_offset(skb));
4168 skb_reset_mac_len(skb);
4169 NAPI_GRO_CB(skb)->same_flow = 0;
4170 NAPI_GRO_CB(skb)->flush = 0;
4171 NAPI_GRO_CB(skb)->free = 0;
4172 NAPI_GRO_CB(skb)->udp_mark = 0;
4173 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4175 /* Setup for GRO checksum validation */
4176 switch (skb->ip_summed) {
4177 case CHECKSUM_COMPLETE:
4178 NAPI_GRO_CB(skb)->csum = skb->csum;
4179 NAPI_GRO_CB(skb)->csum_valid = 1;
4180 NAPI_GRO_CB(skb)->csum_cnt = 0;
4182 case CHECKSUM_UNNECESSARY:
4183 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4184 NAPI_GRO_CB(skb)->csum_valid = 0;
4187 NAPI_GRO_CB(skb)->csum_cnt = 0;
4188 NAPI_GRO_CB(skb)->csum_valid = 0;
4191 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4196 if (&ptype->list == head)
4199 same_flow = NAPI_GRO_CB(skb)->same_flow;
4200 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4203 struct sk_buff *nskb = *pp;
4207 napi_gro_complete(nskb);
4214 if (NAPI_GRO_CB(skb)->flush)
4217 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4218 struct sk_buff *nskb = napi->gro_list;
4220 /* locate the end of the list to select the 'oldest' flow */
4221 while (nskb->next) {
4227 napi_gro_complete(nskb);
4231 NAPI_GRO_CB(skb)->count = 1;
4232 NAPI_GRO_CB(skb)->age = jiffies;
4233 NAPI_GRO_CB(skb)->last = skb;
4234 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4235 skb->next = napi->gro_list;
4236 napi->gro_list = skb;
4240 grow = skb_gro_offset(skb) - skb_headlen(skb);
4242 gro_pull_from_frag0(skb, grow);
4251 struct packet_offload *gro_find_receive_by_type(__be16 type)
4253 struct list_head *offload_head = &offload_base;
4254 struct packet_offload *ptype;
4256 list_for_each_entry_rcu(ptype, offload_head, list) {
4257 if (ptype->type != type || !ptype->callbacks.gro_receive)
4263 EXPORT_SYMBOL(gro_find_receive_by_type);
4265 struct packet_offload *gro_find_complete_by_type(__be16 type)
4267 struct list_head *offload_head = &offload_base;
4268 struct packet_offload *ptype;
4270 list_for_each_entry_rcu(ptype, offload_head, list) {
4271 if (ptype->type != type || !ptype->callbacks.gro_complete)
4277 EXPORT_SYMBOL(gro_find_complete_by_type);
4279 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4283 if (netif_receive_skb_internal(skb))
4291 case GRO_MERGED_FREE:
4292 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4293 kmem_cache_free(skbuff_head_cache, skb);
4306 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4308 trace_napi_gro_receive_entry(skb);
4310 skb_gro_reset_offset(skb);
4312 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4314 EXPORT_SYMBOL(napi_gro_receive);
4316 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4318 if (unlikely(skb->pfmemalloc)) {
4322 __skb_pull(skb, skb_headlen(skb));
4323 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4324 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4326 skb->dev = napi->dev;
4328 skb->encapsulation = 0;
4329 skb_shinfo(skb)->gso_type = 0;
4330 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4335 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4337 struct sk_buff *skb = napi->skb;
4340 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4345 EXPORT_SYMBOL(napi_get_frags);
4347 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4348 struct sk_buff *skb,
4354 __skb_push(skb, ETH_HLEN);
4355 skb->protocol = eth_type_trans(skb, skb->dev);
4356 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4361 case GRO_MERGED_FREE:
4362 napi_reuse_skb(napi, skb);
4372 /* Upper GRO stack assumes network header starts at gro_offset=0
4373 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4374 * We copy ethernet header into skb->data to have a common layout.
4376 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4378 struct sk_buff *skb = napi->skb;
4379 const struct ethhdr *eth;
4380 unsigned int hlen = sizeof(*eth);
4384 skb_reset_mac_header(skb);
4385 skb_gro_reset_offset(skb);
4387 eth = skb_gro_header_fast(skb, 0);
4388 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4389 eth = skb_gro_header_slow(skb, hlen, 0);
4390 if (unlikely(!eth)) {
4391 napi_reuse_skb(napi, skb);
4395 gro_pull_from_frag0(skb, hlen);
4396 NAPI_GRO_CB(skb)->frag0 += hlen;
4397 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4399 __skb_pull(skb, hlen);
4402 * This works because the only protocols we care about don't require
4404 * We'll fix it up properly in napi_frags_finish()
4406 skb->protocol = eth->h_proto;
4411 gro_result_t napi_gro_frags(struct napi_struct *napi)
4413 struct sk_buff *skb = napi_frags_skb(napi);
4418 trace_napi_gro_frags_entry(skb);
4420 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4422 EXPORT_SYMBOL(napi_gro_frags);
4424 /* Compute the checksum from gro_offset and return the folded value
4425 * after adding in any pseudo checksum.
4427 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4432 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4434 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4435 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4437 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4438 !skb->csum_complete_sw)
4439 netdev_rx_csum_fault(skb->dev);
4442 NAPI_GRO_CB(skb)->csum = wsum;
4443 NAPI_GRO_CB(skb)->csum_valid = 1;
4447 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4450 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4451 * Note: called with local irq disabled, but exits with local irq enabled.
4453 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4456 struct softnet_data *remsd = sd->rps_ipi_list;
4459 sd->rps_ipi_list = NULL;
4463 /* Send pending IPI's to kick RPS processing on remote cpus. */
4465 struct softnet_data *next = remsd->rps_ipi_next;
4467 if (cpu_online(remsd->cpu))
4468 smp_call_function_single_async(remsd->cpu,
4477 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4480 return sd->rps_ipi_list != NULL;
4486 static int process_backlog(struct napi_struct *napi, int quota)
4489 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4491 /* Check if we have pending ipi, its better to send them now,
4492 * not waiting net_rx_action() end.
4494 if (sd_has_rps_ipi_waiting(sd)) {
4495 local_irq_disable();
4496 net_rps_action_and_irq_enable(sd);
4499 napi->weight = weight_p;
4500 local_irq_disable();
4502 struct sk_buff *skb;
4504 while ((skb = __skb_dequeue(&sd->process_queue))) {
4506 __netif_receive_skb(skb);
4507 local_irq_disable();
4508 input_queue_head_incr(sd);
4509 if (++work >= quota) {
4516 if (skb_queue_empty(&sd->input_pkt_queue)) {
4518 * Inline a custom version of __napi_complete().
4519 * only current cpu owns and manipulates this napi,
4520 * and NAPI_STATE_SCHED is the only possible flag set
4522 * We can use a plain write instead of clear_bit(),
4523 * and we dont need an smp_mb() memory barrier.
4531 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4532 &sd->process_queue);
4541 * __napi_schedule - schedule for receive
4542 * @n: entry to schedule
4544 * The entry's receive function will be scheduled to run.
