1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
111 #include <linux/highmem.h>
112 #include <linux/init.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <net/mpls.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <trace/events/net.h>
134 #include <trace/events/skb.h>
135 #include <trace/events/qdisc.h>
136 #include <trace/events/xdp.h>
137 #include <linux/inetdevice.h>
138 #include <linux/cpu_rmap.h>
139 #include <linux/static_key.h>
140 #include <linux/hashtable.h>
141 #include <linux/vmalloc.h>
142 #include <linux/if_macvlan.h>
143 #include <linux/errqueue.h>
144 #include <linux/hrtimer.h>
145 #include <linux/netfilter_netdev.h>
146 #include <linux/crash_dump.h>
147 #include <linux/sctp.h>
148 #include <net/udp_tunnel.h>
149 #include <linux/net_namespace.h>
150 #include <linux/indirect_call_wrapper.h>
151 #include <net/devlink.h>
152 #include <linux/pm_runtime.h>
153 #include <linux/prandom.h>
154 #include <linux/once_lite.h>
155 #include <net/netdev_rx_queue.h>
158 #include "net-sysfs.h"
160 static DEFINE_SPINLOCK(ptype_lock);
161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
162 struct list_head ptype_all __read_mostly; /* Taps */
164 static int netif_rx_internal(struct sk_buff *skb);
165 static int call_netdevice_notifiers_extack(unsigned long val,
166 struct net_device *dev,
167 struct netlink_ext_ack *extack);
168 static struct napi_struct *napi_by_id(unsigned int napi_id);
171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
176 * Writers must hold the rtnl semaphore while they loop through the
177 * dev_base_head list, and hold dev_base_lock for writing when they do the
178 * actual updates. This allows pure readers to access the list even
179 * while a writer is preparing to update it.
181 * To put it another way, dev_base_lock is held for writing only to
182 * protect against pure readers; the rtnl semaphore provides the
183 * protection against other writers.
185 * See, for example usages, register_netdevice() and
186 * unregister_netdevice(), which must be called with the rtnl
189 DEFINE_RWLOCK(dev_base_lock);
190 EXPORT_SYMBOL(dev_base_lock);
192 static DEFINE_MUTEX(ifalias_mutex);
194 /* protects napi_hash addition/deletion and napi_gen_id */
195 static DEFINE_SPINLOCK(napi_hash_lock);
197 static unsigned int napi_gen_id = NR_CPUS;
198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
200 static DECLARE_RWSEM(devnet_rename_sem);
202 static inline void dev_base_seq_inc(struct net *net)
204 while (++net->dev_base_seq == 0)
208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
220 static inline void rps_lock_irqsave(struct softnet_data *sd,
221 unsigned long *flags)
223 if (IS_ENABLED(CONFIG_RPS))
224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
226 local_irq_save(*flags);
229 static inline void rps_lock_irq_disable(struct softnet_data *sd)
231 if (IS_ENABLED(CONFIG_RPS))
232 spin_lock_irq(&sd->input_pkt_queue.lock);
233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
237 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
238 unsigned long *flags)
240 if (IS_ENABLED(CONFIG_RPS))
241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
243 local_irq_restore(*flags);
246 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
248 if (IS_ENABLED(CONFIG_RPS))
249 spin_unlock_irq(&sd->input_pkt_queue.lock);
250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
257 struct netdev_name_node *name_node;
259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
262 INIT_HLIST_NODE(&name_node->hlist);
263 name_node->dev = dev;
264 name_node->name = name;
268 static struct netdev_name_node *
269 netdev_name_node_head_alloc(struct net_device *dev)
271 struct netdev_name_node *name_node;
273 name_node = netdev_name_node_alloc(dev, dev->name);
276 INIT_LIST_HEAD(&name_node->list);
280 static void netdev_name_node_free(struct netdev_name_node *name_node)
285 static void netdev_name_node_add(struct net *net,
286 struct netdev_name_node *name_node)
288 hlist_add_head_rcu(&name_node->hlist,
289 dev_name_hash(net, name_node->name));
292 static void netdev_name_node_del(struct netdev_name_node *name_node)
294 hlist_del_rcu(&name_node->hlist);
297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
300 struct hlist_head *head = dev_name_hash(net, name);
301 struct netdev_name_node *name_node;
303 hlist_for_each_entry(name_node, head, hlist)
304 if (!strcmp(name_node->name, name))
309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
312 struct hlist_head *head = dev_name_hash(net, name);
313 struct netdev_name_node *name_node;
315 hlist_for_each_entry_rcu(name_node, head, hlist)
316 if (!strcmp(name_node->name, name))
321 bool netdev_name_in_use(struct net *net, const char *name)
323 return netdev_name_node_lookup(net, name);
325 EXPORT_SYMBOL(netdev_name_in_use);
327 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
329 struct netdev_name_node *name_node;
330 struct net *net = dev_net(dev);
332 name_node = netdev_name_node_lookup(net, name);
335 name_node = netdev_name_node_alloc(dev, name);
338 netdev_name_node_add(net, name_node);
339 /* The node that holds dev->name acts as a head of per-device list. */
340 list_add_tail(&name_node->list, &dev->name_node->list);
345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
347 list_del(&name_node->list);
348 netdev_name_node_del(name_node);
349 kfree(name_node->name);
350 netdev_name_node_free(name_node);
353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
355 struct netdev_name_node *name_node;
356 struct net *net = dev_net(dev);
358 name_node = netdev_name_node_lookup(net, name);
361 /* lookup might have found our primary name or a name belonging
364 if (name_node == dev->name_node || name_node->dev != dev)
367 __netdev_name_node_alt_destroy(name_node);
372 static void netdev_name_node_alt_flush(struct net_device *dev)
374 struct netdev_name_node *name_node, *tmp;
376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
377 __netdev_name_node_alt_destroy(name_node);
380 /* Device list insertion */
381 static void list_netdevice(struct net_device *dev)
383 struct net *net = dev_net(dev);
387 write_lock(&dev_base_lock);
388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
389 netdev_name_node_add(net, dev->name_node);
390 hlist_add_head_rcu(&dev->index_hlist,
391 dev_index_hash(net, dev->ifindex));
392 write_unlock(&dev_base_lock);
394 dev_base_seq_inc(net);
397 /* Device list removal
398 * caller must respect a RCU grace period before freeing/reusing dev
400 static void unlist_netdevice(struct net_device *dev, bool lock)
404 /* Unlink dev from the device chain */
406 write_lock(&dev_base_lock);
407 list_del_rcu(&dev->dev_list);
408 netdev_name_node_del(dev->name_node);
409 hlist_del_rcu(&dev->index_hlist);
411 write_unlock(&dev_base_lock);
413 dev_base_seq_inc(dev_net(dev));
420 static RAW_NOTIFIER_HEAD(netdev_chain);
423 * Device drivers call our routines to queue packets here. We empty the
424 * queue in the local softnet handler.
427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
428 EXPORT_PER_CPU_SYMBOL(softnet_data);
430 #ifdef CONFIG_LOCKDEP
432 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
433 * according to dev->type
435 static const unsigned short netdev_lock_type[] = {
436 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
437 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
438 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
439 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
440 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
441 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
442 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
443 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
444 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
445 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
446 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
447 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
448 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
449 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
450 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
452 static const char *const netdev_lock_name[] = {
453 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
454 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
455 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
456 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
457 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
458 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
459 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
460 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
461 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
462 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
463 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
464 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
465 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
466 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
467 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
472 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
476 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
477 if (netdev_lock_type[i] == dev_type)
479 /* the last key is used by default */
480 return ARRAY_SIZE(netdev_lock_type) - 1;
483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
484 unsigned short dev_type)
488 i = netdev_lock_pos(dev_type);
489 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
490 netdev_lock_name[i]);
493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
497 i = netdev_lock_pos(dev->type);
498 lockdep_set_class_and_name(&dev->addr_list_lock,
499 &netdev_addr_lock_key[i],
500 netdev_lock_name[i]);
503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
504 unsigned short dev_type)
508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
513 /*******************************************************************************
515 * Protocol management and registration routines
517 *******************************************************************************/
521 * Add a protocol ID to the list. Now that the input handler is
522 * smarter we can dispense with all the messy stuff that used to be
525 * BEWARE!!! Protocol handlers, mangling input packets,
526 * MUST BE last in hash buckets and checking protocol handlers
527 * MUST start from promiscuous ptype_all chain in net_bh.
528 * It is true now, do not change it.
529 * Explanation follows: if protocol handler, mangling packet, will
530 * be the first on list, it is not able to sense, that packet
531 * is cloned and should be copied-on-write, so that it will
532 * change it and subsequent readers will get broken packet.
536 static inline struct list_head *ptype_head(const struct packet_type *pt)
538 if (pt->type == htons(ETH_P_ALL))
539 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
541 return pt->dev ? &pt->dev->ptype_specific :
542 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
546 * dev_add_pack - add packet handler
547 * @pt: packet type declaration
549 * Add a protocol handler to the networking stack. The passed &packet_type
550 * is linked into kernel lists and may not be freed until it has been
551 * removed from the kernel lists.
553 * This call does not sleep therefore it can not
554 * guarantee all CPU's that are in middle of receiving packets
555 * will see the new packet type (until the next received packet).
558 void dev_add_pack(struct packet_type *pt)
560 struct list_head *head = ptype_head(pt);
562 spin_lock(&ptype_lock);
563 list_add_rcu(&pt->list, head);
564 spin_unlock(&ptype_lock);
566 EXPORT_SYMBOL(dev_add_pack);
569 * __dev_remove_pack - remove packet handler
570 * @pt: packet type declaration
572 * Remove a protocol handler that was previously added to the kernel
573 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
574 * from the kernel lists and can be freed or reused once this function
577 * The packet type might still be in use by receivers
578 * and must not be freed until after all the CPU's have gone
579 * through a quiescent state.
581 void __dev_remove_pack(struct packet_type *pt)
583 struct list_head *head = ptype_head(pt);
584 struct packet_type *pt1;
586 spin_lock(&ptype_lock);
588 list_for_each_entry(pt1, head, list) {
590 list_del_rcu(&pt->list);
595 pr_warn("dev_remove_pack: %p not found\n", pt);
597 spin_unlock(&ptype_lock);
599 EXPORT_SYMBOL(__dev_remove_pack);
602 * dev_remove_pack - remove packet handler
603 * @pt: packet type declaration
605 * Remove a protocol handler that was previously added to the kernel
606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
607 * from the kernel lists and can be freed or reused once this function
610 * This call sleeps to guarantee that no CPU is looking at the packet
613 void dev_remove_pack(struct packet_type *pt)
615 __dev_remove_pack(pt);
619 EXPORT_SYMBOL(dev_remove_pack);
622 /*******************************************************************************
624 * Device Interface Subroutines
626 *******************************************************************************/
629 * dev_get_iflink - get 'iflink' value of a interface
630 * @dev: targeted interface
632 * Indicates the ifindex the interface is linked to.
633 * Physical interfaces have the same 'ifindex' and 'iflink' values.
636 int dev_get_iflink(const struct net_device *dev)
638 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
639 return dev->netdev_ops->ndo_get_iflink(dev);
643 EXPORT_SYMBOL(dev_get_iflink);
646 * dev_fill_metadata_dst - Retrieve tunnel egress information.
647 * @dev: targeted interface
650 * For better visibility of tunnel traffic OVS needs to retrieve
651 * egress tunnel information for a packet. Following API allows
652 * user to get this info.
654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
656 struct ip_tunnel_info *info;
658 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
661 info = skb_tunnel_info_unclone(skb);
664 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
667 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
673 int k = stack->num_paths++;
675 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
678 return &stack->path[k];
681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
682 struct net_device_path_stack *stack)
684 const struct net_device *last_dev;
685 struct net_device_path_ctx ctx = {
688 struct net_device_path *path;
691 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
692 stack->num_paths = 0;
693 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
695 path = dev_fwd_path(stack);
699 memset(path, 0, sizeof(struct net_device_path));
700 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
704 if (WARN_ON_ONCE(last_dev == ctx.dev))
711 path = dev_fwd_path(stack);
714 path->type = DEV_PATH_ETHERNET;
719 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
722 * __dev_get_by_name - find a device by its name
723 * @net: the applicable net namespace
724 * @name: name to find
726 * Find an interface by name. Must be called under RTNL semaphore
727 * or @dev_base_lock. If the name is found a pointer to the device
728 * is returned. If the name is not found then %NULL is returned. The
729 * reference counters are not incremented so the caller must be
730 * careful with locks.
733 struct net_device *__dev_get_by_name(struct net *net, const char *name)
735 struct netdev_name_node *node_name;
737 node_name = netdev_name_node_lookup(net, name);
738 return node_name ? node_name->dev : NULL;
740 EXPORT_SYMBOL(__dev_get_by_name);
743 * dev_get_by_name_rcu - find a device by its name
744 * @net: the applicable net namespace
745 * @name: name to find
747 * Find an interface by name.
748 * If the name is found a pointer to the device is returned.
749 * If the name is not found then %NULL is returned.
750 * The reference counters are not incremented so the caller must be
751 * careful with locks. The caller must hold RCU lock.
754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
756 struct netdev_name_node *node_name;
758 node_name = netdev_name_node_lookup_rcu(net, name);
759 return node_name ? node_name->dev : NULL;
761 EXPORT_SYMBOL(dev_get_by_name_rcu);
763 /* Deprecated for new users, call netdev_get_by_name() instead */
764 struct net_device *dev_get_by_name(struct net *net, const char *name)
766 struct net_device *dev;
769 dev = dev_get_by_name_rcu(net, name);
774 EXPORT_SYMBOL(dev_get_by_name);
777 * netdev_get_by_name() - find a device by its name
778 * @net: the applicable net namespace
779 * @name: name to find
780 * @tracker: tracking object for the acquired reference
781 * @gfp: allocation flags for the tracker
783 * Find an interface by name. This can be called from any
784 * context and does its own locking. The returned handle has
785 * the usage count incremented and the caller must use netdev_put() to
786 * release it when it is no longer needed. %NULL is returned if no
787 * matching device is found.
789 struct net_device *netdev_get_by_name(struct net *net, const char *name,
790 netdevice_tracker *tracker, gfp_t gfp)
792 struct net_device *dev;
794 dev = dev_get_by_name(net, name);
796 netdev_tracker_alloc(dev, tracker, gfp);
799 EXPORT_SYMBOL(netdev_get_by_name);
802 * __dev_get_by_index - find a device by its ifindex
803 * @net: the applicable net namespace
804 * @ifindex: index of device
806 * Search for an interface by index. Returns %NULL if the device
807 * is not found or a pointer to the device. The device has not
808 * had its reference counter increased so the caller must be careful
809 * about locking. The caller must hold either the RTNL semaphore
813 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
815 struct net_device *dev;
816 struct hlist_head *head = dev_index_hash(net, ifindex);
818 hlist_for_each_entry(dev, head, index_hlist)
819 if (dev->ifindex == ifindex)
824 EXPORT_SYMBOL(__dev_get_by_index);
827 * dev_get_by_index_rcu - find a device by its ifindex
828 * @net: the applicable net namespace
829 * @ifindex: index of device
831 * Search for an interface by index. Returns %NULL if the device
832 * is not found or a pointer to the device. The device has not
833 * had its reference counter increased so the caller must be careful
834 * about locking. The caller must hold RCU lock.
837 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
839 struct net_device *dev;
840 struct hlist_head *head = dev_index_hash(net, ifindex);
842 hlist_for_each_entry_rcu(dev, head, index_hlist)
843 if (dev->ifindex == ifindex)
848 EXPORT_SYMBOL(dev_get_by_index_rcu);
850 /* Deprecated for new users, call netdev_get_by_index() instead */
851 struct net_device *dev_get_by_index(struct net *net, int ifindex)
853 struct net_device *dev;
856 dev = dev_get_by_index_rcu(net, ifindex);
861 EXPORT_SYMBOL(dev_get_by_index);
864 * netdev_get_by_index() - find a device by its ifindex
865 * @net: the applicable net namespace
866 * @ifindex: index of device
867 * @tracker: tracking object for the acquired reference
868 * @gfp: allocation flags for the tracker
870 * Search for an interface by index. Returns NULL if the device
871 * is not found or a pointer to the device. The device returned has
872 * had a reference added and the pointer is safe until the user calls
873 * netdev_put() to indicate they have finished with it.
875 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
876 netdevice_tracker *tracker, gfp_t gfp)
878 struct net_device *dev;
880 dev = dev_get_by_index(net, ifindex);
882 netdev_tracker_alloc(dev, tracker, gfp);
885 EXPORT_SYMBOL(netdev_get_by_index);
888 * dev_get_by_napi_id - find a device by napi_id
889 * @napi_id: ID of the NAPI struct
891 * Search for an interface by NAPI ID. Returns %NULL if the device
892 * is not found or a pointer to the device. The device has not had
893 * its reference counter increased so the caller must be careful
894 * about locking. The caller must hold RCU lock.
897 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
899 struct napi_struct *napi;
901 WARN_ON_ONCE(!rcu_read_lock_held());
903 if (napi_id < MIN_NAPI_ID)
906 napi = napi_by_id(napi_id);
908 return napi ? napi->dev : NULL;
910 EXPORT_SYMBOL(dev_get_by_napi_id);
913 * netdev_get_name - get a netdevice name, knowing its ifindex.
914 * @net: network namespace
915 * @name: a pointer to the buffer where the name will be stored.
916 * @ifindex: the ifindex of the interface to get the name from.
918 int netdev_get_name(struct net *net, char *name, int ifindex)
920 struct net_device *dev;
923 down_read(&devnet_rename_sem);
926 dev = dev_get_by_index_rcu(net, ifindex);
932 strcpy(name, dev->name);
937 up_read(&devnet_rename_sem);
942 * dev_getbyhwaddr_rcu - find a device by its hardware address
943 * @net: the applicable net namespace
944 * @type: media type of device
945 * @ha: hardware address
947 * Search for an interface by MAC address. Returns NULL if the device
948 * is not found or a pointer to the device.
949 * The caller must hold RCU or RTNL.
950 * The returned device has not had its ref count increased
951 * and the caller must therefore be careful about locking
955 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
958 struct net_device *dev;
960 for_each_netdev_rcu(net, dev)
961 if (dev->type == type &&
962 !memcmp(dev->dev_addr, ha, dev->addr_len))
967 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
969 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
971 struct net_device *dev, *ret = NULL;
974 for_each_netdev_rcu(net, dev)
975 if (dev->type == type) {
983 EXPORT_SYMBOL(dev_getfirstbyhwtype);
986 * __dev_get_by_flags - find any device with given flags
987 * @net: the applicable net namespace
988 * @if_flags: IFF_* values
989 * @mask: bitmask of bits in if_flags to check
991 * Search for any interface with the given flags. Returns NULL if a device
992 * is not found or a pointer to the device. Must be called inside
993 * rtnl_lock(), and result refcount is unchanged.
996 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
999 struct net_device *dev, *ret;
1004 for_each_netdev(net, dev) {
1005 if (((dev->flags ^ if_flags) & mask) == 0) {
1012 EXPORT_SYMBOL(__dev_get_by_flags);
1015 * dev_valid_name - check if name is okay for network device
1016 * @name: name string
1018 * Network device names need to be valid file names to
1019 * allow sysfs to work. We also disallow any kind of
1022 bool dev_valid_name(const char *name)
1026 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1028 if (!strcmp(name, ".") || !strcmp(name, ".."))
1032 if (*name == '/' || *name == ':' || isspace(*name))
1038 EXPORT_SYMBOL(dev_valid_name);
1041 * __dev_alloc_name - allocate a name for a device
1042 * @net: network namespace to allocate the device name in
1043 * @name: name format string
1044 * @buf: scratch buffer and result name string
1046 * Passed a format string - eg "lt%d" it will try and find a suitable
1047 * id. It scans list of devices to build up a free map, then chooses
1048 * the first empty slot. The caller must hold the dev_base or rtnl lock
1049 * while allocating the name and adding the device in order to avoid
1051 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1052 * Returns the number of the unit assigned or a negative errno code.
1055 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1059 const int max_netdevices = 8*PAGE_SIZE;
1060 unsigned long *inuse;
1061 struct net_device *d;
1063 if (!dev_valid_name(name))
1066 p = strchr(name, '%');
1069 * Verify the string as this thing may have come from
1070 * the user. There must be either one "%d" and no other "%"
1073 if (p[1] != 'd' || strchr(p + 2, '%'))
1076 /* Use one page as a bit array of possible slots */
1077 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1081 for_each_netdev(net, d) {
1082 struct netdev_name_node *name_node;
1083 list_for_each_entry(name_node, &d->name_node->list, list) {
1084 if (!sscanf(name_node->name, name, &i))
1086 if (i < 0 || i >= max_netdevices)
1089 /* avoid cases where sscanf is not exact inverse of printf */
1090 snprintf(buf, IFNAMSIZ, name, i);
1091 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1092 __set_bit(i, inuse);
1094 if (!sscanf(d->name, name, &i))
1096 if (i < 0 || i >= max_netdevices)
1099 /* avoid cases where sscanf is not exact inverse of printf */
1100 snprintf(buf, IFNAMSIZ, name, i);
1101 if (!strncmp(buf, d->name, IFNAMSIZ))
1102 __set_bit(i, inuse);
1105 i = find_first_zero_bit(inuse, max_netdevices);
1106 free_page((unsigned long) inuse);
1109 snprintf(buf, IFNAMSIZ, name, i);
1110 if (!netdev_name_in_use(net, buf))
1113 /* It is possible to run out of possible slots
1114 * when the name is long and there isn't enough space left
1115 * for the digits, or if all bits are used.
1120 static int dev_alloc_name_ns(struct net *net,
1121 struct net_device *dev,
1128 ret = __dev_alloc_name(net, name, buf);
1130 strscpy(dev->name, buf, IFNAMSIZ);
1135 * dev_alloc_name - allocate a name for a device
1137 * @name: name format string
1139 * Passed a format string - eg "lt%d" it will try and find a suitable
1140 * id. It scans list of devices to build up a free map, then chooses
1141 * the first empty slot. The caller must hold the dev_base or rtnl lock
1142 * while allocating the name and adding the device in order to avoid
1144 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1145 * Returns the number of the unit assigned or a negative errno code.
1148 int dev_alloc_name(struct net_device *dev, const char *name)
1150 return dev_alloc_name_ns(dev_net(dev), dev, name);
1152 EXPORT_SYMBOL(dev_alloc_name);
1154 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1159 if (!dev_valid_name(name))
1162 if (strchr(name, '%'))
1163 return dev_alloc_name_ns(net, dev, name);
1164 else if (netdev_name_in_use(net, name))
1166 else if (dev->name != name)
1167 strscpy(dev->name, name, IFNAMSIZ);
1173 * dev_change_name - change name of a device
1175 * @newname: name (or format string) must be at least IFNAMSIZ
1177 * Change name of a device, can pass format strings "eth%d".
1180 int dev_change_name(struct net_device *dev, const char *newname)
1182 unsigned char old_assign_type;
1183 char oldname[IFNAMSIZ];
1189 BUG_ON(!dev_net(dev));
1193 down_write(&devnet_rename_sem);
1195 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1196 up_write(&devnet_rename_sem);
1200 memcpy(oldname, dev->name, IFNAMSIZ);
1202 err = dev_get_valid_name(net, dev, newname);
1204 up_write(&devnet_rename_sem);
1208 if (oldname[0] && !strchr(oldname, '%'))
1209 netdev_info(dev, "renamed from %s%s\n", oldname,
1210 dev->flags & IFF_UP ? " (while UP)" : "");
1212 old_assign_type = dev->name_assign_type;
1213 dev->name_assign_type = NET_NAME_RENAMED;
1216 ret = device_rename(&dev->dev, dev->name);
1218 memcpy(dev->name, oldname, IFNAMSIZ);
1219 dev->name_assign_type = old_assign_type;
1220 up_write(&devnet_rename_sem);
1224 up_write(&devnet_rename_sem);
1226 netdev_adjacent_rename_links(dev, oldname);
1228 write_lock(&dev_base_lock);
1229 netdev_name_node_del(dev->name_node);
1230 write_unlock(&dev_base_lock);
1234 write_lock(&dev_base_lock);
1235 netdev_name_node_add(net, dev->name_node);
1236 write_unlock(&dev_base_lock);
1238 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1239 ret = notifier_to_errno(ret);
1242 /* err >= 0 after dev_alloc_name() or stores the first errno */
1245 down_write(&devnet_rename_sem);
1246 memcpy(dev->name, oldname, IFNAMSIZ);
1247 memcpy(oldname, newname, IFNAMSIZ);
1248 dev->name_assign_type = old_assign_type;
1249 old_assign_type = NET_NAME_RENAMED;
1252 netdev_err(dev, "name change rollback failed: %d\n",
1261 * dev_set_alias - change ifalias of a device
1263 * @alias: name up to IFALIASZ
1264 * @len: limit of bytes to copy from info
1266 * Set ifalias for a device,
1268 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1270 struct dev_ifalias *new_alias = NULL;
1272 if (len >= IFALIASZ)
1276 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1280 memcpy(new_alias->ifalias, alias, len);
1281 new_alias->ifalias[len] = 0;
1284 mutex_lock(&ifalias_mutex);
1285 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1286 mutex_is_locked(&ifalias_mutex));
1287 mutex_unlock(&ifalias_mutex);
1290 kfree_rcu(new_alias, rcuhead);
1294 EXPORT_SYMBOL(dev_set_alias);
1297 * dev_get_alias - get ifalias of a device
1299 * @name: buffer to store name of ifalias
1300 * @len: size of buffer
1302 * get ifalias for a device. Caller must make sure dev cannot go
1303 * away, e.g. rcu read lock or own a reference count to device.
1305 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1307 const struct dev_ifalias *alias;
1311 alias = rcu_dereference(dev->ifalias);
1313 ret = snprintf(name, len, "%s", alias->ifalias);
1320 * netdev_features_change - device changes features
1321 * @dev: device to cause notification
1323 * Called to indicate a device has changed features.
1325 void netdev_features_change(struct net_device *dev)
1327 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1329 EXPORT_SYMBOL(netdev_features_change);
1332 * netdev_state_change - device changes state
1333 * @dev: device to cause notification
1335 * Called to indicate a device has changed state. This function calls
1336 * the notifier chains for netdev_chain and sends a NEWLINK message
1337 * to the routing socket.
1339 void netdev_state_change(struct net_device *dev)
1341 if (dev->flags & IFF_UP) {
1342 struct netdev_notifier_change_info change_info = {
1346 call_netdevice_notifiers_info(NETDEV_CHANGE,
1348 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1351 EXPORT_SYMBOL(netdev_state_change);
1354 * __netdev_notify_peers - notify network peers about existence of @dev,
1355 * to be called when rtnl lock is already held.
1356 * @dev: network device
1358 * Generate traffic such that interested network peers are aware of
1359 * @dev, such as by generating a gratuitous ARP. This may be used when
1360 * a device wants to inform the rest of the network about some sort of
1361 * reconfiguration such as a failover event or virtual machine
1364 void __netdev_notify_peers(struct net_device *dev)
1367 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1368 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1370 EXPORT_SYMBOL(__netdev_notify_peers);
1373 * netdev_notify_peers - notify network peers about existence of @dev
1374 * @dev: network device
1376 * Generate traffic such that interested network peers are aware of
1377 * @dev, such as by generating a gratuitous ARP. This may be used when
1378 * a device wants to inform the rest of the network about some sort of
1379 * reconfiguration such as a failover event or virtual machine
1382 void netdev_notify_peers(struct net_device *dev)
1385 __netdev_notify_peers(dev);
1388 EXPORT_SYMBOL(netdev_notify_peers);
1390 static int napi_threaded_poll(void *data);
1392 static int napi_kthread_create(struct napi_struct *n)
1396 /* Create and wake up the kthread once to put it in
1397 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1398 * warning and work with loadavg.
1400 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1401 n->dev->name, n->napi_id);
1402 if (IS_ERR(n->thread)) {
1403 err = PTR_ERR(n->thread);
1404 pr_err("kthread_run failed with err %d\n", err);
1411 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1413 const struct net_device_ops *ops = dev->netdev_ops;
1417 dev_addr_check(dev);
1419 if (!netif_device_present(dev)) {
1420 /* may be detached because parent is runtime-suspended */
1421 if (dev->dev.parent)
1422 pm_runtime_resume(dev->dev.parent);
1423 if (!netif_device_present(dev))
1427 /* Block netpoll from trying to do any rx path servicing.
1428 * If we don't do this there is a chance ndo_poll_controller
1429 * or ndo_poll may be running while we open the device
1431 netpoll_poll_disable(dev);
1433 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1434 ret = notifier_to_errno(ret);
1438 set_bit(__LINK_STATE_START, &dev->state);
1440 if (ops->ndo_validate_addr)
1441 ret = ops->ndo_validate_addr(dev);
1443 if (!ret && ops->ndo_open)
1444 ret = ops->ndo_open(dev);
1446 netpoll_poll_enable(dev);
1449 clear_bit(__LINK_STATE_START, &dev->state);
1451 dev->flags |= IFF_UP;
1452 dev_set_rx_mode(dev);
1454 add_device_randomness(dev->dev_addr, dev->addr_len);
1461 * dev_open - prepare an interface for use.
1462 * @dev: device to open
1463 * @extack: netlink extended ack
1465 * Takes a device from down to up state. The device's private open
1466 * function is invoked and then the multicast lists are loaded. Finally
1467 * the device is moved into the up state and a %NETDEV_UP message is
1468 * sent to the netdev notifier chain.
1470 * Calling this function on an active interface is a nop. On a failure
1471 * a negative errno code is returned.
