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 netdev_name_node_del(name_node);
349 kfree(name_node->name);
350 netdev_name_node_free(name_node);
353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
355 struct netdev_name_node *name_node;
356 struct net *net = dev_net(dev);
358 name_node = netdev_name_node_lookup(net, name);
361 /* lookup might have found our primary name or a name belonging
364 if (name_node == dev->name_node || name_node->dev != dev)
367 __netdev_name_node_alt_destroy(name_node);
372 static void netdev_name_node_alt_flush(struct net_device *dev)
374 struct netdev_name_node *name_node, *tmp;
376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
377 __netdev_name_node_alt_destroy(name_node);
380 /* Device list insertion */
381 static void list_netdevice(struct net_device *dev)
383 struct net *net = dev_net(dev);
387 write_lock(&dev_base_lock);
388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
389 netdev_name_node_add(net, dev->name_node);
390 hlist_add_head_rcu(&dev->index_hlist,
391 dev_index_hash(net, dev->ifindex));
392 write_unlock(&dev_base_lock);
394 dev_base_seq_inc(net);
397 /* Device list removal
398 * caller must respect a RCU grace period before freeing/reusing dev
400 static void unlist_netdevice(struct net_device *dev, bool lock)
404 /* Unlink dev from the device chain */
406 write_lock(&dev_base_lock);
407 list_del_rcu(&dev->dev_list);
408 netdev_name_node_del(dev->name_node);
409 hlist_del_rcu(&dev->index_hlist);
411 write_unlock(&dev_base_lock);
413 dev_base_seq_inc(dev_net(dev));
420 static RAW_NOTIFIER_HEAD(netdev_chain);
423 * Device drivers call our routines to queue packets here. We empty the
424 * queue in the local softnet handler.
427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
428 EXPORT_PER_CPU_SYMBOL(softnet_data);
430 #ifdef CONFIG_LOCKDEP
432 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
433 * according to dev->type
435 static const unsigned short netdev_lock_type[] = {
436 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
437 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
438 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
439 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
440 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
441 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
442 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
443 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
444 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
445 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
446 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
447 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
448 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
449 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
450 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
452 static const char *const netdev_lock_name[] = {
453 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
454 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
455 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
456 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
457 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
458 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
459 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
460 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
461 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
462 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
463 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
464 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
465 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
466 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
467 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
472 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
476 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
477 if (netdev_lock_type[i] == dev_type)
479 /* the last key is used by default */
480 return ARRAY_SIZE(netdev_lock_type) - 1;
483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
484 unsigned short dev_type)
488 i = netdev_lock_pos(dev_type);
489 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
490 netdev_lock_name[i]);
493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
497 i = netdev_lock_pos(dev->type);
498 lockdep_set_class_and_name(&dev->addr_list_lock,
499 &netdev_addr_lock_key[i],
500 netdev_lock_name[i]);
503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
504 unsigned short dev_type)
508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
513 /*******************************************************************************
515 * Protocol management and registration routines
517 *******************************************************************************/
521 * Add a protocol ID to the list. Now that the input handler is
522 * smarter we can dispense with all the messy stuff that used to be
525 * BEWARE!!! Protocol handlers, mangling input packets,
526 * MUST BE last in hash buckets and checking protocol handlers
527 * MUST start from promiscuous ptype_all chain in net_bh.
528 * It is true now, do not change it.
529 * Explanation follows: if protocol handler, mangling packet, will
530 * be the first on list, it is not able to sense, that packet
531 * is cloned and should be copied-on-write, so that it will
532 * change it and subsequent readers will get broken packet.
536 static inline struct list_head *ptype_head(const struct packet_type *pt)
538 if (pt->type == htons(ETH_P_ALL))
539 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
541 return pt->dev ? &pt->dev->ptype_specific :
542 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
546 * dev_add_pack - add packet handler
547 * @pt: packet type declaration
549 * Add a protocol handler to the networking stack. The passed &packet_type
550 * is linked into kernel lists and may not be freed until it has been
551 * removed from the kernel lists.
553 * This call does not sleep therefore it can not
554 * guarantee all CPU's that are in middle of receiving packets
555 * will see the new packet type (until the next received packet).
558 void dev_add_pack(struct packet_type *pt)
560 struct list_head *head = ptype_head(pt);
562 spin_lock(&ptype_lock);
563 list_add_rcu(&pt->list, head);
564 spin_unlock(&ptype_lock);
566 EXPORT_SYMBOL(dev_add_pack);
569 * __dev_remove_pack - remove packet handler
570 * @pt: packet type declaration
572 * Remove a protocol handler that was previously added to the kernel
573 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
574 * from the kernel lists and can be freed or reused once this function
577 * The packet type might still be in use by receivers
578 * and must not be freed until after all the CPU's have gone
579 * through a quiescent state.
581 void __dev_remove_pack(struct packet_type *pt)
583 struct list_head *head = ptype_head(pt);
584 struct packet_type *pt1;
586 spin_lock(&ptype_lock);
588 list_for_each_entry(pt1, head, list) {
590 list_del_rcu(&pt->list);
595 pr_warn("dev_remove_pack: %p not found\n", pt);
597 spin_unlock(&ptype_lock);
599 EXPORT_SYMBOL(__dev_remove_pack);
602 * dev_remove_pack - remove packet handler
603 * @pt: packet type declaration
605 * Remove a protocol handler that was previously added to the kernel
606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
607 * from the kernel lists and can be freed or reused once this function
610 * This call sleeps to guarantee that no CPU is looking at the packet
613 void dev_remove_pack(struct packet_type *pt)
615 __dev_remove_pack(pt);
619 EXPORT_SYMBOL(dev_remove_pack);
622 /*******************************************************************************
624 * Device Interface Subroutines
626 *******************************************************************************/
629 * dev_get_iflink - get 'iflink' value of a interface
630 * @dev: targeted interface
632 * Indicates the ifindex the interface is linked to.
633 * Physical interfaces have the same 'ifindex' and 'iflink' values.
636 int dev_get_iflink(const struct net_device *dev)
638 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
639 return dev->netdev_ops->ndo_get_iflink(dev);
643 EXPORT_SYMBOL(dev_get_iflink);
646 * dev_fill_metadata_dst - Retrieve tunnel egress information.
647 * @dev: targeted interface
650 * For better visibility of tunnel traffic OVS needs to retrieve
651 * egress tunnel information for a packet. Following API allows
652 * user to get this info.
654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
656 struct ip_tunnel_info *info;
658 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
661 info = skb_tunnel_info_unclone(skb);
664 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
667 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
673 int k = stack->num_paths++;
675 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
678 return &stack->path[k];
681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
682 struct net_device_path_stack *stack)
684 const struct net_device *last_dev;
685 struct net_device_path_ctx ctx = {
688 struct net_device_path *path;
691 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
692 stack->num_paths = 0;
693 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
695 path = dev_fwd_path(stack);
699 memset(path, 0, sizeof(struct net_device_path));
700 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
704 if (WARN_ON_ONCE(last_dev == ctx.dev))
711 path = dev_fwd_path(stack);
714 path->type = DEV_PATH_ETHERNET;
719 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
722 * __dev_get_by_name - find a device by its name
723 * @net: the applicable net namespace
724 * @name: name to find
726 * Find an interface by name. Must be called under RTNL semaphore
727 * or @dev_base_lock. If the name is found a pointer to the device
728 * is returned. If the name is not found then %NULL is returned. The
729 * reference counters are not incremented so the caller must be
730 * careful with locks.
733 struct net_device *__dev_get_by_name(struct net *net, const char *name)
735 struct netdev_name_node *node_name;
737 node_name = netdev_name_node_lookup(net, name);
738 return node_name ? node_name->dev : NULL;
740 EXPORT_SYMBOL(__dev_get_by_name);
743 * dev_get_by_name_rcu - find a device by its name
744 * @net: the applicable net namespace
745 * @name: name to find
747 * Find an interface by name.
748 * If the name is found a pointer to the device is returned.
749 * If the name is not found then %NULL is returned.
750 * The reference counters are not incremented so the caller must be
751 * careful with locks. The caller must hold RCU lock.
754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
756 struct netdev_name_node *node_name;
758 node_name = netdev_name_node_lookup_rcu(net, name);
759 return node_name ? node_name->dev : NULL;
761 EXPORT_SYMBOL(dev_get_by_name_rcu);
764 * dev_get_by_name - find a device by its name
765 * @net: the applicable net namespace
766 * @name: name to find
768 * Find an interface by name. This can be called from any
769 * context and does its own locking. The returned handle has
770 * the usage count incremented and the caller must use dev_put() to
771 * release it when it is no longer needed. %NULL is returned if no
772 * matching device is found.
775 struct net_device *dev_get_by_name(struct net *net, const char *name)
777 struct net_device *dev;
780 dev = dev_get_by_name_rcu(net, name);
785 EXPORT_SYMBOL(dev_get_by_name);
788 * __dev_get_by_index - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns %NULL if the device
793 * is not found or a pointer to the device. The device has not
794 * had its reference counter increased so the caller must be careful
795 * about locking. The caller must hold either the RTNL semaphore
799 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
801 struct net_device *dev;
802 struct hlist_head *head = dev_index_hash(net, ifindex);
804 hlist_for_each_entry(dev, head, index_hlist)
805 if (dev->ifindex == ifindex)
810 EXPORT_SYMBOL(__dev_get_by_index);
813 * dev_get_by_index_rcu - find a device by its ifindex
814 * @net: the applicable net namespace
815 * @ifindex: index of device
817 * Search for an interface by index. Returns %NULL if the device
818 * is not found or a pointer to the device. The device has not
819 * had its reference counter increased so the caller must be careful
820 * about locking. The caller must hold RCU lock.
823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
825 struct net_device *dev;
826 struct hlist_head *head = dev_index_hash(net, ifindex);
828 hlist_for_each_entry_rcu(dev, head, index_hlist)
829 if (dev->ifindex == ifindex)
834 EXPORT_SYMBOL(dev_get_by_index_rcu);
838 * dev_get_by_index - find a device by its ifindex
839 * @net: the applicable net namespace
840 * @ifindex: index of device
842 * Search for an interface by index. Returns NULL if the device
843 * is not found or a pointer to the device. The device returned has
844 * had a reference added and the pointer is safe until the user calls
845 * dev_put to indicate they have finished with it.
848 struct net_device *dev_get_by_index(struct net *net, int ifindex)
850 struct net_device *dev;
853 dev = dev_get_by_index_rcu(net, ifindex);
858 EXPORT_SYMBOL(dev_get_by_index);
861 * dev_get_by_napi_id - find a device by napi_id
862 * @napi_id: ID of the NAPI struct
864 * Search for an interface by NAPI ID. Returns %NULL if the device
865 * is not found or a pointer to the device. The device has not had
866 * its reference counter increased so the caller must be careful
867 * about locking. The caller must hold RCU lock.
870 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
872 struct napi_struct *napi;
874 WARN_ON_ONCE(!rcu_read_lock_held());
876 if (napi_id < MIN_NAPI_ID)
879 napi = napi_by_id(napi_id);
881 return napi ? napi->dev : NULL;
883 EXPORT_SYMBOL(dev_get_by_napi_id);
886 * netdev_get_name - get a netdevice name, knowing its ifindex.
887 * @net: network namespace
888 * @name: a pointer to the buffer where the name will be stored.
889 * @ifindex: the ifindex of the interface to get the name from.
891 int netdev_get_name(struct net *net, char *name, int ifindex)
893 struct net_device *dev;
896 down_read(&devnet_rename_sem);
899 dev = dev_get_by_index_rcu(net, ifindex);
905 strcpy(name, dev->name);
910 up_read(&devnet_rename_sem);
915 * dev_getbyhwaddr_rcu - find a device by its hardware address
916 * @net: the applicable net namespace
917 * @type: media type of device
918 * @ha: hardware address
920 * Search for an interface by MAC address. Returns NULL if the device
921 * is not found or a pointer to the device.
922 * The caller must hold RCU or RTNL.
923 * The returned device has not had its ref count increased
924 * and the caller must therefore be careful about locking
928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
931 struct net_device *dev;
933 for_each_netdev_rcu(net, dev)
934 if (dev->type == type &&
935 !memcmp(dev->dev_addr, ha, dev->addr_len))
940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
944 struct net_device *dev, *ret = NULL;
947 for_each_netdev_rcu(net, dev)
948 if (dev->type == type) {
956 EXPORT_SYMBOL(dev_getfirstbyhwtype);
959 * __dev_get_by_flags - find any device with given flags
960 * @net: the applicable net namespace
961 * @if_flags: IFF_* values
962 * @mask: bitmask of bits in if_flags to check
964 * Search for any interface with the given flags. Returns NULL if a device
965 * is not found or a pointer to the device. Must be called inside
966 * rtnl_lock(), and result refcount is unchanged.
969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
972 struct net_device *dev, *ret;
977 for_each_netdev(net, dev) {
978 if (((dev->flags ^ if_flags) & mask) == 0) {
985 EXPORT_SYMBOL(__dev_get_by_flags);
988 * dev_valid_name - check if name is okay for network device
991 * Network device names need to be valid file names to
992 * allow sysfs to work. We also disallow any kind of
995 bool dev_valid_name(const char *name)
999 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1001 if (!strcmp(name, ".") || !strcmp(name, ".."))
1005 if (*name == '/' || *name == ':' || isspace(*name))
1011 EXPORT_SYMBOL(dev_valid_name);
1014 * __dev_alloc_name - allocate a name for a device
1015 * @net: network namespace to allocate the device name in
1016 * @name: name format string
1017 * @buf: scratch buffer and result name string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1032 const int max_netdevices = 8*PAGE_SIZE;
1033 unsigned long *inuse;
1034 struct net_device *d;
1036 if (!dev_valid_name(name))
1039 p = strchr(name, '%');
1042 * Verify the string as this thing may have come from
1043 * the user. There must be either one "%d" and no other "%"
1046 if (p[1] != 'd' || strchr(p + 2, '%'))
1049 /* Use one page as a bit array of possible slots */
1050 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1054 for_each_netdev(net, d) {
1055 struct netdev_name_node *name_node;
1056 list_for_each_entry(name_node, &d->name_node->list, list) {
1057 if (!sscanf(name_node->name, name, &i))
1059 if (i < 0 || i >= max_netdevices)
1062 /* avoid cases where sscanf is not exact inverse of printf */
1063 snprintf(buf, IFNAMSIZ, name, i);
1064 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1065 __set_bit(i, inuse);
1067 if (!sscanf(d->name, name, &i))
1069 if (i < 0 || i >= max_netdevices)
1072 /* avoid cases where sscanf is not exact inverse of printf */
1073 snprintf(buf, IFNAMSIZ, name, i);
1074 if (!strncmp(buf, d->name, IFNAMSIZ))
1075 __set_bit(i, inuse);
1078 i = find_first_zero_bit(inuse, max_netdevices);
1079 free_page((unsigned long) inuse);
1082 snprintf(buf, IFNAMSIZ, name, i);
1083 if (!netdev_name_in_use(net, buf))
1086 /* It is possible to run out of possible slots
1087 * when the name is long and there isn't enough space left
1088 * for the digits, or if all bits are used.
1093 static int dev_alloc_name_ns(struct net *net,
1094 struct net_device *dev,
1101 ret = __dev_alloc_name(net, name, buf);
1103 strscpy(dev->name, buf, IFNAMSIZ);
1108 * dev_alloc_name - allocate a name for a device
1110 * @name: name format string
1112 * Passed a format string - eg "lt%d" it will try and find a suitable
1113 * id. It scans list of devices to build up a free map, then chooses
1114 * the first empty slot. The caller must hold the dev_base or rtnl lock
1115 * while allocating the name and adding the device in order to avoid
1117 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1118 * Returns the number of the unit assigned or a negative errno code.
1121 int dev_alloc_name(struct net_device *dev, const char *name)
1123 return dev_alloc_name_ns(dev_net(dev), dev, name);
1125 EXPORT_SYMBOL(dev_alloc_name);
1127 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1132 if (!dev_valid_name(name))
1135 if (strchr(name, '%'))
1136 return dev_alloc_name_ns(net, dev, name);
1137 else if (netdev_name_in_use(net, name))
1139 else if (dev->name != name)
1140 strscpy(dev->name, name, IFNAMSIZ);
1146 * dev_change_name - change name of a device
1148 * @newname: name (or format string) must be at least IFNAMSIZ
1150 * Change name of a device, can pass format strings "eth%d".
1153 int dev_change_name(struct net_device *dev, const char *newname)
1155 unsigned char old_assign_type;
1156 char oldname[IFNAMSIZ];
1162 BUG_ON(!dev_net(dev));
1166 /* Some auto-enslaved devices e.g. failover slaves are
1167 * special, as userspace might rename the device after
1168 * the interface had been brought up and running since
1169 * the point kernel initiated auto-enslavement. Allow
1170 * live name change even when these slave devices are
1173 * Typically, users of these auto-enslaving devices
1174 * don't actually care about slave name change, as
1175 * they are supposed to operate on master interface
1178 if (dev->flags & IFF_UP &&
1179 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1182 down_write(&devnet_rename_sem);
1184 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1185 up_write(&devnet_rename_sem);
1189 memcpy(oldname, dev->name, IFNAMSIZ);
1191 err = dev_get_valid_name(net, dev, newname);
1193 up_write(&devnet_rename_sem);
1197 if (oldname[0] && !strchr(oldname, '%'))
1198 netdev_info(dev, "renamed from %s\n", oldname);
1200 old_assign_type = dev->name_assign_type;
1201 dev->name_assign_type = NET_NAME_RENAMED;
1204 ret = device_rename(&dev->dev, dev->name);
1206 memcpy(dev->name, oldname, IFNAMSIZ);
1207 dev->name_assign_type = old_assign_type;
1208 up_write(&devnet_rename_sem);
1212 up_write(&devnet_rename_sem);
1214 netdev_adjacent_rename_links(dev, oldname);
1216 write_lock(&dev_base_lock);
1217 netdev_name_node_del(dev->name_node);
1218 write_unlock(&dev_base_lock);
1222 write_lock(&dev_base_lock);
1223 netdev_name_node_add(net, dev->name_node);
1224 write_unlock(&dev_base_lock);
1226 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1227 ret = notifier_to_errno(ret);
1230 /* err >= 0 after dev_alloc_name() or stores the first errno */
1233 down_write(&devnet_rename_sem);
1234 memcpy(dev->name, oldname, IFNAMSIZ);
1235 memcpy(oldname, newname, IFNAMSIZ);
1236 dev->name_assign_type = old_assign_type;
1237 old_assign_type = NET_NAME_RENAMED;
1240 netdev_err(dev, "name change rollback failed: %d\n",
1249 * dev_set_alias - change ifalias of a device
1251 * @alias: name up to IFALIASZ
1252 * @len: limit of bytes to copy from info
1254 * Set ifalias for a device,
1256 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1258 struct dev_ifalias *new_alias = NULL;
1260 if (len >= IFALIASZ)
1264 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1268 memcpy(new_alias->ifalias, alias, len);
1269 new_alias->ifalias[len] = 0;
1272 mutex_lock(&ifalias_mutex);
1273 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1274 mutex_is_locked(&ifalias_mutex));
1275 mutex_unlock(&ifalias_mutex);
1278 kfree_rcu(new_alias, rcuhead);
1282 EXPORT_SYMBOL(dev_set_alias);
1285 * dev_get_alias - get ifalias of a device
1287 * @name: buffer to store name of ifalias
1288 * @len: size of buffer
1290 * get ifalias for a device. Caller must make sure dev cannot go
1291 * away, e.g. rcu read lock or own a reference count to device.
1293 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1295 const struct dev_ifalias *alias;
1299 alias = rcu_dereference(dev->ifalias);
1301 ret = snprintf(name, len, "%s", alias->ifalias);
1308 * netdev_features_change - device changes features
1309 * @dev: device to cause notification
1311 * Called to indicate a device has changed features.
1313 void netdev_features_change(struct net_device *dev)
1315 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1317 EXPORT_SYMBOL(netdev_features_change);
1320 * netdev_state_change - device changes state
1321 * @dev: device to cause notification
1323 * Called to indicate a device has changed state. This function calls
1324 * the notifier chains for netdev_chain and sends a NEWLINK message
1325 * to the routing socket.
1327 void netdev_state_change(struct net_device *dev)
1329 if (dev->flags & IFF_UP) {
1330 struct netdev_notifier_change_info change_info = {
1334 call_netdevice_notifiers_info(NETDEV_CHANGE,
1336 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1339 EXPORT_SYMBOL(netdev_state_change);
1342 * __netdev_notify_peers - notify network peers about existence of @dev,
1343 * to be called when rtnl lock is already held.
1344 * @dev: network device
1346 * Generate traffic such that interested network peers are aware of
1347 * @dev, such as by generating a gratuitous ARP. This may be used when
1348 * a device wants to inform the rest of the network about some sort of
1349 * reconfiguration such as a failover event or virtual machine
1352 void __netdev_notify_peers(struct net_device *dev)
1355 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1356 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1358 EXPORT_SYMBOL(__netdev_notify_peers);
1361 * netdev_notify_peers - notify network peers about existence of @dev
1362 * @dev: network device
1364 * Generate traffic such that interested network peers are aware of
1365 * @dev, such as by generating a gratuitous ARP. This may be used when
1366 * a device wants to inform the rest of the network about some sort of
1367 * reconfiguration such as a failover event or virtual machine
1370 void netdev_notify_peers(struct net_device *dev)
1373 __netdev_notify_peers(dev);
1376 EXPORT_SYMBOL(netdev_notify_peers);
1378 static int napi_threaded_poll(void *data);
1380 static int napi_kthread_create(struct napi_struct *n)
1384 /* Create and wake up the kthread once to put it in
1385 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1386 * warning and work with loadavg.
1388 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1389 n->dev->name, n->napi_id);
1390 if (IS_ERR(n->thread)) {
1391 err = PTR_ERR(n->thread);
1392 pr_err("kthread_run failed with err %d\n", err);
1399 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1401 const struct net_device_ops *ops = dev->netdev_ops;
1405 dev_addr_check(dev);
1407 if (!netif_device_present(dev)) {
1408 /* may be detached because parent is runtime-suspended */
1409 if (dev->dev.parent)
1410 pm_runtime_resume(dev->dev.parent);
1411 if (!netif_device_present(dev))
1415 /* Block netpoll from trying to do any rx path servicing.
1416 * If we don't do this there is a chance ndo_poll_controller
1417 * or ndo_poll may be running while we open the device
1419 netpoll_poll_disable(dev);
1421 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1422 ret = notifier_to_errno(ret);
1426 set_bit(__LINK_STATE_START, &dev->state);
1428 if (ops->ndo_validate_addr)
1429 ret = ops->ndo_validate_addr(dev);
1431 if (!ret && ops->ndo_open)
1432 ret = ops->ndo_open(dev);
1434 netpoll_poll_enable(dev);
1437 clear_bit(__LINK_STATE_START, &dev->state);
1439 dev->flags |= IFF_UP;
1440 dev_set_rx_mode(dev);
1442 add_device_randomness(dev->dev_addr, dev->addr_len);
1449 * dev_open - prepare an interface for use.
1450 * @dev: device to open
1451 * @extack: netlink extended ack
1453 * Takes a device from down to up state. The device's private open
1454 * function is invoked and then the multicast lists are loaded. Finally
1455 * the device is moved into the up state and a %NETDEV_UP message is
1456 * sent to the netdev notifier chain.
1458 * Calling this function on an active interface is a nop. On a failure
1459 * a negative errno code is returned.
1461 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1465 if (dev->flags & IFF_UP)
1468 ret = __dev_open(dev, extack);
1472 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1473 call_netdevice_notifiers(NETDEV_UP, dev);
1477 EXPORT_SYMBOL(dev_open);
1479 static void __dev_close_many(struct list_head *head)
1481 struct net_device *dev;
1486 list_for_each_entry(dev, head, close_list) {
1487 /* Temporarily disable netpoll until the interface is down */
1488 netpoll_poll_disable(dev);
1490 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1492 clear_bit(__LINK_STATE_START, &dev->state);
1494 /* Synchronize to scheduled poll. We cannot touch poll list, it
1495 * can be even on different cpu. So just clear netif_running().
1497 * dev->stop() will invoke napi_disable() on all of it's
1498 * napi_struct instances on this device.
1500 smp_mb__after_atomic(); /* Commit netif_running(). */
1503 dev_deactivate_many(head);
1505 list_for_each_entry(dev, head, close_list) {
1506 const struct net_device_ops *ops = dev->netdev_ops;
1509 * Call the device specific close. This cannot fail.
1510 * Only if device is UP
1512 * We allow it to be called even after a DETACH hot-plug
1518 dev->flags &= ~IFF_UP;
1519 netpoll_poll_enable(dev);
1523 static void __dev_close(struct net_device *dev)
1527 list_add(&dev->close_list, &single);
1528 __dev_close_many(&single);
1532 void dev_close_many(struct list_head *head, bool unlink)
1534 struct net_device *dev, *tmp;
1536 /* Remove the devices that don't need to be closed */
1537 list_for_each_entry_safe(dev, tmp, head, close_list)
1538 if (!(dev->flags & IFF_UP))
1539 list_del_init(&dev->close_list);
1541 __dev_close_many(head);
1543 list_for_each_entry_safe(dev, tmp, head, close_list) {
1544 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1545 call_netdevice_notifiers(NETDEV_DOWN, dev);
1547 list_del_init(&dev->close_list);
1550 EXPORT_SYMBOL(dev_close_many);
1553 * dev_close - shutdown an interface.
1554 * @dev: device to shutdown
1556 * This function moves an active device into down state. A
1557 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1558 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1561 void dev_close(struct net_device *dev)
1563 if (dev->flags & IFF_UP) {
1566 list_add(&dev->close_list, &single);
1567 dev_close_many(&single, true);
1571 EXPORT_SYMBOL(dev_close);
1575 * dev_disable_lro - disable Large Receive Offload on a device
1578 * Disable Large Receive Offload (LRO) on a net device. Must be
1579 * called under RTNL. This is needed if received packets may be
1580 * forwarded to another interface.
1582 void dev_disable_lro(struct net_device *dev)
1584 struct net_device *lower_dev;
1585 struct list_head *iter;
1587 dev->wanted_features &= ~NETIF_F_LRO;
1588 netdev_update_features(dev);
1590 if (unlikely(dev->features & NETIF_F_LRO))
1591 netdev_WARN(dev, "failed to disable LRO!\n");
1593 netdev_for_each_lower_dev(dev, lower_dev, iter)
1594 dev_disable_lro(lower_dev);
1596 EXPORT_SYMBOL(dev_disable_lro);
1599 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1602 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1603 * called under RTNL. This is needed if Generic XDP is installed on
1606 static void dev_disable_gro_hw(struct net_device *dev)
1608 dev->wanted_features &= ~NETIF_F_GRO_HW;
1609 netdev_update_features(dev);
1611 if (unlikely(dev->features & NETIF_F_GRO_HW))
1612 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1615 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1618 case NETDEV_##val: \
1619 return "NETDEV_" __stringify(val);
1621 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1622 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1623 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1624 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1625 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1626 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1627 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1628 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1629 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1630 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1631 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1634 return "UNKNOWN_NETDEV_EVENT";
1636 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1638 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1639 struct net_device *dev)
1641 struct netdev_notifier_info info = {
1645 return nb->notifier_call(nb, val, &info);
1648 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1649 struct net_device *dev)
1653 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1654 err = notifier_to_errno(err);
1658 if (!(dev->flags & IFF_UP))
1661 call_netdevice_notifier(nb, NETDEV_UP, dev);
1665 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1666 struct net_device *dev)
1668 if (dev->flags & IFF_UP) {
1669 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1671 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1673 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1676 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1679 struct net_device *dev;
1682 for_each_netdev(net, dev) {
1683 err = call_netdevice_register_notifiers(nb, dev);
1690 for_each_netdev_continue_reverse(net, dev)
1691 call_netdevice_unregister_notifiers(nb, dev);
1695 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1698 struct net_device *dev;
1700 for_each_netdev(net, dev)
1701 call_netdevice_unregister_notifiers(nb, dev);
1704 static int dev_boot_phase = 1;
1707 * register_netdevice_notifier - register a network notifier block
1710 * Register a notifier to be called when network device events occur.
1711 * The notifier passed is linked into the kernel structures and must
1712 * not be reused until it has been unregistered. A negative errno code
1713 * is returned on a failure.
1715 * When registered all registration and up events are replayed
1716 * to the new notifier to allow device to have a race free
1717 * view of the network device list.
