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>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_netdev.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
155 #include "net-sysfs.h"
158 static DEFINE_SPINLOCK(ptype_lock);
159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 struct list_head ptype_all __read_mostly; /* Taps */
162 static int netif_rx_internal(struct sk_buff *skb);
163 static int call_netdevice_notifiers_info(unsigned long val,
164 struct netdev_notifier_info *info);
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 kfree(name_node->name);
349 netdev_name_node_free(name_node);
352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
354 struct netdev_name_node *name_node;
355 struct net *net = dev_net(dev);
357 name_node = netdev_name_node_lookup(net, name);
360 /* lookup might have found our primary name or a name belonging
363 if (name_node == dev->name_node || name_node->dev != dev)
366 netdev_name_node_del(name_node);
368 __netdev_name_node_alt_destroy(name_node);
373 static void netdev_name_node_alt_flush(struct net_device *dev)
375 struct netdev_name_node *name_node, *tmp;
377 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
378 __netdev_name_node_alt_destroy(name_node);
381 /* Device list insertion */
382 static void list_netdevice(struct net_device *dev)
384 struct netdev_name_node *name_node;
385 struct net *net = dev_net(dev);
389 write_lock(&dev_base_lock);
390 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
391 netdev_name_node_add(net, dev->name_node);
392 hlist_add_head_rcu(&dev->index_hlist,
393 dev_index_hash(net, dev->ifindex));
394 write_unlock(&dev_base_lock);
396 netdev_for_each_altname(dev, name_node)
397 netdev_name_node_add(net, name_node);
399 dev_base_seq_inc(net);
402 /* Device list removal
403 * caller must respect a RCU grace period before freeing/reusing dev
405 static void unlist_netdevice(struct net_device *dev, bool lock)
407 struct netdev_name_node *name_node;
411 netdev_for_each_altname(dev, name_node)
412 netdev_name_node_del(name_node);
414 /* Unlink dev from the device chain */
416 write_lock(&dev_base_lock);
417 list_del_rcu(&dev->dev_list);
418 netdev_name_node_del(dev->name_node);
419 hlist_del_rcu(&dev->index_hlist);
421 write_unlock(&dev_base_lock);
423 dev_base_seq_inc(dev_net(dev));
430 static RAW_NOTIFIER_HEAD(netdev_chain);
433 * Device drivers call our routines to queue packets here. We empty the
434 * queue in the local softnet handler.
437 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
438 EXPORT_PER_CPU_SYMBOL(softnet_data);
440 #ifdef CONFIG_LOCKDEP
442 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
443 * according to dev->type
445 static const unsigned short netdev_lock_type[] = {
446 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
447 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
448 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
449 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
450 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
451 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
452 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
453 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
454 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
455 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
456 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
457 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
458 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
459 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
460 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
462 static const char *const netdev_lock_name[] = {
463 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
464 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
465 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
466 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
467 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
468 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
469 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
470 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
471 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
472 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
473 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
474 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
475 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
476 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
477 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
479 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
480 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
482 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
486 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
487 if (netdev_lock_type[i] == dev_type)
489 /* the last key is used by default */
490 return ARRAY_SIZE(netdev_lock_type) - 1;
493 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
494 unsigned short dev_type)
498 i = netdev_lock_pos(dev_type);
499 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
500 netdev_lock_name[i]);
503 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
507 i = netdev_lock_pos(dev->type);
508 lockdep_set_class_and_name(&dev->addr_list_lock,
509 &netdev_addr_lock_key[i],
510 netdev_lock_name[i]);
513 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
514 unsigned short dev_type)
518 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
523 /*******************************************************************************
525 * Protocol management and registration routines
527 *******************************************************************************/
531 * Add a protocol ID to the list. Now that the input handler is
532 * smarter we can dispense with all the messy stuff that used to be
535 * BEWARE!!! Protocol handlers, mangling input packets,
536 * MUST BE last in hash buckets and checking protocol handlers
537 * MUST start from promiscuous ptype_all chain in net_bh.
538 * It is true now, do not change it.
539 * Explanation follows: if protocol handler, mangling packet, will
540 * be the first on list, it is not able to sense, that packet
541 * is cloned and should be copied-on-write, so that it will
542 * change it and subsequent readers will get broken packet.
546 static inline struct list_head *ptype_head(const struct packet_type *pt)
548 if (pt->type == htons(ETH_P_ALL))
549 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
551 return pt->dev ? &pt->dev->ptype_specific :
552 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
556 * dev_add_pack - add packet handler
557 * @pt: packet type declaration
559 * Add a protocol handler to the networking stack. The passed &packet_type
560 * is linked into kernel lists and may not be freed until it has been
561 * removed from the kernel lists.
563 * This call does not sleep therefore it can not
564 * guarantee all CPU's that are in middle of receiving packets
565 * will see the new packet type (until the next received packet).
568 void dev_add_pack(struct packet_type *pt)
570 struct list_head *head = ptype_head(pt);
572 spin_lock(&ptype_lock);
573 list_add_rcu(&pt->list, head);
574 spin_unlock(&ptype_lock);
576 EXPORT_SYMBOL(dev_add_pack);
579 * __dev_remove_pack - remove packet handler
580 * @pt: packet type declaration
582 * Remove a protocol handler that was previously added to the kernel
583 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
584 * from the kernel lists and can be freed or reused once this function
587 * The packet type might still be in use by receivers
588 * and must not be freed until after all the CPU's have gone
589 * through a quiescent state.
591 void __dev_remove_pack(struct packet_type *pt)
593 struct list_head *head = ptype_head(pt);
594 struct packet_type *pt1;
596 spin_lock(&ptype_lock);
598 list_for_each_entry(pt1, head, list) {
600 list_del_rcu(&pt->list);
605 pr_warn("dev_remove_pack: %p not found\n", pt);
607 spin_unlock(&ptype_lock);
609 EXPORT_SYMBOL(__dev_remove_pack);
612 * dev_remove_pack - remove packet handler
613 * @pt: packet type declaration
615 * Remove a protocol handler that was previously added to the kernel
616 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
617 * from the kernel lists and can be freed or reused once this function
620 * This call sleeps to guarantee that no CPU is looking at the packet
623 void dev_remove_pack(struct packet_type *pt)
625 __dev_remove_pack(pt);
629 EXPORT_SYMBOL(dev_remove_pack);
632 /*******************************************************************************
634 * Device Interface Subroutines
636 *******************************************************************************/
639 * dev_get_iflink - get 'iflink' value of a interface
640 * @dev: targeted interface
642 * Indicates the ifindex the interface is linked to.
643 * Physical interfaces have the same 'ifindex' and 'iflink' values.
646 int dev_get_iflink(const struct net_device *dev)
648 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
649 return dev->netdev_ops->ndo_get_iflink(dev);
653 EXPORT_SYMBOL(dev_get_iflink);
656 * dev_fill_metadata_dst - Retrieve tunnel egress information.
657 * @dev: targeted interface
660 * For better visibility of tunnel traffic OVS needs to retrieve
661 * egress tunnel information for a packet. Following API allows
662 * user to get this info.
664 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
666 struct ip_tunnel_info *info;
668 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
671 info = skb_tunnel_info_unclone(skb);
674 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
677 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
679 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
681 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
683 int k = stack->num_paths++;
685 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
688 return &stack->path[k];
691 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
692 struct net_device_path_stack *stack)
694 const struct net_device *last_dev;
695 struct net_device_path_ctx ctx = {
698 struct net_device_path *path;
701 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
702 stack->num_paths = 0;
703 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
705 path = dev_fwd_path(stack);
709 memset(path, 0, sizeof(struct net_device_path));
710 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
714 if (WARN_ON_ONCE(last_dev == ctx.dev))
721 path = dev_fwd_path(stack);
724 path->type = DEV_PATH_ETHERNET;
729 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
732 * __dev_get_by_name - find a device by its name
733 * @net: the applicable net namespace
734 * @name: name to find
736 * Find an interface by name. Must be called under RTNL semaphore
737 * or @dev_base_lock. If the name is found a pointer to the device
738 * is returned. If the name is not found then %NULL is returned. The
739 * reference counters are not incremented so the caller must be
740 * careful with locks.
743 struct net_device *__dev_get_by_name(struct net *net, const char *name)
745 struct netdev_name_node *node_name;
747 node_name = netdev_name_node_lookup(net, name);
748 return node_name ? node_name->dev : NULL;
750 EXPORT_SYMBOL(__dev_get_by_name);
753 * dev_get_by_name_rcu - find a device by its name
754 * @net: the applicable net namespace
755 * @name: name to find
757 * Find an interface by name.
758 * If the name is found a pointer to the device is returned.
759 * If the name is not found then %NULL is returned.
760 * The reference counters are not incremented so the caller must be
761 * careful with locks. The caller must hold RCU lock.
764 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
766 struct netdev_name_node *node_name;
768 node_name = netdev_name_node_lookup_rcu(net, name);
769 return node_name ? node_name->dev : NULL;
771 EXPORT_SYMBOL(dev_get_by_name_rcu);
774 * dev_get_by_name - find a device by its name
775 * @net: the applicable net namespace
776 * @name: name to find
778 * Find an interface by name. This can be called from any
779 * context and does its own locking. The returned handle has
780 * the usage count incremented and the caller must use dev_put() to
781 * release it when it is no longer needed. %NULL is returned if no
782 * matching device is found.
785 struct net_device *dev_get_by_name(struct net *net, const char *name)
787 struct net_device *dev;
790 dev = dev_get_by_name_rcu(net, name);
795 EXPORT_SYMBOL(dev_get_by_name);
798 * __dev_get_by_index - find a device by its ifindex
799 * @net: the applicable net namespace
800 * @ifindex: index of device
802 * Search for an interface by index. Returns %NULL if the device
803 * is not found or a pointer to the device. The device has not
804 * had its reference counter increased so the caller must be careful
805 * about locking. The caller must hold either the RTNL semaphore
809 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
811 struct net_device *dev;
812 struct hlist_head *head = dev_index_hash(net, ifindex);
814 hlist_for_each_entry(dev, head, index_hlist)
815 if (dev->ifindex == ifindex)
820 EXPORT_SYMBOL(__dev_get_by_index);
823 * dev_get_by_index_rcu - find a device by its ifindex
824 * @net: the applicable net namespace
825 * @ifindex: index of device
827 * Search for an interface by index. Returns %NULL if the device
828 * is not found or a pointer to the device. The device has not
829 * had its reference counter increased so the caller must be careful
830 * about locking. The caller must hold RCU lock.
833 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
835 struct net_device *dev;
836 struct hlist_head *head = dev_index_hash(net, ifindex);
838 hlist_for_each_entry_rcu(dev, head, index_hlist)
839 if (dev->ifindex == ifindex)
844 EXPORT_SYMBOL(dev_get_by_index_rcu);
848 * dev_get_by_index - find a device by its ifindex
849 * @net: the applicable net namespace
850 * @ifindex: index of device
852 * Search for an interface by index. Returns NULL if the device
853 * is not found or a pointer to the device. The device returned has
854 * had a reference added and the pointer is safe until the user calls
855 * dev_put to indicate they have finished with it.
858 struct net_device *dev_get_by_index(struct net *net, int ifindex)
860 struct net_device *dev;
863 dev = dev_get_by_index_rcu(net, ifindex);
868 EXPORT_SYMBOL(dev_get_by_index);
871 * dev_get_by_napi_id - find a device by napi_id
872 * @napi_id: ID of the NAPI struct
874 * Search for an interface by NAPI ID. Returns %NULL if the device
875 * is not found or a pointer to the device. The device has not had
876 * its reference counter increased so the caller must be careful
877 * about locking. The caller must hold RCU lock.
880 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
882 struct napi_struct *napi;
884 WARN_ON_ONCE(!rcu_read_lock_held());
886 if (napi_id < MIN_NAPI_ID)
889 napi = napi_by_id(napi_id);
891 return napi ? napi->dev : NULL;
893 EXPORT_SYMBOL(dev_get_by_napi_id);
896 * netdev_get_name - get a netdevice name, knowing its ifindex.
897 * @net: network namespace
898 * @name: a pointer to the buffer where the name will be stored.
899 * @ifindex: the ifindex of the interface to get the name from.
901 int netdev_get_name(struct net *net, char *name, int ifindex)
903 struct net_device *dev;
906 down_read(&devnet_rename_sem);
909 dev = dev_get_by_index_rcu(net, ifindex);
915 strcpy(name, dev->name);
920 up_read(&devnet_rename_sem);
925 * dev_getbyhwaddr_rcu - find a device by its hardware address
926 * @net: the applicable net namespace
927 * @type: media type of device
928 * @ha: hardware address
930 * Search for an interface by MAC address. Returns NULL if the device
931 * is not found or a pointer to the device.
932 * The caller must hold RCU or RTNL.
933 * The returned device has not had its ref count increased
934 * and the caller must therefore be careful about locking
938 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
941 struct net_device *dev;
943 for_each_netdev_rcu(net, dev)
944 if (dev->type == type &&
945 !memcmp(dev->dev_addr, ha, dev->addr_len))
950 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
952 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
954 struct net_device *dev, *ret = NULL;
957 for_each_netdev_rcu(net, dev)
958 if (dev->type == type) {
966 EXPORT_SYMBOL(dev_getfirstbyhwtype);
969 * __dev_get_by_flags - find any device with given flags
970 * @net: the applicable net namespace
971 * @if_flags: IFF_* values
972 * @mask: bitmask of bits in if_flags to check
974 * Search for any interface with the given flags. Returns NULL if a device
975 * is not found or a pointer to the device. Must be called inside
976 * rtnl_lock(), and result refcount is unchanged.
979 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
982 struct net_device *dev, *ret;
987 for_each_netdev(net, dev) {
988 if (((dev->flags ^ if_flags) & mask) == 0) {
995 EXPORT_SYMBOL(__dev_get_by_flags);
998 * dev_valid_name - check if name is okay for network device
1001 * Network device names need to be valid file names to
1002 * allow sysfs to work. We also disallow any kind of
1005 bool dev_valid_name(const char *name)
1009 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1011 if (!strcmp(name, ".") || !strcmp(name, ".."))
1015 if (*name == '/' || *name == ':' || isspace(*name))
1021 EXPORT_SYMBOL(dev_valid_name);
1024 * __dev_alloc_name - allocate a name for a device
1025 * @net: network namespace to allocate the device name in
1026 * @name: name format string
1027 * @buf: scratch buffer and result name string
1029 * Passed a format string - eg "lt%d" it will try and find a suitable
1030 * id. It scans list of devices to build up a free map, then chooses
1031 * the first empty slot. The caller must hold the dev_base or rtnl lock
1032 * while allocating the name and adding the device in order to avoid
1034 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1035 * Returns the number of the unit assigned or a negative errno code.
1038 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1042 const int max_netdevices = 8*PAGE_SIZE;
1043 unsigned long *inuse;
1044 struct net_device *d;
1046 if (!dev_valid_name(name))
1049 p = strchr(name, '%');
1052 * Verify the string as this thing may have come from
1053 * the user. There must be either one "%d" and no other "%"
1056 if (p[1] != 'd' || strchr(p + 2, '%'))
1059 /* Use one page as a bit array of possible slots */
1060 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1064 for_each_netdev(net, d) {
1065 struct netdev_name_node *name_node;
1067 netdev_for_each_altname(d, name_node) {
1068 if (!sscanf(name_node->name, name, &i))
1070 if (i < 0 || i >= max_netdevices)
1073 /* avoid cases where sscanf is not exact inverse of printf */
1074 snprintf(buf, IFNAMSIZ, name, i);
1075 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1076 __set_bit(i, inuse);
1078 if (!sscanf(d->name, name, &i))
1080 if (i < 0 || i >= max_netdevices)
1083 /* avoid cases where sscanf is not exact inverse of printf */
1084 snprintf(buf, IFNAMSIZ, name, i);
1085 if (!strncmp(buf, d->name, IFNAMSIZ))
1086 __set_bit(i, inuse);
1089 i = find_first_zero_bit(inuse, max_netdevices);
1090 free_page((unsigned long) inuse);
1093 snprintf(buf, IFNAMSIZ, name, i);
1094 if (!netdev_name_in_use(net, buf))
1097 /* It is possible to run out of possible slots
1098 * when the name is long and there isn't enough space left
1099 * for the digits, or if all bits are used.
1104 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1105 const char *want_name, char *out_name)
1109 if (!dev_valid_name(want_name))
1112 if (strchr(want_name, '%')) {
1113 ret = __dev_alloc_name(net, want_name, out_name);
1114 return ret < 0 ? ret : 0;
1115 } else if (netdev_name_in_use(net, want_name)) {
1117 } else if (out_name != want_name) {
1118 strscpy(out_name, want_name, IFNAMSIZ);
1124 static int dev_alloc_name_ns(struct net *net,
1125 struct net_device *dev,
1132 ret = __dev_alloc_name(net, name, buf);
1134 strscpy(dev->name, buf, IFNAMSIZ);
1139 * dev_alloc_name - allocate a name for a device
1141 * @name: name format string
1143 * Passed a format string - eg "lt%d" it will try and find a suitable
1144 * id. It scans list of devices to build up a free map, then chooses
1145 * the first empty slot. The caller must hold the dev_base or rtnl lock
1146 * while allocating the name and adding the device in order to avoid
1148 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1149 * Returns the number of the unit assigned or a negative errno code.
1152 int dev_alloc_name(struct net_device *dev, const char *name)
1154 return dev_alloc_name_ns(dev_net(dev), dev, name);
1156 EXPORT_SYMBOL(dev_alloc_name);
1158 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1164 ret = dev_prep_valid_name(net, dev, name, buf);
1166 strscpy(dev->name, buf, IFNAMSIZ);
1171 * dev_change_name - change name of a device
1173 * @newname: name (or format string) must be at least IFNAMSIZ
1175 * Change name of a device, can pass format strings "eth%d".
1178 int dev_change_name(struct net_device *dev, const char *newname)
1180 unsigned char old_assign_type;
1181 char oldname[IFNAMSIZ];
1187 BUG_ON(!dev_net(dev));
1191 /* Some auto-enslaved devices e.g. failover slaves are
1192 * special, as userspace might rename the device after
1193 * the interface had been brought up and running since
1194 * the point kernel initiated auto-enslavement. Allow
1195 * live name change even when these slave devices are
1198 * Typically, users of these auto-enslaving devices
1199 * don't actually care about slave name change, as
1200 * they are supposed to operate on master interface
1203 if (dev->flags & IFF_UP &&
1204 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1207 down_write(&devnet_rename_sem);
1209 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1210 up_write(&devnet_rename_sem);
1214 memcpy(oldname, dev->name, IFNAMSIZ);
1216 err = dev_get_valid_name(net, dev, newname);
1218 up_write(&devnet_rename_sem);
1222 if (oldname[0] && !strchr(oldname, '%'))
1223 netdev_info(dev, "renamed from %s\n", oldname);
1225 old_assign_type = dev->name_assign_type;
1226 dev->name_assign_type = NET_NAME_RENAMED;
1229 ret = device_rename(&dev->dev, dev->name);
1231 memcpy(dev->name, oldname, IFNAMSIZ);
1232 dev->name_assign_type = old_assign_type;
1233 up_write(&devnet_rename_sem);
1237 up_write(&devnet_rename_sem);
1239 netdev_adjacent_rename_links(dev, oldname);
1241 write_lock(&dev_base_lock);
1242 netdev_name_node_del(dev->name_node);
1243 write_unlock(&dev_base_lock);
1247 write_lock(&dev_base_lock);
1248 netdev_name_node_add(net, dev->name_node);
1249 write_unlock(&dev_base_lock);
1251 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1252 ret = notifier_to_errno(ret);
1255 /* err >= 0 after dev_alloc_name() or stores the first errno */
1258 down_write(&devnet_rename_sem);
1259 memcpy(dev->name, oldname, IFNAMSIZ);
1260 memcpy(oldname, newname, IFNAMSIZ);
1261 dev->name_assign_type = old_assign_type;
1262 old_assign_type = NET_NAME_RENAMED;
1265 netdev_err(dev, "name change rollback failed: %d\n",
1274 * dev_set_alias - change ifalias of a device
1276 * @alias: name up to IFALIASZ
1277 * @len: limit of bytes to copy from info
1279 * Set ifalias for a device,
1281 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1283 struct dev_ifalias *new_alias = NULL;
1285 if (len >= IFALIASZ)
1289 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1293 memcpy(new_alias->ifalias, alias, len);
1294 new_alias->ifalias[len] = 0;
1297 mutex_lock(&ifalias_mutex);
1298 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1299 mutex_is_locked(&ifalias_mutex));
1300 mutex_unlock(&ifalias_mutex);
1303 kfree_rcu(new_alias, rcuhead);
1307 EXPORT_SYMBOL(dev_set_alias);
1310 * dev_get_alias - get ifalias of a device
1312 * @name: buffer to store name of ifalias
1313 * @len: size of buffer
1315 * get ifalias for a device. Caller must make sure dev cannot go
1316 * away, e.g. rcu read lock or own a reference count to device.
1318 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1320 const struct dev_ifalias *alias;
1324 alias = rcu_dereference(dev->ifalias);
1326 ret = snprintf(name, len, "%s", alias->ifalias);
1333 * netdev_features_change - device changes features
1334 * @dev: device to cause notification
1336 * Called to indicate a device has changed features.
1338 void netdev_features_change(struct net_device *dev)
1340 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1342 EXPORT_SYMBOL(netdev_features_change);
1345 * netdev_state_change - device changes state
1346 * @dev: device to cause notification
1348 * Called to indicate a device has changed state. This function calls
1349 * the notifier chains for netdev_chain and sends a NEWLINK message
1350 * to the routing socket.
1352 void netdev_state_change(struct net_device *dev)
1354 if (dev->flags & IFF_UP) {
1355 struct netdev_notifier_change_info change_info = {
1359 call_netdevice_notifiers_info(NETDEV_CHANGE,
1361 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1364 EXPORT_SYMBOL(netdev_state_change);
1367 * __netdev_notify_peers - notify network peers about existence of @dev,
1368 * to be called when rtnl lock is already held.
1369 * @dev: network device
1371 * Generate traffic such that interested network peers are aware of
1372 * @dev, such as by generating a gratuitous ARP. This may be used when
1373 * a device wants to inform the rest of the network about some sort of
1374 * reconfiguration such as a failover event or virtual machine
1377 void __netdev_notify_peers(struct net_device *dev)
1380 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1381 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1383 EXPORT_SYMBOL(__netdev_notify_peers);
1386 * netdev_notify_peers - notify network peers about existence of @dev
1387 * @dev: network device
1389 * Generate traffic such that interested network peers are aware of
1390 * @dev, such as by generating a gratuitous ARP. This may be used when
1391 * a device wants to inform the rest of the network about some sort of
1392 * reconfiguration such as a failover event or virtual machine
1395 void netdev_notify_peers(struct net_device *dev)
1398 __netdev_notify_peers(dev);
1401 EXPORT_SYMBOL(netdev_notify_peers);
1403 static int napi_threaded_poll(void *data);
1405 static int napi_kthread_create(struct napi_struct *n)
1409 /* Create and wake up the kthread once to put it in
1410 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1411 * warning and work with loadavg.
1413 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1414 n->dev->name, n->napi_id);
1415 if (IS_ERR(n->thread)) {
1416 err = PTR_ERR(n->thread);
1417 pr_err("kthread_run failed with err %d\n", err);
1424 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1426 const struct net_device_ops *ops = dev->netdev_ops;
1430 dev_addr_check(dev);
1432 if (!netif_device_present(dev)) {
1433 /* may be detached because parent is runtime-suspended */
1434 if (dev->dev.parent)
1435 pm_runtime_resume(dev->dev.parent);
1436 if (!netif_device_present(dev))
1440 /* Block netpoll from trying to do any rx path servicing.
1441 * If we don't do this there is a chance ndo_poll_controller
1442 * or ndo_poll may be running while we open the device
1444 netpoll_poll_disable(dev);
1446 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1447 ret = notifier_to_errno(ret);
1451 set_bit(__LINK_STATE_START, &dev->state);
1453 if (ops->ndo_validate_addr)
1454 ret = ops->ndo_validate_addr(dev);
1456 if (!ret && ops->ndo_open)
1457 ret = ops->ndo_open(dev);
1459 netpoll_poll_enable(dev);
1462 clear_bit(__LINK_STATE_START, &dev->state);
1464 dev->flags |= IFF_UP;
1465 dev_set_rx_mode(dev);
1467 add_device_randomness(dev->dev_addr, dev->addr_len);
1474 * dev_open - prepare an interface for use.
1475 * @dev: device to open
1476 * @extack: netlink extended ack
1478 * Takes a device from down to up state. The device's private open
1479 * function is invoked and then the multicast lists are loaded. Finally
1480 * the device is moved into the up state and a %NETDEV_UP message is
1481 * sent to the netdev notifier chain.
1483 * Calling this function on an active interface is a nop. On a failure
1484 * a negative errno code is returned.
1486 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1490 if (dev->flags & IFF_UP)
1493 ret = __dev_open(dev, extack);
1497 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1498 call_netdevice_notifiers(NETDEV_UP, dev);
1502 EXPORT_SYMBOL(dev_open);
1504 static void __dev_close_many(struct list_head *head)
1506 struct net_device *dev;
1511 list_for_each_entry(dev, head, close_list) {
1512 /* Temporarily disable netpoll until the interface is down */
1513 netpoll_poll_disable(dev);
1515 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1517 clear_bit(__LINK_STATE_START, &dev->state);
1519 /* Synchronize to scheduled poll. We cannot touch poll list, it
1520 * can be even on different cpu. So just clear netif_running().
1522 * dev->stop() will invoke napi_disable() on all of it's
1523 * napi_struct instances on this device.
1525 smp_mb__after_atomic(); /* Commit netif_running(). */
1528 dev_deactivate_many(head);
1530 list_for_each_entry(dev, head, close_list) {
1531 const struct net_device_ops *ops = dev->netdev_ops;
1534 * Call the device specific close. This cannot fail.
1535 * Only if device is UP
1537 * We allow it to be called even after a DETACH hot-plug
1543 dev->flags &= ~IFF_UP;
1544 netpoll_poll_enable(dev);
1548 static void __dev_close(struct net_device *dev)
1552 list_add(&dev->close_list, &single);
1553 __dev_close_many(&single);
1557 void dev_close_many(struct list_head *head, bool unlink)
1559 struct net_device *dev, *tmp;
1561 /* Remove the devices that don't need to be closed */
1562 list_for_each_entry_safe(dev, tmp, head, close_list)
1563 if (!(dev->flags & IFF_UP))
1564 list_del_init(&dev->close_list);
1566 __dev_close_many(head);
1568 list_for_each_entry_safe(dev, tmp, head, close_list) {
1569 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1570 call_netdevice_notifiers(NETDEV_DOWN, dev);
1572 list_del_init(&dev->close_list);
1575 EXPORT_SYMBOL(dev_close_many);
1578 * dev_close - shutdown an interface.
1579 * @dev: device to shutdown
1581 * This function moves an active device into down state. A
1582 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1583 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1586 void dev_close(struct net_device *dev)
1588 if (dev->flags & IFF_UP) {
1591 list_add(&dev->close_list, &single);
1592 dev_close_many(&single, true);
1596 EXPORT_SYMBOL(dev_close);
1600 * dev_disable_lro - disable Large Receive Offload on a device
1603 * Disable Large Receive Offload (LRO) on a net device. Must be
1604 * called under RTNL. This is needed if received packets may be
1605 * forwarded to another interface.
1607 void dev_disable_lro(struct net_device *dev)
1609 struct net_device *lower_dev;
1610 struct list_head *iter;
1612 dev->wanted_features &= ~NETIF_F_LRO;
1613 netdev_update_features(dev);
1615 if (unlikely(dev->features & NETIF_F_LRO))
1616 netdev_WARN(dev, "failed to disable LRO!\n");
1618 netdev_for_each_lower_dev(dev, lower_dev, iter)
1619 dev_disable_lro(lower_dev);
1621 EXPORT_SYMBOL(dev_disable_lro);
1624 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1627 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1628 * called under RTNL. This is needed if Generic XDP is installed on
1631 static void dev_disable_gro_hw(struct net_device *dev)
1633 dev->wanted_features &= ~NETIF_F_GRO_HW;
1634 netdev_update_features(dev);
1636 if (unlikely(dev->features & NETIF_F_GRO_HW))
1637 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1640 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1643 case NETDEV_##val: \
1644 return "NETDEV_" __stringify(val);
1646 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1647 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1648 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1649 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1650 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1651 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1652 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1653 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1654 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1655 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1656 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1659 return "UNKNOWN_NETDEV_EVENT";
1661 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1663 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1664 struct net_device *dev)
1666 struct netdev_notifier_info info = {
1670 return nb->notifier_call(nb, val, &info);
1673 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1674 struct net_device *dev)
1678 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1679 err = notifier_to_errno(err);
1683 if (!(dev->flags & IFF_UP))
1686 call_netdevice_notifier(nb, NETDEV_UP, dev);
1690 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1691 struct net_device *dev)
1693 if (dev->flags & IFF_UP) {
1694 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1696 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1698 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1701 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1704 struct net_device *dev;
1707 for_each_netdev(net, dev) {
1708 err = call_netdevice_register_notifiers(nb, dev);
1715 for_each_netdev_continue_reverse(net, dev)
1716 call_netdevice_unregister_notifiers(nb, dev);
1720 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1723 struct net_device *dev;
1725 for_each_netdev(net, dev)
1726 call_netdevice_unregister_notifiers(nb, dev);
1729 static int dev_boot_phase = 1;
1732 * register_netdevice_notifier - register a network notifier block
1735 * Register a notifier to be called when network device events occur.