4545 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4547 void __napi_schedule(struct napi_struct *n)
4549 unsigned long flags;
4551 local_irq_save(flags);
4552 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4553 local_irq_restore(flags);
4555 EXPORT_SYMBOL(__napi_schedule);
4558 * __napi_schedule_irqoff - schedule for receive
4559 * @n: entry to schedule
4561 * Variant of __napi_schedule() assuming hard irqs are masked
4563 void __napi_schedule_irqoff(struct napi_struct *n)
4565 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4567 EXPORT_SYMBOL(__napi_schedule_irqoff);
4569 void __napi_complete(struct napi_struct *n)
4571 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4573 list_del_init(&n->poll_list);
4574 smp_mb__before_atomic();
4575 clear_bit(NAPI_STATE_SCHED, &n->state);
4577 EXPORT_SYMBOL(__napi_complete);
4579 void napi_complete_done(struct napi_struct *n, int work_done)
4581 unsigned long flags;
4584 * don't let napi dequeue from the cpu poll list
4585 * just in case its running on a different cpu
4587 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4591 unsigned long timeout = 0;
4594 timeout = n->dev->gro_flush_timeout;
4597 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4598 HRTIMER_MODE_REL_PINNED);
4600 napi_gro_flush(n, false);
4602 if (likely(list_empty(&n->poll_list))) {
4603 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4605 /* If n->poll_list is not empty, we need to mask irqs */
4606 local_irq_save(flags);
4608 local_irq_restore(flags);
4611 EXPORT_SYMBOL(napi_complete_done);
4613 /* must be called under rcu_read_lock(), as we dont take a reference */
4614 struct napi_struct *napi_by_id(unsigned int napi_id)
4616 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4617 struct napi_struct *napi;
4619 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4620 if (napi->napi_id == napi_id)
4625 EXPORT_SYMBOL_GPL(napi_by_id);
4627 void napi_hash_add(struct napi_struct *napi)
4629 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4631 spin_lock(&napi_hash_lock);
4633 /* 0 is not a valid id, we also skip an id that is taken
4634 * we expect both events to be extremely rare
4637 while (!napi->napi_id) {
4638 napi->napi_id = ++napi_gen_id;
4639 if (napi_by_id(napi->napi_id))
4643 hlist_add_head_rcu(&napi->napi_hash_node,
4644 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4646 spin_unlock(&napi_hash_lock);
4649 EXPORT_SYMBOL_GPL(napi_hash_add);
4651 /* Warning : caller is responsible to make sure rcu grace period
4652 * is respected before freeing memory containing @napi
4654 void napi_hash_del(struct napi_struct *napi)
4656 spin_lock(&napi_hash_lock);
4658 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4659 hlist_del_rcu(&napi->napi_hash_node);
4661 spin_unlock(&napi_hash_lock);
4663 EXPORT_SYMBOL_GPL(napi_hash_del);
4665 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4667 struct napi_struct *napi;
4669 napi = container_of(timer, struct napi_struct, timer);
4671 napi_schedule(napi);
4673 return HRTIMER_NORESTART;
4676 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4677 int (*poll)(struct napi_struct *, int), int weight)
4679 INIT_LIST_HEAD(&napi->poll_list);
4680 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4681 napi->timer.function = napi_watchdog;
4682 napi->gro_count = 0;
4683 napi->gro_list = NULL;
4686 if (weight > NAPI_POLL_WEIGHT)
4687 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4689 napi->weight = weight;
4690 list_add(&napi->dev_list, &dev->napi_list);
4692 #ifdef CONFIG_NETPOLL
4693 spin_lock_init(&napi->poll_lock);
4694 napi->poll_owner = -1;
4696 set_bit(NAPI_STATE_SCHED, &napi->state);
4698 EXPORT_SYMBOL(netif_napi_add);
4700 void napi_disable(struct napi_struct *n)
4703 set_bit(NAPI_STATE_DISABLE, &n->state);
4705 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4708 hrtimer_cancel(&n->timer);
4710 clear_bit(NAPI_STATE_DISABLE, &n->state);
4712 EXPORT_SYMBOL(napi_disable);
4714 void netif_napi_del(struct napi_struct *napi)
4716 list_del_init(&napi->dev_list);
4717 napi_free_frags(napi);
4719 kfree_skb_list(napi->gro_list);
4720 napi->gro_list = NULL;
4721 napi->gro_count = 0;
4723 EXPORT_SYMBOL(netif_napi_del);
4725 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4730 list_del_init(&n->poll_list);
4732 have = netpoll_poll_lock(n);
4736 /* This NAPI_STATE_SCHED test is for avoiding a race
4737 * with netpoll's poll_napi(). Only the entity which
4738 * obtains the lock and sees NAPI_STATE_SCHED set will
4739 * actually make the ->poll() call. Therefore we avoid
4740 * accidentally calling ->poll() when NAPI is not scheduled.
4743 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4744 work = n->poll(n, weight);
4748 WARN_ON_ONCE(work > weight);
4750 if (likely(work < weight))
4753 /* Drivers must not modify the NAPI state if they
4754 * consume the entire weight. In such cases this code
4755 * still "owns" the NAPI instance and therefore can
4756 * move the instance around on the list at-will.
4758 if (unlikely(napi_disable_pending(n))) {
4764 /* flush too old packets
4765 * If HZ < 1000, flush all packets.
4767 napi_gro_flush(n, HZ >= 1000);
4770 /* Some drivers may have called napi_schedule
4771 * prior to exhausting their budget.
4773 if (unlikely(!list_empty(&n->poll_list))) {
4774 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4775 n->dev ? n->dev->name : "backlog");
4779 list_add_tail(&n->poll_list, repoll);
4782 netpoll_poll_unlock(have);
4787 static void net_rx_action(struct softirq_action *h)
4789 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4790 unsigned long time_limit = jiffies + 2;
4791 int budget = netdev_budget;
4795 local_irq_disable();
4796 list_splice_init(&sd->poll_list, &list);
4800 struct napi_struct *n;
4802 if (list_empty(&list)) {
4803 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4808 n = list_first_entry(&list, struct napi_struct, poll_list);
4809 budget -= napi_poll(n, &repoll);
4811 /* If softirq window is exhausted then punt.
4812 * Allow this to run for 2 jiffies since which will allow
4813 * an average latency of 1.5/HZ.
4815 if (unlikely(budget <= 0 ||
4816 time_after_eq(jiffies, time_limit))) {
4822 local_irq_disable();
4824 list_splice_tail_init(&sd->poll_list, &list);
4825 list_splice_tail(&repoll, &list);
4826 list_splice(&list, &sd->poll_list);
4827 if (!list_empty(&sd->poll_list))
4828 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4830 net_rps_action_and_irq_enable(sd);
4833 struct netdev_adjacent {
4834 struct net_device *dev;
4836 /* upper master flag, there can only be one master device per list */
4839 /* counter for the number of times this device was added to us */
4842 /* private field for the users */
4845 struct list_head list;
4846 struct rcu_head rcu;
4849 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4850 struct net_device *adj_dev,
4851 struct list_head *adj_list)
4853 struct netdev_adjacent *adj;
4855 list_for_each_entry(adj, adj_list, list) {
4856 if (adj->dev == adj_dev)
4863 * netdev_has_upper_dev - Check if device is linked to an upper device
4865 * @upper_dev: upper device to check
4867 * Find out if a device is linked to specified upper device and return true
4868 * in case it is. Note that this checks only immediate upper device,
4869 * not through a complete stack of devices. The caller must hold the RTNL lock.
4871 bool netdev_has_upper_dev(struct net_device *dev,
4872 struct net_device *upper_dev)
4876 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4878 EXPORT_SYMBOL(netdev_has_upper_dev);
4881 * netdev_has_any_upper_dev - Check if device is linked to some device
4884 * Find out if a device is linked to an upper device and return true in case
4885 * it is. The caller must hold the RTNL lock.
4887 static bool netdev_has_any_upper_dev(struct net_device *dev)
4891 return !list_empty(&dev->all_adj_list.upper);
4895 * netdev_master_upper_dev_get - Get master upper device
4898 * Find a master upper device and return pointer to it or NULL in case
4899 * it's not there. The caller must hold the RTNL lock.
4901 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4903 struct netdev_adjacent *upper;
4907 if (list_empty(&dev->adj_list.upper))
4910 upper = list_first_entry(&dev->adj_list.upper,
4911 struct netdev_adjacent, list);
4912 if (likely(upper->master))
4916 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4918 void *netdev_adjacent_get_private(struct list_head *adj_list)
4920 struct netdev_adjacent *adj;
4922 adj = list_entry(adj_list, struct netdev_adjacent, list);
4924 return adj->private;
4926 EXPORT_SYMBOL(netdev_adjacent_get_private);
4929 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4931 * @iter: list_head ** of the current position
4933 * Gets the next device from the dev's upper list, starting from iter
4934 * position. The caller must hold RCU read lock.
4936 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4937 struct list_head **iter)
4939 struct netdev_adjacent *upper;
4941 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4943 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4945 if (&upper->list == &dev->adj_list.upper)
4948 *iter = &upper->list;
4952 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4955 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4957 * @iter: list_head ** of the current position
4959 * Gets the next device from the dev's upper list, starting from iter
4960 * position. The caller must hold RCU read lock.
4962 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4963 struct list_head **iter)
4965 struct netdev_adjacent *upper;
4967 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4969 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4971 if (&upper->list == &dev->all_adj_list.upper)
4974 *iter = &upper->list;
4978 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4981 * netdev_lower_get_next_private - Get the next ->private from the
4982 * lower neighbour list
4984 * @iter: list_head ** of the current position
4986 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4987 * list, starting from iter position. The caller must hold either hold the
4988 * RTNL lock or its own locking that guarantees that the neighbour lower
4989 * list will remain unchainged.
4991 void *netdev_lower_get_next_private(struct net_device *dev,
4992 struct list_head **iter)
4994 struct netdev_adjacent *lower;
4996 lower = list_entry(*iter, struct netdev_adjacent, list);
4998 if (&lower->list == &dev->adj_list.lower)
5001 *iter = lower->list.next;
5003 return lower->private;
5005 EXPORT_SYMBOL(netdev_lower_get_next_private);
5008 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5009 * lower neighbour list, RCU
5012 * @iter: list_head ** of the current position
5014 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5015 * list, starting from iter position. The caller must hold RCU read lock.
5017 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5018 struct list_head **iter)
5020 struct netdev_adjacent *lower;
5022 WARN_ON_ONCE(!rcu_read_lock_held());
5024 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5026 if (&lower->list == &dev->adj_list.lower)
5029 *iter = &lower->list;
5031 return lower->private;
5033 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5036 * netdev_lower_get_next - Get the next device from the lower neighbour
5039 * @iter: list_head ** of the current position
5041 * Gets the next netdev_adjacent from the dev's lower neighbour
5042 * list, starting from iter position. The caller must hold RTNL lock or
5043 * its own locking that guarantees that the neighbour lower
5044 * list will remain unchainged.
5046 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5048 struct netdev_adjacent *lower;
5050 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5052 if (&lower->list == &dev->adj_list.lower)
5055 *iter = &lower->list;
5059 EXPORT_SYMBOL(netdev_lower_get_next);
5062 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5063 * lower neighbour list, RCU
5067 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5068 * list. The caller must hold RCU read lock.
5070 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5072 struct netdev_adjacent *lower;
5074 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5075 struct netdev_adjacent, list);
5077 return lower->private;
5080 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5083 * netdev_master_upper_dev_get_rcu - Get master upper device
5086 * Find a master upper device and return pointer to it or NULL in case
5087 * it's not there. The caller must hold the RCU read lock.