1720 int register_netdevice_notifier(struct notifier_block *nb)
1725 /* Close race with setup_net() and cleanup_net() */
1726 down_write(&pernet_ops_rwsem);
1728 err = raw_notifier_chain_register(&netdev_chain, nb);
1734 err = call_netdevice_register_net_notifiers(nb, net);
1741 up_write(&pernet_ops_rwsem);
1745 for_each_net_continue_reverse(net)
1746 call_netdevice_unregister_net_notifiers(nb, net);
1748 raw_notifier_chain_unregister(&netdev_chain, nb);
1751 EXPORT_SYMBOL(register_netdevice_notifier);
1754 * unregister_netdevice_notifier - unregister a network notifier block
1757 * Unregister a notifier previously registered by
1758 * register_netdevice_notifier(). The notifier is unlinked into the
1759 * kernel structures and may then be reused. A negative errno code
1760 * is returned on a failure.
1762 * After unregistering unregister and down device events are synthesized
1763 * for all devices on the device list to the removed notifier to remove
1764 * the need for special case cleanup code.
1767 int unregister_netdevice_notifier(struct notifier_block *nb)
1772 /* Close race with setup_net() and cleanup_net() */
1773 down_write(&pernet_ops_rwsem);
1775 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1780 call_netdevice_unregister_net_notifiers(nb, net);
1784 up_write(&pernet_ops_rwsem);
1787 EXPORT_SYMBOL(unregister_netdevice_notifier);
1789 static int __register_netdevice_notifier_net(struct net *net,
1790 struct notifier_block *nb,
1791 bool ignore_call_fail)
1795 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1801 err = call_netdevice_register_net_notifiers(nb, net);
1802 if (err && !ignore_call_fail)
1803 goto chain_unregister;
1808 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1812 static int __unregister_netdevice_notifier_net(struct net *net,
1813 struct notifier_block *nb)
1817 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1821 call_netdevice_unregister_net_notifiers(nb, net);
1826 * register_netdevice_notifier_net - register a per-netns network notifier block
1827 * @net: network namespace
1830 * Register a notifier to be called when network device events occur.
1831 * The notifier passed is linked into the kernel structures and must
1832 * not be reused until it has been unregistered. A negative errno code
1833 * is returned on a failure.
1835 * When registered all registration and up events are replayed
1836 * to the new notifier to allow device to have a race free
1837 * view of the network device list.
1840 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1845 err = __register_netdevice_notifier_net(net, nb, false);
1849 EXPORT_SYMBOL(register_netdevice_notifier_net);
1852 * unregister_netdevice_notifier_net - unregister a per-netns
1853 * network notifier block
1854 * @net: network namespace
1857 * Unregister a notifier previously registered by
1858 * register_netdevice_notifier(). The notifier is unlinked into the
1859 * kernel structures and may then be reused. A negative errno code
1860 * is returned on a failure.
1862 * After unregistering unregister and down device events are synthesized
1863 * for all devices on the device list to the removed notifier to remove
1864 * the need for special case cleanup code.
1867 int unregister_netdevice_notifier_net(struct net *net,
1868 struct notifier_block *nb)
1873 err = __unregister_netdevice_notifier_net(net, nb);
1877 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1879 int register_netdevice_notifier_dev_net(struct net_device *dev,
1880 struct notifier_block *nb,
1881 struct netdev_net_notifier *nn)
1886 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1889 list_add(&nn->list, &dev->net_notifier_list);
1894 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1896 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1897 struct notifier_block *nb,
1898 struct netdev_net_notifier *nn)
1903 list_del(&nn->list);
1904 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1908 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1910 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1913 struct netdev_net_notifier *nn;
1915 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1916 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1917 __register_netdevice_notifier_net(net, nn->nb, true);
1922 * call_netdevice_notifiers_info - call all network notifier blocks
1923 * @val: value passed unmodified to notifier function
1924 * @info: notifier information data
1926 * Call all network notifier blocks. Parameters and return value
1927 * are as for raw_notifier_call_chain().
1930 static int call_netdevice_notifiers_info(unsigned long val,
1931 struct netdev_notifier_info *info)
1933 struct net *net = dev_net(info->dev);
1938 /* Run per-netns notifier block chain first, then run the global one.
1939 * Hopefully, one day, the global one is going to be removed after
1940 * all notifier block registrators get converted to be per-netns.
1942 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1943 if (ret & NOTIFY_STOP_MASK)
1945 return raw_notifier_call_chain(&netdev_chain, val, info);
1949 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1950 * for and rollback on error
1951 * @val_up: value passed unmodified to notifier function
1952 * @val_down: value passed unmodified to the notifier function when
1953 * recovering from an error on @val_up
1954 * @info: notifier information data
1956 * Call all per-netns network notifier blocks, but not notifier blocks on
1957 * the global notifier chain. Parameters and return value are as for
1958 * raw_notifier_call_chain_robust().
1962 call_netdevice_notifiers_info_robust(unsigned long val_up,
1963 unsigned long val_down,
1964 struct netdev_notifier_info *info)
1966 struct net *net = dev_net(info->dev);
1970 return raw_notifier_call_chain_robust(&net->netdev_chain,
1971 val_up, val_down, info);
1974 static int call_netdevice_notifiers_extack(unsigned long val,
1975 struct net_device *dev,
1976 struct netlink_ext_ack *extack)
1978 struct netdev_notifier_info info = {
1983 return call_netdevice_notifiers_info(val, &info);
1987 * call_netdevice_notifiers - call all network notifier blocks
1988 * @val: value passed unmodified to notifier function
1989 * @dev: net_device pointer passed unmodified to notifier function
1991 * Call all network notifier blocks. Parameters and return value
1992 * are as for raw_notifier_call_chain().
1995 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1997 return call_netdevice_notifiers_extack(val, dev, NULL);
1999 EXPORT_SYMBOL(call_netdevice_notifiers);
2002 * call_netdevice_notifiers_mtu - call all network notifier blocks
2003 * @val: value passed unmodified to notifier function
2004 * @dev: net_device pointer passed unmodified to notifier function
2005 * @arg: additional u32 argument passed to the notifier function
2007 * Call all network notifier blocks. Parameters and return value
2008 * are as for raw_notifier_call_chain().
2010 static int call_netdevice_notifiers_mtu(unsigned long val,
2011 struct net_device *dev, u32 arg)
2013 struct netdev_notifier_info_ext info = {
2018 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2020 return call_netdevice_notifiers_info(val, &info.info);
2023 #ifdef CONFIG_NET_INGRESS
2024 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2026 void net_inc_ingress_queue(void)
2028 static_branch_inc(&ingress_needed_key);
2030 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2032 void net_dec_ingress_queue(void)
2034 static_branch_dec(&ingress_needed_key);
2036 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2039 #ifdef CONFIG_NET_EGRESS
2040 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2042 void net_inc_egress_queue(void)
2044 static_branch_inc(&egress_needed_key);
2046 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2048 void net_dec_egress_queue(void)
2050 static_branch_dec(&egress_needed_key);
2052 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2055 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2056 EXPORT_SYMBOL(netstamp_needed_key);
2057 #ifdef CONFIG_JUMP_LABEL
2058 static atomic_t netstamp_needed_deferred;
2059 static atomic_t netstamp_wanted;
2060 static void netstamp_clear(struct work_struct *work)
2062 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2065 wanted = atomic_add_return(deferred, &netstamp_wanted);
2067 static_branch_enable(&netstamp_needed_key);
2069 static_branch_disable(&netstamp_needed_key);
2071 static DECLARE_WORK(netstamp_work, netstamp_clear);
2074 void net_enable_timestamp(void)
2076 #ifdef CONFIG_JUMP_LABEL
2080 wanted = atomic_read(&netstamp_wanted);
2083 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2086 atomic_inc(&netstamp_needed_deferred);
2087 schedule_work(&netstamp_work);
2089 static_branch_inc(&netstamp_needed_key);
2092 EXPORT_SYMBOL(net_enable_timestamp);
2094 void net_disable_timestamp(void)
2096 #ifdef CONFIG_JUMP_LABEL
2100 wanted = atomic_read(&netstamp_wanted);
2103 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2106 atomic_dec(&netstamp_needed_deferred);
2107 schedule_work(&netstamp_work);
2109 static_branch_dec(&netstamp_needed_key);
2112 EXPORT_SYMBOL(net_disable_timestamp);
2114 static inline void net_timestamp_set(struct sk_buff *skb)
2117 skb->mono_delivery_time = 0;
2118 if (static_branch_unlikely(&netstamp_needed_key))
2119 skb->tstamp = ktime_get_real();
2122 #define net_timestamp_check(COND, SKB) \
2123 if (static_branch_unlikely(&netstamp_needed_key)) { \
2124 if ((COND) && !(SKB)->tstamp) \
2125 (SKB)->tstamp = ktime_get_real(); \
2128 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2130 return __is_skb_forwardable(dev, skb, true);
2132 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2134 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2137 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2140 skb->protocol = eth_type_trans(skb, dev);
2141 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2147 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2149 return __dev_forward_skb2(dev, skb, true);
2151 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2154 * dev_forward_skb - loopback an skb to another netif
2156 * @dev: destination network device
2157 * @skb: buffer to forward
2160 * NET_RX_SUCCESS (no congestion)
2161 * NET_RX_DROP (packet was dropped, but freed)
2163 * dev_forward_skb can be used for injecting an skb from the
2164 * start_xmit function of one device into the receive queue
2165 * of another device.
2167 * The receiving device may be in another namespace, so
2168 * we have to clear all information in the skb that could
2169 * impact namespace isolation.
2171 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2173 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2175 EXPORT_SYMBOL_GPL(dev_forward_skb);
2177 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2179 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2182 static inline int deliver_skb(struct sk_buff *skb,
2183 struct packet_type *pt_prev,
2184 struct net_device *orig_dev)
2186 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2188 refcount_inc(&skb->users);
2189 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2192 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2193 struct packet_type **pt,
2194 struct net_device *orig_dev,
2196 struct list_head *ptype_list)
2198 struct packet_type *ptype, *pt_prev = *pt;
2200 list_for_each_entry_rcu(ptype, ptype_list, list) {
2201 if (ptype->type != type)
2204 deliver_skb(skb, pt_prev, orig_dev);
2210 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2212 if (!ptype->af_packet_priv || !skb->sk)
2215 if (ptype->id_match)
2216 return ptype->id_match(ptype, skb->sk);
2217 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2224 * dev_nit_active - return true if any network interface taps are in use
2226 * @dev: network device to check for the presence of taps
2228 bool dev_nit_active(struct net_device *dev)
2230 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2232 EXPORT_SYMBOL_GPL(dev_nit_active);
2235 * Support routine. Sends outgoing frames to any network
2236 * taps currently in use.
2239 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2241 struct packet_type *ptype;
2242 struct sk_buff *skb2 = NULL;
2243 struct packet_type *pt_prev = NULL;
2244 struct list_head *ptype_list = &ptype_all;
2248 list_for_each_entry_rcu(ptype, ptype_list, list) {
2249 if (ptype->ignore_outgoing)
2252 /* Never send packets back to the socket
2253 * they originated from - MvS (miquels@drinkel.ow.org)
2255 if (skb_loop_sk(ptype, skb))
2259 deliver_skb(skb2, pt_prev, skb->dev);
2264 /* need to clone skb, done only once */
2265 skb2 = skb_clone(skb, GFP_ATOMIC);
2269 net_timestamp_set(skb2);
2271 /* skb->nh should be correctly
2272 * set by sender, so that the second statement is
2273 * just protection against buggy protocols.
2275 skb_reset_mac_header(skb2);
2277 if (skb_network_header(skb2) < skb2->data ||
2278 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2279 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2280 ntohs(skb2->protocol),
2282 skb_reset_network_header(skb2);
2285 skb2->transport_header = skb2->network_header;
2286 skb2->pkt_type = PACKET_OUTGOING;
2290 if (ptype_list == &ptype_all) {
2291 ptype_list = &dev->ptype_all;
2296 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2297 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2303 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2306 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2307 * @dev: Network device
2308 * @txq: number of queues available
2310 * If real_num_tx_queues is changed the tc mappings may no longer be
2311 * valid. To resolve this verify the tc mapping remains valid and if
2312 * not NULL the mapping. With no priorities mapping to this
2313 * offset/count pair it will no longer be used. In the worst case TC0
2314 * is invalid nothing can be done so disable priority mappings. If is
2315 * expected that drivers will fix this mapping if they can before
2316 * calling netif_set_real_num_tx_queues.
2318 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2321 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2323 /* If TC0 is invalidated disable TC mapping */
2324 if (tc->offset + tc->count > txq) {
2325 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2330 /* Invalidated prio to tc mappings set to TC0 */
2331 for (i = 1; i < TC_BITMASK + 1; i++) {
2332 int q = netdev_get_prio_tc_map(dev, i);
2334 tc = &dev->tc_to_txq[q];
2335 if (tc->offset + tc->count > txq) {
2336 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",
2338 netdev_set_prio_tc_map(dev, i, 0);
2343 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2346 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2349 /* walk through the TCs and see if it falls into any of them */
2350 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2351 if ((txq - tc->offset) < tc->count)
2355 /* didn't find it, just return -1 to indicate no match */
2361 EXPORT_SYMBOL(netdev_txq_to_tc);
2364 static struct static_key xps_needed __read_mostly;
2365 static struct static_key xps_rxqs_needed __read_mostly;
2366 static DEFINE_MUTEX(xps_map_mutex);
2367 #define xmap_dereference(P) \
2368 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2370 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2371 struct xps_dev_maps *old_maps, int tci, u16 index)
2373 struct xps_map *map = NULL;
2377 map = xmap_dereference(dev_maps->attr_map[tci]);
2381 for (pos = map->len; pos--;) {
2382 if (map->queues[pos] != index)
2386 map->queues[pos] = map->queues[--map->len];
2391 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2392 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2393 kfree_rcu(map, rcu);
2400 static bool remove_xps_queue_cpu(struct net_device *dev,
2401 struct xps_dev_maps *dev_maps,
2402 int cpu, u16 offset, u16 count)
2404 int num_tc = dev_maps->num_tc;
2405 bool active = false;
2408 for (tci = cpu * num_tc; num_tc--; tci++) {
2411 for (i = count, j = offset; i--; j++) {
2412 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2422 static void reset_xps_maps(struct net_device *dev,
2423 struct xps_dev_maps *dev_maps,
2424 enum xps_map_type type)
2426 static_key_slow_dec_cpuslocked(&xps_needed);
2427 if (type == XPS_RXQS)
2428 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2430 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2432 kfree_rcu(dev_maps, rcu);
2435 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2436 u16 offset, u16 count)
2438 struct xps_dev_maps *dev_maps;
2439 bool active = false;
2442 dev_maps = xmap_dereference(dev->xps_maps[type]);
2446 for (j = 0; j < dev_maps->nr_ids; j++)
2447 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2449 reset_xps_maps(dev, dev_maps, type);
2451 if (type == XPS_CPUS) {
2452 for (i = offset + (count - 1); count--; i--)
2453 netdev_queue_numa_node_write(
2454 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2458 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2461 if (!static_key_false(&xps_needed))
2465 mutex_lock(&xps_map_mutex);
2467 if (static_key_false(&xps_rxqs_needed))
2468 clean_xps_maps(dev, XPS_RXQS, offset, count);
2470 clean_xps_maps(dev, XPS_CPUS, offset, count);
2472 mutex_unlock(&xps_map_mutex);
2476 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2478 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2481 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2482 u16 index, bool is_rxqs_map)
2484 struct xps_map *new_map;
2485 int alloc_len = XPS_MIN_MAP_ALLOC;
2488 for (pos = 0; map && pos < map->len; pos++) {
2489 if (map->queues[pos] != index)
2494 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2496 if (pos < map->alloc_len)
2499 alloc_len = map->alloc_len * 2;
2502 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2506 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2508 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2509 cpu_to_node(attr_index));
2513 for (i = 0; i < pos; i++)
2514 new_map->queues[i] = map->queues[i];
2515 new_map->alloc_len = alloc_len;
2521 /* Copy xps maps at a given index */
2522 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2523 struct xps_dev_maps *new_dev_maps, int index,
2524 int tc, bool skip_tc)
2526 int i, tci = index * dev_maps->num_tc;
2527 struct xps_map *map;
2529 /* copy maps belonging to foreign traffic classes */
2530 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2531 if (i == tc && skip_tc)
2534 /* fill in the new device map from the old device map */
2535 map = xmap_dereference(dev_maps->attr_map[tci]);
2536 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2540 /* Must be called under cpus_read_lock */
2541 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2542 u16 index, enum xps_map_type type)
2544 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2545 const unsigned long *online_mask = NULL;
2546 bool active = false, copy = false;
2547 int i, j, tci, numa_node_id = -2;
2548 int maps_sz, num_tc = 1, tc = 0;
2549 struct xps_map *map, *new_map;
2550 unsigned int nr_ids;
2553 /* Do not allow XPS on subordinate device directly */
2554 num_tc = dev->num_tc;
2558 /* If queue belongs to subordinate dev use its map */
2559 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2561 tc = netdev_txq_to_tc(dev, index);
2566 mutex_lock(&xps_map_mutex);
2568 dev_maps = xmap_dereference(dev->xps_maps[type]);
2569 if (type == XPS_RXQS) {
2570 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2571 nr_ids = dev->num_rx_queues;
2573 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2574 if (num_possible_cpus() > 1)
2575 online_mask = cpumask_bits(cpu_online_mask);
2576 nr_ids = nr_cpu_ids;
2579 if (maps_sz < L1_CACHE_BYTES)
2580 maps_sz = L1_CACHE_BYTES;
2582 /* The old dev_maps could be larger or smaller than the one we're
2583 * setting up now, as dev->num_tc or nr_ids could have been updated in
2584 * between. We could try to be smart, but let's be safe instead and only
2585 * copy foreign traffic classes if the two map sizes match.
2588 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2591 /* allocate memory for queue storage */
2592 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2594 if (!new_dev_maps) {
2595 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2596 if (!new_dev_maps) {
2597 mutex_unlock(&xps_map_mutex);
2601 new_dev_maps->nr_ids = nr_ids;
2602 new_dev_maps->num_tc = num_tc;
2605 tci = j * num_tc + tc;
2606 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2608 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2612 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2616 goto out_no_new_maps;
2619 /* Increment static keys at most once per type */
2620 static_key_slow_inc_cpuslocked(&xps_needed);
2621 if (type == XPS_RXQS)
2622 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2625 for (j = 0; j < nr_ids; j++) {
2626 bool skip_tc = false;
2628 tci = j * num_tc + tc;
2629 if (netif_attr_test_mask(j, mask, nr_ids) &&
2630 netif_attr_test_online(j, online_mask, nr_ids)) {
2631 /* add tx-queue to CPU/rx-queue maps */
2636 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2637 while ((pos < map->len) && (map->queues[pos] != index))
2640 if (pos == map->len)
2641 map->queues[map->len++] = index;
2643 if (type == XPS_CPUS) {
2644 if (numa_node_id == -2)
2645 numa_node_id = cpu_to_node(j);
2646 else if (numa_node_id != cpu_to_node(j))
2653 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2657 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2659 /* Cleanup old maps */
2661 goto out_no_old_maps;
2663 for (j = 0; j < dev_maps->nr_ids; j++) {
2664 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2665 map = xmap_dereference(dev_maps->attr_map[tci]);
2670 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2675 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2676 kfree_rcu(map, rcu);
2680 old_dev_maps = dev_maps;
2683 dev_maps = new_dev_maps;
2687 if (type == XPS_CPUS)
2688 /* update Tx queue numa node */
2689 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2690 (numa_node_id >= 0) ?
2691 numa_node_id : NUMA_NO_NODE);
2696 /* removes tx-queue from unused CPUs/rx-queues */
2697 for (j = 0; j < dev_maps->nr_ids; j++) {
2698 tci = j * dev_maps->num_tc;
2700 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2702 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2703 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2706 active |= remove_xps_queue(dev_maps,
2707 copy ? old_dev_maps : NULL,
2713 kfree_rcu(old_dev_maps, rcu);
2715 /* free map if not active */
2717 reset_xps_maps(dev, dev_maps, type);
2720 mutex_unlock(&xps_map_mutex);
2724 /* remove any maps that we added */
2725 for (j = 0; j < nr_ids; j++) {
2726 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2727 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2729 xmap_dereference(dev_maps->attr_map[tci]) :
2731 if (new_map && new_map != map)
2736 mutex_unlock(&xps_map_mutex);
2738 kfree(new_dev_maps);
2741 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2743 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2749 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2754 EXPORT_SYMBOL(netif_set_xps_queue);
2757 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2759 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2761 /* Unbind any subordinate channels */
2762 while (txq-- != &dev->_tx[0]) {
2764 netdev_unbind_sb_channel(dev, txq->sb_dev);
2768 void netdev_reset_tc(struct net_device *dev)
2771 netif_reset_xps_queues_gt(dev, 0);
2773 netdev_unbind_all_sb_channels(dev);
2775 /* Reset TC configuration of device */
2777 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2778 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2780 EXPORT_SYMBOL(netdev_reset_tc);
2782 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2784 if (tc >= dev->num_tc)
2788 netif_reset_xps_queues(dev, offset, count);
2790 dev->tc_to_txq[tc].count = count;
2791 dev->tc_to_txq[tc].offset = offset;
2794 EXPORT_SYMBOL(netdev_set_tc_queue);
2796 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2798 if (num_tc > TC_MAX_QUEUE)
2802 netif_reset_xps_queues_gt(dev, 0);
2804 netdev_unbind_all_sb_channels(dev);
2806 dev->num_tc = num_tc;
2809 EXPORT_SYMBOL(netdev_set_num_tc);
2811 void netdev_unbind_sb_channel(struct net_device *dev,
2812 struct net_device *sb_dev)
2814 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2817 netif_reset_xps_queues_gt(sb_dev, 0);
2819 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2820 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2822 while (txq-- != &dev->_tx[0]) {
2823 if (txq->sb_dev == sb_dev)
2827 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2829 int netdev_bind_sb_channel_queue(struct net_device *dev,
2830 struct net_device *sb_dev,
2831 u8 tc, u16 count, u16 offset)
2833 /* Make certain the sb_dev and dev are already configured */
2834 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2837 /* We cannot hand out queues we don't have */
2838 if ((offset + count) > dev->real_num_tx_queues)
2841 /* Record the mapping */
2842 sb_dev->tc_to_txq[tc].count = count;
2843 sb_dev->tc_to_txq[tc].offset = offset;
2845 /* Provide a way for Tx queue to find the tc_to_txq map or
2846 * XPS map for itself.
2849 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2853 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2855 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2857 /* Do not use a multiqueue device to represent a subordinate channel */
2858 if (netif_is_multiqueue(dev))
2861 /* We allow channels 1 - 32767 to be used for subordinate channels.
2862 * Channel 0 is meant to be "native" mode and used only to represent
2863 * the main root device. We allow writing 0 to reset the device back
2864 * to normal mode after being used as a subordinate channel.
2866 if (channel > S16_MAX)
2869 dev->num_tc = -channel;
2873 EXPORT_SYMBOL(netdev_set_sb_channel);
2876 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2877 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2879 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2884 disabling = txq < dev->real_num_tx_queues;
2886 if (txq < 1 || txq > dev->num_tx_queues)
2889 if (dev->reg_state == NETREG_REGISTERED ||
2890 dev->reg_state == NETREG_UNREGISTERING) {
2893 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2899 netif_setup_tc(dev, txq);
2901 dev_qdisc_change_real_num_tx(dev, txq);
2903 dev->real_num_tx_queues = txq;
2907 qdisc_reset_all_tx_gt(dev, txq);
2909 netif_reset_xps_queues_gt(dev, txq);
2913 dev->real_num_tx_queues = txq;
2918 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2922 * netif_set_real_num_rx_queues - set actual number of RX queues used
2923 * @dev: Network device
2924 * @rxq: Actual number of RX queues
2926 * This must be called either with the rtnl_lock held or before
2927 * registration of the net device. Returns 0 on success, or a
2928 * negative error code. If called before registration, it always
2931 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2935 if (rxq < 1 || rxq > dev->num_rx_queues)
2938 if (dev->reg_state == NETREG_REGISTERED) {
2941 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2947 dev->real_num_rx_queues = rxq;
2950 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2954 * netif_set_real_num_queues - set actual number of RX and TX queues used
2955 * @dev: Network device
2956 * @txq: Actual number of TX queues
2957 * @rxq: Actual number of RX queues
2959 * Set the real number of both TX and RX queues.
2960 * Does nothing if the number of queues is already correct.
2962 int netif_set_real_num_queues(struct net_device *dev,
2963 unsigned int txq, unsigned int rxq)
2965 unsigned int old_rxq = dev->real_num_rx_queues;
2968 if (txq < 1 || txq > dev->num_tx_queues ||
2969 rxq < 1 || rxq > dev->num_rx_queues)
2972 /* Start from increases, so the error path only does decreases -
2973 * decreases can't fail.
2975 if (rxq > dev->real_num_rx_queues) {
2976 err = netif_set_real_num_rx_queues(dev, rxq);
2980 if (txq > dev->real_num_tx_queues) {
2981 err = netif_set_real_num_tx_queues(dev, txq);
2985 if (rxq < dev->real_num_rx_queues)
2986 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2987 if (txq < dev->real_num_tx_queues)
2988 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2992 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2995 EXPORT_SYMBOL(netif_set_real_num_queues);
2998 * netif_set_tso_max_size() - set the max size of TSO frames supported
2999 * @dev: netdev to update
3000 * @size: max skb->len of a TSO frame
3002 * Set the limit on the size of TSO super-frames the device can handle.
3003 * Unless explicitly set the stack will assume the value of
3004 * %GSO_LEGACY_MAX_SIZE.
3006 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3008 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3009 if (size < READ_ONCE(dev->gso_max_size))
3010 netif_set_gso_max_size(dev, size);
3012 EXPORT_SYMBOL(netif_set_tso_max_size);
3015 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3016 * @dev: netdev to update
3017 * @segs: max number of TCP segments
3019 * Set the limit on the number of TCP segments the device can generate from
3020 * a single TSO super-frame.
3021 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3023 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3025 dev->tso_max_segs = segs;
3026 if (segs < READ_ONCE(dev->gso_max_segs))
3027 netif_set_gso_max_segs(dev, segs);
3029 EXPORT_SYMBOL(netif_set_tso_max_segs);
3032 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3033 * @to: netdev to update
3034 * @from: netdev from which to copy the limits
3036 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3038 netif_set_tso_max_size(to, from->tso_max_size);
3039 netif_set_tso_max_segs(to, from->tso_max_segs);
3041 EXPORT_SYMBOL(netif_inherit_tso_max);
3044 * netif_get_num_default_rss_queues - default number of RSS queues
3046 * Default value is the number of physical cores if there are only 1 or 2, or
3047 * divided by 2 if there are more.
3049 int netif_get_num_default_rss_queues(void)
3054 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3057 cpumask_copy(cpus, cpu_online_mask);
3058 for_each_cpu(cpu, cpus) {
3060 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3062 free_cpumask_var(cpus);
3064 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3066 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3068 static void __netif_reschedule(struct Qdisc *q)
3070 struct softnet_data *sd;
3071 unsigned long flags;
3073 local_irq_save(flags);
3074 sd = this_cpu_ptr(&softnet_data);
3075 q->next_sched = NULL;
3076 *sd->output_queue_tailp = q;
3077 sd->output_queue_tailp = &q->next_sched;
3078 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3079 local_irq_restore(flags);
3082 void __netif_schedule(struct Qdisc *q)
3084 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3085 __netif_reschedule(q);
3087 EXPORT_SYMBOL(__netif_schedule);
3089 struct dev_kfree_skb_cb {
3090 enum skb_free_reason reason;
3093 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3095 return (struct dev_kfree_skb_cb *)skb->cb;
3098 void netif_schedule_queue(struct netdev_queue *txq)
3101 if (!netif_xmit_stopped(txq)) {
3102 struct Qdisc *q = rcu_dereference(txq->qdisc);
3104 __netif_schedule(q);
3108 EXPORT_SYMBOL(netif_schedule_queue);
3110 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3112 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3116 q = rcu_dereference(dev_queue->qdisc);
3117 __netif_schedule(q);
3121 EXPORT_SYMBOL(netif_tx_wake_queue);
3123 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3125 unsigned long flags;
3130 if (likely(refcount_read(&skb->users) == 1)) {
3132 refcount_set(&skb->users, 0);
3133 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3136 get_kfree_skb_cb(skb)->reason = reason;
3137 local_irq_save(flags);
3138 skb->next = __this_cpu_read(softnet_data.completion_queue);
3139 __this_cpu_write(softnet_data.completion_queue, skb);
3140 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3141 local_irq_restore(flags);
3143 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3145 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3147 if (in_hardirq() || irqs_disabled())
3148 __dev_kfree_skb_irq(skb, reason);
3152 EXPORT_SYMBOL(__dev_kfree_skb_any);
3156 * netif_device_detach - mark device as removed
3157 * @dev: network device
3159 * Mark device as removed from system and therefore no longer available.