1473 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1477 if (dev->flags & IFF_UP)
1480 ret = __dev_open(dev, extack);
1484 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1485 call_netdevice_notifiers(NETDEV_UP, dev);
1489 EXPORT_SYMBOL(dev_open);
1491 static void __dev_close_many(struct list_head *head)
1493 struct net_device *dev;
1498 list_for_each_entry(dev, head, close_list) {
1499 /* Temporarily disable netpoll until the interface is down */
1500 netpoll_poll_disable(dev);
1502 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1504 clear_bit(__LINK_STATE_START, &dev->state);
1506 /* Synchronize to scheduled poll. We cannot touch poll list, it
1507 * can be even on different cpu. So just clear netif_running().
1509 * dev->stop() will invoke napi_disable() on all of it's
1510 * napi_struct instances on this device.
1512 smp_mb__after_atomic(); /* Commit netif_running(). */
1515 dev_deactivate_many(head);
1517 list_for_each_entry(dev, head, close_list) {
1518 const struct net_device_ops *ops = dev->netdev_ops;
1521 * Call the device specific close. This cannot fail.
1522 * Only if device is UP
1524 * We allow it to be called even after a DETACH hot-plug
1530 dev->flags &= ~IFF_UP;
1531 netpoll_poll_enable(dev);
1535 static void __dev_close(struct net_device *dev)
1539 list_add(&dev->close_list, &single);
1540 __dev_close_many(&single);
1544 void dev_close_many(struct list_head *head, bool unlink)
1546 struct net_device *dev, *tmp;
1548 /* Remove the devices that don't need to be closed */
1549 list_for_each_entry_safe(dev, tmp, head, close_list)
1550 if (!(dev->flags & IFF_UP))
1551 list_del_init(&dev->close_list);
1553 __dev_close_many(head);
1555 list_for_each_entry_safe(dev, tmp, head, close_list) {
1556 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1557 call_netdevice_notifiers(NETDEV_DOWN, dev);
1559 list_del_init(&dev->close_list);
1562 EXPORT_SYMBOL(dev_close_many);
1565 * dev_close - shutdown an interface.
1566 * @dev: device to shutdown
1568 * This function moves an active device into down state. A
1569 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1570 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1573 void dev_close(struct net_device *dev)
1575 if (dev->flags & IFF_UP) {
1578 list_add(&dev->close_list, &single);
1579 dev_close_many(&single, true);
1583 EXPORT_SYMBOL(dev_close);
1587 * dev_disable_lro - disable Large Receive Offload on a device
1590 * Disable Large Receive Offload (LRO) on a net device. Must be
1591 * called under RTNL. This is needed if received packets may be
1592 * forwarded to another interface.
1594 void dev_disable_lro(struct net_device *dev)
1596 struct net_device *lower_dev;
1597 struct list_head *iter;
1599 dev->wanted_features &= ~NETIF_F_LRO;
1600 netdev_update_features(dev);
1602 if (unlikely(dev->features & NETIF_F_LRO))
1603 netdev_WARN(dev, "failed to disable LRO!\n");
1605 netdev_for_each_lower_dev(dev, lower_dev, iter)
1606 dev_disable_lro(lower_dev);
1608 EXPORT_SYMBOL(dev_disable_lro);
1611 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1614 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1615 * called under RTNL. This is needed if Generic XDP is installed on
1618 static void dev_disable_gro_hw(struct net_device *dev)
1620 dev->wanted_features &= ~NETIF_F_GRO_HW;
1621 netdev_update_features(dev);
1623 if (unlikely(dev->features & NETIF_F_GRO_HW))
1624 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1627 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1630 case NETDEV_##val: \
1631 return "NETDEV_" __stringify(val);
1633 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1634 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1635 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1636 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1637 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1638 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1639 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1640 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1641 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1642 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1643 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1647 return "UNKNOWN_NETDEV_EVENT";
1649 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1651 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1652 struct net_device *dev)
1654 struct netdev_notifier_info info = {
1658 return nb->notifier_call(nb, val, &info);
1661 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1662 struct net_device *dev)
1666 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1667 err = notifier_to_errno(err);
1671 if (!(dev->flags & IFF_UP))
1674 call_netdevice_notifier(nb, NETDEV_UP, dev);
1678 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1679 struct net_device *dev)
1681 if (dev->flags & IFF_UP) {
1682 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1684 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1686 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1689 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1692 struct net_device *dev;
1695 for_each_netdev(net, dev) {
1696 err = call_netdevice_register_notifiers(nb, dev);
1703 for_each_netdev_continue_reverse(net, dev)
1704 call_netdevice_unregister_notifiers(nb, dev);
1708 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1711 struct net_device *dev;
1713 for_each_netdev(net, dev)
1714 call_netdevice_unregister_notifiers(nb, dev);
1717 static int dev_boot_phase = 1;
1720 * register_netdevice_notifier - register a network notifier block
1723 * Register a notifier to be called when network device events occur.
1724 * The notifier passed is linked into the kernel structures and must
1725 * not be reused until it has been unregistered. A negative errno code
1726 * is returned on a failure.
1728 * When registered all registration and up events are replayed
1729 * to the new notifier to allow device to have a race free
1730 * view of the network device list.
1733 int register_netdevice_notifier(struct notifier_block *nb)
1738 /* Close race with setup_net() and cleanup_net() */
1739 down_write(&pernet_ops_rwsem);
1741 err = raw_notifier_chain_register(&netdev_chain, nb);
1747 err = call_netdevice_register_net_notifiers(nb, net);
1754 up_write(&pernet_ops_rwsem);
1758 for_each_net_continue_reverse(net)
1759 call_netdevice_unregister_net_notifiers(nb, net);
1761 raw_notifier_chain_unregister(&netdev_chain, nb);
1764 EXPORT_SYMBOL(register_netdevice_notifier);
1767 * unregister_netdevice_notifier - unregister a network notifier block
1770 * Unregister a notifier previously registered by
1771 * register_netdevice_notifier(). The notifier is unlinked into the
1772 * kernel structures and may then be reused. A negative errno code
1773 * is returned on a failure.
1775 * After unregistering unregister and down device events are synthesized
1776 * for all devices on the device list to the removed notifier to remove
1777 * the need for special case cleanup code.
1780 int unregister_netdevice_notifier(struct notifier_block *nb)
1785 /* Close race with setup_net() and cleanup_net() */
1786 down_write(&pernet_ops_rwsem);
1788 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1793 call_netdevice_unregister_net_notifiers(nb, net);
1797 up_write(&pernet_ops_rwsem);
1800 EXPORT_SYMBOL(unregister_netdevice_notifier);
1802 static int __register_netdevice_notifier_net(struct net *net,
1803 struct notifier_block *nb,
1804 bool ignore_call_fail)
1808 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1814 err = call_netdevice_register_net_notifiers(nb, net);
1815 if (err && !ignore_call_fail)
1816 goto chain_unregister;
1821 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1825 static int __unregister_netdevice_notifier_net(struct net *net,
1826 struct notifier_block *nb)
1830 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1834 call_netdevice_unregister_net_notifiers(nb, net);
1839 * register_netdevice_notifier_net - register a per-netns network notifier block
1840 * @net: network namespace
1843 * Register a notifier to be called when network device events occur.
1844 * The notifier passed is linked into the kernel structures and must
1845 * not be reused until it has been unregistered. A negative errno code
1846 * is returned on a failure.
1848 * When registered all registration and up events are replayed
1849 * to the new notifier to allow device to have a race free
1850 * view of the network device list.
1853 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1858 err = __register_netdevice_notifier_net(net, nb, false);
1862 EXPORT_SYMBOL(register_netdevice_notifier_net);
1865 * unregister_netdevice_notifier_net - unregister a per-netns
1866 * network notifier block
1867 * @net: network namespace
1870 * Unregister a notifier previously registered by
1871 * register_netdevice_notifier_net(). The notifier is unlinked from the
1872 * kernel structures and may then be reused. A negative errno code
1873 * is returned on a failure.
1875 * After unregistering unregister and down device events are synthesized
1876 * for all devices on the device list to the removed notifier to remove
1877 * the need for special case cleanup code.
1880 int unregister_netdevice_notifier_net(struct net *net,
1881 struct notifier_block *nb)
1886 err = __unregister_netdevice_notifier_net(net, nb);
1890 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1892 static void __move_netdevice_notifier_net(struct net *src_net,
1893 struct net *dst_net,
1894 struct notifier_block *nb)
1896 __unregister_netdevice_notifier_net(src_net, nb);
1897 __register_netdevice_notifier_net(dst_net, nb, true);
1900 int register_netdevice_notifier_dev_net(struct net_device *dev,
1901 struct notifier_block *nb,
1902 struct netdev_net_notifier *nn)
1907 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1910 list_add(&nn->list, &dev->net_notifier_list);
1915 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1917 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1918 struct notifier_block *nb,
1919 struct netdev_net_notifier *nn)
1924 list_del(&nn->list);
1925 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1929 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1931 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1934 struct netdev_net_notifier *nn;
1936 list_for_each_entry(nn, &dev->net_notifier_list, list)
1937 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1941 * call_netdevice_notifiers_info - call all network notifier blocks
1942 * @val: value passed unmodified to notifier function
1943 * @info: notifier information data
1945 * Call all network notifier blocks. Parameters and return value
1946 * are as for raw_notifier_call_chain().
1949 int call_netdevice_notifiers_info(unsigned long val,
1950 struct netdev_notifier_info *info)
1952 struct net *net = dev_net(info->dev);
1957 /* Run per-netns notifier block chain first, then run the global one.
1958 * Hopefully, one day, the global one is going to be removed after
1959 * all notifier block registrators get converted to be per-netns.
1961 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1962 if (ret & NOTIFY_STOP_MASK)
1964 return raw_notifier_call_chain(&netdev_chain, val, info);
1968 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1969 * for and rollback on error
1970 * @val_up: value passed unmodified to notifier function
1971 * @val_down: value passed unmodified to the notifier function when
1972 * recovering from an error on @val_up
1973 * @info: notifier information data
1975 * Call all per-netns network notifier blocks, but not notifier blocks on
1976 * the global notifier chain. Parameters and return value are as for
1977 * raw_notifier_call_chain_robust().
1981 call_netdevice_notifiers_info_robust(unsigned long val_up,
1982 unsigned long val_down,
1983 struct netdev_notifier_info *info)
1985 struct net *net = dev_net(info->dev);
1989 return raw_notifier_call_chain_robust(&net->netdev_chain,
1990 val_up, val_down, info);
1993 static int call_netdevice_notifiers_extack(unsigned long val,
1994 struct net_device *dev,
1995 struct netlink_ext_ack *extack)
1997 struct netdev_notifier_info info = {
2002 return call_netdevice_notifiers_info(val, &info);
2006 * call_netdevice_notifiers - call all network notifier blocks
2007 * @val: value passed unmodified to notifier function
2008 * @dev: net_device pointer passed unmodified to notifier function
2010 * Call all network notifier blocks. Parameters and return value
2011 * are as for raw_notifier_call_chain().
2014 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2016 return call_netdevice_notifiers_extack(val, dev, NULL);
2018 EXPORT_SYMBOL(call_netdevice_notifiers);
2021 * call_netdevice_notifiers_mtu - call all network notifier blocks
2022 * @val: value passed unmodified to notifier function
2023 * @dev: net_device pointer passed unmodified to notifier function
2024 * @arg: additional u32 argument passed to the notifier function
2026 * Call all network notifier blocks. Parameters and return value
2027 * are as for raw_notifier_call_chain().
2029 static int call_netdevice_notifiers_mtu(unsigned long val,
2030 struct net_device *dev, u32 arg)
2032 struct netdev_notifier_info_ext info = {
2037 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2039 return call_netdevice_notifiers_info(val, &info.info);
2042 #ifdef CONFIG_NET_INGRESS
2043 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2045 void net_inc_ingress_queue(void)
2047 static_branch_inc(&ingress_needed_key);
2049 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2051 void net_dec_ingress_queue(void)
2053 static_branch_dec(&ingress_needed_key);
2055 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2058 #ifdef CONFIG_NET_EGRESS
2059 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2061 void net_inc_egress_queue(void)
2063 static_branch_inc(&egress_needed_key);
2065 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2067 void net_dec_egress_queue(void)
2069 static_branch_dec(&egress_needed_key);
2071 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2074 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2075 EXPORT_SYMBOL(netstamp_needed_key);
2076 #ifdef CONFIG_JUMP_LABEL
2077 static atomic_t netstamp_needed_deferred;
2078 static atomic_t netstamp_wanted;
2079 static void netstamp_clear(struct work_struct *work)
2081 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2084 wanted = atomic_add_return(deferred, &netstamp_wanted);
2086 static_branch_enable(&netstamp_needed_key);
2088 static_branch_disable(&netstamp_needed_key);
2090 static DECLARE_WORK(netstamp_work, netstamp_clear);
2093 void net_enable_timestamp(void)
2095 #ifdef CONFIG_JUMP_LABEL
2096 int wanted = atomic_read(&netstamp_wanted);
2098 while (wanted > 0) {
2099 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2102 atomic_inc(&netstamp_needed_deferred);
2103 schedule_work(&netstamp_work);
2105 static_branch_inc(&netstamp_needed_key);
2108 EXPORT_SYMBOL(net_enable_timestamp);
2110 void net_disable_timestamp(void)
2112 #ifdef CONFIG_JUMP_LABEL
2113 int wanted = atomic_read(&netstamp_wanted);
2115 while (wanted > 1) {
2116 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2119 atomic_dec(&netstamp_needed_deferred);
2120 schedule_work(&netstamp_work);
2122 static_branch_dec(&netstamp_needed_key);
2125 EXPORT_SYMBOL(net_disable_timestamp);
2127 static inline void net_timestamp_set(struct sk_buff *skb)
2130 skb->mono_delivery_time = 0;
2131 if (static_branch_unlikely(&netstamp_needed_key))
2132 skb->tstamp = ktime_get_real();
2135 #define net_timestamp_check(COND, SKB) \
2136 if (static_branch_unlikely(&netstamp_needed_key)) { \
2137 if ((COND) && !(SKB)->tstamp) \
2138 (SKB)->tstamp = ktime_get_real(); \
2141 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2143 return __is_skb_forwardable(dev, skb, true);
2145 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2147 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2150 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2153 skb->protocol = eth_type_trans(skb, dev);
2154 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2160 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2162 return __dev_forward_skb2(dev, skb, true);
2164 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2167 * dev_forward_skb - loopback an skb to another netif
2169 * @dev: destination network device
2170 * @skb: buffer to forward
2173 * NET_RX_SUCCESS (no congestion)
2174 * NET_RX_DROP (packet was dropped, but freed)
2176 * dev_forward_skb can be used for injecting an skb from the
2177 * start_xmit function of one device into the receive queue
2178 * of another device.
2180 * The receiving device may be in another namespace, so
2181 * we have to clear all information in the skb that could
2182 * impact namespace isolation.
2184 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2186 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2188 EXPORT_SYMBOL_GPL(dev_forward_skb);
2190 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2192 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2195 static inline int deliver_skb(struct sk_buff *skb,
2196 struct packet_type *pt_prev,
2197 struct net_device *orig_dev)
2199 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2201 refcount_inc(&skb->users);
2202 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2205 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2206 struct packet_type **pt,
2207 struct net_device *orig_dev,
2209 struct list_head *ptype_list)
2211 struct packet_type *ptype, *pt_prev = *pt;
2213 list_for_each_entry_rcu(ptype, ptype_list, list) {
2214 if (ptype->type != type)
2217 deliver_skb(skb, pt_prev, orig_dev);
2223 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2225 if (!ptype->af_packet_priv || !skb->sk)
2228 if (ptype->id_match)
2229 return ptype->id_match(ptype, skb->sk);
2230 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2237 * dev_nit_active - return true if any network interface taps are in use
2239 * @dev: network device to check for the presence of taps
2241 bool dev_nit_active(struct net_device *dev)
2243 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2245 EXPORT_SYMBOL_GPL(dev_nit_active);
2248 * Support routine. Sends outgoing frames to any network
2249 * taps currently in use.
2252 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2254 struct packet_type *ptype;
2255 struct sk_buff *skb2 = NULL;
2256 struct packet_type *pt_prev = NULL;
2257 struct list_head *ptype_list = &ptype_all;
2261 list_for_each_entry_rcu(ptype, ptype_list, list) {
2262 if (ptype->ignore_outgoing)
2265 /* Never send packets back to the socket
2266 * they originated from - MvS (miquels@drinkel.ow.org)
2268 if (skb_loop_sk(ptype, skb))
2272 deliver_skb(skb2, pt_prev, skb->dev);
2277 /* need to clone skb, done only once */
2278 skb2 = skb_clone(skb, GFP_ATOMIC);
2282 net_timestamp_set(skb2);
2284 /* skb->nh should be correctly
2285 * set by sender, so that the second statement is
2286 * just protection against buggy protocols.
2288 skb_reset_mac_header(skb2);
2290 if (skb_network_header(skb2) < skb2->data ||
2291 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2292 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2293 ntohs(skb2->protocol),
2295 skb_reset_network_header(skb2);
2298 skb2->transport_header = skb2->network_header;
2299 skb2->pkt_type = PACKET_OUTGOING;
2303 if (ptype_list == &ptype_all) {
2304 ptype_list = &dev->ptype_all;
2309 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2310 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2316 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2319 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2320 * @dev: Network device
2321 * @txq: number of queues available
2323 * If real_num_tx_queues is changed the tc mappings may no longer be
2324 * valid. To resolve this verify the tc mapping remains valid and if
2325 * not NULL the mapping. With no priorities mapping to this
2326 * offset/count pair it will no longer be used. In the worst case TC0
2327 * is invalid nothing can be done so disable priority mappings. If is
2328 * expected that drivers will fix this mapping if they can before
2329 * calling netif_set_real_num_tx_queues.
2331 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2334 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2336 /* If TC0 is invalidated disable TC mapping */
2337 if (tc->offset + tc->count > txq) {
2338 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2343 /* Invalidated prio to tc mappings set to TC0 */
2344 for (i = 1; i < TC_BITMASK + 1; i++) {
2345 int q = netdev_get_prio_tc_map(dev, i);
2347 tc = &dev->tc_to_txq[q];
2348 if (tc->offset + tc->count > txq) {
2349 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2351 netdev_set_prio_tc_map(dev, i, 0);
2356 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2359 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2362 /* walk through the TCs and see if it falls into any of them */
2363 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2364 if ((txq - tc->offset) < tc->count)
2368 /* didn't find it, just return -1 to indicate no match */
2374 EXPORT_SYMBOL(netdev_txq_to_tc);
2377 static struct static_key xps_needed __read_mostly;
2378 static struct static_key xps_rxqs_needed __read_mostly;
2379 static DEFINE_MUTEX(xps_map_mutex);
2380 #define xmap_dereference(P) \
2381 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2383 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2384 struct xps_dev_maps *old_maps, int tci, u16 index)
2386 struct xps_map *map = NULL;
2390 map = xmap_dereference(dev_maps->attr_map[tci]);
2394 for (pos = map->len; pos--;) {
2395 if (map->queues[pos] != index)
2399 map->queues[pos] = map->queues[--map->len];
2404 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2405 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2406 kfree_rcu(map, rcu);
2413 static bool remove_xps_queue_cpu(struct net_device *dev,
2414 struct xps_dev_maps *dev_maps,
2415 int cpu, u16 offset, u16 count)
2417 int num_tc = dev_maps->num_tc;
2418 bool active = false;
2421 for (tci = cpu * num_tc; num_tc--; tci++) {
2424 for (i = count, j = offset; i--; j++) {
2425 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2435 static void reset_xps_maps(struct net_device *dev,
2436 struct xps_dev_maps *dev_maps,
2437 enum xps_map_type type)
2439 static_key_slow_dec_cpuslocked(&xps_needed);
2440 if (type == XPS_RXQS)
2441 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2443 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2445 kfree_rcu(dev_maps, rcu);
2448 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2449 u16 offset, u16 count)
2451 struct xps_dev_maps *dev_maps;
2452 bool active = false;
2455 dev_maps = xmap_dereference(dev->xps_maps[type]);
2459 for (j = 0; j < dev_maps->nr_ids; j++)
2460 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2462 reset_xps_maps(dev, dev_maps, type);
2464 if (type == XPS_CPUS) {
2465 for (i = offset + (count - 1); count--; i--)
2466 netdev_queue_numa_node_write(
2467 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2471 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2474 if (!static_key_false(&xps_needed))
2478 mutex_lock(&xps_map_mutex);
2480 if (static_key_false(&xps_rxqs_needed))
2481 clean_xps_maps(dev, XPS_RXQS, offset, count);
2483 clean_xps_maps(dev, XPS_CPUS, offset, count);
2485 mutex_unlock(&xps_map_mutex);
2489 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2491 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2494 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2495 u16 index, bool is_rxqs_map)
2497 struct xps_map *new_map;
2498 int alloc_len = XPS_MIN_MAP_ALLOC;
2501 for (pos = 0; map && pos < map->len; pos++) {
2502 if (map->queues[pos] != index)
2507 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2509 if (pos < map->alloc_len)
2512 alloc_len = map->alloc_len * 2;
2515 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2519 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2521 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2522 cpu_to_node(attr_index));
2526 for (i = 0; i < pos; i++)
2527 new_map->queues[i] = map->queues[i];
2528 new_map->alloc_len = alloc_len;
2534 /* Copy xps maps at a given index */
2535 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2536 struct xps_dev_maps *new_dev_maps, int index,
2537 int tc, bool skip_tc)
2539 int i, tci = index * dev_maps->num_tc;
2540 struct xps_map *map;
2542 /* copy maps belonging to foreign traffic classes */
2543 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2544 if (i == tc && skip_tc)
2547 /* fill in the new device map from the old device map */
2548 map = xmap_dereference(dev_maps->attr_map[tci]);
2549 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2553 /* Must be called under cpus_read_lock */
2554 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2555 u16 index, enum xps_map_type type)
2557 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2558 const unsigned long *online_mask = NULL;
2559 bool active = false, copy = false;
2560 int i, j, tci, numa_node_id = -2;
2561 int maps_sz, num_tc = 1, tc = 0;
2562 struct xps_map *map, *new_map;
2563 unsigned int nr_ids;
2565 WARN_ON_ONCE(index >= dev->num_tx_queues);
2568 /* Do not allow XPS on subordinate device directly */
2569 num_tc = dev->num_tc;
2573 /* If queue belongs to subordinate dev use its map */
2574 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2576 tc = netdev_txq_to_tc(dev, index);
2581 mutex_lock(&xps_map_mutex);
2583 dev_maps = xmap_dereference(dev->xps_maps[type]);
2584 if (type == XPS_RXQS) {
2585 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2586 nr_ids = dev->num_rx_queues;
2588 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2589 if (num_possible_cpus() > 1)
2590 online_mask = cpumask_bits(cpu_online_mask);
2591 nr_ids = nr_cpu_ids;
2594 if (maps_sz < L1_CACHE_BYTES)
2595 maps_sz = L1_CACHE_BYTES;
2597 /* The old dev_maps could be larger or smaller than the one we're
2598 * setting up now, as dev->num_tc or nr_ids could have been updated in
2599 * between. We could try to be smart, but let's be safe instead and only
2600 * copy foreign traffic classes if the two map sizes match.
2603 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2606 /* allocate memory for queue storage */
2607 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2609 if (!new_dev_maps) {
2610 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2611 if (!new_dev_maps) {
2612 mutex_unlock(&xps_map_mutex);
2616 new_dev_maps->nr_ids = nr_ids;
2617 new_dev_maps->num_tc = num_tc;
2620 tci = j * num_tc + tc;
2621 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2623 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2627 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2631 goto out_no_new_maps;
2634 /* Increment static keys at most once per type */
2635 static_key_slow_inc_cpuslocked(&xps_needed);
2636 if (type == XPS_RXQS)
2637 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2640 for (j = 0; j < nr_ids; j++) {
2641 bool skip_tc = false;
2643 tci = j * num_tc + tc;
2644 if (netif_attr_test_mask(j, mask, nr_ids) &&
2645 netif_attr_test_online(j, online_mask, nr_ids)) {
2646 /* add tx-queue to CPU/rx-queue maps */
2651 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2652 while ((pos < map->len) && (map->queues[pos] != index))
2655 if (pos == map->len)
2656 map->queues[map->len++] = index;
2658 if (type == XPS_CPUS) {
2659 if (numa_node_id == -2)
2660 numa_node_id = cpu_to_node(j);
2661 else if (numa_node_id != cpu_to_node(j))
2668 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2672 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2674 /* Cleanup old maps */
2676 goto out_no_old_maps;
2678 for (j = 0; j < dev_maps->nr_ids; j++) {
2679 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2680 map = xmap_dereference(dev_maps->attr_map[tci]);
2685 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2690 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2691 kfree_rcu(map, rcu);
2695 old_dev_maps = dev_maps;
2698 dev_maps = new_dev_maps;
2702 if (type == XPS_CPUS)
2703 /* update Tx queue numa node */
2704 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2705 (numa_node_id >= 0) ?
2706 numa_node_id : NUMA_NO_NODE);
2711 /* removes tx-queue from unused CPUs/rx-queues */
2712 for (j = 0; j < dev_maps->nr_ids; j++) {
2713 tci = j * dev_maps->num_tc;
2715 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2717 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2718 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2721 active |= remove_xps_queue(dev_maps,
2722 copy ? old_dev_maps : NULL,
2728 kfree_rcu(old_dev_maps, rcu);
2730 /* free map if not active */
2732 reset_xps_maps(dev, dev_maps, type);
2735 mutex_unlock(&xps_map_mutex);
2739 /* remove any maps that we added */
2740 for (j = 0; j < nr_ids; j++) {
2741 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2742 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2744 xmap_dereference(dev_maps->attr_map[tci]) :
2746 if (new_map && new_map != map)
2751 mutex_unlock(&xps_map_mutex);
2753 kfree(new_dev_maps);
2756 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2758 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2764 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2769 EXPORT_SYMBOL(netif_set_xps_queue);
2772 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2774 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2776 /* Unbind any subordinate channels */
2777 while (txq-- != &dev->_tx[0]) {
2779 netdev_unbind_sb_channel(dev, txq->sb_dev);
2783 void netdev_reset_tc(struct net_device *dev)
2786 netif_reset_xps_queues_gt(dev, 0);
2788 netdev_unbind_all_sb_channels(dev);
2790 /* Reset TC configuration of device */
2792 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2793 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2795 EXPORT_SYMBOL(netdev_reset_tc);
2797 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2799 if (tc >= dev->num_tc)
2803 netif_reset_xps_queues(dev, offset, count);
2805 dev->tc_to_txq[tc].count = count;
2806 dev->tc_to_txq[tc].offset = offset;
2809 EXPORT_SYMBOL(netdev_set_tc_queue);
2811 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2813 if (num_tc > TC_MAX_QUEUE)
2817 netif_reset_xps_queues_gt(dev, 0);
2819 netdev_unbind_all_sb_channels(dev);
2821 dev->num_tc = num_tc;
2824 EXPORT_SYMBOL(netdev_set_num_tc);
2826 void netdev_unbind_sb_channel(struct net_device *dev,
2827 struct net_device *sb_dev)
2829 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2832 netif_reset_xps_queues_gt(sb_dev, 0);
2834 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2835 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2837 while (txq-- != &dev->_tx[0]) {
2838 if (txq->sb_dev == sb_dev)
2842 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2844 int netdev_bind_sb_channel_queue(struct net_device *dev,
2845 struct net_device *sb_dev,
2846 u8 tc, u16 count, u16 offset)
2848 /* Make certain the sb_dev and dev are already configured */
2849 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2852 /* We cannot hand out queues we don't have */
2853 if ((offset + count) > dev->real_num_tx_queues)
2856 /* Record the mapping */
2857 sb_dev->tc_to_txq[tc].count = count;
2858 sb_dev->tc_to_txq[tc].offset = offset;
2860 /* Provide a way for Tx queue to find the tc_to_txq map or
2861 * XPS map for itself.
2864 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2868 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2870 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2872 /* Do not use a multiqueue device to represent a subordinate channel */
2873 if (netif_is_multiqueue(dev))
2876 /* We allow channels 1 - 32767 to be used for subordinate channels.
2877 * Channel 0 is meant to be "native" mode and used only to represent
2878 * the main root device. We allow writing 0 to reset the device back
2879 * to normal mode after being used as a subordinate channel.
2881 if (channel > S16_MAX)
2884 dev->num_tc = -channel;
2888 EXPORT_SYMBOL(netdev_set_sb_channel);
2891 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2892 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2894 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2899 disabling = txq < dev->real_num_tx_queues;
2901 if (txq < 1 || txq > dev->num_tx_queues)
2904 if (dev->reg_state == NETREG_REGISTERED ||
2905 dev->reg_state == NETREG_UNREGISTERING) {
2908 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2914 netif_setup_tc(dev, txq);
2916 dev_qdisc_change_real_num_tx(dev, txq);
2918 dev->real_num_tx_queues = txq;
2922 qdisc_reset_all_tx_gt(dev, txq);
2924 netif_reset_xps_queues_gt(dev, txq);
2928 dev->real_num_tx_queues = txq;
2933 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2937 * netif_set_real_num_rx_queues - set actual number of RX queues used
2938 * @dev: Network device
2939 * @rxq: Actual number of RX queues
2941 * This must be called either with the rtnl_lock held or before
2942 * registration of the net device. Returns 0 on success, or a
2943 * negative error code. If called before registration, it always
2946 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2950 if (rxq < 1 || rxq > dev->num_rx_queues)
2953 if (dev->reg_state == NETREG_REGISTERED) {
2956 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2962 dev->real_num_rx_queues = rxq;
2965 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2969 * netif_set_real_num_queues - set actual number of RX and TX queues used
2970 * @dev: Network device
2971 * @txq: Actual number of TX queues
2972 * @rxq: Actual number of RX queues
2974 * Set the real number of both TX and RX queues.
2975 * Does nothing if the number of queues is already correct.