1736 * The notifier passed is linked into the kernel structures and must
1737 * not be reused until it has been unregistered. A negative errno code
1738 * is returned on a failure.
1740 * When registered all registration and up events are replayed
1741 * to the new notifier to allow device to have a race free
1742 * view of the network device list.
1745 int register_netdevice_notifier(struct notifier_block *nb)
1750 /* Close race with setup_net() and cleanup_net() */
1751 down_write(&pernet_ops_rwsem);
1753 err = raw_notifier_chain_register(&netdev_chain, nb);
1759 err = call_netdevice_register_net_notifiers(nb, net);
1766 up_write(&pernet_ops_rwsem);
1770 for_each_net_continue_reverse(net)
1771 call_netdevice_unregister_net_notifiers(nb, net);
1773 raw_notifier_chain_unregister(&netdev_chain, nb);
1776 EXPORT_SYMBOL(register_netdevice_notifier);
1779 * unregister_netdevice_notifier - unregister a network notifier block
1782 * Unregister a notifier previously registered by
1783 * register_netdevice_notifier(). The notifier is unlinked into the
1784 * kernel structures and may then be reused. A negative errno code
1785 * is returned on a failure.
1787 * After unregistering unregister and down device events are synthesized
1788 * for all devices on the device list to the removed notifier to remove
1789 * the need for special case cleanup code.
1792 int unregister_netdevice_notifier(struct notifier_block *nb)
1797 /* Close race with setup_net() and cleanup_net() */
1798 down_write(&pernet_ops_rwsem);
1800 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1805 call_netdevice_unregister_net_notifiers(nb, net);
1809 up_write(&pernet_ops_rwsem);
1812 EXPORT_SYMBOL(unregister_netdevice_notifier);
1814 static int __register_netdevice_notifier_net(struct net *net,
1815 struct notifier_block *nb,
1816 bool ignore_call_fail)
1820 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1826 err = call_netdevice_register_net_notifiers(nb, net);
1827 if (err && !ignore_call_fail)
1828 goto chain_unregister;
1833 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1837 static int __unregister_netdevice_notifier_net(struct net *net,
1838 struct notifier_block *nb)
1842 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1846 call_netdevice_unregister_net_notifiers(nb, net);
1851 * register_netdevice_notifier_net - register a per-netns network notifier block
1852 * @net: network namespace
1855 * Register a notifier to be called when network device events occur.
1856 * The notifier passed is linked into the kernel structures and must
1857 * not be reused until it has been unregistered. A negative errno code
1858 * is returned on a failure.
1860 * When registered all registration and up events are replayed
1861 * to the new notifier to allow device to have a race free
1862 * view of the network device list.
1865 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1870 err = __register_netdevice_notifier_net(net, nb, false);
1874 EXPORT_SYMBOL(register_netdevice_notifier_net);
1877 * unregister_netdevice_notifier_net - unregister a per-netns
1878 * network notifier block
1879 * @net: network namespace
1882 * Unregister a notifier previously registered by
1883 * register_netdevice_notifier(). The notifier is unlinked into the
1884 * kernel structures and may then be reused. A negative errno code
1885 * is returned on a failure.
1887 * After unregistering unregister and down device events are synthesized
1888 * for all devices on the device list to the removed notifier to remove
1889 * the need for special case cleanup code.
1892 int unregister_netdevice_notifier_net(struct net *net,
1893 struct notifier_block *nb)
1898 err = __unregister_netdevice_notifier_net(net, nb);
1902 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1904 int register_netdevice_notifier_dev_net(struct net_device *dev,
1905 struct notifier_block *nb,
1906 struct netdev_net_notifier *nn)
1911 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1914 list_add(&nn->list, &dev->net_notifier_list);
1919 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1921 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1922 struct notifier_block *nb,
1923 struct netdev_net_notifier *nn)
1928 list_del(&nn->list);
1929 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1933 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1935 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1938 struct netdev_net_notifier *nn;
1940 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1941 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1942 __register_netdevice_notifier_net(net, nn->nb, true);
1947 * call_netdevice_notifiers_info - call all network notifier blocks
1948 * @val: value passed unmodified to notifier function
1949 * @info: notifier information data
1951 * Call all network notifier blocks. Parameters and return value
1952 * are as for raw_notifier_call_chain().
1955 static int call_netdevice_notifiers_info(unsigned long val,
1956 struct netdev_notifier_info *info)
1958 struct net *net = dev_net(info->dev);
1963 /* Run per-netns notifier block chain first, then run the global one.
1964 * Hopefully, one day, the global one is going to be removed after
1965 * all notifier block registrators get converted to be per-netns.
1967 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1968 if (ret & NOTIFY_STOP_MASK)
1970 return raw_notifier_call_chain(&netdev_chain, val, info);
1974 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1975 * for and rollback on error
1976 * @val_up: value passed unmodified to notifier function
1977 * @val_down: value passed unmodified to the notifier function when
1978 * recovering from an error on @val_up
1979 * @info: notifier information data
1981 * Call all per-netns network notifier blocks, but not notifier blocks on
1982 * the global notifier chain. Parameters and return value are as for
1983 * raw_notifier_call_chain_robust().
1987 call_netdevice_notifiers_info_robust(unsigned long val_up,
1988 unsigned long val_down,
1989 struct netdev_notifier_info *info)
1991 struct net *net = dev_net(info->dev);
1995 return raw_notifier_call_chain_robust(&net->netdev_chain,
1996 val_up, val_down, info);
1999 static int call_netdevice_notifiers_extack(unsigned long val,
2000 struct net_device *dev,
2001 struct netlink_ext_ack *extack)
2003 struct netdev_notifier_info info = {
2008 return call_netdevice_notifiers_info(val, &info);
2012 * call_netdevice_notifiers - call all network notifier blocks
2013 * @val: value passed unmodified to notifier function
2014 * @dev: net_device pointer passed unmodified to notifier function
2016 * Call all network notifier blocks. Parameters and return value
2017 * are as for raw_notifier_call_chain().
2020 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2022 return call_netdevice_notifiers_extack(val, dev, NULL);
2024 EXPORT_SYMBOL(call_netdevice_notifiers);
2027 * call_netdevice_notifiers_mtu - call all network notifier blocks
2028 * @val: value passed unmodified to notifier function
2029 * @dev: net_device pointer passed unmodified to notifier function
2030 * @arg: additional u32 argument passed to the notifier function
2032 * Call all network notifier blocks. Parameters and return value
2033 * are as for raw_notifier_call_chain().
2035 static int call_netdevice_notifiers_mtu(unsigned long val,
2036 struct net_device *dev, u32 arg)
2038 struct netdev_notifier_info_ext info = {
2043 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2045 return call_netdevice_notifiers_info(val, &info.info);
2048 #ifdef CONFIG_NET_INGRESS
2049 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2051 void net_inc_ingress_queue(void)
2053 static_branch_inc(&ingress_needed_key);
2055 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2057 void net_dec_ingress_queue(void)
2059 static_branch_dec(&ingress_needed_key);
2061 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2064 #ifdef CONFIG_NET_EGRESS
2065 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2067 void net_inc_egress_queue(void)
2069 static_branch_inc(&egress_needed_key);
2071 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2073 void net_dec_egress_queue(void)
2075 static_branch_dec(&egress_needed_key);
2077 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2080 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2081 EXPORT_SYMBOL(netstamp_needed_key);
2082 #ifdef CONFIG_JUMP_LABEL
2083 static atomic_t netstamp_needed_deferred;
2084 static atomic_t netstamp_wanted;
2085 static void netstamp_clear(struct work_struct *work)
2087 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2090 wanted = atomic_add_return(deferred, &netstamp_wanted);
2092 static_branch_enable(&netstamp_needed_key);
2094 static_branch_disable(&netstamp_needed_key);
2096 static DECLARE_WORK(netstamp_work, netstamp_clear);
2099 void net_enable_timestamp(void)
2101 #ifdef CONFIG_JUMP_LABEL
2105 wanted = atomic_read(&netstamp_wanted);
2108 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2111 atomic_inc(&netstamp_needed_deferred);
2112 schedule_work(&netstamp_work);
2114 static_branch_inc(&netstamp_needed_key);
2117 EXPORT_SYMBOL(net_enable_timestamp);
2119 void net_disable_timestamp(void)
2121 #ifdef CONFIG_JUMP_LABEL
2125 wanted = atomic_read(&netstamp_wanted);
2128 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2131 atomic_dec(&netstamp_needed_deferred);
2132 schedule_work(&netstamp_work);
2134 static_branch_dec(&netstamp_needed_key);
2137 EXPORT_SYMBOL(net_disable_timestamp);
2139 static inline void net_timestamp_set(struct sk_buff *skb)
2142 skb->mono_delivery_time = 0;
2143 if (static_branch_unlikely(&netstamp_needed_key))
2144 skb->tstamp = ktime_get_real();
2147 #define net_timestamp_check(COND, SKB) \
2148 if (static_branch_unlikely(&netstamp_needed_key)) { \
2149 if ((COND) && !(SKB)->tstamp) \
2150 (SKB)->tstamp = ktime_get_real(); \
2153 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2155 return __is_skb_forwardable(dev, skb, true);
2157 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2159 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2162 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2165 skb->protocol = eth_type_trans(skb, dev);
2166 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2172 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2174 return __dev_forward_skb2(dev, skb, true);
2176 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2179 * dev_forward_skb - loopback an skb to another netif
2181 * @dev: destination network device
2182 * @skb: buffer to forward
2185 * NET_RX_SUCCESS (no congestion)
2186 * NET_RX_DROP (packet was dropped, but freed)
2188 * dev_forward_skb can be used for injecting an skb from the
2189 * start_xmit function of one device into the receive queue
2190 * of another device.
2192 * The receiving device may be in another namespace, so
2193 * we have to clear all information in the skb that could
2194 * impact namespace isolation.
2196 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2198 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2200 EXPORT_SYMBOL_GPL(dev_forward_skb);
2202 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2204 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2207 static inline int deliver_skb(struct sk_buff *skb,
2208 struct packet_type *pt_prev,
2209 struct net_device *orig_dev)
2211 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2213 refcount_inc(&skb->users);
2214 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2217 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2218 struct packet_type **pt,
2219 struct net_device *orig_dev,
2221 struct list_head *ptype_list)
2223 struct packet_type *ptype, *pt_prev = *pt;
2225 list_for_each_entry_rcu(ptype, ptype_list, list) {
2226 if (ptype->type != type)
2229 deliver_skb(skb, pt_prev, orig_dev);
2235 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2237 if (!ptype->af_packet_priv || !skb->sk)
2240 if (ptype->id_match)
2241 return ptype->id_match(ptype, skb->sk);
2242 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2249 * dev_nit_active - return true if any network interface taps are in use
2251 * @dev: network device to check for the presence of taps
2253 bool dev_nit_active(struct net_device *dev)
2255 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2257 EXPORT_SYMBOL_GPL(dev_nit_active);
2260 * Support routine. Sends outgoing frames to any network
2261 * taps currently in use.
2264 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2266 struct packet_type *ptype;
2267 struct sk_buff *skb2 = NULL;
2268 struct packet_type *pt_prev = NULL;
2269 struct list_head *ptype_list = &ptype_all;
2273 list_for_each_entry_rcu(ptype, ptype_list, list) {
2274 if (READ_ONCE(ptype->ignore_outgoing))
2277 /* Never send packets back to the socket
2278 * they originated from - MvS (miquels@drinkel.ow.org)
2280 if (skb_loop_sk(ptype, skb))
2284 deliver_skb(skb2, pt_prev, skb->dev);
2289 /* need to clone skb, done only once */
2290 skb2 = skb_clone(skb, GFP_ATOMIC);
2294 net_timestamp_set(skb2);
2296 /* skb->nh should be correctly
2297 * set by sender, so that the second statement is
2298 * just protection against buggy protocols.
2300 skb_reset_mac_header(skb2);
2302 if (skb_network_header(skb2) < skb2->data ||
2303 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2304 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2305 ntohs(skb2->protocol),
2307 skb_reset_network_header(skb2);
2310 skb2->transport_header = skb2->network_header;
2311 skb2->pkt_type = PACKET_OUTGOING;
2315 if (ptype_list == &ptype_all) {
2316 ptype_list = &dev->ptype_all;
2321 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2322 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2328 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2331 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2332 * @dev: Network device
2333 * @txq: number of queues available
2335 * If real_num_tx_queues is changed the tc mappings may no longer be
2336 * valid. To resolve this verify the tc mapping remains valid and if
2337 * not NULL the mapping. With no priorities mapping to this
2338 * offset/count pair it will no longer be used. In the worst case TC0
2339 * is invalid nothing can be done so disable priority mappings. If is
2340 * expected that drivers will fix this mapping if they can before
2341 * calling netif_set_real_num_tx_queues.
2343 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2346 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2348 /* If TC0 is invalidated disable TC mapping */
2349 if (tc->offset + tc->count > txq) {
2350 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2355 /* Invalidated prio to tc mappings set to TC0 */
2356 for (i = 1; i < TC_BITMASK + 1; i++) {
2357 int q = netdev_get_prio_tc_map(dev, i);
2359 tc = &dev->tc_to_txq[q];
2360 if (tc->offset + tc->count > txq) {
2361 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",
2363 netdev_set_prio_tc_map(dev, i, 0);
2368 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2371 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2374 /* walk through the TCs and see if it falls into any of them */
2375 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2376 if ((txq - tc->offset) < tc->count)
2380 /* didn't find it, just return -1 to indicate no match */
2386 EXPORT_SYMBOL(netdev_txq_to_tc);
2389 static struct static_key xps_needed __read_mostly;
2390 static struct static_key xps_rxqs_needed __read_mostly;
2391 static DEFINE_MUTEX(xps_map_mutex);
2392 #define xmap_dereference(P) \
2393 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2395 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2396 struct xps_dev_maps *old_maps, int tci, u16 index)
2398 struct xps_map *map = NULL;
2402 map = xmap_dereference(dev_maps->attr_map[tci]);
2406 for (pos = map->len; pos--;) {
2407 if (map->queues[pos] != index)
2411 map->queues[pos] = map->queues[--map->len];
2416 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2417 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2418 kfree_rcu(map, rcu);
2425 static bool remove_xps_queue_cpu(struct net_device *dev,
2426 struct xps_dev_maps *dev_maps,
2427 int cpu, u16 offset, u16 count)
2429 int num_tc = dev_maps->num_tc;
2430 bool active = false;
2433 for (tci = cpu * num_tc; num_tc--; tci++) {
2436 for (i = count, j = offset; i--; j++) {
2437 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2447 static void reset_xps_maps(struct net_device *dev,
2448 struct xps_dev_maps *dev_maps,
2449 enum xps_map_type type)
2451 static_key_slow_dec_cpuslocked(&xps_needed);
2452 if (type == XPS_RXQS)
2453 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2455 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2457 kfree_rcu(dev_maps, rcu);
2460 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2461 u16 offset, u16 count)
2463 struct xps_dev_maps *dev_maps;
2464 bool active = false;
2467 dev_maps = xmap_dereference(dev->xps_maps[type]);
2471 for (j = 0; j < dev_maps->nr_ids; j++)
2472 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2474 reset_xps_maps(dev, dev_maps, type);
2476 if (type == XPS_CPUS) {
2477 for (i = offset + (count - 1); count--; i--)
2478 netdev_queue_numa_node_write(
2479 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2483 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2486 if (!static_key_false(&xps_needed))
2490 mutex_lock(&xps_map_mutex);
2492 if (static_key_false(&xps_rxqs_needed))
2493 clean_xps_maps(dev, XPS_RXQS, offset, count);
2495 clean_xps_maps(dev, XPS_CPUS, offset, count);
2497 mutex_unlock(&xps_map_mutex);
2501 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2503 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2506 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2507 u16 index, bool is_rxqs_map)
2509 struct xps_map *new_map;
2510 int alloc_len = XPS_MIN_MAP_ALLOC;
2513 for (pos = 0; map && pos < map->len; pos++) {
2514 if (map->queues[pos] != index)
2519 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2521 if (pos < map->alloc_len)
2524 alloc_len = map->alloc_len * 2;
2527 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2531 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2533 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2534 cpu_to_node(attr_index));
2538 for (i = 0; i < pos; i++)
2539 new_map->queues[i] = map->queues[i];
2540 new_map->alloc_len = alloc_len;
2546 /* Copy xps maps at a given index */
2547 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2548 struct xps_dev_maps *new_dev_maps, int index,
2549 int tc, bool skip_tc)
2551 int i, tci = index * dev_maps->num_tc;
2552 struct xps_map *map;
2554 /* copy maps belonging to foreign traffic classes */
2555 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2556 if (i == tc && skip_tc)
2559 /* fill in the new device map from the old device map */
2560 map = xmap_dereference(dev_maps->attr_map[tci]);
2561 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2565 /* Must be called under cpus_read_lock */
2566 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2567 u16 index, enum xps_map_type type)
2569 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2570 const unsigned long *online_mask = NULL;
2571 bool active = false, copy = false;
2572 int i, j, tci, numa_node_id = -2;
2573 int maps_sz, num_tc = 1, tc = 0;
2574 struct xps_map *map, *new_map;
2575 unsigned int nr_ids;
2577 WARN_ON_ONCE(index >= dev->num_tx_queues);
2580 /* Do not allow XPS on subordinate device directly */
2581 num_tc = dev->num_tc;
2585 /* If queue belongs to subordinate dev use its map */
2586 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2588 tc = netdev_txq_to_tc(dev, index);
2593 mutex_lock(&xps_map_mutex);
2595 dev_maps = xmap_dereference(dev->xps_maps[type]);
2596 if (type == XPS_RXQS) {
2597 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2598 nr_ids = dev->num_rx_queues;
2600 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2601 if (num_possible_cpus() > 1)
2602 online_mask = cpumask_bits(cpu_online_mask);
2603 nr_ids = nr_cpu_ids;
2606 if (maps_sz < L1_CACHE_BYTES)
2607 maps_sz = L1_CACHE_BYTES;
2609 /* The old dev_maps could be larger or smaller than the one we're
2610 * setting up now, as dev->num_tc or nr_ids could have been updated in
2611 * between. We could try to be smart, but let's be safe instead and only
2612 * copy foreign traffic classes if the two map sizes match.
2615 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2618 /* allocate memory for queue storage */
2619 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2621 if (!new_dev_maps) {
2622 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2623 if (!new_dev_maps) {
2624 mutex_unlock(&xps_map_mutex);
2628 new_dev_maps->nr_ids = nr_ids;
2629 new_dev_maps->num_tc = num_tc;
2632 tci = j * num_tc + tc;
2633 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2635 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2639 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2643 goto out_no_new_maps;
2646 /* Increment static keys at most once per type */
2647 static_key_slow_inc_cpuslocked(&xps_needed);
2648 if (type == XPS_RXQS)
2649 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2652 for (j = 0; j < nr_ids; j++) {
2653 bool skip_tc = false;
2655 tci = j * num_tc + tc;
2656 if (netif_attr_test_mask(j, mask, nr_ids) &&
2657 netif_attr_test_online(j, online_mask, nr_ids)) {
2658 /* add tx-queue to CPU/rx-queue maps */
2663 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2664 while ((pos < map->len) && (map->queues[pos] != index))
2667 if (pos == map->len)
2668 map->queues[map->len++] = index;
2670 if (type == XPS_CPUS) {
2671 if (numa_node_id == -2)
2672 numa_node_id = cpu_to_node(j);
2673 else if (numa_node_id != cpu_to_node(j))
2680 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2684 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2686 /* Cleanup old maps */
2688 goto out_no_old_maps;
2690 for (j = 0; j < dev_maps->nr_ids; j++) {
2691 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2692 map = xmap_dereference(dev_maps->attr_map[tci]);
2697 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2702 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2703 kfree_rcu(map, rcu);
2707 old_dev_maps = dev_maps;
2710 dev_maps = new_dev_maps;
2714 if (type == XPS_CPUS)
2715 /* update Tx queue numa node */
2716 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2717 (numa_node_id >= 0) ?
2718 numa_node_id : NUMA_NO_NODE);
2723 /* removes tx-queue from unused CPUs/rx-queues */
2724 for (j = 0; j < dev_maps->nr_ids; j++) {
2725 tci = j * dev_maps->num_tc;
2727 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2729 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2730 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2733 active |= remove_xps_queue(dev_maps,
2734 copy ? old_dev_maps : NULL,
2740 kfree_rcu(old_dev_maps, rcu);
2742 /* free map if not active */
2744 reset_xps_maps(dev, dev_maps, type);
2747 mutex_unlock(&xps_map_mutex);
2751 /* remove any maps that we added */
2752 for (j = 0; j < nr_ids; j++) {
2753 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2754 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2756 xmap_dereference(dev_maps->attr_map[tci]) :
2758 if (new_map && new_map != map)
2763 mutex_unlock(&xps_map_mutex);
2765 kfree(new_dev_maps);
2768 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2770 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2776 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2781 EXPORT_SYMBOL(netif_set_xps_queue);
2784 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2786 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2788 /* Unbind any subordinate channels */
2789 while (txq-- != &dev->_tx[0]) {
2791 netdev_unbind_sb_channel(dev, txq->sb_dev);
2795 void netdev_reset_tc(struct net_device *dev)
2798 netif_reset_xps_queues_gt(dev, 0);
2800 netdev_unbind_all_sb_channels(dev);
2802 /* Reset TC configuration of device */
2804 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2805 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2807 EXPORT_SYMBOL(netdev_reset_tc);
2809 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2811 if (tc >= dev->num_tc)
2815 netif_reset_xps_queues(dev, offset, count);
2817 dev->tc_to_txq[tc].count = count;
2818 dev->tc_to_txq[tc].offset = offset;
2821 EXPORT_SYMBOL(netdev_set_tc_queue);
2823 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2825 if (num_tc > TC_MAX_QUEUE)
2829 netif_reset_xps_queues_gt(dev, 0);
2831 netdev_unbind_all_sb_channels(dev);
2833 dev->num_tc = num_tc;
2836 EXPORT_SYMBOL(netdev_set_num_tc);
2838 void netdev_unbind_sb_channel(struct net_device *dev,
2839 struct net_device *sb_dev)
2841 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2844 netif_reset_xps_queues_gt(sb_dev, 0);
2846 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2847 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2849 while (txq-- != &dev->_tx[0]) {
2850 if (txq->sb_dev == sb_dev)
2854 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2856 int netdev_bind_sb_channel_queue(struct net_device *dev,
2857 struct net_device *sb_dev,
2858 u8 tc, u16 count, u16 offset)
2860 /* Make certain the sb_dev and dev are already configured */
2861 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2864 /* We cannot hand out queues we don't have */
2865 if ((offset + count) > dev->real_num_tx_queues)
2868 /* Record the mapping */
2869 sb_dev->tc_to_txq[tc].count = count;
2870 sb_dev->tc_to_txq[tc].offset = offset;
2872 /* Provide a way for Tx queue to find the tc_to_txq map or
2873 * XPS map for itself.
2876 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2880 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2882 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2884 /* Do not use a multiqueue device to represent a subordinate channel */
2885 if (netif_is_multiqueue(dev))
2888 /* We allow channels 1 - 32767 to be used for subordinate channels.
2889 * Channel 0 is meant to be "native" mode and used only to represent
2890 * the main root device. We allow writing 0 to reset the device back
2891 * to normal mode after being used as a subordinate channel.
2893 if (channel > S16_MAX)
2896 dev->num_tc = -channel;
2900 EXPORT_SYMBOL(netdev_set_sb_channel);
2903 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2904 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2906 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2911 disabling = txq < dev->real_num_tx_queues;
2913 if (txq < 1 || txq > dev->num_tx_queues)
2916 if (dev->reg_state == NETREG_REGISTERED ||
2917 dev->reg_state == NETREG_UNREGISTERING) {
2920 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2926 netif_setup_tc(dev, txq);
2928 dev_qdisc_change_real_num_tx(dev, txq);
2930 dev->real_num_tx_queues = txq;
2934 qdisc_reset_all_tx_gt(dev, txq);
2936 netif_reset_xps_queues_gt(dev, txq);
2940 dev->real_num_tx_queues = txq;
2945 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2949 * netif_set_real_num_rx_queues - set actual number of RX queues used
2950 * @dev: Network device
2951 * @rxq: Actual number of RX queues
2953 * This must be called either with the rtnl_lock held or before
2954 * registration of the net device. Returns 0 on success, or a
2955 * negative error code. If called before registration, it always
2958 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2962 if (rxq < 1 || rxq > dev->num_rx_queues)
2965 if (dev->reg_state == NETREG_REGISTERED) {
2968 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2974 dev->real_num_rx_queues = rxq;
2977 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2981 * netif_set_real_num_queues - set actual number of RX and TX queues used
2982 * @dev: Network device
2983 * @txq: Actual number of TX queues
2984 * @rxq: Actual number of RX queues
2986 * Set the real number of both TX and RX queues.
2987 * Does nothing if the number of queues is already correct.
2989 int netif_set_real_num_queues(struct net_device *dev,
2990 unsigned int txq, unsigned int rxq)
2992 unsigned int old_rxq = dev->real_num_rx_queues;
2995 if (txq < 1 || txq > dev->num_tx_queues ||
2996 rxq < 1 || rxq > dev->num_rx_queues)
2999 /* Start from increases, so the error path only does decreases -
3000 * decreases can't fail.
3002 if (rxq > dev->real_num_rx_queues) {
3003 err = netif_set_real_num_rx_queues(dev, rxq);
3007 if (txq > dev->real_num_tx_queues) {
3008 err = netif_set_real_num_tx_queues(dev, txq);
3012 if (rxq < dev->real_num_rx_queues)
3013 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3014 if (txq < dev->real_num_tx_queues)
3015 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3019 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3022 EXPORT_SYMBOL(netif_set_real_num_queues);
3025 * netif_set_tso_max_size() - set the max size of TSO frames supported
3026 * @dev: netdev to update
3027 * @size: max skb->len of a TSO frame
3029 * Set the limit on the size of TSO super-frames the device can handle.
3030 * Unless explicitly set the stack will assume the value of
3031 * %GSO_LEGACY_MAX_SIZE.
3033 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3035 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3036 if (size < READ_ONCE(dev->gso_max_size))
3037 netif_set_gso_max_size(dev, size);
3039 EXPORT_SYMBOL(netif_set_tso_max_size);
3042 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3043 * @dev: netdev to update
3044 * @segs: max number of TCP segments
3046 * Set the limit on the number of TCP segments the device can generate from
3047 * a single TSO super-frame.
3048 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3050 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3052 dev->tso_max_segs = segs;
3053 if (segs < READ_ONCE(dev->gso_max_segs))
3054 netif_set_gso_max_segs(dev, segs);
3056 EXPORT_SYMBOL(netif_set_tso_max_segs);
3059 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3060 * @to: netdev to update
3061 * @from: netdev from which to copy the limits
3063 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3065 netif_set_tso_max_size(to, from->tso_max_size);
3066 netif_set_tso_max_segs(to, from->tso_max_segs);
3068 EXPORT_SYMBOL(netif_inherit_tso_max);
3071 * netif_get_num_default_rss_queues - default number of RSS queues
3073 * Default value is the number of physical cores if there are only 1 or 2, or
3074 * divided by 2 if there are more.
3076 int netif_get_num_default_rss_queues(void)
3081 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3084 cpumask_copy(cpus, cpu_online_mask);
3085 for_each_cpu(cpu, cpus) {
3087 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3089 free_cpumask_var(cpus);
3091 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3093 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3095 static void __netif_reschedule(struct Qdisc *q)
3097 struct softnet_data *sd;
3098 unsigned long flags;
3100 local_irq_save(flags);
3101 sd = this_cpu_ptr(&softnet_data);
3102 q->next_sched = NULL;
3103 *sd->output_queue_tailp = q;
3104 sd->output_queue_tailp = &q->next_sched;
3105 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3106 local_irq_restore(flags);
3109 void __netif_schedule(struct Qdisc *q)
3111 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3112 __netif_reschedule(q);
3114 EXPORT_SYMBOL(__netif_schedule);
3116 struct dev_kfree_skb_cb {
3117 enum skb_free_reason reason;
3120 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3122 return (struct dev_kfree_skb_cb *)skb->cb;
3125 void netif_schedule_queue(struct netdev_queue *txq)
3128 if (!netif_xmit_stopped(txq)) {
3129 struct Qdisc *q = rcu_dereference(txq->qdisc);
3131 __netif_schedule(q);
3135 EXPORT_SYMBOL(netif_schedule_queue);
3137 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3139 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3143 q = rcu_dereference(dev_queue->qdisc);
3144 __netif_schedule(q);
3148 EXPORT_SYMBOL(netif_tx_wake_queue);
3150 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3152 unsigned long flags;
3157 if (likely(refcount_read(&skb->users) == 1)) {
3159 refcount_set(&skb->users, 0);
3160 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3163 get_kfree_skb_cb(skb)->reason = reason;
3164 local_irq_save(flags);
3165 skb->next = __this_cpu_read(softnet_data.completion_queue);
3166 __this_cpu_write(softnet_data.completion_queue, skb);
3167 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3168 local_irq_restore(flags);
3170 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3172 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3174 if (in_hardirq() || irqs_disabled())
3175 __dev_kfree_skb_irq(skb, reason);
3176 else if (unlikely(reason == SKB_REASON_DROPPED))
3181 EXPORT_SYMBOL(__dev_kfree_skb_any);
3185 * netif_device_detach - mark device as removed
3186 * @dev: network device
3188 * Mark device as removed from system and therefore no longer available.