3161 void netif_device_detach(struct net_device *dev)
3163 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3164 netif_running(dev)) {
3165 netif_tx_stop_all_queues(dev);
3168 EXPORT_SYMBOL(netif_device_detach);
3171 * netif_device_attach - mark device as attached
3172 * @dev: network device
3174 * Mark device as attached from system and restart if needed.
3176 void netif_device_attach(struct net_device *dev)
3178 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3179 netif_running(dev)) {
3180 netif_tx_wake_all_queues(dev);
3181 __netdev_watchdog_up(dev);
3184 EXPORT_SYMBOL(netif_device_attach);
3187 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3188 * to be used as a distribution range.
3190 static u16 skb_tx_hash(const struct net_device *dev,
3191 const struct net_device *sb_dev,
3192 struct sk_buff *skb)
3196 u16 qcount = dev->real_num_tx_queues;
3199 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3201 qoffset = sb_dev->tc_to_txq[tc].offset;
3202 qcount = sb_dev->tc_to_txq[tc].count;
3203 if (unlikely(!qcount)) {
3204 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3205 sb_dev->name, qoffset, tc);
3207 qcount = dev->real_num_tx_queues;
3211 if (skb_rx_queue_recorded(skb)) {
3212 hash = skb_get_rx_queue(skb);
3213 if (hash >= qoffset)
3215 while (unlikely(hash >= qcount))
3217 return hash + qoffset;
3220 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3223 static void skb_warn_bad_offload(const struct sk_buff *skb)
3225 static const netdev_features_t null_features;
3226 struct net_device *dev = skb->dev;
3227 const char *name = "";
3229 if (!net_ratelimit())
3233 if (dev->dev.parent)
3234 name = dev_driver_string(dev->dev.parent);
3236 name = netdev_name(dev);
3238 skb_dump(KERN_WARNING, skb, false);
3239 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3240 name, dev ? &dev->features : &null_features,
3241 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3245 * Invalidate hardware checksum when packet is to be mangled, and
3246 * complete checksum manually on outgoing path.
3248 int skb_checksum_help(struct sk_buff *skb)
3251 int ret = 0, offset;
3253 if (skb->ip_summed == CHECKSUM_COMPLETE)
3254 goto out_set_summed;
3256 if (unlikely(skb_is_gso(skb))) {
3257 skb_warn_bad_offload(skb);
3261 /* Before computing a checksum, we should make sure no frag could
3262 * be modified by an external entity : checksum could be wrong.
3264 if (skb_has_shared_frag(skb)) {
3265 ret = __skb_linearize(skb);
3270 offset = skb_checksum_start_offset(skb);
3272 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3273 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3276 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3278 offset += skb->csum_offset;
3279 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3280 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3283 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3287 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3289 skb->ip_summed = CHECKSUM_NONE;
3293 EXPORT_SYMBOL(skb_checksum_help);
3295 int skb_crc32c_csum_help(struct sk_buff *skb)
3298 int ret = 0, offset, start;
3300 if (skb->ip_summed != CHECKSUM_PARTIAL)
3303 if (unlikely(skb_is_gso(skb)))
3306 /* Before computing a checksum, we should make sure no frag could
3307 * be modified by an external entity : checksum could be wrong.
3309 if (unlikely(skb_has_shared_frag(skb))) {
3310 ret = __skb_linearize(skb);
3314 start = skb_checksum_start_offset(skb);
3315 offset = start + offsetof(struct sctphdr, checksum);
3316 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3321 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3325 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3326 skb->len - start, ~(__u32)0,
3328 *(__le32 *)(skb->data + offset) = crc32c_csum;
3329 skb->ip_summed = CHECKSUM_NONE;
3330 skb->csum_not_inet = 0;
3335 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3337 __be16 type = skb->protocol;
3339 /* Tunnel gso handlers can set protocol to ethernet. */
3340 if (type == htons(ETH_P_TEB)) {
3343 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3346 eth = (struct ethhdr *)skb->data;
3347 type = eth->h_proto;
3350 return __vlan_get_protocol(skb, type, depth);
3353 /* openvswitch calls this on rx path, so we need a different check.
3355 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3358 return skb->ip_summed != CHECKSUM_PARTIAL &&
3359 skb->ip_summed != CHECKSUM_UNNECESSARY;
3361 return skb->ip_summed == CHECKSUM_NONE;
3365 * __skb_gso_segment - Perform segmentation on skb.
3366 * @skb: buffer to segment
3367 * @features: features for the output path (see dev->features)
3368 * @tx_path: whether it is called in TX path
3370 * This function segments the given skb and returns a list of segments.
3372 * It may return NULL if the skb requires no segmentation. This is
3373 * only possible when GSO is used for verifying header integrity.
3375 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3377 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3378 netdev_features_t features, bool tx_path)
3380 struct sk_buff *segs;
3382 if (unlikely(skb_needs_check(skb, tx_path))) {
3385 /* We're going to init ->check field in TCP or UDP header */
3386 err = skb_cow_head(skb, 0);
3388 return ERR_PTR(err);
3391 /* Only report GSO partial support if it will enable us to
3392 * support segmentation on this frame without needing additional
3395 if (features & NETIF_F_GSO_PARTIAL) {
3396 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3397 struct net_device *dev = skb->dev;
3399 partial_features |= dev->features & dev->gso_partial_features;
3400 if (!skb_gso_ok(skb, features | partial_features))
3401 features &= ~NETIF_F_GSO_PARTIAL;
3404 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3405 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3407 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3408 SKB_GSO_CB(skb)->encap_level = 0;
3410 skb_reset_mac_header(skb);
3411 skb_reset_mac_len(skb);
3413 segs = skb_mac_gso_segment(skb, features);
3415 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3416 skb_warn_bad_offload(skb);
3420 EXPORT_SYMBOL(__skb_gso_segment);
3422 /* Take action when hardware reception checksum errors are detected. */
3424 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3426 netdev_err(dev, "hw csum failure\n");
3427 skb_dump(KERN_ERR, skb, true);
3431 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3433 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3435 EXPORT_SYMBOL(netdev_rx_csum_fault);
3438 /* XXX: check that highmem exists at all on the given machine. */
3439 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3441 #ifdef CONFIG_HIGHMEM
3444 if (!(dev->features & NETIF_F_HIGHDMA)) {
3445 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3446 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3448 if (PageHighMem(skb_frag_page(frag)))
3456 /* If MPLS offload request, verify we are testing hardware MPLS features
3457 * instead of standard features for the netdev.
3459 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3460 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3461 netdev_features_t features,
3464 if (eth_p_mpls(type))
3465 features &= skb->dev->mpls_features;
3470 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3471 netdev_features_t features,
3478 static netdev_features_t harmonize_features(struct sk_buff *skb,
3479 netdev_features_t features)
3483 type = skb_network_protocol(skb, NULL);
3484 features = net_mpls_features(skb, features, type);
3486 if (skb->ip_summed != CHECKSUM_NONE &&
3487 !can_checksum_protocol(features, type)) {
3488 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3490 if (illegal_highdma(skb->dev, skb))
3491 features &= ~NETIF_F_SG;
3496 netdev_features_t passthru_features_check(struct sk_buff *skb,
3497 struct net_device *dev,
3498 netdev_features_t features)
3502 EXPORT_SYMBOL(passthru_features_check);
3504 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3505 struct net_device *dev,
3506 netdev_features_t features)
3508 return vlan_features_check(skb, features);
3511 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3512 struct net_device *dev,
3513 netdev_features_t features)
3515 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3517 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3518 return features & ~NETIF_F_GSO_MASK;
3520 if (!skb_shinfo(skb)->gso_type) {
3521 skb_warn_bad_offload(skb);
3522 return features & ~NETIF_F_GSO_MASK;
3525 /* Support for GSO partial features requires software
3526 * intervention before we can actually process the packets
3527 * so we need to strip support for any partial features now
3528 * and we can pull them back in after we have partially
3529 * segmented the frame.
3531 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3532 features &= ~dev->gso_partial_features;
3534 /* Make sure to clear the IPv4 ID mangling feature if the
3535 * IPv4 header has the potential to be fragmented.
3537 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3538 struct iphdr *iph = skb->encapsulation ?
3539 inner_ip_hdr(skb) : ip_hdr(skb);
3541 if (!(iph->frag_off & htons(IP_DF)))
3542 features &= ~NETIF_F_TSO_MANGLEID;
3548 netdev_features_t netif_skb_features(struct sk_buff *skb)
3550 struct net_device *dev = skb->dev;
3551 netdev_features_t features = dev->features;
3553 if (skb_is_gso(skb))
3554 features = gso_features_check(skb, dev, features);
3556 /* If encapsulation offload request, verify we are testing
3557 * hardware encapsulation features instead of standard
3558 * features for the netdev
3560 if (skb->encapsulation)
3561 features &= dev->hw_enc_features;
3563 if (skb_vlan_tagged(skb))
3564 features = netdev_intersect_features(features,
3565 dev->vlan_features |
3566 NETIF_F_HW_VLAN_CTAG_TX |
3567 NETIF_F_HW_VLAN_STAG_TX);
3569 if (dev->netdev_ops->ndo_features_check)
3570 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3573 features &= dflt_features_check(skb, dev, features);
3575 return harmonize_features(skb, features);
3577 EXPORT_SYMBOL(netif_skb_features);
3579 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3580 struct netdev_queue *txq, bool more)
3585 if (dev_nit_active(dev))
3586 dev_queue_xmit_nit(skb, dev);
3589 trace_net_dev_start_xmit(skb, dev);
3590 rc = netdev_start_xmit(skb, dev, txq, more);
3591 trace_net_dev_xmit(skb, rc, dev, len);
3596 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3597 struct netdev_queue *txq, int *ret)
3599 struct sk_buff *skb = first;
3600 int rc = NETDEV_TX_OK;
3603 struct sk_buff *next = skb->next;
3605 skb_mark_not_on_list(skb);
3606 rc = xmit_one(skb, dev, txq, next != NULL);
3607 if (unlikely(!dev_xmit_complete(rc))) {
3613 if (netif_tx_queue_stopped(txq) && skb) {
3614 rc = NETDEV_TX_BUSY;
3624 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3625 netdev_features_t features)
3627 if (skb_vlan_tag_present(skb) &&
3628 !vlan_hw_offload_capable(features, skb->vlan_proto))
3629 skb = __vlan_hwaccel_push_inside(skb);
3633 int skb_csum_hwoffload_help(struct sk_buff *skb,
3634 const netdev_features_t features)
3636 if (unlikely(skb_csum_is_sctp(skb)))
3637 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3638 skb_crc32c_csum_help(skb);
3640 if (features & NETIF_F_HW_CSUM)
3643 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3644 switch (skb->csum_offset) {
3645 case offsetof(struct tcphdr, check):
3646 case offsetof(struct udphdr, check):
3651 return skb_checksum_help(skb);
3653 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3655 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3657 netdev_features_t features;
3659 features = netif_skb_features(skb);
3660 skb = validate_xmit_vlan(skb, features);
3664 skb = sk_validate_xmit_skb(skb, dev);
3668 if (netif_needs_gso(skb, features)) {
3669 struct sk_buff *segs;
3671 segs = skb_gso_segment(skb, features);
3679 if (skb_needs_linearize(skb, features) &&
3680 __skb_linearize(skb))
3683 /* If packet is not checksummed and device does not
3684 * support checksumming for this protocol, complete
3685 * checksumming here.
3687 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3688 if (skb->encapsulation)
3689 skb_set_inner_transport_header(skb,
3690 skb_checksum_start_offset(skb));
3692 skb_set_transport_header(skb,
3693 skb_checksum_start_offset(skb));
3694 if (skb_csum_hwoffload_help(skb, features))
3699 skb = validate_xmit_xfrm(skb, features, again);
3706 dev_core_stats_tx_dropped_inc(dev);
3710 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3712 struct sk_buff *next, *head = NULL, *tail;
3714 for (; skb != NULL; skb = next) {
3716 skb_mark_not_on_list(skb);
3718 /* in case skb wont be segmented, point to itself */
3721 skb = validate_xmit_skb(skb, dev, again);
3729 /* If skb was segmented, skb->prev points to
3730 * the last segment. If not, it still contains skb.
3736 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3738 static void qdisc_pkt_len_init(struct sk_buff *skb)
3740 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3742 qdisc_skb_cb(skb)->pkt_len = skb->len;
3744 /* To get more precise estimation of bytes sent on wire,
3745 * we add to pkt_len the headers size of all segments
3747 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3748 unsigned int hdr_len;
3749 u16 gso_segs = shinfo->gso_segs;
3751 /* mac layer + network layer */
3752 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3754 /* + transport layer */
3755 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3756 const struct tcphdr *th;
3757 struct tcphdr _tcphdr;
3759 th = skb_header_pointer(skb, skb_transport_offset(skb),
3760 sizeof(_tcphdr), &_tcphdr);
3762 hdr_len += __tcp_hdrlen(th);
3764 struct udphdr _udphdr;
3766 if (skb_header_pointer(skb, skb_transport_offset(skb),
3767 sizeof(_udphdr), &_udphdr))
3768 hdr_len += sizeof(struct udphdr);
3771 if (shinfo->gso_type & SKB_GSO_DODGY)
3772 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3775 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3779 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3780 struct sk_buff **to_free,
3781 struct netdev_queue *txq)
3785 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3786 if (rc == NET_XMIT_SUCCESS)
3787 trace_qdisc_enqueue(q, txq, skb);
3791 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3792 struct net_device *dev,
3793 struct netdev_queue *txq)
3795 spinlock_t *root_lock = qdisc_lock(q);
3796 struct sk_buff *to_free = NULL;
3800 qdisc_calculate_pkt_len(skb, q);
3802 if (q->flags & TCQ_F_NOLOCK) {
3803 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3804 qdisc_run_begin(q)) {
3805 /* Retest nolock_qdisc_is_empty() within the protection
3806 * of q->seqlock to protect from racing with requeuing.
3808 if (unlikely(!nolock_qdisc_is_empty(q))) {
3809 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3816 qdisc_bstats_cpu_update(q, skb);
3817 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3818 !nolock_qdisc_is_empty(q))
3822 return NET_XMIT_SUCCESS;
3825 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3829 if (unlikely(to_free))
3830 kfree_skb_list_reason(to_free,
3831 SKB_DROP_REASON_QDISC_DROP);
3836 * Heuristic to force contended enqueues to serialize on a
3837 * separate lock before trying to get qdisc main lock.
3838 * This permits qdisc->running owner to get the lock more
3839 * often and dequeue packets faster.
3840 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3841 * and then other tasks will only enqueue packets. The packets will be
3842 * sent after the qdisc owner is scheduled again. To prevent this
3843 * scenario the task always serialize on the lock.
3845 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3846 if (unlikely(contended))
3847 spin_lock(&q->busylock);
3849 spin_lock(root_lock);
3850 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3851 __qdisc_drop(skb, &to_free);
3853 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3854 qdisc_run_begin(q)) {
3856 * This is a work-conserving queue; there are no old skbs
3857 * waiting to be sent out; and the qdisc is not running -
3858 * xmit the skb directly.
3861 qdisc_bstats_update(q, skb);
3863 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3864 if (unlikely(contended)) {
3865 spin_unlock(&q->busylock);
3872 rc = NET_XMIT_SUCCESS;
3874 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3875 if (qdisc_run_begin(q)) {
3876 if (unlikely(contended)) {
3877 spin_unlock(&q->busylock);
3884 spin_unlock(root_lock);
3885 if (unlikely(to_free))
3886 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3887 if (unlikely(contended))
3888 spin_unlock(&q->busylock);
3892 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3893 static void skb_update_prio(struct sk_buff *skb)
3895 const struct netprio_map *map;
3896 const struct sock *sk;
3897 unsigned int prioidx;
3901 map = rcu_dereference_bh(skb->dev->priomap);
3904 sk = skb_to_full_sk(skb);
3908 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3910 if (prioidx < map->priomap_len)
3911 skb->priority = map->priomap[prioidx];
3914 #define skb_update_prio(skb)
3918 * dev_loopback_xmit - loop back @skb
3919 * @net: network namespace this loopback is happening in
3920 * @sk: sk needed to be a netfilter okfn
3921 * @skb: buffer to transmit
3923 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3925 skb_reset_mac_header(skb);
3926 __skb_pull(skb, skb_network_offset(skb));
3927 skb->pkt_type = PACKET_LOOPBACK;
3928 if (skb->ip_summed == CHECKSUM_NONE)
3929 skb->ip_summed = CHECKSUM_UNNECESSARY;
3930 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3935 EXPORT_SYMBOL(dev_loopback_xmit);
3937 #ifdef CONFIG_NET_EGRESS
3938 static struct sk_buff *
3939 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3941 #ifdef CONFIG_NET_CLS_ACT
3942 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3943 struct tcf_result cl_res;
3948 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3949 tc_skb_cb(skb)->mru = 0;
3950 tc_skb_cb(skb)->post_ct = false;
3951 mini_qdisc_bstats_cpu_update(miniq, skb);
3953 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3955 case TC_ACT_RECLASSIFY:
3956 skb->tc_index = TC_H_MIN(cl_res.classid);
3959 mini_qdisc_qstats_cpu_drop(miniq);
3960 *ret = NET_XMIT_DROP;
3961 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3966 *ret = NET_XMIT_SUCCESS;
3969 case TC_ACT_REDIRECT:
3970 /* No need to push/pop skb's mac_header here on egress! */
3971 skb_do_redirect(skb);
3972 *ret = NET_XMIT_SUCCESS;
3977 #endif /* CONFIG_NET_CLS_ACT */
3982 static struct netdev_queue *
3983 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3985 int qm = skb_get_queue_mapping(skb);
3987 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3990 static bool netdev_xmit_txqueue_skipped(void)
3992 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3995 void netdev_xmit_skip_txqueue(bool skip)
3997 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3999 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
4000 #endif /* CONFIG_NET_EGRESS */
4003 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4004 struct xps_dev_maps *dev_maps, unsigned int tci)
4006 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4007 struct xps_map *map;
4008 int queue_index = -1;
4010 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4013 tci *= dev_maps->num_tc;
4016 map = rcu_dereference(dev_maps->attr_map[tci]);
4019 queue_index = map->queues[0];
4021 queue_index = map->queues[reciprocal_scale(
4022 skb_get_hash(skb), map->len)];
4023 if (unlikely(queue_index >= dev->real_num_tx_queues))
4030 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4031 struct sk_buff *skb)
4034 struct xps_dev_maps *dev_maps;
4035 struct sock *sk = skb->sk;
4036 int queue_index = -1;
4038 if (!static_key_false(&xps_needed))
4042 if (!static_key_false(&xps_rxqs_needed))
4045 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4047 int tci = sk_rx_queue_get(sk);
4050 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4055 if (queue_index < 0) {
4056 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4058 unsigned int tci = skb->sender_cpu - 1;
4060 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4072 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4073 struct net_device *sb_dev)
4077 EXPORT_SYMBOL(dev_pick_tx_zero);
4079 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4080 struct net_device *sb_dev)
4082 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4084 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4086 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4087 struct net_device *sb_dev)
4089 struct sock *sk = skb->sk;
4090 int queue_index = sk_tx_queue_get(sk);
4092 sb_dev = sb_dev ? : dev;
4094 if (queue_index < 0 || skb->ooo_okay ||
4095 queue_index >= dev->real_num_tx_queues) {
4096 int new_index = get_xps_queue(dev, sb_dev, skb);
4099 new_index = skb_tx_hash(dev, sb_dev, skb);
4101 if (queue_index != new_index && sk &&
4103 rcu_access_pointer(sk->sk_dst_cache))
4104 sk_tx_queue_set(sk, new_index);
4106 queue_index = new_index;
4111 EXPORT_SYMBOL(netdev_pick_tx);
4113 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4114 struct sk_buff *skb,
4115 struct net_device *sb_dev)
4117 int queue_index = 0;
4120 u32 sender_cpu = skb->sender_cpu - 1;
4122 if (sender_cpu >= (u32)NR_CPUS)
4123 skb->sender_cpu = raw_smp_processor_id() + 1;
4126 if (dev->real_num_tx_queues != 1) {
4127 const struct net_device_ops *ops = dev->netdev_ops;
4129 if (ops->ndo_select_queue)
4130 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4132 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4134 queue_index = netdev_cap_txqueue(dev, queue_index);
4137 skb_set_queue_mapping(skb, queue_index);
4138 return netdev_get_tx_queue(dev, queue_index);
4142 * __dev_queue_xmit() - transmit a buffer
4143 * @skb: buffer to transmit
4144 * @sb_dev: suboordinate device used for L2 forwarding offload
4146 * Queue a buffer for transmission to a network device. The caller must
4147 * have set the device and priority and built the buffer before calling
4148 * this function. The function can be called from an interrupt.
4150 * When calling this method, interrupts MUST be enabled. This is because
4151 * the BH enable code must have IRQs enabled so that it will not deadlock.
4153 * Regardless of the return value, the skb is consumed, so it is currently
4154 * difficult to retry a send to this method. (You can bump the ref count
4155 * before sending to hold a reference for retry if you are careful.)
4158 * * 0 - buffer successfully transmitted
4159 * * positive qdisc return code - NET_XMIT_DROP etc.
4160 * * negative errno - other errors
4162 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4164 struct net_device *dev = skb->dev;
4165 struct netdev_queue *txq = NULL;
4170 skb_reset_mac_header(skb);
4171 skb_assert_len(skb);
4173 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4174 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4176 /* Disable soft irqs for various locks below. Also
4177 * stops preemption for RCU.
4181 skb_update_prio(skb);
4183 qdisc_pkt_len_init(skb);
4184 #ifdef CONFIG_NET_CLS_ACT
4185 skb->tc_at_ingress = 0;
4187 #ifdef CONFIG_NET_EGRESS
4188 if (static_branch_unlikely(&egress_needed_key)) {
4189 if (nf_hook_egress_active()) {
4190 skb = nf_hook_egress(skb, &rc, dev);
4195 netdev_xmit_skip_txqueue(false);
4197 nf_skip_egress(skb, true);
4198 skb = sch_handle_egress(skb, &rc, dev);
4201 nf_skip_egress(skb, false);
4203 if (netdev_xmit_txqueue_skipped())
4204 txq = netdev_tx_queue_mapping(dev, skb);
4207 /* If device/qdisc don't need skb->dst, release it right now while
4208 * its hot in this cpu cache.
4210 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4216 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4218 q = rcu_dereference_bh(txq->qdisc);
4220 trace_net_dev_queue(skb);
4222 rc = __dev_xmit_skb(skb, q, dev, txq);
4226 /* The device has no queue. Common case for software devices:
4227 * loopback, all the sorts of tunnels...
4229 * Really, it is unlikely that netif_tx_lock protection is necessary
4230 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4232 * However, it is possible, that they rely on protection
4235 * Check this and shot the lock. It is not prone from deadlocks.
4236 *Either shot noqueue qdisc, it is even simpler 8)
4238 if (dev->flags & IFF_UP) {
4239 int cpu = smp_processor_id(); /* ok because BHs are off */
4241 /* Other cpus might concurrently change txq->xmit_lock_owner
4242 * to -1 or to their cpu id, but not to our id.
4244 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4245 if (dev_xmit_recursion())
4246 goto recursion_alert;
4248 skb = validate_xmit_skb(skb, dev, &again);
4252 HARD_TX_LOCK(dev, txq, cpu);
4254 if (!netif_xmit_stopped(txq)) {
4255 dev_xmit_recursion_inc();
4256 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4257 dev_xmit_recursion_dec();
4258 if (dev_xmit_complete(rc)) {
4259 HARD_TX_UNLOCK(dev, txq);
4263 HARD_TX_UNLOCK(dev, txq);
4264 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4267 /* Recursion is detected! It is possible,
4271 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4277 rcu_read_unlock_bh();
4279 dev_core_stats_tx_dropped_inc(dev);
4280 kfree_skb_list(skb);
4283 rcu_read_unlock_bh();
4286 EXPORT_SYMBOL(__dev_queue_xmit);
4288 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4290 struct net_device *dev = skb->dev;
4291 struct sk_buff *orig_skb = skb;
4292 struct netdev_queue *txq;
4293 int ret = NETDEV_TX_BUSY;
4296 if (unlikely(!netif_running(dev) ||
4297 !netif_carrier_ok(dev)))
4300 skb = validate_xmit_skb_list(skb, dev, &again);
4301 if (skb != orig_skb)
4304 skb_set_queue_mapping(skb, queue_id);
4305 txq = skb_get_tx_queue(dev, skb);
4309 dev_xmit_recursion_inc();
4310 HARD_TX_LOCK(dev, txq, smp_processor_id());
4311 if (!netif_xmit_frozen_or_drv_stopped(txq))
4312 ret = netdev_start_xmit(skb, dev, txq, false);
4313 HARD_TX_UNLOCK(dev, txq);
4314 dev_xmit_recursion_dec();
4319 dev_core_stats_tx_dropped_inc(dev);
4320 kfree_skb_list(skb);
4321 return NET_XMIT_DROP;
4323 EXPORT_SYMBOL(__dev_direct_xmit);
4325 /*************************************************************************
4327 *************************************************************************/
4329 int netdev_max_backlog __read_mostly = 1000;
4330 EXPORT_SYMBOL(netdev_max_backlog);
4332 int netdev_tstamp_prequeue __read_mostly = 1;
4333 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4334 int netdev_budget __read_mostly = 300;
4335 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4336 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4337 int weight_p __read_mostly = 64; /* old backlog weight */
4338 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4339 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4340 int dev_rx_weight __read_mostly = 64;
4341 int dev_tx_weight __read_mostly = 64;
4343 /* Called with irq disabled */
4344 static inline void ____napi_schedule(struct softnet_data *sd,
4345 struct napi_struct *napi)
4347 struct task_struct *thread;
4349 lockdep_assert_irqs_disabled();
4351 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4352 /* Paired with smp_mb__before_atomic() in
4353 * napi_enable()/dev_set_threaded().
4354 * Use READ_ONCE() to guarantee a complete
4355 * read on napi->thread. Only call
4356 * wake_up_process() when it's not NULL.
4358 thread = READ_ONCE(napi->thread);
4360 /* Avoid doing set_bit() if the thread is in
4361 * INTERRUPTIBLE state, cause napi_thread_wait()
4362 * makes sure to proceed with napi polling
4363 * if the thread is explicitly woken from here.
4365 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4366 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4367 wake_up_process(thread);
4372 list_add_tail(&napi->poll_list, &sd->poll_list);
4373 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4378 /* One global table that all flow-based protocols share. */
4379 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4380 EXPORT_SYMBOL(rps_sock_flow_table);
4381 u32 rps_cpu_mask __read_mostly;
4382 EXPORT_SYMBOL(rps_cpu_mask);
4384 struct static_key_false rps_needed __read_mostly;
4385 EXPORT_SYMBOL(rps_needed);
4386 struct static_key_false rfs_needed __read_mostly;
4387 EXPORT_SYMBOL(rfs_needed);
4389 static struct rps_dev_flow *
4390 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4391 struct rps_dev_flow *rflow, u16 next_cpu)
4393 if (next_cpu < nr_cpu_ids) {
4394 #ifdef CONFIG_RFS_ACCEL
4395 struct netdev_rx_queue *rxqueue;
4396 struct rps_dev_flow_table *flow_table;
4397 struct rps_dev_flow *old_rflow;
4402 /* Should we steer this flow to a different hardware queue? */
4403 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4404 !(dev->features & NETIF_F_NTUPLE))
4406 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4407 if (rxq_index == skb_get_rx_queue(skb))
4410 rxqueue = dev->_rx + rxq_index;
4411 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4414 flow_id = skb_get_hash(skb) & flow_table->mask;
4415 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4416 rxq_index, flow_id);
4420 rflow = &flow_table->flows[flow_id];
4422 if (old_rflow->filter == rflow->filter)
4423 old_rflow->filter = RPS_NO_FILTER;
4427 per_cpu(softnet_data, next_cpu).input_queue_head;
4430 rflow->cpu = next_cpu;
4435 * get_rps_cpu is called from netif_receive_skb and returns the target
4436 * CPU from the RPS map of the receiving queue for a given skb.
4437 * rcu_read_lock must be held on entry.
