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);
3149 else if (unlikely(reason == SKB_REASON_DROPPED))
3154 EXPORT_SYMBOL(__dev_kfree_skb_any);
3158 * netif_device_detach - mark device as removed
3159 * @dev: network device
3161 * Mark device as removed from system and therefore no longer available.
3163 void netif_device_detach(struct net_device *dev)
3165 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3166 netif_running(dev)) {
3167 netif_tx_stop_all_queues(dev);
3170 EXPORT_SYMBOL(netif_device_detach);
3173 * netif_device_attach - mark device as attached
3174 * @dev: network device
3176 * Mark device as attached from system and restart if needed.
3178 void netif_device_attach(struct net_device *dev)
3180 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3181 netif_running(dev)) {
3182 netif_tx_wake_all_queues(dev);
3183 __netdev_watchdog_up(dev);
3186 EXPORT_SYMBOL(netif_device_attach);
3189 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3190 * to be used as a distribution range.
3192 static u16 skb_tx_hash(const struct net_device *dev,
3193 const struct net_device *sb_dev,
3194 struct sk_buff *skb)
3198 u16 qcount = dev->real_num_tx_queues;
3201 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3203 qoffset = sb_dev->tc_to_txq[tc].offset;
3204 qcount = sb_dev->tc_to_txq[tc].count;
3205 if (unlikely(!qcount)) {
3206 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3207 sb_dev->name, qoffset, tc);
3209 qcount = dev->real_num_tx_queues;
3213 if (skb_rx_queue_recorded(skb)) {
3214 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3215 hash = skb_get_rx_queue(skb);
3216 if (hash >= qoffset)
3218 while (unlikely(hash >= qcount))
3220 return hash + qoffset;
3223 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3226 static void skb_warn_bad_offload(const struct sk_buff *skb)
3228 static const netdev_features_t null_features;
3229 struct net_device *dev = skb->dev;
3230 const char *name = "";
3232 if (!net_ratelimit())
3236 if (dev->dev.parent)
3237 name = dev_driver_string(dev->dev.parent);
3239 name = netdev_name(dev);
3241 skb_dump(KERN_WARNING, skb, false);
3242 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3243 name, dev ? &dev->features : &null_features,
3244 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3248 * Invalidate hardware checksum when packet is to be mangled, and
3249 * complete checksum manually on outgoing path.
3251 int skb_checksum_help(struct sk_buff *skb)
3254 int ret = 0, offset;
3256 if (skb->ip_summed == CHECKSUM_COMPLETE)
3257 goto out_set_summed;
3259 if (unlikely(skb_is_gso(skb))) {
3260 skb_warn_bad_offload(skb);
3264 /* Before computing a checksum, we should make sure no frag could
3265 * be modified by an external entity : checksum could be wrong.
3267 if (skb_has_shared_frag(skb)) {
3268 ret = __skb_linearize(skb);
3273 offset = skb_checksum_start_offset(skb);
3275 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3276 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3279 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3281 offset += skb->csum_offset;
3282 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3283 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3286 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3290 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3292 skb->ip_summed = CHECKSUM_NONE;
3296 EXPORT_SYMBOL(skb_checksum_help);
3298 int skb_crc32c_csum_help(struct sk_buff *skb)
3301 int ret = 0, offset, start;
3303 if (skb->ip_summed != CHECKSUM_PARTIAL)
3306 if (unlikely(skb_is_gso(skb)))
3309 /* Before computing a checksum, we should make sure no frag could
3310 * be modified by an external entity : checksum could be wrong.
3312 if (unlikely(skb_has_shared_frag(skb))) {
3313 ret = __skb_linearize(skb);
3317 start = skb_checksum_start_offset(skb);
3318 offset = start + offsetof(struct sctphdr, checksum);
3319 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3324 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3328 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3329 skb->len - start, ~(__u32)0,
3331 *(__le32 *)(skb->data + offset) = crc32c_csum;
3332 skb->ip_summed = CHECKSUM_NONE;
3333 skb->csum_not_inet = 0;
3338 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3340 __be16 type = skb->protocol;
3342 /* Tunnel gso handlers can set protocol to ethernet. */
3343 if (type == htons(ETH_P_TEB)) {
3346 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3349 eth = (struct ethhdr *)skb->data;
3350 type = eth->h_proto;
3353 return __vlan_get_protocol(skb, type, depth);
3356 /* openvswitch calls this on rx path, so we need a different check.
3358 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3361 return skb->ip_summed != CHECKSUM_PARTIAL &&
3362 skb->ip_summed != CHECKSUM_UNNECESSARY;
3364 return skb->ip_summed == CHECKSUM_NONE;
3368 * __skb_gso_segment - Perform segmentation on skb.
3369 * @skb: buffer to segment
3370 * @features: features for the output path (see dev->features)
3371 * @tx_path: whether it is called in TX path
3373 * This function segments the given skb and returns a list of segments.
3375 * It may return NULL if the skb requires no segmentation. This is
3376 * only possible when GSO is used for verifying header integrity.
3378 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3380 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3381 netdev_features_t features, bool tx_path)
3383 struct sk_buff *segs;
3385 if (unlikely(skb_needs_check(skb, tx_path))) {
3388 /* We're going to init ->check field in TCP or UDP header */
3389 err = skb_cow_head(skb, 0);
3391 return ERR_PTR(err);
3394 /* Only report GSO partial support if it will enable us to
3395 * support segmentation on this frame without needing additional
3398 if (features & NETIF_F_GSO_PARTIAL) {
3399 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3400 struct net_device *dev = skb->dev;
3402 partial_features |= dev->features & dev->gso_partial_features;
3403 if (!skb_gso_ok(skb, features | partial_features))
3404 features &= ~NETIF_F_GSO_PARTIAL;
3407 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3408 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3410 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3411 SKB_GSO_CB(skb)->encap_level = 0;
3413 skb_reset_mac_header(skb);
3414 skb_reset_mac_len(skb);
3416 segs = skb_mac_gso_segment(skb, features);
3418 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3419 skb_warn_bad_offload(skb);
3423 EXPORT_SYMBOL(__skb_gso_segment);
3425 /* Take action when hardware reception checksum errors are detected. */
3427 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3429 netdev_err(dev, "hw csum failure\n");
3430 skb_dump(KERN_ERR, skb, true);
3434 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3436 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3438 EXPORT_SYMBOL(netdev_rx_csum_fault);
3441 /* XXX: check that highmem exists at all on the given machine. */
3442 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3444 #ifdef CONFIG_HIGHMEM
3447 if (!(dev->features & NETIF_F_HIGHDMA)) {
3448 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3449 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3451 if (PageHighMem(skb_frag_page(frag)))
3459 /* If MPLS offload request, verify we are testing hardware MPLS features
3460 * instead of standard features for the netdev.
3462 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3463 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3464 netdev_features_t features,
3467 if (eth_p_mpls(type))
3468 features &= skb->dev->mpls_features;
3473 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3474 netdev_features_t features,
3481 static netdev_features_t harmonize_features(struct sk_buff *skb,
3482 netdev_features_t features)
3486 type = skb_network_protocol(skb, NULL);
3487 features = net_mpls_features(skb, features, type);
3489 if (skb->ip_summed != CHECKSUM_NONE &&
3490 !can_checksum_protocol(features, type)) {
3491 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3493 if (illegal_highdma(skb->dev, skb))
3494 features &= ~NETIF_F_SG;
3499 netdev_features_t passthru_features_check(struct sk_buff *skb,
3500 struct net_device *dev,
3501 netdev_features_t features)
3505 EXPORT_SYMBOL(passthru_features_check);
3507 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3508 struct net_device *dev,
3509 netdev_features_t features)
3511 return vlan_features_check(skb, features);
3514 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3515 struct net_device *dev,
3516 netdev_features_t features)
3518 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3520 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3521 return features & ~NETIF_F_GSO_MASK;
3523 if (!skb_shinfo(skb)->gso_type) {
3524 skb_warn_bad_offload(skb);
3525 return features & ~NETIF_F_GSO_MASK;
3528 /* Support for GSO partial features requires software
3529 * intervention before we can actually process the packets
3530 * so we need to strip support for any partial features now
3531 * and we can pull them back in after we have partially
3532 * segmented the frame.
3534 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3535 features &= ~dev->gso_partial_features;
3537 /* Make sure to clear the IPv4 ID mangling feature if the
3538 * IPv4 header has the potential to be fragmented.
3540 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3541 struct iphdr *iph = skb->encapsulation ?
3542 inner_ip_hdr(skb) : ip_hdr(skb);
3544 if (!(iph->frag_off & htons(IP_DF)))
3545 features &= ~NETIF_F_TSO_MANGLEID;
3551 netdev_features_t netif_skb_features(struct sk_buff *skb)
3553 struct net_device *dev = skb->dev;
3554 netdev_features_t features = dev->features;
3556 if (skb_is_gso(skb))
3557 features = gso_features_check(skb, dev, features);
3559 /* If encapsulation offload request, verify we are testing
3560 * hardware encapsulation features instead of standard
3561 * features for the netdev
3563 if (skb->encapsulation)
3564 features &= dev->hw_enc_features;
3566 if (skb_vlan_tagged(skb))
3567 features = netdev_intersect_features(features,
3568 dev->vlan_features |
3569 NETIF_F_HW_VLAN_CTAG_TX |
3570 NETIF_F_HW_VLAN_STAG_TX);
3572 if (dev->netdev_ops->ndo_features_check)
3573 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3576 features &= dflt_features_check(skb, dev, features);
3578 return harmonize_features(skb, features);
3580 EXPORT_SYMBOL(netif_skb_features);
3582 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3583 struct netdev_queue *txq, bool more)
3588 if (dev_nit_active(dev))
3589 dev_queue_xmit_nit(skb, dev);
3592 trace_net_dev_start_xmit(skb, dev);
3593 rc = netdev_start_xmit(skb, dev, txq, more);
3594 trace_net_dev_xmit(skb, rc, dev, len);
3599 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3600 struct netdev_queue *txq, int *ret)
3602 struct sk_buff *skb = first;
3603 int rc = NETDEV_TX_OK;
3606 struct sk_buff *next = skb->next;
3608 skb_mark_not_on_list(skb);
3609 rc = xmit_one(skb, dev, txq, next != NULL);
3610 if (unlikely(!dev_xmit_complete(rc))) {
3616 if (netif_tx_queue_stopped(txq) && skb) {
3617 rc = NETDEV_TX_BUSY;
3627 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3628 netdev_features_t features)
3630 if (skb_vlan_tag_present(skb) &&
3631 !vlan_hw_offload_capable(features, skb->vlan_proto))
3632 skb = __vlan_hwaccel_push_inside(skb);
3636 int skb_csum_hwoffload_help(struct sk_buff *skb,
3637 const netdev_features_t features)
3639 if (unlikely(skb_csum_is_sctp(skb)))
3640 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3641 skb_crc32c_csum_help(skb);
3643 if (features & NETIF_F_HW_CSUM)
3646 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3647 switch (skb->csum_offset) {
3648 case offsetof(struct tcphdr, check):
3649 case offsetof(struct udphdr, check):
3654 return skb_checksum_help(skb);
3656 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3658 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3660 netdev_features_t features;
3662 features = netif_skb_features(skb);
3663 skb = validate_xmit_vlan(skb, features);
3667 skb = sk_validate_xmit_skb(skb, dev);
3671 if (netif_needs_gso(skb, features)) {
3672 struct sk_buff *segs;
3674 segs = skb_gso_segment(skb, features);
3682 if (skb_needs_linearize(skb, features) &&
3683 __skb_linearize(skb))
3686 /* If packet is not checksummed and device does not
3687 * support checksumming for this protocol, complete
3688 * checksumming here.
3690 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3691 if (skb->encapsulation)
3692 skb_set_inner_transport_header(skb,
3693 skb_checksum_start_offset(skb));
3695 skb_set_transport_header(skb,
3696 skb_checksum_start_offset(skb));
3697 if (skb_csum_hwoffload_help(skb, features))
3702 skb = validate_xmit_xfrm(skb, features, again);
3709 dev_core_stats_tx_dropped_inc(dev);
3713 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3715 struct sk_buff *next, *head = NULL, *tail;
3717 for (; skb != NULL; skb = next) {
3719 skb_mark_not_on_list(skb);
3721 /* in case skb wont be segmented, point to itself */
3724 skb = validate_xmit_skb(skb, dev, again);
3732 /* If skb was segmented, skb->prev points to
3733 * the last segment. If not, it still contains skb.
3739 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3741 static void qdisc_pkt_len_init(struct sk_buff *skb)
3743 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3745 qdisc_skb_cb(skb)->pkt_len = skb->len;
3747 /* To get more precise estimation of bytes sent on wire,
3748 * we add to pkt_len the headers size of all segments
3750 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3751 unsigned int hdr_len;
3752 u16 gso_segs = shinfo->gso_segs;
3754 /* mac layer + network layer */
3755 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3757 /* + transport layer */
3758 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3759 const struct tcphdr *th;
3760 struct tcphdr _tcphdr;
3762 th = skb_header_pointer(skb, skb_transport_offset(skb),
3763 sizeof(_tcphdr), &_tcphdr);
3765 hdr_len += __tcp_hdrlen(th);
3767 struct udphdr _udphdr;
3769 if (skb_header_pointer(skb, skb_transport_offset(skb),
3770 sizeof(_udphdr), &_udphdr))
3771 hdr_len += sizeof(struct udphdr);
3774 if (shinfo->gso_type & SKB_GSO_DODGY)
3775 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3778 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3782 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3783 struct sk_buff **to_free,
3784 struct netdev_queue *txq)
3788 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3789 if (rc == NET_XMIT_SUCCESS)
3790 trace_qdisc_enqueue(q, txq, skb);
3794 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3795 struct net_device *dev,
3796 struct netdev_queue *txq)
3798 spinlock_t *root_lock = qdisc_lock(q);
3799 struct sk_buff *to_free = NULL;
3803 qdisc_calculate_pkt_len(skb, q);
3805 if (q->flags & TCQ_F_NOLOCK) {
3806 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3807 qdisc_run_begin(q)) {
3808 /* Retest nolock_qdisc_is_empty() within the protection
3809 * of q->seqlock to protect from racing with requeuing.
3811 if (unlikely(!nolock_qdisc_is_empty(q))) {
3812 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3819 qdisc_bstats_cpu_update(q, skb);
3820 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3821 !nolock_qdisc_is_empty(q))
3825 return NET_XMIT_SUCCESS;
3828 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3832 if (unlikely(to_free))
3833 kfree_skb_list_reason(to_free,
3834 SKB_DROP_REASON_QDISC_DROP);
3839 * Heuristic to force contended enqueues to serialize on a
3840 * separate lock before trying to get qdisc main lock.
3841 * This permits qdisc->running owner to get the lock more
3842 * often and dequeue packets faster.
3843 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3844 * and then other tasks will only enqueue packets. The packets will be
3845 * sent after the qdisc owner is scheduled again. To prevent this
3846 * scenario the task always serialize on the lock.
3848 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3849 if (unlikely(contended))
3850 spin_lock(&q->busylock);
3852 spin_lock(root_lock);
3853 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3854 __qdisc_drop(skb, &to_free);
3856 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3857 qdisc_run_begin(q)) {
3859 * This is a work-conserving queue; there are no old skbs
3860 * waiting to be sent out; and the qdisc is not running -
3861 * xmit the skb directly.