2977 int netif_set_real_num_queues(struct net_device *dev,
2978 unsigned int txq, unsigned int rxq)
2980 unsigned int old_rxq = dev->real_num_rx_queues;
2983 if (txq < 1 || txq > dev->num_tx_queues ||
2984 rxq < 1 || rxq > dev->num_rx_queues)
2987 /* Start from increases, so the error path only does decreases -
2988 * decreases can't fail.
2990 if (rxq > dev->real_num_rx_queues) {
2991 err = netif_set_real_num_rx_queues(dev, rxq);
2995 if (txq > dev->real_num_tx_queues) {
2996 err = netif_set_real_num_tx_queues(dev, txq);
3000 if (rxq < dev->real_num_rx_queues)
3001 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3002 if (txq < dev->real_num_tx_queues)
3003 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3007 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3010 EXPORT_SYMBOL(netif_set_real_num_queues);
3013 * netif_set_tso_max_size() - set the max size of TSO frames supported
3014 * @dev: netdev to update
3015 * @size: max skb->len of a TSO frame
3017 * Set the limit on the size of TSO super-frames the device can handle.
3018 * Unless explicitly set the stack will assume the value of
3019 * %GSO_LEGACY_MAX_SIZE.
3021 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3023 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3024 if (size < READ_ONCE(dev->gso_max_size))
3025 netif_set_gso_max_size(dev, size);
3026 if (size < READ_ONCE(dev->gso_ipv4_max_size))
3027 netif_set_gso_ipv4_max_size(dev, size);
3029 EXPORT_SYMBOL(netif_set_tso_max_size);
3032 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3033 * @dev: netdev to update
3034 * @segs: max number of TCP segments
3036 * Set the limit on the number of TCP segments the device can generate from
3037 * a single TSO super-frame.
3038 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3040 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3042 dev->tso_max_segs = segs;
3043 if (segs < READ_ONCE(dev->gso_max_segs))
3044 netif_set_gso_max_segs(dev, segs);
3046 EXPORT_SYMBOL(netif_set_tso_max_segs);
3049 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3050 * @to: netdev to update
3051 * @from: netdev from which to copy the limits
3053 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3055 netif_set_tso_max_size(to, from->tso_max_size);
3056 netif_set_tso_max_segs(to, from->tso_max_segs);
3058 EXPORT_SYMBOL(netif_inherit_tso_max);
3061 * netif_get_num_default_rss_queues - default number of RSS queues
3063 * Default value is the number of physical cores if there are only 1 or 2, or
3064 * divided by 2 if there are more.
3066 int netif_get_num_default_rss_queues(void)
3071 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3074 cpumask_copy(cpus, cpu_online_mask);
3075 for_each_cpu(cpu, cpus) {
3077 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3079 free_cpumask_var(cpus);
3081 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3083 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3085 static void __netif_reschedule(struct Qdisc *q)
3087 struct softnet_data *sd;
3088 unsigned long flags;
3090 local_irq_save(flags);
3091 sd = this_cpu_ptr(&softnet_data);
3092 q->next_sched = NULL;
3093 *sd->output_queue_tailp = q;
3094 sd->output_queue_tailp = &q->next_sched;
3095 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3096 local_irq_restore(flags);
3099 void __netif_schedule(struct Qdisc *q)
3101 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3102 __netif_reschedule(q);
3104 EXPORT_SYMBOL(__netif_schedule);
3106 struct dev_kfree_skb_cb {
3107 enum skb_drop_reason reason;
3110 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3112 return (struct dev_kfree_skb_cb *)skb->cb;
3115 void netif_schedule_queue(struct netdev_queue *txq)
3118 if (!netif_xmit_stopped(txq)) {
3119 struct Qdisc *q = rcu_dereference(txq->qdisc);
3121 __netif_schedule(q);
3125 EXPORT_SYMBOL(netif_schedule_queue);
3127 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3129 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3133 q = rcu_dereference(dev_queue->qdisc);
3134 __netif_schedule(q);
3138 EXPORT_SYMBOL(netif_tx_wake_queue);
3140 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3142 unsigned long flags;
3147 if (likely(refcount_read(&skb->users) == 1)) {
3149 refcount_set(&skb->users, 0);
3150 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3153 get_kfree_skb_cb(skb)->reason = reason;
3154 local_irq_save(flags);
3155 skb->next = __this_cpu_read(softnet_data.completion_queue);
3156 __this_cpu_write(softnet_data.completion_queue, skb);
3157 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3158 local_irq_restore(flags);
3160 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3162 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3164 if (in_hardirq() || irqs_disabled())
3165 dev_kfree_skb_irq_reason(skb, reason);
3167 kfree_skb_reason(skb, reason);
3169 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3173 * netif_device_detach - mark device as removed
3174 * @dev: network device
3176 * Mark device as removed from system and therefore no longer available.
3178 void netif_device_detach(struct net_device *dev)
3180 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3181 netif_running(dev)) {
3182 netif_tx_stop_all_queues(dev);
3185 EXPORT_SYMBOL(netif_device_detach);
3188 * netif_device_attach - mark device as attached
3189 * @dev: network device
3191 * Mark device as attached from system and restart if needed.
3193 void netif_device_attach(struct net_device *dev)
3195 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3196 netif_running(dev)) {
3197 netif_tx_wake_all_queues(dev);
3198 __netdev_watchdog_up(dev);
3201 EXPORT_SYMBOL(netif_device_attach);
3204 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3205 * to be used as a distribution range.
3207 static u16 skb_tx_hash(const struct net_device *dev,
3208 const struct net_device *sb_dev,
3209 struct sk_buff *skb)
3213 u16 qcount = dev->real_num_tx_queues;
3216 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3218 qoffset = sb_dev->tc_to_txq[tc].offset;
3219 qcount = sb_dev->tc_to_txq[tc].count;
3220 if (unlikely(!qcount)) {
3221 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3222 sb_dev->name, qoffset, tc);
3224 qcount = dev->real_num_tx_queues;
3228 if (skb_rx_queue_recorded(skb)) {
3229 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3230 hash = skb_get_rx_queue(skb);
3231 if (hash >= qoffset)
3233 while (unlikely(hash >= qcount))
3235 return hash + qoffset;
3238 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3241 void skb_warn_bad_offload(const struct sk_buff *skb)
3243 static const netdev_features_t null_features;
3244 struct net_device *dev = skb->dev;
3245 const char *name = "";
3247 if (!net_ratelimit())
3251 if (dev->dev.parent)
3252 name = dev_driver_string(dev->dev.parent);
3254 name = netdev_name(dev);
3256 skb_dump(KERN_WARNING, skb, false);
3257 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3258 name, dev ? &dev->features : &null_features,
3259 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3263 * Invalidate hardware checksum when packet is to be mangled, and
3264 * complete checksum manually on outgoing path.
3266 int skb_checksum_help(struct sk_buff *skb)
3269 int ret = 0, offset;
3271 if (skb->ip_summed == CHECKSUM_COMPLETE)
3272 goto out_set_summed;
3274 if (unlikely(skb_is_gso(skb))) {
3275 skb_warn_bad_offload(skb);
3279 /* Before computing a checksum, we should make sure no frag could
3280 * be modified by an external entity : checksum could be wrong.
3282 if (skb_has_shared_frag(skb)) {
3283 ret = __skb_linearize(skb);
3288 offset = skb_checksum_start_offset(skb);
3290 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3291 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3294 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3296 offset += skb->csum_offset;
3297 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3298 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3301 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3305 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3307 skb->ip_summed = CHECKSUM_NONE;
3311 EXPORT_SYMBOL(skb_checksum_help);
3313 int skb_crc32c_csum_help(struct sk_buff *skb)
3316 int ret = 0, offset, start;
3318 if (skb->ip_summed != CHECKSUM_PARTIAL)
3321 if (unlikely(skb_is_gso(skb)))
3324 /* Before computing a checksum, we should make sure no frag could
3325 * be modified by an external entity : checksum could be wrong.
3327 if (unlikely(skb_has_shared_frag(skb))) {
3328 ret = __skb_linearize(skb);
3332 start = skb_checksum_start_offset(skb);
3333 offset = start + offsetof(struct sctphdr, checksum);
3334 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3339 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3343 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3344 skb->len - start, ~(__u32)0,
3346 *(__le32 *)(skb->data + offset) = crc32c_csum;
3347 skb_reset_csum_not_inet(skb);
3352 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3354 __be16 type = skb->protocol;
3356 /* Tunnel gso handlers can set protocol to ethernet. */
3357 if (type == htons(ETH_P_TEB)) {
3360 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3363 eth = (struct ethhdr *)skb->data;
3364 type = eth->h_proto;
3367 return vlan_get_protocol_and_depth(skb, type, depth);
3371 /* Take action when hardware reception checksum errors are detected. */
3373 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3375 netdev_err(dev, "hw csum failure\n");
3376 skb_dump(KERN_ERR, skb, true);
3380 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3382 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3384 EXPORT_SYMBOL(netdev_rx_csum_fault);
3387 /* XXX: check that highmem exists at all on the given machine. */
3388 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3390 #ifdef CONFIG_HIGHMEM
3393 if (!(dev->features & NETIF_F_HIGHDMA)) {
3394 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3395 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3397 if (PageHighMem(skb_frag_page(frag)))
3405 /* If MPLS offload request, verify we are testing hardware MPLS features
3406 * instead of standard features for the netdev.
3408 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3409 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3410 netdev_features_t features,
3413 if (eth_p_mpls(type))
3414 features &= skb->dev->mpls_features;
3419 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3420 netdev_features_t features,
3427 static netdev_features_t harmonize_features(struct sk_buff *skb,
3428 netdev_features_t features)
3432 type = skb_network_protocol(skb, NULL);
3433 features = net_mpls_features(skb, features, type);
3435 if (skb->ip_summed != CHECKSUM_NONE &&
3436 !can_checksum_protocol(features, type)) {
3437 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3439 if (illegal_highdma(skb->dev, skb))
3440 features &= ~NETIF_F_SG;
3445 netdev_features_t passthru_features_check(struct sk_buff *skb,
3446 struct net_device *dev,
3447 netdev_features_t features)
3451 EXPORT_SYMBOL(passthru_features_check);
3453 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3454 struct net_device *dev,
3455 netdev_features_t features)
3457 return vlan_features_check(skb, features);
3460 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3461 struct net_device *dev,
3462 netdev_features_t features)
3464 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3466 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3467 return features & ~NETIF_F_GSO_MASK;
3469 if (!skb_shinfo(skb)->gso_type) {
3470 skb_warn_bad_offload(skb);
3471 return features & ~NETIF_F_GSO_MASK;
3474 /* Support for GSO partial features requires software
3475 * intervention before we can actually process the packets
3476 * so we need to strip support for any partial features now
3477 * and we can pull them back in after we have partially
3478 * segmented the frame.
3480 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3481 features &= ~dev->gso_partial_features;
3483 /* Make sure to clear the IPv4 ID mangling feature if the
3484 * IPv4 header has the potential to be fragmented.
3486 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3487 struct iphdr *iph = skb->encapsulation ?
3488 inner_ip_hdr(skb) : ip_hdr(skb);
3490 if (!(iph->frag_off & htons(IP_DF)))
3491 features &= ~NETIF_F_TSO_MANGLEID;
3497 netdev_features_t netif_skb_features(struct sk_buff *skb)
3499 struct net_device *dev = skb->dev;
3500 netdev_features_t features = dev->features;
3502 if (skb_is_gso(skb))
3503 features = gso_features_check(skb, dev, features);
3505 /* If encapsulation offload request, verify we are testing
3506 * hardware encapsulation features instead of standard
3507 * features for the netdev
3509 if (skb->encapsulation)
3510 features &= dev->hw_enc_features;
3512 if (skb_vlan_tagged(skb))
3513 features = netdev_intersect_features(features,
3514 dev->vlan_features |
3515 NETIF_F_HW_VLAN_CTAG_TX |
3516 NETIF_F_HW_VLAN_STAG_TX);
3518 if (dev->netdev_ops->ndo_features_check)
3519 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3522 features &= dflt_features_check(skb, dev, features);
3524 return harmonize_features(skb, features);
3526 EXPORT_SYMBOL(netif_skb_features);
3528 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3529 struct netdev_queue *txq, bool more)
3534 if (dev_nit_active(dev))
3535 dev_queue_xmit_nit(skb, dev);
3538 trace_net_dev_start_xmit(skb, dev);
3539 rc = netdev_start_xmit(skb, dev, txq, more);
3540 trace_net_dev_xmit(skb, rc, dev, len);
3545 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3546 struct netdev_queue *txq, int *ret)
3548 struct sk_buff *skb = first;
3549 int rc = NETDEV_TX_OK;
3552 struct sk_buff *next = skb->next;
3554 skb_mark_not_on_list(skb);
3555 rc = xmit_one(skb, dev, txq, next != NULL);
3556 if (unlikely(!dev_xmit_complete(rc))) {
3562 if (netif_tx_queue_stopped(txq) && skb) {
3563 rc = NETDEV_TX_BUSY;
3573 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3574 netdev_features_t features)
3576 if (skb_vlan_tag_present(skb) &&
3577 !vlan_hw_offload_capable(features, skb->vlan_proto))
3578 skb = __vlan_hwaccel_push_inside(skb);
3582 int skb_csum_hwoffload_help(struct sk_buff *skb,
3583 const netdev_features_t features)
3585 if (unlikely(skb_csum_is_sctp(skb)))
3586 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3587 skb_crc32c_csum_help(skb);
3589 if (features & NETIF_F_HW_CSUM)
3592 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3593 switch (skb->csum_offset) {
3594 case offsetof(struct tcphdr, check):
3595 case offsetof(struct udphdr, check):
3600 return skb_checksum_help(skb);
3602 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3604 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3606 netdev_features_t features;
3608 features = netif_skb_features(skb);
3609 skb = validate_xmit_vlan(skb, features);
3613 skb = sk_validate_xmit_skb(skb, dev);
3617 if (netif_needs_gso(skb, features)) {
3618 struct sk_buff *segs;
3620 segs = skb_gso_segment(skb, features);
3628 if (skb_needs_linearize(skb, features) &&
3629 __skb_linearize(skb))
3632 /* If packet is not checksummed and device does not
3633 * support checksumming for this protocol, complete
3634 * checksumming here.
3636 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3637 if (skb->encapsulation)
3638 skb_set_inner_transport_header(skb,
3639 skb_checksum_start_offset(skb));
3641 skb_set_transport_header(skb,
3642 skb_checksum_start_offset(skb));
3643 if (skb_csum_hwoffload_help(skb, features))
3648 skb = validate_xmit_xfrm(skb, features, again);
3655 dev_core_stats_tx_dropped_inc(dev);
3659 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3661 struct sk_buff *next, *head = NULL, *tail;
3663 for (; skb != NULL; skb = next) {
3665 skb_mark_not_on_list(skb);
3667 /* in case skb wont be segmented, point to itself */
3670 skb = validate_xmit_skb(skb, dev, again);
3678 /* If skb was segmented, skb->prev points to
3679 * the last segment. If not, it still contains skb.
3685 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3687 static void qdisc_pkt_len_init(struct sk_buff *skb)
3689 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3691 qdisc_skb_cb(skb)->pkt_len = skb->len;
3693 /* To get more precise estimation of bytes sent on wire,
3694 * we add to pkt_len the headers size of all segments
3696 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3697 u16 gso_segs = shinfo->gso_segs;
3698 unsigned int hdr_len;
3700 /* mac layer + network layer */
3701 hdr_len = skb_transport_offset(skb);
3703 /* + transport layer */
3704 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3705 const struct tcphdr *th;
3706 struct tcphdr _tcphdr;
3708 th = skb_header_pointer(skb, hdr_len,
3709 sizeof(_tcphdr), &_tcphdr);
3711 hdr_len += __tcp_hdrlen(th);
3713 struct udphdr _udphdr;
3715 if (skb_header_pointer(skb, hdr_len,
3716 sizeof(_udphdr), &_udphdr))
3717 hdr_len += sizeof(struct udphdr);
3720 if (shinfo->gso_type & SKB_GSO_DODGY)
3721 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3724 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3728 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3729 struct sk_buff **to_free,
3730 struct netdev_queue *txq)
3734 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3735 if (rc == NET_XMIT_SUCCESS)
3736 trace_qdisc_enqueue(q, txq, skb);
3740 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3741 struct net_device *dev,
3742 struct netdev_queue *txq)
3744 spinlock_t *root_lock = qdisc_lock(q);
3745 struct sk_buff *to_free = NULL;
3749 qdisc_calculate_pkt_len(skb, q);
3751 if (q->flags & TCQ_F_NOLOCK) {
3752 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3753 qdisc_run_begin(q)) {
3754 /* Retest nolock_qdisc_is_empty() within the protection
3755 * of q->seqlock to protect from racing with requeuing.
3757 if (unlikely(!nolock_qdisc_is_empty(q))) {
3758 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3765 qdisc_bstats_cpu_update(q, skb);
3766 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3767 !nolock_qdisc_is_empty(q))
3771 return NET_XMIT_SUCCESS;
3774 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3778 if (unlikely(to_free))
3779 kfree_skb_list_reason(to_free,
3780 SKB_DROP_REASON_QDISC_DROP);
3785 * Heuristic to force contended enqueues to serialize on a
3786 * separate lock before trying to get qdisc main lock.
3787 * This permits qdisc->running owner to get the lock more
3788 * often and dequeue packets faster.
3789 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3790 * and then other tasks will only enqueue packets. The packets will be
3791 * sent after the qdisc owner is scheduled again. To prevent this
3792 * scenario the task always serialize on the lock.
3794 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3795 if (unlikely(contended))
3796 spin_lock(&q->busylock);
3798 spin_lock(root_lock);
3799 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3800 __qdisc_drop(skb, &to_free);
3802 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3803 qdisc_run_begin(q)) {
3805 * This is a work-conserving queue; there are no old skbs
3806 * waiting to be sent out; and the qdisc is not running -
3807 * xmit the skb directly.
3810 qdisc_bstats_update(q, skb);
3812 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3813 if (unlikely(contended)) {
3814 spin_unlock(&q->busylock);
3821 rc = NET_XMIT_SUCCESS;
3823 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3824 if (qdisc_run_begin(q)) {
3825 if (unlikely(contended)) {
3826 spin_unlock(&q->busylock);
3833 spin_unlock(root_lock);
3834 if (unlikely(to_free))
3835 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3836 if (unlikely(contended))
3837 spin_unlock(&q->busylock);
3841 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3842 static void skb_update_prio(struct sk_buff *skb)
3844 const struct netprio_map *map;
3845 const struct sock *sk;
3846 unsigned int prioidx;
3850 map = rcu_dereference_bh(skb->dev->priomap);
3853 sk = skb_to_full_sk(skb);
3857 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3859 if (prioidx < map->priomap_len)
3860 skb->priority = map->priomap[prioidx];
3863 #define skb_update_prio(skb)
3867 * dev_loopback_xmit - loop back @skb
3868 * @net: network namespace this loopback is happening in
3869 * @sk: sk needed to be a netfilter okfn
3870 * @skb: buffer to transmit
3872 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3874 skb_reset_mac_header(skb);
3875 __skb_pull(skb, skb_network_offset(skb));
3876 skb->pkt_type = PACKET_LOOPBACK;
3877 if (skb->ip_summed == CHECKSUM_NONE)
3878 skb->ip_summed = CHECKSUM_UNNECESSARY;
3879 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3884 EXPORT_SYMBOL(dev_loopback_xmit);
3886 #ifdef CONFIG_NET_EGRESS
3887 static struct netdev_queue *
3888 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3890 int qm = skb_get_queue_mapping(skb);
3892 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3895 static bool netdev_xmit_txqueue_skipped(void)
3897 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3900 void netdev_xmit_skip_txqueue(bool skip)
3902 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3904 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3905 #endif /* CONFIG_NET_EGRESS */
3907 #ifdef CONFIG_NET_XGRESS
3908 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb)
3910 int ret = TC_ACT_UNSPEC;
3911 #ifdef CONFIG_NET_CLS_ACT
3912 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
3913 struct tcf_result res;
3918 tc_skb_cb(skb)->mru = 0;
3919 tc_skb_cb(skb)->post_ct = false;
3921 mini_qdisc_bstats_cpu_update(miniq, skb);
3922 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
3923 /* Only tcf related quirks below. */
3926 mini_qdisc_qstats_cpu_drop(miniq);
3929 case TC_ACT_RECLASSIFY:
3930 skb->tc_index = TC_H_MIN(res.classid);
3933 #endif /* CONFIG_NET_CLS_ACT */
3937 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
3941 static_branch_inc(&tcx_needed_key);
3946 static_branch_dec(&tcx_needed_key);
3949 static __always_inline enum tcx_action_base
3950 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
3951 const bool needs_mac)
3953 const struct bpf_mprog_fp *fp;
3954 const struct bpf_prog *prog;
3958 __skb_push(skb, skb->mac_len);
3959 bpf_mprog_foreach_prog(entry, fp, prog) {
3960 bpf_compute_data_pointers(skb);
3961 ret = bpf_prog_run(prog, skb);
3962 if (ret != TCX_NEXT)
3966 __skb_pull(skb, skb->mac_len);
3967 return tcx_action_code(skb, ret);
3970 static __always_inline struct sk_buff *
3971 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3972 struct net_device *orig_dev, bool *another)
3974 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
3980 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3984 qdisc_skb_cb(skb)->pkt_len = skb->len;
3985 tcx_set_ingress(skb, true);
3987 if (static_branch_unlikely(&tcx_needed_key)) {
3988 sch_ret = tcx_run(entry, skb, true);
3989 if (sch_ret != TC_ACT_UNSPEC)
3990 goto ingress_verdict;
3992 sch_ret = tc_run(tcx_entry(entry), skb);
3995 case TC_ACT_REDIRECT:
3996 /* skb_mac_header check was done by BPF, so we can safely
3997 * push the L2 header back before redirecting to another
4000 __skb_push(skb, skb->mac_len);
4001 if (skb_do_redirect(skb) == -EAGAIN) {
4002 __skb_pull(skb, skb->mac_len);
4006 *ret = NET_RX_SUCCESS;
4009 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
4012 /* used by tc_run */
4018 case TC_ACT_CONSUMED:
4019 *ret = NET_RX_SUCCESS;
4026 static __always_inline struct sk_buff *
4027 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4029 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4035 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4036 * already set by the caller.
4038 if (static_branch_unlikely(&tcx_needed_key)) {
4039 sch_ret = tcx_run(entry, skb, false);
4040 if (sch_ret != TC_ACT_UNSPEC)
4041 goto egress_verdict;
4043 sch_ret = tc_run(tcx_entry(entry), skb);
4046 case TC_ACT_REDIRECT:
4047 /* No need to push/pop skb's mac_header here on egress! */
4048 skb_do_redirect(skb);
4049 *ret = NET_XMIT_SUCCESS;
4052 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
4053 *ret = NET_XMIT_DROP;
4055 /* used by tc_run */
4059 *ret = NET_XMIT_SUCCESS;
4066 static __always_inline struct sk_buff *
4067 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4068 struct net_device *orig_dev, bool *another)
4073 static __always_inline struct sk_buff *
4074 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4078 #endif /* CONFIG_NET_XGRESS */
4081 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4082 struct xps_dev_maps *dev_maps, unsigned int tci)
4084 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4085 struct xps_map *map;
4086 int queue_index = -1;
4088 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4091 tci *= dev_maps->num_tc;
4094 map = rcu_dereference(dev_maps->attr_map[tci]);
4097 queue_index = map->queues[0];
4099 queue_index = map->queues[reciprocal_scale(
4100 skb_get_hash(skb), map->len)];
4101 if (unlikely(queue_index >= dev->real_num_tx_queues))
4108 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4109 struct sk_buff *skb)
4112 struct xps_dev_maps *dev_maps;
4113 struct sock *sk = skb->sk;
4114 int queue_index = -1;
4116 if (!static_key_false(&xps_needed))
4120 if (!static_key_false(&xps_rxqs_needed))
4123 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4125 int tci = sk_rx_queue_get(sk);
4128 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4133 if (queue_index < 0) {
4134 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4136 unsigned int tci = skb->sender_cpu - 1;
4138 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4150 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4151 struct net_device *sb_dev)
4155 EXPORT_SYMBOL(dev_pick_tx_zero);
4157 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4158 struct net_device *sb_dev)
4160 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4162 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4164 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4165 struct net_device *sb_dev)
4167 struct sock *sk = skb->sk;
4168 int queue_index = sk_tx_queue_get(sk);
4170 sb_dev = sb_dev ? : dev;
4172 if (queue_index < 0 || skb->ooo_okay ||
4173 queue_index >= dev->real_num_tx_queues) {
4174 int new_index = get_xps_queue(dev, sb_dev, skb);
4177 new_index = skb_tx_hash(dev, sb_dev, skb);
4179 if (queue_index != new_index && sk &&
4181 rcu_access_pointer(sk->sk_dst_cache))
4182 sk_tx_queue_set(sk, new_index);
4184 queue_index = new_index;
4189 EXPORT_SYMBOL(netdev_pick_tx);
4191 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4192 struct sk_buff *skb,
4193 struct net_device *sb_dev)
4195 int queue_index = 0;
4198 u32 sender_cpu = skb->sender_cpu - 1;
4200 if (sender_cpu >= (u32)NR_CPUS)
4201 skb->sender_cpu = raw_smp_processor_id() + 1;
4204 if (dev->real_num_tx_queues != 1) {
4205 const struct net_device_ops *ops = dev->netdev_ops;
4207 if (ops->ndo_select_queue)
4208 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4210 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4212 queue_index = netdev_cap_txqueue(dev, queue_index);
4215 skb_set_queue_mapping(skb, queue_index);
4216 return netdev_get_tx_queue(dev, queue_index);
4220 * __dev_queue_xmit() - transmit a buffer
4221 * @skb: buffer to transmit
4222 * @sb_dev: suboordinate device used for L2 forwarding offload
4224 * Queue a buffer for transmission to a network device. The caller must
4225 * have set the device and priority and built the buffer before calling
4226 * this function. The function can be called from an interrupt.
4228 * When calling this method, interrupts MUST be enabled. This is because
4229 * the BH enable code must have IRQs enabled so that it will not deadlock.
4231 * Regardless of the return value, the skb is consumed, so it is currently
4232 * difficult to retry a send to this method. (You can bump the ref count
4233 * before sending to hold a reference for retry if you are careful.)
4236 * * 0 - buffer successfully transmitted
4237 * * positive qdisc return code - NET_XMIT_DROP etc.
4238 * * negative errno - other errors
4240 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4242 struct net_device *dev = skb->dev;
4243 struct netdev_queue *txq = NULL;
4248 skb_reset_mac_header(skb);
4249 skb_assert_len(skb);
4251 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4252 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4254 /* Disable soft irqs for various locks below. Also
4255 * stops preemption for RCU.
4259 skb_update_prio(skb);
4261 qdisc_pkt_len_init(skb);
4262 tcx_set_ingress(skb, false);
4263 #ifdef CONFIG_NET_EGRESS
4264 if (static_branch_unlikely(&egress_needed_key)) {
4265 if (nf_hook_egress_active()) {
4266 skb = nf_hook_egress(skb, &rc, dev);
4271 netdev_xmit_skip_txqueue(false);
4273 nf_skip_egress(skb, true);
4274 skb = sch_handle_egress(skb, &rc, dev);
4277 nf_skip_egress(skb, false);
4279 if (netdev_xmit_txqueue_skipped())
4280 txq = netdev_tx_queue_mapping(dev, skb);
4283 /* If device/qdisc don't need skb->dst, release it right now while
4284 * its hot in this cpu cache.
4286 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4292 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4294 q = rcu_dereference_bh(txq->qdisc);
4296 trace_net_dev_queue(skb);
4298 rc = __dev_xmit_skb(skb, q, dev, txq);
4302 /* The device has no queue. Common case for software devices:
4303 * loopback, all the sorts of tunnels...
4305 * Really, it is unlikely that netif_tx_lock protection is necessary
4306 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4308 * However, it is possible, that they rely on protection
4311 * Check this and shot the lock. It is not prone from deadlocks.
4312 *Either shot noqueue qdisc, it is even simpler 8)
4314 if (dev->flags & IFF_UP) {
4315 int cpu = smp_processor_id(); /* ok because BHs are off */
4317 /* Other cpus might concurrently change txq->xmit_lock_owner
4318 * to -1 or to their cpu id, but not to our id.
4320 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4321 if (dev_xmit_recursion())
4322 goto recursion_alert;
4324 skb = validate_xmit_skb(skb, dev, &again);
4328 HARD_TX_LOCK(dev, txq, cpu);
4330 if (!netif_xmit_stopped(txq)) {
4331 dev_xmit_recursion_inc();
4332 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4333 dev_xmit_recursion_dec();
4334 if (dev_xmit_complete(rc)) {
4335 HARD_TX_UNLOCK(dev, txq);
4339 HARD_TX_UNLOCK(dev, txq);
4340 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4343 /* Recursion is detected! It is possible,
4347 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4353 rcu_read_unlock_bh();
4355 dev_core_stats_tx_dropped_inc(dev);
4356 kfree_skb_list(skb);
4359 rcu_read_unlock_bh();
4362 EXPORT_SYMBOL(__dev_queue_xmit);
4364 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4366 struct net_device *dev = skb->dev;
4367 struct sk_buff *orig_skb = skb;
4368 struct netdev_queue *txq;
4369 int ret = NETDEV_TX_BUSY;
4372 if (unlikely(!netif_running(dev) ||
4373 !netif_carrier_ok(dev)))
4376 skb = validate_xmit_skb_list(skb, dev, &again);
4377 if (skb != orig_skb)
4380 skb_set_queue_mapping(skb, queue_id);
4381 txq = skb_get_tx_queue(dev, skb);
4385 dev_xmit_recursion_inc();
4386 HARD_TX_LOCK(dev, txq, smp_processor_id());
4387 if (!netif_xmit_frozen_or_drv_stopped(txq))
4388 ret = netdev_start_xmit(skb, dev, txq, false);
4389 HARD_TX_UNLOCK(dev, txq);
4390 dev_xmit_recursion_dec();
4395 dev_core_stats_tx_dropped_inc(dev);
4396 kfree_skb_list(skb);
4397 return NET_XMIT_DROP;
4399 EXPORT_SYMBOL(__dev_direct_xmit);
4401 /*************************************************************************
4403 *************************************************************************/
4405 int netdev_max_backlog __read_mostly = 1000;
4406 EXPORT_SYMBOL(netdev_max_backlog);
4408 int netdev_tstamp_prequeue __read_mostly = 1;
4409 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4410 int netdev_budget __read_mostly = 300;
4411 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4412 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4413 int weight_p __read_mostly = 64; /* old backlog weight */
4414 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4415 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4416 int dev_rx_weight __read_mostly = 64;
4417 int dev_tx_weight __read_mostly = 64;
4419 /* Called with irq disabled */
4420 static inline void ____napi_schedule(struct softnet_data *sd,
4421 struct napi_struct *napi)
4423 struct task_struct *thread;
4425 lockdep_assert_irqs_disabled();
4427 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4428 /* Paired with smp_mb__before_atomic() in
4429 * napi_enable()/dev_set_threaded().