5089 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5091 struct netdev_adjacent *upper;
5093 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5094 struct netdev_adjacent, list);
5095 if (upper && likely(upper->master))
5099 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5101 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5102 struct net_device *adj_dev,
5103 struct list_head *dev_list)
5105 char linkname[IFNAMSIZ+7];
5106 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5107 "upper_%s" : "lower_%s", adj_dev->name);
5108 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5111 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5113 struct list_head *dev_list)
5115 char linkname[IFNAMSIZ+7];
5116 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5117 "upper_%s" : "lower_%s", name);
5118 sysfs_remove_link(&(dev->dev.kobj), linkname);
5121 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5122 struct net_device *adj_dev,
5123 struct list_head *dev_list)
5125 return (dev_list == &dev->adj_list.upper ||
5126 dev_list == &dev->adj_list.lower) &&
5127 net_eq(dev_net(dev), dev_net(adj_dev));
5130 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5131 struct net_device *adj_dev,
5132 struct list_head *dev_list,
5133 void *private, bool master)
5135 struct netdev_adjacent *adj;
5138 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5145 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5150 adj->master = master;
5152 adj->private = private;
5155 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5156 adj_dev->name, dev->name, adj_dev->name);
5158 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5159 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5164 /* Ensure that master link is always the first item in list. */
5166 ret = sysfs_create_link(&(dev->dev.kobj),
5167 &(adj_dev->dev.kobj), "master");
5169 goto remove_symlinks;
5171 list_add_rcu(&adj->list, dev_list);
5173 list_add_tail_rcu(&adj->list, dev_list);
5179 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5180 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5188 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5189 struct net_device *adj_dev,
5190 struct list_head *dev_list)
5192 struct netdev_adjacent *adj;
5194 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5197 pr_err("tried to remove device %s from %s\n",
5198 dev->name, adj_dev->name);
5202 if (adj->ref_nr > 1) {
5203 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5210 sysfs_remove_link(&(dev->dev.kobj), "master");
5212 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5213 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5215 list_del_rcu(&adj->list);
5216 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5217 adj_dev->name, dev->name, adj_dev->name);
5219 kfree_rcu(adj, rcu);
5222 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5223 struct net_device *upper_dev,
5224 struct list_head *up_list,
5225 struct list_head *down_list,
5226 void *private, bool master)
5230 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5235 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5238 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5245 static int __netdev_adjacent_dev_link(struct net_device *dev,
5246 struct net_device *upper_dev)
5248 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5249 &dev->all_adj_list.upper,
5250 &upper_dev->all_adj_list.lower,
5254 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5255 struct net_device *upper_dev,
5256 struct list_head *up_list,
5257 struct list_head *down_list)
5259 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5260 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5263 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5264 struct net_device *upper_dev)
5266 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5267 &dev->all_adj_list.upper,
5268 &upper_dev->all_adj_list.lower);
5271 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5272 struct net_device *upper_dev,
5273 void *private, bool master)
5275 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5280 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5281 &dev->adj_list.upper,
5282 &upper_dev->adj_list.lower,
5285 __netdev_adjacent_dev_unlink(dev, upper_dev);
5292 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5293 struct net_device *upper_dev)
5295 __netdev_adjacent_dev_unlink(dev, upper_dev);
5296 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5297 &dev->adj_list.upper,
5298 &upper_dev->adj_list.lower);
5301 static int __netdev_upper_dev_link(struct net_device *dev,
5302 struct net_device *upper_dev, bool master,
5305 struct netdev_adjacent *i, *j, *to_i, *to_j;
5310 if (dev == upper_dev)
5313 /* To prevent loops, check if dev is not upper device to upper_dev. */
5314 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5317 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper))
5320 if (master && netdev_master_upper_dev_get(dev))
5323 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5328 /* Now that we linked these devs, make all the upper_dev's
5329 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5330 * versa, and don't forget the devices itself. All of these
5331 * links are non-neighbours.
5333 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5334 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5335 pr_debug("Interlinking %s with %s, non-neighbour\n",
5336 i->dev->name, j->dev->name);
5337 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5343 /* add dev to every upper_dev's upper device */
5344 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5345 pr_debug("linking %s's upper device %s with %s\n",
5346 upper_dev->name, i->dev->name, dev->name);
5347 ret = __netdev_adjacent_dev_link(dev, i->dev);
5349 goto rollback_upper_mesh;
5352 /* add upper_dev to every dev's lower device */
5353 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5354 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5355 i->dev->name, upper_dev->name);
5356 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5358 goto rollback_lower_mesh;
5361 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5364 rollback_lower_mesh:
5366 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5369 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5374 rollback_upper_mesh:
5376 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5379 __netdev_adjacent_dev_unlink(dev, i->dev);
5387 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5388 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5389 if (i == to_i && j == to_j)
5391 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5397 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5403 * netdev_upper_dev_link - Add a link to the upper device
5405 * @upper_dev: new upper device
5407 * Adds a link to device which is upper to this one. The caller must hold
5408 * the RTNL lock. On a failure a negative errno code is returned.
5409 * On success the reference counts are adjusted and the function
5412 int netdev_upper_dev_link(struct net_device *dev,
5413 struct net_device *upper_dev)
5415 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5417 EXPORT_SYMBOL(netdev_upper_dev_link);
5420 * netdev_master_upper_dev_link - Add a master link to the upper device
5422 * @upper_dev: new upper device
5424 * Adds a link to device which is upper to this one. In this case, only
5425 * one master upper device can be linked, although other non-master devices
5426 * might be linked as well. The caller must hold the RTNL lock.
5427 * On a failure a negative errno code is returned. On success the reference
5428 * counts are adjusted and the function returns zero.
5430 int netdev_master_upper_dev_link(struct net_device *dev,
5431 struct net_device *upper_dev)
5433 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5435 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5437 int netdev_master_upper_dev_link_private(struct net_device *dev,
5438 struct net_device *upper_dev,
5441 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5443 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5446 * netdev_upper_dev_unlink - Removes a link to upper device
5448 * @upper_dev: new upper device
5450 * Removes a link to device which is upper to this one. The caller must hold
5453 void netdev_upper_dev_unlink(struct net_device *dev,
5454 struct net_device *upper_dev)
5456 struct netdev_adjacent *i, *j;
5459 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5461 /* Here is the tricky part. We must remove all dev's lower
5462 * devices from all upper_dev's upper devices and vice
5463 * versa, to maintain the graph relationship.
5465 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5466 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5467 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5469 /* remove also the devices itself from lower/upper device
5472 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5473 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5475 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5476 __netdev_adjacent_dev_unlink(dev, i->dev);
5478 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5480 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5483 * netdev_bonding_info_change - Dispatch event about slave change
5485 * @bonding_info: info to dispatch
5487 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5488 * The caller must hold the RTNL lock.
5490 void netdev_bonding_info_change(struct net_device *dev,
5491 struct netdev_bonding_info *bonding_info)
5493 struct netdev_notifier_bonding_info info;
5495 memcpy(&info.bonding_info, bonding_info,
5496 sizeof(struct netdev_bonding_info));
5497 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5500 EXPORT_SYMBOL(netdev_bonding_info_change);
5502 static void netdev_adjacent_add_links(struct net_device *dev)
5504 struct netdev_adjacent *iter;
5506 struct net *net = dev_net(dev);
5508 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5509 if (!net_eq(net,dev_net(iter->dev)))
5511 netdev_adjacent_sysfs_add(iter->dev, dev,
5512 &iter->dev->adj_list.lower);
5513 netdev_adjacent_sysfs_add(dev, iter->dev,
5514 &dev->adj_list.upper);
5517 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5518 if (!net_eq(net,dev_net(iter->dev)))
5520 netdev_adjacent_sysfs_add(iter->dev, dev,
5521 &iter->dev->adj_list.upper);
5522 netdev_adjacent_sysfs_add(dev, iter->dev,
5523 &dev->adj_list.lower);
5527 static void netdev_adjacent_del_links(struct net_device *dev)
5529 struct netdev_adjacent *iter;
5531 struct net *net = dev_net(dev);
5533 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5534 if (!net_eq(net,dev_net(iter->dev)))
5536 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5537 &iter->dev->adj_list.lower);
5538 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5539 &dev->adj_list.upper);
5542 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5543 if (!net_eq(net,dev_net(iter->dev)))
5545 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5546 &iter->dev->adj_list.upper);
5547 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5548 &dev->adj_list.lower);
5552 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5554 struct netdev_adjacent *iter;
5556 struct net *net = dev_net(dev);
5558 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5559 if (!net_eq(net,dev_net(iter->dev)))
5561 netdev_adjacent_sysfs_del(iter->dev, oldname,
5562 &iter->dev->adj_list.lower);
5563 netdev_adjacent_sysfs_add(iter->dev, dev,
5564 &iter->dev->adj_list.lower);
5567 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5568 if (!net_eq(net,dev_net(iter->dev)))
5570 netdev_adjacent_sysfs_del(iter->dev, oldname,
5571 &iter->dev->adj_list.upper);
5572 netdev_adjacent_sysfs_add(iter->dev, dev,
5573 &iter->dev->adj_list.upper);
5577 void *netdev_lower_dev_get_private(struct net_device *dev,
5578 struct net_device *lower_dev)
5580 struct netdev_adjacent *lower;
5584 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5588 return lower->private;
5590 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5593 int dev_get_nest_level(struct net_device *dev,
5594 bool (*type_check)(struct net_device *dev))
5596 struct net_device *lower = NULL;
5597 struct list_head *iter;
5603 netdev_for_each_lower_dev(dev, lower, iter) {
5604 nest = dev_get_nest_level(lower, type_check);
5605 if (max_nest < nest)
5609 if (type_check(dev))
5614 EXPORT_SYMBOL(dev_get_nest_level);
5616 static void dev_change_rx_flags(struct net_device *dev, int flags)
5618 const struct net_device_ops *ops = dev->netdev_ops;
5620 if (ops->ndo_change_rx_flags)
5621 ops->ndo_change_rx_flags(dev, flags);
5624 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5626 unsigned int old_flags = dev->flags;
5632 dev->flags |= IFF_PROMISC;
5633 dev->promiscuity += inc;
5634 if (dev->promiscuity == 0) {
5637 * If inc causes overflow, untouch promisc and return error.