3864 qdisc_bstats_update(q, skb);
3866 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3867 if (unlikely(contended)) {
3868 spin_unlock(&q->busylock);
3875 rc = NET_XMIT_SUCCESS;
3877 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3878 if (qdisc_run_begin(q)) {
3879 if (unlikely(contended)) {
3880 spin_unlock(&q->busylock);
3887 spin_unlock(root_lock);
3888 if (unlikely(to_free))
3889 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3890 if (unlikely(contended))
3891 spin_unlock(&q->busylock);
3895 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3896 static void skb_update_prio(struct sk_buff *skb)
3898 const struct netprio_map *map;
3899 const struct sock *sk;
3900 unsigned int prioidx;
3904 map = rcu_dereference_bh(skb->dev->priomap);
3907 sk = skb_to_full_sk(skb);
3911 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3913 if (prioidx < map->priomap_len)
3914 skb->priority = map->priomap[prioidx];
3917 #define skb_update_prio(skb)
3921 * dev_loopback_xmit - loop back @skb
3922 * @net: network namespace this loopback is happening in
3923 * @sk: sk needed to be a netfilter okfn
3924 * @skb: buffer to transmit
3926 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3928 skb_reset_mac_header(skb);
3929 __skb_pull(skb, skb_network_offset(skb));
3930 skb->pkt_type = PACKET_LOOPBACK;
3931 if (skb->ip_summed == CHECKSUM_NONE)
3932 skb->ip_summed = CHECKSUM_UNNECESSARY;
3933 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3938 EXPORT_SYMBOL(dev_loopback_xmit);
3940 #ifdef CONFIG_NET_EGRESS
3941 static struct sk_buff *
3942 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3944 #ifdef CONFIG_NET_CLS_ACT
3945 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3946 struct tcf_result cl_res;
3951 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3952 tc_skb_cb(skb)->mru = 0;
3953 tc_skb_cb(skb)->post_ct = false;
3954 mini_qdisc_bstats_cpu_update(miniq, skb);
3956 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3958 case TC_ACT_RECLASSIFY:
3959 skb->tc_index = TC_H_MIN(cl_res.classid);
3962 mini_qdisc_qstats_cpu_drop(miniq);
3963 *ret = NET_XMIT_DROP;
3964 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3969 *ret = NET_XMIT_SUCCESS;
3972 case TC_ACT_REDIRECT:
3973 /* No need to push/pop skb's mac_header here on egress! */
3974 skb_do_redirect(skb);
3975 *ret = NET_XMIT_SUCCESS;
3980 #endif /* CONFIG_NET_CLS_ACT */
3985 static struct netdev_queue *
3986 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3988 int qm = skb_get_queue_mapping(skb);
3990 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3993 static bool netdev_xmit_txqueue_skipped(void)
3995 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3998 void netdev_xmit_skip_txqueue(bool skip)
4000 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
4002 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
4003 #endif /* CONFIG_NET_EGRESS */
4006 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4007 struct xps_dev_maps *dev_maps, unsigned int tci)
4009 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4010 struct xps_map *map;
4011 int queue_index = -1;
4013 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4016 tci *= dev_maps->num_tc;
4019 map = rcu_dereference(dev_maps->attr_map[tci]);
4022 queue_index = map->queues[0];
4024 queue_index = map->queues[reciprocal_scale(
4025 skb_get_hash(skb), map->len)];
4026 if (unlikely(queue_index >= dev->real_num_tx_queues))
4033 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4034 struct sk_buff *skb)
4037 struct xps_dev_maps *dev_maps;
4038 struct sock *sk = skb->sk;
4039 int queue_index = -1;
4041 if (!static_key_false(&xps_needed))
4045 if (!static_key_false(&xps_rxqs_needed))
4048 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4050 int tci = sk_rx_queue_get(sk);
4053 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4058 if (queue_index < 0) {
4059 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4061 unsigned int tci = skb->sender_cpu - 1;
4063 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4075 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4076 struct net_device *sb_dev)
4080 EXPORT_SYMBOL(dev_pick_tx_zero);
4082 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4083 struct net_device *sb_dev)
4085 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4087 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4089 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4090 struct net_device *sb_dev)
4092 struct sock *sk = skb->sk;
4093 int queue_index = sk_tx_queue_get(sk);
4095 sb_dev = sb_dev ? : dev;
4097 if (queue_index < 0 || skb->ooo_okay ||
4098 queue_index >= dev->real_num_tx_queues) {
4099 int new_index = get_xps_queue(dev, sb_dev, skb);
4102 new_index = skb_tx_hash(dev, sb_dev, skb);
4104 if (queue_index != new_index && sk &&
4106 rcu_access_pointer(sk->sk_dst_cache))
4107 sk_tx_queue_set(sk, new_index);
4109 queue_index = new_index;
4114 EXPORT_SYMBOL(netdev_pick_tx);
4116 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4117 struct sk_buff *skb,
4118 struct net_device *sb_dev)
4120 int queue_index = 0;
4123 u32 sender_cpu = skb->sender_cpu - 1;
4125 if (sender_cpu >= (u32)NR_CPUS)
4126 skb->sender_cpu = raw_smp_processor_id() + 1;
4129 if (dev->real_num_tx_queues != 1) {
4130 const struct net_device_ops *ops = dev->netdev_ops;
4132 if (ops->ndo_select_queue)
4133 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4135 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4137 queue_index = netdev_cap_txqueue(dev, queue_index);
4140 skb_set_queue_mapping(skb, queue_index);
4141 return netdev_get_tx_queue(dev, queue_index);
4145 * __dev_queue_xmit() - transmit a buffer
4146 * @skb: buffer to transmit
4147 * @sb_dev: suboordinate device used for L2 forwarding offload
4149 * Queue a buffer for transmission to a network device. The caller must
4150 * have set the device and priority and built the buffer before calling
4151 * this function. The function can be called from an interrupt.
4153 * When calling this method, interrupts MUST be enabled. This is because
4154 * the BH enable code must have IRQs enabled so that it will not deadlock.
4156 * Regardless of the return value, the skb is consumed, so it is currently
4157 * difficult to retry a send to this method. (You can bump the ref count
4158 * before sending to hold a reference for retry if you are careful.)
4161 * * 0 - buffer successfully transmitted
4162 * * positive qdisc return code - NET_XMIT_DROP etc.
4163 * * negative errno - other errors
4165 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4167 struct net_device *dev = skb->dev;
4168 struct netdev_queue *txq = NULL;
4173 skb_reset_mac_header(skb);
4174 skb_assert_len(skb);
4176 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4177 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4179 /* Disable soft irqs for various locks below. Also
4180 * stops preemption for RCU.
4184 skb_update_prio(skb);
4186 qdisc_pkt_len_init(skb);
4187 #ifdef CONFIG_NET_CLS_ACT
4188 skb->tc_at_ingress = 0;
4190 #ifdef CONFIG_NET_EGRESS
4191 if (static_branch_unlikely(&egress_needed_key)) {
4192 if (nf_hook_egress_active()) {
4193 skb = nf_hook_egress(skb, &rc, dev);
4198 netdev_xmit_skip_txqueue(false);
4200 nf_skip_egress(skb, true);
4201 skb = sch_handle_egress(skb, &rc, dev);
4204 nf_skip_egress(skb, false);
4206 if (netdev_xmit_txqueue_skipped())
4207 txq = netdev_tx_queue_mapping(dev, skb);
4210 /* If device/qdisc don't need skb->dst, release it right now while
4211 * its hot in this cpu cache.
4213 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4219 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4221 q = rcu_dereference_bh(txq->qdisc);
4223 trace_net_dev_queue(skb);
4225 rc = __dev_xmit_skb(skb, q, dev, txq);
4229 /* The device has no queue. Common case for software devices:
4230 * loopback, all the sorts of tunnels...
4232 * Really, it is unlikely that netif_tx_lock protection is necessary
4233 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4235 * However, it is possible, that they rely on protection
4238 * Check this and shot the lock. It is not prone from deadlocks.
4239 *Either shot noqueue qdisc, it is even simpler 8)
4241 if (dev->flags & IFF_UP) {
4242 int cpu = smp_processor_id(); /* ok because BHs are off */
4244 /* Other cpus might concurrently change txq->xmit_lock_owner
4245 * to -1 or to their cpu id, but not to our id.
4247 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4248 if (dev_xmit_recursion())
4249 goto recursion_alert;
4251 skb = validate_xmit_skb(skb, dev, &again);
4255 HARD_TX_LOCK(dev, txq, cpu);
4257 if (!netif_xmit_stopped(txq)) {
4258 dev_xmit_recursion_inc();
4259 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4260 dev_xmit_recursion_dec();
4261 if (dev_xmit_complete(rc)) {
4262 HARD_TX_UNLOCK(dev, txq);
4266 HARD_TX_UNLOCK(dev, txq);
4267 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4270 /* Recursion is detected! It is possible,
4274 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4280 rcu_read_unlock_bh();
4282 dev_core_stats_tx_dropped_inc(dev);
4283 kfree_skb_list(skb);
4286 rcu_read_unlock_bh();
4289 EXPORT_SYMBOL(__dev_queue_xmit);
4291 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4293 struct net_device *dev = skb->dev;
4294 struct sk_buff *orig_skb = skb;
4295 struct netdev_queue *txq;
4296 int ret = NETDEV_TX_BUSY;
4299 if (unlikely(!netif_running(dev) ||
4300 !netif_carrier_ok(dev)))
4303 skb = validate_xmit_skb_list(skb, dev, &again);
4304 if (skb != orig_skb)
4307 skb_set_queue_mapping(skb, queue_id);
4308 txq = skb_get_tx_queue(dev, skb);
4312 dev_xmit_recursion_inc();
4313 HARD_TX_LOCK(dev, txq, smp_processor_id());
4314 if (!netif_xmit_frozen_or_drv_stopped(txq))
4315 ret = netdev_start_xmit(skb, dev, txq, false);
4316 HARD_TX_UNLOCK(dev, txq);
4317 dev_xmit_recursion_dec();
4322 dev_core_stats_tx_dropped_inc(dev);
4323 kfree_skb_list(skb);
4324 return NET_XMIT_DROP;
4326 EXPORT_SYMBOL(__dev_direct_xmit);
4328 /*************************************************************************
4330 *************************************************************************/
4332 int netdev_max_backlog __read_mostly = 1000;
4333 EXPORT_SYMBOL(netdev_max_backlog);
4335 int netdev_tstamp_prequeue __read_mostly = 1;
4336 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4337 int netdev_budget __read_mostly = 300;
4338 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4339 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4340 int weight_p __read_mostly = 64; /* old backlog weight */
4341 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4342 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4343 int dev_rx_weight __read_mostly = 64;
4344 int dev_tx_weight __read_mostly = 64;
4346 /* Called with irq disabled */
4347 static inline void ____napi_schedule(struct softnet_data *sd,
4348 struct napi_struct *napi)
4350 struct task_struct *thread;
4352 lockdep_assert_irqs_disabled();
4354 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4355 /* Paired with smp_mb__before_atomic() in
4356 * napi_enable()/dev_set_threaded().
4357 * Use READ_ONCE() to guarantee a complete
4358 * read on napi->thread. Only call
4359 * wake_up_process() when it's not NULL.
4361 thread = READ_ONCE(napi->thread);
4363 /* Avoid doing set_bit() if the thread is in
4364 * INTERRUPTIBLE state, cause napi_thread_wait()
4365 * makes sure to proceed with napi polling
4366 * if the thread is explicitly woken from here.
4368 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4369 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4370 wake_up_process(thread);
4375 list_add_tail(&napi->poll_list, &sd->poll_list);
4376 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4381 /* One global table that all flow-based protocols share. */
4382 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4383 EXPORT_SYMBOL(rps_sock_flow_table);
4384 u32 rps_cpu_mask __read_mostly;
4385 EXPORT_SYMBOL(rps_cpu_mask);
4387 struct static_key_false rps_needed __read_mostly;
4388 EXPORT_SYMBOL(rps_needed);
4389 struct static_key_false rfs_needed __read_mostly;
4390 EXPORT_SYMBOL(rfs_needed);
4392 static struct rps_dev_flow *
4393 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4394 struct rps_dev_flow *rflow, u16 next_cpu)
4396 if (next_cpu < nr_cpu_ids) {
4397 #ifdef CONFIG_RFS_ACCEL
4398 struct netdev_rx_queue *rxqueue;
4399 struct rps_dev_flow_table *flow_table;
4400 struct rps_dev_flow *old_rflow;
4405 /* Should we steer this flow to a different hardware queue? */
4406 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4407 !(dev->features & NETIF_F_NTUPLE))
4409 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4410 if (rxq_index == skb_get_rx_queue(skb))
4413 rxqueue = dev->_rx + rxq_index;
4414 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4417 flow_id = skb_get_hash(skb) & flow_table->mask;
4418 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4419 rxq_index, flow_id);
4423 rflow = &flow_table->flows[flow_id];
4425 if (old_rflow->filter == rflow->filter)
4426 old_rflow->filter = RPS_NO_FILTER;
4430 per_cpu(softnet_data, next_cpu).input_queue_head;
4433 rflow->cpu = next_cpu;
4438 * get_rps_cpu is called from netif_receive_skb and returns the target
4439 * CPU from the RPS map of the receiving queue for a given skb.
4440 * rcu_read_lock must be held on entry.
4442 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4443 struct rps_dev_flow **rflowp)
4445 const struct rps_sock_flow_table *sock_flow_table;
4446 struct netdev_rx_queue *rxqueue = dev->_rx;
4447 struct rps_dev_flow_table *flow_table;
4448 struct rps_map *map;
4453 if (skb_rx_queue_recorded(skb)) {
4454 u16 index = skb_get_rx_queue(skb);
4456 if (unlikely(index >= dev->real_num_rx_queues)) {
4457 WARN_ONCE(dev->real_num_rx_queues > 1,
4458 "%s received packet on queue %u, but number "
4459 "of RX queues is %u\n",
4460 dev->name, index, dev->real_num_rx_queues);
4466 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4468 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4469 map = rcu_dereference(rxqueue->rps_map);
4470 if (!flow_table && !map)
4473 skb_reset_network_header(skb);
4474 hash = skb_get_hash(skb);
4478 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4479 if (flow_table && sock_flow_table) {
4480 struct rps_dev_flow *rflow;
4484 /* First check into global flow table if there is a match */
4485 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4486 if ((ident ^ hash) & ~rps_cpu_mask)
4489 next_cpu = ident & rps_cpu_mask;
4491 /* OK, now we know there is a match,
4492 * we can look at the local (per receive queue) flow table
4494 rflow = &flow_table->flows[hash & flow_table->mask];
4498 * If the desired CPU (where last recvmsg was done) is
4499 * different from current CPU (one in the rx-queue flow
4500 * table entry), switch if one of the following holds:
4501 * - Current CPU is unset (>= nr_cpu_ids).
4502 * - Current CPU is offline.
4503 * - The current CPU's queue tail has advanced beyond the
4504 * last packet that was enqueued using this table entry.
4505 * This guarantees that all previous packets for the flow
4506 * have been dequeued, thus preserving in order delivery.
4508 if (unlikely(tcpu != next_cpu) &&
4509 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4510 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4511 rflow->last_qtail)) >= 0)) {
4513 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4516 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4526 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4527 if (cpu_online(tcpu)) {
4537 #ifdef CONFIG_RFS_ACCEL
4540 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4541 * @dev: Device on which the filter was set
4542 * @rxq_index: RX queue index
4543 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4544 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4546 * Drivers that implement ndo_rx_flow_steer() should periodically call
4547 * this function for each installed filter and remove the filters for
4548 * which it returns %true.