4430 * Use READ_ONCE() to guarantee a complete
4431 * read on napi->thread. Only call
4432 * wake_up_process() when it's not NULL.
4434 thread = READ_ONCE(napi->thread);
4436 /* Avoid doing set_bit() if the thread is in
4437 * INTERRUPTIBLE state, cause napi_thread_wait()
4438 * makes sure to proceed with napi polling
4439 * if the thread is explicitly woken from here.
4441 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4442 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4443 wake_up_process(thread);
4448 list_add_tail(&napi->poll_list, &sd->poll_list);
4449 WRITE_ONCE(napi->list_owner, smp_processor_id());
4450 /* If not called from net_rx_action()
4451 * we have to raise NET_RX_SOFTIRQ.
4453 if (!sd->in_net_rx_action)
4454 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4459 /* One global table that all flow-based protocols share. */
4460 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4461 EXPORT_SYMBOL(rps_sock_flow_table);
4462 u32 rps_cpu_mask __read_mostly;
4463 EXPORT_SYMBOL(rps_cpu_mask);
4465 struct static_key_false rps_needed __read_mostly;
4466 EXPORT_SYMBOL(rps_needed);
4467 struct static_key_false rfs_needed __read_mostly;
4468 EXPORT_SYMBOL(rfs_needed);
4470 static struct rps_dev_flow *
4471 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4472 struct rps_dev_flow *rflow, u16 next_cpu)
4474 if (next_cpu < nr_cpu_ids) {
4475 #ifdef CONFIG_RFS_ACCEL
4476 struct netdev_rx_queue *rxqueue;
4477 struct rps_dev_flow_table *flow_table;
4478 struct rps_dev_flow *old_rflow;
4483 /* Should we steer this flow to a different hardware queue? */
4484 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4485 !(dev->features & NETIF_F_NTUPLE))
4487 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4488 if (rxq_index == skb_get_rx_queue(skb))
4491 rxqueue = dev->_rx + rxq_index;
4492 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4495 flow_id = skb_get_hash(skb) & flow_table->mask;
4496 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4497 rxq_index, flow_id);
4501 rflow = &flow_table->flows[flow_id];
4503 if (old_rflow->filter == rflow->filter)
4504 old_rflow->filter = RPS_NO_FILTER;
4508 per_cpu(softnet_data, next_cpu).input_queue_head;
4511 rflow->cpu = next_cpu;
4516 * get_rps_cpu is called from netif_receive_skb and returns the target
4517 * CPU from the RPS map of the receiving queue for a given skb.
4518 * rcu_read_lock must be held on entry.
4520 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4521 struct rps_dev_flow **rflowp)
4523 const struct rps_sock_flow_table *sock_flow_table;
4524 struct netdev_rx_queue *rxqueue = dev->_rx;
4525 struct rps_dev_flow_table *flow_table;
4526 struct rps_map *map;
4531 if (skb_rx_queue_recorded(skb)) {
4532 u16 index = skb_get_rx_queue(skb);
4534 if (unlikely(index >= dev->real_num_rx_queues)) {
4535 WARN_ONCE(dev->real_num_rx_queues > 1,
4536 "%s received packet on queue %u, but number "
4537 "of RX queues is %u\n",
4538 dev->name, index, dev->real_num_rx_queues);
4544 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4546 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4547 map = rcu_dereference(rxqueue->rps_map);
4548 if (!flow_table && !map)
4551 skb_reset_network_header(skb);
4552 hash = skb_get_hash(skb);
4556 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4557 if (flow_table && sock_flow_table) {
4558 struct rps_dev_flow *rflow;
4562 /* First check into global flow table if there is a match.
4563 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4565 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4566 if ((ident ^ hash) & ~rps_cpu_mask)
4569 next_cpu = ident & rps_cpu_mask;
4571 /* OK, now we know there is a match,
4572 * we can look at the local (per receive queue) flow table
4574 rflow = &flow_table->flows[hash & flow_table->mask];
4578 * If the desired CPU (where last recvmsg was done) is
4579 * different from current CPU (one in the rx-queue flow
4580 * table entry), switch if one of the following holds:
4581 * - Current CPU is unset (>= nr_cpu_ids).
4582 * - Current CPU is offline.
4583 * - The current CPU's queue tail has advanced beyond the
4584 * last packet that was enqueued using this table entry.
4585 * This guarantees that all previous packets for the flow
4586 * have been dequeued, thus preserving in order delivery.
4588 if (unlikely(tcpu != next_cpu) &&
4589 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4590 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4591 rflow->last_qtail)) >= 0)) {
4593 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4596 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4606 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4607 if (cpu_online(tcpu)) {
4617 #ifdef CONFIG_RFS_ACCEL
4620 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4621 * @dev: Device on which the filter was set
4622 * @rxq_index: RX queue index
4623 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4624 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4626 * Drivers that implement ndo_rx_flow_steer() should periodically call
4627 * this function for each installed filter and remove the filters for
4628 * which it returns %true.
4630 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4631 u32 flow_id, u16 filter_id)
4633 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4634 struct rps_dev_flow_table *flow_table;
4635 struct rps_dev_flow *rflow;
4640 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4641 if (flow_table && flow_id <= flow_table->mask) {
4642 rflow = &flow_table->flows[flow_id];
4643 cpu = READ_ONCE(rflow->cpu);
4644 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4645 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4646 rflow->last_qtail) <
4647 (int)(10 * flow_table->mask)))
4653 EXPORT_SYMBOL(rps_may_expire_flow);
4655 #endif /* CONFIG_RFS_ACCEL */
4657 /* Called from hardirq (IPI) context */
4658 static void rps_trigger_softirq(void *data)
4660 struct softnet_data *sd = data;
4662 ____napi_schedule(sd, &sd->backlog);
4666 #endif /* CONFIG_RPS */
4668 /* Called from hardirq (IPI) context */
4669 static void trigger_rx_softirq(void *data)
4671 struct softnet_data *sd = data;
4673 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4674 smp_store_release(&sd->defer_ipi_scheduled, 0);
4678 * After we queued a packet into sd->input_pkt_queue,
4679 * we need to make sure this queue is serviced soon.
4681 * - If this is another cpu queue, link it to our rps_ipi_list,
4682 * and make sure we will process rps_ipi_list from net_rx_action().
4684 * - If this is our own queue, NAPI schedule our backlog.
4685 * Note that this also raises NET_RX_SOFTIRQ.
4687 static void napi_schedule_rps(struct softnet_data *sd)
4689 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4693 sd->rps_ipi_next = mysd->rps_ipi_list;
4694 mysd->rps_ipi_list = sd;
4696 /* If not called from net_rx_action() or napi_threaded_poll()
4697 * we have to raise NET_RX_SOFTIRQ.
4699 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4700 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4703 #endif /* CONFIG_RPS */
4704 __napi_schedule_irqoff(&mysd->backlog);
4707 #ifdef CONFIG_NET_FLOW_LIMIT
4708 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4711 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4713 #ifdef CONFIG_NET_FLOW_LIMIT
4714 struct sd_flow_limit *fl;
4715 struct softnet_data *sd;
4716 unsigned int old_flow, new_flow;
4718 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4721 sd = this_cpu_ptr(&softnet_data);
4724 fl = rcu_dereference(sd->flow_limit);
4726 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4727 old_flow = fl->history[fl->history_head];
4728 fl->history[fl->history_head] = new_flow;
4731 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4733 if (likely(fl->buckets[old_flow]))
4734 fl->buckets[old_flow]--;
4736 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4748 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4749 * queue (may be a remote CPU queue).
4751 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4752 unsigned int *qtail)
4754 enum skb_drop_reason reason;
4755 struct softnet_data *sd;
4756 unsigned long flags;
4759 reason = SKB_DROP_REASON_NOT_SPECIFIED;
4760 sd = &per_cpu(softnet_data, cpu);
4762 rps_lock_irqsave(sd, &flags);
4763 if (!netif_running(skb->dev))
4765 qlen = skb_queue_len(&sd->input_pkt_queue);
4766 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4769 __skb_queue_tail(&sd->input_pkt_queue, skb);
4770 input_queue_tail_incr_save(sd, qtail);
4771 rps_unlock_irq_restore(sd, &flags);
4772 return NET_RX_SUCCESS;
4775 /* Schedule NAPI for backlog device
4776 * We can use non atomic operation since we own the queue lock
4778 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4779 napi_schedule_rps(sd);
4782 reason = SKB_DROP_REASON_CPU_BACKLOG;
4786 rps_unlock_irq_restore(sd, &flags);
4788 dev_core_stats_rx_dropped_inc(skb->dev);
4789 kfree_skb_reason(skb, reason);
4793 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4795 struct net_device *dev = skb->dev;
4796 struct netdev_rx_queue *rxqueue;
4800 if (skb_rx_queue_recorded(skb)) {
4801 u16 index = skb_get_rx_queue(skb);
4803 if (unlikely(index >= dev->real_num_rx_queues)) {
4804 WARN_ONCE(dev->real_num_rx_queues > 1,
4805 "%s received packet on queue %u, but number "
4806 "of RX queues is %u\n",
4807 dev->name, index, dev->real_num_rx_queues);
4809 return rxqueue; /* Return first rxqueue */
4816 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4817 struct bpf_prog *xdp_prog)
4819 void *orig_data, *orig_data_end, *hard_start;
4820 struct netdev_rx_queue *rxqueue;
4821 bool orig_bcast, orig_host;
4822 u32 mac_len, frame_sz;
4823 __be16 orig_eth_type;
4828 /* The XDP program wants to see the packet starting at the MAC
4831 mac_len = skb->data - skb_mac_header(skb);
4832 hard_start = skb->data - skb_headroom(skb);
4834 /* SKB "head" area always have tailroom for skb_shared_info */
4835 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4836 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4838 rxqueue = netif_get_rxqueue(skb);
4839 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4840 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4841 skb_headlen(skb) + mac_len, true);
4843 orig_data_end = xdp->data_end;
4844 orig_data = xdp->data;
4845 eth = (struct ethhdr *)xdp->data;
4846 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4847 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4848 orig_eth_type = eth->h_proto;
4850 act = bpf_prog_run_xdp(xdp_prog, xdp);
4852 /* check if bpf_xdp_adjust_head was used */
4853 off = xdp->data - orig_data;
4856 __skb_pull(skb, off);
4858 __skb_push(skb, -off);
4860 skb->mac_header += off;
4861 skb_reset_network_header(skb);
4864 /* check if bpf_xdp_adjust_tail was used */
4865 off = xdp->data_end - orig_data_end;
4867 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4868 skb->len += off; /* positive on grow, negative on shrink */
4871 /* check if XDP changed eth hdr such SKB needs update */
4872 eth = (struct ethhdr *)xdp->data;
4873 if ((orig_eth_type != eth->h_proto) ||
4874 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4875 skb->dev->dev_addr)) ||
4876 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4877 __skb_push(skb, ETH_HLEN);
4878 skb->pkt_type = PACKET_HOST;
4879 skb->protocol = eth_type_trans(skb, skb->dev);
4882 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4883 * before calling us again on redirect path. We do not call do_redirect
4884 * as we leave that up to the caller.
4886 * Caller is responsible for managing lifetime of skb (i.e. calling
4887 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4892 __skb_push(skb, mac_len);
4895 metalen = xdp->data - xdp->data_meta;
4897 skb_metadata_set(skb, metalen);
4904 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4905 struct xdp_buff *xdp,
4906 struct bpf_prog *xdp_prog)
4910 /* Reinjected packets coming from act_mirred or similar should
4911 * not get XDP generic processing.
4913 if (skb_is_redirected(skb))
4916 /* XDP packets must be linear and must have sufficient headroom
4917 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4918 * native XDP provides, thus we need to do it here as well.
4920 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4921 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4922 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4923 int troom = skb->tail + skb->data_len - skb->end;
4925 /* In case we have to go down the path and also linearize,
4926 * then lets do the pskb_expand_head() work just once here.
4928 if (pskb_expand_head(skb,
4929 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4930 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4932 if (skb_linearize(skb))
4936 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4943 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4946 trace_xdp_exception(skb->dev, xdp_prog, act);
4957 /* When doing generic XDP we have to bypass the qdisc layer and the
4958 * network taps in order to match in-driver-XDP behavior. This also means
4959 * that XDP packets are able to starve other packets going through a qdisc,
4960 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4961 * queues, so they do not have this starvation issue.
4963 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4965 struct net_device *dev = skb->dev;
4966 struct netdev_queue *txq;
4967 bool free_skb = true;
4970 txq = netdev_core_pick_tx(dev, skb, NULL);
4971 cpu = smp_processor_id();
4972 HARD_TX_LOCK(dev, txq, cpu);
4973 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4974 rc = netdev_start_xmit(skb, dev, txq, 0);
4975 if (dev_xmit_complete(rc))
4978 HARD_TX_UNLOCK(dev, txq);
4980 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4981 dev_core_stats_tx_dropped_inc(dev);
4986 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4988 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4991 struct xdp_buff xdp;
4995 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4996 if (act != XDP_PASS) {
4999 err = xdp_do_generic_redirect(skb->dev, skb,
5005 generic_xdp_tx(skb, xdp_prog);
5013 kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
5016 EXPORT_SYMBOL_GPL(do_xdp_generic);
5018 static int netif_rx_internal(struct sk_buff *skb)
5022 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5024 trace_netif_rx(skb);
5027 if (static_branch_unlikely(&rps_needed)) {
5028 struct rps_dev_flow voidflow, *rflow = &voidflow;
5033 cpu = get_rps_cpu(skb->dev, skb, &rflow);
5035 cpu = smp_processor_id();
5037 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5045 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5051 * __netif_rx - Slightly optimized version of netif_rx
5052 * @skb: buffer to post
5054 * This behaves as netif_rx except that it does not disable bottom halves.
5055 * As a result this function may only be invoked from the interrupt context
5056 * (either hard or soft interrupt).
5058 int __netif_rx(struct sk_buff *skb)
5062 lockdep_assert_once(hardirq_count() | softirq_count());
5064 trace_netif_rx_entry(skb);
5065 ret = netif_rx_internal(skb);
5066 trace_netif_rx_exit(ret);
5069 EXPORT_SYMBOL(__netif_rx);
5072 * netif_rx - post buffer to the network code
5073 * @skb: buffer to post
5075 * This function receives a packet from a device driver and queues it for
5076 * the upper (protocol) levels to process via the backlog NAPI device. It
5077 * always succeeds. The buffer may be dropped during processing for
5078 * congestion control or by the protocol layers.
5079 * The network buffer is passed via the backlog NAPI device. Modern NIC
5080 * driver should use NAPI and GRO.
5081 * This function can used from interrupt and from process context. The
5082 * caller from process context must not disable interrupts before invoking
5086 * NET_RX_SUCCESS (no congestion)
5087 * NET_RX_DROP (packet was dropped)
5090 int netif_rx(struct sk_buff *skb)
5092 bool need_bh_off = !(hardirq_count() | softirq_count());
5097 trace_netif_rx_entry(skb);
5098 ret = netif_rx_internal(skb);
5099 trace_netif_rx_exit(ret);
5104 EXPORT_SYMBOL(netif_rx);
5106 static __latent_entropy void net_tx_action(struct softirq_action *h)
5108 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5110 if (sd->completion_queue) {
5111 struct sk_buff *clist;
5113 local_irq_disable();
5114 clist = sd->completion_queue;
5115 sd->completion_queue = NULL;
5119 struct sk_buff *skb = clist;
5121 clist = clist->next;
5123 WARN_ON(refcount_read(&skb->users));
5124 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5125 trace_consume_skb(skb, net_tx_action);
5127 trace_kfree_skb(skb, net_tx_action,
5128 get_kfree_skb_cb(skb)->reason);
5130 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5133 __napi_kfree_skb(skb,
5134 get_kfree_skb_cb(skb)->reason);
5138 if (sd->output_queue) {
5141 local_irq_disable();
5142 head = sd->output_queue;
5143 sd->output_queue = NULL;
5144 sd->output_queue_tailp = &sd->output_queue;
5150 struct Qdisc *q = head;
5151 spinlock_t *root_lock = NULL;
5153 head = head->next_sched;
5155 /* We need to make sure head->next_sched is read
5156 * before clearing __QDISC_STATE_SCHED
5158 smp_mb__before_atomic();
5160 if (!(q->flags & TCQ_F_NOLOCK)) {
5161 root_lock = qdisc_lock(q);
5162 spin_lock(root_lock);
5163 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5165 /* There is a synchronize_net() between
5166 * STATE_DEACTIVATED flag being set and
5167 * qdisc_reset()/some_qdisc_is_busy() in
5168 * dev_deactivate(), so we can safely bail out
5169 * early here to avoid data race between
5170 * qdisc_deactivate() and some_qdisc_is_busy()
5171 * for lockless qdisc.
5173 clear_bit(__QDISC_STATE_SCHED, &q->state);
5177 clear_bit(__QDISC_STATE_SCHED, &q->state);
5180 spin_unlock(root_lock);
5186 xfrm_dev_backlog(sd);
5189 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5190 /* This hook is defined here for ATM LANE */
5191 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5192 unsigned char *addr) __read_mostly;
5193 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5197 * netdev_is_rx_handler_busy - check if receive handler is registered
5198 * @dev: device to check
5200 * Check if a receive handler is already registered for a given device.
5201 * Return true if there one.
5203 * The caller must hold the rtnl_mutex.
5205 bool netdev_is_rx_handler_busy(struct net_device *dev)
5208 return dev && rtnl_dereference(dev->rx_handler);
5210 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5213 * netdev_rx_handler_register - register receive handler
5214 * @dev: device to register a handler for
5215 * @rx_handler: receive handler to register
5216 * @rx_handler_data: data pointer that is used by rx handler
5218 * Register a receive handler for a device. This handler will then be
5219 * called from __netif_receive_skb. A negative errno code is returned
5222 * The caller must hold the rtnl_mutex.
5224 * For a general description of rx_handler, see enum rx_handler_result.
5226 int netdev_rx_handler_register(struct net_device *dev,
5227 rx_handler_func_t *rx_handler,
5228 void *rx_handler_data)
5230 if (netdev_is_rx_handler_busy(dev))
5233 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5236 /* Note: rx_handler_data must be set before rx_handler */
5237 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5238 rcu_assign_pointer(dev->rx_handler, rx_handler);
5242 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5245 * netdev_rx_handler_unregister - unregister receive handler
5246 * @dev: device to unregister a handler from
5248 * Unregister a receive handler from a device.
5250 * The caller must hold the rtnl_mutex.
5252 void netdev_rx_handler_unregister(struct net_device *dev)
5256 RCU_INIT_POINTER(dev->rx_handler, NULL);
5257 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5258 * section has a guarantee to see a non NULL rx_handler_data
5262 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5264 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5267 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5268 * the special handling of PFMEMALLOC skbs.
5270 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5272 switch (skb->protocol) {
5273 case htons(ETH_P_ARP):
5274 case htons(ETH_P_IP):
5275 case htons(ETH_P_IPV6):
5276 case htons(ETH_P_8021Q):
5277 case htons(ETH_P_8021AD):
5284 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5285 int *ret, struct net_device *orig_dev)
5287 if (nf_hook_ingress_active(skb)) {
5291 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5296 ingress_retval = nf_hook_ingress(skb);
5298 return ingress_retval;
5303 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5304 struct packet_type **ppt_prev)
5306 struct packet_type *ptype, *pt_prev;
5307 rx_handler_func_t *rx_handler;
5308 struct sk_buff *skb = *pskb;
5309 struct net_device *orig_dev;
5310 bool deliver_exact = false;
5311 int ret = NET_RX_DROP;
5314 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5316 trace_netif_receive_skb(skb);
5318 orig_dev = skb->dev;
5320 skb_reset_network_header(skb);
5321 if (!skb_transport_header_was_set(skb))
5322 skb_reset_transport_header(skb);
5323 skb_reset_mac_len(skb);
5328 skb->skb_iif = skb->dev->ifindex;
5330 __this_cpu_inc(softnet_data.processed);
5332 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5336 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5339 if (ret2 != XDP_PASS) {
5345 if (eth_type_vlan(skb->protocol)) {
5346 skb = skb_vlan_untag(skb);
5351 if (skb_skip_tc_classify(skb))
5357 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5359 ret = deliver_skb(skb, pt_prev, orig_dev);
5363 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5365 ret = deliver_skb(skb, pt_prev, orig_dev);
5370 #ifdef CONFIG_NET_INGRESS
5371 if (static_branch_unlikely(&ingress_needed_key)) {
5372 bool another = false;
5374 nf_skip_egress(skb, true);
5375 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5382 nf_skip_egress(skb, false);
5383 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5387 skb_reset_redirect(skb);
5389 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5392 if (skb_vlan_tag_present(skb)) {
5394 ret = deliver_skb(skb, pt_prev, orig_dev);
5397 if (vlan_do_receive(&skb))
5399 else if (unlikely(!skb))
5403 rx_handler = rcu_dereference(skb->dev->rx_handler);
5406 ret = deliver_skb(skb, pt_prev, orig_dev);
5409 switch (rx_handler(&skb)) {
5410 case RX_HANDLER_CONSUMED:
5411 ret = NET_RX_SUCCESS;
5413 case RX_HANDLER_ANOTHER:
5415 case RX_HANDLER_EXACT:
5416 deliver_exact = true;
5418 case RX_HANDLER_PASS:
5425 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5427 if (skb_vlan_tag_get_id(skb)) {
5428 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5431 skb->pkt_type = PACKET_OTHERHOST;
5432 } else if (eth_type_vlan(skb->protocol)) {
5433 /* Outer header is 802.1P with vlan 0, inner header is
5434 * 802.1Q or 802.1AD and vlan_do_receive() above could
5435 * not find vlan dev for vlan id 0.
5437 __vlan_hwaccel_clear_tag(skb);
5438 skb = skb_vlan_untag(skb);
5441 if (vlan_do_receive(&skb))
5442 /* After stripping off 802.1P header with vlan 0
5443 * vlan dev is found for inner header.
5446 else if (unlikely(!skb))
5449 /* We have stripped outer 802.1P vlan 0 header.
5450 * But could not find vlan dev.
5451 * check again for vlan id to set OTHERHOST.
5455 /* Note: we might in the future use prio bits
5456 * and set skb->priority like in vlan_do_receive()
5457 * For the time being, just ignore Priority Code Point
5459 __vlan_hwaccel_clear_tag(skb);
5462 type = skb->protocol;
5464 /* deliver only exact match when indicated */
5465 if (likely(!deliver_exact)) {
5466 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5467 &ptype_base[ntohs(type) &
5471 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5472 &orig_dev->ptype_specific);
5474 if (unlikely(skb->dev != orig_dev)) {
5475 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5476 &skb->dev->ptype_specific);
5480 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5482 *ppt_prev = pt_prev;
5486 dev_core_stats_rx_dropped_inc(skb->dev);
5488 dev_core_stats_rx_nohandler_inc(skb->dev);
5489 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5490 /* Jamal, now you will not able to escape explaining
5491 * me how you were going to use this. :-)
5497 /* The invariant here is that if *ppt_prev is not NULL
5498 * then skb should also be non-NULL.
5500 * Apparently *ppt_prev assignment above holds this invariant due to
5501 * skb dereferencing near it.
5507 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5509 struct net_device *orig_dev = skb->dev;
5510 struct packet_type *pt_prev = NULL;
5513 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5515 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5516 skb->dev, pt_prev, orig_dev);
5521 * netif_receive_skb_core - special purpose version of netif_receive_skb
5522 * @skb: buffer to process
5524 * More direct receive version of netif_receive_skb(). It should
5525 * only be used by callers that have a need to skip RPS and Generic XDP.
5526 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5528 * This function may only be called from softirq context and interrupts
5529 * should be enabled.
5531 * Return values (usually ignored):
5532 * NET_RX_SUCCESS: no congestion
5533 * NET_RX_DROP: packet was dropped
5535 int netif_receive_skb_core(struct sk_buff *skb)
5540 ret = __netif_receive_skb_one_core(skb, false);
5545 EXPORT_SYMBOL(netif_receive_skb_core);
5547 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5548 struct packet_type *pt_prev,
5549 struct net_device *orig_dev)
5551 struct sk_buff *skb, *next;
5555 if (list_empty(head))
5557 if (pt_prev->list_func != NULL)
5558 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5559 ip_list_rcv, head, pt_prev, orig_dev);
5561 list_for_each_entry_safe(skb, next, head, list) {
5562 skb_list_del_init(skb);
5563 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5567 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5569 /* Fast-path assumptions:
5570 * - There is no RX handler.
5571 * - Only one packet_type matches.
5572 * If either of these fails, we will end up doing some per-packet
5573 * processing in-line, then handling the 'last ptype' for the whole
5574 * sublist. This can't cause out-of-order delivery to any single ptype,
5575 * because the 'last ptype' must be constant across the sublist, and all
5576 * other ptypes are handled per-packet.
5578 /* Current (common) ptype of sublist */
5579 struct packet_type *pt_curr = NULL;
5580 /* Current (common) orig_dev of sublist */
5581 struct net_device *od_curr = NULL;
5582 struct list_head sublist;
5583 struct sk_buff *skb, *next;
5585 INIT_LIST_HEAD(&sublist);
5586 list_for_each_entry_safe(skb, next, head, list) {
5587 struct net_device *orig_dev = skb->dev;
5588 struct packet_type *pt_prev = NULL;
5590 skb_list_del_init(skb);
5591 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5594 if (pt_curr != pt_prev || od_curr != orig_dev) {
5595 /* dispatch old sublist */
5596 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5597 /* start new sublist */
5598 INIT_LIST_HEAD(&sublist);
5602 list_add_tail(&skb->list, &sublist);
5605 /* dispatch final sublist */
5606 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5609 static int __netif_receive_skb(struct sk_buff *skb)
5613 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5614 unsigned int noreclaim_flag;
5617 * PFMEMALLOC skbs are special, they should
5618 * - be delivered to SOCK_MEMALLOC sockets only
5619 * - stay away from userspace
5620 * - have bounded memory usage
5622 * Use PF_MEMALLOC as this saves us from propagating the allocation
5623 * context down to all allocation sites.
5625 noreclaim_flag = memalloc_noreclaim_save();
5626 ret = __netif_receive_skb_one_core(skb, true);
5627 memalloc_noreclaim_restore(noreclaim_flag);
5629 ret = __netif_receive_skb_one_core(skb, false);
5634 static void __netif_receive_skb_list(struct list_head *head)
5636 unsigned long noreclaim_flag = 0;
5637 struct sk_buff *skb, *next;
5638 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5640 list_for_each_entry_safe(skb, next, head, list) {
5641 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5642 struct list_head sublist;
5644 /* Handle the previous sublist */
5645 list_cut_before(&sublist, head, &skb->list);
5646 if (!list_empty(&sublist))
5647 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5648 pfmemalloc = !pfmemalloc;
5649 /* See comments in __netif_receive_skb */
5651 noreclaim_flag = memalloc_noreclaim_save();
5653 memalloc_noreclaim_restore(noreclaim_flag);
5656 /* Handle the remaining sublist */
5657 if (!list_empty(head))
5658 __netif_receive_skb_list_core(head, pfmemalloc);
5659 /* Restore pflags */
5661 memalloc_noreclaim_restore(noreclaim_flag);
5664 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5666 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5667 struct bpf_prog *new = xdp->prog;
5670 switch (xdp->command) {
5671 case XDP_SETUP_PROG:
5672 rcu_assign_pointer(dev->xdp_prog, new);
5677 static_branch_dec(&generic_xdp_needed_key);
5678 } else if (new && !old) {
5679 static_branch_inc(&generic_xdp_needed_key);
5680 dev_disable_lro(dev);
5681 dev_disable_gro_hw(dev);
5693 static int netif_receive_skb_internal(struct sk_buff *skb)
5697 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5699 if (skb_defer_rx_timestamp(skb))
5700 return NET_RX_SUCCESS;
5704 if (static_branch_unlikely(&rps_needed)) {
5705 struct rps_dev_flow voidflow, *rflow = &voidflow;
5706 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5709 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5715 ret = __netif_receive_skb(skb);
5720 void netif_receive_skb_list_internal(struct list_head *head)
5722 struct sk_buff *skb, *next;
5723 struct list_head sublist;
5725 INIT_LIST_HEAD(&sublist);
5726 list_for_each_entry_safe(skb, next, head, list) {
5727 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5728 skb_list_del_init(skb);
5729 if (!skb_defer_rx_timestamp(skb))
5730 list_add_tail(&skb->list, &sublist);
5732 list_splice_init(&sublist, head);
5736 if (static_branch_unlikely(&rps_needed)) {
5737 list_for_each_entry_safe(skb, next, head, list) {
5738 struct rps_dev_flow voidflow, *rflow = &voidflow;
5739 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5742 /* Will be handled, remove from list */
5743 skb_list_del_init(skb);
5744 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5749 __netif_receive_skb_list(head);
5754 * netif_receive_skb - process receive buffer from network
5755 * @skb: buffer to process
5757 * netif_receive_skb() is the main receive data processing function.