3190 void netif_device_detach(struct net_device *dev)
3192 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3193 netif_running(dev)) {
3194 netif_tx_stop_all_queues(dev);
3197 EXPORT_SYMBOL(netif_device_detach);
3200 * netif_device_attach - mark device as attached
3201 * @dev: network device
3203 * Mark device as attached from system and restart if needed.
3205 void netif_device_attach(struct net_device *dev)
3207 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3208 netif_running(dev)) {
3209 netif_tx_wake_all_queues(dev);
3210 __netdev_watchdog_up(dev);
3213 EXPORT_SYMBOL(netif_device_attach);
3216 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3217 * to be used as a distribution range.
3219 static u16 skb_tx_hash(const struct net_device *dev,
3220 const struct net_device *sb_dev,
3221 struct sk_buff *skb)
3225 u16 qcount = dev->real_num_tx_queues;
3228 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3230 qoffset = sb_dev->tc_to_txq[tc].offset;
3231 qcount = sb_dev->tc_to_txq[tc].count;
3232 if (unlikely(!qcount)) {
3233 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3234 sb_dev->name, qoffset, tc);
3236 qcount = dev->real_num_tx_queues;
3240 if (skb_rx_queue_recorded(skb)) {
3241 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3242 hash = skb_get_rx_queue(skb);
3243 if (hash >= qoffset)
3245 while (unlikely(hash >= qcount))
3247 return hash + qoffset;
3250 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3253 static void skb_warn_bad_offload(const struct sk_buff *skb)
3255 static const netdev_features_t null_features;
3256 struct net_device *dev = skb->dev;
3257 const char *name = "";
3259 if (!net_ratelimit())
3263 if (dev->dev.parent)
3264 name = dev_driver_string(dev->dev.parent);
3266 name = netdev_name(dev);
3268 skb_dump(KERN_WARNING, skb, false);
3269 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3270 name, dev ? &dev->features : &null_features,
3271 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3275 * Invalidate hardware checksum when packet is to be mangled, and
3276 * complete checksum manually on outgoing path.
3278 int skb_checksum_help(struct sk_buff *skb)
3281 int ret = 0, offset;
3283 if (skb->ip_summed == CHECKSUM_COMPLETE)
3284 goto out_set_summed;
3286 if (unlikely(skb_is_gso(skb))) {
3287 skb_warn_bad_offload(skb);
3291 /* Before computing a checksum, we should make sure no frag could
3292 * be modified by an external entity : checksum could be wrong.
3294 if (skb_has_shared_frag(skb)) {
3295 ret = __skb_linearize(skb);
3300 offset = skb_checksum_start_offset(skb);
3302 if (unlikely(offset >= skb_headlen(skb))) {
3303 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3304 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3305 offset, skb_headlen(skb));
3308 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3310 offset += skb->csum_offset;
3311 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3312 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3313 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3314 offset + sizeof(__sum16), skb_headlen(skb));
3317 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3321 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3323 skb->ip_summed = CHECKSUM_NONE;
3327 EXPORT_SYMBOL(skb_checksum_help);
3329 int skb_crc32c_csum_help(struct sk_buff *skb)
3332 int ret = 0, offset, start;
3334 if (skb->ip_summed != CHECKSUM_PARTIAL)
3337 if (unlikely(skb_is_gso(skb)))
3340 /* Before computing a checksum, we should make sure no frag could
3341 * be modified by an external entity : checksum could be wrong.
3343 if (unlikely(skb_has_shared_frag(skb))) {
3344 ret = __skb_linearize(skb);
3348 start = skb_checksum_start_offset(skb);
3349 offset = start + offsetof(struct sctphdr, checksum);
3350 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3355 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3359 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3360 skb->len - start, ~(__u32)0,
3362 *(__le32 *)(skb->data + offset) = crc32c_csum;
3363 skb->ip_summed = CHECKSUM_NONE;
3364 skb->csum_not_inet = 0;
3369 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3371 __be16 type = skb->protocol;
3373 /* Tunnel gso handlers can set protocol to ethernet. */
3374 if (type == htons(ETH_P_TEB)) {
3377 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3380 eth = (struct ethhdr *)skb->data;
3381 type = eth->h_proto;
3384 return vlan_get_protocol_and_depth(skb, type, depth);
3387 /* openvswitch calls this on rx path, so we need a different check.
3389 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3392 return skb->ip_summed != CHECKSUM_PARTIAL &&
3393 skb->ip_summed != CHECKSUM_UNNECESSARY;
3395 return skb->ip_summed == CHECKSUM_NONE;
3399 * __skb_gso_segment - Perform segmentation on skb.
3400 * @skb: buffer to segment
3401 * @features: features for the output path (see dev->features)
3402 * @tx_path: whether it is called in TX path
3404 * This function segments the given skb and returns a list of segments.
3406 * It may return NULL if the skb requires no segmentation. This is
3407 * only possible when GSO is used for verifying header integrity.
3409 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3411 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3412 netdev_features_t features, bool tx_path)
3414 struct sk_buff *segs;
3416 if (unlikely(skb_needs_check(skb, tx_path))) {
3419 /* We're going to init ->check field in TCP or UDP header */
3420 err = skb_cow_head(skb, 0);
3422 return ERR_PTR(err);
3425 /* Only report GSO partial support if it will enable us to
3426 * support segmentation on this frame without needing additional
3429 if (features & NETIF_F_GSO_PARTIAL) {
3430 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3431 struct net_device *dev = skb->dev;
3433 partial_features |= dev->features & dev->gso_partial_features;
3434 if (!skb_gso_ok(skb, features | partial_features))
3435 features &= ~NETIF_F_GSO_PARTIAL;
3438 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3439 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3441 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3442 SKB_GSO_CB(skb)->encap_level = 0;
3444 skb_reset_mac_header(skb);
3445 skb_reset_mac_len(skb);
3447 segs = skb_mac_gso_segment(skb, features);
3449 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3450 skb_warn_bad_offload(skb);
3454 EXPORT_SYMBOL(__skb_gso_segment);
3456 /* Take action when hardware reception checksum errors are detected. */
3458 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3460 netdev_err(dev, "hw csum failure\n");
3461 skb_dump(KERN_ERR, skb, true);
3465 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3467 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3469 EXPORT_SYMBOL(netdev_rx_csum_fault);
3472 /* XXX: check that highmem exists at all on the given machine. */
3473 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3475 #ifdef CONFIG_HIGHMEM
3478 if (!(dev->features & NETIF_F_HIGHDMA)) {
3479 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3480 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3482 if (PageHighMem(skb_frag_page(frag)))
3490 /* If MPLS offload request, verify we are testing hardware MPLS features
3491 * instead of standard features for the netdev.
3493 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3494 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3495 netdev_features_t features,
3498 if (eth_p_mpls(type))
3499 features &= skb->dev->mpls_features;
3504 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3505 netdev_features_t features,
3512 static netdev_features_t harmonize_features(struct sk_buff *skb,
3513 netdev_features_t features)
3517 type = skb_network_protocol(skb, NULL);
3518 features = net_mpls_features(skb, features, type);
3520 if (skb->ip_summed != CHECKSUM_NONE &&
3521 !can_checksum_protocol(features, type)) {
3522 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3524 if (illegal_highdma(skb->dev, skb))
3525 features &= ~NETIF_F_SG;
3530 netdev_features_t passthru_features_check(struct sk_buff *skb,
3531 struct net_device *dev,
3532 netdev_features_t features)
3536 EXPORT_SYMBOL(passthru_features_check);
3538 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3539 struct net_device *dev,
3540 netdev_features_t features)
3542 return vlan_features_check(skb, features);
3545 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3546 struct net_device *dev,
3547 netdev_features_t features)
3549 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3551 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3552 return features & ~NETIF_F_GSO_MASK;
3554 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size)))
3555 return features & ~NETIF_F_GSO_MASK;
3557 if (!skb_shinfo(skb)->gso_type) {
3558 skb_warn_bad_offload(skb);
3559 return features & ~NETIF_F_GSO_MASK;
3562 /* Support for GSO partial features requires software
3563 * intervention before we can actually process the packets
3564 * so we need to strip support for any partial features now
3565 * and we can pull them back in after we have partially
3566 * segmented the frame.
3568 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3569 features &= ~dev->gso_partial_features;
3571 /* Make sure to clear the IPv4 ID mangling feature if the
3572 * IPv4 header has the potential to be fragmented.
3574 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3575 struct iphdr *iph = skb->encapsulation ?
3576 inner_ip_hdr(skb) : ip_hdr(skb);
3578 if (!(iph->frag_off & htons(IP_DF)))
3579 features &= ~NETIF_F_TSO_MANGLEID;
3585 netdev_features_t netif_skb_features(struct sk_buff *skb)
3587 struct net_device *dev = skb->dev;
3588 netdev_features_t features = dev->features;
3590 if (skb_is_gso(skb))
3591 features = gso_features_check(skb, dev, features);
3593 /* If encapsulation offload request, verify we are testing
3594 * hardware encapsulation features instead of standard
3595 * features for the netdev
3597 if (skb->encapsulation)
3598 features &= dev->hw_enc_features;
3600 if (skb_vlan_tagged(skb))
3601 features = netdev_intersect_features(features,
3602 dev->vlan_features |
3603 NETIF_F_HW_VLAN_CTAG_TX |
3604 NETIF_F_HW_VLAN_STAG_TX);
3606 if (dev->netdev_ops->ndo_features_check)
3607 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3610 features &= dflt_features_check(skb, dev, features);
3612 return harmonize_features(skb, features);
3614 EXPORT_SYMBOL(netif_skb_features);
3616 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3617 struct netdev_queue *txq, bool more)
3622 if (dev_nit_active(dev))
3623 dev_queue_xmit_nit(skb, dev);
3626 trace_net_dev_start_xmit(skb, dev);
3627 rc = netdev_start_xmit(skb, dev, txq, more);
3628 trace_net_dev_xmit(skb, rc, dev, len);
3633 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3634 struct netdev_queue *txq, int *ret)
3636 struct sk_buff *skb = first;
3637 int rc = NETDEV_TX_OK;
3640 struct sk_buff *next = skb->next;
3642 skb_mark_not_on_list(skb);
3643 rc = xmit_one(skb, dev, txq, next != NULL);
3644 if (unlikely(!dev_xmit_complete(rc))) {
3650 if (netif_tx_queue_stopped(txq) && skb) {
3651 rc = NETDEV_TX_BUSY;
3661 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3662 netdev_features_t features)
3664 if (skb_vlan_tag_present(skb) &&
3665 !vlan_hw_offload_capable(features, skb->vlan_proto))
3666 skb = __vlan_hwaccel_push_inside(skb);
3670 int skb_csum_hwoffload_help(struct sk_buff *skb,
3671 const netdev_features_t features)
3673 if (unlikely(skb_csum_is_sctp(skb)))
3674 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3675 skb_crc32c_csum_help(skb);
3677 if (features & NETIF_F_HW_CSUM)
3680 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3681 switch (skb->csum_offset) {
3682 case offsetof(struct tcphdr, check):
3683 case offsetof(struct udphdr, check):
3688 return skb_checksum_help(skb);
3690 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3692 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3694 netdev_features_t features;
3696 features = netif_skb_features(skb);
3697 skb = validate_xmit_vlan(skb, features);
3701 skb = sk_validate_xmit_skb(skb, dev);
3705 if (netif_needs_gso(skb, features)) {
3706 struct sk_buff *segs;
3708 segs = skb_gso_segment(skb, features);
3716 if (skb_needs_linearize(skb, features) &&
3717 __skb_linearize(skb))
3720 /* If packet is not checksummed and device does not
3721 * support checksumming for this protocol, complete
3722 * checksumming here.
3724 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3725 if (skb->encapsulation)
3726 skb_set_inner_transport_header(skb,
3727 skb_checksum_start_offset(skb));
3729 skb_set_transport_header(skb,
3730 skb_checksum_start_offset(skb));
3731 if (skb_csum_hwoffload_help(skb, features))
3736 skb = validate_xmit_xfrm(skb, features, again);
3743 dev_core_stats_tx_dropped_inc(dev);
3747 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3749 struct sk_buff *next, *head = NULL, *tail;
3751 for (; skb != NULL; skb = next) {
3753 skb_mark_not_on_list(skb);
3755 /* in case skb wont be segmented, point to itself */
3758 skb = validate_xmit_skb(skb, dev, again);
3766 /* If skb was segmented, skb->prev points to
3767 * the last segment. If not, it still contains skb.
3773 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3775 static void qdisc_pkt_len_init(struct sk_buff *skb)
3777 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3779 qdisc_skb_cb(skb)->pkt_len = skb->len;
3781 /* To get more precise estimation of bytes sent on wire,
3782 * we add to pkt_len the headers size of all segments
3784 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3785 unsigned int hdr_len;
3786 u16 gso_segs = shinfo->gso_segs;
3788 /* mac layer + network layer */
3789 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3791 /* + transport layer */
3792 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3793 const struct tcphdr *th;
3794 struct tcphdr _tcphdr;
3796 th = skb_header_pointer(skb, skb_transport_offset(skb),
3797 sizeof(_tcphdr), &_tcphdr);
3799 hdr_len += __tcp_hdrlen(th);
3801 struct udphdr _udphdr;
3803 if (skb_header_pointer(skb, skb_transport_offset(skb),
3804 sizeof(_udphdr), &_udphdr))
3805 hdr_len += sizeof(struct udphdr);
3808 if (shinfo->gso_type & SKB_GSO_DODGY)
3809 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3812 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3816 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3817 struct sk_buff **to_free,
3818 struct netdev_queue *txq)
3822 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3823 if (rc == NET_XMIT_SUCCESS)
3824 trace_qdisc_enqueue(q, txq, skb);
3828 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3829 struct net_device *dev,
3830 struct netdev_queue *txq)
3832 spinlock_t *root_lock = qdisc_lock(q);
3833 struct sk_buff *to_free = NULL;
3837 qdisc_calculate_pkt_len(skb, q);
3839 if (q->flags & TCQ_F_NOLOCK) {
3840 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3841 qdisc_run_begin(q)) {
3842 /* Retest nolock_qdisc_is_empty() within the protection
3843 * of q->seqlock to protect from racing with requeuing.
3845 if (unlikely(!nolock_qdisc_is_empty(q))) {
3846 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3853 qdisc_bstats_cpu_update(q, skb);
3854 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3855 !nolock_qdisc_is_empty(q))
3859 return NET_XMIT_SUCCESS;
3862 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3866 if (unlikely(to_free))
3867 kfree_skb_list_reason(to_free,
3868 SKB_DROP_REASON_QDISC_DROP);
3873 * Heuristic to force contended enqueues to serialize on a
3874 * separate lock before trying to get qdisc main lock.
3875 * This permits qdisc->running owner to get the lock more
3876 * often and dequeue packets faster.
3877 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3878 * and then other tasks will only enqueue packets. The packets will be
3879 * sent after the qdisc owner is scheduled again. To prevent this
3880 * scenario the task always serialize on the lock.
3882 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3883 if (unlikely(contended))
3884 spin_lock(&q->busylock);
3886 spin_lock(root_lock);
3887 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3888 __qdisc_drop(skb, &to_free);
3890 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3891 qdisc_run_begin(q)) {
3893 * This is a work-conserving queue; there are no old skbs
3894 * waiting to be sent out; and the qdisc is not running -
3895 * xmit the skb directly.
3898 qdisc_bstats_update(q, skb);
3900 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3901 if (unlikely(contended)) {
3902 spin_unlock(&q->busylock);
3909 rc = NET_XMIT_SUCCESS;
3911 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3912 if (qdisc_run_begin(q)) {
3913 if (unlikely(contended)) {
3914 spin_unlock(&q->busylock);
3921 spin_unlock(root_lock);
3922 if (unlikely(to_free))
3923 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3924 if (unlikely(contended))
3925 spin_unlock(&q->busylock);
3929 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3930 static void skb_update_prio(struct sk_buff *skb)
3932 const struct netprio_map *map;
3933 const struct sock *sk;
3934 unsigned int prioidx;
3938 map = rcu_dereference_bh(skb->dev->priomap);
3941 sk = skb_to_full_sk(skb);
3945 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3947 if (prioidx < map->priomap_len)
3948 skb->priority = map->priomap[prioidx];
3951 #define skb_update_prio(skb)
3955 * dev_loopback_xmit - loop back @skb
3956 * @net: network namespace this loopback is happening in
3957 * @sk: sk needed to be a netfilter okfn
3958 * @skb: buffer to transmit
3960 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3962 skb_reset_mac_header(skb);
3963 __skb_pull(skb, skb_network_offset(skb));
3964 skb->pkt_type = PACKET_LOOPBACK;
3965 if (skb->ip_summed == CHECKSUM_NONE)
3966 skb->ip_summed = CHECKSUM_UNNECESSARY;
3967 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3972 EXPORT_SYMBOL(dev_loopback_xmit);
3974 #ifdef CONFIG_NET_EGRESS
3975 static struct sk_buff *
3976 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3978 #ifdef CONFIG_NET_CLS_ACT
3979 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3980 struct tcf_result cl_res;
3985 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3986 tc_skb_cb(skb)->mru = 0;
3987 tc_skb_cb(skb)->post_ct = false;
3988 mini_qdisc_bstats_cpu_update(miniq, skb);
3990 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3992 case TC_ACT_RECLASSIFY:
3993 skb->tc_index = TC_H_MIN(cl_res.classid);
3996 mini_qdisc_qstats_cpu_drop(miniq);
3997 *ret = NET_XMIT_DROP;
3998 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
4003 *ret = NET_XMIT_SUCCESS;
4006 case TC_ACT_REDIRECT:
4007 /* No need to push/pop skb's mac_header here on egress! */
4008 skb_do_redirect(skb);
4009 *ret = NET_XMIT_SUCCESS;
4014 #endif /* CONFIG_NET_CLS_ACT */
4019 static struct netdev_queue *
4020 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
4022 int qm = skb_get_queue_mapping(skb);
4024 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
4027 static bool netdev_xmit_txqueue_skipped(void)
4029 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
4032 void netdev_xmit_skip_txqueue(bool skip)
4034 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
4036 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
4037 #endif /* CONFIG_NET_EGRESS */
4040 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4041 struct xps_dev_maps *dev_maps, unsigned int tci)
4043 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4044 struct xps_map *map;
4045 int queue_index = -1;
4047 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4050 tci *= dev_maps->num_tc;
4053 map = rcu_dereference(dev_maps->attr_map[tci]);
4056 queue_index = map->queues[0];
4058 queue_index = map->queues[reciprocal_scale(
4059 skb_get_hash(skb), map->len)];
4060 if (unlikely(queue_index >= dev->real_num_tx_queues))
4067 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4068 struct sk_buff *skb)
4071 struct xps_dev_maps *dev_maps;
4072 struct sock *sk = skb->sk;
4073 int queue_index = -1;
4075 if (!static_key_false(&xps_needed))
4079 if (!static_key_false(&xps_rxqs_needed))
4082 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4084 int tci = sk_rx_queue_get(sk);
4087 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4092 if (queue_index < 0) {
4093 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4095 unsigned int tci = skb->sender_cpu - 1;
4097 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4109 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4110 struct net_device *sb_dev)
4114 EXPORT_SYMBOL(dev_pick_tx_zero);
4116 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4117 struct net_device *sb_dev)
4119 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4121 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4123 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4124 struct net_device *sb_dev)
4126 struct sock *sk = skb->sk;
4127 int queue_index = sk_tx_queue_get(sk);
4129 sb_dev = sb_dev ? : dev;
4131 if (queue_index < 0 || skb->ooo_okay ||
4132 queue_index >= dev->real_num_tx_queues) {
4133 int new_index = get_xps_queue(dev, sb_dev, skb);
4136 new_index = skb_tx_hash(dev, sb_dev, skb);
4138 if (queue_index != new_index && sk &&
4140 rcu_access_pointer(sk->sk_dst_cache))
4141 sk_tx_queue_set(sk, new_index);
4143 queue_index = new_index;
4148 EXPORT_SYMBOL(netdev_pick_tx);
4150 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4151 struct sk_buff *skb,
4152 struct net_device *sb_dev)
4154 int queue_index = 0;
4157 u32 sender_cpu = skb->sender_cpu - 1;
4159 if (sender_cpu >= (u32)NR_CPUS)
4160 skb->sender_cpu = raw_smp_processor_id() + 1;
4163 if (dev->real_num_tx_queues != 1) {
4164 const struct net_device_ops *ops = dev->netdev_ops;
4166 if (ops->ndo_select_queue)
4167 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4169 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4171 queue_index = netdev_cap_txqueue(dev, queue_index);
4174 skb_set_queue_mapping(skb, queue_index);
4175 return netdev_get_tx_queue(dev, queue_index);
4179 * __dev_queue_xmit() - transmit a buffer
4180 * @skb: buffer to transmit
4181 * @sb_dev: suboordinate device used for L2 forwarding offload
4183 * Queue a buffer for transmission to a network device. The caller must
4184 * have set the device and priority and built the buffer before calling
4185 * this function. The function can be called from an interrupt.
4187 * When calling this method, interrupts MUST be enabled. This is because
4188 * the BH enable code must have IRQs enabled so that it will not deadlock.
4190 * Regardless of the return value, the skb is consumed, so it is currently
4191 * difficult to retry a send to this method. (You can bump the ref count
4192 * before sending to hold a reference for retry if you are careful.)
4195 * * 0 - buffer successfully transmitted
4196 * * positive qdisc return code - NET_XMIT_DROP etc.
4197 * * negative errno - other errors
4199 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4201 struct net_device *dev = skb->dev;
4202 struct netdev_queue *txq = NULL;
4207 skb_reset_mac_header(skb);
4208 skb_assert_len(skb);
4210 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4211 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4213 /* Disable soft irqs for various locks below. Also
4214 * stops preemption for RCU.
4218 skb_update_prio(skb);
4220 qdisc_pkt_len_init(skb);
4221 #ifdef CONFIG_NET_CLS_ACT
4222 skb->tc_at_ingress = 0;
4224 #ifdef CONFIG_NET_EGRESS
4225 if (static_branch_unlikely(&egress_needed_key)) {
4226 if (nf_hook_egress_active()) {
4227 skb = nf_hook_egress(skb, &rc, dev);
4232 netdev_xmit_skip_txqueue(false);
4234 nf_skip_egress(skb, true);
4235 skb = sch_handle_egress(skb, &rc, dev);
4238 nf_skip_egress(skb, false);
4240 if (netdev_xmit_txqueue_skipped())
4241 txq = netdev_tx_queue_mapping(dev, skb);
4244 /* If device/qdisc don't need skb->dst, release it right now while
4245 * its hot in this cpu cache.
4247 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4253 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4255 q = rcu_dereference_bh(txq->qdisc);
4257 trace_net_dev_queue(skb);
4259 rc = __dev_xmit_skb(skb, q, dev, txq);
4263 /* The device has no queue. Common case for software devices:
4264 * loopback, all the sorts of tunnels...
4266 * Really, it is unlikely that netif_tx_lock protection is necessary
4267 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4269 * However, it is possible, that they rely on protection
4272 * Check this and shot the lock. It is not prone from deadlocks.
4273 *Either shot noqueue qdisc, it is even simpler 8)
4275 if (dev->flags & IFF_UP) {
4276 int cpu = smp_processor_id(); /* ok because BHs are off */
4278 /* Other cpus might concurrently change txq->xmit_lock_owner
4279 * to -1 or to their cpu id, but not to our id.
4281 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4282 if (dev_xmit_recursion())
4283 goto recursion_alert;
4285 skb = validate_xmit_skb(skb, dev, &again);
4289 HARD_TX_LOCK(dev, txq, cpu);
4291 if (!netif_xmit_stopped(txq)) {
4292 dev_xmit_recursion_inc();
4293 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4294 dev_xmit_recursion_dec();
4295 if (dev_xmit_complete(rc)) {
4296 HARD_TX_UNLOCK(dev, txq);
4300 HARD_TX_UNLOCK(dev, txq);
4301 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4304 /* Recursion is detected! It is possible,
4308 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4314 rcu_read_unlock_bh();
4316 dev_core_stats_tx_dropped_inc(dev);
4317 kfree_skb_list(skb);
4320 rcu_read_unlock_bh();
4323 EXPORT_SYMBOL(__dev_queue_xmit);
4325 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4327 struct net_device *dev = skb->dev;
4328 struct sk_buff *orig_skb = skb;
4329 struct netdev_queue *txq;
4330 int ret = NETDEV_TX_BUSY;
4333 if (unlikely(!netif_running(dev) ||
4334 !netif_carrier_ok(dev)))
4337 skb = validate_xmit_skb_list(skb, dev, &again);
4338 if (skb != orig_skb)
4341 skb_set_queue_mapping(skb, queue_id);
4342 txq = skb_get_tx_queue(dev, skb);
4346 dev_xmit_recursion_inc();
4347 HARD_TX_LOCK(dev, txq, smp_processor_id());
4348 if (!netif_xmit_frozen_or_drv_stopped(txq))
4349 ret = netdev_start_xmit(skb, dev, txq, false);
4350 HARD_TX_UNLOCK(dev, txq);
4351 dev_xmit_recursion_dec();
4356 dev_core_stats_tx_dropped_inc(dev);
4357 kfree_skb_list(skb);
4358 return NET_XMIT_DROP;
4360 EXPORT_SYMBOL(__dev_direct_xmit);
4362 /*************************************************************************
4364 *************************************************************************/
4366 int netdev_max_backlog __read_mostly = 1000;
4367 EXPORT_SYMBOL(netdev_max_backlog);
4369 int netdev_tstamp_prequeue __read_mostly = 1;
4370 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4371 int netdev_budget __read_mostly = 300;
4372 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4373 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4374 int weight_p __read_mostly = 64; /* old backlog weight */
4375 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4376 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4377 int dev_rx_weight __read_mostly = 64;
4378 int dev_tx_weight __read_mostly = 64;
4380 /* Called with irq disabled */
4381 static inline void ____napi_schedule(struct softnet_data *sd,
4382 struct napi_struct *napi)
4384 struct task_struct *thread;
4386 lockdep_assert_irqs_disabled();
4388 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4389 /* Paired with smp_mb__before_atomic() in
4390 * napi_enable()/dev_set_threaded().
4391 * Use READ_ONCE() to guarantee a complete
4392 * read on napi->thread. Only call
4393 * wake_up_process() when it's not NULL.
4395 thread = READ_ONCE(napi->thread);
4397 /* Avoid doing set_bit() if the thread is in
4398 * INTERRUPTIBLE state, cause napi_thread_wait()
4399 * makes sure to proceed with napi polling
4400 * if the thread is explicitly woken from here.
4402 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4403 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4404 wake_up_process(thread);
4409 list_add_tail(&napi->poll_list, &sd->poll_list);
4410 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4415 /* One global table that all flow-based protocols share. */
4416 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4417 EXPORT_SYMBOL(rps_sock_flow_table);
4418 u32 rps_cpu_mask __read_mostly;
4419 EXPORT_SYMBOL(rps_cpu_mask);
4421 struct static_key_false rps_needed __read_mostly;
4422 EXPORT_SYMBOL(rps_needed);
4423 struct static_key_false rfs_needed __read_mostly;
4424 EXPORT_SYMBOL(rfs_needed);
4426 static struct rps_dev_flow *
4427 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4428 struct rps_dev_flow *rflow, u16 next_cpu)
4430 if (next_cpu < nr_cpu_ids) {
4431 #ifdef CONFIG_RFS_ACCEL
4432 struct netdev_rx_queue *rxqueue;
4433 struct rps_dev_flow_table *flow_table;
4434 struct rps_dev_flow *old_rflow;
4439 /* Should we steer this flow to a different hardware queue? */
4440 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4441 !(dev->features & NETIF_F_NTUPLE))
4443 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4444 if (rxq_index == skb_get_rx_queue(skb))
4447 rxqueue = dev->_rx + rxq_index;
4448 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4451 flow_id = skb_get_hash(skb) & flow_table->mask;
4452 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4453 rxq_index, flow_id);
4457 rflow = &flow_table->flows[flow_id];
4459 if (old_rflow->filter == rflow->filter)
4460 old_rflow->filter = RPS_NO_FILTER;
4464 per_cpu(softnet_data, next_cpu).input_queue_head;
4467 rflow->cpu = next_cpu;
4472 * get_rps_cpu is called from netif_receive_skb and returns the target
4473 * CPU from the RPS map of the receiving queue for a given skb.