4550 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4551 u32 flow_id, u16 filter_id)
4553 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4554 struct rps_dev_flow_table *flow_table;
4555 struct rps_dev_flow *rflow;
4560 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4561 if (flow_table && flow_id <= flow_table->mask) {
4562 rflow = &flow_table->flows[flow_id];
4563 cpu = READ_ONCE(rflow->cpu);
4564 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4565 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4566 rflow->last_qtail) <
4567 (int)(10 * flow_table->mask)))
4573 EXPORT_SYMBOL(rps_may_expire_flow);
4575 #endif /* CONFIG_RFS_ACCEL */
4577 /* Called from hardirq (IPI) context */
4578 static void rps_trigger_softirq(void *data)
4580 struct softnet_data *sd = data;
4582 ____napi_schedule(sd, &sd->backlog);
4586 #endif /* CONFIG_RPS */
4588 /* Called from hardirq (IPI) context */
4589 static void trigger_rx_softirq(void *data)
4591 struct softnet_data *sd = data;
4593 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4594 smp_store_release(&sd->defer_ipi_scheduled, 0);
4598 * Check if this softnet_data structure is another cpu one
4599 * If yes, queue it to our IPI list and return 1
4602 static int napi_schedule_rps(struct softnet_data *sd)
4604 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4608 sd->rps_ipi_next = mysd->rps_ipi_list;
4609 mysd->rps_ipi_list = sd;
4611 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4614 #endif /* CONFIG_RPS */
4615 __napi_schedule_irqoff(&mysd->backlog);
4619 #ifdef CONFIG_NET_FLOW_LIMIT
4620 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4623 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4625 #ifdef CONFIG_NET_FLOW_LIMIT
4626 struct sd_flow_limit *fl;
4627 struct softnet_data *sd;
4628 unsigned int old_flow, new_flow;
4630 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4633 sd = this_cpu_ptr(&softnet_data);
4636 fl = rcu_dereference(sd->flow_limit);
4638 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4639 old_flow = fl->history[fl->history_head];
4640 fl->history[fl->history_head] = new_flow;
4643 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4645 if (likely(fl->buckets[old_flow]))
4646 fl->buckets[old_flow]--;
4648 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4660 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4661 * queue (may be a remote CPU queue).
4663 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4664 unsigned int *qtail)
4666 enum skb_drop_reason reason;
4667 struct softnet_data *sd;
4668 unsigned long flags;
4671 reason = SKB_DROP_REASON_NOT_SPECIFIED;
4672 sd = &per_cpu(softnet_data, cpu);
4674 rps_lock_irqsave(sd, &flags);
4675 if (!netif_running(skb->dev))
4677 qlen = skb_queue_len(&sd->input_pkt_queue);
4678 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4681 __skb_queue_tail(&sd->input_pkt_queue, skb);
4682 input_queue_tail_incr_save(sd, qtail);
4683 rps_unlock_irq_restore(sd, &flags);
4684 return NET_RX_SUCCESS;
4687 /* Schedule NAPI for backlog device
4688 * We can use non atomic operation since we own the queue lock
4690 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4691 napi_schedule_rps(sd);
4694 reason = SKB_DROP_REASON_CPU_BACKLOG;
4698 rps_unlock_irq_restore(sd, &flags);
4700 dev_core_stats_rx_dropped_inc(skb->dev);
4701 kfree_skb_reason(skb, reason);
4705 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4707 struct net_device *dev = skb->dev;
4708 struct netdev_rx_queue *rxqueue;
4712 if (skb_rx_queue_recorded(skb)) {
4713 u16 index = skb_get_rx_queue(skb);
4715 if (unlikely(index >= dev->real_num_rx_queues)) {
4716 WARN_ONCE(dev->real_num_rx_queues > 1,
4717 "%s received packet on queue %u, but number "
4718 "of RX queues is %u\n",
4719 dev->name, index, dev->real_num_rx_queues);
4721 return rxqueue; /* Return first rxqueue */
4728 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4729 struct bpf_prog *xdp_prog)
4731 void *orig_data, *orig_data_end, *hard_start;
4732 struct netdev_rx_queue *rxqueue;
4733 bool orig_bcast, orig_host;
4734 u32 mac_len, frame_sz;
4735 __be16 orig_eth_type;
4740 /* The XDP program wants to see the packet starting at the MAC
4743 mac_len = skb->data - skb_mac_header(skb);
4744 hard_start = skb->data - skb_headroom(skb);
4746 /* SKB "head" area always have tailroom for skb_shared_info */
4747 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4748 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4750 rxqueue = netif_get_rxqueue(skb);
4751 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4752 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4753 skb_headlen(skb) + mac_len, true);
4755 orig_data_end = xdp->data_end;
4756 orig_data = xdp->data;
4757 eth = (struct ethhdr *)xdp->data;
4758 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4759 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4760 orig_eth_type = eth->h_proto;
4762 act = bpf_prog_run_xdp(xdp_prog, xdp);
4764 /* check if bpf_xdp_adjust_head was used */
4765 off = xdp->data - orig_data;
4768 __skb_pull(skb, off);
4770 __skb_push(skb, -off);
4772 skb->mac_header += off;
4773 skb_reset_network_header(skb);
4776 /* check if bpf_xdp_adjust_tail was used */
4777 off = xdp->data_end - orig_data_end;
4779 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4780 skb->len += off; /* positive on grow, negative on shrink */
4783 /* check if XDP changed eth hdr such SKB needs update */
4784 eth = (struct ethhdr *)xdp->data;
4785 if ((orig_eth_type != eth->h_proto) ||
4786 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4787 skb->dev->dev_addr)) ||
4788 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4789 __skb_push(skb, ETH_HLEN);
4790 skb->pkt_type = PACKET_HOST;
4791 skb->protocol = eth_type_trans(skb, skb->dev);
4794 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4795 * before calling us again on redirect path. We do not call do_redirect
4796 * as we leave that up to the caller.
4798 * Caller is responsible for managing lifetime of skb (i.e. calling
4799 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4804 __skb_push(skb, mac_len);
4807 metalen = xdp->data - xdp->data_meta;
4809 skb_metadata_set(skb, metalen);
4816 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4817 struct xdp_buff *xdp,
4818 struct bpf_prog *xdp_prog)
4822 /* Reinjected packets coming from act_mirred or similar should
4823 * not get XDP generic processing.
4825 if (skb_is_redirected(skb))
4828 /* XDP packets must be linear and must have sufficient headroom
4829 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4830 * native XDP provides, thus we need to do it here as well.
4832 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4833 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4834 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4835 int troom = skb->tail + skb->data_len - skb->end;
4837 /* In case we have to go down the path and also linearize,
4838 * then lets do the pskb_expand_head() work just once here.
4840 if (pskb_expand_head(skb,
4841 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4842 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4844 if (skb_linearize(skb))
4848 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4855 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4858 trace_xdp_exception(skb->dev, xdp_prog, act);
4869 /* When doing generic XDP we have to bypass the qdisc layer and the
4870 * network taps in order to match in-driver-XDP behavior. This also means
4871 * that XDP packets are able to starve other packets going through a qdisc,
4872 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4873 * queues, so they do not have this starvation issue.
4875 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4877 struct net_device *dev = skb->dev;
4878 struct netdev_queue *txq;
4879 bool free_skb = true;
4882 txq = netdev_core_pick_tx(dev, skb, NULL);
4883 cpu = smp_processor_id();
4884 HARD_TX_LOCK(dev, txq, cpu);
4885 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4886 rc = netdev_start_xmit(skb, dev, txq, 0);
4887 if (dev_xmit_complete(rc))
4890 HARD_TX_UNLOCK(dev, txq);
4892 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4893 dev_core_stats_tx_dropped_inc(dev);
4898 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4900 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4903 struct xdp_buff xdp;
4907 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4908 if (act != XDP_PASS) {
4911 err = xdp_do_generic_redirect(skb->dev, skb,
4917 generic_xdp_tx(skb, xdp_prog);
4925 kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4928 EXPORT_SYMBOL_GPL(do_xdp_generic);
4930 static int netif_rx_internal(struct sk_buff *skb)
4934 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4936 trace_netif_rx(skb);
4939 if (static_branch_unlikely(&rps_needed)) {
4940 struct rps_dev_flow voidflow, *rflow = &voidflow;
4945 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4947 cpu = smp_processor_id();
4949 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4957 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4963 * __netif_rx - Slightly optimized version of netif_rx
4964 * @skb: buffer to post
4966 * This behaves as netif_rx except that it does not disable bottom halves.
4967 * As a result this function may only be invoked from the interrupt context
4968 * (either hard or soft interrupt).
4970 int __netif_rx(struct sk_buff *skb)
4974 lockdep_assert_once(hardirq_count() | softirq_count());
4976 trace_netif_rx_entry(skb);
4977 ret = netif_rx_internal(skb);
4978 trace_netif_rx_exit(ret);
4981 EXPORT_SYMBOL(__netif_rx);
4984 * netif_rx - post buffer to the network code
4985 * @skb: buffer to post
4987 * This function receives a packet from a device driver and queues it for
4988 * the upper (protocol) levels to process via the backlog NAPI device. It
4989 * always succeeds. The buffer may be dropped during processing for
4990 * congestion control or by the protocol layers.
4991 * The network buffer is passed via the backlog NAPI device. Modern NIC
4992 * driver should use NAPI and GRO.
4993 * This function can used from interrupt and from process context. The
4994 * caller from process context must not disable interrupts before invoking
4998 * NET_RX_SUCCESS (no congestion)
4999 * NET_RX_DROP (packet was dropped)
5002 int netif_rx(struct sk_buff *skb)
5004 bool need_bh_off = !(hardirq_count() | softirq_count());
5009 trace_netif_rx_entry(skb);
5010 ret = netif_rx_internal(skb);
5011 trace_netif_rx_exit(ret);
5016 EXPORT_SYMBOL(netif_rx);
5018 static __latent_entropy void net_tx_action(struct softirq_action *h)
5020 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5022 if (sd->completion_queue) {
5023 struct sk_buff *clist;
5025 local_irq_disable();
5026 clist = sd->completion_queue;
5027 sd->completion_queue = NULL;
5031 struct sk_buff *skb = clist;
5033 clist = clist->next;
5035 WARN_ON(refcount_read(&skb->users));
5036 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5037 trace_consume_skb(skb);
5039 trace_kfree_skb(skb, net_tx_action,
5040 SKB_DROP_REASON_NOT_SPECIFIED);
5042 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5045 __kfree_skb_defer(skb);
5049 if (sd->output_queue) {
5052 local_irq_disable();
5053 head = sd->output_queue;
5054 sd->output_queue = NULL;
5055 sd->output_queue_tailp = &sd->output_queue;
5061 struct Qdisc *q = head;
5062 spinlock_t *root_lock = NULL;
5064 head = head->next_sched;
5066 /* We need to make sure head->next_sched is read
5067 * before clearing __QDISC_STATE_SCHED
5069 smp_mb__before_atomic();
5071 if (!(q->flags & TCQ_F_NOLOCK)) {
5072 root_lock = qdisc_lock(q);
5073 spin_lock(root_lock);
5074 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5076 /* There is a synchronize_net() between
5077 * STATE_DEACTIVATED flag being set and
5078 * qdisc_reset()/some_qdisc_is_busy() in
5079 * dev_deactivate(), so we can safely bail out
5080 * early here to avoid data race between
5081 * qdisc_deactivate() and some_qdisc_is_busy()
5082 * for lockless qdisc.
5084 clear_bit(__QDISC_STATE_SCHED, &q->state);
5088 clear_bit(__QDISC_STATE_SCHED, &q->state);
5091 spin_unlock(root_lock);
5097 xfrm_dev_backlog(sd);
5100 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5101 /* This hook is defined here for ATM LANE */
5102 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5103 unsigned char *addr) __read_mostly;
5104 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5107 static inline struct sk_buff *
5108 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5109 struct net_device *orig_dev, bool *another)
5111 #ifdef CONFIG_NET_CLS_ACT
5112 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5113 struct tcf_result cl_res;
5115 /* If there's at least one ingress present somewhere (so
5116 * we get here via enabled static key), remaining devices
5117 * that are not configured with an ingress qdisc will bail
5124 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5128 qdisc_skb_cb(skb)->pkt_len = skb->len;
5129 tc_skb_cb(skb)->mru = 0;
5130 tc_skb_cb(skb)->post_ct = false;
5131 skb->tc_at_ingress = 1;
5132 mini_qdisc_bstats_cpu_update(miniq, skb);
5134 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5136 case TC_ACT_RECLASSIFY:
5137 skb->tc_index = TC_H_MIN(cl_res.classid);
5140 mini_qdisc_qstats_cpu_drop(miniq);
5141 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5148 *ret = NET_RX_SUCCESS;
5150 case TC_ACT_REDIRECT:
5151 /* skb_mac_header check was done by cls/act_bpf, so
5152 * we can safely push the L2 header back before
5153 * redirecting to another netdev
5155 __skb_push(skb, skb->mac_len);
5156 if (skb_do_redirect(skb) == -EAGAIN) {
5157 __skb_pull(skb, skb->mac_len);
5161 *ret = NET_RX_SUCCESS;
5163 case TC_ACT_CONSUMED:
5164 *ret = NET_RX_SUCCESS;
5169 #endif /* CONFIG_NET_CLS_ACT */
5174 * netdev_is_rx_handler_busy - check if receive handler is registered
5175 * @dev: device to check
5177 * Check if a receive handler is already registered for a given device.
5178 * Return true if there one.
5180 * The caller must hold the rtnl_mutex.
5182 bool netdev_is_rx_handler_busy(struct net_device *dev)
5185 return dev && rtnl_dereference(dev->rx_handler);
5187 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5190 * netdev_rx_handler_register - register receive handler
5191 * @dev: device to register a handler for
5192 * @rx_handler: receive handler to register
5193 * @rx_handler_data: data pointer that is used by rx handler
5195 * Register a receive handler for a device. This handler will then be
5196 * called from __netif_receive_skb. A negative errno code is returned
5199 * The caller must hold the rtnl_mutex.
5201 * For a general description of rx_handler, see enum rx_handler_result.
5203 int netdev_rx_handler_register(struct net_device *dev,
5204 rx_handler_func_t *rx_handler,
5205 void *rx_handler_data)
5207 if (netdev_is_rx_handler_busy(dev))
5210 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5213 /* Note: rx_handler_data must be set before rx_handler */
5214 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5215 rcu_assign_pointer(dev->rx_handler, rx_handler);
5219 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5222 * netdev_rx_handler_unregister - unregister receive handler
5223 * @dev: device to unregister a handler from
5225 * Unregister a receive handler from a device.
5227 * The caller must hold the rtnl_mutex.
5229 void netdev_rx_handler_unregister(struct net_device *dev)
5233 RCU_INIT_POINTER(dev->rx_handler, NULL);
5234 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5235 * section has a guarantee to see a non NULL rx_handler_data
5239 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5241 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5244 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5245 * the special handling of PFMEMALLOC skbs.