5758 * It always succeeds. The buffer may be dropped during processing
5759 * for congestion control or by the protocol layers.
5761 * This function may only be called from softirq context and interrupts
5762 * should be enabled.
5764 * Return values (usually ignored):
5765 * NET_RX_SUCCESS: no congestion
5766 * NET_RX_DROP: packet was dropped
5768 int netif_receive_skb(struct sk_buff *skb)
5772 trace_netif_receive_skb_entry(skb);
5774 ret = netif_receive_skb_internal(skb);
5775 trace_netif_receive_skb_exit(ret);
5779 EXPORT_SYMBOL(netif_receive_skb);
5782 * netif_receive_skb_list - process many receive buffers from network
5783 * @head: list of skbs to process.
5785 * Since return value of netif_receive_skb() is normally ignored, and
5786 * wouldn't be meaningful for a list, this function returns void.
5788 * This function may only be called from softirq context and interrupts
5789 * should be enabled.
5791 void netif_receive_skb_list(struct list_head *head)
5793 struct sk_buff *skb;
5795 if (list_empty(head))
5797 if (trace_netif_receive_skb_list_entry_enabled()) {
5798 list_for_each_entry(skb, head, list)
5799 trace_netif_receive_skb_list_entry(skb);
5801 netif_receive_skb_list_internal(head);
5802 trace_netif_receive_skb_list_exit(0);
5804 EXPORT_SYMBOL(netif_receive_skb_list);
5806 static DEFINE_PER_CPU(struct work_struct, flush_works);
5808 /* Network device is going away, flush any packets still pending */
5809 static void flush_backlog(struct work_struct *work)
5811 struct sk_buff *skb, *tmp;
5812 struct softnet_data *sd;
5815 sd = this_cpu_ptr(&softnet_data);
5817 rps_lock_irq_disable(sd);
5818 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5819 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5820 __skb_unlink(skb, &sd->input_pkt_queue);
5821 dev_kfree_skb_irq(skb);
5822 input_queue_head_incr(sd);
5825 rps_unlock_irq_enable(sd);
5827 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5828 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5829 __skb_unlink(skb, &sd->process_queue);
5831 input_queue_head_incr(sd);
5837 static bool flush_required(int cpu)
5839 #if IS_ENABLED(CONFIG_RPS)
5840 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5843 rps_lock_irq_disable(sd);
5845 /* as insertion into process_queue happens with the rps lock held,
5846 * process_queue access may race only with dequeue
5848 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5849 !skb_queue_empty_lockless(&sd->process_queue);
5850 rps_unlock_irq_enable(sd);
5854 /* without RPS we can't safely check input_pkt_queue: during a
5855 * concurrent remote skb_queue_splice() we can detect as empty both
5856 * input_pkt_queue and process_queue even if the latter could end-up
5857 * containing a lot of packets.
5862 static void flush_all_backlogs(void)
5864 static cpumask_t flush_cpus;
5867 /* since we are under rtnl lock protection we can use static data
5868 * for the cpumask and avoid allocating on stack the possibly
5875 cpumask_clear(&flush_cpus);
5876 for_each_online_cpu(cpu) {
5877 if (flush_required(cpu)) {
5878 queue_work_on(cpu, system_highpri_wq,
5879 per_cpu_ptr(&flush_works, cpu));
5880 cpumask_set_cpu(cpu, &flush_cpus);
5884 /* we can have in flight packet[s] on the cpus we are not flushing,
5885 * synchronize_net() in unregister_netdevice_many() will take care of
5888 for_each_cpu(cpu, &flush_cpus)
5889 flush_work(per_cpu_ptr(&flush_works, cpu));
5894 static void net_rps_send_ipi(struct softnet_data *remsd)
5898 struct softnet_data *next = remsd->rps_ipi_next;
5900 if (cpu_online(remsd->cpu))
5901 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5908 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5909 * Note: called with local irq disabled, but exits with local irq enabled.
5911 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5914 struct softnet_data *remsd = sd->rps_ipi_list;
5917 sd->rps_ipi_list = NULL;
5921 /* Send pending IPI's to kick RPS processing on remote cpus. */
5922 net_rps_send_ipi(remsd);
5928 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5931 return sd->rps_ipi_list != NULL;
5937 static int process_backlog(struct napi_struct *napi, int quota)
5939 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5943 /* Check if we have pending ipi, its better to send them now,
5944 * not waiting net_rx_action() end.
5946 if (sd_has_rps_ipi_waiting(sd)) {
5947 local_irq_disable();
5948 net_rps_action_and_irq_enable(sd);
5951 napi->weight = READ_ONCE(dev_rx_weight);
5953 struct sk_buff *skb;
5955 while ((skb = __skb_dequeue(&sd->process_queue))) {
5957 __netif_receive_skb(skb);
5959 input_queue_head_incr(sd);
5960 if (++work >= quota)
5965 rps_lock_irq_disable(sd);
5966 if (skb_queue_empty(&sd->input_pkt_queue)) {
5968 * Inline a custom version of __napi_complete().
5969 * only current cpu owns and manipulates this napi,
5970 * and NAPI_STATE_SCHED is the only possible flag set
5972 * We can use a plain write instead of clear_bit(),
5973 * and we dont need an smp_mb() memory barrier.
5978 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5979 &sd->process_queue);
5981 rps_unlock_irq_enable(sd);
5988 * __napi_schedule - schedule for receive
5989 * @n: entry to schedule
5991 * The entry's receive function will be scheduled to run.
5992 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5994 void __napi_schedule(struct napi_struct *n)
5996 unsigned long flags;
5998 local_irq_save(flags);
5999 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6000 local_irq_restore(flags);
6002 EXPORT_SYMBOL(__napi_schedule);
6005 * napi_schedule_prep - check if napi can be scheduled
6008 * Test if NAPI routine is already running, and if not mark
6009 * it as running. This is used as a condition variable to
6010 * insure only one NAPI poll instance runs. We also make
6011 * sure there is no pending NAPI disable.
6013 bool napi_schedule_prep(struct napi_struct *n)
6015 unsigned long new, val = READ_ONCE(n->state);
6018 if (unlikely(val & NAPIF_STATE_DISABLE))
6020 new = val | NAPIF_STATE_SCHED;
6022 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6023 * This was suggested by Alexander Duyck, as compiler
6024 * emits better code than :
6025 * if (val & NAPIF_STATE_SCHED)
6026 * new |= NAPIF_STATE_MISSED;
6028 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6030 } while (!try_cmpxchg(&n->state, &val, new));
6032 return !(val & NAPIF_STATE_SCHED);
6034 EXPORT_SYMBOL(napi_schedule_prep);
6037 * __napi_schedule_irqoff - schedule for receive
6038 * @n: entry to schedule
6040 * Variant of __napi_schedule() assuming hard irqs are masked.
6042 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6043 * because the interrupt disabled assumption might not be true
6044 * due to force-threaded interrupts and spinlock substitution.
6046 void __napi_schedule_irqoff(struct napi_struct *n)
6048 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6049 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6053 EXPORT_SYMBOL(__napi_schedule_irqoff);
6055 bool napi_complete_done(struct napi_struct *n, int work_done)
6057 unsigned long flags, val, new, timeout = 0;
6061 * 1) Don't let napi dequeue from the cpu poll list
6062 * just in case its running on a different cpu.
6063 * 2) If we are busy polling, do nothing here, we have
6064 * the guarantee we will be called later.
6066 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6067 NAPIF_STATE_IN_BUSY_POLL)))
6072 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6073 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6075 if (n->defer_hard_irqs_count > 0) {
6076 n->defer_hard_irqs_count--;
6077 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6081 if (n->gro_bitmask) {
6082 /* When the NAPI instance uses a timeout and keeps postponing
6083 * it, we need to bound somehow the time packets are kept in
6086 napi_gro_flush(n, !!timeout);
6091 if (unlikely(!list_empty(&n->poll_list))) {
6092 /* If n->poll_list is not empty, we need to mask irqs */
6093 local_irq_save(flags);
6094 list_del_init(&n->poll_list);
6095 local_irq_restore(flags);
6097 WRITE_ONCE(n->list_owner, -1);
6099 val = READ_ONCE(n->state);
6101 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6103 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6104 NAPIF_STATE_SCHED_THREADED |
6105 NAPIF_STATE_PREFER_BUSY_POLL);
6107 /* If STATE_MISSED was set, leave STATE_SCHED set,
6108 * because we will call napi->poll() one more time.
6109 * This C code was suggested by Alexander Duyck to help gcc.
6111 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6113 } while (!try_cmpxchg(&n->state, &val, new));
6115 if (unlikely(val & NAPIF_STATE_MISSED)) {
6121 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6122 HRTIMER_MODE_REL_PINNED);
6125 EXPORT_SYMBOL(napi_complete_done);
6127 /* must be called under rcu_read_lock(), as we dont take a reference */
6128 static struct napi_struct *napi_by_id(unsigned int napi_id)
6130 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6131 struct napi_struct *napi;
6133 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6134 if (napi->napi_id == napi_id)
6140 #if defined(CONFIG_NET_RX_BUSY_POLL)
6142 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6144 if (!skip_schedule) {
6145 gro_normal_list(napi);
6146 __napi_schedule(napi);
6150 if (napi->gro_bitmask) {
6151 /* flush too old packets
6152 * If HZ < 1000, flush all packets.
6154 napi_gro_flush(napi, HZ >= 1000);
6157 gro_normal_list(napi);
6158 clear_bit(NAPI_STATE_SCHED, &napi->state);
6161 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6164 bool skip_schedule = false;
6165 unsigned long timeout;
6168 /* Busy polling means there is a high chance device driver hard irq
6169 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6170 * set in napi_schedule_prep().
6171 * Since we are about to call napi->poll() once more, we can safely
6172 * clear NAPI_STATE_MISSED.
6174 * Note: x86 could use a single "lock and ..." instruction
6175 * to perform these two clear_bit()
6177 clear_bit(NAPI_STATE_MISSED, &napi->state);
6178 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6182 if (prefer_busy_poll) {
6183 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6184 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6185 if (napi->defer_hard_irqs_count && timeout) {
6186 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6187 skip_schedule = true;
6191 /* All we really want here is to re-enable device interrupts.
6192 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6194 rc = napi->poll(napi, budget);
6195 /* We can't gro_normal_list() here, because napi->poll() might have
6196 * rearmed the napi (napi_complete_done()) in which case it could
6197 * already be running on another CPU.
6199 trace_napi_poll(napi, rc, budget);
6200 netpoll_poll_unlock(have_poll_lock);
6202 __busy_poll_stop(napi, skip_schedule);
6206 void napi_busy_loop(unsigned int napi_id,
6207 bool (*loop_end)(void *, unsigned long),
6208 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6210 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6211 int (*napi_poll)(struct napi_struct *napi, int budget);
6212 void *have_poll_lock = NULL;
6213 struct napi_struct *napi;
6220 napi = napi_by_id(napi_id);
6224 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6231 unsigned long val = READ_ONCE(napi->state);
6233 /* If multiple threads are competing for this napi,
6234 * we avoid dirtying napi->state as much as we can.
6236 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6237 NAPIF_STATE_IN_BUSY_POLL)) {
6238 if (prefer_busy_poll)
6239 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6242 if (cmpxchg(&napi->state, val,
6243 val | NAPIF_STATE_IN_BUSY_POLL |
6244 NAPIF_STATE_SCHED) != val) {
6245 if (prefer_busy_poll)
6246 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6249 have_poll_lock = netpoll_poll_lock(napi);
6250 napi_poll = napi->poll;
6252 work = napi_poll(napi, budget);
6253 trace_napi_poll(napi, work, budget);
6254 gro_normal_list(napi);
6257 __NET_ADD_STATS(dev_net(napi->dev),
6258 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6261 if (!loop_end || loop_end(loop_end_arg, start_time))
6264 if (unlikely(need_resched())) {
6266 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6267 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6271 if (loop_end(loop_end_arg, start_time))
6278 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6279 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6284 EXPORT_SYMBOL(napi_busy_loop);
6286 #endif /* CONFIG_NET_RX_BUSY_POLL */
6288 static void napi_hash_add(struct napi_struct *napi)
6290 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6293 spin_lock(&napi_hash_lock);
6295 /* 0..NR_CPUS range is reserved for sender_cpu use */
6297 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6298 napi_gen_id = MIN_NAPI_ID;
6299 } while (napi_by_id(napi_gen_id));
6300 napi->napi_id = napi_gen_id;
6302 hlist_add_head_rcu(&napi->napi_hash_node,
6303 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6305 spin_unlock(&napi_hash_lock);
6308 /* Warning : caller is responsible to make sure rcu grace period
6309 * is respected before freeing memory containing @napi
6311 static void napi_hash_del(struct napi_struct *napi)
6313 spin_lock(&napi_hash_lock);
6315 hlist_del_init_rcu(&napi->napi_hash_node);
6317 spin_unlock(&napi_hash_lock);
6320 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6322 struct napi_struct *napi;
6324 napi = container_of(timer, struct napi_struct, timer);
6326 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6327 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6329 if (!napi_disable_pending(napi) &&
6330 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6331 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6332 __napi_schedule_irqoff(napi);
6335 return HRTIMER_NORESTART;
6338 static void init_gro_hash(struct napi_struct *napi)
6342 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6343 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6344 napi->gro_hash[i].count = 0;
6346 napi->gro_bitmask = 0;
6349 int dev_set_threaded(struct net_device *dev, bool threaded)
6351 struct napi_struct *napi;
6354 if (dev->threaded == threaded)
6358 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6359 if (!napi->thread) {
6360 err = napi_kthread_create(napi);
6369 dev->threaded = threaded;
6371 /* Make sure kthread is created before THREADED bit
6374 smp_mb__before_atomic();
6376 /* Setting/unsetting threaded mode on a napi might not immediately
6377 * take effect, if the current napi instance is actively being
6378 * polled. In this case, the switch between threaded mode and
6379 * softirq mode will happen in the next round of napi_schedule().
6380 * This should not cause hiccups/stalls to the live traffic.
6382 list_for_each_entry(napi, &dev->napi_list, dev_list)
6383 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6387 EXPORT_SYMBOL(dev_set_threaded);
6389 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6390 int (*poll)(struct napi_struct *, int), int weight)
6392 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6395 INIT_LIST_HEAD(&napi->poll_list);
6396 INIT_HLIST_NODE(&napi->napi_hash_node);
6397 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6398 napi->timer.function = napi_watchdog;
6399 init_gro_hash(napi);
6401 INIT_LIST_HEAD(&napi->rx_list);
6404 if (weight > NAPI_POLL_WEIGHT)
6405 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6407 napi->weight = weight;
6409 #ifdef CONFIG_NETPOLL
6410 napi->poll_owner = -1;
6412 napi->list_owner = -1;
6413 set_bit(NAPI_STATE_SCHED, &napi->state);
6414 set_bit(NAPI_STATE_NPSVC, &napi->state);
6415 list_add_rcu(&napi->dev_list, &dev->napi_list);
6416 napi_hash_add(napi);
6417 napi_get_frags_check(napi);
6418 /* Create kthread for this napi if dev->threaded is set.
6419 * Clear dev->threaded if kthread creation failed so that
6420 * threaded mode will not be enabled in napi_enable().
6422 if (dev->threaded && napi_kthread_create(napi))
6425 EXPORT_SYMBOL(netif_napi_add_weight);
6427 void napi_disable(struct napi_struct *n)
6429 unsigned long val, new;
6432 set_bit(NAPI_STATE_DISABLE, &n->state);
6434 val = READ_ONCE(n->state);
6436 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6437 usleep_range(20, 200);
6438 val = READ_ONCE(n->state);
6441 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6442 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6443 } while (!try_cmpxchg(&n->state, &val, new));
6445 hrtimer_cancel(&n->timer);
6447 clear_bit(NAPI_STATE_DISABLE, &n->state);
6449 EXPORT_SYMBOL(napi_disable);
6452 * napi_enable - enable NAPI scheduling
6455 * Resume NAPI from being scheduled on this context.
6456 * Must be paired with napi_disable.
6458 void napi_enable(struct napi_struct *n)
6460 unsigned long new, val = READ_ONCE(n->state);
6463 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6465 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6466 if (n->dev->threaded && n->thread)
6467 new |= NAPIF_STATE_THREADED;
6468 } while (!try_cmpxchg(&n->state, &val, new));
6470 EXPORT_SYMBOL(napi_enable);
6472 static void flush_gro_hash(struct napi_struct *napi)
6476 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6477 struct sk_buff *skb, *n;
6479 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6481 napi->gro_hash[i].count = 0;
6485 /* Must be called in process context */
6486 void __netif_napi_del(struct napi_struct *napi)
6488 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6491 napi_hash_del(napi);
6492 list_del_rcu(&napi->dev_list);
6493 napi_free_frags(napi);
6495 flush_gro_hash(napi);
6496 napi->gro_bitmask = 0;
6499 kthread_stop(napi->thread);
6500 napi->thread = NULL;
6503 EXPORT_SYMBOL(__netif_napi_del);
6505 static int __napi_poll(struct napi_struct *n, bool *repoll)
6511 /* This NAPI_STATE_SCHED test is for avoiding a race
6512 * with netpoll's poll_napi(). Only the entity which
6513 * obtains the lock and sees NAPI_STATE_SCHED set will
6514 * actually make the ->poll() call. Therefore we avoid
6515 * accidentally calling ->poll() when NAPI is not scheduled.
6518 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6519 work = n->poll(n, weight);
6520 trace_napi_poll(n, work, weight);
6523 if (unlikely(work > weight))
6524 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6525 n->poll, work, weight);
6527 if (likely(work < weight))
6530 /* Drivers must not modify the NAPI state if they
6531 * consume the entire weight. In such cases this code
6532 * still "owns" the NAPI instance and therefore can
6533 * move the instance around on the list at-will.
6535 if (unlikely(napi_disable_pending(n))) {
6540 /* The NAPI context has more processing work, but busy-polling
6541 * is preferred. Exit early.
6543 if (napi_prefer_busy_poll(n)) {
6544 if (napi_complete_done(n, work)) {
6545 /* If timeout is not set, we need to make sure
6546 * that the NAPI is re-scheduled.
6553 if (n->gro_bitmask) {
6554 /* flush too old packets
6555 * If HZ < 1000, flush all packets.
6557 napi_gro_flush(n, HZ >= 1000);
6562 /* Some drivers may have called napi_schedule
6563 * prior to exhausting their budget.
6565 if (unlikely(!list_empty(&n->poll_list))) {
6566 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6567 n->dev ? n->dev->name : "backlog");
6576 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6578 bool do_repoll = false;
6582 list_del_init(&n->poll_list);
6584 have = netpoll_poll_lock(n);
6586 work = __napi_poll(n, &do_repoll);
6589 list_add_tail(&n->poll_list, repoll);
6591 netpoll_poll_unlock(have);
6596 static int napi_thread_wait(struct napi_struct *napi)
6600 set_current_state(TASK_INTERRUPTIBLE);
6602 while (!kthread_should_stop()) {
6603 /* Testing SCHED_THREADED bit here to make sure the current
6604 * kthread owns this napi and could poll on this napi.
6605 * Testing SCHED bit is not enough because SCHED bit might be
6606 * set by some other busy poll thread or by napi_disable().
6608 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6609 WARN_ON(!list_empty(&napi->poll_list));
6610 __set_current_state(TASK_RUNNING);
6615 /* woken being true indicates this thread owns this napi. */
6617 set_current_state(TASK_INTERRUPTIBLE);
6619 __set_current_state(TASK_RUNNING);
6624 static void skb_defer_free_flush(struct softnet_data *sd)
6626 struct sk_buff *skb, *next;
6628 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6629 if (!READ_ONCE(sd->defer_list))
6632 spin_lock(&sd->defer_lock);
6633 skb = sd->defer_list;
6634 sd->defer_list = NULL;
6635 sd->defer_count = 0;
6636 spin_unlock(&sd->defer_lock);
6638 while (skb != NULL) {
6640 napi_consume_skb(skb, 1);
6645 static int napi_threaded_poll(void *data)
6647 struct napi_struct *napi = data;
6648 struct softnet_data *sd;
6651 while (!napi_thread_wait(napi)) {
6653 bool repoll = false;
6656 sd = this_cpu_ptr(&softnet_data);
6657 sd->in_napi_threaded_poll = true;
6659 have = netpoll_poll_lock(napi);
6660 __napi_poll(napi, &repoll);
6661 netpoll_poll_unlock(have);
6663 sd->in_napi_threaded_poll = false;
6666 if (sd_has_rps_ipi_waiting(sd)) {
6667 local_irq_disable();
6668 net_rps_action_and_irq_enable(sd);
6670 skb_defer_free_flush(sd);
6682 static __latent_entropy void net_rx_action(struct softirq_action *h)
6684 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6685 unsigned long time_limit = jiffies +
6686 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6687 int budget = READ_ONCE(netdev_budget);
6692 sd->in_net_rx_action = true;
6693 local_irq_disable();
6694 list_splice_init(&sd->poll_list, &list);
6698 struct napi_struct *n;
6700 skb_defer_free_flush(sd);
6702 if (list_empty(&list)) {
6703 if (list_empty(&repoll)) {
6704 sd->in_net_rx_action = false;
6706 /* We need to check if ____napi_schedule()
6707 * had refilled poll_list while
6708 * sd->in_net_rx_action was true.
6710 if (!list_empty(&sd->poll_list))
6712 if (!sd_has_rps_ipi_waiting(sd))
6718 n = list_first_entry(&list, struct napi_struct, poll_list);
6719 budget -= napi_poll(n, &repoll);
6721 /* If softirq window is exhausted then punt.
6722 * Allow this to run for 2 jiffies since which will allow
6723 * an average latency of 1.5/HZ.
6725 if (unlikely(budget <= 0 ||
6726 time_after_eq(jiffies, time_limit))) {
6732 local_irq_disable();
6734 list_splice_tail_init(&sd->poll_list, &list);
6735 list_splice_tail(&repoll, &list);
6736 list_splice(&list, &sd->poll_list);
6737 if (!list_empty(&sd->poll_list))
6738 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6740 sd->in_net_rx_action = false;
6742 net_rps_action_and_irq_enable(sd);
6746 struct netdev_adjacent {
6747 struct net_device *dev;
6748 netdevice_tracker dev_tracker;
6750 /* upper master flag, there can only be one master device per list */
6753 /* lookup ignore flag */
6756 /* counter for the number of times this device was added to us */
6759 /* private field for the users */
6762 struct list_head list;
6763 struct rcu_head rcu;
6766 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6767 struct list_head *adj_list)
6769 struct netdev_adjacent *adj;
6771 list_for_each_entry(adj, adj_list, list) {
6772 if (adj->dev == adj_dev)
6778 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6779 struct netdev_nested_priv *priv)
6781 struct net_device *dev = (struct net_device *)priv->data;
6783 return upper_dev == dev;
6787 * netdev_has_upper_dev - Check if device is linked to an upper device
6789 * @upper_dev: upper device to check
6791 * Find out if a device is linked to specified upper device and return true
6792 * in case it is. Note that this checks only immediate upper device,
6793 * not through a complete stack of devices. The caller must hold the RTNL lock.
6795 bool netdev_has_upper_dev(struct net_device *dev,
6796 struct net_device *upper_dev)
6798 struct netdev_nested_priv priv = {
6799 .data = (void *)upper_dev,
6804 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6807 EXPORT_SYMBOL(netdev_has_upper_dev);
6810 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6812 * @upper_dev: upper device to check
6814 * Find out if a device is linked to specified upper device and return true
6815 * in case it is. Note that this checks the entire upper device chain.
6816 * The caller must hold rcu lock.
6819 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6820 struct net_device *upper_dev)
6822 struct netdev_nested_priv priv = {
6823 .data = (void *)upper_dev,
6826 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6829 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6832 * netdev_has_any_upper_dev - Check if device is linked to some device
6835 * Find out if a device is linked to an upper device and return true in case
6836 * it is. The caller must hold the RTNL lock.
6838 bool netdev_has_any_upper_dev(struct net_device *dev)
6842 return !list_empty(&dev->adj_list.upper);
6844 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6847 * netdev_master_upper_dev_get - Get master upper device
6850 * Find a master upper device and return pointer to it or NULL in case
6851 * it's not there. The caller must hold the RTNL lock.
6853 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6855 struct netdev_adjacent *upper;
6859 if (list_empty(&dev->adj_list.upper))
6862 upper = list_first_entry(&dev->adj_list.upper,
6863 struct netdev_adjacent, list);
6864 if (likely(upper->master))
6868 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6870 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6872 struct netdev_adjacent *upper;
6876 if (list_empty(&dev->adj_list.upper))
6879 upper = list_first_entry(&dev->adj_list.upper,
6880 struct netdev_adjacent, list);
6881 if (likely(upper->master) && !upper->ignore)
6887 * netdev_has_any_lower_dev - Check if device is linked to some device
6890 * Find out if a device is linked to a lower device and return true in case
6891 * it is. The caller must hold the RTNL lock.
6893 static bool netdev_has_any_lower_dev(struct net_device *dev)
6897 return !list_empty(&dev->adj_list.lower);
6900 void *netdev_adjacent_get_private(struct list_head *adj_list)
6902 struct netdev_adjacent *adj;
6904 adj = list_entry(adj_list, struct netdev_adjacent, list);
6906 return adj->private;
6908 EXPORT_SYMBOL(netdev_adjacent_get_private);
6911 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6913 * @iter: list_head ** of the current position
6915 * Gets the next device from the dev's upper list, starting from iter
6916 * position. The caller must hold RCU read lock.
6918 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6919 struct list_head **iter)
6921 struct netdev_adjacent *upper;
6923 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6925 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6927 if (&upper->list == &dev->adj_list.upper)
6930 *iter = &upper->list;
6934 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6936 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6937 struct list_head **iter,
6940 struct netdev_adjacent *upper;
6942 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6944 if (&upper->list == &dev->adj_list.upper)
6947 *iter = &upper->list;
6948 *ignore = upper->ignore;
6953 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6954 struct list_head **iter)
6956 struct netdev_adjacent *upper;
6958 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6960 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6962 if (&upper->list == &dev->adj_list.upper)
6965 *iter = &upper->list;
6970 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6971 int (*fn)(struct net_device *dev,
6972 struct netdev_nested_priv *priv),
6973 struct netdev_nested_priv *priv)
6975 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6976 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6981 iter = &dev->adj_list.upper;
6985 ret = fn(now, priv);
6992 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6999 niter = &udev->adj_list.upper;
7000 dev_stack[cur] = now;
7001 iter_stack[cur++] = iter;
7008 next = dev_stack[--cur];
7009 niter = iter_stack[cur];
7019 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7020 int (*fn)(struct net_device *dev,
7021 struct netdev_nested_priv *priv),
7022 struct netdev_nested_priv *priv)
7024 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7025 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7029 iter = &dev->adj_list.upper;
7033 ret = fn(now, priv);
7040 udev = netdev_next_upper_dev_rcu(now, &iter);
7045 niter = &udev->adj_list.upper;
7046 dev_stack[cur] = now;
7047 iter_stack[cur++] = iter;
7054 next = dev_stack[--cur];
7055 niter = iter_stack[cur];
7064 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7066 static bool __netdev_has_upper_dev(struct net_device *dev,
7067 struct net_device *upper_dev)
7069 struct netdev_nested_priv priv = {
7071 .data = (void *)upper_dev,
7076 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7081 * netdev_lower_get_next_private - Get the next ->private from the
7082 * lower neighbour list
7084 * @iter: list_head ** of the current position
7086 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7087 * list, starting from iter position. The caller must hold either hold the
7088 * RTNL lock or its own locking that guarantees that the neighbour lower
7089 * list will remain unchanged.
7091 void *netdev_lower_get_next_private(struct net_device *dev,
7092 struct list_head **iter)
7094 struct netdev_adjacent *lower;
7096 lower = list_entry(*iter, struct netdev_adjacent, list);
7098 if (&lower->list == &dev->adj_list.lower)
7101 *iter = lower->list.next;
7103 return lower->private;
7105 EXPORT_SYMBOL(netdev_lower_get_next_private);
7108 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7109 * lower neighbour list, RCU
7112 * @iter: list_head ** of the current position
7114 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7115 * list, starting from iter position. The caller must hold RCU read lock.
7117 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7118 struct list_head **iter)
7120 struct netdev_adjacent *lower;
7122 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7124 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7126 if (&lower->list == &dev->adj_list.lower)
7129 *iter = &lower->list;
7131 return lower->private;
7133 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7136 * netdev_lower_get_next - Get the next device from the lower neighbour
7139 * @iter: list_head ** of the current position
7141 * Gets the next netdev_adjacent from the dev's lower neighbour
7142 * list, starting from iter position. The caller must hold RTNL lock or
7143 * its own locking that guarantees that the neighbour lower
7144 * list will remain unchanged.