4439 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4440 struct rps_dev_flow **rflowp)
4442 const struct rps_sock_flow_table *sock_flow_table;
4443 struct netdev_rx_queue *rxqueue = dev->_rx;
4444 struct rps_dev_flow_table *flow_table;
4445 struct rps_map *map;
4450 if (skb_rx_queue_recorded(skb)) {
4451 u16 index = skb_get_rx_queue(skb);
4453 if (unlikely(index >= dev->real_num_rx_queues)) {
4454 WARN_ONCE(dev->real_num_rx_queues > 1,
4455 "%s received packet on queue %u, but number "
4456 "of RX queues is %u\n",
4457 dev->name, index, dev->real_num_rx_queues);
4463 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4465 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4466 map = rcu_dereference(rxqueue->rps_map);
4467 if (!flow_table && !map)
4470 skb_reset_network_header(skb);
4471 hash = skb_get_hash(skb);
4475 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4476 if (flow_table && sock_flow_table) {
4477 struct rps_dev_flow *rflow;
4481 /* First check into global flow table if there is a match */
4482 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4483 if ((ident ^ hash) & ~rps_cpu_mask)
4486 next_cpu = ident & rps_cpu_mask;
4488 /* OK, now we know there is a match,
4489 * we can look at the local (per receive queue) flow table
4491 rflow = &flow_table->flows[hash & flow_table->mask];
4495 * If the desired CPU (where last recvmsg was done) is
4496 * different from current CPU (one in the rx-queue flow
4497 * table entry), switch if one of the following holds:
4498 * - Current CPU is unset (>= nr_cpu_ids).
4499 * - Current CPU is offline.
4500 * - The current CPU's queue tail has advanced beyond the
4501 * last packet that was enqueued using this table entry.
4502 * This guarantees that all previous packets for the flow
4503 * have been dequeued, thus preserving in order delivery.
4505 if (unlikely(tcpu != next_cpu) &&
4506 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4507 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4508 rflow->last_qtail)) >= 0)) {
4510 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4513 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4523 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4524 if (cpu_online(tcpu)) {
4534 #ifdef CONFIG_RFS_ACCEL
4537 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4538 * @dev: Device on which the filter was set
4539 * @rxq_index: RX queue index
4540 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4541 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4543 * Drivers that implement ndo_rx_flow_steer() should periodically call
4544 * this function for each installed filter and remove the filters for
4545 * which it returns %true.
4547 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4548 u32 flow_id, u16 filter_id)
4550 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4551 struct rps_dev_flow_table *flow_table;
4552 struct rps_dev_flow *rflow;
4557 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4558 if (flow_table && flow_id <= flow_table->mask) {
4559 rflow = &flow_table->flows[flow_id];
4560 cpu = READ_ONCE(rflow->cpu);
4561 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4562 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4563 rflow->last_qtail) <
4564 (int)(10 * flow_table->mask)))
4570 EXPORT_SYMBOL(rps_may_expire_flow);
4572 #endif /* CONFIG_RFS_ACCEL */
4574 /* Called from hardirq (IPI) context */
4575 static void rps_trigger_softirq(void *data)
4577 struct softnet_data *sd = data;
4579 ____napi_schedule(sd, &sd->backlog);
4583 #endif /* CONFIG_RPS */
4585 /* Called from hardirq (IPI) context */
4586 static void trigger_rx_softirq(void *data)
4588 struct softnet_data *sd = data;
4590 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4591 smp_store_release(&sd->defer_ipi_scheduled, 0);
4595 * Check if this softnet_data structure is another cpu one
4596 * If yes, queue it to our IPI list and return 1
4599 static int napi_schedule_rps(struct softnet_data *sd)
4601 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4605 sd->rps_ipi_next = mysd->rps_ipi_list;
4606 mysd->rps_ipi_list = sd;
4608 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4611 #endif /* CONFIG_RPS */
4612 __napi_schedule_irqoff(&mysd->backlog);
4616 #ifdef CONFIG_NET_FLOW_LIMIT
4617 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4620 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4622 #ifdef CONFIG_NET_FLOW_LIMIT
4623 struct sd_flow_limit *fl;
4624 struct softnet_data *sd;
4625 unsigned int old_flow, new_flow;
4627 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4630 sd = this_cpu_ptr(&softnet_data);
4633 fl = rcu_dereference(sd->flow_limit);
4635 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4636 old_flow = fl->history[fl->history_head];
4637 fl->history[fl->history_head] = new_flow;
4640 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4642 if (likely(fl->buckets[old_flow]))
4643 fl->buckets[old_flow]--;
4645 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4657 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4658 * queue (may be a remote CPU queue).
4660 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4661 unsigned int *qtail)
4663 enum skb_drop_reason reason;
4664 struct softnet_data *sd;
4665 unsigned long flags;
4668 reason = SKB_DROP_REASON_NOT_SPECIFIED;
4669 sd = &per_cpu(softnet_data, cpu);
4671 rps_lock_irqsave(sd, &flags);
4672 if (!netif_running(skb->dev))
4674 qlen = skb_queue_len(&sd->input_pkt_queue);
4675 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4678 __skb_queue_tail(&sd->input_pkt_queue, skb);
4679 input_queue_tail_incr_save(sd, qtail);
4680 rps_unlock_irq_restore(sd, &flags);
4681 return NET_RX_SUCCESS;
4684 /* Schedule NAPI for backlog device
4685 * We can use non atomic operation since we own the queue lock
4687 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4688 napi_schedule_rps(sd);
4691 reason = SKB_DROP_REASON_CPU_BACKLOG;
4695 rps_unlock_irq_restore(sd, &flags);
4697 dev_core_stats_rx_dropped_inc(skb->dev);
4698 kfree_skb_reason(skb, reason);
4702 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4704 struct net_device *dev = skb->dev;
4705 struct netdev_rx_queue *rxqueue;
4709 if (skb_rx_queue_recorded(skb)) {
4710 u16 index = skb_get_rx_queue(skb);
4712 if (unlikely(index >= dev->real_num_rx_queues)) {
4713 WARN_ONCE(dev->real_num_rx_queues > 1,
4714 "%s received packet on queue %u, but number "
4715 "of RX queues is %u\n",
4716 dev->name, index, dev->real_num_rx_queues);
4718 return rxqueue; /* Return first rxqueue */
4725 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4726 struct bpf_prog *xdp_prog)
4728 void *orig_data, *orig_data_end, *hard_start;
4729 struct netdev_rx_queue *rxqueue;
4730 bool orig_bcast, orig_host;
4731 u32 mac_len, frame_sz;
4732 __be16 orig_eth_type;
4737 /* The XDP program wants to see the packet starting at the MAC
4740 mac_len = skb->data - skb_mac_header(skb);
4741 hard_start = skb->data - skb_headroom(skb);
4743 /* SKB "head" area always have tailroom for skb_shared_info */
4744 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4745 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4747 rxqueue = netif_get_rxqueue(skb);
4748 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4749 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4750 skb_headlen(skb) + mac_len, true);
4752 orig_data_end = xdp->data_end;
4753 orig_data = xdp->data;
4754 eth = (struct ethhdr *)xdp->data;
4755 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4756 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4757 orig_eth_type = eth->h_proto;
4759 act = bpf_prog_run_xdp(xdp_prog, xdp);
4761 /* check if bpf_xdp_adjust_head was used */
4762 off = xdp->data - orig_data;
4765 __skb_pull(skb, off);
4767 __skb_push(skb, -off);
4769 skb->mac_header += off;
4770 skb_reset_network_header(skb);
4773 /* check if bpf_xdp_adjust_tail was used */
4774 off = xdp->data_end - orig_data_end;
4776 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4777 skb->len += off; /* positive on grow, negative on shrink */
4780 /* check if XDP changed eth hdr such SKB needs update */
4781 eth = (struct ethhdr *)xdp->data;
4782 if ((orig_eth_type != eth->h_proto) ||
4783 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4784 skb->dev->dev_addr)) ||
4785 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4786 __skb_push(skb, ETH_HLEN);
4787 skb->pkt_type = PACKET_HOST;
4788 skb->protocol = eth_type_trans(skb, skb->dev);
4791 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4792 * before calling us again on redirect path. We do not call do_redirect
4793 * as we leave that up to the caller.
4795 * Caller is responsible for managing lifetime of skb (i.e. calling
4796 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4801 __skb_push(skb, mac_len);
4804 metalen = xdp->data - xdp->data_meta;
4806 skb_metadata_set(skb, metalen);
4813 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4814 struct xdp_buff *xdp,
4815 struct bpf_prog *xdp_prog)
4819 /* Reinjected packets coming from act_mirred or similar should
4820 * not get XDP generic processing.
4822 if (skb_is_redirected(skb))
4825 /* XDP packets must be linear and must have sufficient headroom
4826 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4827 * native XDP provides, thus we need to do it here as well.
4829 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4830 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4831 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4832 int troom = skb->tail + skb->data_len - skb->end;
4834 /* In case we have to go down the path and also linearize,
4835 * then lets do the pskb_expand_head() work just once here.
4837 if (pskb_expand_head(skb,
4838 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4839 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4841 if (skb_linearize(skb))
4845 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4852 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4855 trace_xdp_exception(skb->dev, xdp_prog, act);
4866 /* When doing generic XDP we have to bypass the qdisc layer and the
4867 * network taps in order to match in-driver-XDP behavior. This also means
4868 * that XDP packets are able to starve other packets going through a qdisc,
4869 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4870 * queues, so they do not have this starvation issue.
4872 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4874 struct net_device *dev = skb->dev;
4875 struct netdev_queue *txq;
4876 bool free_skb = true;
4879 txq = netdev_core_pick_tx(dev, skb, NULL);
4880 cpu = smp_processor_id();
4881 HARD_TX_LOCK(dev, txq, cpu);
4882 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4883 rc = netdev_start_xmit(skb, dev, txq, 0);
4884 if (dev_xmit_complete(rc))
4887 HARD_TX_UNLOCK(dev, txq);
4889 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4890 dev_core_stats_tx_dropped_inc(dev);
4895 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4897 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4900 struct xdp_buff xdp;
4904 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4905 if (act != XDP_PASS) {
4908 err = xdp_do_generic_redirect(skb->dev, skb,
4914 generic_xdp_tx(skb, xdp_prog);
4922 kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4925 EXPORT_SYMBOL_GPL(do_xdp_generic);
4927 static int netif_rx_internal(struct sk_buff *skb)
4931 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4933 trace_netif_rx(skb);
4936 if (static_branch_unlikely(&rps_needed)) {
4937 struct rps_dev_flow voidflow, *rflow = &voidflow;
4942 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4944 cpu = smp_processor_id();
4946 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4954 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4960 * __netif_rx - Slightly optimized version of netif_rx
4961 * @skb: buffer to post
4963 * This behaves as netif_rx except that it does not disable bottom halves.
4964 * As a result this function may only be invoked from the interrupt context
4965 * (either hard or soft interrupt).
4967 int __netif_rx(struct sk_buff *skb)
4971 lockdep_assert_once(hardirq_count() | softirq_count());
4973 trace_netif_rx_entry(skb);
4974 ret = netif_rx_internal(skb);
4975 trace_netif_rx_exit(ret);
4978 EXPORT_SYMBOL(__netif_rx);
4981 * netif_rx - post buffer to the network code
4982 * @skb: buffer to post
4984 * This function receives a packet from a device driver and queues it for
4985 * the upper (protocol) levels to process via the backlog NAPI device. It
4986 * always succeeds. The buffer may be dropped during processing for
4987 * congestion control or by the protocol layers.
4988 * The network buffer is passed via the backlog NAPI device. Modern NIC
4989 * driver should use NAPI and GRO.
4990 * This function can used from interrupt and from process context. The
4991 * caller from process context must not disable interrupts before invoking
4995 * NET_RX_SUCCESS (no congestion)
4996 * NET_RX_DROP (packet was dropped)
4999 int netif_rx(struct sk_buff *skb)
5001 bool need_bh_off = !(hardirq_count() | softirq_count());
5006 trace_netif_rx_entry(skb);
5007 ret = netif_rx_internal(skb);
5008 trace_netif_rx_exit(ret);
5013 EXPORT_SYMBOL(netif_rx);
5015 static __latent_entropy void net_tx_action(struct softirq_action *h)
5017 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5019 if (sd->completion_queue) {
5020 struct sk_buff *clist;
5022 local_irq_disable();
5023 clist = sd->completion_queue;
5024 sd->completion_queue = NULL;
5028 struct sk_buff *skb = clist;
5030 clist = clist->next;
5032 WARN_ON(refcount_read(&skb->users));
5033 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5034 trace_consume_skb(skb);
5036 trace_kfree_skb(skb, net_tx_action,
5037 SKB_DROP_REASON_NOT_SPECIFIED);
5039 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5042 __kfree_skb_defer(skb);
5046 if (sd->output_queue) {
5049 local_irq_disable();
5050 head = sd->output_queue;
5051 sd->output_queue = NULL;
5052 sd->output_queue_tailp = &sd->output_queue;
5058 struct Qdisc *q = head;
5059 spinlock_t *root_lock = NULL;
5061 head = head->next_sched;
5063 /* We need to make sure head->next_sched is read
5064 * before clearing __QDISC_STATE_SCHED
5066 smp_mb__before_atomic();
5068 if (!(q->flags & TCQ_F_NOLOCK)) {
5069 root_lock = qdisc_lock(q);
5070 spin_lock(root_lock);
5071 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5073 /* There is a synchronize_net() between
5074 * STATE_DEACTIVATED flag being set and
5075 * qdisc_reset()/some_qdisc_is_busy() in
5076 * dev_deactivate(), so we can safely bail out
5077 * early here to avoid data race between
5078 * qdisc_deactivate() and some_qdisc_is_busy()
5079 * for lockless qdisc.
5081 clear_bit(__QDISC_STATE_SCHED, &q->state);
5085 clear_bit(__QDISC_STATE_SCHED, &q->state);
5088 spin_unlock(root_lock);
5094 xfrm_dev_backlog(sd);
5097 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5098 /* This hook is defined here for ATM LANE */
5099 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5100 unsigned char *addr) __read_mostly;
5101 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5104 static inline struct sk_buff *
5105 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5106 struct net_device *orig_dev, bool *another)
5108 #ifdef CONFIG_NET_CLS_ACT
5109 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5110 struct tcf_result cl_res;
5112 /* If there's at least one ingress present somewhere (so
5113 * we get here via enabled static key), remaining devices
5114 * that are not configured with an ingress qdisc will bail
5121 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5125 qdisc_skb_cb(skb)->pkt_len = skb->len;
5126 tc_skb_cb(skb)->mru = 0;
5127 tc_skb_cb(skb)->post_ct = false;
5128 skb->tc_at_ingress = 1;
5129 mini_qdisc_bstats_cpu_update(miniq, skb);
5131 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5133 case TC_ACT_RECLASSIFY:
5134 skb->tc_index = TC_H_MIN(cl_res.classid);
5137 mini_qdisc_qstats_cpu_drop(miniq);
5138 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5145 *ret = NET_RX_SUCCESS;
5147 case TC_ACT_REDIRECT:
5148 /* skb_mac_header check was done by cls/act_bpf, so
5149 * we can safely push the L2 header back before
5150 * redirecting to another netdev
5152 __skb_push(skb, skb->mac_len);
5153 if (skb_do_redirect(skb) == -EAGAIN) {
5154 __skb_pull(skb, skb->mac_len);
5158 *ret = NET_RX_SUCCESS;
5160 case TC_ACT_CONSUMED:
5161 *ret = NET_RX_SUCCESS;
5166 #endif /* CONFIG_NET_CLS_ACT */
5171 * netdev_is_rx_handler_busy - check if receive handler is registered
5172 * @dev: device to check
5174 * Check if a receive handler is already registered for a given device.
5175 * Return true if there one.
5177 * The caller must hold the rtnl_mutex.
5179 bool netdev_is_rx_handler_busy(struct net_device *dev)
5182 return dev && rtnl_dereference(dev->rx_handler);
5184 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5187 * netdev_rx_handler_register - register receive handler
5188 * @dev: device to register a handler for
5189 * @rx_handler: receive handler to register
5190 * @rx_handler_data: data pointer that is used by rx handler
5192 * Register a receive handler for a device. This handler will then be
5193 * called from __netif_receive_skb. A negative errno code is returned
5196 * The caller must hold the rtnl_mutex.
5198 * For a general description of rx_handler, see enum rx_handler_result.
5200 int netdev_rx_handler_register(struct net_device *dev,
5201 rx_handler_func_t *rx_handler,
5202 void *rx_handler_data)
5204 if (netdev_is_rx_handler_busy(dev))
5207 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5210 /* Note: rx_handler_data must be set before rx_handler */
5211 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5212 rcu_assign_pointer(dev->rx_handler, rx_handler);
5216 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5219 * netdev_rx_handler_unregister - unregister receive handler
5220 * @dev: device to unregister a handler from
5222 * Unregister a receive handler from a device.
5224 * The caller must hold the rtnl_mutex.
5226 void netdev_rx_handler_unregister(struct net_device *dev)
5230 RCU_INIT_POINTER(dev->rx_handler, NULL);
5231 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5232 * section has a guarantee to see a non NULL rx_handler_data
5236 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5238 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5241 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5242 * the special handling of PFMEMALLOC skbs.
5244 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5246 switch (skb->protocol) {
5247 case htons(ETH_P_ARP):
5248 case htons(ETH_P_IP):
5249 case htons(ETH_P_IPV6):
5250 case htons(ETH_P_8021Q):
5251 case htons(ETH_P_8021AD):
5258 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5259 int *ret, struct net_device *orig_dev)
5261 if (nf_hook_ingress_active(skb)) {
5265 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5270 ingress_retval = nf_hook_ingress(skb);
5272 return ingress_retval;
5277 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5278 struct packet_type **ppt_prev)
5280 struct packet_type *ptype, *pt_prev;
5281 rx_handler_func_t *rx_handler;
5282 struct sk_buff *skb = *pskb;
5283 struct net_device *orig_dev;
5284 bool deliver_exact = false;
5285 int ret = NET_RX_DROP;
5288 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5290 trace_netif_receive_skb(skb);
5292 orig_dev = skb->dev;
5294 skb_reset_network_header(skb);
5295 if (!skb_transport_header_was_set(skb))
5296 skb_reset_transport_header(skb);
5297 skb_reset_mac_len(skb);
5302 skb->skb_iif = skb->dev->ifindex;
5304 __this_cpu_inc(softnet_data.processed);
5306 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5310 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5313 if (ret2 != XDP_PASS) {
5319 if (eth_type_vlan(skb->protocol)) {
5320 skb = skb_vlan_untag(skb);
5325 if (skb_skip_tc_classify(skb))
5331 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5333 ret = deliver_skb(skb, pt_prev, orig_dev);
5337 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5339 ret = deliver_skb(skb, pt_prev, orig_dev);
5344 #ifdef CONFIG_NET_INGRESS
5345 if (static_branch_unlikely(&ingress_needed_key)) {
5346 bool another = false;
5348 nf_skip_egress(skb, true);
5349 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5356 nf_skip_egress(skb, false);
5357 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5361 skb_reset_redirect(skb);
5363 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5366 if (skb_vlan_tag_present(skb)) {
5368 ret = deliver_skb(skb, pt_prev, orig_dev);
5371 if (vlan_do_receive(&skb))
5373 else if (unlikely(!skb))
5377 rx_handler = rcu_dereference(skb->dev->rx_handler);
5380 ret = deliver_skb(skb, pt_prev, orig_dev);
5383 switch (rx_handler(&skb)) {
5384 case RX_HANDLER_CONSUMED:
5385 ret = NET_RX_SUCCESS;
5387 case RX_HANDLER_ANOTHER:
5389 case RX_HANDLER_EXACT:
5390 deliver_exact = true;
5392 case RX_HANDLER_PASS:
5399 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5401 if (skb_vlan_tag_get_id(skb)) {
5402 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5405 skb->pkt_type = PACKET_OTHERHOST;
5406 } else if (eth_type_vlan(skb->protocol)) {
5407 /* Outer header is 802.1P with vlan 0, inner header is
5408 * 802.1Q or 802.1AD and vlan_do_receive() above could
5409 * not find vlan dev for vlan id 0.
5411 __vlan_hwaccel_clear_tag(skb);
5412 skb = skb_vlan_untag(skb);
5415 if (vlan_do_receive(&skb))
5416 /* After stripping off 802.1P header with vlan 0
5417 * vlan dev is found for inner header.
5420 else if (unlikely(!skb))
5423 /* We have stripped outer 802.1P vlan 0 header.
5424 * But could not find vlan dev.
5425 * check again for vlan id to set OTHERHOST.
5429 /* Note: we might in the future use prio bits
5430 * and set skb->priority like in vlan_do_receive()
5431 * For the time being, just ignore Priority Code Point
5433 __vlan_hwaccel_clear_tag(skb);
5436 type = skb->protocol;
5438 /* deliver only exact match when indicated */
5439 if (likely(!deliver_exact)) {
5440 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5441 &ptype_base[ntohs(type) &
5445 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5446 &orig_dev->ptype_specific);
5448 if (unlikely(skb->dev != orig_dev)) {
5449 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5450 &skb->dev->ptype_specific);
5454 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5456 *ppt_prev = pt_prev;
5460 dev_core_stats_rx_dropped_inc(skb->dev);
5462 dev_core_stats_rx_nohandler_inc(skb->dev);
5463 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5464 /* Jamal, now you will not able to escape explaining
5465 * me how you were going to use this. :-)
5471 /* The invariant here is that if *ppt_prev is not NULL
5472 * then skb should also be non-NULL.
5474 * Apparently *ppt_prev assignment above holds this invariant due to
5475 * skb dereferencing near it.
5481 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5483 struct net_device *orig_dev = skb->dev;
5484 struct packet_type *pt_prev = NULL;
5487 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5489 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5490 skb->dev, pt_prev, orig_dev);
5495 * netif_receive_skb_core - special purpose version of netif_receive_skb
5496 * @skb: buffer to process
5498 * More direct receive version of netif_receive_skb(). It should
5499 * only be used by callers that have a need to skip RPS and Generic XDP.
5500 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5502 * This function may only be called from softirq context and interrupts
5503 * should be enabled.
5505 * Return values (usually ignored):
5506 * NET_RX_SUCCESS: no congestion
5507 * NET_RX_DROP: packet was dropped
5509 int netif_receive_skb_core(struct sk_buff *skb)
5514 ret = __netif_receive_skb_one_core(skb, false);
5519 EXPORT_SYMBOL(netif_receive_skb_core);
5521 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5522 struct packet_type *pt_prev,
5523 struct net_device *orig_dev)
5525 struct sk_buff *skb, *next;
5529 if (list_empty(head))
5531 if (pt_prev->list_func != NULL)
5532 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5533 ip_list_rcv, head, pt_prev, orig_dev);
5535 list_for_each_entry_safe(skb, next, head, list) {
5536 skb_list_del_init(skb);
5537 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5541 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5543 /* Fast-path assumptions:
5544 * - There is no RX handler.
5545 * - Only one packet_type matches.
5546 * If either of these fails, we will end up doing some per-packet
5547 * processing in-line, then handling the 'last ptype' for the whole
5548 * sublist. This can't cause out-of-order delivery to any single ptype,
5549 * because the 'last ptype' must be constant across the sublist, and all
5550 * other ptypes are handled per-packet.
5552 /* Current (common) ptype of sublist */
5553 struct packet_type *pt_curr = NULL;
5554 /* Current (common) orig_dev of sublist */
5555 struct net_device *od_curr = NULL;
5556 struct list_head sublist;
5557 struct sk_buff *skb, *next;
5559 INIT_LIST_HEAD(&sublist);
5560 list_for_each_entry_safe(skb, next, head, list) {
5561 struct net_device *orig_dev = skb->dev;
5562 struct packet_type *pt_prev = NULL;
5564 skb_list_del_init(skb);
5565 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5568 if (pt_curr != pt_prev || od_curr != orig_dev) {
5569 /* dispatch old sublist */
5570 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5571 /* start new sublist */
5572 INIT_LIST_HEAD(&sublist);
5576 list_add_tail(&skb->list, &sublist);
5579 /* dispatch final sublist */
5580 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5583 static int __netif_receive_skb(struct sk_buff *skb)
5587 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5588 unsigned int noreclaim_flag;
5591 * PFMEMALLOC skbs are special, they should
5592 * - be delivered to SOCK_MEMALLOC sockets only
5593 * - stay away from userspace
5594 * - have bounded memory usage
5596 * Use PF_MEMALLOC as this saves us from propagating the allocation
5597 * context down to all allocation sites.
5599 noreclaim_flag = memalloc_noreclaim_save();
5600 ret = __netif_receive_skb_one_core(skb, true);
5601 memalloc_noreclaim_restore(noreclaim_flag);
5603 ret = __netif_receive_skb_one_core(skb, false);
5608 static void __netif_receive_skb_list(struct list_head *head)
5610 unsigned long noreclaim_flag = 0;
5611 struct sk_buff *skb, *next;
5612 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5614 list_for_each_entry_safe(skb, next, head, list) {
5615 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5616 struct list_head sublist;
5618 /* Handle the previous sublist */
5619 list_cut_before(&sublist, head, &skb->list);
5620 if (!list_empty(&sublist))
5621 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5622 pfmemalloc = !pfmemalloc;
5623 /* See comments in __netif_receive_skb */
5625 noreclaim_flag = memalloc_noreclaim_save();
5627 memalloc_noreclaim_restore(noreclaim_flag);
5630 /* Handle the remaining sublist */
5631 if (!list_empty(head))
5632 __netif_receive_skb_list_core(head, pfmemalloc);
5633 /* Restore pflags */
5635 memalloc_noreclaim_restore(noreclaim_flag);
5638 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5640 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5641 struct bpf_prog *new = xdp->prog;
5644 switch (xdp->command) {
5645 case XDP_SETUP_PROG:
5646 rcu_assign_pointer(dev->xdp_prog, new);
5651 static_branch_dec(&generic_xdp_needed_key);
5652 } else if (new && !old) {
5653 static_branch_inc(&generic_xdp_needed_key);
5654 dev_disable_lro(dev);
5655 dev_disable_gro_hw(dev);
5667 static int netif_receive_skb_internal(struct sk_buff *skb)
5671 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5673 if (skb_defer_rx_timestamp(skb))
5674 return NET_RX_SUCCESS;
5678 if (static_branch_unlikely(&rps_needed)) {
5679 struct rps_dev_flow voidflow, *rflow = &voidflow;
5680 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5683 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5689 ret = __netif_receive_skb(skb);
5694 void netif_receive_skb_list_internal(struct list_head *head)
5696 struct sk_buff *skb, *next;
5697 struct list_head sublist;
5699 INIT_LIST_HEAD(&sublist);
5700 list_for_each_entry_safe(skb, next, head, list) {
5701 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5702 skb_list_del_init(skb);
5703 if (!skb_defer_rx_timestamp(skb))
5704 list_add_tail(&skb->list, &sublist);
5706 list_splice_init(&sublist, head);
5710 if (static_branch_unlikely(&rps_needed)) {
5711 list_for_each_entry_safe(skb, next, head, list) {
5712 struct rps_dev_flow voidflow, *rflow = &voidflow;
5713 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5716 /* Will be handled, remove from list */
5717 skb_list_del_init(skb);
5718 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5723 __netif_receive_skb_list(head);
5728 * netif_receive_skb - process receive buffer from network
5729 * @skb: buffer to process
5731 * netif_receive_skb() is the main receive data processing function.
5732 * It always succeeds. The buffer may be dropped during processing
5733 * for congestion control or by the protocol layers.
5735 * This function may only be called from softirq context and interrupts
5736 * should be enabled.
5738 * Return values (usually ignored):
5739 * NET_RX_SUCCESS: no congestion
5740 * NET_RX_DROP: packet was dropped
5742 int netif_receive_skb(struct sk_buff *skb)
5746 trace_netif_receive_skb_entry(skb);
5748 ret = netif_receive_skb_internal(skb);
5749 trace_netif_receive_skb_exit(ret);
5753 EXPORT_SYMBOL(netif_receive_skb);
5756 * netif_receive_skb_list - process many receive buffers from network
5757 * @head: list of skbs to process.
5759 * Since return value of netif_receive_skb() is normally ignored, and
5760 * wouldn't be meaningful for a list, this function returns void.
5762 * This function may only be called from softirq context and interrupts
5763 * should be enabled.