5247 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5249 switch (skb->protocol) {
5250 case htons(ETH_P_ARP):
5251 case htons(ETH_P_IP):
5252 case htons(ETH_P_IPV6):
5253 case htons(ETH_P_8021Q):
5254 case htons(ETH_P_8021AD):
5261 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5262 int *ret, struct net_device *orig_dev)
5264 if (nf_hook_ingress_active(skb)) {
5268 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5273 ingress_retval = nf_hook_ingress(skb);
5275 return ingress_retval;
5280 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5281 struct packet_type **ppt_prev)
5283 struct packet_type *ptype, *pt_prev;
5284 rx_handler_func_t *rx_handler;
5285 struct sk_buff *skb = *pskb;
5286 struct net_device *orig_dev;
5287 bool deliver_exact = false;
5288 int ret = NET_RX_DROP;
5291 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5293 trace_netif_receive_skb(skb);
5295 orig_dev = skb->dev;
5297 skb_reset_network_header(skb);
5298 if (!skb_transport_header_was_set(skb))
5299 skb_reset_transport_header(skb);
5300 skb_reset_mac_len(skb);
5305 skb->skb_iif = skb->dev->ifindex;
5307 __this_cpu_inc(softnet_data.processed);
5309 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5313 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5316 if (ret2 != XDP_PASS) {
5322 if (eth_type_vlan(skb->protocol)) {
5323 skb = skb_vlan_untag(skb);
5328 if (skb_skip_tc_classify(skb))
5334 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5336 ret = deliver_skb(skb, pt_prev, orig_dev);
5340 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5342 ret = deliver_skb(skb, pt_prev, orig_dev);
5347 #ifdef CONFIG_NET_INGRESS
5348 if (static_branch_unlikely(&ingress_needed_key)) {
5349 bool another = false;
5351 nf_skip_egress(skb, true);
5352 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5359 nf_skip_egress(skb, false);
5360 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5364 skb_reset_redirect(skb);
5366 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5369 if (skb_vlan_tag_present(skb)) {
5371 ret = deliver_skb(skb, pt_prev, orig_dev);
5374 if (vlan_do_receive(&skb))
5376 else if (unlikely(!skb))
5380 rx_handler = rcu_dereference(skb->dev->rx_handler);
5383 ret = deliver_skb(skb, pt_prev, orig_dev);
5386 switch (rx_handler(&skb)) {
5387 case RX_HANDLER_CONSUMED:
5388 ret = NET_RX_SUCCESS;
5390 case RX_HANDLER_ANOTHER:
5392 case RX_HANDLER_EXACT:
5393 deliver_exact = true;
5395 case RX_HANDLER_PASS:
5402 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5404 if (skb_vlan_tag_get_id(skb)) {
5405 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5408 skb->pkt_type = PACKET_OTHERHOST;
5409 } else if (eth_type_vlan(skb->protocol)) {
5410 /* Outer header is 802.1P with vlan 0, inner header is
5411 * 802.1Q or 802.1AD and vlan_do_receive() above could
5412 * not find vlan dev for vlan id 0.
5414 __vlan_hwaccel_clear_tag(skb);
5415 skb = skb_vlan_untag(skb);
5418 if (vlan_do_receive(&skb))
5419 /* After stripping off 802.1P header with vlan 0
5420 * vlan dev is found for inner header.
5423 else if (unlikely(!skb))
5426 /* We have stripped outer 802.1P vlan 0 header.
5427 * But could not find vlan dev.
5428 * check again for vlan id to set OTHERHOST.
5432 /* Note: we might in the future use prio bits
5433 * and set skb->priority like in vlan_do_receive()
5434 * For the time being, just ignore Priority Code Point
5436 __vlan_hwaccel_clear_tag(skb);
5439 type = skb->protocol;
5441 /* deliver only exact match when indicated */
5442 if (likely(!deliver_exact)) {
5443 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5444 &ptype_base[ntohs(type) &
5448 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5449 &orig_dev->ptype_specific);
5451 if (unlikely(skb->dev != orig_dev)) {
5452 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5453 &skb->dev->ptype_specific);
5457 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5459 *ppt_prev = pt_prev;
5463 dev_core_stats_rx_dropped_inc(skb->dev);
5465 dev_core_stats_rx_nohandler_inc(skb->dev);
5466 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5467 /* Jamal, now you will not able to escape explaining
5468 * me how you were going to use this. :-)
5474 /* The invariant here is that if *ppt_prev is not NULL
5475 * then skb should also be non-NULL.
5477 * Apparently *ppt_prev assignment above holds this invariant due to
5478 * skb dereferencing near it.
5484 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5486 struct net_device *orig_dev = skb->dev;
5487 struct packet_type *pt_prev = NULL;
5490 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5492 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5493 skb->dev, pt_prev, orig_dev);
5498 * netif_receive_skb_core - special purpose version of netif_receive_skb
5499 * @skb: buffer to process
5501 * More direct receive version of netif_receive_skb(). It should
5502 * only be used by callers that have a need to skip RPS and Generic XDP.
5503 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5505 * This function may only be called from softirq context and interrupts
5506 * should be enabled.
5508 * Return values (usually ignored):
5509 * NET_RX_SUCCESS: no congestion
5510 * NET_RX_DROP: packet was dropped
5512 int netif_receive_skb_core(struct sk_buff *skb)
5517 ret = __netif_receive_skb_one_core(skb, false);
5522 EXPORT_SYMBOL(netif_receive_skb_core);
5524 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5525 struct packet_type *pt_prev,
5526 struct net_device *orig_dev)
5528 struct sk_buff *skb, *next;
5532 if (list_empty(head))
5534 if (pt_prev->list_func != NULL)
5535 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5536 ip_list_rcv, head, pt_prev, orig_dev);
5538 list_for_each_entry_safe(skb, next, head, list) {
5539 skb_list_del_init(skb);
5540 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5544 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5546 /* Fast-path assumptions:
5547 * - There is no RX handler.
5548 * - Only one packet_type matches.
5549 * If either of these fails, we will end up doing some per-packet
5550 * processing in-line, then handling the 'last ptype' for the whole
5551 * sublist. This can't cause out-of-order delivery to any single ptype,
5552 * because the 'last ptype' must be constant across the sublist, and all
5553 * other ptypes are handled per-packet.
5555 /* Current (common) ptype of sublist */
5556 struct packet_type *pt_curr = NULL;
5557 /* Current (common) orig_dev of sublist */
5558 struct net_device *od_curr = NULL;
5559 struct list_head sublist;
5560 struct sk_buff *skb, *next;
5562 INIT_LIST_HEAD(&sublist);
5563 list_for_each_entry_safe(skb, next, head, list) {
5564 struct net_device *orig_dev = skb->dev;
5565 struct packet_type *pt_prev = NULL;
5567 skb_list_del_init(skb);
5568 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5571 if (pt_curr != pt_prev || od_curr != orig_dev) {
5572 /* dispatch old sublist */
5573 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5574 /* start new sublist */
5575 INIT_LIST_HEAD(&sublist);
5579 list_add_tail(&skb->list, &sublist);
5582 /* dispatch final sublist */
5583 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5586 static int __netif_receive_skb(struct sk_buff *skb)
5590 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5591 unsigned int noreclaim_flag;
5594 * PFMEMALLOC skbs are special, they should
5595 * - be delivered to SOCK_MEMALLOC sockets only
5596 * - stay away from userspace
5597 * - have bounded memory usage
5599 * Use PF_MEMALLOC as this saves us from propagating the allocation
5600 * context down to all allocation sites.
5602 noreclaim_flag = memalloc_noreclaim_save();
5603 ret = __netif_receive_skb_one_core(skb, true);
5604 memalloc_noreclaim_restore(noreclaim_flag);
5606 ret = __netif_receive_skb_one_core(skb, false);
5611 static void __netif_receive_skb_list(struct list_head *head)
5613 unsigned long noreclaim_flag = 0;
5614 struct sk_buff *skb, *next;
5615 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5617 list_for_each_entry_safe(skb, next, head, list) {
5618 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5619 struct list_head sublist;
5621 /* Handle the previous sublist */
5622 list_cut_before(&sublist, head, &skb->list);
5623 if (!list_empty(&sublist))
5624 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5625 pfmemalloc = !pfmemalloc;
5626 /* See comments in __netif_receive_skb */
5628 noreclaim_flag = memalloc_noreclaim_save();
5630 memalloc_noreclaim_restore(noreclaim_flag);
5633 /* Handle the remaining sublist */
5634 if (!list_empty(head))
5635 __netif_receive_skb_list_core(head, pfmemalloc);
5636 /* Restore pflags */
5638 memalloc_noreclaim_restore(noreclaim_flag);
5641 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5643 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5644 struct bpf_prog *new = xdp->prog;
5647 switch (xdp->command) {
5648 case XDP_SETUP_PROG:
5649 rcu_assign_pointer(dev->xdp_prog, new);
5654 static_branch_dec(&generic_xdp_needed_key);
5655 } else if (new && !old) {
5656 static_branch_inc(&generic_xdp_needed_key);
5657 dev_disable_lro(dev);
5658 dev_disable_gro_hw(dev);
5670 static int netif_receive_skb_internal(struct sk_buff *skb)
5674 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5676 if (skb_defer_rx_timestamp(skb))
5677 return NET_RX_SUCCESS;
5681 if (static_branch_unlikely(&rps_needed)) {
5682 struct rps_dev_flow voidflow, *rflow = &voidflow;
5683 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5686 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5692 ret = __netif_receive_skb(skb);
5697 void netif_receive_skb_list_internal(struct list_head *head)
5699 struct sk_buff *skb, *next;
5700 struct list_head sublist;
5702 INIT_LIST_HEAD(&sublist);
5703 list_for_each_entry_safe(skb, next, head, list) {
5704 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5705 skb_list_del_init(skb);
5706 if (!skb_defer_rx_timestamp(skb))
5707 list_add_tail(&skb->list, &sublist);
5709 list_splice_init(&sublist, head);
5713 if (static_branch_unlikely(&rps_needed)) {
5714 list_for_each_entry_safe(skb, next, head, list) {
5715 struct rps_dev_flow voidflow, *rflow = &voidflow;
5716 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5719 /* Will be handled, remove from list */
5720 skb_list_del_init(skb);
5721 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5726 __netif_receive_skb_list(head);
5731 * netif_receive_skb - process receive buffer from network
5732 * @skb: buffer to process
5734 * netif_receive_skb() is the main receive data processing function.
5735 * It always succeeds. The buffer may be dropped during processing
5736 * for congestion control or by the protocol layers.
5738 * This function may only be called from softirq context and interrupts
5739 * should be enabled.
5741 * Return values (usually ignored):
5742 * NET_RX_SUCCESS: no congestion
5743 * NET_RX_DROP: packet was dropped
5745 int netif_receive_skb(struct sk_buff *skb)
5749 trace_netif_receive_skb_entry(skb);
5751 ret = netif_receive_skb_internal(skb);
5752 trace_netif_receive_skb_exit(ret);
5756 EXPORT_SYMBOL(netif_receive_skb);
5759 * netif_receive_skb_list - process many receive buffers from network
5760 * @head: list of skbs to process.
5762 * Since return value of netif_receive_skb() is normally ignored, and
5763 * wouldn't be meaningful for a list, this function returns void.
5765 * This function may only be called from softirq context and interrupts
5766 * should be enabled.
5768 void netif_receive_skb_list(struct list_head *head)
5770 struct sk_buff *skb;
5772 if (list_empty(head))
5774 if (trace_netif_receive_skb_list_entry_enabled()) {
5775 list_for_each_entry(skb, head, list)
5776 trace_netif_receive_skb_list_entry(skb);
5778 netif_receive_skb_list_internal(head);
5779 trace_netif_receive_skb_list_exit(0);
5781 EXPORT_SYMBOL(netif_receive_skb_list);
5783 static DEFINE_PER_CPU(struct work_struct, flush_works);
5785 /* Network device is going away, flush any packets still pending */
5786 static void flush_backlog(struct work_struct *work)
5788 struct sk_buff *skb, *tmp;
5789 struct softnet_data *sd;
5792 sd = this_cpu_ptr(&softnet_data);
5794 rps_lock_irq_disable(sd);
5795 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5796 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5797 __skb_unlink(skb, &sd->input_pkt_queue);
5798 dev_kfree_skb_irq(skb);
5799 input_queue_head_incr(sd);
5802 rps_unlock_irq_enable(sd);
5804 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5805 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5806 __skb_unlink(skb, &sd->process_queue);
5808 input_queue_head_incr(sd);
5814 static bool flush_required(int cpu)
5816 #if IS_ENABLED(CONFIG_RPS)
5817 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5820 rps_lock_irq_disable(sd);
5822 /* as insertion into process_queue happens with the rps lock held,
5823 * process_queue access may race only with dequeue
5825 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5826 !skb_queue_empty_lockless(&sd->process_queue);
5827 rps_unlock_irq_enable(sd);
5831 /* without RPS we can't safely check input_pkt_queue: during a
5832 * concurrent remote skb_queue_splice() we can detect as empty both
5833 * input_pkt_queue and process_queue even if the latter could end-up
5834 * containing a lot of packets.
5839 static void flush_all_backlogs(void)
5841 static cpumask_t flush_cpus;
5844 /* since we are under rtnl lock protection we can use static data
5845 * for the cpumask and avoid allocating on stack the possibly
5852 cpumask_clear(&flush_cpus);
5853 for_each_online_cpu(cpu) {
5854 if (flush_required(cpu)) {
5855 queue_work_on(cpu, system_highpri_wq,
5856 per_cpu_ptr(&flush_works, cpu));
5857 cpumask_set_cpu(cpu, &flush_cpus);
5861 /* we can have in flight packet[s] on the cpus we are not flushing,
5862 * synchronize_net() in unregister_netdevice_many() will take care of
5865 for_each_cpu(cpu, &flush_cpus)
5866 flush_work(per_cpu_ptr(&flush_works, cpu));
5871 static void net_rps_send_ipi(struct softnet_data *remsd)
5875 struct softnet_data *next = remsd->rps_ipi_next;
5877 if (cpu_online(remsd->cpu))
5878 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5885 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5886 * Note: called with local irq disabled, but exits with local irq enabled.
5888 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5891 struct softnet_data *remsd = sd->rps_ipi_list;
5894 sd->rps_ipi_list = NULL;
5898 /* Send pending IPI's to kick RPS processing on remote cpus. */
5899 net_rps_send_ipi(remsd);
5905 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5908 return sd->rps_ipi_list != NULL;
5914 static int process_backlog(struct napi_struct *napi, int quota)
5916 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5920 /* Check if we have pending ipi, its better to send them now,
5921 * not waiting net_rx_action() end.
5923 if (sd_has_rps_ipi_waiting(sd)) {
5924 local_irq_disable();
5925 net_rps_action_and_irq_enable(sd);
5928 napi->weight = READ_ONCE(dev_rx_weight);
5930 struct sk_buff *skb;
5932 while ((skb = __skb_dequeue(&sd->process_queue))) {
5934 __netif_receive_skb(skb);
5936 input_queue_head_incr(sd);
5937 if (++work >= quota)
5942 rps_lock_irq_disable(sd);
5943 if (skb_queue_empty(&sd->input_pkt_queue)) {
5945 * Inline a custom version of __napi_complete().
5946 * only current cpu owns and manipulates this napi,
5947 * and NAPI_STATE_SCHED is the only possible flag set
5949 * We can use a plain write instead of clear_bit(),
5950 * and we dont need an smp_mb() memory barrier.
5955 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5956 &sd->process_queue);
5958 rps_unlock_irq_enable(sd);
5965 * __napi_schedule - schedule for receive
5966 * @n: entry to schedule
5968 * The entry's receive function will be scheduled to run.
5969 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5971 void __napi_schedule(struct napi_struct *n)
5973 unsigned long flags;
5975 local_irq_save(flags);
5976 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5977 local_irq_restore(flags);
5979 EXPORT_SYMBOL(__napi_schedule);
5982 * napi_schedule_prep - check if napi can be scheduled
5985 * Test if NAPI routine is already running, and if not mark
5986 * it as running. This is used as a condition variable to
5987 * insure only one NAPI poll instance runs. We also make
5988 * sure there is no pending NAPI disable.