4474 * rcu_read_lock must be held on entry.
4476 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4477 struct rps_dev_flow **rflowp)
4479 const struct rps_sock_flow_table *sock_flow_table;
4480 struct netdev_rx_queue *rxqueue = dev->_rx;
4481 struct rps_dev_flow_table *flow_table;
4482 struct rps_map *map;
4487 if (skb_rx_queue_recorded(skb)) {
4488 u16 index = skb_get_rx_queue(skb);
4490 if (unlikely(index >= dev->real_num_rx_queues)) {
4491 WARN_ONCE(dev->real_num_rx_queues > 1,
4492 "%s received packet on queue %u, but number "
4493 "of RX queues is %u\n",
4494 dev->name, index, dev->real_num_rx_queues);
4500 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4502 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4503 map = rcu_dereference(rxqueue->rps_map);
4504 if (!flow_table && !map)
4507 skb_reset_network_header(skb);
4508 hash = skb_get_hash(skb);
4512 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4513 if (flow_table && sock_flow_table) {
4514 struct rps_dev_flow *rflow;
4518 /* First check into global flow table if there is a match.
4519 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4521 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4522 if ((ident ^ hash) & ~rps_cpu_mask)
4525 next_cpu = ident & rps_cpu_mask;
4527 /* OK, now we know there is a match,
4528 * we can look at the local (per receive queue) flow table
4530 rflow = &flow_table->flows[hash & flow_table->mask];
4534 * If the desired CPU (where last recvmsg was done) is
4535 * different from current CPU (one in the rx-queue flow
4536 * table entry), switch if one of the following holds:
4537 * - Current CPU is unset (>= nr_cpu_ids).
4538 * - Current CPU is offline.
4539 * - The current CPU's queue tail has advanced beyond the
4540 * last packet that was enqueued using this table entry.
4541 * This guarantees that all previous packets for the flow
4542 * have been dequeued, thus preserving in order delivery.
4544 if (unlikely(tcpu != next_cpu) &&
4545 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4546 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4547 rflow->last_qtail)) >= 0)) {
4549 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4552 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4562 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4563 if (cpu_online(tcpu)) {
4573 #ifdef CONFIG_RFS_ACCEL
4576 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4577 * @dev: Device on which the filter was set
4578 * @rxq_index: RX queue index
4579 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4580 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4582 * Drivers that implement ndo_rx_flow_steer() should periodically call
4583 * this function for each installed filter and remove the filters for
4584 * which it returns %true.
4586 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4587 u32 flow_id, u16 filter_id)
4589 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4590 struct rps_dev_flow_table *flow_table;
4591 struct rps_dev_flow *rflow;
4596 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4597 if (flow_table && flow_id <= flow_table->mask) {
4598 rflow = &flow_table->flows[flow_id];
4599 cpu = READ_ONCE(rflow->cpu);
4600 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4601 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4602 rflow->last_qtail) <
4603 (int)(10 * flow_table->mask)))
4609 EXPORT_SYMBOL(rps_may_expire_flow);
4611 #endif /* CONFIG_RFS_ACCEL */
4613 /* Called from hardirq (IPI) context */
4614 static void rps_trigger_softirq(void *data)
4616 struct softnet_data *sd = data;
4618 ____napi_schedule(sd, &sd->backlog);
4622 #endif /* CONFIG_RPS */
4624 /* Called from hardirq (IPI) context */
4625 static void trigger_rx_softirq(void *data)
4627 struct softnet_data *sd = data;
4629 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4630 smp_store_release(&sd->defer_ipi_scheduled, 0);
4634 * Check if this softnet_data structure is another cpu one
4635 * If yes, queue it to our IPI list and return 1
4638 static int napi_schedule_rps(struct softnet_data *sd)
4640 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4644 sd->rps_ipi_next = mysd->rps_ipi_list;
4645 mysd->rps_ipi_list = sd;
4647 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4650 #endif /* CONFIG_RPS */
4651 __napi_schedule_irqoff(&mysd->backlog);
4655 #ifdef CONFIG_NET_FLOW_LIMIT
4656 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4659 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4661 #ifdef CONFIG_NET_FLOW_LIMIT
4662 struct sd_flow_limit *fl;
4663 struct softnet_data *sd;
4664 unsigned int old_flow, new_flow;
4666 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4669 sd = this_cpu_ptr(&softnet_data);
4672 fl = rcu_dereference(sd->flow_limit);
4674 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4675 old_flow = fl->history[fl->history_head];
4676 fl->history[fl->history_head] = new_flow;
4679 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4681 if (likely(fl->buckets[old_flow]))
4682 fl->buckets[old_flow]--;
4684 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4696 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4697 * queue (may be a remote CPU queue).
4699 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4700 unsigned int *qtail)
4702 enum skb_drop_reason reason;
4703 struct softnet_data *sd;
4704 unsigned long flags;
4707 reason = SKB_DROP_REASON_NOT_SPECIFIED;
4708 sd = &per_cpu(softnet_data, cpu);
4710 rps_lock_irqsave(sd, &flags);
4711 if (!netif_running(skb->dev))
4713 qlen = skb_queue_len(&sd->input_pkt_queue);
4714 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4717 __skb_queue_tail(&sd->input_pkt_queue, skb);
4718 input_queue_tail_incr_save(sd, qtail);
4719 rps_unlock_irq_restore(sd, &flags);
4720 return NET_RX_SUCCESS;
4723 /* Schedule NAPI for backlog device
4724 * We can use non atomic operation since we own the queue lock
4726 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4727 napi_schedule_rps(sd);
4730 reason = SKB_DROP_REASON_CPU_BACKLOG;
4734 rps_unlock_irq_restore(sd, &flags);
4736 dev_core_stats_rx_dropped_inc(skb->dev);
4737 kfree_skb_reason(skb, reason);
4741 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4743 struct net_device *dev = skb->dev;
4744 struct netdev_rx_queue *rxqueue;
4748 if (skb_rx_queue_recorded(skb)) {
4749 u16 index = skb_get_rx_queue(skb);
4751 if (unlikely(index >= dev->real_num_rx_queues)) {
4752 WARN_ONCE(dev->real_num_rx_queues > 1,
4753 "%s received packet on queue %u, but number "
4754 "of RX queues is %u\n",
4755 dev->name, index, dev->real_num_rx_queues);
4757 return rxqueue; /* Return first rxqueue */
4764 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4765 struct bpf_prog *xdp_prog)
4767 void *orig_data, *orig_data_end, *hard_start;
4768 struct netdev_rx_queue *rxqueue;
4769 bool orig_bcast, orig_host;
4770 u32 mac_len, frame_sz;
4771 __be16 orig_eth_type;
4776 /* The XDP program wants to see the packet starting at the MAC
4779 mac_len = skb->data - skb_mac_header(skb);
4780 hard_start = skb->data - skb_headroom(skb);
4782 /* SKB "head" area always have tailroom for skb_shared_info */
4783 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4784 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4786 rxqueue = netif_get_rxqueue(skb);
4787 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4788 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4789 skb_headlen(skb) + mac_len, true);
4791 orig_data_end = xdp->data_end;
4792 orig_data = xdp->data;
4793 eth = (struct ethhdr *)xdp->data;
4794 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4795 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4796 orig_eth_type = eth->h_proto;
4798 act = bpf_prog_run_xdp(xdp_prog, xdp);
4800 /* check if bpf_xdp_adjust_head was used */
4801 off = xdp->data - orig_data;
4804 __skb_pull(skb, off);
4806 __skb_push(skb, -off);
4808 skb->mac_header += off;
4809 skb_reset_network_header(skb);
4812 /* check if bpf_xdp_adjust_tail was used */
4813 off = xdp->data_end - orig_data_end;
4815 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4816 skb->len += off; /* positive on grow, negative on shrink */
4819 /* check if XDP changed eth hdr such SKB needs update */
4820 eth = (struct ethhdr *)xdp->data;
4821 if ((orig_eth_type != eth->h_proto) ||
4822 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4823 skb->dev->dev_addr)) ||
4824 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4825 __skb_push(skb, ETH_HLEN);
4826 skb->pkt_type = PACKET_HOST;
4827 skb->protocol = eth_type_trans(skb, skb->dev);
4830 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4831 * before calling us again on redirect path. We do not call do_redirect
4832 * as we leave that up to the caller.
4834 * Caller is responsible for managing lifetime of skb (i.e. calling
4835 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4840 __skb_push(skb, mac_len);
4843 metalen = xdp->data - xdp->data_meta;
4845 skb_metadata_set(skb, metalen);
4852 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4853 struct xdp_buff *xdp,
4854 struct bpf_prog *xdp_prog)
4858 /* Reinjected packets coming from act_mirred or similar should
4859 * not get XDP generic processing.
4861 if (skb_is_redirected(skb))
4864 /* XDP packets must be linear and must have sufficient headroom
4865 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4866 * native XDP provides, thus we need to do it here as well.
4868 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4869 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4870 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4871 int troom = skb->tail + skb->data_len - skb->end;
4873 /* In case we have to go down the path and also linearize,
4874 * then lets do the pskb_expand_head() work just once here.
4876 if (pskb_expand_head(skb,
4877 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4878 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4880 if (skb_linearize(skb))
4884 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4891 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4894 trace_xdp_exception(skb->dev, xdp_prog, act);
4905 /* When doing generic XDP we have to bypass the qdisc layer and the
4906 * network taps in order to match in-driver-XDP behavior. This also means
4907 * that XDP packets are able to starve other packets going through a qdisc,
4908 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4909 * queues, so they do not have this starvation issue.
4911 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4913 struct net_device *dev = skb->dev;
4914 struct netdev_queue *txq;
4915 bool free_skb = true;
4918 txq = netdev_core_pick_tx(dev, skb, NULL);
4919 cpu = smp_processor_id();
4920 HARD_TX_LOCK(dev, txq, cpu);
4921 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4922 rc = netdev_start_xmit(skb, dev, txq, 0);
4923 if (dev_xmit_complete(rc))
4926 HARD_TX_UNLOCK(dev, txq);
4928 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4929 dev_core_stats_tx_dropped_inc(dev);
4934 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4936 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4939 struct xdp_buff xdp;
4943 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4944 if (act != XDP_PASS) {
4947 err = xdp_do_generic_redirect(skb->dev, skb,
4953 generic_xdp_tx(skb, xdp_prog);
4961 kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4964 EXPORT_SYMBOL_GPL(do_xdp_generic);
4966 static int netif_rx_internal(struct sk_buff *skb)
4970 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4972 trace_netif_rx(skb);
4975 if (static_branch_unlikely(&rps_needed)) {
4976 struct rps_dev_flow voidflow, *rflow = &voidflow;
4981 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4983 cpu = smp_processor_id();
4985 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4993 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4999 * __netif_rx - Slightly optimized version of netif_rx
5000 * @skb: buffer to post
5002 * This behaves as netif_rx except that it does not disable bottom halves.
5003 * As a result this function may only be invoked from the interrupt context
5004 * (either hard or soft interrupt).
5006 int __netif_rx(struct sk_buff *skb)
5010 lockdep_assert_once(hardirq_count() | softirq_count());
5012 trace_netif_rx_entry(skb);
5013 ret = netif_rx_internal(skb);
5014 trace_netif_rx_exit(ret);
5017 EXPORT_SYMBOL(__netif_rx);
5020 * netif_rx - post buffer to the network code
5021 * @skb: buffer to post
5023 * This function receives a packet from a device driver and queues it for
5024 * the upper (protocol) levels to process via the backlog NAPI device. It
5025 * always succeeds. The buffer may be dropped during processing for
5026 * congestion control or by the protocol layers.
5027 * The network buffer is passed via the backlog NAPI device. Modern NIC
5028 * driver should use NAPI and GRO.
5029 * This function can used from interrupt and from process context. The
5030 * caller from process context must not disable interrupts before invoking
5034 * NET_RX_SUCCESS (no congestion)
5035 * NET_RX_DROP (packet was dropped)
5038 int netif_rx(struct sk_buff *skb)
5040 bool need_bh_off = !(hardirq_count() | softirq_count());
5045 trace_netif_rx_entry(skb);
5046 ret = netif_rx_internal(skb);
5047 trace_netif_rx_exit(ret);
5052 EXPORT_SYMBOL(netif_rx);
5054 static __latent_entropy void net_tx_action(struct softirq_action *h)
5056 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5058 if (sd->completion_queue) {
5059 struct sk_buff *clist;
5061 local_irq_disable();
5062 clist = sd->completion_queue;
5063 sd->completion_queue = NULL;
5067 struct sk_buff *skb = clist;
5069 clist = clist->next;
5071 WARN_ON(refcount_read(&skb->users));
5072 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5073 trace_consume_skb(skb);
5075 trace_kfree_skb(skb, net_tx_action,
5076 SKB_DROP_REASON_NOT_SPECIFIED);
5078 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5081 __kfree_skb_defer(skb);
5085 if (sd->output_queue) {
5088 local_irq_disable();
5089 head = sd->output_queue;
5090 sd->output_queue = NULL;
5091 sd->output_queue_tailp = &sd->output_queue;
5097 struct Qdisc *q = head;
5098 spinlock_t *root_lock = NULL;
5100 head = head->next_sched;
5102 /* We need to make sure head->next_sched is read
5103 * before clearing __QDISC_STATE_SCHED
5105 smp_mb__before_atomic();
5107 if (!(q->flags & TCQ_F_NOLOCK)) {
5108 root_lock = qdisc_lock(q);
5109 spin_lock(root_lock);
5110 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5112 /* There is a synchronize_net() between
5113 * STATE_DEACTIVATED flag being set and
5114 * qdisc_reset()/some_qdisc_is_busy() in
5115 * dev_deactivate(), so we can safely bail out
5116 * early here to avoid data race between
5117 * qdisc_deactivate() and some_qdisc_is_busy()
5118 * for lockless qdisc.
5120 clear_bit(__QDISC_STATE_SCHED, &q->state);
5124 clear_bit(__QDISC_STATE_SCHED, &q->state);
5127 spin_unlock(root_lock);
5133 xfrm_dev_backlog(sd);
5136 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5137 /* This hook is defined here for ATM LANE */
5138 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5139 unsigned char *addr) __read_mostly;
5140 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5143 static inline struct sk_buff *
5144 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5145 struct net_device *orig_dev, bool *another)
5147 #ifdef CONFIG_NET_CLS_ACT
5148 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5149 struct tcf_result cl_res;
5151 /* If there's at least one ingress present somewhere (so
5152 * we get here via enabled static key), remaining devices
5153 * that are not configured with an ingress qdisc will bail
5160 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5164 qdisc_skb_cb(skb)->pkt_len = skb->len;
5165 tc_skb_cb(skb)->mru = 0;
5166 tc_skb_cb(skb)->post_ct = false;
5167 skb->tc_at_ingress = 1;
5168 mini_qdisc_bstats_cpu_update(miniq, skb);
5170 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5172 case TC_ACT_RECLASSIFY:
5173 skb->tc_index = TC_H_MIN(cl_res.classid);
5176 mini_qdisc_qstats_cpu_drop(miniq);
5177 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5184 *ret = NET_RX_SUCCESS;
5186 case TC_ACT_REDIRECT:
5187 /* skb_mac_header check was done by cls/act_bpf, so
5188 * we can safely push the L2 header back before
5189 * redirecting to another netdev
5191 __skb_push(skb, skb->mac_len);
5192 if (skb_do_redirect(skb) == -EAGAIN) {
5193 __skb_pull(skb, skb->mac_len);
5197 *ret = NET_RX_SUCCESS;
5199 case TC_ACT_CONSUMED:
5200 *ret = NET_RX_SUCCESS;
5205 #endif /* CONFIG_NET_CLS_ACT */
5210 * netdev_is_rx_handler_busy - check if receive handler is registered
5211 * @dev: device to check
5213 * Check if a receive handler is already registered for a given device.
5214 * Return true if there one.
5216 * The caller must hold the rtnl_mutex.
5218 bool netdev_is_rx_handler_busy(struct net_device *dev)
5221 return dev && rtnl_dereference(dev->rx_handler);
5223 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5226 * netdev_rx_handler_register - register receive handler
5227 * @dev: device to register a handler for
5228 * @rx_handler: receive handler to register
5229 * @rx_handler_data: data pointer that is used by rx handler
5231 * Register a receive handler for a device. This handler will then be
5232 * called from __netif_receive_skb. A negative errno code is returned
5235 * The caller must hold the rtnl_mutex.
5237 * For a general description of rx_handler, see enum rx_handler_result.
5239 int netdev_rx_handler_register(struct net_device *dev,
5240 rx_handler_func_t *rx_handler,
5241 void *rx_handler_data)
5243 if (netdev_is_rx_handler_busy(dev))
5246 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5249 /* Note: rx_handler_data must be set before rx_handler */
5250 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5251 rcu_assign_pointer(dev->rx_handler, rx_handler);
5255 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5258 * netdev_rx_handler_unregister - unregister receive handler
5259 * @dev: device to unregister a handler from
5261 * Unregister a receive handler from a device.
5263 * The caller must hold the rtnl_mutex.
5265 void netdev_rx_handler_unregister(struct net_device *dev)
5269 RCU_INIT_POINTER(dev->rx_handler, NULL);
5270 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5271 * section has a guarantee to see a non NULL rx_handler_data
5275 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5277 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5280 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5281 * the special handling of PFMEMALLOC skbs.
5283 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5285 switch (skb->protocol) {
5286 case htons(ETH_P_ARP):
5287 case htons(ETH_P_IP):
5288 case htons(ETH_P_IPV6):
5289 case htons(ETH_P_8021Q):
5290 case htons(ETH_P_8021AD):
5297 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5298 int *ret, struct net_device *orig_dev)
5300 if (nf_hook_ingress_active(skb)) {
5304 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5309 ingress_retval = nf_hook_ingress(skb);
5311 return ingress_retval;
5316 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5317 struct packet_type **ppt_prev)
5319 struct packet_type *ptype, *pt_prev;
5320 rx_handler_func_t *rx_handler;
5321 struct sk_buff *skb = *pskb;
5322 struct net_device *orig_dev;
5323 bool deliver_exact = false;
5324 int ret = NET_RX_DROP;
5327 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5329 trace_netif_receive_skb(skb);
5331 orig_dev = skb->dev;
5333 skb_reset_network_header(skb);
5334 if (!skb_transport_header_was_set(skb))
5335 skb_reset_transport_header(skb);
5336 skb_reset_mac_len(skb);
5341 skb->skb_iif = skb->dev->ifindex;
5343 __this_cpu_inc(softnet_data.processed);
5345 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5349 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5352 if (ret2 != XDP_PASS) {
5358 if (eth_type_vlan(skb->protocol)) {
5359 skb = skb_vlan_untag(skb);
5364 if (skb_skip_tc_classify(skb))
5370 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5372 ret = deliver_skb(skb, pt_prev, orig_dev);
5376 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5378 ret = deliver_skb(skb, pt_prev, orig_dev);
5383 #ifdef CONFIG_NET_INGRESS
5384 if (static_branch_unlikely(&ingress_needed_key)) {
5385 bool another = false;
5387 nf_skip_egress(skb, true);
5388 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5395 nf_skip_egress(skb, false);
5396 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5400 skb_reset_redirect(skb);
5402 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5405 if (skb_vlan_tag_present(skb)) {
5407 ret = deliver_skb(skb, pt_prev, orig_dev);
5410 if (vlan_do_receive(&skb))
5412 else if (unlikely(!skb))
5416 rx_handler = rcu_dereference(skb->dev->rx_handler);
5419 ret = deliver_skb(skb, pt_prev, orig_dev);
5422 switch (rx_handler(&skb)) {
5423 case RX_HANDLER_CONSUMED:
5424 ret = NET_RX_SUCCESS;
5426 case RX_HANDLER_ANOTHER:
5428 case RX_HANDLER_EXACT:
5429 deliver_exact = true;
5431 case RX_HANDLER_PASS:
5438 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5440 if (skb_vlan_tag_get_id(skb)) {
5441 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5444 skb->pkt_type = PACKET_OTHERHOST;
5445 } else if (eth_type_vlan(skb->protocol)) {
5446 /* Outer header is 802.1P with vlan 0, inner header is
5447 * 802.1Q or 802.1AD and vlan_do_receive() above could
5448 * not find vlan dev for vlan id 0.
5450 __vlan_hwaccel_clear_tag(skb);
5451 skb = skb_vlan_untag(skb);
5454 if (vlan_do_receive(&skb))
5455 /* After stripping off 802.1P header with vlan 0
5456 * vlan dev is found for inner header.
5459 else if (unlikely(!skb))
5462 /* We have stripped outer 802.1P vlan 0 header.
5463 * But could not find vlan dev.
5464 * check again for vlan id to set OTHERHOST.
5468 /* Note: we might in the future use prio bits
5469 * and set skb->priority like in vlan_do_receive()
5470 * For the time being, just ignore Priority Code Point
5472 __vlan_hwaccel_clear_tag(skb);
5475 type = skb->protocol;
5477 /* deliver only exact match when indicated */
5478 if (likely(!deliver_exact)) {
5479 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5480 &ptype_base[ntohs(type) &
5484 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5485 &orig_dev->ptype_specific);
5487 if (unlikely(skb->dev != orig_dev)) {
5488 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5489 &skb->dev->ptype_specific);
5493 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5495 *ppt_prev = pt_prev;
5499 dev_core_stats_rx_dropped_inc(skb->dev);
5501 dev_core_stats_rx_nohandler_inc(skb->dev);
5502 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5503 /* Jamal, now you will not able to escape explaining
5504 * me how you were going to use this. :-)
5510 /* The invariant here is that if *ppt_prev is not NULL
5511 * then skb should also be non-NULL.
5513 * Apparently *ppt_prev assignment above holds this invariant due to
5514 * skb dereferencing near it.
5520 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5522 struct net_device *orig_dev = skb->dev;
5523 struct packet_type *pt_prev = NULL;
5526 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5528 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5529 skb->dev, pt_prev, orig_dev);
5534 * netif_receive_skb_core - special purpose version of netif_receive_skb
5535 * @skb: buffer to process
5537 * More direct receive version of netif_receive_skb(). It should
5538 * only be used by callers that have a need to skip RPS and Generic XDP.
5539 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5541 * This function may only be called from softirq context and interrupts
5542 * should be enabled.
5544 * Return values (usually ignored):
5545 * NET_RX_SUCCESS: no congestion
5546 * NET_RX_DROP: packet was dropped
5548 int netif_receive_skb_core(struct sk_buff *skb)
5553 ret = __netif_receive_skb_one_core(skb, false);
5558 EXPORT_SYMBOL(netif_receive_skb_core);
5560 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5561 struct packet_type *pt_prev,
5562 struct net_device *orig_dev)
5564 struct sk_buff *skb, *next;
5568 if (list_empty(head))
5570 if (pt_prev->list_func != NULL)
5571 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5572 ip_list_rcv, head, pt_prev, orig_dev);
5574 list_for_each_entry_safe(skb, next, head, list) {
5575 skb_list_del_init(skb);
5576 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5580 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5582 /* Fast-path assumptions:
5583 * - There is no RX handler.
5584 * - Only one packet_type matches.
5585 * If either of these fails, we will end up doing some per-packet
5586 * processing in-line, then handling the 'last ptype' for the whole
5587 * sublist. This can't cause out-of-order delivery to any single ptype,
5588 * because the 'last ptype' must be constant across the sublist, and all
5589 * other ptypes are handled per-packet.
5591 /* Current (common) ptype of sublist */
5592 struct packet_type *pt_curr = NULL;
5593 /* Current (common) orig_dev of sublist */
5594 struct net_device *od_curr = NULL;
5595 struct list_head sublist;
5596 struct sk_buff *skb, *next;
5598 INIT_LIST_HEAD(&sublist);
5599 list_for_each_entry_safe(skb, next, head, list) {
5600 struct net_device *orig_dev = skb->dev;
5601 struct packet_type *pt_prev = NULL;
5603 skb_list_del_init(skb);
5604 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5607 if (pt_curr != pt_prev || od_curr != orig_dev) {
5608 /* dispatch old sublist */
5609 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5610 /* start new sublist */
5611 INIT_LIST_HEAD(&sublist);
5615 list_add_tail(&skb->list, &sublist);
5618 /* dispatch final sublist */
5619 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5622 static int __netif_receive_skb(struct sk_buff *skb)
5626 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5627 unsigned int noreclaim_flag;
5630 * PFMEMALLOC skbs are special, they should
5631 * - be delivered to SOCK_MEMALLOC sockets only
5632 * - stay away from userspace
5633 * - have bounded memory usage
5635 * Use PF_MEMALLOC as this saves us from propagating the allocation
5636 * context down to all allocation sites.
5638 noreclaim_flag = memalloc_noreclaim_save();
5639 ret = __netif_receive_skb_one_core(skb, true);
5640 memalloc_noreclaim_restore(noreclaim_flag);
5642 ret = __netif_receive_skb_one_core(skb, false);
5647 static void __netif_receive_skb_list(struct list_head *head)
5649 unsigned long noreclaim_flag = 0;
5650 struct sk_buff *skb, *next;
5651 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5653 list_for_each_entry_safe(skb, next, head, list) {
5654 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5655 struct list_head sublist;
5657 /* Handle the previous sublist */
5658 list_cut_before(&sublist, head, &skb->list);
5659 if (!list_empty(&sublist))
5660 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5661 pfmemalloc = !pfmemalloc;
5662 /* See comments in __netif_receive_skb */
5664 noreclaim_flag = memalloc_noreclaim_save();
5666 memalloc_noreclaim_restore(noreclaim_flag);
5669 /* Handle the remaining sublist */
5670 if (!list_empty(head))
5671 __netif_receive_skb_list_core(head, pfmemalloc);
5672 /* Restore pflags */
5674 memalloc_noreclaim_restore(noreclaim_flag);
5677 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5679 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5680 struct bpf_prog *new = xdp->prog;
5683 switch (xdp->command) {
5684 case XDP_SETUP_PROG:
5685 rcu_assign_pointer(dev->xdp_prog, new);
5690 static_branch_dec(&generic_xdp_needed_key);
5691 } else if (new && !old) {
5692 static_branch_inc(&generic_xdp_needed_key);
5693 dev_disable_lro(dev);
5694 dev_disable_gro_hw(dev);
5706 static int netif_receive_skb_internal(struct sk_buff *skb)
5710 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5712 if (skb_defer_rx_timestamp(skb))
5713 return NET_RX_SUCCESS;
5717 if (static_branch_unlikely(&rps_needed)) {
5718 struct rps_dev_flow voidflow, *rflow = &voidflow;
5719 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5722 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5728 ret = __netif_receive_skb(skb);
5733 void netif_receive_skb_list_internal(struct list_head *head)
5735 struct sk_buff *skb, *next;
5736 struct list_head sublist;
5738 INIT_LIST_HEAD(&sublist);
5739 list_for_each_entry_safe(skb, next, head, list) {
5740 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5741 skb_list_del_init(skb);
5742 if (!skb_defer_rx_timestamp(skb))
5743 list_add_tail(&skb->list, &sublist);
5745 list_splice_init(&sublist, head);
5749 if (static_branch_unlikely(&rps_needed)) {
5750 list_for_each_entry_safe(skb, next, head, list) {
5751 struct rps_dev_flow voidflow, *rflow = &voidflow;
5752 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5755 /* Will be handled, remove from list */
5756 skb_list_del_init(skb);
5757 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5762 __netif_receive_skb_list(head);
5767 * netif_receive_skb - process receive buffer from network
5768 * @skb: buffer to process
5770 * netif_receive_skb() is the main receive data processing function.
5771 * It always succeeds. The buffer may be dropped during processing
5772 * for congestion control or by the protocol layers.
5774 * This function may only be called from softirq context and interrupts
5775 * should be enabled.
5777 * Return values (usually ignored):
5778 * NET_RX_SUCCESS: no congestion
5779 * NET_RX_DROP: packet was dropped
5781 int netif_receive_skb(struct sk_buff *skb)
5785 trace_netif_receive_skb_entry(skb);
5787 ret = netif_receive_skb_internal(skb);
5788 trace_netif_receive_skb_exit(ret);
5792 EXPORT_SYMBOL(netif_receive_skb);
5795 * netif_receive_skb_list - process many receive buffers from network
5796 * @head: list of skbs to process.
5798 * Since return value of netif_receive_skb() is normally ignored, and
5799 * wouldn't be meaningful for a list, this function returns void.
5801 * This function may only be called from softirq context and interrupts
5802 * should be enabled.