7146 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7148 struct netdev_adjacent *lower;
7150 lower = list_entry(*iter, struct netdev_adjacent, list);
7152 if (&lower->list == &dev->adj_list.lower)
7155 *iter = lower->list.next;
7159 EXPORT_SYMBOL(netdev_lower_get_next);
7161 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7162 struct list_head **iter)
7164 struct netdev_adjacent *lower;
7166 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7168 if (&lower->list == &dev->adj_list.lower)
7171 *iter = &lower->list;
7176 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7177 struct list_head **iter,
7180 struct netdev_adjacent *lower;
7182 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7184 if (&lower->list == &dev->adj_list.lower)
7187 *iter = &lower->list;
7188 *ignore = lower->ignore;
7193 int netdev_walk_all_lower_dev(struct net_device *dev,
7194 int (*fn)(struct net_device *dev,
7195 struct netdev_nested_priv *priv),
7196 struct netdev_nested_priv *priv)
7198 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7199 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7203 iter = &dev->adj_list.lower;
7207 ret = fn(now, priv);
7214 ldev = netdev_next_lower_dev(now, &iter);
7219 niter = &ldev->adj_list.lower;
7220 dev_stack[cur] = now;
7221 iter_stack[cur++] = iter;
7228 next = dev_stack[--cur];
7229 niter = iter_stack[cur];
7238 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7240 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7241 int (*fn)(struct net_device *dev,
7242 struct netdev_nested_priv *priv),
7243 struct netdev_nested_priv *priv)
7245 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7246 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7251 iter = &dev->adj_list.lower;
7255 ret = fn(now, priv);
7262 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7269 niter = &ldev->adj_list.lower;
7270 dev_stack[cur] = now;
7271 iter_stack[cur++] = iter;
7278 next = dev_stack[--cur];
7279 niter = iter_stack[cur];
7289 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7290 struct list_head **iter)
7292 struct netdev_adjacent *lower;
7294 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7295 if (&lower->list == &dev->adj_list.lower)
7298 *iter = &lower->list;
7302 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7304 static u8 __netdev_upper_depth(struct net_device *dev)
7306 struct net_device *udev;
7307 struct list_head *iter;
7311 for (iter = &dev->adj_list.upper,
7312 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7314 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7317 if (max_depth < udev->upper_level)
7318 max_depth = udev->upper_level;
7324 static u8 __netdev_lower_depth(struct net_device *dev)
7326 struct net_device *ldev;
7327 struct list_head *iter;
7331 for (iter = &dev->adj_list.lower,
7332 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7334 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7337 if (max_depth < ldev->lower_level)
7338 max_depth = ldev->lower_level;
7344 static int __netdev_update_upper_level(struct net_device *dev,
7345 struct netdev_nested_priv *__unused)
7347 dev->upper_level = __netdev_upper_depth(dev) + 1;
7351 #ifdef CONFIG_LOCKDEP
7352 static LIST_HEAD(net_unlink_list);
7354 static void net_unlink_todo(struct net_device *dev)
7356 if (list_empty(&dev->unlink_list))
7357 list_add_tail(&dev->unlink_list, &net_unlink_list);
7361 static int __netdev_update_lower_level(struct net_device *dev,
7362 struct netdev_nested_priv *priv)
7364 dev->lower_level = __netdev_lower_depth(dev) + 1;
7366 #ifdef CONFIG_LOCKDEP
7370 if (priv->flags & NESTED_SYNC_IMM)
7371 dev->nested_level = dev->lower_level - 1;
7372 if (priv->flags & NESTED_SYNC_TODO)
7373 net_unlink_todo(dev);
7378 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7379 int (*fn)(struct net_device *dev,
7380 struct netdev_nested_priv *priv),
7381 struct netdev_nested_priv *priv)
7383 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7384 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7388 iter = &dev->adj_list.lower;
7392 ret = fn(now, priv);
7399 ldev = netdev_next_lower_dev_rcu(now, &iter);
7404 niter = &ldev->adj_list.lower;
7405 dev_stack[cur] = now;
7406 iter_stack[cur++] = iter;
7413 next = dev_stack[--cur];
7414 niter = iter_stack[cur];
7423 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7426 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7427 * lower neighbour list, RCU
7431 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7432 * list. The caller must hold RCU read lock.
7434 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7436 struct netdev_adjacent *lower;
7438 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7439 struct netdev_adjacent, list);
7441 return lower->private;
7444 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7447 * netdev_master_upper_dev_get_rcu - Get master upper device
7450 * Find a master upper device and return pointer to it or NULL in case
7451 * it's not there. The caller must hold the RCU read lock.
7453 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7455 struct netdev_adjacent *upper;
7457 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7458 struct netdev_adjacent, list);
7459 if (upper && likely(upper->master))
7463 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7465 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7466 struct net_device *adj_dev,
7467 struct list_head *dev_list)
7469 char linkname[IFNAMSIZ+7];
7471 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7472 "upper_%s" : "lower_%s", adj_dev->name);
7473 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7476 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7478 struct list_head *dev_list)
7480 char linkname[IFNAMSIZ+7];
7482 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7483 "upper_%s" : "lower_%s", name);
7484 sysfs_remove_link(&(dev->dev.kobj), linkname);
7487 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7488 struct net_device *adj_dev,
7489 struct list_head *dev_list)
7491 return (dev_list == &dev->adj_list.upper ||
7492 dev_list == &dev->adj_list.lower) &&
7493 net_eq(dev_net(dev), dev_net(adj_dev));
7496 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7497 struct net_device *adj_dev,
7498 struct list_head *dev_list,
7499 void *private, bool master)
7501 struct netdev_adjacent *adj;
7504 adj = __netdev_find_adj(adj_dev, dev_list);
7508 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7509 dev->name, adj_dev->name, adj->ref_nr);
7514 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7519 adj->master = master;
7521 adj->private = private;
7522 adj->ignore = false;
7523 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7525 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7526 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7528 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7529 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7534 /* Ensure that master link is always the first item in list. */
7536 ret = sysfs_create_link(&(dev->dev.kobj),
7537 &(adj_dev->dev.kobj), "master");
7539 goto remove_symlinks;
7541 list_add_rcu(&adj->list, dev_list);
7543 list_add_tail_rcu(&adj->list, dev_list);
7549 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7550 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7552 netdev_put(adj_dev, &adj->dev_tracker);
7558 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7559 struct net_device *adj_dev,
7561 struct list_head *dev_list)
7563 struct netdev_adjacent *adj;
7565 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7566 dev->name, adj_dev->name, ref_nr);
7568 adj = __netdev_find_adj(adj_dev, dev_list);
7571 pr_err("Adjacency does not exist for device %s from %s\n",
7572 dev->name, adj_dev->name);
7577 if (adj->ref_nr > ref_nr) {
7578 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7579 dev->name, adj_dev->name, ref_nr,
7580 adj->ref_nr - ref_nr);
7581 adj->ref_nr -= ref_nr;
7586 sysfs_remove_link(&(dev->dev.kobj), "master");
7588 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7589 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7591 list_del_rcu(&adj->list);
7592 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7593 adj_dev->name, dev->name, adj_dev->name);
7594 netdev_put(adj_dev, &adj->dev_tracker);
7595 kfree_rcu(adj, rcu);
7598 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7599 struct net_device *upper_dev,
7600 struct list_head *up_list,
7601 struct list_head *down_list,
7602 void *private, bool master)
7606 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7611 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7614 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7621 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7622 struct net_device *upper_dev,
7624 struct list_head *up_list,
7625 struct list_head *down_list)
7627 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7628 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7631 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7632 struct net_device *upper_dev,
7633 void *private, bool master)
7635 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7636 &dev->adj_list.upper,
7637 &upper_dev->adj_list.lower,
7641 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7642 struct net_device *upper_dev)
7644 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7645 &dev->adj_list.upper,
7646 &upper_dev->adj_list.lower);
7649 static int __netdev_upper_dev_link(struct net_device *dev,
7650 struct net_device *upper_dev, bool master,
7651 void *upper_priv, void *upper_info,
7652 struct netdev_nested_priv *priv,
7653 struct netlink_ext_ack *extack)
7655 struct netdev_notifier_changeupper_info changeupper_info = {
7660 .upper_dev = upper_dev,
7663 .upper_info = upper_info,
7665 struct net_device *master_dev;
7670 if (dev == upper_dev)
7673 /* To prevent loops, check if dev is not upper device to upper_dev. */
7674 if (__netdev_has_upper_dev(upper_dev, dev))
7677 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7681 if (__netdev_has_upper_dev(dev, upper_dev))
7684 master_dev = __netdev_master_upper_dev_get(dev);
7686 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7689 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7690 &changeupper_info.info);
7691 ret = notifier_to_errno(ret);
7695 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7700 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7701 &changeupper_info.info);
7702 ret = notifier_to_errno(ret);
7706 __netdev_update_upper_level(dev, NULL);
7707 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7709 __netdev_update_lower_level(upper_dev, priv);
7710 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7716 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7722 * netdev_upper_dev_link - Add a link to the upper device
7724 * @upper_dev: new upper device
7725 * @extack: netlink extended ack
7727 * Adds a link to device which is upper to this one. The caller must hold
7728 * the RTNL lock. On a failure a negative errno code is returned.
7729 * On success the reference counts are adjusted and the function
7732 int netdev_upper_dev_link(struct net_device *dev,
7733 struct net_device *upper_dev,
7734 struct netlink_ext_ack *extack)
7736 struct netdev_nested_priv priv = {
7737 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7741 return __netdev_upper_dev_link(dev, upper_dev, false,
7742 NULL, NULL, &priv, extack);
7744 EXPORT_SYMBOL(netdev_upper_dev_link);
7747 * netdev_master_upper_dev_link - Add a master link to the upper device
7749 * @upper_dev: new upper device
7750 * @upper_priv: upper device private
7751 * @upper_info: upper info to be passed down via notifier
7752 * @extack: netlink extended ack
7754 * Adds a link to device which is upper to this one. In this case, only
7755 * one master upper device can be linked, although other non-master devices
7756 * might be linked as well. The caller must hold the RTNL lock.
7757 * On a failure a negative errno code is returned. On success the reference
7758 * counts are adjusted and the function returns zero.
7760 int netdev_master_upper_dev_link(struct net_device *dev,
7761 struct net_device *upper_dev,
7762 void *upper_priv, void *upper_info,
7763 struct netlink_ext_ack *extack)
7765 struct netdev_nested_priv priv = {
7766 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7770 return __netdev_upper_dev_link(dev, upper_dev, true,
7771 upper_priv, upper_info, &priv, extack);
7773 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7775 static void __netdev_upper_dev_unlink(struct net_device *dev,
7776 struct net_device *upper_dev,
7777 struct netdev_nested_priv *priv)
7779 struct netdev_notifier_changeupper_info changeupper_info = {
7783 .upper_dev = upper_dev,
7789 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7791 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7792 &changeupper_info.info);
7794 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7796 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7797 &changeupper_info.info);
7799 __netdev_update_upper_level(dev, NULL);
7800 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7802 __netdev_update_lower_level(upper_dev, priv);
7803 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7808 * netdev_upper_dev_unlink - Removes a link to upper device
7810 * @upper_dev: new upper device
7812 * Removes a link to device which is upper to this one. The caller must hold
7815 void netdev_upper_dev_unlink(struct net_device *dev,
7816 struct net_device *upper_dev)
7818 struct netdev_nested_priv priv = {
7819 .flags = NESTED_SYNC_TODO,
7823 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7825 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7827 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7828 struct net_device *lower_dev,
7831 struct netdev_adjacent *adj;
7833 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7837 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7842 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7843 struct net_device *lower_dev)
7845 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7848 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7849 struct net_device *lower_dev)
7851 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7854 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7855 struct net_device *new_dev,
7856 struct net_device *dev,
7857 struct netlink_ext_ack *extack)
7859 struct netdev_nested_priv priv = {
7868 if (old_dev && new_dev != old_dev)
7869 netdev_adjacent_dev_disable(dev, old_dev);
7870 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7873 if (old_dev && new_dev != old_dev)
7874 netdev_adjacent_dev_enable(dev, old_dev);
7880 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7882 void netdev_adjacent_change_commit(struct net_device *old_dev,
7883 struct net_device *new_dev,
7884 struct net_device *dev)
7886 struct netdev_nested_priv priv = {
7887 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7891 if (!new_dev || !old_dev)
7894 if (new_dev == old_dev)
7897 netdev_adjacent_dev_enable(dev, old_dev);
7898 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7900 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7902 void netdev_adjacent_change_abort(struct net_device *old_dev,
7903 struct net_device *new_dev,
7904 struct net_device *dev)
7906 struct netdev_nested_priv priv = {
7914 if (old_dev && new_dev != old_dev)
7915 netdev_adjacent_dev_enable(dev, old_dev);
7917 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7919 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7922 * netdev_bonding_info_change - Dispatch event about slave change
7924 * @bonding_info: info to dispatch
7926 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7927 * The caller must hold the RTNL lock.
7929 void netdev_bonding_info_change(struct net_device *dev,
7930 struct netdev_bonding_info *bonding_info)
7932 struct netdev_notifier_bonding_info info = {
7936 memcpy(&info.bonding_info, bonding_info,
7937 sizeof(struct netdev_bonding_info));
7938 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7941 EXPORT_SYMBOL(netdev_bonding_info_change);
7943 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7944 struct netlink_ext_ack *extack)
7946 struct netdev_notifier_offload_xstats_info info = {
7948 .info.extack = extack,
7949 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7954 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7956 if (!dev->offload_xstats_l3)
7959 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7960 NETDEV_OFFLOAD_XSTATS_DISABLE,
7962 err = notifier_to_errno(rc);
7969 kfree(dev->offload_xstats_l3);
7970 dev->offload_xstats_l3 = NULL;
7974 int netdev_offload_xstats_enable(struct net_device *dev,
7975 enum netdev_offload_xstats_type type,
7976 struct netlink_ext_ack *extack)
7980 if (netdev_offload_xstats_enabled(dev, type))
7984 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7985 return netdev_offload_xstats_enable_l3(dev, extack);
7991 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7993 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7995 struct netdev_notifier_offload_xstats_info info = {
7997 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8000 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
8002 kfree(dev->offload_xstats_l3);
8003 dev->offload_xstats_l3 = NULL;
8006 int netdev_offload_xstats_disable(struct net_device *dev,
8007 enum netdev_offload_xstats_type type)
8011 if (!netdev_offload_xstats_enabled(dev, type))
8015 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8016 netdev_offload_xstats_disable_l3(dev);
8023 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8025 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8027 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8030 static struct rtnl_hw_stats64 *
8031 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8032 enum netdev_offload_xstats_type type)
8035 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8036 return dev->offload_xstats_l3;
8043 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8044 enum netdev_offload_xstats_type type)
8048 return netdev_offload_xstats_get_ptr(dev, type);
8050 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8052 struct netdev_notifier_offload_xstats_ru {
8056 struct netdev_notifier_offload_xstats_rd {
8057 struct rtnl_hw_stats64 stats;
8061 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8062 const struct rtnl_hw_stats64 *src)
8064 dest->rx_packets += src->rx_packets;
8065 dest->tx_packets += src->tx_packets;
8066 dest->rx_bytes += src->rx_bytes;
8067 dest->tx_bytes += src->tx_bytes;
8068 dest->rx_errors += src->rx_errors;
8069 dest->tx_errors += src->tx_errors;
8070 dest->rx_dropped += src->rx_dropped;
8071 dest->tx_dropped += src->tx_dropped;
8072 dest->multicast += src->multicast;
8075 static int netdev_offload_xstats_get_used(struct net_device *dev,
8076 enum netdev_offload_xstats_type type,
8078 struct netlink_ext_ack *extack)
8080 struct netdev_notifier_offload_xstats_ru report_used = {};
8081 struct netdev_notifier_offload_xstats_info info = {
8083 .info.extack = extack,
8085 .report_used = &report_used,
8089 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8090 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8092 *p_used = report_used.used;
8093 return notifier_to_errno(rc);
8096 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8097 enum netdev_offload_xstats_type type,
8098 struct rtnl_hw_stats64 *p_stats,
8100 struct netlink_ext_ack *extack)
8102 struct netdev_notifier_offload_xstats_rd report_delta = {};
8103 struct netdev_notifier_offload_xstats_info info = {
8105 .info.extack = extack,
8107 .report_delta = &report_delta,
8109 struct rtnl_hw_stats64 *stats;
8112 stats = netdev_offload_xstats_get_ptr(dev, type);
8113 if (WARN_ON(!stats))
8116 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8119 /* Cache whatever we got, even if there was an error, otherwise the
8120 * successful stats retrievals would get lost.
8122 netdev_hw_stats64_add(stats, &report_delta.stats);
8126 *p_used = report_delta.used;
8128 return notifier_to_errno(rc);
8131 int netdev_offload_xstats_get(struct net_device *dev,
8132 enum netdev_offload_xstats_type type,
8133 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8134 struct netlink_ext_ack *extack)
8139 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8142 return netdev_offload_xstats_get_used(dev, type, p_used,
8145 EXPORT_SYMBOL(netdev_offload_xstats_get);
8148 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8149 const struct rtnl_hw_stats64 *stats)
8151 report_delta->used = true;
8152 netdev_hw_stats64_add(&report_delta->stats, stats);
8154 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8157 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8159 report_used->used = true;
8161 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8163 void netdev_offload_xstats_push_delta(struct net_device *dev,
8164 enum netdev_offload_xstats_type type,
8165 const struct rtnl_hw_stats64 *p_stats)
8167 struct rtnl_hw_stats64 *stats;
8171 stats = netdev_offload_xstats_get_ptr(dev, type);
8172 if (WARN_ON(!stats))
8175 netdev_hw_stats64_add(stats, p_stats);
8177 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8180 * netdev_get_xmit_slave - Get the xmit slave of master device
8183 * @all_slaves: assume all the slaves are active
8185 * The reference counters are not incremented so the caller must be
8186 * careful with locks. The caller must hold RCU lock.
8187 * %NULL is returned if no slave is found.
8190 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8191 struct sk_buff *skb,
8194 const struct net_device_ops *ops = dev->netdev_ops;
8196 if (!ops->ndo_get_xmit_slave)
8198 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8200 EXPORT_SYMBOL(netdev_get_xmit_slave);
8202 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8205 const struct net_device_ops *ops = dev->netdev_ops;
8207 if (!ops->ndo_sk_get_lower_dev)
8209 return ops->ndo_sk_get_lower_dev(dev, sk);
8213 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8217 * %NULL is returned if no lower device is found.
8220 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8223 struct net_device *lower;
8225 lower = netdev_sk_get_lower_dev(dev, sk);
8228 lower = netdev_sk_get_lower_dev(dev, sk);
8233 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8235 static void netdev_adjacent_add_links(struct net_device *dev)
8237 struct netdev_adjacent *iter;
8239 struct net *net = dev_net(dev);
8241 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8242 if (!net_eq(net, dev_net(iter->dev)))
8244 netdev_adjacent_sysfs_add(iter->dev, dev,
8245 &iter->dev->adj_list.lower);
8246 netdev_adjacent_sysfs_add(dev, iter->dev,
8247 &dev->adj_list.upper);
8250 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8251 if (!net_eq(net, dev_net(iter->dev)))
8253 netdev_adjacent_sysfs_add(iter->dev, dev,
8254 &iter->dev->adj_list.upper);
8255 netdev_adjacent_sysfs_add(dev, iter->dev,
8256 &dev->adj_list.lower);
8260 static void netdev_adjacent_del_links(struct net_device *dev)
8262 struct netdev_adjacent *iter;
8264 struct net *net = dev_net(dev);
8266 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8267 if (!net_eq(net, dev_net(iter->dev)))
8269 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8270 &iter->dev->adj_list.lower);
8271 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8272 &dev->adj_list.upper);
8275 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8276 if (!net_eq(net, dev_net(iter->dev)))
8278 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8279 &iter->dev->adj_list.upper);
8280 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8281 &dev->adj_list.lower);
8285 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8287 struct netdev_adjacent *iter;
8289 struct net *net = dev_net(dev);
8291 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8292 if (!net_eq(net, dev_net(iter->dev)))
8294 netdev_adjacent_sysfs_del(iter->dev, oldname,
8295 &iter->dev->adj_list.lower);
8296 netdev_adjacent_sysfs_add(iter->dev, dev,
8297 &iter->dev->adj_list.lower);
8300 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8301 if (!net_eq(net, dev_net(iter->dev)))
8303 netdev_adjacent_sysfs_del(iter->dev, oldname,
8304 &iter->dev->adj_list.upper);
8305 netdev_adjacent_sysfs_add(iter->dev, dev,
8306 &iter->dev->adj_list.upper);
8310 void *netdev_lower_dev_get_private(struct net_device *dev,
8311 struct net_device *lower_dev)
8313 struct netdev_adjacent *lower;
8317 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8321 return lower->private;
8323 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8327 * netdev_lower_state_changed - Dispatch event about lower device state change
8328 * @lower_dev: device
8329 * @lower_state_info: state to dispatch
8331 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8332 * The caller must hold the RTNL lock.
8334 void netdev_lower_state_changed(struct net_device *lower_dev,
8335 void *lower_state_info)
8337 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8338 .info.dev = lower_dev,
8342 changelowerstate_info.lower_state_info = lower_state_info;
8343 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8344 &changelowerstate_info.info);
8346 EXPORT_SYMBOL(netdev_lower_state_changed);
8348 static void dev_change_rx_flags(struct net_device *dev, int flags)
8350 const struct net_device_ops *ops = dev->netdev_ops;
8352 if (ops->ndo_change_rx_flags)
8353 ops->ndo_change_rx_flags(dev, flags);
8356 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8358 unsigned int old_flags = dev->flags;
8364 dev->flags |= IFF_PROMISC;
8365 dev->promiscuity += inc;
8366 if (dev->promiscuity == 0) {
8369 * If inc causes overflow, untouch promisc and return error.
8372 dev->flags &= ~IFF_PROMISC;
8374 dev->promiscuity -= inc;
8375 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8379 if (dev->flags != old_flags) {
8380 netdev_info(dev, "%s promiscuous mode\n",
8381 dev->flags & IFF_PROMISC ? "entered" : "left");
8382 if (audit_enabled) {
8383 current_uid_gid(&uid, &gid);
8384 audit_log(audit_context(), GFP_ATOMIC,
8385 AUDIT_ANOM_PROMISCUOUS,
8386 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8387 dev->name, (dev->flags & IFF_PROMISC),
8388 (old_flags & IFF_PROMISC),
8389 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8390 from_kuid(&init_user_ns, uid),
8391 from_kgid(&init_user_ns, gid),
8392 audit_get_sessionid(current));
8395 dev_change_rx_flags(dev, IFF_PROMISC);
8398 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8403 * dev_set_promiscuity - update promiscuity count on a device
8407 * Add or remove promiscuity from a device. While the count in the device
8408 * remains above zero the interface remains promiscuous. Once it hits zero
8409 * the device reverts back to normal filtering operation. A negative inc
8410 * value is used to drop promiscuity on the device.
8411 * Return 0 if successful or a negative errno code on error.
8413 int dev_set_promiscuity(struct net_device *dev, int inc)
8415 unsigned int old_flags = dev->flags;
8418 err = __dev_set_promiscuity(dev, inc, true);
8421 if (dev->flags != old_flags)
8422 dev_set_rx_mode(dev);
8425 EXPORT_SYMBOL(dev_set_promiscuity);
8427 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8429 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8433 dev->flags |= IFF_ALLMULTI;
8434 dev->allmulti += inc;
8435 if (dev->allmulti == 0) {
8438 * If inc causes overflow, untouch allmulti and return error.
8441 dev->flags &= ~IFF_ALLMULTI;
8443 dev->allmulti -= inc;
8444 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8448 if (dev->flags ^ old_flags) {
8449 netdev_info(dev, "%s allmulticast mode\n",
8450 dev->flags & IFF_ALLMULTI ? "entered" : "left");
8451 dev_change_rx_flags(dev, IFF_ALLMULTI);
8452 dev_set_rx_mode(dev);
8454 __dev_notify_flags(dev, old_flags,
8455 dev->gflags ^ old_gflags, 0, NULL);
8461 * dev_set_allmulti - update allmulti count on a device
8465 * Add or remove reception of all multicast frames to a device. While the
8466 * count in the device remains above zero the interface remains listening
8467 * to all interfaces. Once it hits zero the device reverts back to normal
8468 * filtering operation. A negative @inc value is used to drop the counter
8469 * when releasing a resource needing all multicasts.
8470 * Return 0 if successful or a negative errno code on error.
8473 int dev_set_allmulti(struct net_device *dev, int inc)
8475 return __dev_set_allmulti(dev, inc, true);
8477 EXPORT_SYMBOL(dev_set_allmulti);
8480 * Upload unicast and multicast address lists to device and
8481 * configure RX filtering. When the device doesn't support unicast
8482 * filtering it is put in promiscuous mode while unicast addresses
8485 void __dev_set_rx_mode(struct net_device *dev)
8487 const struct net_device_ops *ops = dev->netdev_ops;
8489 /* dev_open will call this function so the list will stay sane. */
8490 if (!(dev->flags&IFF_UP))
8493 if (!netif_device_present(dev))
8496 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8497 /* Unicast addresses changes may only happen under the rtnl,
8498 * therefore calling __dev_set_promiscuity here is safe.
8500 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8501 __dev_set_promiscuity(dev, 1, false);
8502 dev->uc_promisc = true;
8503 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8504 __dev_set_promiscuity(dev, -1, false);
8505 dev->uc_promisc = false;
8509 if (ops->ndo_set_rx_mode)
8510 ops->ndo_set_rx_mode(dev);
8513 void dev_set_rx_mode(struct net_device *dev)
8515 netif_addr_lock_bh(dev);
8516 __dev_set_rx_mode(dev);
8517 netif_addr_unlock_bh(dev);
8521 * dev_get_flags - get flags reported to userspace
8524 * Get the combination of flag bits exported through APIs to userspace.
8526 unsigned int dev_get_flags(const struct net_device *dev)
8530 flags = (dev->flags & ~(IFF_PROMISC |
8535 (dev->gflags & (IFF_PROMISC |
8538 if (netif_running(dev)) {
8539 if (netif_oper_up(dev))
8540 flags |= IFF_RUNNING;
8541 if (netif_carrier_ok(dev))
8542 flags |= IFF_LOWER_UP;
8543 if (netif_dormant(dev))
8544 flags |= IFF_DORMANT;
8549 EXPORT_SYMBOL(dev_get_flags);
8551 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8552 struct netlink_ext_ack *extack)
8554 unsigned int old_flags = dev->flags;
8560 * Set the flags on our device.
8563 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8564 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8566 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8570 * Load in the correct multicast list now the flags have changed.
8573 if ((old_flags ^ flags) & IFF_MULTICAST)
8574 dev_change_rx_flags(dev, IFF_MULTICAST);
8576 dev_set_rx_mode(dev);
8579 * Have we downed the interface. We handle IFF_UP ourselves
8580 * according to user attempts to set it, rather than blindly
8585 if ((old_flags ^ flags) & IFF_UP) {
8586 if (old_flags & IFF_UP)
8589 ret = __dev_open(dev, extack);
8592 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8593 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8594 unsigned int old_flags = dev->flags;
8596 dev->gflags ^= IFF_PROMISC;
8598 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8599 if (dev->flags != old_flags)
8600 dev_set_rx_mode(dev);
8603 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8604 * is important. Some (broken) drivers set IFF_PROMISC, when
8605 * IFF_ALLMULTI is requested not asking us and not reporting.
8607 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8608 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8610 dev->gflags ^= IFF_ALLMULTI;
8611 __dev_set_allmulti(dev, inc, false);
8617 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8618 unsigned int gchanges, u32 portid,
8619 const struct nlmsghdr *nlh)
8621 unsigned int changes = dev->flags ^ old_flags;
8624 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8626 if (changes & IFF_UP) {
8627 if (dev->flags & IFF_UP)
8628 call_netdevice_notifiers(NETDEV_UP, dev);
8630 call_netdevice_notifiers(NETDEV_DOWN, dev);
8633 if (dev->flags & IFF_UP &&
8634 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8635 struct netdev_notifier_change_info change_info = {
8639 .flags_changed = changes,
8642 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8647 * dev_change_flags - change device settings
8649 * @flags: device state flags
8650 * @extack: netlink extended ack
8652 * Change settings on device based state flags. The flags are
8653 * in the userspace exported format.
8655 int dev_change_flags(struct net_device *dev, unsigned int flags,
8656 struct netlink_ext_ack *extack)
8659 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8661 ret = __dev_change_flags(dev, flags, extack);
8665 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8666 __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8669 EXPORT_SYMBOL(dev_change_flags);
8671 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8673 const struct net_device_ops *ops = dev->netdev_ops;
8675 if (ops->ndo_change_mtu)
8676 return ops->ndo_change_mtu(dev, new_mtu);
8678 /* Pairs with all the lockless reads of dev->mtu in the stack */
8679 WRITE_ONCE(dev->mtu, new_mtu);
8682 EXPORT_SYMBOL(__dev_set_mtu);
8684 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8685 struct netlink_ext_ack *extack)
8687 /* MTU must be positive, and in range */
8688 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8689 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8693 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8694 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8701 * dev_set_mtu_ext - Change maximum transfer unit
8703 * @new_mtu: new transfer unit
8704 * @extack: netlink extended ack
8706 * Change the maximum transfer size of the network device.
8708 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8709 struct netlink_ext_ack *extack)
8713 if (new_mtu == dev->mtu)
8716 err = dev_validate_mtu(dev, new_mtu, extack);
8720 if (!netif_device_present(dev))
8723 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8724 err = notifier_to_errno(err);
8728 orig_mtu = dev->mtu;
8729 err = __dev_set_mtu(dev, new_mtu);
8732 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8734 err = notifier_to_errno(err);
8736 /* setting mtu back and notifying everyone again,
8737 * so that they have a chance to revert changes.