5990 bool napi_schedule_prep(struct napi_struct *n)
5992 unsigned long val, new;
5995 val = READ_ONCE(n->state);
5996 if (unlikely(val & NAPIF_STATE_DISABLE))
5998 new = val | NAPIF_STATE_SCHED;
6000 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6001 * This was suggested by Alexander Duyck, as compiler
6002 * emits better code than :
6003 * if (val & NAPIF_STATE_SCHED)
6004 * new |= NAPIF_STATE_MISSED;
6006 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6008 } while (cmpxchg(&n->state, val, new) != val);
6010 return !(val & NAPIF_STATE_SCHED);
6012 EXPORT_SYMBOL(napi_schedule_prep);
6015 * __napi_schedule_irqoff - schedule for receive
6016 * @n: entry to schedule
6018 * Variant of __napi_schedule() assuming hard irqs are masked.
6020 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6021 * because the interrupt disabled assumption might not be true
6022 * due to force-threaded interrupts and spinlock substitution.
6024 void __napi_schedule_irqoff(struct napi_struct *n)
6026 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6027 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6031 EXPORT_SYMBOL(__napi_schedule_irqoff);
6033 bool napi_complete_done(struct napi_struct *n, int work_done)
6035 unsigned long flags, val, new, timeout = 0;
6039 * 1) Don't let napi dequeue from the cpu poll list
6040 * just in case its running on a different cpu.
6041 * 2) If we are busy polling, do nothing here, we have
6042 * the guarantee we will be called later.
6044 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6045 NAPIF_STATE_IN_BUSY_POLL)))
6050 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6051 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6053 if (n->defer_hard_irqs_count > 0) {
6054 n->defer_hard_irqs_count--;
6055 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6059 if (n->gro_bitmask) {
6060 /* When the NAPI instance uses a timeout and keeps postponing
6061 * it, we need to bound somehow the time packets are kept in
6064 napi_gro_flush(n, !!timeout);
6069 if (unlikely(!list_empty(&n->poll_list))) {
6070 /* If n->poll_list is not empty, we need to mask irqs */
6071 local_irq_save(flags);
6072 list_del_init(&n->poll_list);
6073 local_irq_restore(flags);
6077 val = READ_ONCE(n->state);
6079 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6081 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6082 NAPIF_STATE_SCHED_THREADED |
6083 NAPIF_STATE_PREFER_BUSY_POLL);
6085 /* If STATE_MISSED was set, leave STATE_SCHED set,
6086 * because we will call napi->poll() one more time.
6087 * This C code was suggested by Alexander Duyck to help gcc.
6089 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6091 } while (cmpxchg(&n->state, val, new) != val);
6093 if (unlikely(val & NAPIF_STATE_MISSED)) {
6099 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6100 HRTIMER_MODE_REL_PINNED);
6103 EXPORT_SYMBOL(napi_complete_done);
6105 /* must be called under rcu_read_lock(), as we dont take a reference */
6106 static struct napi_struct *napi_by_id(unsigned int napi_id)
6108 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6109 struct napi_struct *napi;
6111 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6112 if (napi->napi_id == napi_id)
6118 #if defined(CONFIG_NET_RX_BUSY_POLL)
6120 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6122 if (!skip_schedule) {
6123 gro_normal_list(napi);
6124 __napi_schedule(napi);
6128 if (napi->gro_bitmask) {
6129 /* flush too old packets
6130 * If HZ < 1000, flush all packets.
6132 napi_gro_flush(napi, HZ >= 1000);
6135 gro_normal_list(napi);
6136 clear_bit(NAPI_STATE_SCHED, &napi->state);
6139 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6142 bool skip_schedule = false;
6143 unsigned long timeout;
6146 /* Busy polling means there is a high chance device driver hard irq
6147 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6148 * set in napi_schedule_prep().
6149 * Since we are about to call napi->poll() once more, we can safely
6150 * clear NAPI_STATE_MISSED.
6152 * Note: x86 could use a single "lock and ..." instruction
6153 * to perform these two clear_bit()
6155 clear_bit(NAPI_STATE_MISSED, &napi->state);
6156 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6160 if (prefer_busy_poll) {
6161 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6162 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6163 if (napi->defer_hard_irqs_count && timeout) {
6164 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6165 skip_schedule = true;
6169 /* All we really want here is to re-enable device interrupts.
6170 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6172 rc = napi->poll(napi, budget);
6173 /* We can't gro_normal_list() here, because napi->poll() might have
6174 * rearmed the napi (napi_complete_done()) in which case it could
6175 * already be running on another CPU.
6177 trace_napi_poll(napi, rc, budget);
6178 netpoll_poll_unlock(have_poll_lock);
6180 __busy_poll_stop(napi, skip_schedule);
6184 void napi_busy_loop(unsigned int napi_id,
6185 bool (*loop_end)(void *, unsigned long),
6186 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6188 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6189 int (*napi_poll)(struct napi_struct *napi, int budget);
6190 void *have_poll_lock = NULL;
6191 struct napi_struct *napi;
6198 napi = napi_by_id(napi_id);
6208 unsigned long val = READ_ONCE(napi->state);
6210 /* If multiple threads are competing for this napi,
6211 * we avoid dirtying napi->state as much as we can.
6213 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6214 NAPIF_STATE_IN_BUSY_POLL)) {
6215 if (prefer_busy_poll)
6216 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6219 if (cmpxchg(&napi->state, val,
6220 val | NAPIF_STATE_IN_BUSY_POLL |
6221 NAPIF_STATE_SCHED) != val) {
6222 if (prefer_busy_poll)
6223 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6226 have_poll_lock = netpoll_poll_lock(napi);
6227 napi_poll = napi->poll;
6229 work = napi_poll(napi, budget);
6230 trace_napi_poll(napi, work, budget);
6231 gro_normal_list(napi);
6234 __NET_ADD_STATS(dev_net(napi->dev),
6235 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6238 if (!loop_end || loop_end(loop_end_arg, start_time))
6241 if (unlikely(need_resched())) {
6243 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6247 if (loop_end(loop_end_arg, start_time))
6254 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6259 EXPORT_SYMBOL(napi_busy_loop);
6261 #endif /* CONFIG_NET_RX_BUSY_POLL */
6263 static void napi_hash_add(struct napi_struct *napi)
6265 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6268 spin_lock(&napi_hash_lock);
6270 /* 0..NR_CPUS range is reserved for sender_cpu use */
6272 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6273 napi_gen_id = MIN_NAPI_ID;
6274 } while (napi_by_id(napi_gen_id));
6275 napi->napi_id = napi_gen_id;
6277 hlist_add_head_rcu(&napi->napi_hash_node,
6278 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6280 spin_unlock(&napi_hash_lock);
6283 /* Warning : caller is responsible to make sure rcu grace period
6284 * is respected before freeing memory containing @napi
6286 static void napi_hash_del(struct napi_struct *napi)
6288 spin_lock(&napi_hash_lock);
6290 hlist_del_init_rcu(&napi->napi_hash_node);
6292 spin_unlock(&napi_hash_lock);
6295 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6297 struct napi_struct *napi;
6299 napi = container_of(timer, struct napi_struct, timer);
6301 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6302 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6304 if (!napi_disable_pending(napi) &&
6305 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6306 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6307 __napi_schedule_irqoff(napi);
6310 return HRTIMER_NORESTART;
6313 static void init_gro_hash(struct napi_struct *napi)
6317 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6318 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6319 napi->gro_hash[i].count = 0;
6321 napi->gro_bitmask = 0;
6324 int dev_set_threaded(struct net_device *dev, bool threaded)
6326 struct napi_struct *napi;
6329 if (dev->threaded == threaded)
6333 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6334 if (!napi->thread) {
6335 err = napi_kthread_create(napi);
6344 dev->threaded = threaded;
6346 /* Make sure kthread is created before THREADED bit
6349 smp_mb__before_atomic();
6351 /* Setting/unsetting threaded mode on a napi might not immediately
6352 * take effect, if the current napi instance is actively being
6353 * polled. In this case, the switch between threaded mode and
6354 * softirq mode will happen in the next round of napi_schedule().
6355 * This should not cause hiccups/stalls to the live traffic.
6357 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6359 set_bit(NAPI_STATE_THREADED, &napi->state);
6361 clear_bit(NAPI_STATE_THREADED, &napi->state);
6366 EXPORT_SYMBOL(dev_set_threaded);
6368 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6369 int (*poll)(struct napi_struct *, int), int weight)
6371 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6374 INIT_LIST_HEAD(&napi->poll_list);
6375 INIT_HLIST_NODE(&napi->napi_hash_node);
6376 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6377 napi->timer.function = napi_watchdog;
6378 init_gro_hash(napi);
6380 INIT_LIST_HEAD(&napi->rx_list);
6383 if (weight > NAPI_POLL_WEIGHT)
6384 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6386 napi->weight = weight;
6388 #ifdef CONFIG_NETPOLL
6389 napi->poll_owner = -1;
6391 set_bit(NAPI_STATE_SCHED, &napi->state);
6392 set_bit(NAPI_STATE_NPSVC, &napi->state);
6393 list_add_rcu(&napi->dev_list, &dev->napi_list);
6394 napi_hash_add(napi);
6395 napi_get_frags_check(napi);
6396 /* Create kthread for this napi if dev->threaded is set.
6397 * Clear dev->threaded if kthread creation failed so that
6398 * threaded mode will not be enabled in napi_enable().
6400 if (dev->threaded && napi_kthread_create(napi))
6403 EXPORT_SYMBOL(netif_napi_add_weight);
6405 void napi_disable(struct napi_struct *n)
6407 unsigned long val, new;
6410 set_bit(NAPI_STATE_DISABLE, &n->state);
6413 val = READ_ONCE(n->state);
6414 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6415 usleep_range(20, 200);
6419 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6420 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6422 if (cmpxchg(&n->state, val, new) == val)
6426 hrtimer_cancel(&n->timer);
6428 clear_bit(NAPI_STATE_DISABLE, &n->state);
6430 EXPORT_SYMBOL(napi_disable);
6433 * napi_enable - enable NAPI scheduling
6436 * Resume NAPI from being scheduled on this context.
6437 * Must be paired with napi_disable.
6439 void napi_enable(struct napi_struct *n)
6441 unsigned long val, new;
6444 val = READ_ONCE(n->state);
6445 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6447 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6448 if (n->dev->threaded && n->thread)
6449 new |= NAPIF_STATE_THREADED;
6450 } while (cmpxchg(&n->state, val, new) != val);
6452 EXPORT_SYMBOL(napi_enable);
6454 static void flush_gro_hash(struct napi_struct *napi)
6458 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6459 struct sk_buff *skb, *n;
6461 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6463 napi->gro_hash[i].count = 0;
6467 /* Must be called in process context */
6468 void __netif_napi_del(struct napi_struct *napi)
6470 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6473 napi_hash_del(napi);
6474 list_del_rcu(&napi->dev_list);
6475 napi_free_frags(napi);
6477 flush_gro_hash(napi);
6478 napi->gro_bitmask = 0;
6481 kthread_stop(napi->thread);
6482 napi->thread = NULL;
6485 EXPORT_SYMBOL(__netif_napi_del);
6487 static int __napi_poll(struct napi_struct *n, bool *repoll)
6493 /* This NAPI_STATE_SCHED test is for avoiding a race
6494 * with netpoll's poll_napi(). Only the entity which
6495 * obtains the lock and sees NAPI_STATE_SCHED set will
6496 * actually make the ->poll() call. Therefore we avoid
6497 * accidentally calling ->poll() when NAPI is not scheduled.
6500 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6501 work = n->poll(n, weight);
6502 trace_napi_poll(n, work, weight);
6505 if (unlikely(work > weight))
6506 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6507 n->poll, work, weight);
6509 if (likely(work < weight))
6512 /* Drivers must not modify the NAPI state if they
6513 * consume the entire weight. In such cases this code
6514 * still "owns" the NAPI instance and therefore can
6515 * move the instance around on the list at-will.
6517 if (unlikely(napi_disable_pending(n))) {
6522 /* The NAPI context has more processing work, but busy-polling
6523 * is preferred. Exit early.
6525 if (napi_prefer_busy_poll(n)) {
6526 if (napi_complete_done(n, work)) {
6527 /* If timeout is not set, we need to make sure
6528 * that the NAPI is re-scheduled.
6535 if (n->gro_bitmask) {
6536 /* flush too old packets
6537 * If HZ < 1000, flush all packets.
6539 napi_gro_flush(n, HZ >= 1000);
6544 /* Some drivers may have called napi_schedule
6545 * prior to exhausting their budget.
6547 if (unlikely(!list_empty(&n->poll_list))) {
6548 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6549 n->dev ? n->dev->name : "backlog");
6558 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6560 bool do_repoll = false;
6564 list_del_init(&n->poll_list);
6566 have = netpoll_poll_lock(n);
6568 work = __napi_poll(n, &do_repoll);
6571 list_add_tail(&n->poll_list, repoll);
6573 netpoll_poll_unlock(have);
6578 static int napi_thread_wait(struct napi_struct *napi)
6582 set_current_state(TASK_INTERRUPTIBLE);
6584 while (!kthread_should_stop()) {
6585 /* Testing SCHED_THREADED bit here to make sure the current
6586 * kthread owns this napi and could poll on this napi.
6587 * Testing SCHED bit is not enough because SCHED bit might be
6588 * set by some other busy poll thread or by napi_disable().
6590 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6591 WARN_ON(!list_empty(&napi->poll_list));
6592 __set_current_state(TASK_RUNNING);
6597 /* woken being true indicates this thread owns this napi. */
6599 set_current_state(TASK_INTERRUPTIBLE);
6601 __set_current_state(TASK_RUNNING);
6606 static int napi_threaded_poll(void *data)
6608 struct napi_struct *napi = data;
6611 while (!napi_thread_wait(napi)) {
6613 bool repoll = false;
6617 have = netpoll_poll_lock(napi);
6618 __napi_poll(napi, &repoll);
6619 netpoll_poll_unlock(have);
6632 static void skb_defer_free_flush(struct softnet_data *sd)
6634 struct sk_buff *skb, *next;
6635 unsigned long flags;
6637 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6638 if (!READ_ONCE(sd->defer_list))
6641 spin_lock_irqsave(&sd->defer_lock, flags);
6642 skb = sd->defer_list;
6643 sd->defer_list = NULL;
6644 sd->defer_count = 0;
6645 spin_unlock_irqrestore(&sd->defer_lock, flags);
6647 while (skb != NULL) {
6649 napi_consume_skb(skb, 1);
6654 static __latent_entropy void net_rx_action(struct softirq_action *h)
6656 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6657 unsigned long time_limit = jiffies +
6658 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6659 int budget = READ_ONCE(netdev_budget);
6663 local_irq_disable();
6664 list_splice_init(&sd->poll_list, &list);
6668 struct napi_struct *n;
6670 skb_defer_free_flush(sd);
6672 if (list_empty(&list)) {
6673 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6678 n = list_first_entry(&list, struct napi_struct, poll_list);
6679 budget -= napi_poll(n, &repoll);
6681 /* If softirq window is exhausted then punt.
6682 * Allow this to run for 2 jiffies since which will allow
6683 * an average latency of 1.5/HZ.