5640 dev->flags &= ~IFF_PROMISC;
5642 dev->promiscuity -= inc;
5643 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5648 if (dev->flags != old_flags) {
5649 pr_info("device %s %s promiscuous mode\n",
5651 dev->flags & IFF_PROMISC ? "entered" : "left");
5652 if (audit_enabled) {
5653 current_uid_gid(&uid, &gid);
5654 audit_log(current->audit_context, GFP_ATOMIC,
5655 AUDIT_ANOM_PROMISCUOUS,
5656 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5657 dev->name, (dev->flags & IFF_PROMISC),
5658 (old_flags & IFF_PROMISC),
5659 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5660 from_kuid(&init_user_ns, uid),
5661 from_kgid(&init_user_ns, gid),
5662 audit_get_sessionid(current));
5665 dev_change_rx_flags(dev, IFF_PROMISC);
5668 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5673 * dev_set_promiscuity - update promiscuity count on a device
5677 * Add or remove promiscuity from a device. While the count in the device
5678 * remains above zero the interface remains promiscuous. Once it hits zero
5679 * the device reverts back to normal filtering operation. A negative inc
5680 * value is used to drop promiscuity on the device.
5681 * Return 0 if successful or a negative errno code on error.
5683 int dev_set_promiscuity(struct net_device *dev, int inc)
5685 unsigned int old_flags = dev->flags;
5688 err = __dev_set_promiscuity(dev, inc, true);
5691 if (dev->flags != old_flags)
5692 dev_set_rx_mode(dev);
5695 EXPORT_SYMBOL(dev_set_promiscuity);
5697 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5699 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5703 dev->flags |= IFF_ALLMULTI;
5704 dev->allmulti += inc;
5705 if (dev->allmulti == 0) {
5708 * If inc causes overflow, untouch allmulti and return error.
5711 dev->flags &= ~IFF_ALLMULTI;
5713 dev->allmulti -= inc;
5714 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5719 if (dev->flags ^ old_flags) {
5720 dev_change_rx_flags(dev, IFF_ALLMULTI);
5721 dev_set_rx_mode(dev);
5723 __dev_notify_flags(dev, old_flags,
5724 dev->gflags ^ old_gflags);
5730 * dev_set_allmulti - update allmulti count on a device
5734 * Add or remove reception of all multicast frames to a device. While the
5735 * count in the device remains above zero the interface remains listening
5736 * to all interfaces. Once it hits zero the device reverts back to normal
5737 * filtering operation. A negative @inc value is used to drop the counter
5738 * when releasing a resource needing all multicasts.
5739 * Return 0 if successful or a negative errno code on error.
5742 int dev_set_allmulti(struct net_device *dev, int inc)
5744 return __dev_set_allmulti(dev, inc, true);
5746 EXPORT_SYMBOL(dev_set_allmulti);
5749 * Upload unicast and multicast address lists to device and
5750 * configure RX filtering. When the device doesn't support unicast
5751 * filtering it is put in promiscuous mode while unicast addresses
5754 void __dev_set_rx_mode(struct net_device *dev)
5756 const struct net_device_ops *ops = dev->netdev_ops;
5758 /* dev_open will call this function so the list will stay sane. */
5759 if (!(dev->flags&IFF_UP))
5762 if (!netif_device_present(dev))
5765 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5766 /* Unicast addresses changes may only happen under the rtnl,
5767 * therefore calling __dev_set_promiscuity here is safe.
5769 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5770 __dev_set_promiscuity(dev, 1, false);
5771 dev->uc_promisc = true;
5772 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5773 __dev_set_promiscuity(dev, -1, false);
5774 dev->uc_promisc = false;
5778 if (ops->ndo_set_rx_mode)
5779 ops->ndo_set_rx_mode(dev);
5782 void dev_set_rx_mode(struct net_device *dev)
5784 netif_addr_lock_bh(dev);
5785 __dev_set_rx_mode(dev);
5786 netif_addr_unlock_bh(dev);
5790 * dev_get_flags - get flags reported to userspace
5793 * Get the combination of flag bits exported through APIs to userspace.
5795 unsigned int dev_get_flags(const struct net_device *dev)
5799 flags = (dev->flags & ~(IFF_PROMISC |
5804 (dev->gflags & (IFF_PROMISC |
5807 if (netif_running(dev)) {
5808 if (netif_oper_up(dev))
5809 flags |= IFF_RUNNING;
5810 if (netif_carrier_ok(dev))
5811 flags |= IFF_LOWER_UP;
5812 if (netif_dormant(dev))
5813 flags |= IFF_DORMANT;
5818 EXPORT_SYMBOL(dev_get_flags);
5820 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5822 unsigned int old_flags = dev->flags;
5828 * Set the flags on our device.
5831 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5832 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5834 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5838 * Load in the correct multicast list now the flags have changed.
5841 if ((old_flags ^ flags) & IFF_MULTICAST)
5842 dev_change_rx_flags(dev, IFF_MULTICAST);
5844 dev_set_rx_mode(dev);
5847 * Have we downed the interface. We handle IFF_UP ourselves
5848 * according to user attempts to set it, rather than blindly
5853 if ((old_flags ^ flags) & IFF_UP)
5854 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5856 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5857 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5858 unsigned int old_flags = dev->flags;
5860 dev->gflags ^= IFF_PROMISC;
5862 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5863 if (dev->flags != old_flags)
5864 dev_set_rx_mode(dev);
5867 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5868 is important. Some (broken) drivers set IFF_PROMISC, when
5869 IFF_ALLMULTI is requested not asking us and not reporting.
5871 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5872 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5874 dev->gflags ^= IFF_ALLMULTI;
5875 __dev_set_allmulti(dev, inc, false);
5881 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5882 unsigned int gchanges)
5884 unsigned int changes = dev->flags ^ old_flags;
5887 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5889 if (changes & IFF_UP) {
5890 if (dev->flags & IFF_UP)
5891 call_netdevice_notifiers(NETDEV_UP, dev);
5893 call_netdevice_notifiers(NETDEV_DOWN, dev);
5896 if (dev->flags & IFF_UP &&
5897 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5898 struct netdev_notifier_change_info change_info;
5900 change_info.flags_changed = changes;
5901 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5907 * dev_change_flags - change device settings
5909 * @flags: device state flags
5911 * Change settings on device based state flags. The flags are
5912 * in the userspace exported format.
5914 int dev_change_flags(struct net_device *dev, unsigned int flags)
5917 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5919 ret = __dev_change_flags(dev, flags);
5923 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5924 __dev_notify_flags(dev, old_flags, changes);
5927 EXPORT_SYMBOL(dev_change_flags);
5929 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5931 const struct net_device_ops *ops = dev->netdev_ops;
5933 if (ops->ndo_change_mtu)
5934 return ops->ndo_change_mtu(dev, new_mtu);
5941 * dev_set_mtu - Change maximum transfer unit
5943 * @new_mtu: new transfer unit
5945 * Change the maximum transfer size of the network device.
5947 int dev_set_mtu(struct net_device *dev, int new_mtu)
5951 if (new_mtu == dev->mtu)
5954 /* MTU must be positive. */
5958 if (!netif_device_present(dev))
5961 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5962 err = notifier_to_errno(err);
5966 orig_mtu = dev->mtu;
5967 err = __dev_set_mtu(dev, new_mtu);
5970 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5971 err = notifier_to_errno(err);
5973 /* setting mtu back and notifying everyone again,
5974 * so that they have a chance to revert changes.