5765 void netif_receive_skb_list(struct list_head *head)
5767 struct sk_buff *skb;
5769 if (list_empty(head))
5771 if (trace_netif_receive_skb_list_entry_enabled()) {
5772 list_for_each_entry(skb, head, list)
5773 trace_netif_receive_skb_list_entry(skb);
5775 netif_receive_skb_list_internal(head);
5776 trace_netif_receive_skb_list_exit(0);
5778 EXPORT_SYMBOL(netif_receive_skb_list);
5780 static DEFINE_PER_CPU(struct work_struct, flush_works);
5782 /* Network device is going away, flush any packets still pending */
5783 static void flush_backlog(struct work_struct *work)
5785 struct sk_buff *skb, *tmp;
5786 struct softnet_data *sd;
5789 sd = this_cpu_ptr(&softnet_data);
5791 rps_lock_irq_disable(sd);
5792 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5793 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5794 __skb_unlink(skb, &sd->input_pkt_queue);
5795 dev_kfree_skb_irq(skb);
5796 input_queue_head_incr(sd);
5799 rps_unlock_irq_enable(sd);
5801 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5802 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5803 __skb_unlink(skb, &sd->process_queue);
5805 input_queue_head_incr(sd);
5811 static bool flush_required(int cpu)
5813 #if IS_ENABLED(CONFIG_RPS)
5814 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5817 rps_lock_irq_disable(sd);
5819 /* as insertion into process_queue happens with the rps lock held,
5820 * process_queue access may race only with dequeue
5822 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5823 !skb_queue_empty_lockless(&sd->process_queue);
5824 rps_unlock_irq_enable(sd);
5828 /* without RPS we can't safely check input_pkt_queue: during a
5829 * concurrent remote skb_queue_splice() we can detect as empty both
5830 * input_pkt_queue and process_queue even if the latter could end-up
5831 * containing a lot of packets.
5836 static void flush_all_backlogs(void)
5838 static cpumask_t flush_cpus;
5841 /* since we are under rtnl lock protection we can use static data
5842 * for the cpumask and avoid allocating on stack the possibly
5849 cpumask_clear(&flush_cpus);
5850 for_each_online_cpu(cpu) {
5851 if (flush_required(cpu)) {
5852 queue_work_on(cpu, system_highpri_wq,
5853 per_cpu_ptr(&flush_works, cpu));
5854 cpumask_set_cpu(cpu, &flush_cpus);
5858 /* we can have in flight packet[s] on the cpus we are not flushing,
5859 * synchronize_net() in unregister_netdevice_many() will take care of
5862 for_each_cpu(cpu, &flush_cpus)
5863 flush_work(per_cpu_ptr(&flush_works, cpu));
5868 static void net_rps_send_ipi(struct softnet_data *remsd)
5872 struct softnet_data *next = remsd->rps_ipi_next;
5874 if (cpu_online(remsd->cpu))
5875 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5882 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5883 * Note: called with local irq disabled, but exits with local irq enabled.
5885 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5888 struct softnet_data *remsd = sd->rps_ipi_list;
5891 sd->rps_ipi_list = NULL;
5895 /* Send pending IPI's to kick RPS processing on remote cpus. */
5896 net_rps_send_ipi(remsd);
5902 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5905 return sd->rps_ipi_list != NULL;
5911 static int process_backlog(struct napi_struct *napi, int quota)
5913 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5917 /* Check if we have pending ipi, its better to send them now,
5918 * not waiting net_rx_action() end.
5920 if (sd_has_rps_ipi_waiting(sd)) {
5921 local_irq_disable();
5922 net_rps_action_and_irq_enable(sd);
5925 napi->weight = READ_ONCE(dev_rx_weight);
5927 struct sk_buff *skb;
5929 while ((skb = __skb_dequeue(&sd->process_queue))) {
5931 __netif_receive_skb(skb);
5933 input_queue_head_incr(sd);
5934 if (++work >= quota)
5939 rps_lock_irq_disable(sd);
5940 if (skb_queue_empty(&sd->input_pkt_queue)) {
5942 * Inline a custom version of __napi_complete().
5943 * only current cpu owns and manipulates this napi,
5944 * and NAPI_STATE_SCHED is the only possible flag set
5946 * We can use a plain write instead of clear_bit(),
5947 * and we dont need an smp_mb() memory barrier.
5952 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5953 &sd->process_queue);
5955 rps_unlock_irq_enable(sd);
5962 * __napi_schedule - schedule for receive
5963 * @n: entry to schedule
5965 * The entry's receive function will be scheduled to run.
5966 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5968 void __napi_schedule(struct napi_struct *n)
5970 unsigned long flags;
5972 local_irq_save(flags);
5973 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5974 local_irq_restore(flags);
5976 EXPORT_SYMBOL(__napi_schedule);
5979 * napi_schedule_prep - check if napi can be scheduled
5982 * Test if NAPI routine is already running, and if not mark
5983 * it as running. This is used as a condition variable to
5984 * insure only one NAPI poll instance runs. We also make
5985 * sure there is no pending NAPI disable.
5987 bool napi_schedule_prep(struct napi_struct *n)
5989 unsigned long val, new;
5992 val = READ_ONCE(n->state);
5993 if (unlikely(val & NAPIF_STATE_DISABLE))
5995 new = val | NAPIF_STATE_SCHED;
5997 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5998 * This was suggested by Alexander Duyck, as compiler
5999 * emits better code than :
6000 * if (val & NAPIF_STATE_SCHED)
6001 * new |= NAPIF_STATE_MISSED;
6003 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6005 } while (cmpxchg(&n->state, val, new) != val);
6007 return !(val & NAPIF_STATE_SCHED);
6009 EXPORT_SYMBOL(napi_schedule_prep);
6012 * __napi_schedule_irqoff - schedule for receive
6013 * @n: entry to schedule
6015 * Variant of __napi_schedule() assuming hard irqs are masked.
6017 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6018 * because the interrupt disabled assumption might not be true
6019 * due to force-threaded interrupts and spinlock substitution.
6021 void __napi_schedule_irqoff(struct napi_struct *n)
6023 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6024 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6028 EXPORT_SYMBOL(__napi_schedule_irqoff);
6030 bool napi_complete_done(struct napi_struct *n, int work_done)
6032 unsigned long flags, val, new, timeout = 0;
6036 * 1) Don't let napi dequeue from the cpu poll list
6037 * just in case its running on a different cpu.
6038 * 2) If we are busy polling, do nothing here, we have
6039 * the guarantee we will be called later.
6041 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6042 NAPIF_STATE_IN_BUSY_POLL)))
6047 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6048 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6050 if (n->defer_hard_irqs_count > 0) {
6051 n->defer_hard_irqs_count--;
6052 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6056 if (n->gro_bitmask) {
6057 /* When the NAPI instance uses a timeout and keeps postponing
6058 * it, we need to bound somehow the time packets are kept in
6061 napi_gro_flush(n, !!timeout);
6066 if (unlikely(!list_empty(&n->poll_list))) {
6067 /* If n->poll_list is not empty, we need to mask irqs */
6068 local_irq_save(flags);
6069 list_del_init(&n->poll_list);
6070 local_irq_restore(flags);
6074 val = READ_ONCE(n->state);
6076 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6078 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6079 NAPIF_STATE_SCHED_THREADED |
6080 NAPIF_STATE_PREFER_BUSY_POLL);
6082 /* If STATE_MISSED was set, leave STATE_SCHED set,
6083 * because we will call napi->poll() one more time.
6084 * This C code was suggested by Alexander Duyck to help gcc.
6086 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6088 } while (cmpxchg(&n->state, val, new) != val);
6090 if (unlikely(val & NAPIF_STATE_MISSED)) {
6096 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6097 HRTIMER_MODE_REL_PINNED);
6100 EXPORT_SYMBOL(napi_complete_done);
6102 /* must be called under rcu_read_lock(), as we dont take a reference */
6103 static struct napi_struct *napi_by_id(unsigned int napi_id)
6105 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6106 struct napi_struct *napi;
6108 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6109 if (napi->napi_id == napi_id)
6115 #if defined(CONFIG_NET_RX_BUSY_POLL)
6117 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6119 if (!skip_schedule) {
6120 gro_normal_list(napi);
6121 __napi_schedule(napi);
6125 if (napi->gro_bitmask) {
6126 /* flush too old packets
6127 * If HZ < 1000, flush all packets.
6129 napi_gro_flush(napi, HZ >= 1000);
6132 gro_normal_list(napi);
6133 clear_bit(NAPI_STATE_SCHED, &napi->state);
6136 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6139 bool skip_schedule = false;
6140 unsigned long timeout;
6143 /* Busy polling means there is a high chance device driver hard irq
6144 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6145 * set in napi_schedule_prep().
6146 * Since we are about to call napi->poll() once more, we can safely
6147 * clear NAPI_STATE_MISSED.
6149 * Note: x86 could use a single "lock and ..." instruction
6150 * to perform these two clear_bit()
6152 clear_bit(NAPI_STATE_MISSED, &napi->state);
6153 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6157 if (prefer_busy_poll) {
6158 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6159 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6160 if (napi->defer_hard_irqs_count && timeout) {
6161 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6162 skip_schedule = true;
6166 /* All we really want here is to re-enable device interrupts.
6167 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6169 rc = napi->poll(napi, budget);
6170 /* We can't gro_normal_list() here, because napi->poll() might have
6171 * rearmed the napi (napi_complete_done()) in which case it could
6172 * already be running on another CPU.
6174 trace_napi_poll(napi, rc, budget);
6175 netpoll_poll_unlock(have_poll_lock);
6177 __busy_poll_stop(napi, skip_schedule);
6181 void napi_busy_loop(unsigned int napi_id,
6182 bool (*loop_end)(void *, unsigned long),
6183 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6185 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6186 int (*napi_poll)(struct napi_struct *napi, int budget);
6187 void *have_poll_lock = NULL;
6188 struct napi_struct *napi;
6195 napi = napi_by_id(napi_id);
6205 unsigned long val = READ_ONCE(napi->state);
6207 /* If multiple threads are competing for this napi,
6208 * we avoid dirtying napi->state as much as we can.
6210 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6211 NAPIF_STATE_IN_BUSY_POLL)) {
6212 if (prefer_busy_poll)
6213 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6216 if (cmpxchg(&napi->state, val,
6217 val | NAPIF_STATE_IN_BUSY_POLL |
6218 NAPIF_STATE_SCHED) != val) {
6219 if (prefer_busy_poll)
6220 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6223 have_poll_lock = netpoll_poll_lock(napi);
6224 napi_poll = napi->poll;
6226 work = napi_poll(napi, budget);
6227 trace_napi_poll(napi, work, budget);
6228 gro_normal_list(napi);
6231 __NET_ADD_STATS(dev_net(napi->dev),
6232 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6235 if (!loop_end || loop_end(loop_end_arg, start_time))
6238 if (unlikely(need_resched())) {
6240 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6244 if (loop_end(loop_end_arg, start_time))
6251 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6256 EXPORT_SYMBOL(napi_busy_loop);
6258 #endif /* CONFIG_NET_RX_BUSY_POLL */
6260 static void napi_hash_add(struct napi_struct *napi)
6262 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6265 spin_lock(&napi_hash_lock);
6267 /* 0..NR_CPUS range is reserved for sender_cpu use */
6269 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6270 napi_gen_id = MIN_NAPI_ID;
6271 } while (napi_by_id(napi_gen_id));
6272 napi->napi_id = napi_gen_id;
6274 hlist_add_head_rcu(&napi->napi_hash_node,
6275 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6277 spin_unlock(&napi_hash_lock);
6280 /* Warning : caller is responsible to make sure rcu grace period
6281 * is respected before freeing memory containing @napi
6283 static void napi_hash_del(struct napi_struct *napi)
6285 spin_lock(&napi_hash_lock);
6287 hlist_del_init_rcu(&napi->napi_hash_node);
6289 spin_unlock(&napi_hash_lock);
6292 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6294 struct napi_struct *napi;
6296 napi = container_of(timer, struct napi_struct, timer);
6298 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6299 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6301 if (!napi_disable_pending(napi) &&
6302 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6303 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6304 __napi_schedule_irqoff(napi);
6307 return HRTIMER_NORESTART;
6310 static void init_gro_hash(struct napi_struct *napi)
6314 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6315 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6316 napi->gro_hash[i].count = 0;
6318 napi->gro_bitmask = 0;
6321 int dev_set_threaded(struct net_device *dev, bool threaded)
6323 struct napi_struct *napi;
6326 if (dev->threaded == threaded)
6330 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6331 if (!napi->thread) {
6332 err = napi_kthread_create(napi);
6341 dev->threaded = threaded;
6343 /* Make sure kthread is created before THREADED bit
6346 smp_mb__before_atomic();
6348 /* Setting/unsetting threaded mode on a napi might not immediately
6349 * take effect, if the current napi instance is actively being
6350 * polled. In this case, the switch between threaded mode and
6351 * softirq mode will happen in the next round of napi_schedule().
6352 * This should not cause hiccups/stalls to the live traffic.
6354 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6356 set_bit(NAPI_STATE_THREADED, &napi->state);
6358 clear_bit(NAPI_STATE_THREADED, &napi->state);
6363 EXPORT_SYMBOL(dev_set_threaded);
6365 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6366 int (*poll)(struct napi_struct *, int), int weight)
6368 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6371 INIT_LIST_HEAD(&napi->poll_list);
6372 INIT_HLIST_NODE(&napi->napi_hash_node);
6373 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6374 napi->timer.function = napi_watchdog;
6375 init_gro_hash(napi);
6377 INIT_LIST_HEAD(&napi->rx_list);
6380 if (weight > NAPI_POLL_WEIGHT)
6381 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6383 napi->weight = weight;
6385 #ifdef CONFIG_NETPOLL
6386 napi->poll_owner = -1;
6388 set_bit(NAPI_STATE_SCHED, &napi->state);
6389 set_bit(NAPI_STATE_NPSVC, &napi->state);
6390 list_add_rcu(&napi->dev_list, &dev->napi_list);
6391 napi_hash_add(napi);
6392 napi_get_frags_check(napi);
6393 /* Create kthread for this napi if dev->threaded is set.
6394 * Clear dev->threaded if kthread creation failed so that
6395 * threaded mode will not be enabled in napi_enable().
6397 if (dev->threaded && napi_kthread_create(napi))
6400 EXPORT_SYMBOL(netif_napi_add_weight);
6402 void napi_disable(struct napi_struct *n)
6404 unsigned long val, new;
6407 set_bit(NAPI_STATE_DISABLE, &n->state);
6410 val = READ_ONCE(n->state);
6411 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6412 usleep_range(20, 200);
6416 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6417 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6419 if (cmpxchg(&n->state, val, new) == val)
6423 hrtimer_cancel(&n->timer);
6425 clear_bit(NAPI_STATE_DISABLE, &n->state);
6427 EXPORT_SYMBOL(napi_disable);
6430 * napi_enable - enable NAPI scheduling
6433 * Resume NAPI from being scheduled on this context.
6434 * Must be paired with napi_disable.
6436 void napi_enable(struct napi_struct *n)
6438 unsigned long val, new;
6441 val = READ_ONCE(n->state);
6442 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6444 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6445 if (n->dev->threaded && n->thread)
6446 new |= NAPIF_STATE_THREADED;
6447 } while (cmpxchg(&n->state, val, new) != val);
6449 EXPORT_SYMBOL(napi_enable);
6451 static void flush_gro_hash(struct napi_struct *napi)
6455 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6456 struct sk_buff *skb, *n;
6458 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6460 napi->gro_hash[i].count = 0;
6464 /* Must be called in process context */
6465 void __netif_napi_del(struct napi_struct *napi)
6467 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6470 napi_hash_del(napi);
6471 list_del_rcu(&napi->dev_list);
6472 napi_free_frags(napi);
6474 flush_gro_hash(napi);
6475 napi->gro_bitmask = 0;
6478 kthread_stop(napi->thread);
6479 napi->thread = NULL;
6482 EXPORT_SYMBOL(__netif_napi_del);
6484 static int __napi_poll(struct napi_struct *n, bool *repoll)
6490 /* This NAPI_STATE_SCHED test is for avoiding a race
6491 * with netpoll's poll_napi(). Only the entity which
6492 * obtains the lock and sees NAPI_STATE_SCHED set will
6493 * actually make the ->poll() call. Therefore we avoid
6494 * accidentally calling ->poll() when NAPI is not scheduled.
6497 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6498 work = n->poll(n, weight);
6499 trace_napi_poll(n, work, weight);
6502 if (unlikely(work > weight))
6503 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6504 n->poll, work, weight);
6506 if (likely(work < weight))
6509 /* Drivers must not modify the NAPI state if they
6510 * consume the entire weight. In such cases this code
6511 * still "owns" the NAPI instance and therefore can
6512 * move the instance around on the list at-will.
6514 if (unlikely(napi_disable_pending(n))) {
6519 /* The NAPI context has more processing work, but busy-polling
6520 * is preferred. Exit early.
6522 if (napi_prefer_busy_poll(n)) {
6523 if (napi_complete_done(n, work)) {
6524 /* If timeout is not set, we need to make sure
6525 * that the NAPI is re-scheduled.
6532 if (n->gro_bitmask) {
6533 /* flush too old packets
6534 * If HZ < 1000, flush all packets.
6536 napi_gro_flush(n, HZ >= 1000);
6541 /* Some drivers may have called napi_schedule
6542 * prior to exhausting their budget.
6544 if (unlikely(!list_empty(&n->poll_list))) {
6545 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6546 n->dev ? n->dev->name : "backlog");
6555 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6557 bool do_repoll = false;
6561 list_del_init(&n->poll_list);
6563 have = netpoll_poll_lock(n);
6565 work = __napi_poll(n, &do_repoll);
6568 list_add_tail(&n->poll_list, repoll);
6570 netpoll_poll_unlock(have);
6575 static int napi_thread_wait(struct napi_struct *napi)
6579 set_current_state(TASK_INTERRUPTIBLE);
6581 while (!kthread_should_stop()) {
6582 /* Testing SCHED_THREADED bit here to make sure the current
6583 * kthread owns this napi and could poll on this napi.
6584 * Testing SCHED bit is not enough because SCHED bit might be
6585 * set by some other busy poll thread or by napi_disable().
6587 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6588 WARN_ON(!list_empty(&napi->poll_list));
6589 __set_current_state(TASK_RUNNING);
6594 /* woken being true indicates this thread owns this napi. */
6596 set_current_state(TASK_INTERRUPTIBLE);
6598 __set_current_state(TASK_RUNNING);
6603 static int napi_threaded_poll(void *data)
6605 struct napi_struct *napi = data;
6608 while (!napi_thread_wait(napi)) {
6610 bool repoll = false;
6614 have = netpoll_poll_lock(napi);
6615 __napi_poll(napi, &repoll);
6616 netpoll_poll_unlock(have);
6629 static void skb_defer_free_flush(struct softnet_data *sd)
6631 struct sk_buff *skb, *next;
6632 unsigned long flags;
6634 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6635 if (!READ_ONCE(sd->defer_list))
6638 spin_lock_irqsave(&sd->defer_lock, flags);
6639 skb = sd->defer_list;
6640 sd->defer_list = NULL;
6641 sd->defer_count = 0;
6642 spin_unlock_irqrestore(&sd->defer_lock, flags);
6644 while (skb != NULL) {
6646 napi_consume_skb(skb, 1);
6651 static __latent_entropy void net_rx_action(struct softirq_action *h)
6653 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6654 unsigned long time_limit = jiffies +
6655 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6656 int budget = READ_ONCE(netdev_budget);
6660 local_irq_disable();
6661 list_splice_init(&sd->poll_list, &list);
6665 struct napi_struct *n;
6667 skb_defer_free_flush(sd);
6669 if (list_empty(&list)) {
6670 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6675 n = list_first_entry(&list, struct napi_struct, poll_list);
6676 budget -= napi_poll(n, &repoll);
6678 /* If softirq window is exhausted then punt.
6679 * Allow this to run for 2 jiffies since which will allow
6680 * an average latency of 1.5/HZ.
6682 if (unlikely(budget <= 0 ||
6683 time_after_eq(jiffies, time_limit))) {
6689 local_irq_disable();
6691 list_splice_tail_init(&sd->poll_list, &list);
6692 list_splice_tail(&repoll, &list);
6693 list_splice(&list, &sd->poll_list);
6694 if (!list_empty(&sd->poll_list))
6695 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6697 net_rps_action_and_irq_enable(sd);
6701 struct netdev_adjacent {
6702 struct net_device *dev;
6703 netdevice_tracker dev_tracker;
6705 /* upper master flag, there can only be one master device per list */
6708 /* lookup ignore flag */
6711 /* counter for the number of times this device was added to us */
6714 /* private field for the users */
6717 struct list_head list;
6718 struct rcu_head rcu;
6721 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6722 struct list_head *adj_list)
6724 struct netdev_adjacent *adj;
6726 list_for_each_entry(adj, adj_list, list) {
6727 if (adj->dev == adj_dev)
6733 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6734 struct netdev_nested_priv *priv)
6736 struct net_device *dev = (struct net_device *)priv->data;
6738 return upper_dev == dev;
6742 * netdev_has_upper_dev - Check if device is linked to an upper device
6744 * @upper_dev: upper device to check
6746 * Find out if a device is linked to specified upper device and return true
6747 * in case it is. Note that this checks only immediate upper device,
6748 * not through a complete stack of devices. The caller must hold the RTNL lock.
6750 bool netdev_has_upper_dev(struct net_device *dev,
6751 struct net_device *upper_dev)
6753 struct netdev_nested_priv priv = {
6754 .data = (void *)upper_dev,
6759 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6762 EXPORT_SYMBOL(netdev_has_upper_dev);
6765 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6767 * @upper_dev: upper device to check
6769 * Find out if a device is linked to specified upper device and return true
6770 * in case it is. Note that this checks the entire upper device chain.
6771 * The caller must hold rcu lock.
6774 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6775 struct net_device *upper_dev)
6777 struct netdev_nested_priv priv = {
6778 .data = (void *)upper_dev,
6781 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6784 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6787 * netdev_has_any_upper_dev - Check if device is linked to some device
6790 * Find out if a device is linked to an upper device and return true in case
6791 * it is. The caller must hold the RTNL lock.
6793 bool netdev_has_any_upper_dev(struct net_device *dev)
6797 return !list_empty(&dev->adj_list.upper);
6799 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6802 * netdev_master_upper_dev_get - Get master upper device
6805 * Find a master upper device and return pointer to it or NULL in case
6806 * it's not there. The caller must hold the RTNL lock.
6808 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6810 struct netdev_adjacent *upper;
6814 if (list_empty(&dev->adj_list.upper))
6817 upper = list_first_entry(&dev->adj_list.upper,
6818 struct netdev_adjacent, list);
6819 if (likely(upper->master))
6823 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6825 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6827 struct netdev_adjacent *upper;
6831 if (list_empty(&dev->adj_list.upper))
6834 upper = list_first_entry(&dev->adj_list.upper,
6835 struct netdev_adjacent, list);
6836 if (likely(upper->master) && !upper->ignore)
6842 * netdev_has_any_lower_dev - Check if device is linked to some device
6845 * Find out if a device is linked to a lower device and return true in case
6846 * it is. The caller must hold the RTNL lock.
6848 static bool netdev_has_any_lower_dev(struct net_device *dev)
6852 return !list_empty(&dev->adj_list.lower);
6855 void *netdev_adjacent_get_private(struct list_head *adj_list)
6857 struct netdev_adjacent *adj;
6859 adj = list_entry(adj_list, struct netdev_adjacent, list);
6861 return adj->private;
6863 EXPORT_SYMBOL(netdev_adjacent_get_private);
6866 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6868 * @iter: list_head ** of the current position
6870 * Gets the next device from the dev's upper list, starting from iter
6871 * position. The caller must hold RCU read lock.
6873 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6874 struct list_head **iter)
6876 struct netdev_adjacent *upper;
6878 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6880 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6882 if (&upper->list == &dev->adj_list.upper)
6885 *iter = &upper->list;
6889 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6891 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6892 struct list_head **iter,
6895 struct netdev_adjacent *upper;
6897 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6899 if (&upper->list == &dev->adj_list.upper)
6902 *iter = &upper->list;
6903 *ignore = upper->ignore;
6908 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6909 struct list_head **iter)
6911 struct netdev_adjacent *upper;
6913 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6915 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6917 if (&upper->list == &dev->adj_list.upper)
6920 *iter = &upper->list;
6925 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6926 int (*fn)(struct net_device *dev,
6927 struct netdev_nested_priv *priv),
6928 struct netdev_nested_priv *priv)
6930 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6931 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6936 iter = &dev->adj_list.upper;
6940 ret = fn(now, priv);
6947 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6954 niter = &udev->adj_list.upper;
6955 dev_stack[cur] = now;
6956 iter_stack[cur++] = iter;
6963 next = dev_stack[--cur];
6964 niter = iter_stack[cur];
6974 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6975 int (*fn)(struct net_device *dev,
6976 struct netdev_nested_priv *priv),
6977 struct netdev_nested_priv *priv)
6979 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6980 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6984 iter = &dev->adj_list.upper;
6988 ret = fn(now, priv);
6995 udev = netdev_next_upper_dev_rcu(now, &iter);
7000 niter = &udev->adj_list.upper;
7001 dev_stack[cur] = now;
7002 iter_stack[cur++] = iter;
7009 next = dev_stack[--cur];
7010 niter = iter_stack[cur];
7019 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7021 static bool __netdev_has_upper_dev(struct net_device *dev,
7022 struct net_device *upper_dev)
7024 struct netdev_nested_priv priv = {
7026 .data = (void *)upper_dev,
7031 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7036 * netdev_lower_get_next_private - Get the next ->private from the
7037 * lower neighbour list
7039 * @iter: list_head ** of the current position
7041 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7042 * list, starting from iter position. The caller must hold either hold the
7043 * RTNL lock or its own locking that guarantees that the neighbour lower
7044 * list will remain unchanged.
7046 void *netdev_lower_get_next_private(struct net_device *dev,
7047 struct list_head **iter)
7049 struct netdev_adjacent *lower;
7051 lower = list_entry(*iter, struct netdev_adjacent, list);
7053 if (&lower->list == &dev->adj_list.lower)
7056 *iter = lower->list.next;
7058 return lower->private;
7060 EXPORT_SYMBOL(netdev_lower_get_next_private);
7063 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7064 * lower neighbour list, RCU
7067 * @iter: list_head ** of the current position
7069 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7070 * list, starting from iter position. The caller must hold RCU read lock.
7072 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7073 struct list_head **iter)
7075 struct netdev_adjacent *lower;
7077 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7079 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7081 if (&lower->list == &dev->adj_list.lower)
7084 *iter = &lower->list;
7086 return lower->private;
7088 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7091 * netdev_lower_get_next - Get the next device from the lower neighbour
7094 * @iter: list_head ** of the current position
7096 * Gets the next netdev_adjacent from the dev's lower neighbour
7097 * list, starting from iter position. The caller must hold RTNL lock or
7098 * its own locking that guarantees that the neighbour lower
7099 * list will remain unchanged.
7101 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7103 struct netdev_adjacent *lower;
7105 lower = list_entry(*iter, struct netdev_adjacent, list);
7107 if (&lower->list == &dev->adj_list.lower)
7110 *iter = lower->list.next;
7114 EXPORT_SYMBOL(netdev_lower_get_next);
7116 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7117 struct list_head **iter)
7119 struct netdev_adjacent *lower;
7121 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7123 if (&lower->list == &dev->adj_list.lower)
7126 *iter = &lower->list;
7131 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7132 struct list_head **iter,
7135 struct netdev_adjacent *lower;
7137 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7139 if (&lower->list == &dev->adj_list.lower)
7142 *iter = &lower->list;
7143 *ignore = lower->ignore;
7148 int netdev_walk_all_lower_dev(struct net_device *dev,
7149 int (*fn)(struct net_device *dev,
7150 struct netdev_nested_priv *priv),
7151 struct netdev_nested_priv *priv)
7153 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7154 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7158 iter = &dev->adj_list.lower;
7162 ret = fn(now, priv);
7169 ldev = netdev_next_lower_dev(now, &iter);
7174 niter = &ldev->adj_list.lower;
7175 dev_stack[cur] = now;
7176 iter_stack[cur++] = iter;
7183 next = dev_stack[--cur];
7184 niter = iter_stack[cur];
7193 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7195 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7196 int (*fn)(struct net_device *dev,
7197 struct netdev_nested_priv *priv),
7198 struct netdev_nested_priv *priv)
7200 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7201 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7206 iter = &dev->adj_list.lower;
7210 ret = fn(now, priv);
7217 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7224 niter = &ldev->adj_list.lower;
7225 dev_stack[cur] = now;
7226 iter_stack[cur++] = iter;
7233 next = dev_stack[--cur];
7234 niter = iter_stack[cur];
7244 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7245 struct list_head **iter)
7247 struct netdev_adjacent *lower;
7249 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7250 if (&lower->list == &dev->adj_list.lower)
7253 *iter = &lower->list;
7257 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7259 static u8 __netdev_upper_depth(struct net_device *dev)
7261 struct net_device *udev;
7262 struct list_head *iter;
7266 for (iter = &dev->adj_list.upper,
7267 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7269 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7272 if (max_depth < udev->upper_level)
7273 max_depth = udev->upper_level;
7279 static u8 __netdev_lower_depth(struct net_device *dev)
7281 struct net_device *ldev;
7282 struct list_head *iter;
7286 for (iter = &dev->adj_list.lower,
7287 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7289 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7292 if (max_depth < ldev->lower_level)
7293 max_depth = ldev->lower_level;
7299 static int __netdev_update_upper_level(struct net_device *dev,
7300 struct netdev_nested_priv *__unused)
7302 dev->upper_level = __netdev_upper_depth(dev) + 1;
7306 #ifdef CONFIG_LOCKDEP
7307 static LIST_HEAD(net_unlink_list);
7309 static void net_unlink_todo(struct net_device *dev)
7311 if (list_empty(&dev->unlink_list))
7312 list_add_tail(&dev->unlink_list, &net_unlink_list);
7316 static int __netdev_update_lower_level(struct net_device *dev,
7317 struct netdev_nested_priv *priv)
7319 dev->lower_level = __netdev_lower_depth(dev) + 1;
7321 #ifdef CONFIG_LOCKDEP
7325 if (priv->flags & NESTED_SYNC_IMM)
7326 dev->nested_level = dev->lower_level - 1;
7327 if (priv->flags & NESTED_SYNC_TODO)
7328 net_unlink_todo(dev);
7333 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7334 int (*fn)(struct net_device *dev,
7335 struct netdev_nested_priv *priv),
7336 struct netdev_nested_priv *priv)
7338 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7339 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7343 iter = &dev->adj_list.lower;
7347 ret = fn(now, priv);
7354 ldev = netdev_next_lower_dev_rcu(now, &iter);
7359 niter = &ldev->adj_list.lower;
7360 dev_stack[cur] = now;
7361 iter_stack[cur++] = iter;
7368 next = dev_stack[--cur];
7369 niter = iter_stack[cur];
7378 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7381 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7382 * lower neighbour list, RCU
7386 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7387 * list. The caller must hold RCU read lock.