8739 __dev_set_mtu(dev, orig_mtu);
8740 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8747 int dev_set_mtu(struct net_device *dev, int new_mtu)
8749 struct netlink_ext_ack extack;
8752 memset(&extack, 0, sizeof(extack));
8753 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8754 if (err && extack._msg)
8755 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8758 EXPORT_SYMBOL(dev_set_mtu);
8761 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8763 * @new_len: new tx queue length
8765 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8767 unsigned int orig_len = dev->tx_queue_len;
8770 if (new_len != (unsigned int)new_len)
8773 if (new_len != orig_len) {
8774 dev->tx_queue_len = new_len;
8775 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8776 res = notifier_to_errno(res);
8779 res = dev_qdisc_change_tx_queue_len(dev);
8787 netdev_err(dev, "refused to change device tx_queue_len\n");
8788 dev->tx_queue_len = orig_len;
8793 * dev_set_group - Change group this device belongs to
8795 * @new_group: group this device should belong to
8797 void dev_set_group(struct net_device *dev, int new_group)
8799 dev->group = new_group;
8803 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8805 * @addr: new address
8806 * @extack: netlink extended ack
8808 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8809 struct netlink_ext_ack *extack)
8811 struct netdev_notifier_pre_changeaddr_info info = {
8813 .info.extack = extack,
8818 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8819 return notifier_to_errno(rc);
8821 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8824 * dev_set_mac_address - Change Media Access Control Address
8827 * @extack: netlink extended ack
8829 * Change the hardware (MAC) address of the device
8831 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8832 struct netlink_ext_ack *extack)
8834 const struct net_device_ops *ops = dev->netdev_ops;
8837 if (!ops->ndo_set_mac_address)
8839 if (sa->sa_family != dev->type)
8841 if (!netif_device_present(dev))
8843 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8846 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) {
8847 err = ops->ndo_set_mac_address(dev, sa);
8851 dev->addr_assign_type = NET_ADDR_SET;
8852 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8853 add_device_randomness(dev->dev_addr, dev->addr_len);
8856 EXPORT_SYMBOL(dev_set_mac_address);
8858 static DECLARE_RWSEM(dev_addr_sem);
8860 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8861 struct netlink_ext_ack *extack)
8865 down_write(&dev_addr_sem);
8866 ret = dev_set_mac_address(dev, sa, extack);
8867 up_write(&dev_addr_sem);
8870 EXPORT_SYMBOL(dev_set_mac_address_user);
8872 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8874 size_t size = sizeof(sa->sa_data_min);
8875 struct net_device *dev;
8878 down_read(&dev_addr_sem);
8881 dev = dev_get_by_name_rcu(net, dev_name);
8887 memset(sa->sa_data, 0, size);
8889 memcpy(sa->sa_data, dev->dev_addr,
8890 min_t(size_t, size, dev->addr_len));
8891 sa->sa_family = dev->type;
8895 up_read(&dev_addr_sem);
8898 EXPORT_SYMBOL(dev_get_mac_address);
8901 * dev_change_carrier - Change device carrier
8903 * @new_carrier: new value
8905 * Change device carrier
8907 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8909 const struct net_device_ops *ops = dev->netdev_ops;
8911 if (!ops->ndo_change_carrier)
8913 if (!netif_device_present(dev))
8915 return ops->ndo_change_carrier(dev, new_carrier);
8919 * dev_get_phys_port_id - Get device physical port ID
8923 * Get device physical port ID
8925 int dev_get_phys_port_id(struct net_device *dev,
8926 struct netdev_phys_item_id *ppid)
8928 const struct net_device_ops *ops = dev->netdev_ops;
8930 if (!ops->ndo_get_phys_port_id)
8932 return ops->ndo_get_phys_port_id(dev, ppid);
8936 * dev_get_phys_port_name - Get device physical port name
8939 * @len: limit of bytes to copy to name
8941 * Get device physical port name
8943 int dev_get_phys_port_name(struct net_device *dev,
8944 char *name, size_t len)
8946 const struct net_device_ops *ops = dev->netdev_ops;
8949 if (ops->ndo_get_phys_port_name) {
8950 err = ops->ndo_get_phys_port_name(dev, name, len);
8951 if (err != -EOPNOTSUPP)
8954 return devlink_compat_phys_port_name_get(dev, name, len);
8958 * dev_get_port_parent_id - Get the device's port parent identifier
8959 * @dev: network device
8960 * @ppid: pointer to a storage for the port's parent identifier
8961 * @recurse: allow/disallow recursion to lower devices
8963 * Get the devices's port parent identifier
8965 int dev_get_port_parent_id(struct net_device *dev,
8966 struct netdev_phys_item_id *ppid,
8969 const struct net_device_ops *ops = dev->netdev_ops;
8970 struct netdev_phys_item_id first = { };
8971 struct net_device *lower_dev;
8972 struct list_head *iter;
8975 if (ops->ndo_get_port_parent_id) {
8976 err = ops->ndo_get_port_parent_id(dev, ppid);
8977 if (err != -EOPNOTSUPP)
8981 err = devlink_compat_switch_id_get(dev, ppid);
8982 if (!recurse || err != -EOPNOTSUPP)
8985 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8986 err = dev_get_port_parent_id(lower_dev, ppid, true);
8991 else if (memcmp(&first, ppid, sizeof(*ppid)))
8997 EXPORT_SYMBOL(dev_get_port_parent_id);
9000 * netdev_port_same_parent_id - Indicate if two network devices have
9001 * the same port parent identifier
9002 * @a: first network device
9003 * @b: second network device
9005 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9007 struct netdev_phys_item_id a_id = { };
9008 struct netdev_phys_item_id b_id = { };
9010 if (dev_get_port_parent_id(a, &a_id, true) ||
9011 dev_get_port_parent_id(b, &b_id, true))
9014 return netdev_phys_item_id_same(&a_id, &b_id);
9016 EXPORT_SYMBOL(netdev_port_same_parent_id);
9019 * dev_change_proto_down - set carrier according to proto_down.
9022 * @proto_down: new value
9024 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9026 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
9028 if (!netif_device_present(dev))
9031 netif_carrier_off(dev);
9033 netif_carrier_on(dev);
9034 dev->proto_down = proto_down;
9039 * dev_change_proto_down_reason - proto down reason
9042 * @mask: proto down mask
9043 * @value: proto down value
9045 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9051 dev->proto_down_reason = value;
9053 for_each_set_bit(b, &mask, 32) {
9054 if (value & (1 << b))
9055 dev->proto_down_reason |= BIT(b);
9057 dev->proto_down_reason &= ~BIT(b);
9062 struct bpf_xdp_link {
9063 struct bpf_link link;
9064 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9068 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9070 if (flags & XDP_FLAGS_HW_MODE)
9072 if (flags & XDP_FLAGS_DRV_MODE)
9073 return XDP_MODE_DRV;
9074 if (flags & XDP_FLAGS_SKB_MODE)
9075 return XDP_MODE_SKB;
9076 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9079 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9083 return generic_xdp_install;
9086 return dev->netdev_ops->ndo_bpf;
9092 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9093 enum bpf_xdp_mode mode)
9095 return dev->xdp_state[mode].link;
9098 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9099 enum bpf_xdp_mode mode)
9101 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9104 return link->link.prog;
9105 return dev->xdp_state[mode].prog;
9108 u8 dev_xdp_prog_count(struct net_device *dev)
9113 for (i = 0; i < __MAX_XDP_MODE; i++)
9114 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9118 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9120 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9122 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9124 return prog ? prog->aux->id : 0;
9127 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9128 struct bpf_xdp_link *link)
9130 dev->xdp_state[mode].link = link;
9131 dev->xdp_state[mode].prog = NULL;
9134 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9135 struct bpf_prog *prog)
9137 dev->xdp_state[mode].link = NULL;
9138 dev->xdp_state[mode].prog = prog;
9141 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9142 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9143 u32 flags, struct bpf_prog *prog)
9145 struct netdev_bpf xdp;
9148 memset(&xdp, 0, sizeof(xdp));
9149 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9150 xdp.extack = extack;
9154 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9155 * "moved" into driver), so they don't increment it on their own, but
9156 * they do decrement refcnt when program is detached or replaced.
9157 * Given net_device also owns link/prog, we need to bump refcnt here
9158 * to prevent drivers from underflowing it.
9162 err = bpf_op(dev, &xdp);
9169 if (mode != XDP_MODE_HW)
9170 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9175 static void dev_xdp_uninstall(struct net_device *dev)
9177 struct bpf_xdp_link *link;
9178 struct bpf_prog *prog;
9179 enum bpf_xdp_mode mode;
9184 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9185 prog = dev_xdp_prog(dev, mode);
9189 bpf_op = dev_xdp_bpf_op(dev, mode);
9193 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9195 /* auto-detach link from net device */
9196 link = dev_xdp_link(dev, mode);
9202 dev_xdp_set_link(dev, mode, NULL);
9206 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9207 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9208 struct bpf_prog *old_prog, u32 flags)
9210 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9211 struct bpf_prog *cur_prog;
9212 struct net_device *upper;
9213 struct list_head *iter;
9214 enum bpf_xdp_mode mode;
9220 /* either link or prog attachment, never both */
9221 if (link && (new_prog || old_prog))
9223 /* link supports only XDP mode flags */
9224 if (link && (flags & ~XDP_FLAGS_MODES)) {
9225 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9228 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9229 if (num_modes > 1) {
9230 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9233 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9234 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9235 NL_SET_ERR_MSG(extack,
9236 "More than one program loaded, unset mode is ambiguous");
9239 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9240 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9241 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9245 mode = dev_xdp_mode(dev, flags);
9246 /* can't replace attached link */
9247 if (dev_xdp_link(dev, mode)) {
9248 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9252 /* don't allow if an upper device already has a program */
9253 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9254 if (dev_xdp_prog_count(upper) > 0) {
9255 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9260 cur_prog = dev_xdp_prog(dev, mode);
9261 /* can't replace attached prog with link */
9262 if (link && cur_prog) {
9263 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9266 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9267 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9271 /* put effective new program into new_prog */
9273 new_prog = link->link.prog;
9276 bool offload = mode == XDP_MODE_HW;
9277 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9278 ? XDP_MODE_DRV : XDP_MODE_SKB;
9280 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9281 NL_SET_ERR_MSG(extack, "XDP program already attached");
9284 if (!offload && dev_xdp_prog(dev, other_mode)) {
9285 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9288 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9289 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9292 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9293 NL_SET_ERR_MSG(extack, "Program bound to different device");
9296 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9297 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9300 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9301 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9306 /* don't call drivers if the effective program didn't change */
9307 if (new_prog != cur_prog) {
9308 bpf_op = dev_xdp_bpf_op(dev, mode);
9310 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9314 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9320 dev_xdp_set_link(dev, mode, link);
9322 dev_xdp_set_prog(dev, mode, new_prog);
9324 bpf_prog_put(cur_prog);
9329 static int dev_xdp_attach_link(struct net_device *dev,
9330 struct netlink_ext_ack *extack,
9331 struct bpf_xdp_link *link)
9333 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9336 static int dev_xdp_detach_link(struct net_device *dev,
9337 struct netlink_ext_ack *extack,
9338 struct bpf_xdp_link *link)
9340 enum bpf_xdp_mode mode;
9345 mode = dev_xdp_mode(dev, link->flags);
9346 if (dev_xdp_link(dev, mode) != link)
9349 bpf_op = dev_xdp_bpf_op(dev, mode);
9350 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9351 dev_xdp_set_link(dev, mode, NULL);
9355 static void bpf_xdp_link_release(struct bpf_link *link)
9357 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9361 /* if racing with net_device's tear down, xdp_link->dev might be
9362 * already NULL, in which case link was already auto-detached
9364 if (xdp_link->dev) {
9365 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9366 xdp_link->dev = NULL;
9372 static int bpf_xdp_link_detach(struct bpf_link *link)
9374 bpf_xdp_link_release(link);
9378 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9380 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9385 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9386 struct seq_file *seq)
9388 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9393 ifindex = xdp_link->dev->ifindex;
9396 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9399 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9400 struct bpf_link_info *info)
9402 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9407 ifindex = xdp_link->dev->ifindex;
9410 info->xdp.ifindex = ifindex;
9414 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9415 struct bpf_prog *old_prog)
9417 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9418 enum bpf_xdp_mode mode;
9424 /* link might have been auto-released already, so fail */
9425 if (!xdp_link->dev) {
9430 if (old_prog && link->prog != old_prog) {
9434 old_prog = link->prog;
9435 if (old_prog->type != new_prog->type ||
9436 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9441 if (old_prog == new_prog) {
9442 /* no-op, don't disturb drivers */
9443 bpf_prog_put(new_prog);
9447 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9448 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9449 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9450 xdp_link->flags, new_prog);
9454 old_prog = xchg(&link->prog, new_prog);
9455 bpf_prog_put(old_prog);
9462 static const struct bpf_link_ops bpf_xdp_link_lops = {
9463 .release = bpf_xdp_link_release,
9464 .dealloc = bpf_xdp_link_dealloc,
9465 .detach = bpf_xdp_link_detach,
9466 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9467 .fill_link_info = bpf_xdp_link_fill_link_info,
9468 .update_prog = bpf_xdp_link_update,
9471 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9473 struct net *net = current->nsproxy->net_ns;
9474 struct bpf_link_primer link_primer;
9475 struct netlink_ext_ack extack = {};
9476 struct bpf_xdp_link *link;
9477 struct net_device *dev;
9481 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9487 link = kzalloc(sizeof(*link), GFP_USER);
9493 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9495 link->flags = attr->link_create.flags;
9497 err = bpf_link_prime(&link->link, &link_primer);
9503 err = dev_xdp_attach_link(dev, &extack, link);
9508 bpf_link_cleanup(&link_primer);
9509 trace_bpf_xdp_link_attach_failed(extack._msg);
9513 fd = bpf_link_settle(&link_primer);
9514 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9527 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9529 * @extack: netlink extended ack
9530 * @fd: new program fd or negative value to clear
9531 * @expected_fd: old program fd that userspace expects to replace or clear
9532 * @flags: xdp-related flags
9534 * Set or clear a bpf program for a device
9536 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9537 int fd, int expected_fd, u32 flags)
9539 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9540 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9546 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9547 mode != XDP_MODE_SKB);
9548 if (IS_ERR(new_prog))
9549 return PTR_ERR(new_prog);
9552 if (expected_fd >= 0) {
9553 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9554 mode != XDP_MODE_SKB);
9555 if (IS_ERR(old_prog)) {
9556 err = PTR_ERR(old_prog);
9562 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9565 if (err && new_prog)
9566 bpf_prog_put(new_prog);
9568 bpf_prog_put(old_prog);
9573 * dev_new_index - allocate an ifindex
9574 * @net: the applicable net namespace
9576 * Returns a suitable unique value for a new device interface
9577 * number. The caller must hold the rtnl semaphore or the
9578 * dev_base_lock to be sure it remains unique.
9580 static int dev_new_index(struct net *net)
9582 int ifindex = net->ifindex;
9587 if (!__dev_get_by_index(net, ifindex))
9588 return net->ifindex = ifindex;
9592 /* Delayed registration/unregisteration */
9593 LIST_HEAD(net_todo_list);
9594 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9596 static void net_set_todo(struct net_device *dev)
9598 list_add_tail(&dev->todo_list, &net_todo_list);
9599 atomic_inc(&dev_net(dev)->dev_unreg_count);
9602 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9603 struct net_device *upper, netdev_features_t features)
9605 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9606 netdev_features_t feature;
9609 for_each_netdev_feature(upper_disables, feature_bit) {
9610 feature = __NETIF_F_BIT(feature_bit);
9611 if (!(upper->wanted_features & feature)
9612 && (features & feature)) {
9613 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9614 &feature, upper->name);
9615 features &= ~feature;
9622 static void netdev_sync_lower_features(struct net_device *upper,
9623 struct net_device *lower, netdev_features_t features)
9625 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9626 netdev_features_t feature;
9629 for_each_netdev_feature(upper_disables, feature_bit) {
9630 feature = __NETIF_F_BIT(feature_bit);
9631 if (!(features & feature) && (lower->features & feature)) {
9632 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9633 &feature, lower->name);
9634 lower->wanted_features &= ~feature;
9635 __netdev_update_features(lower);
9637 if (unlikely(lower->features & feature))
9638 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9639 &feature, lower->name);
9641 netdev_features_change(lower);
9646 static netdev_features_t netdev_fix_features(struct net_device *dev,
9647 netdev_features_t features)
9649 /* Fix illegal checksum combinations */
9650 if ((features & NETIF_F_HW_CSUM) &&
9651 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9652 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9653 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9656 /* TSO requires that SG is present as well. */
9657 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9658 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9659 features &= ~NETIF_F_ALL_TSO;
9662 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9663 !(features & NETIF_F_IP_CSUM)) {
9664 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9665 features &= ~NETIF_F_TSO;
9666 features &= ~NETIF_F_TSO_ECN;
9669 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9670 !(features & NETIF_F_IPV6_CSUM)) {
9671 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9672 features &= ~NETIF_F_TSO6;
9675 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9676 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9677 features &= ~NETIF_F_TSO_MANGLEID;
9679 /* TSO ECN requires that TSO is present as well. */
9680 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9681 features &= ~NETIF_F_TSO_ECN;
9683 /* Software GSO depends on SG. */
9684 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9685 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9686 features &= ~NETIF_F_GSO;
9689 /* GSO partial features require GSO partial be set */
9690 if ((features & dev->gso_partial_features) &&
9691 !(features & NETIF_F_GSO_PARTIAL)) {
9693 "Dropping partially supported GSO features since no GSO partial.\n");
9694 features &= ~dev->gso_partial_features;
9697 if (!(features & NETIF_F_RXCSUM)) {
9698 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9699 * successfully merged by hardware must also have the
9700 * checksum verified by hardware. If the user does not
9701 * want to enable RXCSUM, logically, we should disable GRO_HW.
9703 if (features & NETIF_F_GRO_HW) {
9704 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9705 features &= ~NETIF_F_GRO_HW;
9709 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9710 if (features & NETIF_F_RXFCS) {
9711 if (features & NETIF_F_LRO) {
9712 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9713 features &= ~NETIF_F_LRO;
9716 if (features & NETIF_F_GRO_HW) {
9717 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9718 features &= ~NETIF_F_GRO_HW;
9722 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9723 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9724 features &= ~NETIF_F_LRO;
9727 if (features & NETIF_F_HW_TLS_TX) {
9728 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9729 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9730 bool hw_csum = features & NETIF_F_HW_CSUM;
9732 if (!ip_csum && !hw_csum) {
9733 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9734 features &= ~NETIF_F_HW_TLS_TX;
9738 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9739 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9740 features &= ~NETIF_F_HW_TLS_RX;
9746 int __netdev_update_features(struct net_device *dev)
9748 struct net_device *upper, *lower;
9749 netdev_features_t features;
9750 struct list_head *iter;
9755 features = netdev_get_wanted_features(dev);
9757 if (dev->netdev_ops->ndo_fix_features)
9758 features = dev->netdev_ops->ndo_fix_features(dev, features);
9760 /* driver might be less strict about feature dependencies */
9761 features = netdev_fix_features(dev, features);
9763 /* some features can't be enabled if they're off on an upper device */
9764 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9765 features = netdev_sync_upper_features(dev, upper, features);
9767 if (dev->features == features)
9770 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9771 &dev->features, &features);
9773 if (dev->netdev_ops->ndo_set_features)
9774 err = dev->netdev_ops->ndo_set_features(dev, features);
9778 if (unlikely(err < 0)) {
9780 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9781 err, &features, &dev->features);
9782 /* return non-0 since some features might have changed and
9783 * it's better to fire a spurious notification than miss it
9789 /* some features must be disabled on lower devices when disabled
9790 * on an upper device (think: bonding master or bridge)
9792 netdev_for_each_lower_dev(dev, lower, iter)
9793 netdev_sync_lower_features(dev, lower, features);
9796 netdev_features_t diff = features ^ dev->features;
9798 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9799 /* udp_tunnel_{get,drop}_rx_info both need
9800 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9801 * device, or they won't do anything.
9802 * Thus we need to update dev->features
9803 * *before* calling udp_tunnel_get_rx_info,
9804 * but *after* calling udp_tunnel_drop_rx_info.
9806 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9807 dev->features = features;
9808 udp_tunnel_get_rx_info(dev);
9810 udp_tunnel_drop_rx_info(dev);
9814 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9815 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9816 dev->features = features;
9817 err |= vlan_get_rx_ctag_filter_info(dev);
9819 vlan_drop_rx_ctag_filter_info(dev);
9823 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9824 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9825 dev->features = features;
9826 err |= vlan_get_rx_stag_filter_info(dev);
9828 vlan_drop_rx_stag_filter_info(dev);
9832 dev->features = features;
9835 return err < 0 ? 0 : 1;
9839 * netdev_update_features - recalculate device features
9840 * @dev: the device to check
9842 * Recalculate dev->features set and send notifications if it
9843 * has changed. Should be called after driver or hardware dependent
9844 * conditions might have changed that influence the features.
9846 void netdev_update_features(struct net_device *dev)
9848 if (__netdev_update_features(dev))
9849 netdev_features_change(dev);
9851 EXPORT_SYMBOL(netdev_update_features);
9854 * netdev_change_features - recalculate device features
9855 * @dev: the device to check
9857 * Recalculate dev->features set and send notifications even
9858 * if they have not changed. Should be called instead of
9859 * netdev_update_features() if also dev->vlan_features might
9860 * have changed to allow the changes to be propagated to stacked
9863 void netdev_change_features(struct net_device *dev)
9865 __netdev_update_features(dev);
9866 netdev_features_change(dev);
9868 EXPORT_SYMBOL(netdev_change_features);
9871 * netif_stacked_transfer_operstate - transfer operstate
9872 * @rootdev: the root or lower level device to transfer state from
9873 * @dev: the device to transfer operstate to
9875 * Transfer operational state from root to device. This is normally
9876 * called when a stacking relationship exists between the root
9877 * device and the device(a leaf device).
9879 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9880 struct net_device *dev)
9882 if (rootdev->operstate == IF_OPER_DORMANT)
9883 netif_dormant_on(dev);
9885 netif_dormant_off(dev);
9887 if (rootdev->operstate == IF_OPER_TESTING)
9888 netif_testing_on(dev);
9890 netif_testing_off(dev);
9892 if (netif_carrier_ok(rootdev))
9893 netif_carrier_on(dev);
9895 netif_carrier_off(dev);
9897 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9899 static int netif_alloc_rx_queues(struct net_device *dev)
9901 unsigned int i, count = dev->num_rx_queues;
9902 struct netdev_rx_queue *rx;
9903 size_t sz = count * sizeof(*rx);
9908 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9914 for (i = 0; i < count; i++) {
9917 /* XDP RX-queue setup */
9918 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9925 /* Rollback successful reg's and free other resources */
9927 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9933 static void netif_free_rx_queues(struct net_device *dev)
9935 unsigned int i, count = dev->num_rx_queues;
9937 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9941 for (i = 0; i < count; i++)
9942 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9947 static void netdev_init_one_queue(struct net_device *dev,
9948 struct netdev_queue *queue, void *_unused)
9950 /* Initialize queue lock */
9951 spin_lock_init(&queue->_xmit_lock);
9952 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9953 queue->xmit_lock_owner = -1;
9954 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9957 dql_init(&queue->dql, HZ);
9961 static void netif_free_tx_queues(struct net_device *dev)
9966 static int netif_alloc_netdev_queues(struct net_device *dev)
9968 unsigned int count = dev->num_tx_queues;
9969 struct netdev_queue *tx;
9970 size_t sz = count * sizeof(*tx);
9972 if (count < 1 || count > 0xffff)
9975 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9981 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9982 spin_lock_init(&dev->tx_global_lock);
9987 void netif_tx_stop_all_queues(struct net_device *dev)
9991 for (i = 0; i < dev->num_tx_queues; i++) {
9992 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9994 netif_tx_stop_queue(txq);
9997 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10000 * register_netdevice() - register a network device
10001 * @dev: device to register
10003 * Take a prepared network device structure and make it externally accessible.
10004 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
10005 * Callers must hold the rtnl lock - you may want register_netdev()
10008 int register_netdevice(struct net_device *dev)
10011 struct net *net = dev_net(dev);
10013 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10014 NETDEV_FEATURE_COUNT);
10015 BUG_ON(dev_boot_phase);
10020 /* When net_device's are persistent, this will be fatal. */
10021 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10024 ret = ethtool_check_ops(dev->ethtool_ops);
10028 spin_lock_init(&dev->addr_list_lock);
10029 netdev_set_addr_lockdep_class(dev);
10031 ret = dev_get_valid_name(net, dev, dev->name);
10036 dev->name_node = netdev_name_node_head_alloc(dev);
10037 if (!dev->name_node)
10040 /* Init, if this function is available */
10041 if (dev->netdev_ops->ndo_init) {
10042 ret = dev->netdev_ops->ndo_init(dev);
10046 goto err_free_name;
10050 if (((dev->hw_features | dev->features) &
10051 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10052 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10053 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10054 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10061 dev->ifindex = dev_new_index(net);
10062 else if (__dev_get_by_index(net, dev->ifindex))
10065 /* Transfer changeable features to wanted_features and enable
10066 * software offloads (GSO and GRO).
10068 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10069 dev->features |= NETIF_F_SOFT_FEATURES;
10071 if (dev->udp_tunnel_nic_info) {
10072 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10073 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10076 dev->wanted_features = dev->features & dev->hw_features;
10078 if (!(dev->flags & IFF_LOOPBACK))
10079 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10081 /* If IPv4 TCP segmentation offload is supported we should also
10082 * allow the device to enable segmenting the frame with the option
10083 * of ignoring a static IP ID value. This doesn't enable the
10084 * feature itself but allows the user to enable it later.
10086 if (dev->hw_features & NETIF_F_TSO)
10087 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10088 if (dev->vlan_features & NETIF_F_TSO)
10089 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10090 if (dev->mpls_features & NETIF_F_TSO)
10091 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10092 if (dev->hw_enc_features & NETIF_F_TSO)
10093 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10095 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10097 dev->vlan_features |= NETIF_F_HIGHDMA;
10099 /* Make NETIF_F_SG inheritable to tunnel devices.
10101 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10103 /* Make NETIF_F_SG inheritable to MPLS.
10105 dev->mpls_features |= NETIF_F_SG;
10107 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10108 ret = notifier_to_errno(ret);
10112 ret = netdev_register_kobject(dev);
10113 write_lock(&dev_base_lock);
10114 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10115 write_unlock(&dev_base_lock);
10117 goto err_uninit_notify;
10119 __netdev_update_features(dev);
10122 * Default initial state at registry is that the
10123 * device is present.
10126 set_bit(__LINK_STATE_PRESENT, &dev->state);
10128 linkwatch_init_dev(dev);
10130 dev_init_scheduler(dev);
10132 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10133 list_netdevice(dev);
10135 add_device_randomness(dev->dev_addr, dev->addr_len);
10137 /* If the device has permanent device address, driver should
10138 * set dev_addr and also addr_assign_type should be set to
10139 * NET_ADDR_PERM (default value).
10141 if (dev->addr_assign_type == NET_ADDR_PERM)
10142 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10144 /* Notify protocols, that a new device appeared. */
10145 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10146 ret = notifier_to_errno(ret);
10148 /* Expect explicit free_netdev() on failure */
10149 dev->needs_free_netdev = false;
10150 unregister_netdevice_queue(dev, NULL);
10154 * Prevent userspace races by waiting until the network
10155 * device is fully setup before sending notifications.
10157 if (!dev->rtnl_link_ops ||
10158 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10159 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10165 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10167 if (dev->netdev_ops->ndo_uninit)
10168 dev->netdev_ops->ndo_uninit(dev);
10169 if (dev->priv_destructor)
10170 dev->priv_destructor(dev);
10172 netdev_name_node_free(dev->name_node);
10175 EXPORT_SYMBOL(register_netdevice);
10178 * init_dummy_netdev - init a dummy network device for NAPI
10179 * @dev: device to init
10181 * This takes a network device structure and initialize the minimum
10182 * amount of fields so it can be used to schedule NAPI polls without
10183 * registering a full blown interface. This is to be used by drivers
10184 * that need to tie several hardware interfaces to a single NAPI
10185 * poll scheduler due to HW limitations.
10187 int init_dummy_netdev(struct net_device *dev)
10189 /* Clear everything. Note we don't initialize spinlocks
10190 * are they aren't supposed to be taken by any of the
10191 * NAPI code and this dummy netdev is supposed to be
10192 * only ever used for NAPI polls
10194 memset(dev, 0, sizeof(struct net_device));
10196 /* make sure we BUG if trying to hit standard
10197 * register/unregister code path
10199 dev->reg_state = NETREG_DUMMY;
10201 /* NAPI wants this */
10202 INIT_LIST_HEAD(&dev->napi_list);
10204 /* a dummy interface is started by default */
10205 set_bit(__LINK_STATE_PRESENT, &dev->state);
10206 set_bit(__LINK_STATE_START, &dev->state);
10208 /* napi_busy_loop stats accounting wants this */
10209 dev_net_set(dev, &init_net);
10211 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10212 * because users of this 'device' dont need to change
10218 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10222 * register_netdev - register a network device
10223 * @dev: device to register
10225 * Take a completed network device structure and add it to the kernel
10226 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10227 * chain. 0 is returned on success. A negative errno code is returned
10228 * on a failure to set up the device, or if the name is a duplicate.
10230 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10231 * and expands the device name if you passed a format string to
10234 int register_netdev(struct net_device *dev)
10238 if (rtnl_lock_killable())
10240 err = register_netdevice(dev);
10244 EXPORT_SYMBOL(register_netdev);
10246 int netdev_refcnt_read(const struct net_device *dev)
10248 #ifdef CONFIG_PCPU_DEV_REFCNT
10251 for_each_possible_cpu(i)
10252 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10255 return refcount_read(&dev->dev_refcnt);
10258 EXPORT_SYMBOL(netdev_refcnt_read);
10260 int netdev_unregister_timeout_secs __read_mostly = 10;
10262 #define WAIT_REFS_MIN_MSECS 1
10263 #define WAIT_REFS_MAX_MSECS 250
10265 * netdev_wait_allrefs_any - wait until all references are gone.