5976 __dev_set_mtu(dev, orig_mtu);
5977 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5982 EXPORT_SYMBOL(dev_set_mtu);
5985 * dev_set_group - Change group this device belongs to
5987 * @new_group: group this device should belong to
5989 void dev_set_group(struct net_device *dev, int new_group)
5991 dev->group = new_group;
5993 EXPORT_SYMBOL(dev_set_group);
5996 * dev_set_mac_address - Change Media Access Control Address
6000 * Change the hardware (MAC) address of the device
6002 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6004 const struct net_device_ops *ops = dev->netdev_ops;
6007 if (!ops->ndo_set_mac_address)
6009 if (sa->sa_family != dev->type)
6011 if (!netif_device_present(dev))
6013 err = ops->ndo_set_mac_address(dev, sa);
6016 dev->addr_assign_type = NET_ADDR_SET;
6017 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6018 add_device_randomness(dev->dev_addr, dev->addr_len);
6021 EXPORT_SYMBOL(dev_set_mac_address);
6024 * dev_change_carrier - Change device carrier
6026 * @new_carrier: new value
6028 * Change device carrier
6030 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6032 const struct net_device_ops *ops = dev->netdev_ops;
6034 if (!ops->ndo_change_carrier)
6036 if (!netif_device_present(dev))
6038 return ops->ndo_change_carrier(dev, new_carrier);
6040 EXPORT_SYMBOL(dev_change_carrier);
6043 * dev_get_phys_port_id - Get device physical port ID
6047 * Get device physical port ID
6049 int dev_get_phys_port_id(struct net_device *dev,
6050 struct netdev_phys_item_id *ppid)
6052 const struct net_device_ops *ops = dev->netdev_ops;
6054 if (!ops->ndo_get_phys_port_id)
6056 return ops->ndo_get_phys_port_id(dev, ppid);
6058 EXPORT_SYMBOL(dev_get_phys_port_id);
6061 * dev_get_phys_port_name - Get device physical port name
6065 * Get device physical port name
6067 int dev_get_phys_port_name(struct net_device *dev,
6068 char *name, size_t len)
6070 const struct net_device_ops *ops = dev->netdev_ops;
6072 if (!ops->ndo_get_phys_port_name)
6074 return ops->ndo_get_phys_port_name(dev, name, len);
6076 EXPORT_SYMBOL(dev_get_phys_port_name);
6079 * dev_new_index - allocate an ifindex
6080 * @net: the applicable net namespace
6082 * Returns a suitable unique value for a new device interface
6083 * number. The caller must hold the rtnl semaphore or the
6084 * dev_base_lock to be sure it remains unique.
6086 static int dev_new_index(struct net *net)
6088 int ifindex = net->ifindex;
6092 if (!__dev_get_by_index(net, ifindex))
6093 return net->ifindex = ifindex;
6097 /* Delayed registration/unregisteration */
6098 static LIST_HEAD(net_todo_list);
6099 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6101 static void net_set_todo(struct net_device *dev)
6103 list_add_tail(&dev->todo_list, &net_todo_list);
6104 dev_net(dev)->dev_unreg_count++;
6107 static void rollback_registered_many(struct list_head *head)
6109 struct net_device *dev, *tmp;
6110 LIST_HEAD(close_head);
6112 BUG_ON(dev_boot_phase);
6115 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6116 /* Some devices call without registering
6117 * for initialization unwind. Remove those
6118 * devices and proceed with the remaining.
6120 if (dev->reg_state == NETREG_UNINITIALIZED) {
6121 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6125 list_del(&dev->unreg_list);
6128 dev->dismantle = true;
6129 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6132 /* If device is running, close it first. */
6133 list_for_each_entry(dev, head, unreg_list)
6134 list_add_tail(&dev->close_list, &close_head);
6135 dev_close_many(&close_head, true);
6137 list_for_each_entry(dev, head, unreg_list) {
6138 /* And unlink it from device chain. */
6139 unlist_netdevice(dev);
6141 dev->reg_state = NETREG_UNREGISTERING;
6146 list_for_each_entry(dev, head, unreg_list) {
6147 struct sk_buff *skb = NULL;
6149 /* Shutdown queueing discipline. */
6153 /* Notify protocols, that we are about to destroy
6154 this device. They should clean all the things.
6156 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6158 if (!dev->rtnl_link_ops ||
6159 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6160 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6164 * Flush the unicast and multicast chains
6169 if (dev->netdev_ops->ndo_uninit)
6170 dev->netdev_ops->ndo_uninit(dev);
6173 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6175 /* Notifier chain MUST detach us all upper devices. */
6176 WARN_ON(netdev_has_any_upper_dev(dev));
6178 /* Remove entries from kobject tree */
6179 netdev_unregister_kobject(dev);
6181 /* Remove XPS queueing entries */
6182 netif_reset_xps_queues_gt(dev, 0);
6188 list_for_each_entry(dev, head, unreg_list)
6192 static void rollback_registered(struct net_device *dev)
6196 list_add(&dev->unreg_list, &single);
6197 rollback_registered_many(&single);
6201 static netdev_features_t netdev_fix_features(struct net_device *dev,
6202 netdev_features_t features)
6204 /* Fix illegal checksum combinations */
6205 if ((features & NETIF_F_HW_CSUM) &&
6206 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6207 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6208 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6211 /* TSO requires that SG is present as well. */
6212 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6213 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6214 features &= ~NETIF_F_ALL_TSO;
6217 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6218 !(features & NETIF_F_IP_CSUM)) {
6219 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6220 features &= ~NETIF_F_TSO;
6221 features &= ~NETIF_F_TSO_ECN;
6224 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6225 !(features & NETIF_F_IPV6_CSUM)) {
6226 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6227 features &= ~NETIF_F_TSO6;
6230 /* TSO ECN requires that TSO is present as well. */
6231 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6232 features &= ~NETIF_F_TSO_ECN;
6234 /* Software GSO depends on SG. */
6235 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6236 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6237 features &= ~NETIF_F_GSO;
6240 /* UFO needs SG and checksumming */
6241 if (features & NETIF_F_UFO) {
6242 /* maybe split UFO into V4 and V6? */
6243 if (!((features & NETIF_F_GEN_CSUM) ||
6244 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6245 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6247 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6248 features &= ~NETIF_F_UFO;
6251 if (!(features & NETIF_F_SG)) {
6253 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6254 features &= ~NETIF_F_UFO;
6258 #ifdef CONFIG_NET_RX_BUSY_POLL
6259 if (dev->netdev_ops->ndo_busy_poll)
6260 features |= NETIF_F_BUSY_POLL;
6263 features &= ~NETIF_F_BUSY_POLL;
6268 int __netdev_update_features(struct net_device *dev)
6270 netdev_features_t features;
6275 features = netdev_get_wanted_features(dev);
6277 if (dev->netdev_ops->ndo_fix_features)
6278 features = dev->netdev_ops->ndo_fix_features(dev, features);
6280 /* driver might be less strict about feature dependencies */
6281 features = netdev_fix_features(dev, features);
6283 if (dev->features == features)
6286 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6287 &dev->features, &features);
6289 if (dev->netdev_ops->ndo_set_features)
6290 err = dev->netdev_ops->ndo_set_features(dev, features);
6292 if (unlikely(err < 0)) {
6294 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6295 err, &features, &dev->features);
6300 dev->features = features;
6306 * netdev_update_features - recalculate device features
6307 * @dev: the device to check
6309 * Recalculate dev->features set and send notifications if it
6310 * has changed. Should be called after driver or hardware dependent
6311 * conditions might have changed that influence the features.
6313 void netdev_update_features(struct net_device *dev)
6315 if (__netdev_update_features(dev))
6316 netdev_features_change(dev);
6318 EXPORT_SYMBOL(netdev_update_features);
6321 * netdev_change_features - recalculate device features
6322 * @dev: the device to check
6324 * Recalculate dev->features set and send notifications even
6325 * if they have not changed. Should be called instead of
6326 * netdev_update_features() if also dev->vlan_features might
6327 * have changed to allow the changes to be propagated to stacked
6330 void netdev_change_features(struct net_device *dev)
6332 __netdev_update_features(dev);
6333 netdev_features_change(dev);
6335 EXPORT_SYMBOL(netdev_change_features);
6338 * netif_stacked_transfer_operstate - transfer operstate
6339 * @rootdev: the root or lower level device to transfer state from
6340 * @dev: the device to transfer operstate to
6342 * Transfer operational state from root to device. This is normally
6343 * called when a stacking relationship exists between the root
6344 * device and the device(a leaf device).
6346 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6347 struct net_device *dev)
6349 if (rootdev->operstate == IF_OPER_DORMANT)
6350 netif_dormant_on(dev);
6352 netif_dormant_off(dev);
6354 if (netif_carrier_ok(rootdev)) {
6355 if (!netif_carrier_ok(dev))
6356 netif_carrier_on(dev);
6358 if (netif_carrier_ok(dev))
6359 netif_carrier_off(dev);
6362 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6365 static int netif_alloc_rx_queues(struct net_device *dev)
6367 unsigned int i, count = dev->num_rx_queues;
6368 struct netdev_rx_queue *rx;
6369 size_t sz = count * sizeof(*rx);
6373 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6381 for (i = 0; i < count; i++)
6387 static void netdev_init_one_queue(struct net_device *dev,
6388 struct netdev_queue *queue, void *_unused)
6390 /* Initialize queue lock */
6391 spin_lock_init(&queue->_xmit_lock);
6392 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6393 queue->xmit_lock_owner = -1;
6394 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6397 dql_init(&queue->dql, HZ);
6401 static void netif_free_tx_queues(struct net_device *dev)
6406 static int netif_alloc_netdev_queues(struct net_device *dev)
6408 unsigned int count = dev->num_tx_queues;
6409 struct netdev_queue *tx;
6410 size_t sz = count * sizeof(*tx);
6412 BUG_ON(count < 1 || count > 0xffff);
6414 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6422 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6423 spin_lock_init(&dev->tx_global_lock);
6428 void netif_tx_stop_all_queues(struct net_device *dev)
6432 for (i = 0; i < dev->num_tx_queues; i++) {
6433 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6434 netif_tx_stop_queue(txq);
6437 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6440 * register_netdevice - register a network device
6441 * @dev: device to register
6443 * Take a completed network device structure and add it to the kernel
6444 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6445 * chain. 0 is returned on success. A negative errno code is returned
6446 * on a failure to set up the device, or if the name is a duplicate.