5804 void netif_receive_skb_list(struct list_head *head)
5806 struct sk_buff *skb;
5808 if (list_empty(head))
5810 if (trace_netif_receive_skb_list_entry_enabled()) {
5811 list_for_each_entry(skb, head, list)
5812 trace_netif_receive_skb_list_entry(skb);
5814 netif_receive_skb_list_internal(head);
5815 trace_netif_receive_skb_list_exit(0);
5817 EXPORT_SYMBOL(netif_receive_skb_list);
5819 static DEFINE_PER_CPU(struct work_struct, flush_works);
5821 /* Network device is going away, flush any packets still pending */
5822 static void flush_backlog(struct work_struct *work)
5824 struct sk_buff *skb, *tmp;
5825 struct softnet_data *sd;
5828 sd = this_cpu_ptr(&softnet_data);
5830 rps_lock_irq_disable(sd);
5831 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5832 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5833 __skb_unlink(skb, &sd->input_pkt_queue);
5834 dev_kfree_skb_irq(skb);
5835 input_queue_head_incr(sd);
5838 rps_unlock_irq_enable(sd);
5840 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5841 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5842 __skb_unlink(skb, &sd->process_queue);
5844 input_queue_head_incr(sd);
5850 static bool flush_required(int cpu)
5852 #if IS_ENABLED(CONFIG_RPS)
5853 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5856 rps_lock_irq_disable(sd);
5858 /* as insertion into process_queue happens with the rps lock held,
5859 * process_queue access may race only with dequeue
5861 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5862 !skb_queue_empty_lockless(&sd->process_queue);
5863 rps_unlock_irq_enable(sd);
5867 /* without RPS we can't safely check input_pkt_queue: during a
5868 * concurrent remote skb_queue_splice() we can detect as empty both
5869 * input_pkt_queue and process_queue even if the latter could end-up
5870 * containing a lot of packets.
5875 static void flush_all_backlogs(void)
5877 static cpumask_t flush_cpus;
5880 /* since we are under rtnl lock protection we can use static data
5881 * for the cpumask and avoid allocating on stack the possibly
5888 cpumask_clear(&flush_cpus);
5889 for_each_online_cpu(cpu) {
5890 if (flush_required(cpu)) {
5891 queue_work_on(cpu, system_highpri_wq,
5892 per_cpu_ptr(&flush_works, cpu));
5893 cpumask_set_cpu(cpu, &flush_cpus);
5897 /* we can have in flight packet[s] on the cpus we are not flushing,
5898 * synchronize_net() in unregister_netdevice_many() will take care of
5901 for_each_cpu(cpu, &flush_cpus)
5902 flush_work(per_cpu_ptr(&flush_works, cpu));
5907 static void net_rps_send_ipi(struct softnet_data *remsd)
5911 struct softnet_data *next = remsd->rps_ipi_next;
5913 if (cpu_online(remsd->cpu))
5914 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5921 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5922 * Note: called with local irq disabled, but exits with local irq enabled.
5924 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5927 struct softnet_data *remsd = sd->rps_ipi_list;
5930 sd->rps_ipi_list = NULL;
5934 /* Send pending IPI's to kick RPS processing on remote cpus. */
5935 net_rps_send_ipi(remsd);
5941 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5944 return sd->rps_ipi_list != NULL;
5950 static int process_backlog(struct napi_struct *napi, int quota)
5952 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5956 /* Check if we have pending ipi, its better to send them now,
5957 * not waiting net_rx_action() end.
5959 if (sd_has_rps_ipi_waiting(sd)) {
5960 local_irq_disable();
5961 net_rps_action_and_irq_enable(sd);
5964 napi->weight = READ_ONCE(dev_rx_weight);
5966 struct sk_buff *skb;
5968 while ((skb = __skb_dequeue(&sd->process_queue))) {
5970 __netif_receive_skb(skb);
5972 input_queue_head_incr(sd);
5973 if (++work >= quota)
5978 rps_lock_irq_disable(sd);
5979 if (skb_queue_empty(&sd->input_pkt_queue)) {
5981 * Inline a custom version of __napi_complete().
5982 * only current cpu owns and manipulates this napi,
5983 * and NAPI_STATE_SCHED is the only possible flag set
5985 * We can use a plain write instead of clear_bit(),
5986 * and we dont need an smp_mb() memory barrier.
5991 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5992 &sd->process_queue);
5994 rps_unlock_irq_enable(sd);
6001 * __napi_schedule - schedule for receive
6002 * @n: entry to schedule
6004 * The entry's receive function will be scheduled to run.
6005 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6007 void __napi_schedule(struct napi_struct *n)
6009 unsigned long flags;
6011 local_irq_save(flags);
6012 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6013 local_irq_restore(flags);
6015 EXPORT_SYMBOL(__napi_schedule);
6018 * napi_schedule_prep - check if napi can be scheduled
6021 * Test if NAPI routine is already running, and if not mark
6022 * it as running. This is used as a condition variable to
6023 * insure only one NAPI poll instance runs. We also make
6024 * sure there is no pending NAPI disable.
6026 bool napi_schedule_prep(struct napi_struct *n)
6028 unsigned long val, new;
6031 val = READ_ONCE(n->state);
6032 if (unlikely(val & NAPIF_STATE_DISABLE))
6034 new = val | NAPIF_STATE_SCHED;
6036 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6037 * This was suggested by Alexander Duyck, as compiler
6038 * emits better code than :
6039 * if (val & NAPIF_STATE_SCHED)
6040 * new |= NAPIF_STATE_MISSED;
6042 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6044 } while (cmpxchg(&n->state, val, new) != val);
6046 return !(val & NAPIF_STATE_SCHED);
6048 EXPORT_SYMBOL(napi_schedule_prep);
6051 * __napi_schedule_irqoff - schedule for receive
6052 * @n: entry to schedule
6054 * Variant of __napi_schedule() assuming hard irqs are masked.
6056 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6057 * because the interrupt disabled assumption might not be true
6058 * due to force-threaded interrupts and spinlock substitution.
6060 void __napi_schedule_irqoff(struct napi_struct *n)
6062 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6063 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6067 EXPORT_SYMBOL(__napi_schedule_irqoff);
6069 bool napi_complete_done(struct napi_struct *n, int work_done)
6071 unsigned long flags, val, new, timeout = 0;
6075 * 1) Don't let napi dequeue from the cpu poll list
6076 * just in case its running on a different cpu.
6077 * 2) If we are busy polling, do nothing here, we have
6078 * the guarantee we will be called later.
6080 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6081 NAPIF_STATE_IN_BUSY_POLL)))
6086 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6087 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6089 if (n->defer_hard_irqs_count > 0) {
6090 n->defer_hard_irqs_count--;
6091 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6095 if (n->gro_bitmask) {
6096 /* When the NAPI instance uses a timeout and keeps postponing
6097 * it, we need to bound somehow the time packets are kept in
6100 napi_gro_flush(n, !!timeout);
6105 if (unlikely(!list_empty(&n->poll_list))) {
6106 /* If n->poll_list is not empty, we need to mask irqs */
6107 local_irq_save(flags);
6108 list_del_init(&n->poll_list);
6109 local_irq_restore(flags);
6113 val = READ_ONCE(n->state);
6115 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6117 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6118 NAPIF_STATE_SCHED_THREADED |
6119 NAPIF_STATE_PREFER_BUSY_POLL);
6121 /* If STATE_MISSED was set, leave STATE_SCHED set,
6122 * because we will call napi->poll() one more time.
6123 * This C code was suggested by Alexander Duyck to help gcc.
6125 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6127 } while (cmpxchg(&n->state, val, new) != val);
6129 if (unlikely(val & NAPIF_STATE_MISSED)) {
6135 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6136 HRTIMER_MODE_REL_PINNED);
6139 EXPORT_SYMBOL(napi_complete_done);
6141 /* must be called under rcu_read_lock(), as we dont take a reference */
6142 static struct napi_struct *napi_by_id(unsigned int napi_id)
6144 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6145 struct napi_struct *napi;
6147 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6148 if (napi->napi_id == napi_id)
6154 #if defined(CONFIG_NET_RX_BUSY_POLL)
6156 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6158 if (!skip_schedule) {
6159 gro_normal_list(napi);
6160 __napi_schedule(napi);
6164 if (napi->gro_bitmask) {
6165 /* flush too old packets
6166 * If HZ < 1000, flush all packets.
6168 napi_gro_flush(napi, HZ >= 1000);
6171 gro_normal_list(napi);
6172 clear_bit(NAPI_STATE_SCHED, &napi->state);
6175 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6178 bool skip_schedule = false;
6179 unsigned long timeout;
6182 /* Busy polling means there is a high chance device driver hard irq
6183 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6184 * set in napi_schedule_prep().
6185 * Since we are about to call napi->poll() once more, we can safely
6186 * clear NAPI_STATE_MISSED.
6188 * Note: x86 could use a single "lock and ..." instruction
6189 * to perform these two clear_bit()
6191 clear_bit(NAPI_STATE_MISSED, &napi->state);
6192 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6196 if (prefer_busy_poll) {
6197 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6198 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6199 if (napi->defer_hard_irqs_count && timeout) {
6200 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6201 skip_schedule = true;
6205 /* All we really want here is to re-enable device interrupts.
6206 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6208 rc = napi->poll(napi, budget);
6209 /* We can't gro_normal_list() here, because napi->poll() might have
6210 * rearmed the napi (napi_complete_done()) in which case it could
6211 * already be running on another CPU.
6213 trace_napi_poll(napi, rc, budget);
6214 netpoll_poll_unlock(have_poll_lock);
6216 __busy_poll_stop(napi, skip_schedule);
6220 void napi_busy_loop(unsigned int napi_id,
6221 bool (*loop_end)(void *, unsigned long),
6222 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6224 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6225 int (*napi_poll)(struct napi_struct *napi, int budget);
6226 void *have_poll_lock = NULL;
6227 struct napi_struct *napi;
6234 napi = napi_by_id(napi_id);
6244 unsigned long val = READ_ONCE(napi->state);
6246 /* If multiple threads are competing for this napi,
6247 * we avoid dirtying napi->state as much as we can.
6249 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6250 NAPIF_STATE_IN_BUSY_POLL)) {
6251 if (prefer_busy_poll)
6252 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6255 if (cmpxchg(&napi->state, val,
6256 val | NAPIF_STATE_IN_BUSY_POLL |
6257 NAPIF_STATE_SCHED) != val) {
6258 if (prefer_busy_poll)
6259 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6262 have_poll_lock = netpoll_poll_lock(napi);
6263 napi_poll = napi->poll;
6265 work = napi_poll(napi, budget);
6266 trace_napi_poll(napi, work, budget);
6267 gro_normal_list(napi);
6270 __NET_ADD_STATS(dev_net(napi->dev),
6271 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6274 if (!loop_end || loop_end(loop_end_arg, start_time))
6277 if (unlikely(need_resched())) {
6279 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6283 if (loop_end(loop_end_arg, start_time))
6290 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6295 EXPORT_SYMBOL(napi_busy_loop);
6297 #endif /* CONFIG_NET_RX_BUSY_POLL */
6299 static void napi_hash_add(struct napi_struct *napi)
6301 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6304 spin_lock(&napi_hash_lock);
6306 /* 0..NR_CPUS range is reserved for sender_cpu use */
6308 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6309 napi_gen_id = MIN_NAPI_ID;
6310 } while (napi_by_id(napi_gen_id));
6311 napi->napi_id = napi_gen_id;
6313 hlist_add_head_rcu(&napi->napi_hash_node,
6314 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6316 spin_unlock(&napi_hash_lock);
6319 /* Warning : caller is responsible to make sure rcu grace period
6320 * is respected before freeing memory containing @napi
6322 static void napi_hash_del(struct napi_struct *napi)
6324 spin_lock(&napi_hash_lock);
6326 hlist_del_init_rcu(&napi->napi_hash_node);
6328 spin_unlock(&napi_hash_lock);
6331 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6333 struct napi_struct *napi;
6335 napi = container_of(timer, struct napi_struct, timer);
6337 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6338 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6340 if (!napi_disable_pending(napi) &&
6341 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6342 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6343 __napi_schedule_irqoff(napi);
6346 return HRTIMER_NORESTART;
6349 static void init_gro_hash(struct napi_struct *napi)
6353 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6354 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6355 napi->gro_hash[i].count = 0;
6357 napi->gro_bitmask = 0;
6360 int dev_set_threaded(struct net_device *dev, bool threaded)
6362 struct napi_struct *napi;
6365 if (dev->threaded == threaded)
6369 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6370 if (!napi->thread) {
6371 err = napi_kthread_create(napi);
6380 dev->threaded = threaded;
6382 /* Make sure kthread is created before THREADED bit
6385 smp_mb__before_atomic();
6387 /* Setting/unsetting threaded mode on a napi might not immediately
6388 * take effect, if the current napi instance is actively being
6389 * polled. In this case, the switch between threaded mode and
6390 * softirq mode will happen in the next round of napi_schedule().
6391 * This should not cause hiccups/stalls to the live traffic.
6393 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6395 set_bit(NAPI_STATE_THREADED, &napi->state);
6397 clear_bit(NAPI_STATE_THREADED, &napi->state);
6402 EXPORT_SYMBOL(dev_set_threaded);
6404 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6405 int (*poll)(struct napi_struct *, int), int weight)
6407 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6410 INIT_LIST_HEAD(&napi->poll_list);
6411 INIT_HLIST_NODE(&napi->napi_hash_node);
6412 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6413 napi->timer.function = napi_watchdog;
6414 init_gro_hash(napi);
6416 INIT_LIST_HEAD(&napi->rx_list);
6419 if (weight > NAPI_POLL_WEIGHT)
6420 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6422 napi->weight = weight;
6424 #ifdef CONFIG_NETPOLL
6425 napi->poll_owner = -1;
6427 set_bit(NAPI_STATE_SCHED, &napi->state);
6428 set_bit(NAPI_STATE_NPSVC, &napi->state);
6429 list_add_rcu(&napi->dev_list, &dev->napi_list);
6430 napi_hash_add(napi);
6431 napi_get_frags_check(napi);
6432 /* Create kthread for this napi if dev->threaded is set.
6433 * Clear dev->threaded if kthread creation failed so that
6434 * threaded mode will not be enabled in napi_enable().
6436 if (dev->threaded && napi_kthread_create(napi))
6439 EXPORT_SYMBOL(netif_napi_add_weight);
6441 void napi_disable(struct napi_struct *n)
6443 unsigned long val, new;
6446 set_bit(NAPI_STATE_DISABLE, &n->state);
6449 val = READ_ONCE(n->state);
6450 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6451 usleep_range(20, 200);
6455 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6456 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6458 if (cmpxchg(&n->state, val, new) == val)
6462 hrtimer_cancel(&n->timer);
6464 clear_bit(NAPI_STATE_DISABLE, &n->state);
6466 EXPORT_SYMBOL(napi_disable);
6469 * napi_enable - enable NAPI scheduling
6472 * Resume NAPI from being scheduled on this context.
6473 * Must be paired with napi_disable.
6475 void napi_enable(struct napi_struct *n)
6477 unsigned long val, new;
6480 val = READ_ONCE(n->state);
6481 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6483 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6484 if (n->dev->threaded && n->thread)
6485 new |= NAPIF_STATE_THREADED;
6486 } while (cmpxchg(&n->state, val, new) != val);
6488 EXPORT_SYMBOL(napi_enable);
6490 static void flush_gro_hash(struct napi_struct *napi)
6494 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6495 struct sk_buff *skb, *n;
6497 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6499 napi->gro_hash[i].count = 0;
6503 /* Must be called in process context */
6504 void __netif_napi_del(struct napi_struct *napi)
6506 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6509 napi_hash_del(napi);
6510 list_del_rcu(&napi->dev_list);
6511 napi_free_frags(napi);
6513 flush_gro_hash(napi);
6514 napi->gro_bitmask = 0;
6517 kthread_stop(napi->thread);
6518 napi->thread = NULL;
6521 EXPORT_SYMBOL(__netif_napi_del);
6523 static int __napi_poll(struct napi_struct *n, bool *repoll)
6529 /* This NAPI_STATE_SCHED test is for avoiding a race
6530 * with netpoll's poll_napi(). Only the entity which
6531 * obtains the lock and sees NAPI_STATE_SCHED set will
6532 * actually make the ->poll() call. Therefore we avoid
6533 * accidentally calling ->poll() when NAPI is not scheduled.
6536 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6537 work = n->poll(n, weight);
6538 trace_napi_poll(n, work, weight);
6541 if (unlikely(work > weight))
6542 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6543 n->poll, work, weight);
6545 if (likely(work < weight))
6548 /* Drivers must not modify the NAPI state if they
6549 * consume the entire weight. In such cases this code
6550 * still "owns" the NAPI instance and therefore can
6551 * move the instance around on the list at-will.
6553 if (unlikely(napi_disable_pending(n))) {
6558 /* The NAPI context has more processing work, but busy-polling
6559 * is preferred. Exit early.
6561 if (napi_prefer_busy_poll(n)) {
6562 if (napi_complete_done(n, work)) {
6563 /* If timeout is not set, we need to make sure
6564 * that the NAPI is re-scheduled.
6571 if (n->gro_bitmask) {
6572 /* flush too old packets
6573 * If HZ < 1000, flush all packets.
6575 napi_gro_flush(n, HZ >= 1000);
6580 /* Some drivers may have called napi_schedule
6581 * prior to exhausting their budget.
6583 if (unlikely(!list_empty(&n->poll_list))) {
6584 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6585 n->dev ? n->dev->name : "backlog");
6594 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6596 bool do_repoll = false;
6600 list_del_init(&n->poll_list);
6602 have = netpoll_poll_lock(n);
6604 work = __napi_poll(n, &do_repoll);
6607 list_add_tail(&n->poll_list, repoll);
6609 netpoll_poll_unlock(have);
6614 static int napi_thread_wait(struct napi_struct *napi)
6618 set_current_state(TASK_INTERRUPTIBLE);
6620 while (!kthread_should_stop()) {
6621 /* Testing SCHED_THREADED bit here to make sure the current
6622 * kthread owns this napi and could poll on this napi.
6623 * Testing SCHED bit is not enough because SCHED bit might be
6624 * set by some other busy poll thread or by napi_disable().
6626 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6627 WARN_ON(!list_empty(&napi->poll_list));
6628 __set_current_state(TASK_RUNNING);
6633 /* woken being true indicates this thread owns this napi. */
6635 set_current_state(TASK_INTERRUPTIBLE);
6637 __set_current_state(TASK_RUNNING);
6642 static int napi_threaded_poll(void *data)
6644 struct napi_struct *napi = data;
6647 while (!napi_thread_wait(napi)) {
6648 unsigned long last_qs = jiffies;
6651 bool repoll = false;
6655 have = netpoll_poll_lock(napi);
6656 __napi_poll(napi, &repoll);
6657 netpoll_poll_unlock(have);
6664 rcu_softirq_qs_periodic(last_qs);
6671 static void skb_defer_free_flush(struct softnet_data *sd)
6673 struct sk_buff *skb, *next;
6674 unsigned long flags;
6676 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6677 if (!READ_ONCE(sd->defer_list))
6680 spin_lock_irqsave(&sd->defer_lock, flags);
6681 skb = sd->defer_list;
6682 sd->defer_list = NULL;
6683 sd->defer_count = 0;
6684 spin_unlock_irqrestore(&sd->defer_lock, flags);
6686 while (skb != NULL) {
6688 napi_consume_skb(skb, 1);
6693 static __latent_entropy void net_rx_action(struct softirq_action *h)
6695 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6696 unsigned long time_limit = jiffies +
6697 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6698 int budget = READ_ONCE(netdev_budget);
6702 local_irq_disable();
6703 list_splice_init(&sd->poll_list, &list);
6707 struct napi_struct *n;
6709 skb_defer_free_flush(sd);
6711 if (list_empty(&list)) {
6712 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6717 n = list_first_entry(&list, struct napi_struct, poll_list);
6718 budget -= napi_poll(n, &repoll);
6720 /* If softirq window is exhausted then punt.
6721 * Allow this to run for 2 jiffies since which will allow
6722 * an average latency of 1.5/HZ.
6724 if (unlikely(budget <= 0 ||
6725 time_after_eq(jiffies, time_limit))) {
6731 local_irq_disable();
6733 list_splice_tail_init(&sd->poll_list, &list);
6734 list_splice_tail(&repoll, &list);
6735 list_splice(&list, &sd->poll_list);
6736 if (!list_empty(&sd->poll_list))
6737 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6739 net_rps_action_and_irq_enable(sd);
6743 struct netdev_adjacent {
6744 struct net_device *dev;
6745 netdevice_tracker dev_tracker;
6747 /* upper master flag, there can only be one master device per list */
6750 /* lookup ignore flag */
6753 /* counter for the number of times this device was added to us */
6756 /* private field for the users */
6759 struct list_head list;
6760 struct rcu_head rcu;
6763 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6764 struct list_head *adj_list)
6766 struct netdev_adjacent *adj;
6768 list_for_each_entry(adj, adj_list, list) {
6769 if (adj->dev == adj_dev)
6775 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6776 struct netdev_nested_priv *priv)
6778 struct net_device *dev = (struct net_device *)priv->data;
6780 return upper_dev == dev;
6784 * netdev_has_upper_dev - Check if device is linked to an upper device
6786 * @upper_dev: upper device to check
6788 * Find out if a device is linked to specified upper device and return true
6789 * in case it is. Note that this checks only immediate upper device,
6790 * not through a complete stack of devices. The caller must hold the RTNL lock.
6792 bool netdev_has_upper_dev(struct net_device *dev,
6793 struct net_device *upper_dev)
6795 struct netdev_nested_priv priv = {
6796 .data = (void *)upper_dev,
6801 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6804 EXPORT_SYMBOL(netdev_has_upper_dev);
6807 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6809 * @upper_dev: upper device to check
6811 * Find out if a device is linked to specified upper device and return true
6812 * in case it is. Note that this checks the entire upper device chain.
6813 * The caller must hold rcu lock.
6816 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6817 struct net_device *upper_dev)
6819 struct netdev_nested_priv priv = {
6820 .data = (void *)upper_dev,
6823 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6826 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6829 * netdev_has_any_upper_dev - Check if device is linked to some device
6832 * Find out if a device is linked to an upper device and return true in case
6833 * it is. The caller must hold the RTNL lock.
6835 bool netdev_has_any_upper_dev(struct net_device *dev)
6839 return !list_empty(&dev->adj_list.upper);
6841 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6844 * netdev_master_upper_dev_get - Get master upper device
6847 * Find a master upper device and return pointer to it or NULL in case
6848 * it's not there. The caller must hold the RTNL lock.
6850 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6852 struct netdev_adjacent *upper;
6856 if (list_empty(&dev->adj_list.upper))
6859 upper = list_first_entry(&dev->adj_list.upper,
6860 struct netdev_adjacent, list);
6861 if (likely(upper->master))
6865 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6867 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6869 struct netdev_adjacent *upper;
6873 if (list_empty(&dev->adj_list.upper))
6876 upper = list_first_entry(&dev->adj_list.upper,
6877 struct netdev_adjacent, list);
6878 if (likely(upper->master) && !upper->ignore)
6884 * netdev_has_any_lower_dev - Check if device is linked to some device
6887 * Find out if a device is linked to a lower device and return true in case
6888 * it is. The caller must hold the RTNL lock.
6890 static bool netdev_has_any_lower_dev(struct net_device *dev)
6894 return !list_empty(&dev->adj_list.lower);
6897 void *netdev_adjacent_get_private(struct list_head *adj_list)
6899 struct netdev_adjacent *adj;
6901 adj = list_entry(adj_list, struct netdev_adjacent, list);
6903 return adj->private;
6905 EXPORT_SYMBOL(netdev_adjacent_get_private);
6908 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6910 * @iter: list_head ** of the current position
6912 * Gets the next device from the dev's upper list, starting from iter
6913 * position. The caller must hold RCU read lock.
6915 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6916 struct list_head **iter)
6918 struct netdev_adjacent *upper;
6920 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6922 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6924 if (&upper->list == &dev->adj_list.upper)
6927 *iter = &upper->list;
6931 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6933 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6934 struct list_head **iter,
6937 struct netdev_adjacent *upper;
6939 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6941 if (&upper->list == &dev->adj_list.upper)
6944 *iter = &upper->list;
6945 *ignore = upper->ignore;
6950 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6951 struct list_head **iter)
6953 struct netdev_adjacent *upper;
6955 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6957 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6959 if (&upper->list == &dev->adj_list.upper)
6962 *iter = &upper->list;
6967 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6968 int (*fn)(struct net_device *dev,
6969 struct netdev_nested_priv *priv),
6970 struct netdev_nested_priv *priv)
6972 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6973 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6978 iter = &dev->adj_list.upper;
6982 ret = fn(now, priv);
6989 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6996 niter = &udev->adj_list.upper;
6997 dev_stack[cur] = now;
6998 iter_stack[cur++] = iter;
7005 next = dev_stack[--cur];
7006 niter = iter_stack[cur];
7016 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7017 int (*fn)(struct net_device *dev,
7018 struct netdev_nested_priv *priv),
7019 struct netdev_nested_priv *priv)
7021 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7022 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7026 iter = &dev->adj_list.upper;
7030 ret = fn(now, priv);
7037 udev = netdev_next_upper_dev_rcu(now, &iter);
7042 niter = &udev->adj_list.upper;
7043 dev_stack[cur] = now;
7044 iter_stack[cur++] = iter;
7051 next = dev_stack[--cur];
7052 niter = iter_stack[cur];
7061 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7063 static bool __netdev_has_upper_dev(struct net_device *dev,
7064 struct net_device *upper_dev)
7066 struct netdev_nested_priv priv = {
7068 .data = (void *)upper_dev,
7073 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7078 * netdev_lower_get_next_private - Get the next ->private from the
7079 * lower neighbour list
7081 * @iter: list_head ** of the current position
7083 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7084 * list, starting from iter position. The caller must hold either hold the
7085 * RTNL lock or its own locking that guarantees that the neighbour lower
7086 * list will remain unchanged.
7088 void *netdev_lower_get_next_private(struct net_device *dev,
7089 struct list_head **iter)
7091 struct netdev_adjacent *lower;
7093 lower = list_entry(*iter, struct netdev_adjacent, list);
7095 if (&lower->list == &dev->adj_list.lower)
7098 *iter = lower->list.next;
7100 return lower->private;
7102 EXPORT_SYMBOL(netdev_lower_get_next_private);
7105 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7106 * lower neighbour list, RCU
7109 * @iter: list_head ** of the current position
7111 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7112 * list, starting from iter position. The caller must hold RCU read lock.
7114 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7115 struct list_head **iter)
7117 struct netdev_adjacent *lower;
7119 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7121 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7123 if (&lower->list == &dev->adj_list.lower)
7126 *iter = &lower->list;
7128 return lower->private;
7130 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7133 * netdev_lower_get_next - Get the next device from the lower neighbour
7136 * @iter: list_head ** of the current position
7138 * Gets the next netdev_adjacent from the dev's lower neighbour
7139 * list, starting from iter position. The caller must hold RTNL lock or
7140 * its own locking that guarantees that the neighbour lower
7141 * list will remain unchanged.
7143 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7145 struct netdev_adjacent *lower;
7147 lower = list_entry(*iter, struct netdev_adjacent, list);
7149 if (&lower->list == &dev->adj_list.lower)
7152 *iter = lower->list.next;
7156 EXPORT_SYMBOL(netdev_lower_get_next);
7158 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7159 struct list_head **iter)
7161 struct netdev_adjacent *lower;
7163 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7165 if (&lower->list == &dev->adj_list.lower)
7168 *iter = &lower->list;
7173 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7174 struct list_head **iter,
7177 struct netdev_adjacent *lower;
7179 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7181 if (&lower->list == &dev->adj_list.lower)
7184 *iter = &lower->list;
7185 *ignore = lower->ignore;
7190 int netdev_walk_all_lower_dev(struct net_device *dev,
7191 int (*fn)(struct net_device *dev,
7192 struct netdev_nested_priv *priv),
7193 struct netdev_nested_priv *priv)
7195 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7196 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7200 iter = &dev->adj_list.lower;
7204 ret = fn(now, priv);
7211 ldev = netdev_next_lower_dev(now, &iter);
7216 niter = &ldev->adj_list.lower;
7217 dev_stack[cur] = now;
7218 iter_stack[cur++] = iter;
7225 next = dev_stack[--cur];
7226 niter = iter_stack[cur];
7235 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7237 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7238 int (*fn)(struct net_device *dev,
7239 struct netdev_nested_priv *priv),
7240 struct netdev_nested_priv *priv)
7242 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7243 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7248 iter = &dev->adj_list.lower;
7252 ret = fn(now, priv);
7259 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7266 niter = &ldev->adj_list.lower;
7267 dev_stack[cur] = now;
7268 iter_stack[cur++] = iter;
7275 next = dev_stack[--cur];
7276 niter = iter_stack[cur];
7286 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7287 struct list_head **iter)
7289 struct netdev_adjacent *lower;
7291 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7292 if (&lower->list == &dev->adj_list.lower)
7295 *iter = &lower->list;
7299 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7301 static u8 __netdev_upper_depth(struct net_device *dev)
7303 struct net_device *udev;
7304 struct list_head *iter;
7308 for (iter = &dev->adj_list.upper,
7309 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7311 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7314 if (max_depth < udev->upper_level)
7315 max_depth = udev->upper_level;
7321 static u8 __netdev_lower_depth(struct net_device *dev)
7323 struct net_device *ldev;
7324 struct list_head *iter;
7328 for (iter = &dev->adj_list.lower,
7329 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7331 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7334 if (max_depth < ldev->lower_level)
7335 max_depth = ldev->lower_level;
7341 static int __netdev_update_upper_level(struct net_device *dev,
7342 struct netdev_nested_priv *__unused)
7344 dev->upper_level = __netdev_upper_depth(dev) + 1;
7348 #ifdef CONFIG_LOCKDEP
7349 static LIST_HEAD(net_unlink_list);
7351 static void net_unlink_todo(struct net_device *dev)
7353 if (list_empty(&dev->unlink_list))
7354 list_add_tail(&dev->unlink_list, &net_unlink_list);
7358 static int __netdev_update_lower_level(struct net_device *dev,
7359 struct netdev_nested_priv *priv)
7361 dev->lower_level = __netdev_lower_depth(dev) + 1;
7363 #ifdef CONFIG_LOCKDEP
7367 if (priv->flags & NESTED_SYNC_IMM)
7368 dev->nested_level = dev->lower_level - 1;
7369 if (priv->flags & NESTED_SYNC_TODO)
7370 net_unlink_todo(dev);
7375 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7376 int (*fn)(struct net_device *dev,
7377 struct netdev_nested_priv *priv),
7378 struct netdev_nested_priv *priv)
7380 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7381 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7385 iter = &dev->adj_list.lower;
7389 ret = fn(now, priv);
7396 ldev = netdev_next_lower_dev_rcu(now, &iter);
7401 niter = &ldev->adj_list.lower;
7402 dev_stack[cur] = now;
7403 iter_stack[cur++] = iter;
7410 next = dev_stack[--cur];
7411 niter = iter_stack[cur];
7420 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7423 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7424 * lower neighbour list, RCU
7428 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7429 * list. The caller must hold RCU read lock.