7389 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7391 struct netdev_adjacent *lower;
7393 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7394 struct netdev_adjacent, list);
7396 return lower->private;
7399 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7402 * netdev_master_upper_dev_get_rcu - Get master upper device
7405 * Find a master upper device and return pointer to it or NULL in case
7406 * it's not there. The caller must hold the RCU read lock.
7408 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7410 struct netdev_adjacent *upper;
7412 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7413 struct netdev_adjacent, list);
7414 if (upper && likely(upper->master))
7418 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7420 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7421 struct net_device *adj_dev,
7422 struct list_head *dev_list)
7424 char linkname[IFNAMSIZ+7];
7426 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7427 "upper_%s" : "lower_%s", adj_dev->name);
7428 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7431 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7433 struct list_head *dev_list)
7435 char linkname[IFNAMSIZ+7];
7437 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7438 "upper_%s" : "lower_%s", name);
7439 sysfs_remove_link(&(dev->dev.kobj), linkname);
7442 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7443 struct net_device *adj_dev,
7444 struct list_head *dev_list)
7446 return (dev_list == &dev->adj_list.upper ||
7447 dev_list == &dev->adj_list.lower) &&
7448 net_eq(dev_net(dev), dev_net(adj_dev));
7451 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7452 struct net_device *adj_dev,
7453 struct list_head *dev_list,
7454 void *private, bool master)
7456 struct netdev_adjacent *adj;
7459 adj = __netdev_find_adj(adj_dev, dev_list);
7463 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7464 dev->name, adj_dev->name, adj->ref_nr);
7469 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7474 adj->master = master;
7476 adj->private = private;
7477 adj->ignore = false;
7478 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7480 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7481 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7483 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7484 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7489 /* Ensure that master link is always the first item in list. */
7491 ret = sysfs_create_link(&(dev->dev.kobj),
7492 &(adj_dev->dev.kobj), "master");
7494 goto remove_symlinks;
7496 list_add_rcu(&adj->list, dev_list);
7498 list_add_tail_rcu(&adj->list, dev_list);
7504 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7505 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7507 netdev_put(adj_dev, &adj->dev_tracker);
7513 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7514 struct net_device *adj_dev,
7516 struct list_head *dev_list)
7518 struct netdev_adjacent *adj;
7520 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7521 dev->name, adj_dev->name, ref_nr);
7523 adj = __netdev_find_adj(adj_dev, dev_list);
7526 pr_err("Adjacency does not exist for device %s from %s\n",
7527 dev->name, adj_dev->name);
7532 if (adj->ref_nr > ref_nr) {
7533 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7534 dev->name, adj_dev->name, ref_nr,
7535 adj->ref_nr - ref_nr);
7536 adj->ref_nr -= ref_nr;
7541 sysfs_remove_link(&(dev->dev.kobj), "master");
7543 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7544 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7546 list_del_rcu(&adj->list);
7547 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7548 adj_dev->name, dev->name, adj_dev->name);
7549 netdev_put(adj_dev, &adj->dev_tracker);
7550 kfree_rcu(adj, rcu);
7553 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7554 struct net_device *upper_dev,
7555 struct list_head *up_list,
7556 struct list_head *down_list,
7557 void *private, bool master)
7561 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7566 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7569 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7576 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7577 struct net_device *upper_dev,
7579 struct list_head *up_list,
7580 struct list_head *down_list)
7582 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7583 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7586 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7587 struct net_device *upper_dev,
7588 void *private, bool master)
7590 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7591 &dev->adj_list.upper,
7592 &upper_dev->adj_list.lower,
7596 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7597 struct net_device *upper_dev)
7599 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7600 &dev->adj_list.upper,
7601 &upper_dev->adj_list.lower);
7604 static int __netdev_upper_dev_link(struct net_device *dev,
7605 struct net_device *upper_dev, bool master,
7606 void *upper_priv, void *upper_info,
7607 struct netdev_nested_priv *priv,
7608 struct netlink_ext_ack *extack)
7610 struct netdev_notifier_changeupper_info changeupper_info = {
7615 .upper_dev = upper_dev,
7618 .upper_info = upper_info,
7620 struct net_device *master_dev;
7625 if (dev == upper_dev)
7628 /* To prevent loops, check if dev is not upper device to upper_dev. */
7629 if (__netdev_has_upper_dev(upper_dev, dev))
7632 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7636 if (__netdev_has_upper_dev(dev, upper_dev))
7639 master_dev = __netdev_master_upper_dev_get(dev);
7641 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7644 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7645 &changeupper_info.info);
7646 ret = notifier_to_errno(ret);
7650 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7655 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7656 &changeupper_info.info);
7657 ret = notifier_to_errno(ret);
7661 __netdev_update_upper_level(dev, NULL);
7662 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7664 __netdev_update_lower_level(upper_dev, priv);
7665 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7671 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7677 * netdev_upper_dev_link - Add a link to the upper device
7679 * @upper_dev: new upper device
7680 * @extack: netlink extended ack
7682 * Adds a link to device which is upper to this one. The caller must hold
7683 * the RTNL lock. On a failure a negative errno code is returned.
7684 * On success the reference counts are adjusted and the function
7687 int netdev_upper_dev_link(struct net_device *dev,
7688 struct net_device *upper_dev,
7689 struct netlink_ext_ack *extack)
7691 struct netdev_nested_priv priv = {
7692 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7696 return __netdev_upper_dev_link(dev, upper_dev, false,
7697 NULL, NULL, &priv, extack);
7699 EXPORT_SYMBOL(netdev_upper_dev_link);
7702 * netdev_master_upper_dev_link - Add a master link to the upper device
7704 * @upper_dev: new upper device
7705 * @upper_priv: upper device private
7706 * @upper_info: upper info to be passed down via notifier
7707 * @extack: netlink extended ack
7709 * Adds a link to device which is upper to this one. In this case, only
7710 * one master upper device can be linked, although other non-master devices
7711 * might be linked as well. The caller must hold the RTNL lock.
7712 * On a failure a negative errno code is returned. On success the reference
7713 * counts are adjusted and the function returns zero.
7715 int netdev_master_upper_dev_link(struct net_device *dev,
7716 struct net_device *upper_dev,
7717 void *upper_priv, void *upper_info,
7718 struct netlink_ext_ack *extack)
7720 struct netdev_nested_priv priv = {
7721 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7725 return __netdev_upper_dev_link(dev, upper_dev, true,
7726 upper_priv, upper_info, &priv, extack);
7728 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7730 static void __netdev_upper_dev_unlink(struct net_device *dev,
7731 struct net_device *upper_dev,
7732 struct netdev_nested_priv *priv)
7734 struct netdev_notifier_changeupper_info changeupper_info = {
7738 .upper_dev = upper_dev,
7744 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7746 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7747 &changeupper_info.info);
7749 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7751 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7752 &changeupper_info.info);
7754 __netdev_update_upper_level(dev, NULL);
7755 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7757 __netdev_update_lower_level(upper_dev, priv);
7758 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7763 * netdev_upper_dev_unlink - Removes a link to upper device
7765 * @upper_dev: new upper device
7767 * Removes a link to device which is upper to this one. The caller must hold
7770 void netdev_upper_dev_unlink(struct net_device *dev,
7771 struct net_device *upper_dev)
7773 struct netdev_nested_priv priv = {
7774 .flags = NESTED_SYNC_TODO,
7778 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7780 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7782 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7783 struct net_device *lower_dev,
7786 struct netdev_adjacent *adj;
7788 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7792 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7797 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7798 struct net_device *lower_dev)
7800 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7803 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7804 struct net_device *lower_dev)
7806 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7809 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7810 struct net_device *new_dev,
7811 struct net_device *dev,
7812 struct netlink_ext_ack *extack)
7814 struct netdev_nested_priv priv = {
7823 if (old_dev && new_dev != old_dev)
7824 netdev_adjacent_dev_disable(dev, old_dev);
7825 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7828 if (old_dev && new_dev != old_dev)
7829 netdev_adjacent_dev_enable(dev, old_dev);
7835 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7837 void netdev_adjacent_change_commit(struct net_device *old_dev,
7838 struct net_device *new_dev,
7839 struct net_device *dev)
7841 struct netdev_nested_priv priv = {
7842 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7846 if (!new_dev || !old_dev)
7849 if (new_dev == old_dev)
7852 netdev_adjacent_dev_enable(dev, old_dev);
7853 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7855 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7857 void netdev_adjacent_change_abort(struct net_device *old_dev,
7858 struct net_device *new_dev,
7859 struct net_device *dev)
7861 struct netdev_nested_priv priv = {
7869 if (old_dev && new_dev != old_dev)
7870 netdev_adjacent_dev_enable(dev, old_dev);
7872 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7874 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7877 * netdev_bonding_info_change - Dispatch event about slave change
7879 * @bonding_info: info to dispatch
7881 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7882 * The caller must hold the RTNL lock.
7884 void netdev_bonding_info_change(struct net_device *dev,
7885 struct netdev_bonding_info *bonding_info)
7887 struct netdev_notifier_bonding_info info = {
7891 memcpy(&info.bonding_info, bonding_info,
7892 sizeof(struct netdev_bonding_info));
7893 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7896 EXPORT_SYMBOL(netdev_bonding_info_change);
7898 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7899 struct netlink_ext_ack *extack)
7901 struct netdev_notifier_offload_xstats_info info = {
7903 .info.extack = extack,
7904 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7909 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7911 if (!dev->offload_xstats_l3)
7914 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7915 NETDEV_OFFLOAD_XSTATS_DISABLE,
7917 err = notifier_to_errno(rc);
7924 kfree(dev->offload_xstats_l3);
7925 dev->offload_xstats_l3 = NULL;
7929 int netdev_offload_xstats_enable(struct net_device *dev,
7930 enum netdev_offload_xstats_type type,
7931 struct netlink_ext_ack *extack)
7935 if (netdev_offload_xstats_enabled(dev, type))
7939 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7940 return netdev_offload_xstats_enable_l3(dev, extack);
7946 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7948 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7950 struct netdev_notifier_offload_xstats_info info = {
7952 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7955 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7957 kfree(dev->offload_xstats_l3);
7958 dev->offload_xstats_l3 = NULL;
7961 int netdev_offload_xstats_disable(struct net_device *dev,
7962 enum netdev_offload_xstats_type type)
7966 if (!netdev_offload_xstats_enabled(dev, type))
7970 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7971 netdev_offload_xstats_disable_l3(dev);
7978 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7980 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7982 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7985 static struct rtnl_hw_stats64 *
7986 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7987 enum netdev_offload_xstats_type type)
7990 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7991 return dev->offload_xstats_l3;
7998 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7999 enum netdev_offload_xstats_type type)
8003 return netdev_offload_xstats_get_ptr(dev, type);
8005 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8007 struct netdev_notifier_offload_xstats_ru {
8011 struct netdev_notifier_offload_xstats_rd {
8012 struct rtnl_hw_stats64 stats;
8016 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8017 const struct rtnl_hw_stats64 *src)
8019 dest->rx_packets += src->rx_packets;
8020 dest->tx_packets += src->tx_packets;
8021 dest->rx_bytes += src->rx_bytes;
8022 dest->tx_bytes += src->tx_bytes;
8023 dest->rx_errors += src->rx_errors;
8024 dest->tx_errors += src->tx_errors;
8025 dest->rx_dropped += src->rx_dropped;
8026 dest->tx_dropped += src->tx_dropped;
8027 dest->multicast += src->multicast;
8030 static int netdev_offload_xstats_get_used(struct net_device *dev,
8031 enum netdev_offload_xstats_type type,
8033 struct netlink_ext_ack *extack)
8035 struct netdev_notifier_offload_xstats_ru report_used = {};
8036 struct netdev_notifier_offload_xstats_info info = {
8038 .info.extack = extack,
8040 .report_used = &report_used,
8044 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8045 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8047 *p_used = report_used.used;
8048 return notifier_to_errno(rc);
8051 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8052 enum netdev_offload_xstats_type type,
8053 struct rtnl_hw_stats64 *p_stats,
8055 struct netlink_ext_ack *extack)
8057 struct netdev_notifier_offload_xstats_rd report_delta = {};
8058 struct netdev_notifier_offload_xstats_info info = {
8060 .info.extack = extack,
8062 .report_delta = &report_delta,
8064 struct rtnl_hw_stats64 *stats;
8067 stats = netdev_offload_xstats_get_ptr(dev, type);
8068 if (WARN_ON(!stats))
8071 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8074 /* Cache whatever we got, even if there was an error, otherwise the
8075 * successful stats retrievals would get lost.
8077 netdev_hw_stats64_add(stats, &report_delta.stats);
8081 *p_used = report_delta.used;
8083 return notifier_to_errno(rc);
8086 int netdev_offload_xstats_get(struct net_device *dev,
8087 enum netdev_offload_xstats_type type,
8088 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8089 struct netlink_ext_ack *extack)
8094 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8097 return netdev_offload_xstats_get_used(dev, type, p_used,
8100 EXPORT_SYMBOL(netdev_offload_xstats_get);
8103 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8104 const struct rtnl_hw_stats64 *stats)
8106 report_delta->used = true;
8107 netdev_hw_stats64_add(&report_delta->stats, stats);
8109 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8112 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8114 report_used->used = true;
8116 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8118 void netdev_offload_xstats_push_delta(struct net_device *dev,
8119 enum netdev_offload_xstats_type type,
8120 const struct rtnl_hw_stats64 *p_stats)
8122 struct rtnl_hw_stats64 *stats;
8126 stats = netdev_offload_xstats_get_ptr(dev, type);
8127 if (WARN_ON(!stats))
8130 netdev_hw_stats64_add(stats, p_stats);
8132 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8135 * netdev_get_xmit_slave - Get the xmit slave of master device
8138 * @all_slaves: assume all the slaves are active
8140 * The reference counters are not incremented so the caller must be
8141 * careful with locks. The caller must hold RCU lock.
8142 * %NULL is returned if no slave is found.
8145 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8146 struct sk_buff *skb,
8149 const struct net_device_ops *ops = dev->netdev_ops;
8151 if (!ops->ndo_get_xmit_slave)
8153 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8155 EXPORT_SYMBOL(netdev_get_xmit_slave);
8157 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8160 const struct net_device_ops *ops = dev->netdev_ops;
8162 if (!ops->ndo_sk_get_lower_dev)
8164 return ops->ndo_sk_get_lower_dev(dev, sk);
8168 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8172 * %NULL is returned if no lower device is found.
8175 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8178 struct net_device *lower;
8180 lower = netdev_sk_get_lower_dev(dev, sk);
8183 lower = netdev_sk_get_lower_dev(dev, sk);
8188 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8190 static void netdev_adjacent_add_links(struct net_device *dev)
8192 struct netdev_adjacent *iter;
8194 struct net *net = dev_net(dev);
8196 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8197 if (!net_eq(net, dev_net(iter->dev)))
8199 netdev_adjacent_sysfs_add(iter->dev, dev,
8200 &iter->dev->adj_list.lower);
8201 netdev_adjacent_sysfs_add(dev, iter->dev,
8202 &dev->adj_list.upper);
8205 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8206 if (!net_eq(net, dev_net(iter->dev)))
8208 netdev_adjacent_sysfs_add(iter->dev, dev,
8209 &iter->dev->adj_list.upper);
8210 netdev_adjacent_sysfs_add(dev, iter->dev,
8211 &dev->adj_list.lower);
8215 static void netdev_adjacent_del_links(struct net_device *dev)
8217 struct netdev_adjacent *iter;
8219 struct net *net = dev_net(dev);
8221 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8222 if (!net_eq(net, dev_net(iter->dev)))
8224 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8225 &iter->dev->adj_list.lower);
8226 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8227 &dev->adj_list.upper);
8230 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8231 if (!net_eq(net, dev_net(iter->dev)))
8233 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8234 &iter->dev->adj_list.upper);
8235 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8236 &dev->adj_list.lower);
8240 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8242 struct netdev_adjacent *iter;
8244 struct net *net = dev_net(dev);
8246 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8247 if (!net_eq(net, dev_net(iter->dev)))
8249 netdev_adjacent_sysfs_del(iter->dev, oldname,
8250 &iter->dev->adj_list.lower);
8251 netdev_adjacent_sysfs_add(iter->dev, dev,
8252 &iter->dev->adj_list.lower);
8255 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8256 if (!net_eq(net, dev_net(iter->dev)))
8258 netdev_adjacent_sysfs_del(iter->dev, oldname,
8259 &iter->dev->adj_list.upper);
8260 netdev_adjacent_sysfs_add(iter->dev, dev,
8261 &iter->dev->adj_list.upper);
8265 void *netdev_lower_dev_get_private(struct net_device *dev,
8266 struct net_device *lower_dev)
8268 struct netdev_adjacent *lower;
8272 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8276 return lower->private;
8278 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8282 * netdev_lower_state_changed - Dispatch event about lower device state change
8283 * @lower_dev: device
8284 * @lower_state_info: state to dispatch
8286 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8287 * The caller must hold the RTNL lock.
8289 void netdev_lower_state_changed(struct net_device *lower_dev,
8290 void *lower_state_info)
8292 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8293 .info.dev = lower_dev,
8297 changelowerstate_info.lower_state_info = lower_state_info;
8298 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8299 &changelowerstate_info.info);
8301 EXPORT_SYMBOL(netdev_lower_state_changed);
8303 static void dev_change_rx_flags(struct net_device *dev, int flags)
8305 const struct net_device_ops *ops = dev->netdev_ops;
8307 if (ops->ndo_change_rx_flags)
8308 ops->ndo_change_rx_flags(dev, flags);
8311 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8313 unsigned int old_flags = dev->flags;
8319 dev->flags |= IFF_PROMISC;
8320 dev->promiscuity += inc;
8321 if (dev->promiscuity == 0) {
8324 * If inc causes overflow, untouch promisc and return error.
8327 dev->flags &= ~IFF_PROMISC;
8329 dev->promiscuity -= inc;
8330 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8334 if (dev->flags != old_flags) {
8335 pr_info("device %s %s promiscuous mode\n",
8337 dev->flags & IFF_PROMISC ? "entered" : "left");
8338 if (audit_enabled) {
8339 current_uid_gid(&uid, &gid);
8340 audit_log(audit_context(), GFP_ATOMIC,
8341 AUDIT_ANOM_PROMISCUOUS,
8342 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8343 dev->name, (dev->flags & IFF_PROMISC),
8344 (old_flags & IFF_PROMISC),
8345 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8346 from_kuid(&init_user_ns, uid),
8347 from_kgid(&init_user_ns, gid),
8348 audit_get_sessionid(current));
8351 dev_change_rx_flags(dev, IFF_PROMISC);
8354 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8359 * dev_set_promiscuity - update promiscuity count on a device
8363 * Add or remove promiscuity from a device. While the count in the device
8364 * remains above zero the interface remains promiscuous. Once it hits zero
8365 * the device reverts back to normal filtering operation. A negative inc
8366 * value is used to drop promiscuity on the device.
8367 * Return 0 if successful or a negative errno code on error.
8369 int dev_set_promiscuity(struct net_device *dev, int inc)
8371 unsigned int old_flags = dev->flags;
8374 err = __dev_set_promiscuity(dev, inc, true);
8377 if (dev->flags != old_flags)
8378 dev_set_rx_mode(dev);
8381 EXPORT_SYMBOL(dev_set_promiscuity);
8383 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8385 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8389 dev->flags |= IFF_ALLMULTI;
8390 dev->allmulti += inc;
8391 if (dev->allmulti == 0) {
8394 * If inc causes overflow, untouch allmulti and return error.
8397 dev->flags &= ~IFF_ALLMULTI;
8399 dev->allmulti -= inc;
8400 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8404 if (dev->flags ^ old_flags) {
8405 dev_change_rx_flags(dev, IFF_ALLMULTI);
8406 dev_set_rx_mode(dev);
8408 __dev_notify_flags(dev, old_flags,
8409 dev->gflags ^ old_gflags);
8415 * dev_set_allmulti - update allmulti count on a device
8419 * Add or remove reception of all multicast frames to a device. While the
8420 * count in the device remains above zero the interface remains listening
8421 * to all interfaces. Once it hits zero the device reverts back to normal
8422 * filtering operation. A negative @inc value is used to drop the counter
8423 * when releasing a resource needing all multicasts.
8424 * Return 0 if successful or a negative errno code on error.
8427 int dev_set_allmulti(struct net_device *dev, int inc)
8429 return __dev_set_allmulti(dev, inc, true);
8431 EXPORT_SYMBOL(dev_set_allmulti);
8434 * Upload unicast and multicast address lists to device and
8435 * configure RX filtering. When the device doesn't support unicast
8436 * filtering it is put in promiscuous mode while unicast addresses
8439 void __dev_set_rx_mode(struct net_device *dev)
8441 const struct net_device_ops *ops = dev->netdev_ops;
8443 /* dev_open will call this function so the list will stay sane. */
8444 if (!(dev->flags&IFF_UP))
8447 if (!netif_device_present(dev))
8450 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8451 /* Unicast addresses changes may only happen under the rtnl,
8452 * therefore calling __dev_set_promiscuity here is safe.
8454 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8455 __dev_set_promiscuity(dev, 1, false);
8456 dev->uc_promisc = true;
8457 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8458 __dev_set_promiscuity(dev, -1, false);
8459 dev->uc_promisc = false;
8463 if (ops->ndo_set_rx_mode)
8464 ops->ndo_set_rx_mode(dev);
8467 void dev_set_rx_mode(struct net_device *dev)
8469 netif_addr_lock_bh(dev);
8470 __dev_set_rx_mode(dev);
8471 netif_addr_unlock_bh(dev);
8475 * dev_get_flags - get flags reported to userspace
8478 * Get the combination of flag bits exported through APIs to userspace.
8480 unsigned int dev_get_flags(const struct net_device *dev)
8484 flags = (dev->flags & ~(IFF_PROMISC |
8489 (dev->gflags & (IFF_PROMISC |
8492 if (netif_running(dev)) {
8493 if (netif_oper_up(dev))
8494 flags |= IFF_RUNNING;
8495 if (netif_carrier_ok(dev))
8496 flags |= IFF_LOWER_UP;
8497 if (netif_dormant(dev))
8498 flags |= IFF_DORMANT;
8503 EXPORT_SYMBOL(dev_get_flags);
8505 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8506 struct netlink_ext_ack *extack)
8508 unsigned int old_flags = dev->flags;
8514 * Set the flags on our device.
8517 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8518 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8520 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8524 * Load in the correct multicast list now the flags have changed.
8527 if ((old_flags ^ flags) & IFF_MULTICAST)
8528 dev_change_rx_flags(dev, IFF_MULTICAST);
8530 dev_set_rx_mode(dev);
8533 * Have we downed the interface. We handle IFF_UP ourselves
8534 * according to user attempts to set it, rather than blindly
8539 if ((old_flags ^ flags) & IFF_UP) {
8540 if (old_flags & IFF_UP)
8543 ret = __dev_open(dev, extack);
8546 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8547 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8548 unsigned int old_flags = dev->flags;
8550 dev->gflags ^= IFF_PROMISC;
8552 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8553 if (dev->flags != old_flags)
8554 dev_set_rx_mode(dev);
8557 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8558 * is important. Some (broken) drivers set IFF_PROMISC, when
8559 * IFF_ALLMULTI is requested not asking us and not reporting.
8561 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8562 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8564 dev->gflags ^= IFF_ALLMULTI;
8565 __dev_set_allmulti(dev, inc, false);
8571 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8572 unsigned int gchanges)
8574 unsigned int changes = dev->flags ^ old_flags;
8577 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8579 if (changes & IFF_UP) {
8580 if (dev->flags & IFF_UP)
8581 call_netdevice_notifiers(NETDEV_UP, dev);
8583 call_netdevice_notifiers(NETDEV_DOWN, dev);
8586 if (dev->flags & IFF_UP &&
8587 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8588 struct netdev_notifier_change_info change_info = {
8592 .flags_changed = changes,
8595 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8600 * dev_change_flags - change device settings
8602 * @flags: device state flags
8603 * @extack: netlink extended ack
8605 * Change settings on device based state flags. The flags are
8606 * in the userspace exported format.
8608 int dev_change_flags(struct net_device *dev, unsigned int flags,
8609 struct netlink_ext_ack *extack)
8612 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8614 ret = __dev_change_flags(dev, flags, extack);
8618 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8619 __dev_notify_flags(dev, old_flags, changes);
8622 EXPORT_SYMBOL(dev_change_flags);
8624 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8626 const struct net_device_ops *ops = dev->netdev_ops;
8628 if (ops->ndo_change_mtu)
8629 return ops->ndo_change_mtu(dev, new_mtu);
8631 /* Pairs with all the lockless reads of dev->mtu in the stack */
8632 WRITE_ONCE(dev->mtu, new_mtu);
8635 EXPORT_SYMBOL(__dev_set_mtu);
8637 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8638 struct netlink_ext_ack *extack)
8640 /* MTU must be positive, and in range */
8641 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8642 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8646 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8647 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8654 * dev_set_mtu_ext - Change maximum transfer unit
8656 * @new_mtu: new transfer unit
8657 * @extack: netlink extended ack
8659 * Change the maximum transfer size of the network device.
8661 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8662 struct netlink_ext_ack *extack)
8666 if (new_mtu == dev->mtu)
8669 err = dev_validate_mtu(dev, new_mtu, extack);
8673 if (!netif_device_present(dev))
8676 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8677 err = notifier_to_errno(err);
8681 orig_mtu = dev->mtu;
8682 err = __dev_set_mtu(dev, new_mtu);
8685 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8687 err = notifier_to_errno(err);
8689 /* setting mtu back and notifying everyone again,
8690 * so that they have a chance to revert changes.