6448 * Callers must hold the rtnl semaphore. You may want
6449 * register_netdev() instead of this.
6452 * The locking appears insufficient to guarantee two parallel registers
6453 * will not get the same name.
6456 int register_netdevice(struct net_device *dev)
6459 struct net *net = dev_net(dev);
6461 BUG_ON(dev_boot_phase);
6466 /* When net_device's are persistent, this will be fatal. */
6467 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6470 spin_lock_init(&dev->addr_list_lock);
6471 netdev_set_addr_lockdep_class(dev);
6473 ret = dev_get_valid_name(net, dev, dev->name);
6477 /* Init, if this function is available */
6478 if (dev->netdev_ops->ndo_init) {
6479 ret = dev->netdev_ops->ndo_init(dev);
6487 if (((dev->hw_features | dev->features) &
6488 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6489 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6490 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6491 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6498 dev->ifindex = dev_new_index(net);
6499 else if (__dev_get_by_index(net, dev->ifindex))
6502 /* Transfer changeable features to wanted_features and enable
6503 * software offloads (GSO and GRO).
6505 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6506 dev->features |= NETIF_F_SOFT_FEATURES;
6507 dev->wanted_features = dev->features & dev->hw_features;
6509 if (!(dev->flags & IFF_LOOPBACK)) {
6510 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6513 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6515 dev->vlan_features |= NETIF_F_HIGHDMA;
6517 /* Make NETIF_F_SG inheritable to tunnel devices.
6519 dev->hw_enc_features |= NETIF_F_SG;
6521 /* Make NETIF_F_SG inheritable to MPLS.
6523 dev->mpls_features |= NETIF_F_SG;
6525 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6526 ret = notifier_to_errno(ret);
6530 ret = netdev_register_kobject(dev);
6533 dev->reg_state = NETREG_REGISTERED;
6535 __netdev_update_features(dev);
6538 * Default initial state at registry is that the
6539 * device is present.
6542 set_bit(__LINK_STATE_PRESENT, &dev->state);
6544 linkwatch_init_dev(dev);
6546 dev_init_scheduler(dev);
6548 list_netdevice(dev);
6549 add_device_randomness(dev->dev_addr, dev->addr_len);
6551 /* If the device has permanent device address, driver should
6552 * set dev_addr and also addr_assign_type should be set to
6553 * NET_ADDR_PERM (default value).
6555 if (dev->addr_assign_type == NET_ADDR_PERM)
6556 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6558 /* Notify protocols, that a new device appeared. */
6559 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6560 ret = notifier_to_errno(ret);
6562 rollback_registered(dev);
6563 dev->reg_state = NETREG_UNREGISTERED;
6566 * Prevent userspace races by waiting until the network
6567 * device is fully setup before sending notifications.
6569 if (!dev->rtnl_link_ops ||
6570 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6571 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6577 if (dev->netdev_ops->ndo_uninit)
6578 dev->netdev_ops->ndo_uninit(dev);
6581 EXPORT_SYMBOL(register_netdevice);
6584 * init_dummy_netdev - init a dummy network device for NAPI
6585 * @dev: device to init
6587 * This takes a network device structure and initialize the minimum
6588 * amount of fields so it can be used to schedule NAPI polls without
6589 * registering a full blown interface. This is to be used by drivers
6590 * that need to tie several hardware interfaces to a single NAPI
6591 * poll scheduler due to HW limitations.
6593 int init_dummy_netdev(struct net_device *dev)
6595 /* Clear everything. Note we don't initialize spinlocks
6596 * are they aren't supposed to be taken by any of the
6597 * NAPI code and this dummy netdev is supposed to be
6598 * only ever used for NAPI polls
6600 memset(dev, 0, sizeof(struct net_device));
6602 /* make sure we BUG if trying to hit standard
6603 * register/unregister code path
6605 dev->reg_state = NETREG_DUMMY;
6607 /* NAPI wants this */
6608 INIT_LIST_HEAD(&dev->napi_list);
6610 /* a dummy interface is started by default */
6611 set_bit(__LINK_STATE_PRESENT, &dev->state);
6612 set_bit(__LINK_STATE_START, &dev->state);
6614 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6615 * because users of this 'device' dont need to change
6621 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6625 * register_netdev - register a network device
6626 * @dev: device to register
6628 * Take a completed network device structure and add it to the kernel
6629 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6630 * chain. 0 is returned on success. A negative errno code is returned
6631 * on a failure to set up the device, or if the name is a duplicate.
6633 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6634 * and expands the device name if you passed a format string to
6637 int register_netdev(struct net_device *dev)
6642 err = register_netdevice(dev);
6646 EXPORT_SYMBOL(register_netdev);
6648 int netdev_refcnt_read(const struct net_device *dev)
6652 for_each_possible_cpu(i)
6653 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6656 EXPORT_SYMBOL(netdev_refcnt_read);
6659 * netdev_wait_allrefs - wait until all references are gone.
6660 * @dev: target net_device
6662 * This is called when unregistering network devices.
6664 * Any protocol or device that holds a reference should register
6665 * for netdevice notification, and cleanup and put back the
6666 * reference if they receive an UNREGISTER event.
6667 * We can get stuck here if buggy protocols don't correctly
6670 static void netdev_wait_allrefs(struct net_device *dev)
6672 unsigned long rebroadcast_time, warning_time;
6675 linkwatch_forget_dev(dev);
6677 rebroadcast_time = warning_time = jiffies;
6678 refcnt = netdev_refcnt_read(dev);
6680 while (refcnt != 0) {
6681 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6684 /* Rebroadcast unregister notification */
6685 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6691 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6692 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6694 /* We must not have linkwatch events
6695 * pending on unregister. If this
6696 * happens, we simply run the queue
6697 * unscheduled, resulting in a noop
6700 linkwatch_run_queue();
6705 rebroadcast_time = jiffies;
6710 refcnt = netdev_refcnt_read(dev);
6712 if (time_after(jiffies, warning_time + 10 * HZ)) {
6713 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6715 warning_time = jiffies;
6724 * register_netdevice(x1);
6725 * register_netdevice(x2);
6727 * unregister_netdevice(y1);
6728 * unregister_netdevice(y2);
6734 * We are invoked by rtnl_unlock().
6735 * This allows us to deal with problems:
6736 * 1) We can delete sysfs objects which invoke hotplug
6737 * without deadlocking with linkwatch via keventd.
6738 * 2) Since we run with the RTNL semaphore not held, we can sleep
6739 * safely in order to wait for the netdev refcnt to drop to zero.
6741 * We must not return until all unregister events added during
6742 * the interval the lock was held have been completed.
6744 void netdev_run_todo(void)
6746 struct list_head list;
6748 /* Snapshot list, allow later requests */
6749 list_replace_init(&net_todo_list, &list);
6754 /* Wait for rcu callbacks to finish before next phase */
6755 if (!list_empty(&list))
6758 while (!list_empty(&list)) {
6759 struct net_device *dev
6760 = list_first_entry(&list, struct net_device, todo_list);
6761 list_del(&dev->todo_list);
6764 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6767 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6768 pr_err("network todo '%s' but state %d\n",
6769 dev->name, dev->reg_state);
6774 dev->reg_state = NETREG_UNREGISTERED;
6776 on_each_cpu(flush_backlog, dev, 1);
6778 netdev_wait_allrefs(dev);
6781 BUG_ON(netdev_refcnt_read(dev));
6782 BUG_ON(!list_empty(&dev->ptype_all));
6783 BUG_ON(!list_empty(&dev->ptype_specific));
6784 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6785 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6786 WARN_ON(dev->dn_ptr);
6788 if (dev->destructor)
6789 dev->destructor(dev);
6791 /* Report a network device has been unregistered */
6793 dev_net(dev)->dev_unreg_count--;
6795 wake_up(&netdev_unregistering_wq);
6797 /* Free network device */
6798 kobject_put(&dev->dev.kobj);
6802 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6803 * fields in the same order, with only the type differing.
6805 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6806 const struct net_device_stats *netdev_stats)
6808 #if BITS_PER_LONG == 64
6809 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6810 memcpy(stats64, netdev_stats, sizeof(*stats64));
6812 size_t i, n = sizeof(*stats64) / sizeof(u64);
6813 const unsigned long *src = (const unsigned long *)netdev_stats;
6814 u64 *dst = (u64 *)stats64;
6816 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6817 sizeof(*stats64) / sizeof(u64));
6818 for (i = 0; i < n; i++)
6822 EXPORT_SYMBOL(netdev_stats_to_stats64);
6825 * dev_get_stats - get network device statistics
6826 * @dev: device to get statistics from
6827 * @storage: place to store stats
6829 * Get network statistics from device. Return @storage.
6830 * The device driver may provide its own method by setting
6831 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6832 * otherwise the internal statistics structure is used.