7431 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7433 struct netdev_adjacent *lower;
7435 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7436 struct netdev_adjacent, list);
7438 return lower->private;
7441 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7444 * netdev_master_upper_dev_get_rcu - Get master upper device
7447 * Find a master upper device and return pointer to it or NULL in case
7448 * it's not there. The caller must hold the RCU read lock.
7450 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7452 struct netdev_adjacent *upper;
7454 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7455 struct netdev_adjacent, list);
7456 if (upper && likely(upper->master))
7460 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7462 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7463 struct net_device *adj_dev,
7464 struct list_head *dev_list)
7466 char linkname[IFNAMSIZ+7];
7468 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7469 "upper_%s" : "lower_%s", adj_dev->name);
7470 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7473 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7475 struct list_head *dev_list)
7477 char linkname[IFNAMSIZ+7];
7479 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7480 "upper_%s" : "lower_%s", name);
7481 sysfs_remove_link(&(dev->dev.kobj), linkname);
7484 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7485 struct net_device *adj_dev,
7486 struct list_head *dev_list)
7488 return (dev_list == &dev->adj_list.upper ||
7489 dev_list == &dev->adj_list.lower) &&
7490 net_eq(dev_net(dev), dev_net(adj_dev));
7493 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7494 struct net_device *adj_dev,
7495 struct list_head *dev_list,
7496 void *private, bool master)
7498 struct netdev_adjacent *adj;
7501 adj = __netdev_find_adj(adj_dev, dev_list);
7505 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7506 dev->name, adj_dev->name, adj->ref_nr);
7511 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7516 adj->master = master;
7518 adj->private = private;
7519 adj->ignore = false;
7520 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7522 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7523 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7525 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7526 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7531 /* Ensure that master link is always the first item in list. */
7533 ret = sysfs_create_link(&(dev->dev.kobj),
7534 &(adj_dev->dev.kobj), "master");
7536 goto remove_symlinks;
7538 list_add_rcu(&adj->list, dev_list);
7540 list_add_tail_rcu(&adj->list, dev_list);
7546 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7547 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7549 netdev_put(adj_dev, &adj->dev_tracker);
7555 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7556 struct net_device *adj_dev,
7558 struct list_head *dev_list)
7560 struct netdev_adjacent *adj;
7562 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7563 dev->name, adj_dev->name, ref_nr);
7565 adj = __netdev_find_adj(adj_dev, dev_list);
7568 pr_err("Adjacency does not exist for device %s from %s\n",
7569 dev->name, adj_dev->name);
7574 if (adj->ref_nr > ref_nr) {
7575 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7576 dev->name, adj_dev->name, ref_nr,
7577 adj->ref_nr - ref_nr);
7578 adj->ref_nr -= ref_nr;
7583 sysfs_remove_link(&(dev->dev.kobj), "master");
7585 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7586 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7588 list_del_rcu(&adj->list);
7589 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7590 adj_dev->name, dev->name, adj_dev->name);
7591 netdev_put(adj_dev, &adj->dev_tracker);
7592 kfree_rcu(adj, rcu);
7595 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7596 struct net_device *upper_dev,
7597 struct list_head *up_list,
7598 struct list_head *down_list,
7599 void *private, bool master)
7603 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7608 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7611 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7618 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7619 struct net_device *upper_dev,
7621 struct list_head *up_list,
7622 struct list_head *down_list)
7624 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7625 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7628 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7629 struct net_device *upper_dev,
7630 void *private, bool master)
7632 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7633 &dev->adj_list.upper,
7634 &upper_dev->adj_list.lower,
7638 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7639 struct net_device *upper_dev)
7641 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7642 &dev->adj_list.upper,
7643 &upper_dev->adj_list.lower);
7646 static int __netdev_upper_dev_link(struct net_device *dev,
7647 struct net_device *upper_dev, bool master,
7648 void *upper_priv, void *upper_info,
7649 struct netdev_nested_priv *priv,
7650 struct netlink_ext_ack *extack)
7652 struct netdev_notifier_changeupper_info changeupper_info = {
7657 .upper_dev = upper_dev,
7660 .upper_info = upper_info,
7662 struct net_device *master_dev;
7667 if (dev == upper_dev)
7670 /* To prevent loops, check if dev is not upper device to upper_dev. */
7671 if (__netdev_has_upper_dev(upper_dev, dev))
7674 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7678 if (__netdev_has_upper_dev(dev, upper_dev))
7681 master_dev = __netdev_master_upper_dev_get(dev);
7683 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7686 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7687 &changeupper_info.info);
7688 ret = notifier_to_errno(ret);
7692 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7697 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7698 &changeupper_info.info);
7699 ret = notifier_to_errno(ret);
7703 __netdev_update_upper_level(dev, NULL);
7704 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7706 __netdev_update_lower_level(upper_dev, priv);
7707 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7713 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7719 * netdev_upper_dev_link - Add a link to the upper device
7721 * @upper_dev: new upper device
7722 * @extack: netlink extended ack
7724 * Adds a link to device which is upper to this one. The caller must hold
7725 * the RTNL lock. On a failure a negative errno code is returned.
7726 * On success the reference counts are adjusted and the function
7729 int netdev_upper_dev_link(struct net_device *dev,
7730 struct net_device *upper_dev,
7731 struct netlink_ext_ack *extack)
7733 struct netdev_nested_priv priv = {
7734 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7738 return __netdev_upper_dev_link(dev, upper_dev, false,
7739 NULL, NULL, &priv, extack);
7741 EXPORT_SYMBOL(netdev_upper_dev_link);
7744 * netdev_master_upper_dev_link - Add a master link to the upper device
7746 * @upper_dev: new upper device
7747 * @upper_priv: upper device private
7748 * @upper_info: upper info to be passed down via notifier
7749 * @extack: netlink extended ack
7751 * Adds a link to device which is upper to this one. In this case, only
7752 * one master upper device can be linked, although other non-master devices
7753 * might be linked as well. The caller must hold the RTNL lock.
7754 * On a failure a negative errno code is returned. On success the reference
7755 * counts are adjusted and the function returns zero.
7757 int netdev_master_upper_dev_link(struct net_device *dev,
7758 struct net_device *upper_dev,
7759 void *upper_priv, void *upper_info,
7760 struct netlink_ext_ack *extack)
7762 struct netdev_nested_priv priv = {
7763 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7767 return __netdev_upper_dev_link(dev, upper_dev, true,
7768 upper_priv, upper_info, &priv, extack);
7770 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7772 static void __netdev_upper_dev_unlink(struct net_device *dev,
7773 struct net_device *upper_dev,
7774 struct netdev_nested_priv *priv)
7776 struct netdev_notifier_changeupper_info changeupper_info = {
7780 .upper_dev = upper_dev,
7786 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7788 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7789 &changeupper_info.info);
7791 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7793 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7794 &changeupper_info.info);
7796 __netdev_update_upper_level(dev, NULL);
7797 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7799 __netdev_update_lower_level(upper_dev, priv);
7800 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7805 * netdev_upper_dev_unlink - Removes a link to upper device
7807 * @upper_dev: new upper device
7809 * Removes a link to device which is upper to this one. The caller must hold
7812 void netdev_upper_dev_unlink(struct net_device *dev,
7813 struct net_device *upper_dev)
7815 struct netdev_nested_priv priv = {
7816 .flags = NESTED_SYNC_TODO,
7820 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7822 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7824 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7825 struct net_device *lower_dev,
7828 struct netdev_adjacent *adj;
7830 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7834 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7839 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7840 struct net_device *lower_dev)
7842 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7845 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7846 struct net_device *lower_dev)
7848 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7851 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7852 struct net_device *new_dev,
7853 struct net_device *dev,
7854 struct netlink_ext_ack *extack)
7856 struct netdev_nested_priv priv = {
7865 if (old_dev && new_dev != old_dev)
7866 netdev_adjacent_dev_disable(dev, old_dev);
7867 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7870 if (old_dev && new_dev != old_dev)
7871 netdev_adjacent_dev_enable(dev, old_dev);
7877 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7879 void netdev_adjacent_change_commit(struct net_device *old_dev,
7880 struct net_device *new_dev,
7881 struct net_device *dev)
7883 struct netdev_nested_priv priv = {
7884 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7888 if (!new_dev || !old_dev)
7891 if (new_dev == old_dev)
7894 netdev_adjacent_dev_enable(dev, old_dev);
7895 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7897 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7899 void netdev_adjacent_change_abort(struct net_device *old_dev,
7900 struct net_device *new_dev,
7901 struct net_device *dev)
7903 struct netdev_nested_priv priv = {
7911 if (old_dev && new_dev != old_dev)
7912 netdev_adjacent_dev_enable(dev, old_dev);
7914 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7916 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7919 * netdev_bonding_info_change - Dispatch event about slave change
7921 * @bonding_info: info to dispatch
7923 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7924 * The caller must hold the RTNL lock.
7926 void netdev_bonding_info_change(struct net_device *dev,
7927 struct netdev_bonding_info *bonding_info)
7929 struct netdev_notifier_bonding_info info = {
7933 memcpy(&info.bonding_info, bonding_info,
7934 sizeof(struct netdev_bonding_info));
7935 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7938 EXPORT_SYMBOL(netdev_bonding_info_change);
7940 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7941 struct netlink_ext_ack *extack)
7943 struct netdev_notifier_offload_xstats_info info = {
7945 .info.extack = extack,
7946 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7951 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7953 if (!dev->offload_xstats_l3)
7956 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7957 NETDEV_OFFLOAD_XSTATS_DISABLE,
7959 err = notifier_to_errno(rc);
7966 kfree(dev->offload_xstats_l3);
7967 dev->offload_xstats_l3 = NULL;
7971 int netdev_offload_xstats_enable(struct net_device *dev,
7972 enum netdev_offload_xstats_type type,
7973 struct netlink_ext_ack *extack)
7977 if (netdev_offload_xstats_enabled(dev, type))
7981 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7982 return netdev_offload_xstats_enable_l3(dev, extack);
7988 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7990 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7992 struct netdev_notifier_offload_xstats_info info = {
7994 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7997 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7999 kfree(dev->offload_xstats_l3);
8000 dev->offload_xstats_l3 = NULL;
8003 int netdev_offload_xstats_disable(struct net_device *dev,
8004 enum netdev_offload_xstats_type type)
8008 if (!netdev_offload_xstats_enabled(dev, type))
8012 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8013 netdev_offload_xstats_disable_l3(dev);
8020 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8022 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8024 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8027 static struct rtnl_hw_stats64 *
8028 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8029 enum netdev_offload_xstats_type type)
8032 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8033 return dev->offload_xstats_l3;
8040 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8041 enum netdev_offload_xstats_type type)
8045 return netdev_offload_xstats_get_ptr(dev, type);
8047 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8049 struct netdev_notifier_offload_xstats_ru {
8053 struct netdev_notifier_offload_xstats_rd {
8054 struct rtnl_hw_stats64 stats;
8058 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8059 const struct rtnl_hw_stats64 *src)
8061 dest->rx_packets += src->rx_packets;
8062 dest->tx_packets += src->tx_packets;
8063 dest->rx_bytes += src->rx_bytes;
8064 dest->tx_bytes += src->tx_bytes;
8065 dest->rx_errors += src->rx_errors;
8066 dest->tx_errors += src->tx_errors;
8067 dest->rx_dropped += src->rx_dropped;
8068 dest->tx_dropped += src->tx_dropped;
8069 dest->multicast += src->multicast;
8072 static int netdev_offload_xstats_get_used(struct net_device *dev,
8073 enum netdev_offload_xstats_type type,
8075 struct netlink_ext_ack *extack)
8077 struct netdev_notifier_offload_xstats_ru report_used = {};
8078 struct netdev_notifier_offload_xstats_info info = {
8080 .info.extack = extack,
8082 .report_used = &report_used,
8086 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8087 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8089 *p_used = report_used.used;
8090 return notifier_to_errno(rc);
8093 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8094 enum netdev_offload_xstats_type type,
8095 struct rtnl_hw_stats64 *p_stats,
8097 struct netlink_ext_ack *extack)
8099 struct netdev_notifier_offload_xstats_rd report_delta = {};
8100 struct netdev_notifier_offload_xstats_info info = {
8102 .info.extack = extack,
8104 .report_delta = &report_delta,
8106 struct rtnl_hw_stats64 *stats;
8109 stats = netdev_offload_xstats_get_ptr(dev, type);
8110 if (WARN_ON(!stats))
8113 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8116 /* Cache whatever we got, even if there was an error, otherwise the
8117 * successful stats retrievals would get lost.
8119 netdev_hw_stats64_add(stats, &report_delta.stats);
8123 *p_used = report_delta.used;
8125 return notifier_to_errno(rc);
8128 int netdev_offload_xstats_get(struct net_device *dev,
8129 enum netdev_offload_xstats_type type,
8130 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8131 struct netlink_ext_ack *extack)
8136 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8139 return netdev_offload_xstats_get_used(dev, type, p_used,
8142 EXPORT_SYMBOL(netdev_offload_xstats_get);
8145 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8146 const struct rtnl_hw_stats64 *stats)
8148 report_delta->used = true;
8149 netdev_hw_stats64_add(&report_delta->stats, stats);
8151 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8154 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8156 report_used->used = true;
8158 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8160 void netdev_offload_xstats_push_delta(struct net_device *dev,
8161 enum netdev_offload_xstats_type type,
8162 const struct rtnl_hw_stats64 *p_stats)
8164 struct rtnl_hw_stats64 *stats;
8168 stats = netdev_offload_xstats_get_ptr(dev, type);
8169 if (WARN_ON(!stats))
8172 netdev_hw_stats64_add(stats, p_stats);
8174 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8177 * netdev_get_xmit_slave - Get the xmit slave of master device
8180 * @all_slaves: assume all the slaves are active
8182 * The reference counters are not incremented so the caller must be
8183 * careful with locks. The caller must hold RCU lock.
8184 * %NULL is returned if no slave is found.
8187 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8188 struct sk_buff *skb,
8191 const struct net_device_ops *ops = dev->netdev_ops;
8193 if (!ops->ndo_get_xmit_slave)
8195 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8197 EXPORT_SYMBOL(netdev_get_xmit_slave);
8199 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8202 const struct net_device_ops *ops = dev->netdev_ops;
8204 if (!ops->ndo_sk_get_lower_dev)
8206 return ops->ndo_sk_get_lower_dev(dev, sk);
8210 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8214 * %NULL is returned if no lower device is found.
8217 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8220 struct net_device *lower;
8222 lower = netdev_sk_get_lower_dev(dev, sk);
8225 lower = netdev_sk_get_lower_dev(dev, sk);
8230 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8232 static void netdev_adjacent_add_links(struct net_device *dev)
8234 struct netdev_adjacent *iter;
8236 struct net *net = dev_net(dev);
8238 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8239 if (!net_eq(net, dev_net(iter->dev)))
8241 netdev_adjacent_sysfs_add(iter->dev, dev,
8242 &iter->dev->adj_list.lower);
8243 netdev_adjacent_sysfs_add(dev, iter->dev,
8244 &dev->adj_list.upper);
8247 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8248 if (!net_eq(net, dev_net(iter->dev)))
8250 netdev_adjacent_sysfs_add(iter->dev, dev,
8251 &iter->dev->adj_list.upper);
8252 netdev_adjacent_sysfs_add(dev, iter->dev,
8253 &dev->adj_list.lower);
8257 static void netdev_adjacent_del_links(struct net_device *dev)
8259 struct netdev_adjacent *iter;
8261 struct net *net = dev_net(dev);
8263 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8264 if (!net_eq(net, dev_net(iter->dev)))
8266 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8267 &iter->dev->adj_list.lower);
8268 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8269 &dev->adj_list.upper);
8272 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8273 if (!net_eq(net, dev_net(iter->dev)))
8275 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8276 &iter->dev->adj_list.upper);
8277 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8278 &dev->adj_list.lower);
8282 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8284 struct netdev_adjacent *iter;
8286 struct net *net = dev_net(dev);
8288 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8289 if (!net_eq(net, dev_net(iter->dev)))
8291 netdev_adjacent_sysfs_del(iter->dev, oldname,
8292 &iter->dev->adj_list.lower);
8293 netdev_adjacent_sysfs_add(iter->dev, dev,
8294 &iter->dev->adj_list.lower);
8297 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8298 if (!net_eq(net, dev_net(iter->dev)))
8300 netdev_adjacent_sysfs_del(iter->dev, oldname,
8301 &iter->dev->adj_list.upper);
8302 netdev_adjacent_sysfs_add(iter->dev, dev,
8303 &iter->dev->adj_list.upper);
8307 void *netdev_lower_dev_get_private(struct net_device *dev,
8308 struct net_device *lower_dev)
8310 struct netdev_adjacent *lower;
8314 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8318 return lower->private;
8320 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8324 * netdev_lower_state_changed - Dispatch event about lower device state change
8325 * @lower_dev: device
8326 * @lower_state_info: state to dispatch
8328 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8329 * The caller must hold the RTNL lock.
8331 void netdev_lower_state_changed(struct net_device *lower_dev,
8332 void *lower_state_info)
8334 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8335 .info.dev = lower_dev,
8339 changelowerstate_info.lower_state_info = lower_state_info;
8340 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8341 &changelowerstate_info.info);
8343 EXPORT_SYMBOL(netdev_lower_state_changed);
8345 static void dev_change_rx_flags(struct net_device *dev, int flags)
8347 const struct net_device_ops *ops = dev->netdev_ops;
8349 if (ops->ndo_change_rx_flags)
8350 ops->ndo_change_rx_flags(dev, flags);
8353 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8355 unsigned int old_flags = dev->flags;
8361 dev->flags |= IFF_PROMISC;
8362 dev->promiscuity += inc;
8363 if (dev->promiscuity == 0) {
8366 * If inc causes overflow, untouch promisc and return error.
8369 dev->flags &= ~IFF_PROMISC;
8371 dev->promiscuity -= inc;
8372 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8376 if (dev->flags != old_flags) {
8377 pr_info("device %s %s promiscuous mode\n",
8379 dev->flags & IFF_PROMISC ? "entered" : "left");
8380 if (audit_enabled) {
8381 current_uid_gid(&uid, &gid);
8382 audit_log(audit_context(), GFP_ATOMIC,
8383 AUDIT_ANOM_PROMISCUOUS,
8384 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8385 dev->name, (dev->flags & IFF_PROMISC),
8386 (old_flags & IFF_PROMISC),
8387 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8388 from_kuid(&init_user_ns, uid),
8389 from_kgid(&init_user_ns, gid),
8390 audit_get_sessionid(current));
8393 dev_change_rx_flags(dev, IFF_PROMISC);
8396 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8401 * dev_set_promiscuity - update promiscuity count on a device
8405 * Add or remove promiscuity from a device. While the count in the device
8406 * remains above zero the interface remains promiscuous. Once it hits zero
8407 * the device reverts back to normal filtering operation. A negative inc
8408 * value is used to drop promiscuity on the device.
8409 * Return 0 if successful or a negative errno code on error.
8411 int dev_set_promiscuity(struct net_device *dev, int inc)
8413 unsigned int old_flags = dev->flags;
8416 err = __dev_set_promiscuity(dev, inc, true);
8419 if (dev->flags != old_flags)
8420 dev_set_rx_mode(dev);
8423 EXPORT_SYMBOL(dev_set_promiscuity);
8425 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8427 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8431 dev->flags |= IFF_ALLMULTI;
8432 dev->allmulti += inc;
8433 if (dev->allmulti == 0) {
8436 * If inc causes overflow, untouch allmulti and return error.
8439 dev->flags &= ~IFF_ALLMULTI;
8441 dev->allmulti -= inc;
8442 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8446 if (dev->flags ^ old_flags) {
8447 dev_change_rx_flags(dev, IFF_ALLMULTI);
8448 dev_set_rx_mode(dev);
8450 __dev_notify_flags(dev, old_flags,
8451 dev->gflags ^ old_gflags);
8457 * dev_set_allmulti - update allmulti count on a device
8461 * Add or remove reception of all multicast frames to a device. While the
8462 * count in the device remains above zero the interface remains listening
8463 * to all interfaces. Once it hits zero the device reverts back to normal
8464 * filtering operation. A negative @inc value is used to drop the counter
8465 * when releasing a resource needing all multicasts.
8466 * Return 0 if successful or a negative errno code on error.
8469 int dev_set_allmulti(struct net_device *dev, int inc)
8471 return __dev_set_allmulti(dev, inc, true);
8473 EXPORT_SYMBOL(dev_set_allmulti);
8476 * Upload unicast and multicast address lists to device and
8477 * configure RX filtering. When the device doesn't support unicast
8478 * filtering it is put in promiscuous mode while unicast addresses
8481 void __dev_set_rx_mode(struct net_device *dev)
8483 const struct net_device_ops *ops = dev->netdev_ops;
8485 /* dev_open will call this function so the list will stay sane. */
8486 if (!(dev->flags&IFF_UP))
8489 if (!netif_device_present(dev))
8492 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8493 /* Unicast addresses changes may only happen under the rtnl,
8494 * therefore calling __dev_set_promiscuity here is safe.
8496 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8497 __dev_set_promiscuity(dev, 1, false);
8498 dev->uc_promisc = true;
8499 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8500 __dev_set_promiscuity(dev, -1, false);
8501 dev->uc_promisc = false;
8505 if (ops->ndo_set_rx_mode)
8506 ops->ndo_set_rx_mode(dev);
8509 void dev_set_rx_mode(struct net_device *dev)
8511 netif_addr_lock_bh(dev);
8512 __dev_set_rx_mode(dev);
8513 netif_addr_unlock_bh(dev);
8517 * dev_get_flags - get flags reported to userspace
8520 * Get the combination of flag bits exported through APIs to userspace.
8522 unsigned int dev_get_flags(const struct net_device *dev)
8526 flags = (dev->flags & ~(IFF_PROMISC |
8531 (dev->gflags & (IFF_PROMISC |
8534 if (netif_running(dev)) {
8535 if (netif_oper_up(dev))
8536 flags |= IFF_RUNNING;
8537 if (netif_carrier_ok(dev))
8538 flags |= IFF_LOWER_UP;
8539 if (netif_dormant(dev))
8540 flags |= IFF_DORMANT;
8545 EXPORT_SYMBOL(dev_get_flags);
8547 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8548 struct netlink_ext_ack *extack)
8550 unsigned int old_flags = dev->flags;
8556 * Set the flags on our device.
8559 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8560 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8562 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8566 * Load in the correct multicast list now the flags have changed.
8569 if ((old_flags ^ flags) & IFF_MULTICAST)
8570 dev_change_rx_flags(dev, IFF_MULTICAST);
8572 dev_set_rx_mode(dev);
8575 * Have we downed the interface. We handle IFF_UP ourselves
8576 * according to user attempts to set it, rather than blindly
8581 if ((old_flags ^ flags) & IFF_UP) {
8582 if (old_flags & IFF_UP)
8585 ret = __dev_open(dev, extack);
8588 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8589 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8590 unsigned int old_flags = dev->flags;
8592 dev->gflags ^= IFF_PROMISC;
8594 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8595 if (dev->flags != old_flags)
8596 dev_set_rx_mode(dev);
8599 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8600 * is important. Some (broken) drivers set IFF_PROMISC, when
8601 * IFF_ALLMULTI is requested not asking us and not reporting.
8603 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8604 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8606 dev->gflags ^= IFF_ALLMULTI;
8607 __dev_set_allmulti(dev, inc, false);
8613 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8614 unsigned int gchanges)
8616 unsigned int changes = dev->flags ^ old_flags;
8619 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8621 if (changes & IFF_UP) {
8622 if (dev->flags & IFF_UP)
8623 call_netdevice_notifiers(NETDEV_UP, dev);
8625 call_netdevice_notifiers(NETDEV_DOWN, dev);
8628 if (dev->flags & IFF_UP &&
8629 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8630 struct netdev_notifier_change_info change_info = {
8634 .flags_changed = changes,
8637 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8642 * dev_change_flags - change device settings
8644 * @flags: device state flags
8645 * @extack: netlink extended ack
8647 * Change settings on device based state flags. The flags are
8648 * in the userspace exported format.
8650 int dev_change_flags(struct net_device *dev, unsigned int flags,
8651 struct netlink_ext_ack *extack)
8654 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8656 ret = __dev_change_flags(dev, flags, extack);
8660 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8661 __dev_notify_flags(dev, old_flags, changes);
8664 EXPORT_SYMBOL(dev_change_flags);
8666 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8668 const struct net_device_ops *ops = dev->netdev_ops;
8670 if (ops->ndo_change_mtu)
8671 return ops->ndo_change_mtu(dev, new_mtu);
8673 /* Pairs with all the lockless reads of dev->mtu in the stack */
8674 WRITE_ONCE(dev->mtu, new_mtu);
8677 EXPORT_SYMBOL(__dev_set_mtu);
8679 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8680 struct netlink_ext_ack *extack)
8682 /* MTU must be positive, and in range */
8683 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8684 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8688 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8689 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8696 * dev_set_mtu_ext - Change maximum transfer unit
8698 * @new_mtu: new transfer unit
8699 * @extack: netlink extended ack
8701 * Change the maximum transfer size of the network device.
8703 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8704 struct netlink_ext_ack *extack)
8708 if (new_mtu == dev->mtu)
8711 err = dev_validate_mtu(dev, new_mtu, extack);
8715 if (!netif_device_present(dev))
8718 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8719 err = notifier_to_errno(err);
8723 orig_mtu = dev->mtu;
8724 err = __dev_set_mtu(dev, new_mtu);
8727 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8729 err = notifier_to_errno(err);
8731 /* setting mtu back and notifying everyone again,
8732 * so that they have a chance to revert changes.