8692 __dev_set_mtu(dev, orig_mtu);
8693 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8700 int dev_set_mtu(struct net_device *dev, int new_mtu)
8702 struct netlink_ext_ack extack;
8705 memset(&extack, 0, sizeof(extack));
8706 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8707 if (err && extack._msg)
8708 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8711 EXPORT_SYMBOL(dev_set_mtu);
8714 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8716 * @new_len: new tx queue length
8718 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8720 unsigned int orig_len = dev->tx_queue_len;
8723 if (new_len != (unsigned int)new_len)
8726 if (new_len != orig_len) {
8727 dev->tx_queue_len = new_len;
8728 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8729 res = notifier_to_errno(res);
8732 res = dev_qdisc_change_tx_queue_len(dev);
8740 netdev_err(dev, "refused to change device tx_queue_len\n");
8741 dev->tx_queue_len = orig_len;
8746 * dev_set_group - Change group this device belongs to
8748 * @new_group: group this device should belong to
8750 void dev_set_group(struct net_device *dev, int new_group)
8752 dev->group = new_group;
8756 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8758 * @addr: new address
8759 * @extack: netlink extended ack
8761 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8762 struct netlink_ext_ack *extack)
8764 struct netdev_notifier_pre_changeaddr_info info = {
8766 .info.extack = extack,
8771 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8772 return notifier_to_errno(rc);
8774 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8777 * dev_set_mac_address - Change Media Access Control Address
8780 * @extack: netlink extended ack
8782 * Change the hardware (MAC) address of the device
8784 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8785 struct netlink_ext_ack *extack)
8787 const struct net_device_ops *ops = dev->netdev_ops;
8790 if (!ops->ndo_set_mac_address)
8792 if (sa->sa_family != dev->type)
8794 if (!netif_device_present(dev))
8796 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8799 err = ops->ndo_set_mac_address(dev, sa);
8802 dev->addr_assign_type = NET_ADDR_SET;
8803 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8804 add_device_randomness(dev->dev_addr, dev->addr_len);
8807 EXPORT_SYMBOL(dev_set_mac_address);
8809 static DECLARE_RWSEM(dev_addr_sem);
8811 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8812 struct netlink_ext_ack *extack)
8816 down_write(&dev_addr_sem);
8817 ret = dev_set_mac_address(dev, sa, extack);
8818 up_write(&dev_addr_sem);
8821 EXPORT_SYMBOL(dev_set_mac_address_user);
8823 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8825 size_t size = sizeof(sa->sa_data);
8826 struct net_device *dev;
8829 down_read(&dev_addr_sem);
8832 dev = dev_get_by_name_rcu(net, dev_name);
8838 memset(sa->sa_data, 0, size);
8840 memcpy(sa->sa_data, dev->dev_addr,
8841 min_t(size_t, size, dev->addr_len));
8842 sa->sa_family = dev->type;
8846 up_read(&dev_addr_sem);
8849 EXPORT_SYMBOL(dev_get_mac_address);
8852 * dev_change_carrier - Change device carrier
8854 * @new_carrier: new value
8856 * Change device carrier
8858 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8860 const struct net_device_ops *ops = dev->netdev_ops;
8862 if (!ops->ndo_change_carrier)
8864 if (!netif_device_present(dev))
8866 return ops->ndo_change_carrier(dev, new_carrier);
8870 * dev_get_phys_port_id - Get device physical port ID
8874 * Get device physical port ID
8876 int dev_get_phys_port_id(struct net_device *dev,
8877 struct netdev_phys_item_id *ppid)
8879 const struct net_device_ops *ops = dev->netdev_ops;
8881 if (!ops->ndo_get_phys_port_id)
8883 return ops->ndo_get_phys_port_id(dev, ppid);
8887 * dev_get_phys_port_name - Get device physical port name
8890 * @len: limit of bytes to copy to name
8892 * Get device physical port name
8894 int dev_get_phys_port_name(struct net_device *dev,
8895 char *name, size_t len)
8897 const struct net_device_ops *ops = dev->netdev_ops;
8900 if (ops->ndo_get_phys_port_name) {
8901 err = ops->ndo_get_phys_port_name(dev, name, len);
8902 if (err != -EOPNOTSUPP)
8905 return devlink_compat_phys_port_name_get(dev, name, len);
8909 * dev_get_port_parent_id - Get the device's port parent identifier
8910 * @dev: network device
8911 * @ppid: pointer to a storage for the port's parent identifier
8912 * @recurse: allow/disallow recursion to lower devices
8914 * Get the devices's port parent identifier
8916 int dev_get_port_parent_id(struct net_device *dev,
8917 struct netdev_phys_item_id *ppid,
8920 const struct net_device_ops *ops = dev->netdev_ops;
8921 struct netdev_phys_item_id first = { };
8922 struct net_device *lower_dev;
8923 struct list_head *iter;
8926 if (ops->ndo_get_port_parent_id) {
8927 err = ops->ndo_get_port_parent_id(dev, ppid);
8928 if (err != -EOPNOTSUPP)
8932 err = devlink_compat_switch_id_get(dev, ppid);
8933 if (!recurse || err != -EOPNOTSUPP)
8936 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8937 err = dev_get_port_parent_id(lower_dev, ppid, true);
8942 else if (memcmp(&first, ppid, sizeof(*ppid)))
8948 EXPORT_SYMBOL(dev_get_port_parent_id);
8951 * netdev_port_same_parent_id - Indicate if two network devices have
8952 * the same port parent identifier
8953 * @a: first network device
8954 * @b: second network device
8956 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8958 struct netdev_phys_item_id a_id = { };
8959 struct netdev_phys_item_id b_id = { };
8961 if (dev_get_port_parent_id(a, &a_id, true) ||
8962 dev_get_port_parent_id(b, &b_id, true))
8965 return netdev_phys_item_id_same(&a_id, &b_id);
8967 EXPORT_SYMBOL(netdev_port_same_parent_id);
8970 * dev_change_proto_down - set carrier according to proto_down.
8973 * @proto_down: new value
8975 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8977 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8979 if (!netif_device_present(dev))
8982 netif_carrier_off(dev);
8984 netif_carrier_on(dev);
8985 dev->proto_down = proto_down;
8990 * dev_change_proto_down_reason - proto down reason
8993 * @mask: proto down mask
8994 * @value: proto down value
8996 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9002 dev->proto_down_reason = value;
9004 for_each_set_bit(b, &mask, 32) {
9005 if (value & (1 << b))
9006 dev->proto_down_reason |= BIT(b);
9008 dev->proto_down_reason &= ~BIT(b);
9013 struct bpf_xdp_link {
9014 struct bpf_link link;
9015 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9019 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9021 if (flags & XDP_FLAGS_HW_MODE)
9023 if (flags & XDP_FLAGS_DRV_MODE)
9024 return XDP_MODE_DRV;
9025 if (flags & XDP_FLAGS_SKB_MODE)
9026 return XDP_MODE_SKB;
9027 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9030 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9034 return generic_xdp_install;
9037 return dev->netdev_ops->ndo_bpf;
9043 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9044 enum bpf_xdp_mode mode)
9046 return dev->xdp_state[mode].link;
9049 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9050 enum bpf_xdp_mode mode)
9052 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9055 return link->link.prog;
9056 return dev->xdp_state[mode].prog;
9059 u8 dev_xdp_prog_count(struct net_device *dev)
9064 for (i = 0; i < __MAX_XDP_MODE; i++)
9065 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9069 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9071 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9073 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9075 return prog ? prog->aux->id : 0;
9078 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9079 struct bpf_xdp_link *link)
9081 dev->xdp_state[mode].link = link;
9082 dev->xdp_state[mode].prog = NULL;
9085 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9086 struct bpf_prog *prog)
9088 dev->xdp_state[mode].link = NULL;
9089 dev->xdp_state[mode].prog = prog;
9092 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9093 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9094 u32 flags, struct bpf_prog *prog)
9096 struct netdev_bpf xdp;
9099 memset(&xdp, 0, sizeof(xdp));
9100 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9101 xdp.extack = extack;
9105 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9106 * "moved" into driver), so they don't increment it on their own, but
9107 * they do decrement refcnt when program is detached or replaced.
9108 * Given net_device also owns link/prog, we need to bump refcnt here
9109 * to prevent drivers from underflowing it.
9113 err = bpf_op(dev, &xdp);
9120 if (mode != XDP_MODE_HW)
9121 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9126 static void dev_xdp_uninstall(struct net_device *dev)
9128 struct bpf_xdp_link *link;
9129 struct bpf_prog *prog;
9130 enum bpf_xdp_mode mode;
9135 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9136 prog = dev_xdp_prog(dev, mode);
9140 bpf_op = dev_xdp_bpf_op(dev, mode);
9144 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9146 /* auto-detach link from net device */
9147 link = dev_xdp_link(dev, mode);
9153 dev_xdp_set_link(dev, mode, NULL);
9157 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9158 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9159 struct bpf_prog *old_prog, u32 flags)
9161 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9162 struct bpf_prog *cur_prog;
9163 struct net_device *upper;
9164 struct list_head *iter;
9165 enum bpf_xdp_mode mode;
9171 /* either link or prog attachment, never both */
9172 if (link && (new_prog || old_prog))
9174 /* link supports only XDP mode flags */
9175 if (link && (flags & ~XDP_FLAGS_MODES)) {
9176 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9179 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9180 if (num_modes > 1) {
9181 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9184 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9185 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9186 NL_SET_ERR_MSG(extack,
9187 "More than one program loaded, unset mode is ambiguous");
9190 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9191 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9192 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9196 mode = dev_xdp_mode(dev, flags);
9197 /* can't replace attached link */
9198 if (dev_xdp_link(dev, mode)) {
9199 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9203 /* don't allow if an upper device already has a program */
9204 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9205 if (dev_xdp_prog_count(upper) > 0) {
9206 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9211 cur_prog = dev_xdp_prog(dev, mode);
9212 /* can't replace attached prog with link */
9213 if (link && cur_prog) {
9214 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9217 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9218 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9222 /* put effective new program into new_prog */
9224 new_prog = link->link.prog;
9227 bool offload = mode == XDP_MODE_HW;
9228 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9229 ? XDP_MODE_DRV : XDP_MODE_SKB;
9231 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9232 NL_SET_ERR_MSG(extack, "XDP program already attached");
9235 if (!offload && dev_xdp_prog(dev, other_mode)) {
9236 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9239 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9240 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9243 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9244 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9247 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9248 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9253 /* don't call drivers if the effective program didn't change */
9254 if (new_prog != cur_prog) {
9255 bpf_op = dev_xdp_bpf_op(dev, mode);
9257 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9261 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9267 dev_xdp_set_link(dev, mode, link);
9269 dev_xdp_set_prog(dev, mode, new_prog);
9271 bpf_prog_put(cur_prog);
9276 static int dev_xdp_attach_link(struct net_device *dev,
9277 struct netlink_ext_ack *extack,
9278 struct bpf_xdp_link *link)
9280 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9283 static int dev_xdp_detach_link(struct net_device *dev,
9284 struct netlink_ext_ack *extack,
9285 struct bpf_xdp_link *link)
9287 enum bpf_xdp_mode mode;
9292 mode = dev_xdp_mode(dev, link->flags);
9293 if (dev_xdp_link(dev, mode) != link)
9296 bpf_op = dev_xdp_bpf_op(dev, mode);
9297 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9298 dev_xdp_set_link(dev, mode, NULL);
9302 static void bpf_xdp_link_release(struct bpf_link *link)
9304 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9308 /* if racing with net_device's tear down, xdp_link->dev might be
9309 * already NULL, in which case link was already auto-detached
9311 if (xdp_link->dev) {
9312 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9313 xdp_link->dev = NULL;
9319 static int bpf_xdp_link_detach(struct bpf_link *link)
9321 bpf_xdp_link_release(link);
9325 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9327 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9332 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9333 struct seq_file *seq)
9335 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9340 ifindex = xdp_link->dev->ifindex;
9343 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9346 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9347 struct bpf_link_info *info)
9349 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9354 ifindex = xdp_link->dev->ifindex;
9357 info->xdp.ifindex = ifindex;
9361 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9362 struct bpf_prog *old_prog)
9364 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9365 enum bpf_xdp_mode mode;
9371 /* link might have been auto-released already, so fail */
9372 if (!xdp_link->dev) {
9377 if (old_prog && link->prog != old_prog) {
9381 old_prog = link->prog;
9382 if (old_prog->type != new_prog->type ||
9383 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9388 if (old_prog == new_prog) {
9389 /* no-op, don't disturb drivers */
9390 bpf_prog_put(new_prog);
9394 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9395 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9396 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9397 xdp_link->flags, new_prog);
9401 old_prog = xchg(&link->prog, new_prog);
9402 bpf_prog_put(old_prog);
9409 static const struct bpf_link_ops bpf_xdp_link_lops = {
9410 .release = bpf_xdp_link_release,
9411 .dealloc = bpf_xdp_link_dealloc,
9412 .detach = bpf_xdp_link_detach,
9413 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9414 .fill_link_info = bpf_xdp_link_fill_link_info,
9415 .update_prog = bpf_xdp_link_update,
9418 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9420 struct net *net = current->nsproxy->net_ns;
9421 struct bpf_link_primer link_primer;
9422 struct bpf_xdp_link *link;
9423 struct net_device *dev;
9427 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9433 link = kzalloc(sizeof(*link), GFP_USER);
9439 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9441 link->flags = attr->link_create.flags;
9443 err = bpf_link_prime(&link->link, &link_primer);
9449 err = dev_xdp_attach_link(dev, NULL, link);
9454 bpf_link_cleanup(&link_primer);
9458 fd = bpf_link_settle(&link_primer);
9459 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9472 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9474 * @extack: netlink extended ack
9475 * @fd: new program fd or negative value to clear
9476 * @expected_fd: old program fd that userspace expects to replace or clear
9477 * @flags: xdp-related flags
9479 * Set or clear a bpf program for a device
9481 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9482 int fd, int expected_fd, u32 flags)
9484 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9485 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9491 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9492 mode != XDP_MODE_SKB);
9493 if (IS_ERR(new_prog))
9494 return PTR_ERR(new_prog);
9497 if (expected_fd >= 0) {
9498 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9499 mode != XDP_MODE_SKB);
9500 if (IS_ERR(old_prog)) {
9501 err = PTR_ERR(old_prog);
9507 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9510 if (err && new_prog)
9511 bpf_prog_put(new_prog);
9513 bpf_prog_put(old_prog);
9518 * dev_new_index - allocate an ifindex
9519 * @net: the applicable net namespace
9521 * Returns a suitable unique value for a new device interface
9522 * number. The caller must hold the rtnl semaphore or the
9523 * dev_base_lock to be sure it remains unique.
9525 static int dev_new_index(struct net *net)
9527 int ifindex = net->ifindex;
9532 if (!__dev_get_by_index(net, ifindex))
9533 return net->ifindex = ifindex;
9537 /* Delayed registration/unregisteration */
9538 LIST_HEAD(net_todo_list);
9539 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9541 static void net_set_todo(struct net_device *dev)
9543 list_add_tail(&dev->todo_list, &net_todo_list);
9544 atomic_inc(&dev_net(dev)->dev_unreg_count);
9547 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9548 struct net_device *upper, netdev_features_t features)
9550 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9551 netdev_features_t feature;
9554 for_each_netdev_feature(upper_disables, feature_bit) {
9555 feature = __NETIF_F_BIT(feature_bit);
9556 if (!(upper->wanted_features & feature)
9557 && (features & feature)) {
9558 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9559 &feature, upper->name);
9560 features &= ~feature;
9567 static void netdev_sync_lower_features(struct net_device *upper,
9568 struct net_device *lower, netdev_features_t features)
9570 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9571 netdev_features_t feature;
9574 for_each_netdev_feature(upper_disables, feature_bit) {
9575 feature = __NETIF_F_BIT(feature_bit);
9576 if (!(features & feature) && (lower->features & feature)) {
9577 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9578 &feature, lower->name);
9579 lower->wanted_features &= ~feature;
9580 __netdev_update_features(lower);
9582 if (unlikely(lower->features & feature))
9583 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9584 &feature, lower->name);
9586 netdev_features_change(lower);
9591 static netdev_features_t netdev_fix_features(struct net_device *dev,
9592 netdev_features_t features)
9594 /* Fix illegal checksum combinations */
9595 if ((features & NETIF_F_HW_CSUM) &&
9596 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9597 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9598 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9601 /* TSO requires that SG is present as well. */
9602 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9603 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9604 features &= ~NETIF_F_ALL_TSO;
9607 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9608 !(features & NETIF_F_IP_CSUM)) {
9609 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9610 features &= ~NETIF_F_TSO;
9611 features &= ~NETIF_F_TSO_ECN;
9614 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9615 !(features & NETIF_F_IPV6_CSUM)) {
9616 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9617 features &= ~NETIF_F_TSO6;
9620 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9621 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9622 features &= ~NETIF_F_TSO_MANGLEID;
9624 /* TSO ECN requires that TSO is present as well. */
9625 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9626 features &= ~NETIF_F_TSO_ECN;
9628 /* Software GSO depends on SG. */
9629 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9630 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9631 features &= ~NETIF_F_GSO;
9634 /* GSO partial features require GSO partial be set */
9635 if ((features & dev->gso_partial_features) &&
9636 !(features & NETIF_F_GSO_PARTIAL)) {
9638 "Dropping partially supported GSO features since no GSO partial.\n");
9639 features &= ~dev->gso_partial_features;
9642 if (!(features & NETIF_F_RXCSUM)) {
9643 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9644 * successfully merged by hardware must also have the
9645 * checksum verified by hardware. If the user does not
9646 * want to enable RXCSUM, logically, we should disable GRO_HW.
9648 if (features & NETIF_F_GRO_HW) {
9649 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9650 features &= ~NETIF_F_GRO_HW;
9654 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9655 if (features & NETIF_F_RXFCS) {
9656 if (features & NETIF_F_LRO) {
9657 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9658 features &= ~NETIF_F_LRO;
9661 if (features & NETIF_F_GRO_HW) {
9662 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9663 features &= ~NETIF_F_GRO_HW;
9667 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9668 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9669 features &= ~NETIF_F_LRO;
9672 if (features & NETIF_F_HW_TLS_TX) {
9673 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9674 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9675 bool hw_csum = features & NETIF_F_HW_CSUM;
9677 if (!ip_csum && !hw_csum) {
9678 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9679 features &= ~NETIF_F_HW_TLS_TX;
9683 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9684 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9685 features &= ~NETIF_F_HW_TLS_RX;
9691 int __netdev_update_features(struct net_device *dev)
9693 struct net_device *upper, *lower;
9694 netdev_features_t features;
9695 struct list_head *iter;
9700 features = netdev_get_wanted_features(dev);
9702 if (dev->netdev_ops->ndo_fix_features)
9703 features = dev->netdev_ops->ndo_fix_features(dev, features);
9705 /* driver might be less strict about feature dependencies */
9706 features = netdev_fix_features(dev, features);
9708 /* some features can't be enabled if they're off on an upper device */
9709 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9710 features = netdev_sync_upper_features(dev, upper, features);
9712 if (dev->features == features)
9715 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9716 &dev->features, &features);
9718 if (dev->netdev_ops->ndo_set_features)
9719 err = dev->netdev_ops->ndo_set_features(dev, features);
9723 if (unlikely(err < 0)) {
9725 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9726 err, &features, &dev->features);
9727 /* return non-0 since some features might have changed and
9728 * it's better to fire a spurious notification than miss it
9734 /* some features must be disabled on lower devices when disabled
9735 * on an upper device (think: bonding master or bridge)
9737 netdev_for_each_lower_dev(dev, lower, iter)
9738 netdev_sync_lower_features(dev, lower, features);
9741 netdev_features_t diff = features ^ dev->features;
9743 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9744 /* udp_tunnel_{get,drop}_rx_info both need
9745 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9746 * device, or they won't do anything.
9747 * Thus we need to update dev->features
9748 * *before* calling udp_tunnel_get_rx_info,
9749 * but *after* calling udp_tunnel_drop_rx_info.
9751 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9752 dev->features = features;
9753 udp_tunnel_get_rx_info(dev);
9755 udp_tunnel_drop_rx_info(dev);
9759 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9760 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9761 dev->features = features;
9762 err |= vlan_get_rx_ctag_filter_info(dev);
9764 vlan_drop_rx_ctag_filter_info(dev);
9768 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9769 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9770 dev->features = features;
9771 err |= vlan_get_rx_stag_filter_info(dev);
9773 vlan_drop_rx_stag_filter_info(dev);
9777 dev->features = features;
9780 return err < 0 ? 0 : 1;
9784 * netdev_update_features - recalculate device features
9785 * @dev: the device to check
9787 * Recalculate dev->features set and send notifications if it
9788 * has changed. Should be called after driver or hardware dependent
9789 * conditions might have changed that influence the features.
9791 void netdev_update_features(struct net_device *dev)
9793 if (__netdev_update_features(dev))
9794 netdev_features_change(dev);
9796 EXPORT_SYMBOL(netdev_update_features);
9799 * netdev_change_features - recalculate device features
9800 * @dev: the device to check
9802 * Recalculate dev->features set and send notifications even
9803 * if they have not changed. Should be called instead of
9804 * netdev_update_features() if also dev->vlan_features might
9805 * have changed to allow the changes to be propagated to stacked
9808 void netdev_change_features(struct net_device *dev)
9810 __netdev_update_features(dev);
9811 netdev_features_change(dev);
9813 EXPORT_SYMBOL(netdev_change_features);
9816 * netif_stacked_transfer_operstate - transfer operstate
9817 * @rootdev: the root or lower level device to transfer state from
9818 * @dev: the device to transfer operstate to
9820 * Transfer operational state from root to device. This is normally
9821 * called when a stacking relationship exists between the root
9822 * device and the device(a leaf device).
9824 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9825 struct net_device *dev)
9827 if (rootdev->operstate == IF_OPER_DORMANT)
9828 netif_dormant_on(dev);
9830 netif_dormant_off(dev);
9832 if (rootdev->operstate == IF_OPER_TESTING)
9833 netif_testing_on(dev);
9835 netif_testing_off(dev);
9837 if (netif_carrier_ok(rootdev))
9838 netif_carrier_on(dev);
9840 netif_carrier_off(dev);
9842 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9844 static int netif_alloc_rx_queues(struct net_device *dev)
9846 unsigned int i, count = dev->num_rx_queues;
9847 struct netdev_rx_queue *rx;
9848 size_t sz = count * sizeof(*rx);
9853 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9859 for (i = 0; i < count; i++) {
9862 /* XDP RX-queue setup */
9863 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9870 /* Rollback successful reg's and free other resources */
9872 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9878 static void netif_free_rx_queues(struct net_device *dev)
9880 unsigned int i, count = dev->num_rx_queues;
9882 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9886 for (i = 0; i < count; i++)
9887 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9892 static void netdev_init_one_queue(struct net_device *dev,
9893 struct netdev_queue *queue, void *_unused)
9895 /* Initialize queue lock */
9896 spin_lock_init(&queue->_xmit_lock);
9897 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9898 queue->xmit_lock_owner = -1;
9899 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9902 dql_init(&queue->dql, HZ);
9906 static void netif_free_tx_queues(struct net_device *dev)
9911 static int netif_alloc_netdev_queues(struct net_device *dev)
9913 unsigned int count = dev->num_tx_queues;
9914 struct netdev_queue *tx;
9915 size_t sz = count * sizeof(*tx);
9917 if (count < 1 || count > 0xffff)
9920 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9926 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9927 spin_lock_init(&dev->tx_global_lock);
9932 void netif_tx_stop_all_queues(struct net_device *dev)
9936 for (i = 0; i < dev->num_tx_queues; i++) {
9937 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9939 netif_tx_stop_queue(txq);
9942 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9945 * register_netdevice() - register a network device
9946 * @dev: device to register
9948 * Take a prepared network device structure and make it externally accessible.
9949 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9950 * Callers must hold the rtnl lock - you may want register_netdev()
9953 int register_netdevice(struct net_device *dev)
9956 struct net *net = dev_net(dev);
9958 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9959 NETDEV_FEATURE_COUNT);
9960 BUG_ON(dev_boot_phase);
9965 /* When net_device's are persistent, this will be fatal. */
9966 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9969 ret = ethtool_check_ops(dev->ethtool_ops);
9973 spin_lock_init(&dev->addr_list_lock);
9974 netdev_set_addr_lockdep_class(dev);
9976 ret = dev_get_valid_name(net, dev, dev->name);
9981 dev->name_node = netdev_name_node_head_alloc(dev);
9982 if (!dev->name_node)
9985 /* Init, if this function is available */
9986 if (dev->netdev_ops->ndo_init) {
9987 ret = dev->netdev_ops->ndo_init(dev);
9995 if (((dev->hw_features | dev->features) &
9996 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9997 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9998 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9999 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10006 dev->ifindex = dev_new_index(net);
10007 else if (__dev_get_by_index(net, dev->ifindex))
10010 /* Transfer changeable features to wanted_features and enable
10011 * software offloads (GSO and GRO).
10013 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10014 dev->features |= NETIF_F_SOFT_FEATURES;
10016 if (dev->udp_tunnel_nic_info) {
10017 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10018 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10021 dev->wanted_features = dev->features & dev->hw_features;
10023 if (!(dev->flags & IFF_LOOPBACK))
10024 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10026 /* If IPv4 TCP segmentation offload is supported we should also
10027 * allow the device to enable segmenting the frame with the option
10028 * of ignoring a static IP ID value. This doesn't enable the
10029 * feature itself but allows the user to enable it later.
10031 if (dev->hw_features & NETIF_F_TSO)
10032 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10033 if (dev->vlan_features & NETIF_F_TSO)
10034 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10035 if (dev->mpls_features & NETIF_F_TSO)
10036 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10037 if (dev->hw_enc_features & NETIF_F_TSO)
10038 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10040 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10042 dev->vlan_features |= NETIF_F_HIGHDMA;
10044 /* Make NETIF_F_SG inheritable to tunnel devices.
10046 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10048 /* Make NETIF_F_SG inheritable to MPLS.
10050 dev->mpls_features |= NETIF_F_SG;
10052 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10053 ret = notifier_to_errno(ret);
10057 ret = netdev_register_kobject(dev);
10058 write_lock(&dev_base_lock);
10059 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10060 write_unlock(&dev_base_lock);
10064 __netdev_update_features(dev);
10067 * Default initial state at registry is that the
10068 * device is present.
10071 set_bit(__LINK_STATE_PRESENT, &dev->state);
10073 linkwatch_init_dev(dev);
10075 dev_init_scheduler(dev);
10077 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10078 list_netdevice(dev);
10080 add_device_randomness(dev->dev_addr, dev->addr_len);
10082 /* If the device has permanent device address, driver should
10083 * set dev_addr and also addr_assign_type should be set to
10084 * NET_ADDR_PERM (default value).
10086 if (dev->addr_assign_type == NET_ADDR_PERM)
10087 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10089 /* Notify protocols, that a new device appeared. */
10090 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10091 ret = notifier_to_errno(ret);
10093 /* Expect explicit free_netdev() on failure */
10094 dev->needs_free_netdev = false;
10095 unregister_netdevice_queue(dev, NULL);
10099 * Prevent userspace races by waiting until the network
10100 * device is fully setup before sending notifications.
10102 if (!dev->rtnl_link_ops ||
10103 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10104 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10110 if (dev->netdev_ops->ndo_uninit)
10111 dev->netdev_ops->ndo_uninit(dev);
10112 if (dev->priv_destructor)
10113 dev->priv_destructor(dev);
10115 netdev_name_node_free(dev->name_node);
10118 EXPORT_SYMBOL(register_netdevice);
10121 * init_dummy_netdev - init a dummy network device for NAPI
10122 * @dev: device to init
10124 * This takes a network device structure and initialize the minimum
10125 * amount of fields so it can be used to schedule NAPI polls without
10126 * registering a full blown interface. This is to be used by drivers
10127 * that need to tie several hardware interfaces to a single NAPI
10128 * poll scheduler due to HW limitations.
10130 int init_dummy_netdev(struct net_device *dev)
10132 /* Clear everything. Note we don't initialize spinlocks
10133 * are they aren't supposed to be taken by any of the
10134 * NAPI code and this dummy netdev is supposed to be
10135 * only ever used for NAPI polls
10137 memset(dev, 0, sizeof(struct net_device));
10139 /* make sure we BUG if trying to hit standard
10140 * register/unregister code path
10142 dev->reg_state = NETREG_DUMMY;
10144 /* NAPI wants this */
10145 INIT_LIST_HEAD(&dev->napi_list);
10147 /* a dummy interface is started by default */
10148 set_bit(__LINK_STATE_PRESENT, &dev->state);
10149 set_bit(__LINK_STATE_START, &dev->state);
10151 /* napi_busy_loop stats accounting wants this */
10152 dev_net_set(dev, &init_net);
10154 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10155 * because users of this 'device' dont need to change
10161 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10165 * register_netdev - register a network device
10166 * @dev: device to register
10168 * Take a completed network device structure and add it to the kernel
10169 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10170 * chain. 0 is returned on success. A negative errno code is returned
10171 * on a failure to set up the device, or if the name is a duplicate.
10173 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10174 * and expands the device name if you passed a format string to
10177 int register_netdev(struct net_device *dev)
10181 if (rtnl_lock_killable())
10183 err = register_netdevice(dev);
10187 EXPORT_SYMBOL(register_netdev);
10189 int netdev_refcnt_read(const struct net_device *dev)
10191 #ifdef CONFIG_PCPU_DEV_REFCNT
10194 for_each_possible_cpu(i)
10195 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10198 return refcount_read(&dev->dev_refcnt);
10201 EXPORT_SYMBOL(netdev_refcnt_read);
10203 int netdev_unregister_timeout_secs __read_mostly = 10;
10205 #define WAIT_REFS_MIN_MSECS 1
10206 #define WAIT_REFS_MAX_MSECS 250
10208 * netdev_wait_allrefs_any - wait until all references are gone.
10209 * @list: list of net_devices to wait on
10211 * This is called when unregistering network devices.
10213 * Any protocol or device that holds a reference should register
10214 * for netdevice notification, and cleanup and put back the
10215 * reference if they receive an UNREGISTER event.