6834 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6835 struct rtnl_link_stats64 *storage)
6837 const struct net_device_ops *ops = dev->netdev_ops;
6839 if (ops->ndo_get_stats64) {
6840 memset(storage, 0, sizeof(*storage));
6841 ops->ndo_get_stats64(dev, storage);
6842 } else if (ops->ndo_get_stats) {
6843 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6845 netdev_stats_to_stats64(storage, &dev->stats);
6847 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6848 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6851 EXPORT_SYMBOL(dev_get_stats);
6853 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6855 struct netdev_queue *queue = dev_ingress_queue(dev);
6857 #ifdef CONFIG_NET_CLS_ACT
6860 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6863 netdev_init_one_queue(dev, queue, NULL);
6864 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6865 queue->qdisc_sleeping = &noop_qdisc;
6866 rcu_assign_pointer(dev->ingress_queue, queue);
6871 static const struct ethtool_ops default_ethtool_ops;
6873 void netdev_set_default_ethtool_ops(struct net_device *dev,
6874 const struct ethtool_ops *ops)
6876 if (dev->ethtool_ops == &default_ethtool_ops)
6877 dev->ethtool_ops = ops;
6879 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6881 void netdev_freemem(struct net_device *dev)
6883 char *addr = (char *)dev - dev->padded;
6889 * alloc_netdev_mqs - allocate network device
6890 * @sizeof_priv: size of private data to allocate space for
6891 * @name: device name format string
6892 * @name_assign_type: origin of device name
6893 * @setup: callback to initialize device
6894 * @txqs: the number of TX subqueues to allocate
6895 * @rxqs: the number of RX subqueues to allocate
6897 * Allocates a struct net_device with private data area for driver use
6898 * and performs basic initialization. Also allocates subqueue structs
6899 * for each queue on the device.
6901 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6902 unsigned char name_assign_type,
6903 void (*setup)(struct net_device *),
6904 unsigned int txqs, unsigned int rxqs)
6906 struct net_device *dev;
6908 struct net_device *p;
6910 BUG_ON(strlen(name) >= sizeof(dev->name));
6913 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6919 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6924 alloc_size = sizeof(struct net_device);
6926 /* ensure 32-byte alignment of private area */
6927 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6928 alloc_size += sizeof_priv;
6930 /* ensure 32-byte alignment of whole construct */
6931 alloc_size += NETDEV_ALIGN - 1;
6933 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6935 p = vzalloc(alloc_size);
6939 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6940 dev->padded = (char *)dev - (char *)p;
6942 dev->pcpu_refcnt = alloc_percpu(int);
6943 if (!dev->pcpu_refcnt)
6946 if (dev_addr_init(dev))
6952 dev_net_set(dev, &init_net);
6954 dev->gso_max_size = GSO_MAX_SIZE;
6955 dev->gso_max_segs = GSO_MAX_SEGS;
6956 dev->gso_min_segs = 0;
6958 INIT_LIST_HEAD(&dev->napi_list);
6959 INIT_LIST_HEAD(&dev->unreg_list);
6960 INIT_LIST_HEAD(&dev->close_list);
6961 INIT_LIST_HEAD(&dev->link_watch_list);
6962 INIT_LIST_HEAD(&dev->adj_list.upper);
6963 INIT_LIST_HEAD(&dev->adj_list.lower);
6964 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6965 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6966 INIT_LIST_HEAD(&dev->ptype_all);
6967 INIT_LIST_HEAD(&dev->ptype_specific);
6968 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6971 dev->num_tx_queues = txqs;
6972 dev->real_num_tx_queues = txqs;
6973 if (netif_alloc_netdev_queues(dev))
6977 dev->num_rx_queues = rxqs;
6978 dev->real_num_rx_queues = rxqs;
6979 if (netif_alloc_rx_queues(dev))
6983 strcpy(dev->name, name);
6984 dev->name_assign_type = name_assign_type;
6985 dev->group = INIT_NETDEV_GROUP;
6986 if (!dev->ethtool_ops)
6987 dev->ethtool_ops = &default_ethtool_ops;
6989 nf_hook_ingress_init(dev);
6998 free_percpu(dev->pcpu_refcnt);
7000 netdev_freemem(dev);
7003 EXPORT_SYMBOL(alloc_netdev_mqs);
7006 * free_netdev - free network device
7009 * This function does the last stage of destroying an allocated device
7010 * interface. The reference to the device object is released.
7011 * If this is the last reference then it will be freed.
7013 void free_netdev(struct net_device *dev)
7015 struct napi_struct *p, *n;
7017 netif_free_tx_queues(dev);
7022 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7024 /* Flush device addresses */
7025 dev_addr_flush(dev);
7027 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7030 free_percpu(dev->pcpu_refcnt);
7031 dev->pcpu_refcnt = NULL;
7033 /* Compatibility with error handling in drivers */
7034 if (dev->reg_state == NETREG_UNINITIALIZED) {
7035 netdev_freemem(dev);
7039 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7040 dev->reg_state = NETREG_RELEASED;
7042 /* will free via device release */
7043 put_device(&dev->dev);
7045 EXPORT_SYMBOL(free_netdev);
7048 * synchronize_net - Synchronize with packet receive processing
7050 * Wait for packets currently being received to be done.
7051 * Does not block later packets from starting.
7053 void synchronize_net(void)
7056 if (rtnl_is_locked())
7057 synchronize_rcu_expedited();
7061 EXPORT_SYMBOL(synchronize_net);
7064 * unregister_netdevice_queue - remove device from the kernel
7068 * This function shuts down a device interface and removes it
7069 * from the kernel tables.
7070 * If head not NULL, device is queued to be unregistered later.
7072 * Callers must hold the rtnl semaphore. You may want
7073 * unregister_netdev() instead of this.
7076 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7081 list_move_tail(&dev->unreg_list, head);
7083 rollback_registered(dev);
7084 /* Finish processing unregister after unlock */
7088 EXPORT_SYMBOL(unregister_netdevice_queue);
7091 * unregister_netdevice_many - unregister many devices
7092 * @head: list of devices
7094 * Note: As most callers use a stack allocated list_head,
7095 * we force a list_del() to make sure stack wont be corrupted later.
7097 void unregister_netdevice_many(struct list_head *head)
7099 struct net_device *dev;
7101 if (!list_empty(head)) {
7102 rollback_registered_many(head);
7103 list_for_each_entry(dev, head, unreg_list)
7108 EXPORT_SYMBOL(unregister_netdevice_many);
7111 * unregister_netdev - remove device from the kernel
7114 * This function shuts down a device interface and removes it
7115 * from the kernel tables.
7117 * This is just a wrapper for unregister_netdevice that takes
7118 * the rtnl semaphore. In general you want to use this and not
7119 * unregister_netdevice.
7121 void unregister_netdev(struct net_device *dev)
7124 unregister_netdevice(dev);
7127 EXPORT_SYMBOL(unregister_netdev);
7130 * dev_change_net_namespace - move device to different nethost namespace
7132 * @net: network namespace
7133 * @pat: If not NULL name pattern to try if the current device name
7134 * is already taken in the destination network namespace.
7136 * This function shuts down a device interface and moves it
7137 * to a new network namespace. On success 0 is returned, on
7138 * a failure a netagive errno code is returned.
7140 * Callers must hold the rtnl semaphore.
7143 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7149 /* Don't allow namespace local devices to be moved. */
7151 if (dev->features & NETIF_F_NETNS_LOCAL)
7154 /* Ensure the device has been registrered */
7155 if (dev->reg_state != NETREG_REGISTERED)
7158 /* Get out if there is nothing todo */
7160 if (net_eq(dev_net(dev), net))
7163 /* Pick the destination device name, and ensure
7164 * we can use it in the destination network namespace.
7167 if (__dev_get_by_name(net, dev->name)) {
7168 /* We get here if we can't use the current device name */
7171 if (dev_get_valid_name(net, dev, pat) < 0)
7176 * And now a mini version of register_netdevice unregister_netdevice.
7179 /* If device is running close it first. */
7182 /* And unlink it from device chain */
7184 unlist_netdevice(dev);
7188 /* Shutdown queueing discipline. */
7191 /* Notify protocols, that we are about to destroy
7192 this device. They should clean all the things.
7194 Note that dev->reg_state stays at NETREG_REGISTERED.
7195 This is wanted because this way 8021q and macvlan know
7196 the device is just moving and can keep their slaves up.
7198 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7200 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7201 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7204 * Flush the unicast and multicast chains
7209 /* Send a netdev-removed uevent to the old namespace */
7210 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7211 netdev_adjacent_del_links(dev);
7213 /* Actually switch the network namespace */
7214 dev_net_set(dev, net);
7216 /* If there is an ifindex conflict assign a new one */
7217 if (__dev_get_by_index(net, dev->ifindex))
7218 dev->ifindex = dev_new_index(net);
7220 /* Send a netdev-add uevent to the new namespace */
7221 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7222 netdev_adjacent_add_links(dev);
7224 /* Fixup kobjects */
7225 err = device_rename(&dev->dev, dev->name);
7228 /* Add the device back in the hashes */
7229 list_netdevice(dev);
7231 /* Notify protocols, that a new device appeared. */
7232 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7235 * Prevent userspace races by waiting until the network
7236 * device is fully setup before sending notifications.