6685 if (unlikely(budget <= 0 ||
6686 time_after_eq(jiffies, time_limit))) {
6692 local_irq_disable();
6694 list_splice_tail_init(&sd->poll_list, &list);
6695 list_splice_tail(&repoll, &list);
6696 list_splice(&list, &sd->poll_list);
6697 if (!list_empty(&sd->poll_list))
6698 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6700 net_rps_action_and_irq_enable(sd);
6704 struct netdev_adjacent {
6705 struct net_device *dev;
6706 netdevice_tracker dev_tracker;
6708 /* upper master flag, there can only be one master device per list */
6711 /* lookup ignore flag */
6714 /* counter for the number of times this device was added to us */
6717 /* private field for the users */
6720 struct list_head list;
6721 struct rcu_head rcu;
6724 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6725 struct list_head *adj_list)
6727 struct netdev_adjacent *adj;
6729 list_for_each_entry(adj, adj_list, list) {
6730 if (adj->dev == adj_dev)
6736 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6737 struct netdev_nested_priv *priv)
6739 struct net_device *dev = (struct net_device *)priv->data;
6741 return upper_dev == dev;
6745 * netdev_has_upper_dev - Check if device is linked to an upper device
6747 * @upper_dev: upper device to check
6749 * Find out if a device is linked to specified upper device and return true
6750 * in case it is. Note that this checks only immediate upper device,
6751 * not through a complete stack of devices. The caller must hold the RTNL lock.
6753 bool netdev_has_upper_dev(struct net_device *dev,
6754 struct net_device *upper_dev)
6756 struct netdev_nested_priv priv = {
6757 .data = (void *)upper_dev,
6762 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6765 EXPORT_SYMBOL(netdev_has_upper_dev);
6768 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6770 * @upper_dev: upper device to check
6772 * Find out if a device is linked to specified upper device and return true
6773 * in case it is. Note that this checks the entire upper device chain.
6774 * The caller must hold rcu lock.
6777 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6778 struct net_device *upper_dev)
6780 struct netdev_nested_priv priv = {
6781 .data = (void *)upper_dev,
6784 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6787 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6790 * netdev_has_any_upper_dev - Check if device is linked to some device
6793 * Find out if a device is linked to an upper device and return true in case
6794 * it is. The caller must hold the RTNL lock.
6796 bool netdev_has_any_upper_dev(struct net_device *dev)
6800 return !list_empty(&dev->adj_list.upper);
6802 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6805 * netdev_master_upper_dev_get - Get master upper device
6808 * Find a master upper device and return pointer to it or NULL in case
6809 * it's not there. The caller must hold the RTNL lock.
6811 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6813 struct netdev_adjacent *upper;
6817 if (list_empty(&dev->adj_list.upper))
6820 upper = list_first_entry(&dev->adj_list.upper,
6821 struct netdev_adjacent, list);
6822 if (likely(upper->master))
6826 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6828 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6830 struct netdev_adjacent *upper;
6834 if (list_empty(&dev->adj_list.upper))
6837 upper = list_first_entry(&dev->adj_list.upper,
6838 struct netdev_adjacent, list);
6839 if (likely(upper->master) && !upper->ignore)
6845 * netdev_has_any_lower_dev - Check if device is linked to some device
6848 * Find out if a device is linked to a lower device and return true in case
6849 * it is. The caller must hold the RTNL lock.
6851 static bool netdev_has_any_lower_dev(struct net_device *dev)
6855 return !list_empty(&dev->adj_list.lower);
6858 void *netdev_adjacent_get_private(struct list_head *adj_list)
6860 struct netdev_adjacent *adj;
6862 adj = list_entry(adj_list, struct netdev_adjacent, list);
6864 return adj->private;
6866 EXPORT_SYMBOL(netdev_adjacent_get_private);
6869 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6871 * @iter: list_head ** of the current position
6873 * Gets the next device from the dev's upper list, starting from iter
6874 * position. The caller must hold RCU read lock.
6876 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6877 struct list_head **iter)
6879 struct netdev_adjacent *upper;
6881 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6883 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6885 if (&upper->list == &dev->adj_list.upper)
6888 *iter = &upper->list;
6892 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6894 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6895 struct list_head **iter,
6898 struct netdev_adjacent *upper;
6900 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6902 if (&upper->list == &dev->adj_list.upper)
6905 *iter = &upper->list;
6906 *ignore = upper->ignore;
6911 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6912 struct list_head **iter)
6914 struct netdev_adjacent *upper;
6916 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6918 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6920 if (&upper->list == &dev->adj_list.upper)
6923 *iter = &upper->list;
6928 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6929 int (*fn)(struct net_device *dev,
6930 struct netdev_nested_priv *priv),
6931 struct netdev_nested_priv *priv)
6933 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6934 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6939 iter = &dev->adj_list.upper;
6943 ret = fn(now, priv);
6950 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6957 niter = &udev->adj_list.upper;
6958 dev_stack[cur] = now;
6959 iter_stack[cur++] = iter;
6966 next = dev_stack[--cur];
6967 niter = iter_stack[cur];
6977 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6978 int (*fn)(struct net_device *dev,
6979 struct netdev_nested_priv *priv),
6980 struct netdev_nested_priv *priv)
6982 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6983 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6987 iter = &dev->adj_list.upper;
6991 ret = fn(now, priv);
6998 udev = netdev_next_upper_dev_rcu(now, &iter);
7003 niter = &udev->adj_list.upper;
7004 dev_stack[cur] = now;
7005 iter_stack[cur++] = iter;
7012 next = dev_stack[--cur];
7013 niter = iter_stack[cur];
7022 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7024 static bool __netdev_has_upper_dev(struct net_device *dev,
7025 struct net_device *upper_dev)
7027 struct netdev_nested_priv priv = {
7029 .data = (void *)upper_dev,
7034 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7039 * netdev_lower_get_next_private - Get the next ->private from the
7040 * lower neighbour list
7042 * @iter: list_head ** of the current position
7044 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7045 * list, starting from iter position. The caller must hold either hold the
7046 * RTNL lock or its own locking that guarantees that the neighbour lower
7047 * list will remain unchanged.
7049 void *netdev_lower_get_next_private(struct net_device *dev,
7050 struct list_head **iter)
7052 struct netdev_adjacent *lower;
7054 lower = list_entry(*iter, struct netdev_adjacent, list);
7056 if (&lower->list == &dev->adj_list.lower)
7059 *iter = lower->list.next;
7061 return lower->private;
7063 EXPORT_SYMBOL(netdev_lower_get_next_private);
7066 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7067 * lower neighbour list, RCU
7070 * @iter: list_head ** of the current position
7072 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7073 * list, starting from iter position. The caller must hold RCU read lock.
7075 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7076 struct list_head **iter)
7078 struct netdev_adjacent *lower;
7080 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7082 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7084 if (&lower->list == &dev->adj_list.lower)
7087 *iter = &lower->list;
7089 return lower->private;
7091 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7094 * netdev_lower_get_next - Get the next device from the lower neighbour
7097 * @iter: list_head ** of the current position
7099 * Gets the next netdev_adjacent from the dev's lower neighbour
7100 * list, starting from iter position. The caller must hold RTNL lock or
7101 * its own locking that guarantees that the neighbour lower
7102 * list will remain unchanged.
7104 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7106 struct netdev_adjacent *lower;
7108 lower = list_entry(*iter, struct netdev_adjacent, list);
7110 if (&lower->list == &dev->adj_list.lower)
7113 *iter = lower->list.next;
7117 EXPORT_SYMBOL(netdev_lower_get_next);
7119 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7120 struct list_head **iter)
7122 struct netdev_adjacent *lower;
7124 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7126 if (&lower->list == &dev->adj_list.lower)
7129 *iter = &lower->list;
7134 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7135 struct list_head **iter,
7138 struct netdev_adjacent *lower;
7140 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7142 if (&lower->list == &dev->adj_list.lower)
7145 *iter = &lower->list;
7146 *ignore = lower->ignore;
7151 int netdev_walk_all_lower_dev(struct net_device *dev,
7152 int (*fn)(struct net_device *dev,
7153 struct netdev_nested_priv *priv),
7154 struct netdev_nested_priv *priv)
7156 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7157 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7161 iter = &dev->adj_list.lower;
7165 ret = fn(now, priv);
7172 ldev = netdev_next_lower_dev(now, &iter);
7177 niter = &ldev->adj_list.lower;
7178 dev_stack[cur] = now;
7179 iter_stack[cur++] = iter;
7186 next = dev_stack[--cur];
7187 niter = iter_stack[cur];
7196 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7198 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7199 int (*fn)(struct net_device *dev,
7200 struct netdev_nested_priv *priv),
7201 struct netdev_nested_priv *priv)
7203 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7204 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7209 iter = &dev->adj_list.lower;
7213 ret = fn(now, priv);
7220 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7227 niter = &ldev->adj_list.lower;
7228 dev_stack[cur] = now;
7229 iter_stack[cur++] = iter;
7236 next = dev_stack[--cur];
7237 niter = iter_stack[cur];
7247 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7248 struct list_head **iter)
7250 struct netdev_adjacent *lower;
7252 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7253 if (&lower->list == &dev->adj_list.lower)
7256 *iter = &lower->list;
7260 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7262 static u8 __netdev_upper_depth(struct net_device *dev)
7264 struct net_device *udev;
7265 struct list_head *iter;
7269 for (iter = &dev->adj_list.upper,
7270 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7272 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7275 if (max_depth < udev->upper_level)
7276 max_depth = udev->upper_level;
7282 static u8 __netdev_lower_depth(struct net_device *dev)
7284 struct net_device *ldev;
7285 struct list_head *iter;
7289 for (iter = &dev->adj_list.lower,
7290 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7292 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7295 if (max_depth < ldev->lower_level)
7296 max_depth = ldev->lower_level;
7302 static int __netdev_update_upper_level(struct net_device *dev,
7303 struct netdev_nested_priv *__unused)
7305 dev->upper_level = __netdev_upper_depth(dev) + 1;
7309 #ifdef CONFIG_LOCKDEP
7310 static LIST_HEAD(net_unlink_list);
7312 static void net_unlink_todo(struct net_device *dev)
7314 if (list_empty(&dev->unlink_list))
7315 list_add_tail(&dev->unlink_list, &net_unlink_list);
7319 static int __netdev_update_lower_level(struct net_device *dev,
7320 struct netdev_nested_priv *priv)
7322 dev->lower_level = __netdev_lower_depth(dev) + 1;
7324 #ifdef CONFIG_LOCKDEP
7328 if (priv->flags & NESTED_SYNC_IMM)
7329 dev->nested_level = dev->lower_level - 1;
7330 if (priv->flags & NESTED_SYNC_TODO)
7331 net_unlink_todo(dev);
7336 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7337 int (*fn)(struct net_device *dev,
7338 struct netdev_nested_priv *priv),
7339 struct netdev_nested_priv *priv)
7341 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7342 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7346 iter = &dev->adj_list.lower;
7350 ret = fn(now, priv);
7357 ldev = netdev_next_lower_dev_rcu(now, &iter);
7362 niter = &ldev->adj_list.lower;
7363 dev_stack[cur] = now;
7364 iter_stack[cur++] = iter;
7371 next = dev_stack[--cur];
7372 niter = iter_stack[cur];
7381 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7384 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7385 * lower neighbour list, RCU
7389 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7390 * list. The caller must hold RCU read lock.
7392 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7394 struct netdev_adjacent *lower;
7396 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7397 struct netdev_adjacent, list);
7399 return lower->private;
7402 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7405 * netdev_master_upper_dev_get_rcu - Get master upper device
7408 * Find a master upper device and return pointer to it or NULL in case
7409 * it's not there. The caller must hold the RCU read lock.
7411 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7413 struct netdev_adjacent *upper;
7415 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7416 struct netdev_adjacent, list);
7417 if (upper && likely(upper->master))
7421 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7423 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7424 struct net_device *adj_dev,
7425 struct list_head *dev_list)
7427 char linkname[IFNAMSIZ+7];
7429 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7430 "upper_%s" : "lower_%s", adj_dev->name);
7431 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7434 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7436 struct list_head *dev_list)
7438 char linkname[IFNAMSIZ+7];
7440 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7441 "upper_%s" : "lower_%s", name);
7442 sysfs_remove_link(&(dev->dev.kobj), linkname);
7445 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7446 struct net_device *adj_dev,
7447 struct list_head *dev_list)
7449 return (dev_list == &dev->adj_list.upper ||
7450 dev_list == &dev->adj_list.lower) &&
7451 net_eq(dev_net(dev), dev_net(adj_dev));
7454 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7455 struct net_device *adj_dev,
7456 struct list_head *dev_list,
7457 void *private, bool master)
7459 struct netdev_adjacent *adj;
7462 adj = __netdev_find_adj(adj_dev, dev_list);
7466 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7467 dev->name, adj_dev->name, adj->ref_nr);
7472 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7477 adj->master = master;
7479 adj->private = private;
7480 adj->ignore = false;
7481 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7483 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7484 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7486 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7487 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7492 /* Ensure that master link is always the first item in list. */
7494 ret = sysfs_create_link(&(dev->dev.kobj),
7495 &(adj_dev->dev.kobj), "master");
7497 goto remove_symlinks;
7499 list_add_rcu(&adj->list, dev_list);
7501 list_add_tail_rcu(&adj->list, dev_list);
7507 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7508 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7510 netdev_put(adj_dev, &adj->dev_tracker);
7516 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7517 struct net_device *adj_dev,
7519 struct list_head *dev_list)
7521 struct netdev_adjacent *adj;
7523 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7524 dev->name, adj_dev->name, ref_nr);
7526 adj = __netdev_find_adj(adj_dev, dev_list);
7529 pr_err("Adjacency does not exist for device %s from %s\n",
7530 dev->name, adj_dev->name);
7535 if (adj->ref_nr > ref_nr) {
7536 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7537 dev->name, adj_dev->name, ref_nr,
7538 adj->ref_nr - ref_nr);
7539 adj->ref_nr -= ref_nr;
7544 sysfs_remove_link(&(dev->dev.kobj), "master");
7546 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7547 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7549 list_del_rcu(&adj->list);
7550 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7551 adj_dev->name, dev->name, adj_dev->name);
7552 netdev_put(adj_dev, &adj->dev_tracker);
7553 kfree_rcu(adj, rcu);
7556 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7557 struct net_device *upper_dev,
7558 struct list_head *up_list,
7559 struct list_head *down_list,
7560 void *private, bool master)
7564 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7569 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7572 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7579 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7580 struct net_device *upper_dev,
7582 struct list_head *up_list,
7583 struct list_head *down_list)
7585 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7586 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7589 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7590 struct net_device *upper_dev,
7591 void *private, bool master)
7593 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7594 &dev->adj_list.upper,
7595 &upper_dev->adj_list.lower,
7599 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7600 struct net_device *upper_dev)
7602 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7603 &dev->adj_list.upper,
7604 &upper_dev->adj_list.lower);
7607 static int __netdev_upper_dev_link(struct net_device *dev,
7608 struct net_device *upper_dev, bool master,
7609 void *upper_priv, void *upper_info,
7610 struct netdev_nested_priv *priv,
7611 struct netlink_ext_ack *extack)
7613 struct netdev_notifier_changeupper_info changeupper_info = {
7618 .upper_dev = upper_dev,
7621 .upper_info = upper_info,
7623 struct net_device *master_dev;
7628 if (dev == upper_dev)
7631 /* To prevent loops, check if dev is not upper device to upper_dev. */
7632 if (__netdev_has_upper_dev(upper_dev, dev))
7635 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7639 if (__netdev_has_upper_dev(dev, upper_dev))
7642 master_dev = __netdev_master_upper_dev_get(dev);
7644 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7647 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7648 &changeupper_info.info);
7649 ret = notifier_to_errno(ret);
7653 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7658 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7659 &changeupper_info.info);
7660 ret = notifier_to_errno(ret);
7664 __netdev_update_upper_level(dev, NULL);
7665 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7667 __netdev_update_lower_level(upper_dev, priv);
7668 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7674 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7680 * netdev_upper_dev_link - Add a link to the upper device
7682 * @upper_dev: new upper device
7683 * @extack: netlink extended ack
7685 * Adds a link to device which is upper to this one. The caller must hold
7686 * the RTNL lock. On a failure a negative errno code is returned.