10266 * @list: list of net_devices to wait on
10268 * This is called when unregistering network devices.
10270 * Any protocol or device that holds a reference should register
10271 * for netdevice notification, and cleanup and put back the
10272 * reference if they receive an UNREGISTER event.
10273 * We can get stuck here if buggy protocols don't correctly
10276 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10278 unsigned long rebroadcast_time, warning_time;
10279 struct net_device *dev;
10282 rebroadcast_time = warning_time = jiffies;
10284 list_for_each_entry(dev, list, todo_list)
10285 if (netdev_refcnt_read(dev) == 1)
10289 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10292 /* Rebroadcast unregister notification */
10293 list_for_each_entry(dev, list, todo_list)
10294 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10300 list_for_each_entry(dev, list, todo_list)
10301 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10303 /* We must not have linkwatch events
10304 * pending on unregister. If this
10305 * happens, we simply run the queue
10306 * unscheduled, resulting in a noop
10309 linkwatch_run_queue();
10315 rebroadcast_time = jiffies;
10320 wait = WAIT_REFS_MIN_MSECS;
10323 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10326 list_for_each_entry(dev, list, todo_list)
10327 if (netdev_refcnt_read(dev) == 1)
10330 if (time_after(jiffies, warning_time +
10331 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10332 list_for_each_entry(dev, list, todo_list) {
10333 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10334 dev->name, netdev_refcnt_read(dev));
10335 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10338 warning_time = jiffies;
10343 /* The sequence is:
10347 * register_netdevice(x1);
10348 * register_netdevice(x2);
10350 * unregister_netdevice(y1);
10351 * unregister_netdevice(y2);
10357 * We are invoked by rtnl_unlock().
10358 * This allows us to deal with problems:
10359 * 1) We can delete sysfs objects which invoke hotplug
10360 * without deadlocking with linkwatch via keventd.
10361 * 2) Since we run with the RTNL semaphore not held, we can sleep
10362 * safely in order to wait for the netdev refcnt to drop to zero.
10364 * We must not return until all unregister events added during
10365 * the interval the lock was held have been completed.
10367 void netdev_run_todo(void)
10369 struct net_device *dev, *tmp;
10370 struct list_head list;
10371 #ifdef CONFIG_LOCKDEP
10372 struct list_head unlink_list;
10374 list_replace_init(&net_unlink_list, &unlink_list);
10376 while (!list_empty(&unlink_list)) {
10377 struct net_device *dev = list_first_entry(&unlink_list,
10380 list_del_init(&dev->unlink_list);
10381 dev->nested_level = dev->lower_level - 1;
10385 /* Snapshot list, allow later requests */
10386 list_replace_init(&net_todo_list, &list);
10390 /* Wait for rcu callbacks to finish before next phase */
10391 if (!list_empty(&list))
10394 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10395 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10396 netdev_WARN(dev, "run_todo but not unregistering\n");
10397 list_del(&dev->todo_list);
10401 write_lock(&dev_base_lock);
10402 dev->reg_state = NETREG_UNREGISTERED;
10403 write_unlock(&dev_base_lock);
10404 linkwatch_forget_dev(dev);
10407 while (!list_empty(&list)) {
10408 dev = netdev_wait_allrefs_any(&list);
10409 list_del(&dev->todo_list);
10412 BUG_ON(netdev_refcnt_read(dev) != 1);
10413 BUG_ON(!list_empty(&dev->ptype_all));
10414 BUG_ON(!list_empty(&dev->ptype_specific));
10415 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10416 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10418 if (dev->priv_destructor)
10419 dev->priv_destructor(dev);
10420 if (dev->needs_free_netdev)
10423 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10424 wake_up(&netdev_unregistering_wq);
10426 /* Free network device */
10427 kobject_put(&dev->dev.kobj);
10431 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10432 * all the same fields in the same order as net_device_stats, with only
10433 * the type differing, but rtnl_link_stats64 may have additional fields
10434 * at the end for newer counters.
10436 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10437 const struct net_device_stats *netdev_stats)
10439 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10440 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10441 u64 *dst = (u64 *)stats64;
10443 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10444 for (i = 0; i < n; i++)
10445 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10446 /* zero out counters that only exist in rtnl_link_stats64 */
10447 memset((char *)stats64 + n * sizeof(u64), 0,
10448 sizeof(*stats64) - n * sizeof(u64));
10450 EXPORT_SYMBOL(netdev_stats_to_stats64);
10452 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10454 struct net_device_core_stats __percpu *p;
10456 p = alloc_percpu_gfp(struct net_device_core_stats,
10457 GFP_ATOMIC | __GFP_NOWARN);
10459 if (p && cmpxchg(&dev->core_stats, NULL, p))
10462 /* This READ_ONCE() pairs with the cmpxchg() above */
10463 return READ_ONCE(dev->core_stats);
10465 EXPORT_SYMBOL(netdev_core_stats_alloc);
10468 * dev_get_stats - get network device statistics
10469 * @dev: device to get statistics from
10470 * @storage: place to store stats
10472 * Get network statistics from device. Return @storage.
10473 * The device driver may provide its own method by setting
10474 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10475 * otherwise the internal statistics structure is used.
10477 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10478 struct rtnl_link_stats64 *storage)
10480 const struct net_device_ops *ops = dev->netdev_ops;
10481 const struct net_device_core_stats __percpu *p;
10483 if (ops->ndo_get_stats64) {
10484 memset(storage, 0, sizeof(*storage));
10485 ops->ndo_get_stats64(dev, storage);
10486 } else if (ops->ndo_get_stats) {
10487 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10489 netdev_stats_to_stats64(storage, &dev->stats);
10492 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10493 p = READ_ONCE(dev->core_stats);
10495 const struct net_device_core_stats *core_stats;
10498 for_each_possible_cpu(i) {
10499 core_stats = per_cpu_ptr(p, i);
10500 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10501 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10502 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10503 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10508 EXPORT_SYMBOL(dev_get_stats);
10511 * dev_fetch_sw_netstats - get per-cpu network device statistics
10512 * @s: place to store stats
10513 * @netstats: per-cpu network stats to read from
10515 * Read per-cpu network statistics and populate the related fields in @s.
10517 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10518 const struct pcpu_sw_netstats __percpu *netstats)
10522 for_each_possible_cpu(cpu) {
10523 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10524 const struct pcpu_sw_netstats *stats;
10525 unsigned int start;
10527 stats = per_cpu_ptr(netstats, cpu);
10529 start = u64_stats_fetch_begin(&stats->syncp);
10530 rx_packets = u64_stats_read(&stats->rx_packets);
10531 rx_bytes = u64_stats_read(&stats->rx_bytes);
10532 tx_packets = u64_stats_read(&stats->tx_packets);
10533 tx_bytes = u64_stats_read(&stats->tx_bytes);
10534 } while (u64_stats_fetch_retry(&stats->syncp, start));
10536 s->rx_packets += rx_packets;
10537 s->rx_bytes += rx_bytes;
10538 s->tx_packets += tx_packets;
10539 s->tx_bytes += tx_bytes;
10542 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10545 * dev_get_tstats64 - ndo_get_stats64 implementation
10546 * @dev: device to get statistics from
10547 * @s: place to store stats
10549 * Populate @s from dev->stats and dev->tstats. Can be used as
10550 * ndo_get_stats64() callback.
10552 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10554 netdev_stats_to_stats64(s, &dev->stats);
10555 dev_fetch_sw_netstats(s, dev->tstats);
10557 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10559 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10561 struct netdev_queue *queue = dev_ingress_queue(dev);
10563 #ifdef CONFIG_NET_CLS_ACT
10566 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10569 netdev_init_one_queue(dev, queue, NULL);
10570 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10571 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10572 rcu_assign_pointer(dev->ingress_queue, queue);
10577 static const struct ethtool_ops default_ethtool_ops;
10579 void netdev_set_default_ethtool_ops(struct net_device *dev,
10580 const struct ethtool_ops *ops)
10582 if (dev->ethtool_ops == &default_ethtool_ops)
10583 dev->ethtool_ops = ops;
10585 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10588 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10589 * @dev: netdev to enable the IRQ coalescing on
10591 * Sets a conservative default for SW IRQ coalescing. Users can use
10592 * sysfs attributes to override the default values.
10594 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10596 WARN_ON(dev->reg_state == NETREG_REGISTERED);
10598 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
10599 dev->gro_flush_timeout = 20000;
10600 dev->napi_defer_hard_irqs = 1;
10603 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10605 void netdev_freemem(struct net_device *dev)
10607 char *addr = (char *)dev - dev->padded;
10613 * alloc_netdev_mqs - allocate network device
10614 * @sizeof_priv: size of private data to allocate space for
10615 * @name: device name format string
10616 * @name_assign_type: origin of device name
10617 * @setup: callback to initialize device
10618 * @txqs: the number of TX subqueues to allocate
10619 * @rxqs: the number of RX subqueues to allocate
10621 * Allocates a struct net_device with private data area for driver use
10622 * and performs basic initialization. Also allocates subqueue structs
10623 * for each queue on the device.
10625 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10626 unsigned char name_assign_type,
10627 void (*setup)(struct net_device *),
10628 unsigned int txqs, unsigned int rxqs)
10630 struct net_device *dev;
10631 unsigned int alloc_size;
10632 struct net_device *p;
10634 BUG_ON(strlen(name) >= sizeof(dev->name));
10637 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10642 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10646 alloc_size = sizeof(struct net_device);
10648 /* ensure 32-byte alignment of private area */
10649 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10650 alloc_size += sizeof_priv;
10652 /* ensure 32-byte alignment of whole construct */
10653 alloc_size += NETDEV_ALIGN - 1;
10655 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10659 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10660 dev->padded = (char *)dev - (char *)p;
10662 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
10663 #ifdef CONFIG_PCPU_DEV_REFCNT
10664 dev->pcpu_refcnt = alloc_percpu(int);
10665 if (!dev->pcpu_refcnt)
10669 refcount_set(&dev->dev_refcnt, 1);
10672 if (dev_addr_init(dev))
10678 dev_net_set(dev, &init_net);
10680 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10681 dev->xdp_zc_max_segs = 1;
10682 dev->gso_max_segs = GSO_MAX_SEGS;
10683 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10684 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
10685 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
10686 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10687 dev->tso_max_segs = TSO_MAX_SEGS;
10688 dev->upper_level = 1;
10689 dev->lower_level = 1;
10690 #ifdef CONFIG_LOCKDEP
10691 dev->nested_level = 0;
10692 INIT_LIST_HEAD(&dev->unlink_list);
10695 INIT_LIST_HEAD(&dev->napi_list);
10696 INIT_LIST_HEAD(&dev->unreg_list);
10697 INIT_LIST_HEAD(&dev->close_list);
10698 INIT_LIST_HEAD(&dev->link_watch_list);
10699 INIT_LIST_HEAD(&dev->adj_list.upper);
10700 INIT_LIST_HEAD(&dev->adj_list.lower);
10701 INIT_LIST_HEAD(&dev->ptype_all);
10702 INIT_LIST_HEAD(&dev->ptype_specific);
10703 INIT_LIST_HEAD(&dev->net_notifier_list);
10704 #ifdef CONFIG_NET_SCHED
10705 hash_init(dev->qdisc_hash);
10707 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10710 if (!dev->tx_queue_len) {
10711 dev->priv_flags |= IFF_NO_QUEUE;
10712 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10715 dev->num_tx_queues = txqs;
10716 dev->real_num_tx_queues = txqs;
10717 if (netif_alloc_netdev_queues(dev))
10720 dev->num_rx_queues = rxqs;
10721 dev->real_num_rx_queues = rxqs;
10722 if (netif_alloc_rx_queues(dev))
10725 strcpy(dev->name, name);
10726 dev->name_assign_type = name_assign_type;
10727 dev->group = INIT_NETDEV_GROUP;
10728 if (!dev->ethtool_ops)
10729 dev->ethtool_ops = &default_ethtool_ops;
10731 nf_hook_netdev_init(dev);
10740 #ifdef CONFIG_PCPU_DEV_REFCNT
10741 free_percpu(dev->pcpu_refcnt);
10744 netdev_freemem(dev);
10747 EXPORT_SYMBOL(alloc_netdev_mqs);
10750 * free_netdev - free network device
10753 * This function does the last stage of destroying an allocated device
10754 * interface. The reference to the device object is released. If this
10755 * is the last reference then it will be freed.Must be called in process
10758 void free_netdev(struct net_device *dev)
10760 struct napi_struct *p, *n;
10764 /* When called immediately after register_netdevice() failed the unwind
10765 * handling may still be dismantling the device. Handle that case by
10766 * deferring the free.
10768 if (dev->reg_state == NETREG_UNREGISTERING) {
10770 dev->needs_free_netdev = true;
10774 netif_free_tx_queues(dev);
10775 netif_free_rx_queues(dev);
10777 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10779 /* Flush device addresses */
10780 dev_addr_flush(dev);
10782 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10785 ref_tracker_dir_exit(&dev->refcnt_tracker);
10786 #ifdef CONFIG_PCPU_DEV_REFCNT
10787 free_percpu(dev->pcpu_refcnt);
10788 dev->pcpu_refcnt = NULL;
10790 free_percpu(dev->core_stats);
10791 dev->core_stats = NULL;
10792 free_percpu(dev->xdp_bulkq);
10793 dev->xdp_bulkq = NULL;
10795 /* Compatibility with error handling in drivers */
10796 if (dev->reg_state == NETREG_UNINITIALIZED) {
10797 netdev_freemem(dev);
10801 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10802 dev->reg_state = NETREG_RELEASED;
10804 /* will free via device release */
10805 put_device(&dev->dev);
10807 EXPORT_SYMBOL(free_netdev);
10810 * synchronize_net - Synchronize with packet receive processing
10812 * Wait for packets currently being received to be done.
10813 * Does not block later packets from starting.
10815 void synchronize_net(void)
10818 if (rtnl_is_locked())
10819 synchronize_rcu_expedited();
10823 EXPORT_SYMBOL(synchronize_net);
10826 * unregister_netdevice_queue - remove device from the kernel
10830 * This function shuts down a device interface and removes it
10831 * from the kernel tables.
10832 * If head not NULL, device is queued to be unregistered later.
10834 * Callers must hold the rtnl semaphore. You may want
10835 * unregister_netdev() instead of this.
10838 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10843 list_move_tail(&dev->unreg_list, head);
10847 list_add(&dev->unreg_list, &single);
10848 unregister_netdevice_many(&single);
10851 EXPORT_SYMBOL(unregister_netdevice_queue);
10853 void unregister_netdevice_many_notify(struct list_head *head,
10854 u32 portid, const struct nlmsghdr *nlh)
10856 struct net_device *dev, *tmp;
10857 LIST_HEAD(close_head);
10859 BUG_ON(dev_boot_phase);
10862 if (list_empty(head))
10865 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10866 /* Some devices call without registering
10867 * for initialization unwind. Remove those
10868 * devices and proceed with the remaining.
10870 if (dev->reg_state == NETREG_UNINITIALIZED) {
10871 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10875 list_del(&dev->unreg_list);
10878 dev->dismantle = true;
10879 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10882 /* If device is running, close it first. */
10883 list_for_each_entry(dev, head, unreg_list)
10884 list_add_tail(&dev->close_list, &close_head);
10885 dev_close_many(&close_head, true);
10887 list_for_each_entry(dev, head, unreg_list) {
10888 /* And unlink it from device chain. */
10889 write_lock(&dev_base_lock);
10890 unlist_netdevice(dev, false);
10891 dev->reg_state = NETREG_UNREGISTERING;
10892 write_unlock(&dev_base_lock);
10894 flush_all_backlogs();
10898 list_for_each_entry(dev, head, unreg_list) {
10899 struct sk_buff *skb = NULL;
10901 /* Shutdown queueing discipline. */
10903 dev_tcx_uninstall(dev);
10904 dev_xdp_uninstall(dev);
10905 bpf_dev_bound_netdev_unregister(dev);
10907 netdev_offload_xstats_disable_all(dev);
10909 /* Notify protocols, that we are about to destroy
10910 * this device. They should clean all the things.
10912 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10914 if (!dev->rtnl_link_ops ||
10915 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10916 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10917 GFP_KERNEL, NULL, 0,
10921 * Flush the unicast and multicast chains
10926 netdev_name_node_alt_flush(dev);
10927 netdev_name_node_free(dev->name_node);
10929 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10931 if (dev->netdev_ops->ndo_uninit)
10932 dev->netdev_ops->ndo_uninit(dev);
10935 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10937 /* Notifier chain MUST detach us all upper devices. */
10938 WARN_ON(netdev_has_any_upper_dev(dev));
10939 WARN_ON(netdev_has_any_lower_dev(dev));
10941 /* Remove entries from kobject tree */
10942 netdev_unregister_kobject(dev);
10944 /* Remove XPS queueing entries */
10945 netif_reset_xps_queues_gt(dev, 0);
10951 list_for_each_entry(dev, head, unreg_list) {
10952 netdev_put(dev, &dev->dev_registered_tracker);
10960 * unregister_netdevice_many - unregister many devices
10961 * @head: list of devices
10963 * Note: As most callers use a stack allocated list_head,
10964 * we force a list_del() to make sure stack wont be corrupted later.
10966 void unregister_netdevice_many(struct list_head *head)
10968 unregister_netdevice_many_notify(head, 0, NULL);
10970 EXPORT_SYMBOL(unregister_netdevice_many);
10973 * unregister_netdev - remove device from the kernel
10976 * This function shuts down a device interface and removes it
10977 * from the kernel tables.
10979 * This is just a wrapper for unregister_netdevice that takes
10980 * the rtnl semaphore. In general you want to use this and not
10981 * unregister_netdevice.
10983 void unregister_netdev(struct net_device *dev)
10986 unregister_netdevice(dev);
10989 EXPORT_SYMBOL(unregister_netdev);
10992 * __dev_change_net_namespace - move device to different nethost namespace
10994 * @net: network namespace
10995 * @pat: If not NULL name pattern to try if the current device name
10996 * is already taken in the destination network namespace.
10997 * @new_ifindex: If not zero, specifies device index in the target
11000 * This function shuts down a device interface and moves it
11001 * to a new network namespace. On success 0 is returned, on
11002 * a failure a netagive errno code is returned.
11004 * Callers must hold the rtnl semaphore.
11007 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11008 const char *pat, int new_ifindex)
11010 struct net *net_old = dev_net(dev);
11015 /* Don't allow namespace local devices to be moved. */
11017 if (dev->features & NETIF_F_NETNS_LOCAL)
11020 /* Ensure the device has been registrered */
11021 if (dev->reg_state != NETREG_REGISTERED)
11024 /* Get out if there is nothing todo */
11026 if (net_eq(net_old, net))
11029 /* Pick the destination device name, and ensure
11030 * we can use it in the destination network namespace.
11033 if (netdev_name_in_use(net, dev->name)) {
11034 /* We get here if we can't use the current device name */
11037 err = dev_get_valid_name(net, dev, pat);
11042 /* Check that new_ifindex isn't used yet. */
11044 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11048 * And now a mini version of register_netdevice unregister_netdevice.
11051 /* If device is running close it first. */
11054 /* And unlink it from device chain */
11055 unlist_netdevice(dev, true);
11059 /* Shutdown queueing discipline. */
11062 /* Notify protocols, that we are about to destroy
11063 * this device. They should clean all the things.
11065 * Note that dev->reg_state stays at NETREG_REGISTERED.
11066 * This is wanted because this way 8021q and macvlan know
11067 * the device is just moving and can keep their slaves up.
11069 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11072 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11073 /* If there is an ifindex conflict assign a new one */
11074 if (!new_ifindex) {
11075 if (__dev_get_by_index(net, dev->ifindex))
11076 new_ifindex = dev_new_index(net);
11078 new_ifindex = dev->ifindex;
11081 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11085 * Flush the unicast and multicast chains
11090 /* Send a netdev-removed uevent to the old namespace */
11091 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11092 netdev_adjacent_del_links(dev);
11094 /* Move per-net netdevice notifiers that are following the netdevice */
11095 move_netdevice_notifiers_dev_net(dev, net);
11097 /* Actually switch the network namespace */
11098 dev_net_set(dev, net);
11099 dev->ifindex = new_ifindex;
11101 /* Send a netdev-add uevent to the new namespace */
11102 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11103 netdev_adjacent_add_links(dev);
11105 /* Fixup kobjects */
11106 err = device_rename(&dev->dev, dev->name);
11109 /* Adapt owner in case owning user namespace of target network
11110 * namespace is different from the original one.
11112 err = netdev_change_owner(dev, net_old, net);
11115 /* Add the device back in the hashes */
11116 list_netdevice(dev);
11118 /* Notify protocols, that a new device appeared. */
11119 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11122 * Prevent userspace races by waiting until the network
11123 * device is fully setup before sending notifications.
11125 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11132 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11134 static int dev_cpu_dead(unsigned int oldcpu)
11136 struct sk_buff **list_skb;
11137 struct sk_buff *skb;
11139 struct softnet_data *sd, *oldsd, *remsd = NULL;
11141 local_irq_disable();
11142 cpu = smp_processor_id();
11143 sd = &per_cpu(softnet_data, cpu);
11144 oldsd = &per_cpu(softnet_data, oldcpu);
11146 /* Find end of our completion_queue. */
11147 list_skb = &sd->completion_queue;
11149 list_skb = &(*list_skb)->next;
11150 /* Append completion queue from offline CPU. */
11151 *list_skb = oldsd->completion_queue;
11152 oldsd->completion_queue = NULL;
11154 /* Append output queue from offline CPU. */
11155 if (oldsd->output_queue) {
11156 *sd->output_queue_tailp = oldsd->output_queue;
11157 sd->output_queue_tailp = oldsd->output_queue_tailp;
11158 oldsd->output_queue = NULL;
11159 oldsd->output_queue_tailp = &oldsd->output_queue;
11161 /* Append NAPI poll list from offline CPU, with one exception :
11162 * process_backlog() must be called by cpu owning percpu backlog.
11163 * We properly handle process_queue & input_pkt_queue later.
11165 while (!list_empty(&oldsd->poll_list)) {
11166 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11167 struct napi_struct,
11170 list_del_init(&napi->poll_list);
11171 if (napi->poll == process_backlog)
11174 ____napi_schedule(sd, napi);
11177 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11178 local_irq_enable();
11181 remsd = oldsd->rps_ipi_list;
11182 oldsd->rps_ipi_list = NULL;
11184 /* send out pending IPI's on offline CPU */
11185 net_rps_send_ipi(remsd);
11187 /* Process offline CPU's input_pkt_queue */
11188 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11190 input_queue_head_incr(oldsd);
11192 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11194 input_queue_head_incr(oldsd);
11201 * netdev_increment_features - increment feature set by one
11202 * @all: current feature set
11203 * @one: new feature set
11204 * @mask: mask feature set
11206 * Computes a new feature set after adding a device with feature set
11207 * @one to the master device with current feature set @all. Will not
11208 * enable anything that is off in @mask. Returns the new feature set.
11210 netdev_features_t netdev_increment_features(netdev_features_t all,
11211 netdev_features_t one, netdev_features_t mask)
11213 if (mask & NETIF_F_HW_CSUM)
11214 mask |= NETIF_F_CSUM_MASK;
11215 mask |= NETIF_F_VLAN_CHALLENGED;
11217 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11218 all &= one | ~NETIF_F_ALL_FOR_ALL;
11220 /* If one device supports hw checksumming, set for all. */
11221 if (all & NETIF_F_HW_CSUM)
11222 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11226 EXPORT_SYMBOL(netdev_increment_features);
11228 static struct hlist_head * __net_init netdev_create_hash(void)
11231 struct hlist_head *hash;
11233 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11235 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11236 INIT_HLIST_HEAD(&hash[i]);
11241 /* Initialize per network namespace state */
11242 static int __net_init netdev_init(struct net *net)
11244 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11245 8 * sizeof_field(struct napi_struct, gro_bitmask));
11247 INIT_LIST_HEAD(&net->dev_base_head);
11249 net->dev_name_head = netdev_create_hash();
11250 if (net->dev_name_head == NULL)
11253 net->dev_index_head = netdev_create_hash();
11254 if (net->dev_index_head == NULL)
11257 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11262 kfree(net->dev_name_head);
11268 * netdev_drivername - network driver for the device
11269 * @dev: network device
11271 * Determine network driver for device.
11273 const char *netdev_drivername(const struct net_device *dev)
11275 const struct device_driver *driver;
11276 const struct device *parent;
11277 const char *empty = "";
11279 parent = dev->dev.parent;
11283 driver = parent->driver;
11284 if (driver && driver->name)
11285 return driver->name;
11289 static void __netdev_printk(const char *level, const struct net_device *dev,
11290 struct va_format *vaf)
11292 if (dev && dev->dev.parent) {
11293 dev_printk_emit(level[1] - '0',
11296 dev_driver_string(dev->dev.parent),
11297 dev_name(dev->dev.parent),
11298 netdev_name(dev), netdev_reg_state(dev),
11301 printk("%s%s%s: %pV",
11302 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11304 printk("%s(NULL net_device): %pV", level, vaf);
11308 void netdev_printk(const char *level, const struct net_device *dev,
11309 const char *format, ...)
11311 struct va_format vaf;
11314 va_start(args, format);
11319 __netdev_printk(level, dev, &vaf);
11323 EXPORT_SYMBOL(netdev_printk);
11325 #define define_netdev_printk_level(func, level) \
11326 void func(const struct net_device *dev, const char *fmt, ...) \
11328 struct va_format vaf; \
11331 va_start(args, fmt); \
11336 __netdev_printk(level, dev, &vaf); \
11340 EXPORT_SYMBOL(func);
11342 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11343 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11344 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11345 define_netdev_printk_level(netdev_err, KERN_ERR);
11346 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11347 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11348 define_netdev_printk_level(netdev_info, KERN_INFO);
11350 static void __net_exit netdev_exit(struct net *net)
11352 kfree(net->dev_name_head);
11353 kfree(net->dev_index_head);
11354 if (net != &init_net)
11355 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11358 static struct pernet_operations __net_initdata netdev_net_ops = {
11359 .init = netdev_init,
11360 .exit = netdev_exit,
11363 static void __net_exit default_device_exit_net(struct net *net)
11365 struct net_device *dev, *aux;
11367 * Push all migratable network devices back to the
11368 * initial network namespace
11371 for_each_netdev_safe(net, dev, aux) {
11373 char fb_name[IFNAMSIZ];
11375 /* Ignore unmoveable devices (i.e. loopback) */
11376 if (dev->features & NETIF_F_NETNS_LOCAL)
11379 /* Leave virtual devices for the generic cleanup */
11380 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11383 /* Push remaining network devices to init_net */
11384 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11385 if (netdev_name_in_use(&init_net, fb_name))
11386 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11387 err = dev_change_net_namespace(dev, &init_net, fb_name);
11389 pr_emerg("%s: failed to move %s to init_net: %d\n",
11390 __func__, dev->name, err);
11396 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11398 /* At exit all network devices most be removed from a network
11399 * namespace. Do this in the reverse order of registration.
11400 * Do this across as many network namespaces as possible to
11401 * improve batching efficiency.
11403 struct net_device *dev;
11405 LIST_HEAD(dev_kill_list);
11408 list_for_each_entry(net, net_list, exit_list) {
11409 default_device_exit_net(net);
11413 list_for_each_entry(net, net_list, exit_list) {
11414 for_each_netdev_reverse(net, dev) {
11415 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11416 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11418 unregister_netdevice_queue(dev, &dev_kill_list);
11421 unregister_netdevice_many(&dev_kill_list);
11425 static struct pernet_operations __net_initdata default_device_ops = {
11426 .exit_batch = default_device_exit_batch,
11430 * Initialize the DEV module. At boot time this walks the device list and
11431 * unhooks any devices that fail to initialise (normally hardware not
11432 * present) and leaves us with a valid list of present and active devices.
11437 * This is called single threaded during boot, so no need
11438 * to take the rtnl semaphore.
11440 static int __init net_dev_init(void)
11442 int i, rc = -ENOMEM;
11444 BUG_ON(!dev_boot_phase);
11446 if (dev_proc_init())
11449 if (netdev_kobject_init())
11452 INIT_LIST_HEAD(&ptype_all);
11453 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11454 INIT_LIST_HEAD(&ptype_base[i]);
11456 if (register_pernet_subsys(&netdev_net_ops))
11460 * Initialise the packet receive queues.
11463 for_each_possible_cpu(i) {
11464 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11465 struct softnet_data *sd = &per_cpu(softnet_data, i);
11467 INIT_WORK(flush, flush_backlog);
11469 skb_queue_head_init(&sd->input_pkt_queue);
11470 skb_queue_head_init(&sd->process_queue);
11471 #ifdef CONFIG_XFRM_OFFLOAD
11472 skb_queue_head_init(&sd->xfrm_backlog);
11474 INIT_LIST_HEAD(&sd->poll_list);
11475 sd->output_queue_tailp = &sd->output_queue;
11477 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11480 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11481 spin_lock_init(&sd->defer_lock);
11483 init_gro_hash(&sd->backlog);
11484 sd->backlog.poll = process_backlog;
11485 sd->backlog.weight = weight_p;
11488 dev_boot_phase = 0;
11490 /* The loopback device is special if any other network devices
11491 * is present in a network namespace the loopback device must
11492 * be present. Since we now dynamically allocate and free the
11493 * loopback device ensure this invariant is maintained by
11494 * keeping the loopback device as the first device on the
11495 * list of network devices. Ensuring the loopback devices
11496 * is the first device that appears and the last network device
11499 if (register_pernet_device(&loopback_net_ops))
11502 if (register_pernet_device(&default_device_ops))
11505 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11506 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11508 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11509 NULL, dev_cpu_dead);
11516 subsys_initcall(net_dev_init);