7238 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7245 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7247 static int dev_cpu_callback(struct notifier_block *nfb,
7248 unsigned long action,
7251 struct sk_buff **list_skb;
7252 struct sk_buff *skb;
7253 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7254 struct softnet_data *sd, *oldsd;
7256 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7259 local_irq_disable();
7260 cpu = smp_processor_id();
7261 sd = &per_cpu(softnet_data, cpu);
7262 oldsd = &per_cpu(softnet_data, oldcpu);
7264 /* Find end of our completion_queue. */
7265 list_skb = &sd->completion_queue;
7267 list_skb = &(*list_skb)->next;
7268 /* Append completion queue from offline CPU. */
7269 *list_skb = oldsd->completion_queue;
7270 oldsd->completion_queue = NULL;
7272 /* Append output queue from offline CPU. */
7273 if (oldsd->output_queue) {
7274 *sd->output_queue_tailp = oldsd->output_queue;
7275 sd->output_queue_tailp = oldsd->output_queue_tailp;
7276 oldsd->output_queue = NULL;
7277 oldsd->output_queue_tailp = &oldsd->output_queue;
7279 /* Append NAPI poll list from offline CPU, with one exception :
7280 * process_backlog() must be called by cpu owning percpu backlog.
7281 * We properly handle process_queue & input_pkt_queue later.
7283 while (!list_empty(&oldsd->poll_list)) {
7284 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7288 list_del_init(&napi->poll_list);
7289 if (napi->poll == process_backlog)
7292 ____napi_schedule(sd, napi);
7295 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7298 /* Process offline CPU's input_pkt_queue */
7299 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7301 input_queue_head_incr(oldsd);
7303 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7305 input_queue_head_incr(oldsd);
7313 * netdev_increment_features - increment feature set by one
7314 * @all: current feature set
7315 * @one: new feature set
7316 * @mask: mask feature set
7318 * Computes a new feature set after adding a device with feature set
7319 * @one to the master device with current feature set @all. Will not
7320 * enable anything that is off in @mask. Returns the new feature set.
7322 netdev_features_t netdev_increment_features(netdev_features_t all,
7323 netdev_features_t one, netdev_features_t mask)
7325 if (mask & NETIF_F_GEN_CSUM)
7326 mask |= NETIF_F_ALL_CSUM;
7327 mask |= NETIF_F_VLAN_CHALLENGED;
7329 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7330 all &= one | ~NETIF_F_ALL_FOR_ALL;
7332 /* If one device supports hw checksumming, set for all. */
7333 if (all & NETIF_F_GEN_CSUM)
7334 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7338 EXPORT_SYMBOL(netdev_increment_features);
7340 static struct hlist_head * __net_init netdev_create_hash(void)
7343 struct hlist_head *hash;
7345 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7347 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7348 INIT_HLIST_HEAD(&hash[i]);
7353 /* Initialize per network namespace state */
7354 static int __net_init netdev_init(struct net *net)
7356 if (net != &init_net)
7357 INIT_LIST_HEAD(&net->dev_base_head);
7359 net->dev_name_head = netdev_create_hash();
7360 if (net->dev_name_head == NULL)
7363 net->dev_index_head = netdev_create_hash();
7364 if (net->dev_index_head == NULL)
7370 kfree(net->dev_name_head);
7376 * netdev_drivername - network driver for the device
7377 * @dev: network device
7379 * Determine network driver for device.
7381 const char *netdev_drivername(const struct net_device *dev)
7383 const struct device_driver *driver;
7384 const struct device *parent;
7385 const char *empty = "";
7387 parent = dev->dev.parent;
7391 driver = parent->driver;
7392 if (driver && driver->name)
7393 return driver->name;
7397 static void __netdev_printk(const char *level, const struct net_device *dev,
7398 struct va_format *vaf)
7400 if (dev && dev->dev.parent) {
7401 dev_printk_emit(level[1] - '0',
7404 dev_driver_string(dev->dev.parent),
7405 dev_name(dev->dev.parent),
7406 netdev_name(dev), netdev_reg_state(dev),
7409 printk("%s%s%s: %pV",
7410 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7412 printk("%s(NULL net_device): %pV", level, vaf);
7416 void netdev_printk(const char *level, const struct net_device *dev,
7417 const char *format, ...)
7419 struct va_format vaf;
7422 va_start(args, format);
7427 __netdev_printk(level, dev, &vaf);
7431 EXPORT_SYMBOL(netdev_printk);
7433 #define define_netdev_printk_level(func, level) \
7434 void func(const struct net_device *dev, const char *fmt, ...) \
7436 struct va_format vaf; \
7439 va_start(args, fmt); \
7444 __netdev_printk(level, dev, &vaf); \
7448 EXPORT_SYMBOL(func);
7450 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7451 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7452 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7453 define_netdev_printk_level(netdev_err, KERN_ERR);
7454 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7455 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7456 define_netdev_printk_level(netdev_info, KERN_INFO);
7458 static void __net_exit netdev_exit(struct net *net)
7460 kfree(net->dev_name_head);
7461 kfree(net->dev_index_head);
7464 static struct pernet_operations __net_initdata netdev_net_ops = {
7465 .init = netdev_init,
7466 .exit = netdev_exit,
7469 static void __net_exit default_device_exit(struct net *net)
7471 struct net_device *dev, *aux;
7473 * Push all migratable network devices back to the
7474 * initial network namespace
7477 for_each_netdev_safe(net, dev, aux) {
7479 char fb_name[IFNAMSIZ];
7481 /* Ignore unmoveable devices (i.e. loopback) */
7482 if (dev->features & NETIF_F_NETNS_LOCAL)
7485 /* Leave virtual devices for the generic cleanup */
7486 if (dev->rtnl_link_ops)
7489 /* Push remaining network devices to init_net */
7490 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7491 err = dev_change_net_namespace(dev, &init_net, fb_name);
7493 pr_emerg("%s: failed to move %s to init_net: %d\n",
7494 __func__, dev->name, err);
7501 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7503 /* Return with the rtnl_lock held when there are no network
7504 * devices unregistering in any network namespace in net_list.
7508 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7510 add_wait_queue(&netdev_unregistering_wq, &wait);
7512 unregistering = false;
7514 list_for_each_entry(net, net_list, exit_list) {
7515 if (net->dev_unreg_count > 0) {
7516 unregistering = true;
7524 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7526 remove_wait_queue(&netdev_unregistering_wq, &wait);
7529 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7531 /* At exit all network devices most be removed from a network
7532 * namespace. Do this in the reverse order of registration.
7533 * Do this across as many network namespaces as possible to
7534 * improve batching efficiency.
7536 struct net_device *dev;
7538 LIST_HEAD(dev_kill_list);
7540 /* To prevent network device cleanup code from dereferencing
7541 * loopback devices or network devices that have been freed
7542 * wait here for all pending unregistrations to complete,
7543 * before unregistring the loopback device and allowing the
7544 * network namespace be freed.
7546 * The netdev todo list containing all network devices
7547 * unregistrations that happen in default_device_exit_batch
7548 * will run in the rtnl_unlock() at the end of
7549 * default_device_exit_batch.
7551 rtnl_lock_unregistering(net_list);
7552 list_for_each_entry(net, net_list, exit_list) {
7553 for_each_netdev_reverse(net, dev) {
7554 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7555 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7557 unregister_netdevice_queue(dev, &dev_kill_list);
7560 unregister_netdevice_many(&dev_kill_list);
7564 static struct pernet_operations __net_initdata default_device_ops = {
7565 .exit = default_device_exit,
7566 .exit_batch = default_device_exit_batch,
7570 * Initialize the DEV module. At boot time this walks the device list and
7571 * unhooks any devices that fail to initialise (normally hardware not
7572 * present) and leaves us with a valid list of present and active devices.
7577 * This is called single threaded during boot, so no need
7578 * to take the rtnl semaphore.
7580 static int __init net_dev_init(void)
7582 int i, rc = -ENOMEM;
7584 BUG_ON(!dev_boot_phase);
7586 if (dev_proc_init())
7589 if (netdev_kobject_init())
7592 INIT_LIST_HEAD(&ptype_all);
7593 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7594 INIT_LIST_HEAD(&ptype_base[i]);
7596 INIT_LIST_HEAD(&offload_base);
7598 if (register_pernet_subsys(&netdev_net_ops))
7602 * Initialise the packet receive queues.
7605 for_each_possible_cpu(i) {
7606 struct softnet_data *sd = &per_cpu(softnet_data, i);
7608 skb_queue_head_init(&sd->input_pkt_queue);
7609 skb_queue_head_init(&sd->process_queue);
7610 INIT_LIST_HEAD(&sd->poll_list);
7611 sd->output_queue_tailp = &sd->output_queue;
7613 sd->csd.func = rps_trigger_softirq;
7618 sd->backlog.poll = process_backlog;
7619 sd->backlog.weight = weight_p;
7624 /* The loopback device is special if any other network devices
7625 * is present in a network namespace the loopback device must
7626 * be present. Since we now dynamically allocate and free the
7627 * loopback device ensure this invariant is maintained by
7628 * keeping the loopback device as the first device on the
7629 * list of network devices. Ensuring the loopback devices
7630 * is the first device that appears and the last network device
7633 if (register_pernet_device(&loopback_net_ops))
7636 if (register_pernet_device(&default_device_ops))
7639 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7640 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7642 hotcpu_notifier(dev_cpu_callback, 0);
7649 subsys_initcall(net_dev_init);