8734 __dev_set_mtu(dev, orig_mtu);
8735 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8742 int dev_set_mtu(struct net_device *dev, int new_mtu)
8744 struct netlink_ext_ack extack;
8747 memset(&extack, 0, sizeof(extack));
8748 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8749 if (err && extack._msg)
8750 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8753 EXPORT_SYMBOL(dev_set_mtu);
8756 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8758 * @new_len: new tx queue length
8760 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8762 unsigned int orig_len = dev->tx_queue_len;
8765 if (new_len != (unsigned int)new_len)
8768 if (new_len != orig_len) {
8769 dev->tx_queue_len = new_len;
8770 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8771 res = notifier_to_errno(res);
8774 res = dev_qdisc_change_tx_queue_len(dev);
8782 netdev_err(dev, "refused to change device tx_queue_len\n");
8783 dev->tx_queue_len = orig_len;
8788 * dev_set_group - Change group this device belongs to
8790 * @new_group: group this device should belong to
8792 void dev_set_group(struct net_device *dev, int new_group)
8794 dev->group = new_group;
8798 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8800 * @addr: new address
8801 * @extack: netlink extended ack
8803 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8804 struct netlink_ext_ack *extack)
8806 struct netdev_notifier_pre_changeaddr_info info = {
8808 .info.extack = extack,
8813 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8814 return notifier_to_errno(rc);
8816 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8819 * dev_set_mac_address - Change Media Access Control Address
8822 * @extack: netlink extended ack
8824 * Change the hardware (MAC) address of the device
8826 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8827 struct netlink_ext_ack *extack)
8829 const struct net_device_ops *ops = dev->netdev_ops;
8832 if (!ops->ndo_set_mac_address)
8834 if (sa->sa_family != dev->type)
8836 if (!netif_device_present(dev))
8838 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8841 err = ops->ndo_set_mac_address(dev, sa);
8844 dev->addr_assign_type = NET_ADDR_SET;
8845 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8846 add_device_randomness(dev->dev_addr, dev->addr_len);
8849 EXPORT_SYMBOL(dev_set_mac_address);
8851 static DECLARE_RWSEM(dev_addr_sem);
8853 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8854 struct netlink_ext_ack *extack)
8858 down_write(&dev_addr_sem);
8859 ret = dev_set_mac_address(dev, sa, extack);
8860 up_write(&dev_addr_sem);
8863 EXPORT_SYMBOL(dev_set_mac_address_user);
8865 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8867 size_t size = sizeof(sa->sa_data_min);
8868 struct net_device *dev;
8871 down_read(&dev_addr_sem);
8874 dev = dev_get_by_name_rcu(net, dev_name);
8880 memset(sa->sa_data, 0, size);
8882 memcpy(sa->sa_data, dev->dev_addr,
8883 min_t(size_t, size, dev->addr_len));
8884 sa->sa_family = dev->type;
8888 up_read(&dev_addr_sem);
8891 EXPORT_SYMBOL(dev_get_mac_address);
8894 * dev_change_carrier - Change device carrier
8896 * @new_carrier: new value
8898 * Change device carrier
8900 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8902 const struct net_device_ops *ops = dev->netdev_ops;
8904 if (!ops->ndo_change_carrier)
8906 if (!netif_device_present(dev))
8908 return ops->ndo_change_carrier(dev, new_carrier);
8912 * dev_get_phys_port_id - Get device physical port ID
8916 * Get device physical port ID
8918 int dev_get_phys_port_id(struct net_device *dev,
8919 struct netdev_phys_item_id *ppid)
8921 const struct net_device_ops *ops = dev->netdev_ops;
8923 if (!ops->ndo_get_phys_port_id)
8925 return ops->ndo_get_phys_port_id(dev, ppid);
8929 * dev_get_phys_port_name - Get device physical port name
8932 * @len: limit of bytes to copy to name
8934 * Get device physical port name
8936 int dev_get_phys_port_name(struct net_device *dev,
8937 char *name, size_t len)
8939 const struct net_device_ops *ops = dev->netdev_ops;
8942 if (ops->ndo_get_phys_port_name) {
8943 err = ops->ndo_get_phys_port_name(dev, name, len);
8944 if (err != -EOPNOTSUPP)
8947 return devlink_compat_phys_port_name_get(dev, name, len);
8951 * dev_get_port_parent_id - Get the device's port parent identifier
8952 * @dev: network device
8953 * @ppid: pointer to a storage for the port's parent identifier
8954 * @recurse: allow/disallow recursion to lower devices
8956 * Get the devices's port parent identifier
8958 int dev_get_port_parent_id(struct net_device *dev,
8959 struct netdev_phys_item_id *ppid,
8962 const struct net_device_ops *ops = dev->netdev_ops;
8963 struct netdev_phys_item_id first = { };
8964 struct net_device *lower_dev;
8965 struct list_head *iter;
8968 if (ops->ndo_get_port_parent_id) {
8969 err = ops->ndo_get_port_parent_id(dev, ppid);
8970 if (err != -EOPNOTSUPP)
8974 err = devlink_compat_switch_id_get(dev, ppid);
8975 if (!recurse || err != -EOPNOTSUPP)
8978 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8979 err = dev_get_port_parent_id(lower_dev, ppid, true);
8984 else if (memcmp(&first, ppid, sizeof(*ppid)))
8990 EXPORT_SYMBOL(dev_get_port_parent_id);
8993 * netdev_port_same_parent_id - Indicate if two network devices have
8994 * the same port parent identifier
8995 * @a: first network device
8996 * @b: second network device
8998 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9000 struct netdev_phys_item_id a_id = { };
9001 struct netdev_phys_item_id b_id = { };
9003 if (dev_get_port_parent_id(a, &a_id, true) ||
9004 dev_get_port_parent_id(b, &b_id, true))
9007 return netdev_phys_item_id_same(&a_id, &b_id);
9009 EXPORT_SYMBOL(netdev_port_same_parent_id);
9012 * dev_change_proto_down - set carrier according to proto_down.
9015 * @proto_down: new value
9017 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9019 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
9021 if (!netif_device_present(dev))
9024 netif_carrier_off(dev);
9026 netif_carrier_on(dev);
9027 dev->proto_down = proto_down;
9032 * dev_change_proto_down_reason - proto down reason
9035 * @mask: proto down mask
9036 * @value: proto down value
9038 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9044 dev->proto_down_reason = value;
9046 for_each_set_bit(b, &mask, 32) {
9047 if (value & (1 << b))
9048 dev->proto_down_reason |= BIT(b);
9050 dev->proto_down_reason &= ~BIT(b);
9055 struct bpf_xdp_link {
9056 struct bpf_link link;
9057 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9061 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9063 if (flags & XDP_FLAGS_HW_MODE)
9065 if (flags & XDP_FLAGS_DRV_MODE)
9066 return XDP_MODE_DRV;
9067 if (flags & XDP_FLAGS_SKB_MODE)
9068 return XDP_MODE_SKB;
9069 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9072 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9076 return generic_xdp_install;
9079 return dev->netdev_ops->ndo_bpf;
9085 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9086 enum bpf_xdp_mode mode)
9088 return dev->xdp_state[mode].link;
9091 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9092 enum bpf_xdp_mode mode)
9094 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9097 return link->link.prog;
9098 return dev->xdp_state[mode].prog;
9101 u8 dev_xdp_prog_count(struct net_device *dev)
9106 for (i = 0; i < __MAX_XDP_MODE; i++)
9107 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9111 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9113 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9115 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9117 return prog ? prog->aux->id : 0;
9120 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9121 struct bpf_xdp_link *link)
9123 dev->xdp_state[mode].link = link;
9124 dev->xdp_state[mode].prog = NULL;
9127 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9128 struct bpf_prog *prog)
9130 dev->xdp_state[mode].link = NULL;
9131 dev->xdp_state[mode].prog = prog;
9134 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9135 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9136 u32 flags, struct bpf_prog *prog)
9138 struct netdev_bpf xdp;
9141 memset(&xdp, 0, sizeof(xdp));
9142 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9143 xdp.extack = extack;
9147 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9148 * "moved" into driver), so they don't increment it on their own, but
9149 * they do decrement refcnt when program is detached or replaced.
9150 * Given net_device also owns link/prog, we need to bump refcnt here
9151 * to prevent drivers from underflowing it.
9155 err = bpf_op(dev, &xdp);
9162 if (mode != XDP_MODE_HW)
9163 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9168 static void dev_xdp_uninstall(struct net_device *dev)
9170 struct bpf_xdp_link *link;
9171 struct bpf_prog *prog;
9172 enum bpf_xdp_mode mode;
9177 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9178 prog = dev_xdp_prog(dev, mode);
9182 bpf_op = dev_xdp_bpf_op(dev, mode);
9186 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9188 /* auto-detach link from net device */
9189 link = dev_xdp_link(dev, mode);
9195 dev_xdp_set_link(dev, mode, NULL);
9199 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9200 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9201 struct bpf_prog *old_prog, u32 flags)
9203 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9204 struct bpf_prog *cur_prog;
9205 struct net_device *upper;
9206 struct list_head *iter;
9207 enum bpf_xdp_mode mode;
9213 /* either link or prog attachment, never both */
9214 if (link && (new_prog || old_prog))
9216 /* link supports only XDP mode flags */
9217 if (link && (flags & ~XDP_FLAGS_MODES)) {
9218 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9221 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9222 if (num_modes > 1) {
9223 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9226 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9227 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9228 NL_SET_ERR_MSG(extack,
9229 "More than one program loaded, unset mode is ambiguous");
9232 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9233 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9234 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9238 mode = dev_xdp_mode(dev, flags);
9239 /* can't replace attached link */
9240 if (dev_xdp_link(dev, mode)) {
9241 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9245 /* don't allow if an upper device already has a program */
9246 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9247 if (dev_xdp_prog_count(upper) > 0) {
9248 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9253 cur_prog = dev_xdp_prog(dev, mode);
9254 /* can't replace attached prog with link */
9255 if (link && cur_prog) {
9256 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9259 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9260 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9264 /* put effective new program into new_prog */
9266 new_prog = link->link.prog;
9269 bool offload = mode == XDP_MODE_HW;
9270 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9271 ? XDP_MODE_DRV : XDP_MODE_SKB;
9273 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9274 NL_SET_ERR_MSG(extack, "XDP program already attached");
9277 if (!offload && dev_xdp_prog(dev, other_mode)) {
9278 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9281 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9282 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9285 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9286 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9289 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9290 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9295 /* don't call drivers if the effective program didn't change */
9296 if (new_prog != cur_prog) {
9297 bpf_op = dev_xdp_bpf_op(dev, mode);
9299 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9303 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9309 dev_xdp_set_link(dev, mode, link);
9311 dev_xdp_set_prog(dev, mode, new_prog);
9313 bpf_prog_put(cur_prog);
9318 static int dev_xdp_attach_link(struct net_device *dev,
9319 struct netlink_ext_ack *extack,
9320 struct bpf_xdp_link *link)
9322 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9325 static int dev_xdp_detach_link(struct net_device *dev,
9326 struct netlink_ext_ack *extack,
9327 struct bpf_xdp_link *link)
9329 enum bpf_xdp_mode mode;
9334 mode = dev_xdp_mode(dev, link->flags);
9335 if (dev_xdp_link(dev, mode) != link)
9338 bpf_op = dev_xdp_bpf_op(dev, mode);
9339 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9340 dev_xdp_set_link(dev, mode, NULL);
9344 static void bpf_xdp_link_release(struct bpf_link *link)
9346 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9350 /* if racing with net_device's tear down, xdp_link->dev might be
9351 * already NULL, in which case link was already auto-detached
9353 if (xdp_link->dev) {
9354 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9355 xdp_link->dev = NULL;
9361 static int bpf_xdp_link_detach(struct bpf_link *link)
9363 bpf_xdp_link_release(link);
9367 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9369 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9374 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9375 struct seq_file *seq)
9377 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9382 ifindex = xdp_link->dev->ifindex;
9385 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9388 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9389 struct bpf_link_info *info)
9391 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9396 ifindex = xdp_link->dev->ifindex;
9399 info->xdp.ifindex = ifindex;
9403 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9404 struct bpf_prog *old_prog)
9406 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9407 enum bpf_xdp_mode mode;
9413 /* link might have been auto-released already, so fail */
9414 if (!xdp_link->dev) {
9419 if (old_prog && link->prog != old_prog) {
9423 old_prog = link->prog;
9424 if (old_prog->type != new_prog->type ||
9425 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9430 if (old_prog == new_prog) {
9431 /* no-op, don't disturb drivers */
9432 bpf_prog_put(new_prog);
9436 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9437 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9438 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9439 xdp_link->flags, new_prog);
9443 old_prog = xchg(&link->prog, new_prog);
9444 bpf_prog_put(old_prog);
9451 static const struct bpf_link_ops bpf_xdp_link_lops = {
9452 .release = bpf_xdp_link_release,
9453 .dealloc = bpf_xdp_link_dealloc,
9454 .detach = bpf_xdp_link_detach,
9455 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9456 .fill_link_info = bpf_xdp_link_fill_link_info,
9457 .update_prog = bpf_xdp_link_update,
9460 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9462 struct net *net = current->nsproxy->net_ns;
9463 struct bpf_link_primer link_primer;
9464 struct bpf_xdp_link *link;
9465 struct net_device *dev;
9469 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9475 link = kzalloc(sizeof(*link), GFP_USER);
9481 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9483 link->flags = attr->link_create.flags;
9485 err = bpf_link_prime(&link->link, &link_primer);
9491 err = dev_xdp_attach_link(dev, NULL, link);
9496 bpf_link_cleanup(&link_primer);
9500 fd = bpf_link_settle(&link_primer);
9501 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9514 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9516 * @extack: netlink extended ack
9517 * @fd: new program fd or negative value to clear
9518 * @expected_fd: old program fd that userspace expects to replace or clear
9519 * @flags: xdp-related flags
9521 * Set or clear a bpf program for a device
9523 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9524 int fd, int expected_fd, u32 flags)
9526 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9527 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9533 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9534 mode != XDP_MODE_SKB);
9535 if (IS_ERR(new_prog))
9536 return PTR_ERR(new_prog);
9539 if (expected_fd >= 0) {
9540 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9541 mode != XDP_MODE_SKB);
9542 if (IS_ERR(old_prog)) {
9543 err = PTR_ERR(old_prog);
9549 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9552 if (err && new_prog)
9553 bpf_prog_put(new_prog);
9555 bpf_prog_put(old_prog);
9560 * dev_new_index - allocate an ifindex
9561 * @net: the applicable net namespace
9563 * Returns a suitable unique value for a new device interface
9564 * number. The caller must hold the rtnl semaphore or the
9565 * dev_base_lock to be sure it remains unique.
9567 static int dev_new_index(struct net *net)
9569 int ifindex = net->ifindex;
9574 if (!__dev_get_by_index(net, ifindex))
9575 return net->ifindex = ifindex;
9579 /* Delayed registration/unregisteration */
9580 LIST_HEAD(net_todo_list);
9581 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9583 static void net_set_todo(struct net_device *dev)
9585 list_add_tail(&dev->todo_list, &net_todo_list);
9586 atomic_inc(&dev_net(dev)->dev_unreg_count);
9589 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9590 struct net_device *upper, netdev_features_t features)
9592 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9593 netdev_features_t feature;
9596 for_each_netdev_feature(upper_disables, feature_bit) {
9597 feature = __NETIF_F_BIT(feature_bit);
9598 if (!(upper->wanted_features & feature)
9599 && (features & feature)) {
9600 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9601 &feature, upper->name);
9602 features &= ~feature;
9609 static void netdev_sync_lower_features(struct net_device *upper,
9610 struct net_device *lower, netdev_features_t features)
9612 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9613 netdev_features_t feature;
9616 for_each_netdev_feature(upper_disables, feature_bit) {
9617 feature = __NETIF_F_BIT(feature_bit);
9618 if (!(features & feature) && (lower->features & feature)) {
9619 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9620 &feature, lower->name);
9621 lower->wanted_features &= ~feature;
9622 __netdev_update_features(lower);
9624 if (unlikely(lower->features & feature))
9625 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9626 &feature, lower->name);
9628 netdev_features_change(lower);
9633 static netdev_features_t netdev_fix_features(struct net_device *dev,
9634 netdev_features_t features)
9636 /* Fix illegal checksum combinations */
9637 if ((features & NETIF_F_HW_CSUM) &&
9638 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9639 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9640 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9643 /* TSO requires that SG is present as well. */
9644 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9645 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9646 features &= ~NETIF_F_ALL_TSO;
9649 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9650 !(features & NETIF_F_IP_CSUM)) {
9651 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9652 features &= ~NETIF_F_TSO;
9653 features &= ~NETIF_F_TSO_ECN;
9656 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9657 !(features & NETIF_F_IPV6_CSUM)) {
9658 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9659 features &= ~NETIF_F_TSO6;
9662 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9663 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9664 features &= ~NETIF_F_TSO_MANGLEID;
9666 /* TSO ECN requires that TSO is present as well. */
9667 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9668 features &= ~NETIF_F_TSO_ECN;
9670 /* Software GSO depends on SG. */
9671 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9672 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9673 features &= ~NETIF_F_GSO;
9676 /* GSO partial features require GSO partial be set */
9677 if ((features & dev->gso_partial_features) &&
9678 !(features & NETIF_F_GSO_PARTIAL)) {
9680 "Dropping partially supported GSO features since no GSO partial.\n");
9681 features &= ~dev->gso_partial_features;
9684 if (!(features & NETIF_F_RXCSUM)) {
9685 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9686 * successfully merged by hardware must also have the
9687 * checksum verified by hardware. If the user does not
9688 * want to enable RXCSUM, logically, we should disable GRO_HW.
9690 if (features & NETIF_F_GRO_HW) {
9691 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9692 features &= ~NETIF_F_GRO_HW;
9696 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9697 if (features & NETIF_F_RXFCS) {
9698 if (features & NETIF_F_LRO) {
9699 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9700 features &= ~NETIF_F_LRO;
9703 if (features & NETIF_F_GRO_HW) {
9704 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9705 features &= ~NETIF_F_GRO_HW;
9709 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9710 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9711 features &= ~NETIF_F_LRO;
9714 if (features & NETIF_F_HW_TLS_TX) {
9715 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9716 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9717 bool hw_csum = features & NETIF_F_HW_CSUM;
9719 if (!ip_csum && !hw_csum) {
9720 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9721 features &= ~NETIF_F_HW_TLS_TX;
9725 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9726 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9727 features &= ~NETIF_F_HW_TLS_RX;
9733 int __netdev_update_features(struct net_device *dev)
9735 struct net_device *upper, *lower;
9736 netdev_features_t features;
9737 struct list_head *iter;
9742 features = netdev_get_wanted_features(dev);
9744 if (dev->netdev_ops->ndo_fix_features)
9745 features = dev->netdev_ops->ndo_fix_features(dev, features);
9747 /* driver might be less strict about feature dependencies */
9748 features = netdev_fix_features(dev, features);
9750 /* some features can't be enabled if they're off on an upper device */
9751 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9752 features = netdev_sync_upper_features(dev, upper, features);
9754 if (dev->features == features)
9757 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9758 &dev->features, &features);
9760 if (dev->netdev_ops->ndo_set_features)
9761 err = dev->netdev_ops->ndo_set_features(dev, features);
9765 if (unlikely(err < 0)) {
9767 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9768 err, &features, &dev->features);
9769 /* return non-0 since some features might have changed and
9770 * it's better to fire a spurious notification than miss it
9776 /* some features must be disabled on lower devices when disabled
9777 * on an upper device (think: bonding master or bridge)
9779 netdev_for_each_lower_dev(dev, lower, iter)
9780 netdev_sync_lower_features(dev, lower, features);
9783 netdev_features_t diff = features ^ dev->features;
9785 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9786 /* udp_tunnel_{get,drop}_rx_info both need
9787 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9788 * device, or they won't do anything.
9789 * Thus we need to update dev->features
9790 * *before* calling udp_tunnel_get_rx_info,
9791 * but *after* calling udp_tunnel_drop_rx_info.
9793 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9794 dev->features = features;
9795 udp_tunnel_get_rx_info(dev);
9797 udp_tunnel_drop_rx_info(dev);
9801 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9802 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9803 dev->features = features;
9804 err |= vlan_get_rx_ctag_filter_info(dev);
9806 vlan_drop_rx_ctag_filter_info(dev);
9810 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9811 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9812 dev->features = features;
9813 err |= vlan_get_rx_stag_filter_info(dev);
9815 vlan_drop_rx_stag_filter_info(dev);
9819 dev->features = features;
9822 return err < 0 ? 0 : 1;
9826 * netdev_update_features - recalculate device features
9827 * @dev: the device to check
9829 * Recalculate dev->features set and send notifications if it
9830 * has changed. Should be called after driver or hardware dependent
9831 * conditions might have changed that influence the features.
9833 void netdev_update_features(struct net_device *dev)
9835 if (__netdev_update_features(dev))
9836 netdev_features_change(dev);
9838 EXPORT_SYMBOL(netdev_update_features);
9841 * netdev_change_features - recalculate device features
9842 * @dev: the device to check
9844 * Recalculate dev->features set and send notifications even
9845 * if they have not changed. Should be called instead of
9846 * netdev_update_features() if also dev->vlan_features might
9847 * have changed to allow the changes to be propagated to stacked
9850 void netdev_change_features(struct net_device *dev)
9852 __netdev_update_features(dev);
9853 netdev_features_change(dev);
9855 EXPORT_SYMBOL(netdev_change_features);
9858 * netif_stacked_transfer_operstate - transfer operstate
9859 * @rootdev: the root or lower level device to transfer state from
9860 * @dev: the device to transfer operstate to
9862 * Transfer operational state from root to device. This is normally
9863 * called when a stacking relationship exists between the root
9864 * device and the device(a leaf device).
9866 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9867 struct net_device *dev)
9869 if (rootdev->operstate == IF_OPER_DORMANT)
9870 netif_dormant_on(dev);
9872 netif_dormant_off(dev);
9874 if (rootdev->operstate == IF_OPER_TESTING)
9875 netif_testing_on(dev);
9877 netif_testing_off(dev);
9879 if (netif_carrier_ok(rootdev))
9880 netif_carrier_on(dev);
9882 netif_carrier_off(dev);
9884 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9886 static int netif_alloc_rx_queues(struct net_device *dev)
9888 unsigned int i, count = dev->num_rx_queues;
9889 struct netdev_rx_queue *rx;
9890 size_t sz = count * sizeof(*rx);
9895 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9901 for (i = 0; i < count; i++) {
9904 /* XDP RX-queue setup */
9905 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9912 /* Rollback successful reg's and free other resources */
9914 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9920 static void netif_free_rx_queues(struct net_device *dev)
9922 unsigned int i, count = dev->num_rx_queues;
9924 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9928 for (i = 0; i < count; i++)
9929 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9934 static void netdev_init_one_queue(struct net_device *dev,
9935 struct netdev_queue *queue, void *_unused)
9937 /* Initialize queue lock */
9938 spin_lock_init(&queue->_xmit_lock);
9939 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9940 queue->xmit_lock_owner = -1;
9941 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9944 dql_init(&queue->dql, HZ);
9948 static void netif_free_tx_queues(struct net_device *dev)
9953 static int netif_alloc_netdev_queues(struct net_device *dev)
9955 unsigned int count = dev->num_tx_queues;
9956 struct netdev_queue *tx;
9957 size_t sz = count * sizeof(*tx);
9959 if (count < 1 || count > 0xffff)
9962 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9968 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9969 spin_lock_init(&dev->tx_global_lock);
9974 void netif_tx_stop_all_queues(struct net_device *dev)
9978 for (i = 0; i < dev->num_tx_queues; i++) {
9979 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9981 netif_tx_stop_queue(txq);
9984 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9987 * register_netdevice() - register a network device
9988 * @dev: device to register
9990 * Take a prepared network device structure and make it externally accessible.
9991 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9992 * Callers must hold the rtnl lock - you may want register_netdev()
9995 int register_netdevice(struct net_device *dev)
9998 struct net *net = dev_net(dev);
10000 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10001 NETDEV_FEATURE_COUNT);
10002 BUG_ON(dev_boot_phase);
10007 /* When net_device's are persistent, this will be fatal. */
10008 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10011 ret = ethtool_check_ops(dev->ethtool_ops);
10015 spin_lock_init(&dev->addr_list_lock);
10016 netdev_set_addr_lockdep_class(dev);
10018 ret = dev_get_valid_name(net, dev, dev->name);
10023 dev->name_node = netdev_name_node_head_alloc(dev);
10024 if (!dev->name_node)
10027 /* Init, if this function is available */
10028 if (dev->netdev_ops->ndo_init) {
10029 ret = dev->netdev_ops->ndo_init(dev);
10033 goto err_free_name;
10037 if (((dev->hw_features | dev->features) &
10038 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10039 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10040 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10041 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10048 dev->ifindex = dev_new_index(net);
10049 else if (__dev_get_by_index(net, dev->ifindex))
10052 /* Transfer changeable features to wanted_features and enable
10053 * software offloads (GSO and GRO).
10055 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10056 dev->features |= NETIF_F_SOFT_FEATURES;
10058 if (dev->udp_tunnel_nic_info) {
10059 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10060 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10063 dev->wanted_features = dev->features & dev->hw_features;
10065 if (!(dev->flags & IFF_LOOPBACK))
10066 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10068 /* If IPv4 TCP segmentation offload is supported we should also
10069 * allow the device to enable segmenting the frame with the option
10070 * of ignoring a static IP ID value. This doesn't enable the
10071 * feature itself but allows the user to enable it later.
10073 if (dev->hw_features & NETIF_F_TSO)
10074 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10075 if (dev->vlan_features & NETIF_F_TSO)
10076 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10077 if (dev->mpls_features & NETIF_F_TSO)
10078 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10079 if (dev->hw_enc_features & NETIF_F_TSO)
10080 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10082 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10084 dev->vlan_features |= NETIF_F_HIGHDMA;
10086 /* Make NETIF_F_SG inheritable to tunnel devices.
10088 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10090 /* Make NETIF_F_SG inheritable to MPLS.
10092 dev->mpls_features |= NETIF_F_SG;
10094 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10095 ret = notifier_to_errno(ret);
10099 ret = netdev_register_kobject(dev);
10100 write_lock(&dev_base_lock);
10101 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10102 write_unlock(&dev_base_lock);
10106 __netdev_update_features(dev);
10109 * Default initial state at registry is that the
10110 * device is present.
10113 set_bit(__LINK_STATE_PRESENT, &dev->state);
10115 linkwatch_init_dev(dev);
10117 dev_init_scheduler(dev);
10119 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10120 list_netdevice(dev);
10122 add_device_randomness(dev->dev_addr, dev->addr_len);
10124 /* If the device has permanent device address, driver should
10125 * set dev_addr and also addr_assign_type should be set to
10126 * NET_ADDR_PERM (default value).
10128 if (dev->addr_assign_type == NET_ADDR_PERM)
10129 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10131 /* Notify protocols, that a new device appeared. */
10132 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10133 ret = notifier_to_errno(ret);
10135 /* Expect explicit free_netdev() on failure */
10136 dev->needs_free_netdev = false;
10137 unregister_netdevice_queue(dev, NULL);
10141 * Prevent userspace races by waiting until the network
10142 * device is fully setup before sending notifications.
10144 if (!dev->rtnl_link_ops ||
10145 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10146 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10152 if (dev->netdev_ops->ndo_uninit)
10153 dev->netdev_ops->ndo_uninit(dev);
10154 if (dev->priv_destructor)
10155 dev->priv_destructor(dev);
10157 netdev_name_node_free(dev->name_node);
10160 EXPORT_SYMBOL(register_netdevice);
10163 * init_dummy_netdev - init a dummy network device for NAPI
10164 * @dev: device to init
10166 * This takes a network device structure and initialize the minimum
10167 * amount of fields so it can be used to schedule NAPI polls without
10168 * registering a full blown interface. This is to be used by drivers
10169 * that need to tie several hardware interfaces to a single NAPI
10170 * poll scheduler due to HW limitations.
10172 int init_dummy_netdev(struct net_device *dev)
10174 /* Clear everything. Note we don't initialize spinlocks
10175 * are they aren't supposed to be taken by any of the
10176 * NAPI code and this dummy netdev is supposed to be
10177 * only ever used for NAPI polls
10179 memset(dev, 0, sizeof(struct net_device));
10181 /* make sure we BUG if trying to hit standard
10182 * register/unregister code path
10184 dev->reg_state = NETREG_DUMMY;
10186 /* NAPI wants this */
10187 INIT_LIST_HEAD(&dev->napi_list);
10189 /* a dummy interface is started by default */
10190 set_bit(__LINK_STATE_PRESENT, &dev->state);
10191 set_bit(__LINK_STATE_START, &dev->state);
10193 /* napi_busy_loop stats accounting wants this */
10194 dev_net_set(dev, &init_net);
10196 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10197 * because users of this 'device' dont need to change
10203 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10207 * register_netdev - register a network device
10208 * @dev: device to register
10210 * Take a completed network device structure and add it to the kernel
10211 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10212 * chain. 0 is returned on success. A negative errno code is returned
10213 * on a failure to set up the device, or if the name is a duplicate.
10215 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10216 * and expands the device name if you passed a format string to
10219 int register_netdev(struct net_device *dev)
10223 if (rtnl_lock_killable())
10225 err = register_netdevice(dev);
10229 EXPORT_SYMBOL(register_netdev);
10231 int netdev_refcnt_read(const struct net_device *dev)
10233 #ifdef CONFIG_PCPU_DEV_REFCNT
10236 for_each_possible_cpu(i)
10237 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10240 return refcount_read(&dev->dev_refcnt);
10243 EXPORT_SYMBOL(netdev_refcnt_read);
10245 int netdev_unregister_timeout_secs __read_mostly = 10;
10247 #define WAIT_REFS_MIN_MSECS 1
10248 #define WAIT_REFS_MAX_MSECS 250
10250 * netdev_wait_allrefs_any - wait until all references are gone.
10251 * @list: list of net_devices to wait on
10253 * This is called when unregistering network devices.
10255 * Any protocol or device that holds a reference should register
10256 * for netdevice notification, and cleanup and put back the
10257 * reference if they receive an UNREGISTER event.