7687 * On success the reference counts are adjusted and the function
7690 int netdev_upper_dev_link(struct net_device *dev,
7691 struct net_device *upper_dev,
7692 struct netlink_ext_ack *extack)
7694 struct netdev_nested_priv priv = {
7695 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7699 return __netdev_upper_dev_link(dev, upper_dev, false,
7700 NULL, NULL, &priv, extack);
7702 EXPORT_SYMBOL(netdev_upper_dev_link);
7705 * netdev_master_upper_dev_link - Add a master link to the upper device
7707 * @upper_dev: new upper device
7708 * @upper_priv: upper device private
7709 * @upper_info: upper info to be passed down via notifier
7710 * @extack: netlink extended ack
7712 * Adds a link to device which is upper to this one. In this case, only
7713 * one master upper device can be linked, although other non-master devices
7714 * might be linked as well. The caller must hold the RTNL lock.
7715 * On a failure a negative errno code is returned. On success the reference
7716 * counts are adjusted and the function returns zero.
7718 int netdev_master_upper_dev_link(struct net_device *dev,
7719 struct net_device *upper_dev,
7720 void *upper_priv, void *upper_info,
7721 struct netlink_ext_ack *extack)
7723 struct netdev_nested_priv priv = {
7724 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7728 return __netdev_upper_dev_link(dev, upper_dev, true,
7729 upper_priv, upper_info, &priv, extack);
7731 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7733 static void __netdev_upper_dev_unlink(struct net_device *dev,
7734 struct net_device *upper_dev,
7735 struct netdev_nested_priv *priv)
7737 struct netdev_notifier_changeupper_info changeupper_info = {
7741 .upper_dev = upper_dev,
7747 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7749 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7750 &changeupper_info.info);
7752 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7754 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7755 &changeupper_info.info);
7757 __netdev_update_upper_level(dev, NULL);
7758 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7760 __netdev_update_lower_level(upper_dev, priv);
7761 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7766 * netdev_upper_dev_unlink - Removes a link to upper device
7768 * @upper_dev: new upper device
7770 * Removes a link to device which is upper to this one. The caller must hold
7773 void netdev_upper_dev_unlink(struct net_device *dev,
7774 struct net_device *upper_dev)
7776 struct netdev_nested_priv priv = {
7777 .flags = NESTED_SYNC_TODO,
7781 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7783 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7785 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7786 struct net_device *lower_dev,
7789 struct netdev_adjacent *adj;
7791 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7795 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7800 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7801 struct net_device *lower_dev)
7803 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7806 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7807 struct net_device *lower_dev)
7809 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7812 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7813 struct net_device *new_dev,
7814 struct net_device *dev,
7815 struct netlink_ext_ack *extack)
7817 struct netdev_nested_priv priv = {
7826 if (old_dev && new_dev != old_dev)
7827 netdev_adjacent_dev_disable(dev, old_dev);
7828 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7831 if (old_dev && new_dev != old_dev)
7832 netdev_adjacent_dev_enable(dev, old_dev);
7838 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7840 void netdev_adjacent_change_commit(struct net_device *old_dev,
7841 struct net_device *new_dev,
7842 struct net_device *dev)
7844 struct netdev_nested_priv priv = {
7845 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7849 if (!new_dev || !old_dev)
7852 if (new_dev == old_dev)
7855 netdev_adjacent_dev_enable(dev, old_dev);
7856 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7858 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7860 void netdev_adjacent_change_abort(struct net_device *old_dev,
7861 struct net_device *new_dev,
7862 struct net_device *dev)
7864 struct netdev_nested_priv priv = {
7872 if (old_dev && new_dev != old_dev)
7873 netdev_adjacent_dev_enable(dev, old_dev);
7875 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7877 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7880 * netdev_bonding_info_change - Dispatch event about slave change
7882 * @bonding_info: info to dispatch
7884 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7885 * The caller must hold the RTNL lock.
7887 void netdev_bonding_info_change(struct net_device *dev,
7888 struct netdev_bonding_info *bonding_info)
7890 struct netdev_notifier_bonding_info info = {
7894 memcpy(&info.bonding_info, bonding_info,
7895 sizeof(struct netdev_bonding_info));
7896 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7899 EXPORT_SYMBOL(netdev_bonding_info_change);
7901 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7902 struct netlink_ext_ack *extack)
7904 struct netdev_notifier_offload_xstats_info info = {
7906 .info.extack = extack,
7907 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7912 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7914 if (!dev->offload_xstats_l3)
7917 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7918 NETDEV_OFFLOAD_XSTATS_DISABLE,
7920 err = notifier_to_errno(rc);
7927 kfree(dev->offload_xstats_l3);
7928 dev->offload_xstats_l3 = NULL;
7932 int netdev_offload_xstats_enable(struct net_device *dev,
7933 enum netdev_offload_xstats_type type,
7934 struct netlink_ext_ack *extack)
7938 if (netdev_offload_xstats_enabled(dev, type))
7942 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7943 return netdev_offload_xstats_enable_l3(dev, extack);
7949 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7951 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7953 struct netdev_notifier_offload_xstats_info info = {
7955 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7958 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7960 kfree(dev->offload_xstats_l3);
7961 dev->offload_xstats_l3 = NULL;
7964 int netdev_offload_xstats_disable(struct net_device *dev,
7965 enum netdev_offload_xstats_type type)
7969 if (!netdev_offload_xstats_enabled(dev, type))
7973 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7974 netdev_offload_xstats_disable_l3(dev);
7981 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7983 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7985 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7988 static struct rtnl_hw_stats64 *
7989 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7990 enum netdev_offload_xstats_type type)
7993 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7994 return dev->offload_xstats_l3;
8001 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8002 enum netdev_offload_xstats_type type)
8006 return netdev_offload_xstats_get_ptr(dev, type);
8008 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8010 struct netdev_notifier_offload_xstats_ru {
8014 struct netdev_notifier_offload_xstats_rd {
8015 struct rtnl_hw_stats64 stats;
8019 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8020 const struct rtnl_hw_stats64 *src)
8022 dest->rx_packets += src->rx_packets;
8023 dest->tx_packets += src->tx_packets;
8024 dest->rx_bytes += src->rx_bytes;
8025 dest->tx_bytes += src->tx_bytes;
8026 dest->rx_errors += src->rx_errors;
8027 dest->tx_errors += src->tx_errors;
8028 dest->rx_dropped += src->rx_dropped;
8029 dest->tx_dropped += src->tx_dropped;
8030 dest->multicast += src->multicast;
8033 static int netdev_offload_xstats_get_used(struct net_device *dev,
8034 enum netdev_offload_xstats_type type,
8036 struct netlink_ext_ack *extack)
8038 struct netdev_notifier_offload_xstats_ru report_used = {};
8039 struct netdev_notifier_offload_xstats_info info = {
8041 .info.extack = extack,
8043 .report_used = &report_used,
8047 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8048 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8050 *p_used = report_used.used;
8051 return notifier_to_errno(rc);
8054 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8055 enum netdev_offload_xstats_type type,
8056 struct rtnl_hw_stats64 *p_stats,
8058 struct netlink_ext_ack *extack)
8060 struct netdev_notifier_offload_xstats_rd report_delta = {};
8061 struct netdev_notifier_offload_xstats_info info = {
8063 .info.extack = extack,
8065 .report_delta = &report_delta,
8067 struct rtnl_hw_stats64 *stats;
8070 stats = netdev_offload_xstats_get_ptr(dev, type);
8071 if (WARN_ON(!stats))
8074 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8077 /* Cache whatever we got, even if there was an error, otherwise the
8078 * successful stats retrievals would get lost.
8080 netdev_hw_stats64_add(stats, &report_delta.stats);
8084 *p_used = report_delta.used;
8086 return notifier_to_errno(rc);
8089 int netdev_offload_xstats_get(struct net_device *dev,
8090 enum netdev_offload_xstats_type type,
8091 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8092 struct netlink_ext_ack *extack)
8097 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8100 return netdev_offload_xstats_get_used(dev, type, p_used,
8103 EXPORT_SYMBOL(netdev_offload_xstats_get);
8106 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8107 const struct rtnl_hw_stats64 *stats)
8109 report_delta->used = true;
8110 netdev_hw_stats64_add(&report_delta->stats, stats);
8112 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8115 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8117 report_used->used = true;
8119 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8121 void netdev_offload_xstats_push_delta(struct net_device *dev,
8122 enum netdev_offload_xstats_type type,
8123 const struct rtnl_hw_stats64 *p_stats)
8125 struct rtnl_hw_stats64 *stats;
8129 stats = netdev_offload_xstats_get_ptr(dev, type);
8130 if (WARN_ON(!stats))
8133 netdev_hw_stats64_add(stats, p_stats);
8135 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8138 * netdev_get_xmit_slave - Get the xmit slave of master device
8141 * @all_slaves: assume all the slaves are active
8143 * The reference counters are not incremented so the caller must be
8144 * careful with locks. The caller must hold RCU lock.
8145 * %NULL is returned if no slave is found.
8148 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8149 struct sk_buff *skb,
8152 const struct net_device_ops *ops = dev->netdev_ops;
8154 if (!ops->ndo_get_xmit_slave)
8156 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8158 EXPORT_SYMBOL(netdev_get_xmit_slave);
8160 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8163 const struct net_device_ops *ops = dev->netdev_ops;
8165 if (!ops->ndo_sk_get_lower_dev)
8167 return ops->ndo_sk_get_lower_dev(dev, sk);
8171 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8175 * %NULL is returned if no lower device is found.
8178 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8181 struct net_device *lower;
8183 lower = netdev_sk_get_lower_dev(dev, sk);
8186 lower = netdev_sk_get_lower_dev(dev, sk);
8191 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8193 static void netdev_adjacent_add_links(struct net_device *dev)
8195 struct netdev_adjacent *iter;
8197 struct net *net = dev_net(dev);
8199 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8200 if (!net_eq(net, dev_net(iter->dev)))
8202 netdev_adjacent_sysfs_add(iter->dev, dev,
8203 &iter->dev->adj_list.lower);
8204 netdev_adjacent_sysfs_add(dev, iter->dev,
8205 &dev->adj_list.upper);
8208 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8209 if (!net_eq(net, dev_net(iter->dev)))
8211 netdev_adjacent_sysfs_add(iter->dev, dev,
8212 &iter->dev->adj_list.upper);
8213 netdev_adjacent_sysfs_add(dev, iter->dev,
8214 &dev->adj_list.lower);
8218 static void netdev_adjacent_del_links(struct net_device *dev)
8220 struct netdev_adjacent *iter;
8222 struct net *net = dev_net(dev);
8224 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8225 if (!net_eq(net, dev_net(iter->dev)))
8227 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8228 &iter->dev->adj_list.lower);
8229 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8230 &dev->adj_list.upper);
8233 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8234 if (!net_eq(net, dev_net(iter->dev)))
8236 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8237 &iter->dev->adj_list.upper);
8238 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8239 &dev->adj_list.lower);
8243 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8245 struct netdev_adjacent *iter;
8247 struct net *net = dev_net(dev);
8249 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8250 if (!net_eq(net, dev_net(iter->dev)))
8252 netdev_adjacent_sysfs_del(iter->dev, oldname,
8253 &iter->dev->adj_list.lower);
8254 netdev_adjacent_sysfs_add(iter->dev, dev,
8255 &iter->dev->adj_list.lower);
8258 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8259 if (!net_eq(net, dev_net(iter->dev)))
8261 netdev_adjacent_sysfs_del(iter->dev, oldname,
8262 &iter->dev->adj_list.upper);
8263 netdev_adjacent_sysfs_add(iter->dev, dev,
8264 &iter->dev->adj_list.upper);
8268 void *netdev_lower_dev_get_private(struct net_device *dev,
8269 struct net_device *lower_dev)
8271 struct netdev_adjacent *lower;
8275 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8279 return lower->private;
8281 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8285 * netdev_lower_state_changed - Dispatch event about lower device state change
8286 * @lower_dev: device
8287 * @lower_state_info: state to dispatch
8289 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8290 * The caller must hold the RTNL lock.
8292 void netdev_lower_state_changed(struct net_device *lower_dev,
8293 void *lower_state_info)
8295 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8296 .info.dev = lower_dev,
8300 changelowerstate_info.lower_state_info = lower_state_info;
8301 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8302 &changelowerstate_info.info);
8304 EXPORT_SYMBOL(netdev_lower_state_changed);
8306 static void dev_change_rx_flags(struct net_device *dev, int flags)
8308 const struct net_device_ops *ops = dev->netdev_ops;
8310 if (ops->ndo_change_rx_flags)
8311 ops->ndo_change_rx_flags(dev, flags);
8314 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8316 unsigned int old_flags = dev->flags;
8322 dev->flags |= IFF_PROMISC;
8323 dev->promiscuity += inc;
8324 if (dev->promiscuity == 0) {
8327 * If inc causes overflow, untouch promisc and return error.
8330 dev->flags &= ~IFF_PROMISC;
8332 dev->promiscuity -= inc;
8333 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8337 if (dev->flags != old_flags) {
8338 pr_info("device %s %s promiscuous mode\n",
8340 dev->flags & IFF_PROMISC ? "entered" : "left");
8341 if (audit_enabled) {
8342 current_uid_gid(&uid, &gid);
8343 audit_log(audit_context(), GFP_ATOMIC,
8344 AUDIT_ANOM_PROMISCUOUS,
8345 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8346 dev->name, (dev->flags & IFF_PROMISC),
8347 (old_flags & IFF_PROMISC),
8348 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8349 from_kuid(&init_user_ns, uid),
8350 from_kgid(&init_user_ns, gid),
8351 audit_get_sessionid(current));
8354 dev_change_rx_flags(dev, IFF_PROMISC);
8357 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8362 * dev_set_promiscuity - update promiscuity count on a device
8366 * Add or remove promiscuity from a device. While the count in the device
8367 * remains above zero the interface remains promiscuous. Once it hits zero
8368 * the device reverts back to normal filtering operation. A negative inc
8369 * value is used to drop promiscuity on the device.
8370 * Return 0 if successful or a negative errno code on error.
8372 int dev_set_promiscuity(struct net_device *dev, int inc)
8374 unsigned int old_flags = dev->flags;
8377 err = __dev_set_promiscuity(dev, inc, true);
8380 if (dev->flags != old_flags)
8381 dev_set_rx_mode(dev);
8384 EXPORT_SYMBOL(dev_set_promiscuity);
8386 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8388 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8392 dev->flags |= IFF_ALLMULTI;
8393 dev->allmulti += inc;
8394 if (dev->allmulti == 0) {
8397 * If inc causes overflow, untouch allmulti and return error.
8400 dev->flags &= ~IFF_ALLMULTI;
8402 dev->allmulti -= inc;
8403 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8407 if (dev->flags ^ old_flags) {
8408 dev_change_rx_flags(dev, IFF_ALLMULTI);
8409 dev_set_rx_mode(dev);
8411 __dev_notify_flags(dev, old_flags,
8412 dev->gflags ^ old_gflags);
8418 * dev_set_allmulti - update allmulti count on a device
8422 * Add or remove reception of all multicast frames to a device. While the
8423 * count in the device remains above zero the interface remains listening
8424 * to all interfaces. Once it hits zero the device reverts back to normal
8425 * filtering operation. A negative @inc value is used to drop the counter
8426 * when releasing a resource needing all multicasts.
8427 * Return 0 if successful or a negative errno code on error.