10216 * We can get stuck here if buggy protocols don't correctly
10219 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10221 unsigned long rebroadcast_time, warning_time;
10222 struct net_device *dev;
10225 rebroadcast_time = warning_time = jiffies;
10227 list_for_each_entry(dev, list, todo_list)
10228 if (netdev_refcnt_read(dev) == 1)
10232 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10235 /* Rebroadcast unregister notification */
10236 list_for_each_entry(dev, list, todo_list)
10237 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10243 list_for_each_entry(dev, list, todo_list)
10244 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10246 /* We must not have linkwatch events
10247 * pending on unregister. If this
10248 * happens, we simply run the queue
10249 * unscheduled, resulting in a noop
10252 linkwatch_run_queue();
10258 rebroadcast_time = jiffies;
10263 wait = WAIT_REFS_MIN_MSECS;
10266 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10269 list_for_each_entry(dev, list, todo_list)
10270 if (netdev_refcnt_read(dev) == 1)
10273 if (time_after(jiffies, warning_time +
10274 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10275 list_for_each_entry(dev, list, todo_list) {
10276 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10277 dev->name, netdev_refcnt_read(dev));
10278 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10281 warning_time = jiffies;
10286 /* The sequence is:
10290 * register_netdevice(x1);
10291 * register_netdevice(x2);
10293 * unregister_netdevice(y1);
10294 * unregister_netdevice(y2);
10300 * We are invoked by rtnl_unlock().
10301 * This allows us to deal with problems:
10302 * 1) We can delete sysfs objects which invoke hotplug
10303 * without deadlocking with linkwatch via keventd.
10304 * 2) Since we run with the RTNL semaphore not held, we can sleep
10305 * safely in order to wait for the netdev refcnt to drop to zero.
10307 * We must not return until all unregister events added during
10308 * the interval the lock was held have been completed.
10310 void netdev_run_todo(void)
10312 struct net_device *dev, *tmp;
10313 struct list_head list;
10314 #ifdef CONFIG_LOCKDEP
10315 struct list_head unlink_list;
10317 list_replace_init(&net_unlink_list, &unlink_list);
10319 while (!list_empty(&unlink_list)) {
10320 struct net_device *dev = list_first_entry(&unlink_list,
10323 list_del_init(&dev->unlink_list);
10324 dev->nested_level = dev->lower_level - 1;
10328 /* Snapshot list, allow later requests */
10329 list_replace_init(&net_todo_list, &list);
10333 /* Wait for rcu callbacks to finish before next phase */
10334 if (!list_empty(&list))
10337 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10338 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10339 netdev_WARN(dev, "run_todo but not unregistering\n");
10340 list_del(&dev->todo_list);
10344 write_lock(&dev_base_lock);
10345 dev->reg_state = NETREG_UNREGISTERED;
10346 write_unlock(&dev_base_lock);
10347 linkwatch_forget_dev(dev);
10350 while (!list_empty(&list)) {
10351 dev = netdev_wait_allrefs_any(&list);
10352 list_del(&dev->todo_list);
10355 BUG_ON(netdev_refcnt_read(dev) != 1);
10356 BUG_ON(!list_empty(&dev->ptype_all));
10357 BUG_ON(!list_empty(&dev->ptype_specific));
10358 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10359 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10361 if (dev->priv_destructor)
10362 dev->priv_destructor(dev);
10363 if (dev->needs_free_netdev)
10366 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10367 wake_up(&netdev_unregistering_wq);
10369 /* Free network device */
10370 kobject_put(&dev->dev.kobj);
10374 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10375 * all the same fields in the same order as net_device_stats, with only
10376 * the type differing, but rtnl_link_stats64 may have additional fields
10377 * at the end for newer counters.
10379 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10380 const struct net_device_stats *netdev_stats)
10382 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10383 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10384 u64 *dst = (u64 *)stats64;
10386 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10387 for (i = 0; i < n; i++)
10388 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10389 /* zero out counters that only exist in rtnl_link_stats64 */
10390 memset((char *)stats64 + n * sizeof(u64), 0,
10391 sizeof(*stats64) - n * sizeof(u64));
10393 EXPORT_SYMBOL(netdev_stats_to_stats64);
10395 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10397 struct net_device_core_stats __percpu *p;
10399 p = alloc_percpu_gfp(struct net_device_core_stats,
10400 GFP_ATOMIC | __GFP_NOWARN);
10402 if (p && cmpxchg(&dev->core_stats, NULL, p))
10405 /* This READ_ONCE() pairs with the cmpxchg() above */
10406 return READ_ONCE(dev->core_stats);
10408 EXPORT_SYMBOL(netdev_core_stats_alloc);
10411 * dev_get_stats - get network device statistics
10412 * @dev: device to get statistics from
10413 * @storage: place to store stats
10415 * Get network statistics from device. Return @storage.
10416 * The device driver may provide its own method by setting
10417 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10418 * otherwise the internal statistics structure is used.
10420 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10421 struct rtnl_link_stats64 *storage)
10423 const struct net_device_ops *ops = dev->netdev_ops;
10424 const struct net_device_core_stats __percpu *p;
10426 if (ops->ndo_get_stats64) {
10427 memset(storage, 0, sizeof(*storage));
10428 ops->ndo_get_stats64(dev, storage);
10429 } else if (ops->ndo_get_stats) {
10430 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10432 netdev_stats_to_stats64(storage, &dev->stats);
10435 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10436 p = READ_ONCE(dev->core_stats);
10438 const struct net_device_core_stats *core_stats;
10441 for_each_possible_cpu(i) {
10442 core_stats = per_cpu_ptr(p, i);
10443 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10444 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10445 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10446 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10451 EXPORT_SYMBOL(dev_get_stats);
10454 * dev_fetch_sw_netstats - get per-cpu network device statistics
10455 * @s: place to store stats
10456 * @netstats: per-cpu network stats to read from
10458 * Read per-cpu network statistics and populate the related fields in @s.
10460 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10461 const struct pcpu_sw_netstats __percpu *netstats)
10465 for_each_possible_cpu(cpu) {
10466 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10467 const struct pcpu_sw_netstats *stats;
10468 unsigned int start;
10470 stats = per_cpu_ptr(netstats, cpu);
10472 start = u64_stats_fetch_begin_irq(&stats->syncp);
10473 rx_packets = u64_stats_read(&stats->rx_packets);
10474 rx_bytes = u64_stats_read(&stats->rx_bytes);
10475 tx_packets = u64_stats_read(&stats->tx_packets);
10476 tx_bytes = u64_stats_read(&stats->tx_bytes);
10477 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10479 s->rx_packets += rx_packets;
10480 s->rx_bytes += rx_bytes;
10481 s->tx_packets += tx_packets;
10482 s->tx_bytes += tx_bytes;
10485 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10488 * dev_get_tstats64 - ndo_get_stats64 implementation
10489 * @dev: device to get statistics from
10490 * @s: place to store stats
10492 * Populate @s from dev->stats and dev->tstats. Can be used as
10493 * ndo_get_stats64() callback.
10495 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10497 netdev_stats_to_stats64(s, &dev->stats);
10498 dev_fetch_sw_netstats(s, dev->tstats);
10500 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10502 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10504 struct netdev_queue *queue = dev_ingress_queue(dev);
10506 #ifdef CONFIG_NET_CLS_ACT
10509 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10512 netdev_init_one_queue(dev, queue, NULL);
10513 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10514 queue->qdisc_sleeping = &noop_qdisc;
10515 rcu_assign_pointer(dev->ingress_queue, queue);
10520 static const struct ethtool_ops default_ethtool_ops;
10522 void netdev_set_default_ethtool_ops(struct net_device *dev,
10523 const struct ethtool_ops *ops)
10525 if (dev->ethtool_ops == &default_ethtool_ops)
10526 dev->ethtool_ops = ops;
10528 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10530 void netdev_freemem(struct net_device *dev)
10532 char *addr = (char *)dev - dev->padded;
10538 * alloc_netdev_mqs - allocate network device
10539 * @sizeof_priv: size of private data to allocate space for
10540 * @name: device name format string
10541 * @name_assign_type: origin of device name
10542 * @setup: callback to initialize device
10543 * @txqs: the number of TX subqueues to allocate
10544 * @rxqs: the number of RX subqueues to allocate
10546 * Allocates a struct net_device with private data area for driver use
10547 * and performs basic initialization. Also allocates subqueue structs
10548 * for each queue on the device.
10550 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10551 unsigned char name_assign_type,
10552 void (*setup)(struct net_device *),
10553 unsigned int txqs, unsigned int rxqs)
10555 struct net_device *dev;
10556 unsigned int alloc_size;
10557 struct net_device *p;
10559 BUG_ON(strlen(name) >= sizeof(dev->name));
10562 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10567 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10571 alloc_size = sizeof(struct net_device);
10573 /* ensure 32-byte alignment of private area */
10574 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10575 alloc_size += sizeof_priv;
10577 /* ensure 32-byte alignment of whole construct */
10578 alloc_size += NETDEV_ALIGN - 1;
10580 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10584 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10585 dev->padded = (char *)dev - (char *)p;
10587 ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10588 #ifdef CONFIG_PCPU_DEV_REFCNT
10589 dev->pcpu_refcnt = alloc_percpu(int);
10590 if (!dev->pcpu_refcnt)
10594 refcount_set(&dev->dev_refcnt, 1);
10597 if (dev_addr_init(dev))
10603 dev_net_set(dev, &init_net);
10605 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10606 dev->gso_max_segs = GSO_MAX_SEGS;
10607 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10608 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10609 dev->tso_max_segs = TSO_MAX_SEGS;
10610 dev->upper_level = 1;
10611 dev->lower_level = 1;
10612 #ifdef CONFIG_LOCKDEP
10613 dev->nested_level = 0;
10614 INIT_LIST_HEAD(&dev->unlink_list);
10617 INIT_LIST_HEAD(&dev->napi_list);
10618 INIT_LIST_HEAD(&dev->unreg_list);
10619 INIT_LIST_HEAD(&dev->close_list);
10620 INIT_LIST_HEAD(&dev->link_watch_list);
10621 INIT_LIST_HEAD(&dev->adj_list.upper);
10622 INIT_LIST_HEAD(&dev->adj_list.lower);
10623 INIT_LIST_HEAD(&dev->ptype_all);
10624 INIT_LIST_HEAD(&dev->ptype_specific);
10625 INIT_LIST_HEAD(&dev->net_notifier_list);
10626 #ifdef CONFIG_NET_SCHED
10627 hash_init(dev->qdisc_hash);
10629 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10632 if (!dev->tx_queue_len) {
10633 dev->priv_flags |= IFF_NO_QUEUE;
10634 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10637 dev->num_tx_queues = txqs;
10638 dev->real_num_tx_queues = txqs;
10639 if (netif_alloc_netdev_queues(dev))
10642 dev->num_rx_queues = rxqs;
10643 dev->real_num_rx_queues = rxqs;
10644 if (netif_alloc_rx_queues(dev))
10647 strcpy(dev->name, name);
10648 dev->name_assign_type = name_assign_type;
10649 dev->group = INIT_NETDEV_GROUP;
10650 if (!dev->ethtool_ops)
10651 dev->ethtool_ops = &default_ethtool_ops;
10653 nf_hook_netdev_init(dev);
10662 #ifdef CONFIG_PCPU_DEV_REFCNT
10663 free_percpu(dev->pcpu_refcnt);
10666 netdev_freemem(dev);
10669 EXPORT_SYMBOL(alloc_netdev_mqs);
10672 * free_netdev - free network device
10675 * This function does the last stage of destroying an allocated device
10676 * interface. The reference to the device object is released. If this
10677 * is the last reference then it will be freed.Must be called in process
10680 void free_netdev(struct net_device *dev)
10682 struct napi_struct *p, *n;
10686 /* When called immediately after register_netdevice() failed the unwind
10687 * handling may still be dismantling the device. Handle that case by
10688 * deferring the free.
10690 if (dev->reg_state == NETREG_UNREGISTERING) {
10692 dev->needs_free_netdev = true;
10696 netif_free_tx_queues(dev);
10697 netif_free_rx_queues(dev);
10699 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10701 /* Flush device addresses */
10702 dev_addr_flush(dev);
10704 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10707 ref_tracker_dir_exit(&dev->refcnt_tracker);
10708 #ifdef CONFIG_PCPU_DEV_REFCNT
10709 free_percpu(dev->pcpu_refcnt);
10710 dev->pcpu_refcnt = NULL;
10712 free_percpu(dev->core_stats);
10713 dev->core_stats = NULL;
10714 free_percpu(dev->xdp_bulkq);
10715 dev->xdp_bulkq = NULL;
10717 /* Compatibility with error handling in drivers */
10718 if (dev->reg_state == NETREG_UNINITIALIZED) {
10719 netdev_freemem(dev);
10723 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10724 dev->reg_state = NETREG_RELEASED;
10726 /* will free via device release */
10727 put_device(&dev->dev);
10729 EXPORT_SYMBOL(free_netdev);
10732 * synchronize_net - Synchronize with packet receive processing
10734 * Wait for packets currently being received to be done.
10735 * Does not block later packets from starting.
10737 void synchronize_net(void)
10740 if (rtnl_is_locked())
10741 synchronize_rcu_expedited();
10745 EXPORT_SYMBOL(synchronize_net);
10748 * unregister_netdevice_queue - remove device from the kernel
10752 * This function shuts down a device interface and removes it
10753 * from the kernel tables.
10754 * If head not NULL, device is queued to be unregistered later.
10756 * Callers must hold the rtnl semaphore. You may want
10757 * unregister_netdev() instead of this.
10760 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10765 list_move_tail(&dev->unreg_list, head);
10769 list_add(&dev->unreg_list, &single);
10770 unregister_netdevice_many(&single);
10773 EXPORT_SYMBOL(unregister_netdevice_queue);
10776 * unregister_netdevice_many - unregister many devices
10777 * @head: list of devices
10779 * Note: As most callers use a stack allocated list_head,
10780 * we force a list_del() to make sure stack wont be corrupted later.
10782 void unregister_netdevice_many(struct list_head *head)
10784 struct net_device *dev, *tmp;
10785 LIST_HEAD(close_head);
10787 BUG_ON(dev_boot_phase);
10790 if (list_empty(head))
10793 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10794 /* Some devices call without registering
10795 * for initialization unwind. Remove those
10796 * devices and proceed with the remaining.
10798 if (dev->reg_state == NETREG_UNINITIALIZED) {
10799 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10803 list_del(&dev->unreg_list);
10806 dev->dismantle = true;
10807 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10810 /* If device is running, close it first. */
10811 list_for_each_entry(dev, head, unreg_list)
10812 list_add_tail(&dev->close_list, &close_head);
10813 dev_close_many(&close_head, true);
10815 list_for_each_entry(dev, head, unreg_list) {
10816 /* And unlink it from device chain. */
10817 write_lock(&dev_base_lock);
10818 unlist_netdevice(dev, false);
10819 dev->reg_state = NETREG_UNREGISTERING;
10820 write_unlock(&dev_base_lock);
10822 flush_all_backlogs();
10826 list_for_each_entry(dev, head, unreg_list) {
10827 struct sk_buff *skb = NULL;
10829 /* Shutdown queueing discipline. */
10832 dev_xdp_uninstall(dev);
10834 netdev_offload_xstats_disable_all(dev);
10836 /* Notify protocols, that we are about to destroy
10837 * this device. They should clean all the things.
10839 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10841 if (!dev->rtnl_link_ops ||
10842 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10843 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10844 GFP_KERNEL, NULL, 0);
10847 * Flush the unicast and multicast chains
10852 netdev_name_node_alt_flush(dev);
10853 netdev_name_node_free(dev->name_node);
10855 if (dev->netdev_ops->ndo_uninit)
10856 dev->netdev_ops->ndo_uninit(dev);
10859 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10861 /* Notifier chain MUST detach us all upper devices. */
10862 WARN_ON(netdev_has_any_upper_dev(dev));
10863 WARN_ON(netdev_has_any_lower_dev(dev));
10865 /* Remove entries from kobject tree */
10866 netdev_unregister_kobject(dev);
10868 /* Remove XPS queueing entries */
10869 netif_reset_xps_queues_gt(dev, 0);
10875 list_for_each_entry(dev, head, unreg_list) {
10876 netdev_put(dev, &dev->dev_registered_tracker);
10882 EXPORT_SYMBOL(unregister_netdevice_many);
10885 * unregister_netdev - remove device from the kernel
10888 * This function shuts down a device interface and removes it
10889 * from the kernel tables.
10891 * This is just a wrapper for unregister_netdevice that takes
10892 * the rtnl semaphore. In general you want to use this and not
10893 * unregister_netdevice.
10895 void unregister_netdev(struct net_device *dev)
10898 unregister_netdevice(dev);
10901 EXPORT_SYMBOL(unregister_netdev);
10904 * __dev_change_net_namespace - move device to different nethost namespace
10906 * @net: network namespace
10907 * @pat: If not NULL name pattern to try if the current device name
10908 * is already taken in the destination network namespace.
10909 * @new_ifindex: If not zero, specifies device index in the target
10912 * This function shuts down a device interface and moves it
10913 * to a new network namespace. On success 0 is returned, on
10914 * a failure a netagive errno code is returned.
10916 * Callers must hold the rtnl semaphore.
10919 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10920 const char *pat, int new_ifindex)
10922 struct net *net_old = dev_net(dev);
10927 /* Don't allow namespace local devices to be moved. */
10929 if (dev->features & NETIF_F_NETNS_LOCAL)
10932 /* Ensure the device has been registrered */
10933 if (dev->reg_state != NETREG_REGISTERED)
10936 /* Get out if there is nothing todo */
10938 if (net_eq(net_old, net))
10941 /* Pick the destination device name, and ensure
10942 * we can use it in the destination network namespace.
10945 if (netdev_name_in_use(net, dev->name)) {
10946 /* We get here if we can't use the current device name */
10949 err = dev_get_valid_name(net, dev, pat);
10954 /* Check that new_ifindex isn't used yet. */
10956 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10960 * And now a mini version of register_netdevice unregister_netdevice.
10963 /* If device is running close it first. */
10966 /* And unlink it from device chain */
10967 unlist_netdevice(dev, true);
10971 /* Shutdown queueing discipline. */
10974 /* Notify protocols, that we are about to destroy
10975 * this device. They should clean all the things.
10977 * Note that dev->reg_state stays at NETREG_REGISTERED.
10978 * This is wanted because this way 8021q and macvlan know
10979 * the device is just moving and can keep their slaves up.
10981 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10984 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10985 /* If there is an ifindex conflict assign a new one */
10986 if (!new_ifindex) {
10987 if (__dev_get_by_index(net, dev->ifindex))
10988 new_ifindex = dev_new_index(net);
10990 new_ifindex = dev->ifindex;
10993 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10997 * Flush the unicast and multicast chains
11002 /* Send a netdev-removed uevent to the old namespace */
11003 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11004 netdev_adjacent_del_links(dev);
11006 /* Move per-net netdevice notifiers that are following the netdevice */
11007 move_netdevice_notifiers_dev_net(dev, net);
11009 /* Actually switch the network namespace */
11010 dev_net_set(dev, net);
11011 dev->ifindex = new_ifindex;
11013 /* Send a netdev-add uevent to the new namespace */
11014 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11015 netdev_adjacent_add_links(dev);
11017 /* Fixup kobjects */
11018 err = device_rename(&dev->dev, dev->name);
11021 /* Adapt owner in case owning user namespace of target network
11022 * namespace is different from the original one.
11024 err = netdev_change_owner(dev, net_old, net);
11027 /* Add the device back in the hashes */
11028 list_netdevice(dev);
11030 /* Notify protocols, that a new device appeared. */
11031 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11034 * Prevent userspace races by waiting until the network
11035 * device is fully setup before sending notifications.
11037 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11044 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11046 static int dev_cpu_dead(unsigned int oldcpu)
11048 struct sk_buff **list_skb;
11049 struct sk_buff *skb;
11051 struct softnet_data *sd, *oldsd, *remsd = NULL;
11053 local_irq_disable();
11054 cpu = smp_processor_id();
11055 sd = &per_cpu(softnet_data, cpu);
11056 oldsd = &per_cpu(softnet_data, oldcpu);
11058 /* Find end of our completion_queue. */
11059 list_skb = &sd->completion_queue;
11061 list_skb = &(*list_skb)->next;
11062 /* Append completion queue from offline CPU. */
11063 *list_skb = oldsd->completion_queue;
11064 oldsd->completion_queue = NULL;
11066 /* Append output queue from offline CPU. */
11067 if (oldsd->output_queue) {
11068 *sd->output_queue_tailp = oldsd->output_queue;
11069 sd->output_queue_tailp = oldsd->output_queue_tailp;
11070 oldsd->output_queue = NULL;
11071 oldsd->output_queue_tailp = &oldsd->output_queue;
11073 /* Append NAPI poll list from offline CPU, with one exception :
11074 * process_backlog() must be called by cpu owning percpu backlog.
11075 * We properly handle process_queue & input_pkt_queue later.
11077 while (!list_empty(&oldsd->poll_list)) {
11078 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11079 struct napi_struct,
11082 list_del_init(&napi->poll_list);
11083 if (napi->poll == process_backlog)
11086 ____napi_schedule(sd, napi);
11089 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11090 local_irq_enable();
11093 remsd = oldsd->rps_ipi_list;
11094 oldsd->rps_ipi_list = NULL;
11096 /* send out pending IPI's on offline CPU */
11097 net_rps_send_ipi(remsd);
11099 /* Process offline CPU's input_pkt_queue */
11100 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11102 input_queue_head_incr(oldsd);
11104 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11106 input_queue_head_incr(oldsd);
11113 * netdev_increment_features - increment feature set by one
11114 * @all: current feature set
11115 * @one: new feature set
11116 * @mask: mask feature set
11118 * Computes a new feature set after adding a device with feature set
11119 * @one to the master device with current feature set @all. Will not
11120 * enable anything that is off in @mask. Returns the new feature set.
11122 netdev_features_t netdev_increment_features(netdev_features_t all,
11123 netdev_features_t one, netdev_features_t mask)
11125 if (mask & NETIF_F_HW_CSUM)
11126 mask |= NETIF_F_CSUM_MASK;
11127 mask |= NETIF_F_VLAN_CHALLENGED;
11129 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11130 all &= one | ~NETIF_F_ALL_FOR_ALL;
11132 /* If one device supports hw checksumming, set for all. */
11133 if (all & NETIF_F_HW_CSUM)
11134 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11138 EXPORT_SYMBOL(netdev_increment_features);
11140 static struct hlist_head * __net_init netdev_create_hash(void)
11143 struct hlist_head *hash;
11145 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11147 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11148 INIT_HLIST_HEAD(&hash[i]);
11153 /* Initialize per network namespace state */
11154 static int __net_init netdev_init(struct net *net)
11156 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11157 8 * sizeof_field(struct napi_struct, gro_bitmask));
11159 INIT_LIST_HEAD(&net->dev_base_head);
11161 net->dev_name_head = netdev_create_hash();
11162 if (net->dev_name_head == NULL)
11165 net->dev_index_head = netdev_create_hash();
11166 if (net->dev_index_head == NULL)
11169 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11174 kfree(net->dev_name_head);
11180 * netdev_drivername - network driver for the device
11181 * @dev: network device
11183 * Determine network driver for device.
11185 const char *netdev_drivername(const struct net_device *dev)
11187 const struct device_driver *driver;
11188 const struct device *parent;
11189 const char *empty = "";
11191 parent = dev->dev.parent;
11195 driver = parent->driver;
11196 if (driver && driver->name)
11197 return driver->name;
11201 static void __netdev_printk(const char *level, const struct net_device *dev,
11202 struct va_format *vaf)
11204 if (dev && dev->dev.parent) {
11205 dev_printk_emit(level[1] - '0',
11208 dev_driver_string(dev->dev.parent),
11209 dev_name(dev->dev.parent),
11210 netdev_name(dev), netdev_reg_state(dev),
11213 printk("%s%s%s: %pV",
11214 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11216 printk("%s(NULL net_device): %pV", level, vaf);
11220 void netdev_printk(const char *level, const struct net_device *dev,
11221 const char *format, ...)
11223 struct va_format vaf;
11226 va_start(args, format);
11231 __netdev_printk(level, dev, &vaf);
11235 EXPORT_SYMBOL(netdev_printk);
11237 #define define_netdev_printk_level(func, level) \
11238 void func(const struct net_device *dev, const char *fmt, ...) \
11240 struct va_format vaf; \
11243 va_start(args, fmt); \
11248 __netdev_printk(level, dev, &vaf); \
11252 EXPORT_SYMBOL(func);
11254 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11255 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11256 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11257 define_netdev_printk_level(netdev_err, KERN_ERR);
11258 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11259 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11260 define_netdev_printk_level(netdev_info, KERN_INFO);
11262 static void __net_exit netdev_exit(struct net *net)
11264 kfree(net->dev_name_head);
11265 kfree(net->dev_index_head);
11266 if (net != &init_net)
11267 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11270 static struct pernet_operations __net_initdata netdev_net_ops = {
11271 .init = netdev_init,
11272 .exit = netdev_exit,
11275 static void __net_exit default_device_exit_net(struct net *net)
11277 struct net_device *dev, *aux;
11279 * Push all migratable network devices back to the
11280 * initial network namespace
11283 for_each_netdev_safe(net, dev, aux) {
11285 char fb_name[IFNAMSIZ];
11287 /* Ignore unmoveable devices (i.e. loopback) */
11288 if (dev->features & NETIF_F_NETNS_LOCAL)
11291 /* Leave virtual devices for the generic cleanup */
11292 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11295 /* Push remaining network devices to init_net */
11296 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11297 if (netdev_name_in_use(&init_net, fb_name))
11298 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11299 err = dev_change_net_namespace(dev, &init_net, fb_name);
11301 pr_emerg("%s: failed to move %s to init_net: %d\n",
11302 __func__, dev->name, err);
11308 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11310 /* At exit all network devices most be removed from a network
11311 * namespace. Do this in the reverse order of registration.
11312 * Do this across as many network namespaces as possible to
11313 * improve batching efficiency.
11315 struct net_device *dev;
11317 LIST_HEAD(dev_kill_list);
11320 list_for_each_entry(net, net_list, exit_list) {
11321 default_device_exit_net(net);
11325 list_for_each_entry(net, net_list, exit_list) {
11326 for_each_netdev_reverse(net, dev) {
11327 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11328 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11330 unregister_netdevice_queue(dev, &dev_kill_list);
11333 unregister_netdevice_many(&dev_kill_list);
11337 static struct pernet_operations __net_initdata default_device_ops = {
11338 .exit_batch = default_device_exit_batch,
11342 * Initialize the DEV module. At boot time this walks the device list and
11343 * unhooks any devices that fail to initialise (normally hardware not
11344 * present) and leaves us with a valid list of present and active devices.
11349 * This is called single threaded during boot, so no need
11350 * to take the rtnl semaphore.
11352 static int __init net_dev_init(void)
11354 int i, rc = -ENOMEM;
11356 BUG_ON(!dev_boot_phase);
11358 if (dev_proc_init())
11361 if (netdev_kobject_init())
11364 INIT_LIST_HEAD(&ptype_all);
11365 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11366 INIT_LIST_HEAD(&ptype_base[i]);
11368 if (register_pernet_subsys(&netdev_net_ops))
11372 * Initialise the packet receive queues.
11375 for_each_possible_cpu(i) {
11376 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11377 struct softnet_data *sd = &per_cpu(softnet_data, i);
11379 INIT_WORK(flush, flush_backlog);
11381 skb_queue_head_init(&sd->input_pkt_queue);
11382 skb_queue_head_init(&sd->process_queue);
11383 #ifdef CONFIG_XFRM_OFFLOAD
11384 skb_queue_head_init(&sd->xfrm_backlog);
11386 INIT_LIST_HEAD(&sd->poll_list);
11387 sd->output_queue_tailp = &sd->output_queue;
11389 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11392 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11393 spin_lock_init(&sd->defer_lock);
11395 init_gro_hash(&sd->backlog);
11396 sd->backlog.poll = process_backlog;
11397 sd->backlog.weight = weight_p;
11400 dev_boot_phase = 0;
11402 /* The loopback device is special if any other network devices
11403 * is present in a network namespace the loopback device must
11404 * be present. Since we now dynamically allocate and free the
11405 * loopback device ensure this invariant is maintained by
11406 * keeping the loopback device as the first device on the
11407 * list of network devices. Ensuring the loopback devices
11408 * is the first device that appears and the last network device
11411 if (register_pernet_device(&loopback_net_ops))
11414 if (register_pernet_device(&default_device_ops))
11417 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11418 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11420 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11421 NULL, dev_cpu_dead);
11428 subsys_initcall(net_dev_init);