10258 * We can get stuck here if buggy protocols don't correctly
10261 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10263 unsigned long rebroadcast_time, warning_time;
10264 struct net_device *dev;
10267 rebroadcast_time = warning_time = jiffies;
10269 list_for_each_entry(dev, list, todo_list)
10270 if (netdev_refcnt_read(dev) == 1)
10274 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10277 /* Rebroadcast unregister notification */
10278 list_for_each_entry(dev, list, todo_list)
10279 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10285 list_for_each_entry(dev, list, todo_list)
10286 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10288 /* We must not have linkwatch events
10289 * pending on unregister. If this
10290 * happens, we simply run the queue
10291 * unscheduled, resulting in a noop
10294 linkwatch_run_queue();
10300 rebroadcast_time = jiffies;
10305 wait = WAIT_REFS_MIN_MSECS;
10308 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10311 list_for_each_entry(dev, list, todo_list)
10312 if (netdev_refcnt_read(dev) == 1)
10315 if (time_after(jiffies, warning_time +
10316 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10317 list_for_each_entry(dev, list, todo_list) {
10318 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10319 dev->name, netdev_refcnt_read(dev));
10320 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10323 warning_time = jiffies;
10328 /* The sequence is:
10332 * register_netdevice(x1);
10333 * register_netdevice(x2);
10335 * unregister_netdevice(y1);
10336 * unregister_netdevice(y2);
10342 * We are invoked by rtnl_unlock().
10343 * This allows us to deal with problems:
10344 * 1) We can delete sysfs objects which invoke hotplug
10345 * without deadlocking with linkwatch via keventd.
10346 * 2) Since we run with the RTNL semaphore not held, we can sleep
10347 * safely in order to wait for the netdev refcnt to drop to zero.
10349 * We must not return until all unregister events added during
10350 * the interval the lock was held have been completed.
10352 void netdev_run_todo(void)
10354 struct net_device *dev, *tmp;
10355 struct list_head list;
10356 #ifdef CONFIG_LOCKDEP
10357 struct list_head unlink_list;
10359 list_replace_init(&net_unlink_list, &unlink_list);
10361 while (!list_empty(&unlink_list)) {
10362 struct net_device *dev = list_first_entry(&unlink_list,
10365 list_del_init(&dev->unlink_list);
10366 dev->nested_level = dev->lower_level - 1;
10370 /* Snapshot list, allow later requests */
10371 list_replace_init(&net_todo_list, &list);
10375 /* Wait for rcu callbacks to finish before next phase */
10376 if (!list_empty(&list))
10379 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10380 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10381 netdev_WARN(dev, "run_todo but not unregistering\n");
10382 list_del(&dev->todo_list);
10386 write_lock(&dev_base_lock);
10387 dev->reg_state = NETREG_UNREGISTERED;
10388 write_unlock(&dev_base_lock);
10389 linkwatch_forget_dev(dev);
10392 while (!list_empty(&list)) {
10393 dev = netdev_wait_allrefs_any(&list);
10394 list_del(&dev->todo_list);
10397 BUG_ON(netdev_refcnt_read(dev) != 1);
10398 BUG_ON(!list_empty(&dev->ptype_all));
10399 BUG_ON(!list_empty(&dev->ptype_specific));
10400 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10401 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10403 if (dev->priv_destructor)
10404 dev->priv_destructor(dev);
10405 if (dev->needs_free_netdev)
10408 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10409 wake_up(&netdev_unregistering_wq);
10411 /* Free network device */
10412 kobject_put(&dev->dev.kobj);
10416 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10417 * all the same fields in the same order as net_device_stats, with only
10418 * the type differing, but rtnl_link_stats64 may have additional fields
10419 * at the end for newer counters.
10421 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10422 const struct net_device_stats *netdev_stats)
10424 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10425 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10426 u64 *dst = (u64 *)stats64;
10428 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10429 for (i = 0; i < n; i++)
10430 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10431 /* zero out counters that only exist in rtnl_link_stats64 */
10432 memset((char *)stats64 + n * sizeof(u64), 0,
10433 sizeof(*stats64) - n * sizeof(u64));
10435 EXPORT_SYMBOL(netdev_stats_to_stats64);
10437 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10439 struct net_device_core_stats __percpu *p;
10441 p = alloc_percpu_gfp(struct net_device_core_stats,
10442 GFP_ATOMIC | __GFP_NOWARN);
10444 if (p && cmpxchg(&dev->core_stats, NULL, p))
10447 /* This READ_ONCE() pairs with the cmpxchg() above */
10448 return READ_ONCE(dev->core_stats);
10450 EXPORT_SYMBOL(netdev_core_stats_alloc);
10453 * dev_get_stats - get network device statistics
10454 * @dev: device to get statistics from
10455 * @storage: place to store stats
10457 * Get network statistics from device. Return @storage.
10458 * The device driver may provide its own method by setting
10459 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10460 * otherwise the internal statistics structure is used.
10462 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10463 struct rtnl_link_stats64 *storage)
10465 const struct net_device_ops *ops = dev->netdev_ops;
10466 const struct net_device_core_stats __percpu *p;
10468 if (ops->ndo_get_stats64) {
10469 memset(storage, 0, sizeof(*storage));
10470 ops->ndo_get_stats64(dev, storage);
10471 } else if (ops->ndo_get_stats) {
10472 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10474 netdev_stats_to_stats64(storage, &dev->stats);
10477 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10478 p = READ_ONCE(dev->core_stats);
10480 const struct net_device_core_stats *core_stats;
10483 for_each_possible_cpu(i) {
10484 core_stats = per_cpu_ptr(p, i);
10485 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10486 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10487 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10488 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10493 EXPORT_SYMBOL(dev_get_stats);
10496 * dev_fetch_sw_netstats - get per-cpu network device statistics
10497 * @s: place to store stats
10498 * @netstats: per-cpu network stats to read from
10500 * Read per-cpu network statistics and populate the related fields in @s.
10502 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10503 const struct pcpu_sw_netstats __percpu *netstats)
10507 for_each_possible_cpu(cpu) {
10508 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10509 const struct pcpu_sw_netstats *stats;
10510 unsigned int start;
10512 stats = per_cpu_ptr(netstats, cpu);
10514 start = u64_stats_fetch_begin_irq(&stats->syncp);
10515 rx_packets = u64_stats_read(&stats->rx_packets);
10516 rx_bytes = u64_stats_read(&stats->rx_bytes);
10517 tx_packets = u64_stats_read(&stats->tx_packets);
10518 tx_bytes = u64_stats_read(&stats->tx_bytes);
10519 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10521 s->rx_packets += rx_packets;
10522 s->rx_bytes += rx_bytes;
10523 s->tx_packets += tx_packets;
10524 s->tx_bytes += tx_bytes;
10527 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10530 * dev_get_tstats64 - ndo_get_stats64 implementation
10531 * @dev: device to get statistics from
10532 * @s: place to store stats
10534 * Populate @s from dev->stats and dev->tstats. Can be used as
10535 * ndo_get_stats64() callback.
10537 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10539 netdev_stats_to_stats64(s, &dev->stats);
10540 dev_fetch_sw_netstats(s, dev->tstats);
10542 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10544 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10546 struct netdev_queue *queue = dev_ingress_queue(dev);
10548 #ifdef CONFIG_NET_CLS_ACT
10551 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10554 netdev_init_one_queue(dev, queue, NULL);
10555 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10556 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10557 rcu_assign_pointer(dev->ingress_queue, queue);
10562 static const struct ethtool_ops default_ethtool_ops;
10564 void netdev_set_default_ethtool_ops(struct net_device *dev,
10565 const struct ethtool_ops *ops)
10567 if (dev->ethtool_ops == &default_ethtool_ops)
10568 dev->ethtool_ops = ops;
10570 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10572 void netdev_freemem(struct net_device *dev)
10574 char *addr = (char *)dev - dev->padded;
10580 * alloc_netdev_mqs - allocate network device
10581 * @sizeof_priv: size of private data to allocate space for
10582 * @name: device name format string
10583 * @name_assign_type: origin of device name
10584 * @setup: callback to initialize device
10585 * @txqs: the number of TX subqueues to allocate
10586 * @rxqs: the number of RX subqueues to allocate
10588 * Allocates a struct net_device with private data area for driver use
10589 * and performs basic initialization. Also allocates subqueue structs
10590 * for each queue on the device.
10592 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10593 unsigned char name_assign_type,
10594 void (*setup)(struct net_device *),
10595 unsigned int txqs, unsigned int rxqs)
10597 struct net_device *dev;
10598 unsigned int alloc_size;
10599 struct net_device *p;
10601 BUG_ON(strlen(name) >= sizeof(dev->name));
10604 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10609 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10613 alloc_size = sizeof(struct net_device);
10615 /* ensure 32-byte alignment of private area */
10616 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10617 alloc_size += sizeof_priv;
10619 /* ensure 32-byte alignment of whole construct */
10620 alloc_size += NETDEV_ALIGN - 1;
10622 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10626 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10627 dev->padded = (char *)dev - (char *)p;
10629 ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10630 #ifdef CONFIG_PCPU_DEV_REFCNT
10631 dev->pcpu_refcnt = alloc_percpu(int);
10632 if (!dev->pcpu_refcnt)
10636 refcount_set(&dev->dev_refcnt, 1);
10639 if (dev_addr_init(dev))
10645 dev_net_set(dev, &init_net);
10647 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10648 dev->gso_max_segs = GSO_MAX_SEGS;
10649 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10650 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10651 dev->tso_max_segs = TSO_MAX_SEGS;
10652 dev->upper_level = 1;
10653 dev->lower_level = 1;
10654 #ifdef CONFIG_LOCKDEP
10655 dev->nested_level = 0;
10656 INIT_LIST_HEAD(&dev->unlink_list);
10659 INIT_LIST_HEAD(&dev->napi_list);
10660 INIT_LIST_HEAD(&dev->unreg_list);
10661 INIT_LIST_HEAD(&dev->close_list);
10662 INIT_LIST_HEAD(&dev->link_watch_list);
10663 INIT_LIST_HEAD(&dev->adj_list.upper);
10664 INIT_LIST_HEAD(&dev->adj_list.lower);
10665 INIT_LIST_HEAD(&dev->ptype_all);
10666 INIT_LIST_HEAD(&dev->ptype_specific);
10667 INIT_LIST_HEAD(&dev->net_notifier_list);
10668 #ifdef CONFIG_NET_SCHED
10669 hash_init(dev->qdisc_hash);
10671 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10674 if (!dev->tx_queue_len) {
10675 dev->priv_flags |= IFF_NO_QUEUE;
10676 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10679 dev->num_tx_queues = txqs;
10680 dev->real_num_tx_queues = txqs;
10681 if (netif_alloc_netdev_queues(dev))
10684 dev->num_rx_queues = rxqs;
10685 dev->real_num_rx_queues = rxqs;
10686 if (netif_alloc_rx_queues(dev))
10689 strcpy(dev->name, name);
10690 dev->name_assign_type = name_assign_type;
10691 dev->group = INIT_NETDEV_GROUP;
10692 if (!dev->ethtool_ops)
10693 dev->ethtool_ops = &default_ethtool_ops;
10695 nf_hook_netdev_init(dev);
10704 #ifdef CONFIG_PCPU_DEV_REFCNT
10705 free_percpu(dev->pcpu_refcnt);
10708 netdev_freemem(dev);
10711 EXPORT_SYMBOL(alloc_netdev_mqs);
10714 * free_netdev - free network device
10717 * This function does the last stage of destroying an allocated device
10718 * interface. The reference to the device object is released. If this
10719 * is the last reference then it will be freed.Must be called in process
10722 void free_netdev(struct net_device *dev)
10724 struct napi_struct *p, *n;
10728 /* When called immediately after register_netdevice() failed the unwind
10729 * handling may still be dismantling the device. Handle that case by
10730 * deferring the free.
10732 if (dev->reg_state == NETREG_UNREGISTERING) {
10734 dev->needs_free_netdev = true;
10738 netif_free_tx_queues(dev);
10739 netif_free_rx_queues(dev);
10741 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10743 /* Flush device addresses */
10744 dev_addr_flush(dev);
10746 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10749 ref_tracker_dir_exit(&dev->refcnt_tracker);
10750 #ifdef CONFIG_PCPU_DEV_REFCNT
10751 free_percpu(dev->pcpu_refcnt);
10752 dev->pcpu_refcnt = NULL;
10754 free_percpu(dev->core_stats);
10755 dev->core_stats = NULL;
10756 free_percpu(dev->xdp_bulkq);
10757 dev->xdp_bulkq = NULL;
10759 /* Compatibility with error handling in drivers */
10760 if (dev->reg_state == NETREG_UNINITIALIZED) {
10761 netdev_freemem(dev);
10765 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10766 dev->reg_state = NETREG_RELEASED;
10768 /* will free via device release */
10769 put_device(&dev->dev);
10771 EXPORT_SYMBOL(free_netdev);
10774 * synchronize_net - Synchronize with packet receive processing
10776 * Wait for packets currently being received to be done.
10777 * Does not block later packets from starting.
10779 void synchronize_net(void)
10782 if (rtnl_is_locked())
10783 synchronize_rcu_expedited();
10787 EXPORT_SYMBOL(synchronize_net);
10790 * unregister_netdevice_queue - remove device from the kernel
10794 * This function shuts down a device interface and removes it
10795 * from the kernel tables.
10796 * If head not NULL, device is queued to be unregistered later.
10798 * Callers must hold the rtnl semaphore. You may want
10799 * unregister_netdev() instead of this.
10802 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10807 list_move_tail(&dev->unreg_list, head);
10811 list_add(&dev->unreg_list, &single);
10812 unregister_netdevice_many(&single);
10815 EXPORT_SYMBOL(unregister_netdevice_queue);
10818 * unregister_netdevice_many - unregister many devices
10819 * @head: list of devices
10821 * Note: As most callers use a stack allocated list_head,
10822 * we force a list_del() to make sure stack wont be corrupted later.
10824 void unregister_netdevice_many(struct list_head *head)
10826 struct net_device *dev, *tmp;
10827 LIST_HEAD(close_head);
10829 BUG_ON(dev_boot_phase);
10832 if (list_empty(head))
10835 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10836 /* Some devices call without registering
10837 * for initialization unwind. Remove those
10838 * devices and proceed with the remaining.
10840 if (dev->reg_state == NETREG_UNINITIALIZED) {
10841 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10845 list_del(&dev->unreg_list);
10848 dev->dismantle = true;
10849 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10852 /* If device is running, close it first. */
10853 list_for_each_entry(dev, head, unreg_list)
10854 list_add_tail(&dev->close_list, &close_head);
10855 dev_close_many(&close_head, true);
10857 list_for_each_entry(dev, head, unreg_list) {
10858 /* And unlink it from device chain. */
10859 write_lock(&dev_base_lock);
10860 unlist_netdevice(dev, false);
10861 dev->reg_state = NETREG_UNREGISTERING;
10862 write_unlock(&dev_base_lock);
10864 flush_all_backlogs();
10868 list_for_each_entry(dev, head, unreg_list) {
10869 struct sk_buff *skb = NULL;
10871 /* Shutdown queueing discipline. */
10874 dev_xdp_uninstall(dev);
10876 netdev_offload_xstats_disable_all(dev);
10878 /* Notify protocols, that we are about to destroy
10879 * this device. They should clean all the things.
10881 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10883 if (!dev->rtnl_link_ops ||
10884 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10885 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10886 GFP_KERNEL, NULL, 0);
10889 * Flush the unicast and multicast chains
10894 netdev_name_node_alt_flush(dev);
10895 netdev_name_node_free(dev->name_node);
10897 if (dev->netdev_ops->ndo_uninit)
10898 dev->netdev_ops->ndo_uninit(dev);
10901 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10903 /* Notifier chain MUST detach us all upper devices. */
10904 WARN_ON(netdev_has_any_upper_dev(dev));
10905 WARN_ON(netdev_has_any_lower_dev(dev));
10907 /* Remove entries from kobject tree */
10908 netdev_unregister_kobject(dev);
10910 /* Remove XPS queueing entries */
10911 netif_reset_xps_queues_gt(dev, 0);
10917 list_for_each_entry(dev, head, unreg_list) {
10918 netdev_put(dev, &dev->dev_registered_tracker);
10924 EXPORT_SYMBOL(unregister_netdevice_many);
10927 * unregister_netdev - remove device from the kernel
10930 * This function shuts down a device interface and removes it
10931 * from the kernel tables.
10933 * This is just a wrapper for unregister_netdevice that takes
10934 * the rtnl semaphore. In general you want to use this and not
10935 * unregister_netdevice.
10937 void unregister_netdev(struct net_device *dev)
10940 unregister_netdevice(dev);
10943 EXPORT_SYMBOL(unregister_netdev);
10946 * __dev_change_net_namespace - move device to different nethost namespace
10948 * @net: network namespace
10949 * @pat: If not NULL name pattern to try if the current device name
10950 * is already taken in the destination network namespace.
10951 * @new_ifindex: If not zero, specifies device index in the target
10954 * This function shuts down a device interface and moves it
10955 * to a new network namespace. On success 0 is returned, on
10956 * a failure a netagive errno code is returned.
10958 * Callers must hold the rtnl semaphore.
10961 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10962 const char *pat, int new_ifindex)
10964 struct netdev_name_node *name_node;
10965 struct net *net_old = dev_net(dev);
10966 char new_name[IFNAMSIZ] = {};
10971 /* Don't allow namespace local devices to be moved. */
10973 if (dev->features & NETIF_F_NETNS_LOCAL)
10976 /* Ensure the device has been registrered */
10977 if (dev->reg_state != NETREG_REGISTERED)
10980 /* Get out if there is nothing todo */
10982 if (net_eq(net_old, net))
10985 /* Pick the destination device name, and ensure
10986 * we can use it in the destination network namespace.
10989 if (netdev_name_in_use(net, dev->name)) {
10990 /* We get here if we can't use the current device name */
10993 err = dev_prep_valid_name(net, dev, pat, new_name);
10997 /* Check that none of the altnames conflicts. */
10999 netdev_for_each_altname(dev, name_node)
11000 if (netdev_name_in_use(net, name_node->name))
11003 /* Check that new_ifindex isn't used yet. */
11005 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11009 * And now a mini version of register_netdevice unregister_netdevice.
11012 /* If device is running close it first. */
11015 /* And unlink it from device chain */
11016 unlist_netdevice(dev, true);
11020 /* Shutdown queueing discipline. */
11023 /* Notify protocols, that we are about to destroy
11024 * this device. They should clean all the things.
11026 * Note that dev->reg_state stays at NETREG_REGISTERED.
11027 * This is wanted because this way 8021q and macvlan know
11028 * the device is just moving and can keep their slaves up.
11030 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11033 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11034 /* If there is an ifindex conflict assign a new one */
11035 if (!new_ifindex) {
11036 if (__dev_get_by_index(net, dev->ifindex))
11037 new_ifindex = dev_new_index(net);
11039 new_ifindex = dev->ifindex;
11042 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11046 * Flush the unicast and multicast chains
11051 /* Send a netdev-removed uevent to the old namespace */
11052 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11053 netdev_adjacent_del_links(dev);
11055 /* Move per-net netdevice notifiers that are following the netdevice */
11056 move_netdevice_notifiers_dev_net(dev, net);
11058 /* Actually switch the network namespace */
11059 dev_net_set(dev, net);
11060 dev->ifindex = new_ifindex;
11062 /* Send a netdev-add uevent to the new namespace */
11063 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11064 netdev_adjacent_add_links(dev);
11066 if (new_name[0]) /* Rename the netdev to prepared name */
11067 strscpy(dev->name, new_name, IFNAMSIZ);
11069 /* Fixup kobjects */
11070 err = device_rename(&dev->dev, dev->name);
11073 /* Adapt owner in case owning user namespace of target network
11074 * namespace is different from the original one.
11076 err = netdev_change_owner(dev, net_old, net);
11079 /* Add the device back in the hashes */
11080 list_netdevice(dev);
11082 /* Notify protocols, that a new device appeared. */
11083 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11086 * Prevent userspace races by waiting until the network
11087 * device is fully setup before sending notifications.
11089 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11096 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11098 static int dev_cpu_dead(unsigned int oldcpu)
11100 struct sk_buff **list_skb;
11101 struct sk_buff *skb;
11103 struct softnet_data *sd, *oldsd, *remsd = NULL;
11105 local_irq_disable();
11106 cpu = smp_processor_id();
11107 sd = &per_cpu(softnet_data, cpu);
11108 oldsd = &per_cpu(softnet_data, oldcpu);
11110 /* Find end of our completion_queue. */
11111 list_skb = &sd->completion_queue;
11113 list_skb = &(*list_skb)->next;
11114 /* Append completion queue from offline CPU. */
11115 *list_skb = oldsd->completion_queue;
11116 oldsd->completion_queue = NULL;
11118 /* Append output queue from offline CPU. */
11119 if (oldsd->output_queue) {
11120 *sd->output_queue_tailp = oldsd->output_queue;
11121 sd->output_queue_tailp = oldsd->output_queue_tailp;
11122 oldsd->output_queue = NULL;
11123 oldsd->output_queue_tailp = &oldsd->output_queue;
11125 /* Append NAPI poll list from offline CPU, with one exception :
11126 * process_backlog() must be called by cpu owning percpu backlog.
11127 * We properly handle process_queue & input_pkt_queue later.
11129 while (!list_empty(&oldsd->poll_list)) {
11130 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11131 struct napi_struct,
11134 list_del_init(&napi->poll_list);
11135 if (napi->poll == process_backlog)
11138 ____napi_schedule(sd, napi);
11141 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11142 local_irq_enable();
11145 remsd = oldsd->rps_ipi_list;
11146 oldsd->rps_ipi_list = NULL;
11148 /* send out pending IPI's on offline CPU */
11149 net_rps_send_ipi(remsd);
11151 /* Process offline CPU's input_pkt_queue */
11152 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11154 input_queue_head_incr(oldsd);
11156 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11158 input_queue_head_incr(oldsd);
11165 * netdev_increment_features - increment feature set by one
11166 * @all: current feature set
11167 * @one: new feature set
11168 * @mask: mask feature set
11170 * Computes a new feature set after adding a device with feature set
11171 * @one to the master device with current feature set @all. Will not
11172 * enable anything that is off in @mask. Returns the new feature set.
11174 netdev_features_t netdev_increment_features(netdev_features_t all,
11175 netdev_features_t one, netdev_features_t mask)
11177 if (mask & NETIF_F_HW_CSUM)
11178 mask |= NETIF_F_CSUM_MASK;
11179 mask |= NETIF_F_VLAN_CHALLENGED;
11181 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11182 all &= one | ~NETIF_F_ALL_FOR_ALL;
11184 /* If one device supports hw checksumming, set for all. */
11185 if (all & NETIF_F_HW_CSUM)
11186 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11190 EXPORT_SYMBOL(netdev_increment_features);
11192 static struct hlist_head * __net_init netdev_create_hash(void)
11195 struct hlist_head *hash;
11197 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11199 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11200 INIT_HLIST_HEAD(&hash[i]);
11205 /* Initialize per network namespace state */
11206 static int __net_init netdev_init(struct net *net)
11208 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11209 8 * sizeof_field(struct napi_struct, gro_bitmask));
11211 INIT_LIST_HEAD(&net->dev_base_head);
11213 net->dev_name_head = netdev_create_hash();
11214 if (net->dev_name_head == NULL)
11217 net->dev_index_head = netdev_create_hash();
11218 if (net->dev_index_head == NULL)
11221 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11226 kfree(net->dev_name_head);
11232 * netdev_drivername - network driver for the device
11233 * @dev: network device
11235 * Determine network driver for device.
11237 const char *netdev_drivername(const struct net_device *dev)
11239 const struct device_driver *driver;
11240 const struct device *parent;
11241 const char *empty = "";
11243 parent = dev->dev.parent;
11247 driver = parent->driver;
11248 if (driver && driver->name)
11249 return driver->name;
11253 static void __netdev_printk(const char *level, const struct net_device *dev,
11254 struct va_format *vaf)
11256 if (dev && dev->dev.parent) {
11257 dev_printk_emit(level[1] - '0',
11260 dev_driver_string(dev->dev.parent),
11261 dev_name(dev->dev.parent),
11262 netdev_name(dev), netdev_reg_state(dev),
11265 printk("%s%s%s: %pV",
11266 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11268 printk("%s(NULL net_device): %pV", level, vaf);
11272 void netdev_printk(const char *level, const struct net_device *dev,
11273 const char *format, ...)
11275 struct va_format vaf;
11278 va_start(args, format);
11283 __netdev_printk(level, dev, &vaf);
11287 EXPORT_SYMBOL(netdev_printk);
11289 #define define_netdev_printk_level(func, level) \
11290 void func(const struct net_device *dev, const char *fmt, ...) \
11292 struct va_format vaf; \
11295 va_start(args, fmt); \
11300 __netdev_printk(level, dev, &vaf); \
11304 EXPORT_SYMBOL(func);
11306 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11307 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11308 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11309 define_netdev_printk_level(netdev_err, KERN_ERR);
11310 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11311 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11312 define_netdev_printk_level(netdev_info, KERN_INFO);
11314 static void __net_exit netdev_exit(struct net *net)
11316 kfree(net->dev_name_head);
11317 kfree(net->dev_index_head);
11318 if (net != &init_net)
11319 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11322 static struct pernet_operations __net_initdata netdev_net_ops = {
11323 .init = netdev_init,
11324 .exit = netdev_exit,
11327 static void __net_exit default_device_exit_net(struct net *net)
11329 struct netdev_name_node *name_node, *tmp;
11330 struct net_device *dev, *aux;
11332 * Push all migratable network devices back to the
11333 * initial network namespace
11336 for_each_netdev_safe(net, dev, aux) {
11338 char fb_name[IFNAMSIZ];
11340 /* Ignore unmoveable devices (i.e. loopback) */
11341 if (dev->features & NETIF_F_NETNS_LOCAL)
11344 /* Leave virtual devices for the generic cleanup */
11345 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11348 /* Push remaining network devices to init_net */
11349 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11350 if (netdev_name_in_use(&init_net, fb_name))
11351 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11353 netdev_for_each_altname_safe(dev, name_node, tmp)
11354 if (netdev_name_in_use(&init_net, name_node->name)) {
11355 netdev_name_node_del(name_node);
11357 __netdev_name_node_alt_destroy(name_node);
11360 err = dev_change_net_namespace(dev, &init_net, fb_name);
11362 pr_emerg("%s: failed to move %s to init_net: %d\n",
11363 __func__, dev->name, err);
11369 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11371 /* At exit all network devices most be removed from a network
11372 * namespace. Do this in the reverse order of registration.
11373 * Do this across as many network namespaces as possible to
11374 * improve batching efficiency.
11376 struct net_device *dev;
11378 LIST_HEAD(dev_kill_list);
11381 list_for_each_entry(net, net_list, exit_list) {
11382 default_device_exit_net(net);
11386 list_for_each_entry(net, net_list, exit_list) {
11387 for_each_netdev_reverse(net, dev) {
11388 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11389 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11391 unregister_netdevice_queue(dev, &dev_kill_list);
11394 unregister_netdevice_many(&dev_kill_list);
11398 static struct pernet_operations __net_initdata default_device_ops = {
11399 .exit_batch = default_device_exit_batch,
11403 * Initialize the DEV module. At boot time this walks the device list and
11404 * unhooks any devices that fail to initialise (normally hardware not
11405 * present) and leaves us with a valid list of present and active devices.
11410 * This is called single threaded during boot, so no need
11411 * to take the rtnl semaphore.
11413 static int __init net_dev_init(void)
11415 int i, rc = -ENOMEM;
11417 BUG_ON(!dev_boot_phase);
11419 if (dev_proc_init())
11422 if (netdev_kobject_init())
11425 INIT_LIST_HEAD(&ptype_all);
11426 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11427 INIT_LIST_HEAD(&ptype_base[i]);
11429 if (register_pernet_subsys(&netdev_net_ops))
11433 * Initialise the packet receive queues.
11436 for_each_possible_cpu(i) {
11437 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11438 struct softnet_data *sd = &per_cpu(softnet_data, i);
11440 INIT_WORK(flush, flush_backlog);
11442 skb_queue_head_init(&sd->input_pkt_queue);
11443 skb_queue_head_init(&sd->process_queue);
11444 #ifdef CONFIG_XFRM_OFFLOAD
11445 skb_queue_head_init(&sd->xfrm_backlog);
11447 INIT_LIST_HEAD(&sd->poll_list);
11448 sd->output_queue_tailp = &sd->output_queue;
11450 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11453 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11454 spin_lock_init(&sd->defer_lock);
11456 init_gro_hash(&sd->backlog);
11457 sd->backlog.poll = process_backlog;
11458 sd->backlog.weight = weight_p;
11461 dev_boot_phase = 0;
11463 /* The loopback device is special if any other network devices
11464 * is present in a network namespace the loopback device must
11465 * be present. Since we now dynamically allocate and free the
11466 * loopback device ensure this invariant is maintained by
11467 * keeping the loopback device as the first device on the
11468 * list of network devices. Ensuring the loopback devices
11469 * is the first device that appears and the last network device
11472 if (register_pernet_device(&loopback_net_ops))
11475 if (register_pernet_device(&default_device_ops))
11478 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11479 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11481 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11482 NULL, dev_cpu_dead);
11489 subsys_initcall(net_dev_init);