8430 int dev_set_allmulti(struct net_device *dev, int inc)
8432 return __dev_set_allmulti(dev, inc, true);
8434 EXPORT_SYMBOL(dev_set_allmulti);
8437 * Upload unicast and multicast address lists to device and
8438 * configure RX filtering. When the device doesn't support unicast
8439 * filtering it is put in promiscuous mode while unicast addresses
8442 void __dev_set_rx_mode(struct net_device *dev)
8444 const struct net_device_ops *ops = dev->netdev_ops;
8446 /* dev_open will call this function so the list will stay sane. */
8447 if (!(dev->flags&IFF_UP))
8450 if (!netif_device_present(dev))
8453 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8454 /* Unicast addresses changes may only happen under the rtnl,
8455 * therefore calling __dev_set_promiscuity here is safe.
8457 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8458 __dev_set_promiscuity(dev, 1, false);
8459 dev->uc_promisc = true;
8460 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8461 __dev_set_promiscuity(dev, -1, false);
8462 dev->uc_promisc = false;
8466 if (ops->ndo_set_rx_mode)
8467 ops->ndo_set_rx_mode(dev);
8470 void dev_set_rx_mode(struct net_device *dev)
8472 netif_addr_lock_bh(dev);
8473 __dev_set_rx_mode(dev);
8474 netif_addr_unlock_bh(dev);
8478 * dev_get_flags - get flags reported to userspace
8481 * Get the combination of flag bits exported through APIs to userspace.
8483 unsigned int dev_get_flags(const struct net_device *dev)
8487 flags = (dev->flags & ~(IFF_PROMISC |
8492 (dev->gflags & (IFF_PROMISC |
8495 if (netif_running(dev)) {
8496 if (netif_oper_up(dev))
8497 flags |= IFF_RUNNING;
8498 if (netif_carrier_ok(dev))
8499 flags |= IFF_LOWER_UP;
8500 if (netif_dormant(dev))
8501 flags |= IFF_DORMANT;
8506 EXPORT_SYMBOL(dev_get_flags);
8508 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8509 struct netlink_ext_ack *extack)
8511 unsigned int old_flags = dev->flags;
8517 * Set the flags on our device.
8520 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8521 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8523 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8527 * Load in the correct multicast list now the flags have changed.
8530 if ((old_flags ^ flags) & IFF_MULTICAST)
8531 dev_change_rx_flags(dev, IFF_MULTICAST);
8533 dev_set_rx_mode(dev);
8536 * Have we downed the interface. We handle IFF_UP ourselves
8537 * according to user attempts to set it, rather than blindly
8542 if ((old_flags ^ flags) & IFF_UP) {
8543 if (old_flags & IFF_UP)
8546 ret = __dev_open(dev, extack);
8549 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8550 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8551 unsigned int old_flags = dev->flags;
8553 dev->gflags ^= IFF_PROMISC;
8555 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8556 if (dev->flags != old_flags)
8557 dev_set_rx_mode(dev);
8560 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8561 * is important. Some (broken) drivers set IFF_PROMISC, when
8562 * IFF_ALLMULTI is requested not asking us and not reporting.
8564 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8565 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8567 dev->gflags ^= IFF_ALLMULTI;
8568 __dev_set_allmulti(dev, inc, false);
8574 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8575 unsigned int gchanges)
8577 unsigned int changes = dev->flags ^ old_flags;
8580 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8582 if (changes & IFF_UP) {
8583 if (dev->flags & IFF_UP)
8584 call_netdevice_notifiers(NETDEV_UP, dev);
8586 call_netdevice_notifiers(NETDEV_DOWN, dev);
8589 if (dev->flags & IFF_UP &&
8590 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8591 struct netdev_notifier_change_info change_info = {
8595 .flags_changed = changes,
8598 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8603 * dev_change_flags - change device settings
8605 * @flags: device state flags
8606 * @extack: netlink extended ack
8608 * Change settings on device based state flags. The flags are
8609 * in the userspace exported format.
8611 int dev_change_flags(struct net_device *dev, unsigned int flags,
8612 struct netlink_ext_ack *extack)
8615 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8617 ret = __dev_change_flags(dev, flags, extack);
8621 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8622 __dev_notify_flags(dev, old_flags, changes);
8625 EXPORT_SYMBOL(dev_change_flags);
8627 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8629 const struct net_device_ops *ops = dev->netdev_ops;
8631 if (ops->ndo_change_mtu)
8632 return ops->ndo_change_mtu(dev, new_mtu);
8634 /* Pairs with all the lockless reads of dev->mtu in the stack */
8635 WRITE_ONCE(dev->mtu, new_mtu);
8638 EXPORT_SYMBOL(__dev_set_mtu);
8640 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8641 struct netlink_ext_ack *extack)
8643 /* MTU must be positive, and in range */
8644 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8645 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8649 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8650 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8657 * dev_set_mtu_ext - Change maximum transfer unit
8659 * @new_mtu: new transfer unit
8660 * @extack: netlink extended ack
8662 * Change the maximum transfer size of the network device.
8664 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8665 struct netlink_ext_ack *extack)
8669 if (new_mtu == dev->mtu)
8672 err = dev_validate_mtu(dev, new_mtu, extack);
8676 if (!netif_device_present(dev))
8679 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8680 err = notifier_to_errno(err);
8684 orig_mtu = dev->mtu;
8685 err = __dev_set_mtu(dev, new_mtu);
8688 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8690 err = notifier_to_errno(err);
8692 /* setting mtu back and notifying everyone again,
8693 * so that they have a chance to revert changes.
8695 __dev_set_mtu(dev, orig_mtu);
8696 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8703 int dev_set_mtu(struct net_device *dev, int new_mtu)
8705 struct netlink_ext_ack extack;
8708 memset(&extack, 0, sizeof(extack));
8709 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8710 if (err && extack._msg)
8711 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8714 EXPORT_SYMBOL(dev_set_mtu);
8717 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8719 * @new_len: new tx queue length
8721 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8723 unsigned int orig_len = dev->tx_queue_len;
8726 if (new_len != (unsigned int)new_len)
8729 if (new_len != orig_len) {
8730 dev->tx_queue_len = new_len;
8731 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8732 res = notifier_to_errno(res);
8735 res = dev_qdisc_change_tx_queue_len(dev);
8743 netdev_err(dev, "refused to change device tx_queue_len\n");
8744 dev->tx_queue_len = orig_len;
8749 * dev_set_group - Change group this device belongs to
8751 * @new_group: group this device should belong to
8753 void dev_set_group(struct net_device *dev, int new_group)
8755 dev->group = new_group;
8759 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8761 * @addr: new address
8762 * @extack: netlink extended ack
8764 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8765 struct netlink_ext_ack *extack)
8767 struct netdev_notifier_pre_changeaddr_info info = {
8769 .info.extack = extack,
8774 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8775 return notifier_to_errno(rc);
8777 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8780 * dev_set_mac_address - Change Media Access Control Address
8783 * @extack: netlink extended ack
8785 * Change the hardware (MAC) address of the device
8787 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8788 struct netlink_ext_ack *extack)
8790 const struct net_device_ops *ops = dev->netdev_ops;
8793 if (!ops->ndo_set_mac_address)
8795 if (sa->sa_family != dev->type)
8797 if (!netif_device_present(dev))
8799 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8802 err = ops->ndo_set_mac_address(dev, sa);
8805 dev->addr_assign_type = NET_ADDR_SET;
8806 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8807 add_device_randomness(dev->dev_addr, dev->addr_len);
8810 EXPORT_SYMBOL(dev_set_mac_address);
8812 static DECLARE_RWSEM(dev_addr_sem);
8814 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8815 struct netlink_ext_ack *extack)
8819 down_write(&dev_addr_sem);
8820 ret = dev_set_mac_address(dev, sa, extack);
8821 up_write(&dev_addr_sem);
8824 EXPORT_SYMBOL(dev_set_mac_address_user);
8826 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8828 size_t size = sizeof(sa->sa_data);
8829 struct net_device *dev;
8832 down_read(&dev_addr_sem);
8835 dev = dev_get_by_name_rcu(net, dev_name);
8841 memset(sa->sa_data, 0, size);
8843 memcpy(sa->sa_data, dev->dev_addr,
8844 min_t(size_t, size, dev->addr_len));
8845 sa->sa_family = dev->type;
8849 up_read(&dev_addr_sem);
8852 EXPORT_SYMBOL(dev_get_mac_address);
8855 * dev_change_carrier - Change device carrier
8857 * @new_carrier: new value
8859 * Change device carrier
8861 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8863 const struct net_device_ops *ops = dev->netdev_ops;
8865 if (!ops->ndo_change_carrier)
8867 if (!netif_device_present(dev))
8869 return ops->ndo_change_carrier(dev, new_carrier);
8873 * dev_get_phys_port_id - Get device physical port ID
8877 * Get device physical port ID
8879 int dev_get_phys_port_id(struct net_device *dev,
8880 struct netdev_phys_item_id *ppid)
8882 const struct net_device_ops *ops = dev->netdev_ops;
8884 if (!ops->ndo_get_phys_port_id)
8886 return ops->ndo_get_phys_port_id(dev, ppid);
8890 * dev_get_phys_port_name - Get device physical port name
8893 * @len: limit of bytes to copy to name
8895 * Get device physical port name
8897 int dev_get_phys_port_name(struct net_device *dev,
8898 char *name, size_t len)
8900 const struct net_device_ops *ops = dev->netdev_ops;
8903 if (ops->ndo_get_phys_port_name) {
8904 err = ops->ndo_get_phys_port_name(dev, name, len);
8905 if (err != -EOPNOTSUPP)
8908 return devlink_compat_phys_port_name_get(dev, name, len);
8912 * dev_get_port_parent_id - Get the device's port parent identifier
8913 * @dev: network device
8914 * @ppid: pointer to a storage for the port's parent identifier
8915 * @recurse: allow/disallow recursion to lower devices
8917 * Get the devices's port parent identifier
8919 int dev_get_port_parent_id(struct net_device *dev,
8920 struct netdev_phys_item_id *ppid,
8923 const struct net_device_ops *ops = dev->netdev_ops;
8924 struct netdev_phys_item_id first = { };
8925 struct net_device *lower_dev;
8926 struct list_head *iter;
8929 if (ops->ndo_get_port_parent_id) {
8930 err = ops->ndo_get_port_parent_id(dev, ppid);
8931 if (err != -EOPNOTSUPP)
8935 err = devlink_compat_switch_id_get(dev, ppid);
8936 if (!recurse || err != -EOPNOTSUPP)
8939 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8940 err = dev_get_port_parent_id(lower_dev, ppid, true);
8945 else if (memcmp(&first, ppid, sizeof(*ppid)))
8951 EXPORT_SYMBOL(dev_get_port_parent_id);
8954 * netdev_port_same_parent_id - Indicate if two network devices have
8955 * the same port parent identifier
8956 * @a: first network device
8957 * @b: second network device
8959 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8961 struct netdev_phys_item_id a_id = { };
8962 struct netdev_phys_item_id b_id = { };
8964 if (dev_get_port_parent_id(a, &a_id, true) ||
8965 dev_get_port_parent_id(b, &b_id, true))
8968 return netdev_phys_item_id_same(&a_id, &b_id);
8970 EXPORT_SYMBOL(netdev_port_same_parent_id);
8973 * dev_change_proto_down - set carrier according to proto_down.
8976 * @proto_down: new value
8978 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8980 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8982 if (!netif_device_present(dev))
8985 netif_carrier_off(dev);
8987 netif_carrier_on(dev);
8988 dev->proto_down = proto_down;
8993 * dev_change_proto_down_reason - proto down reason
8996 * @mask: proto down mask
8997 * @value: proto down value
8999 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9005 dev->proto_down_reason = value;
9007 for_each_set_bit(b, &mask, 32) {
9008 if (value & (1 << b))
9009 dev->proto_down_reason |= BIT(b);
9011 dev->proto_down_reason &= ~BIT(b);
9016 struct bpf_xdp_link {
9017 struct bpf_link link;
9018 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9022 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9024 if (flags & XDP_FLAGS_HW_MODE)
9026 if (flags & XDP_FLAGS_DRV_MODE)
9027 return XDP_MODE_DRV;
9028 if (flags & XDP_FLAGS_SKB_MODE)
9029 return XDP_MODE_SKB;
9030 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9033 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9037 return generic_xdp_install;
9040 return dev->netdev_ops->ndo_bpf;
9046 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9047 enum bpf_xdp_mode mode)
9049 return dev->xdp_state[mode].link;
9052 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9053 enum bpf_xdp_mode mode)
9055 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9058 return link->link.prog;
9059 return dev->xdp_state[mode].prog;
9062 u8 dev_xdp_prog_count(struct net_device *dev)
9067 for (i = 0; i < __MAX_XDP_MODE; i++)
9068 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9072 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9074 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9076 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9078 return prog ? prog->aux->id : 0;
9081 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9082 struct bpf_xdp_link *link)
9084 dev->xdp_state[mode].link = link;
9085 dev->xdp_state[mode].prog = NULL;
9088 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9089 struct bpf_prog *prog)
9091 dev->xdp_state[mode].link = NULL;
9092 dev->xdp_state[mode].prog = prog;
9095 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9096 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9097 u32 flags, struct bpf_prog *prog)
9099 struct netdev_bpf xdp;
9102 memset(&xdp, 0, sizeof(xdp));
9103 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9104 xdp.extack = extack;
9108 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9109 * "moved" into driver), so they don't increment it on their own, but
9110 * they do decrement refcnt when program is detached or replaced.
9111 * Given net_device also owns link/prog, we need to bump refcnt here
9112 * to prevent drivers from underflowing it.
9116 err = bpf_op(dev, &xdp);
9123 if (mode != XDP_MODE_HW)
9124 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9129 static void dev_xdp_uninstall(struct net_device *dev)
9131 struct bpf_xdp_link *link;
9132 struct bpf_prog *prog;
9133 enum bpf_xdp_mode mode;
9138 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9139 prog = dev_xdp_prog(dev, mode);
9143 bpf_op = dev_xdp_bpf_op(dev, mode);
9147 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9149 /* auto-detach link from net device */
9150 link = dev_xdp_link(dev, mode);
9156 dev_xdp_set_link(dev, mode, NULL);
9160 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9161 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9162 struct bpf_prog *old_prog, u32 flags)
9164 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9165 struct bpf_prog *cur_prog;
9166 struct net_device *upper;
9167 struct list_head *iter;
9168 enum bpf_xdp_mode mode;
9174 /* either link or prog attachment, never both */
9175 if (link && (new_prog || old_prog))
9177 /* link supports only XDP mode flags */
9178 if (link && (flags & ~XDP_FLAGS_MODES)) {
9179 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9182 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9183 if (num_modes > 1) {
9184 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9187 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9188 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9189 NL_SET_ERR_MSG(extack,
9190 "More than one program loaded, unset mode is ambiguous");
9193 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9194 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9195 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9199 mode = dev_xdp_mode(dev, flags);
9200 /* can't replace attached link */
9201 if (dev_xdp_link(dev, mode)) {
9202 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9206 /* don't allow if an upper device already has a program */
9207 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9208 if (dev_xdp_prog_count(upper) > 0) {
9209 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9214 cur_prog = dev_xdp_prog(dev, mode);
9215 /* can't replace attached prog with link */
9216 if (link && cur_prog) {
9217 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9220 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9221 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9225 /* put effective new program into new_prog */
9227 new_prog = link->link.prog;
9230 bool offload = mode == XDP_MODE_HW;
9231 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9232 ? XDP_MODE_DRV : XDP_MODE_SKB;
9234 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9235 NL_SET_ERR_MSG(extack, "XDP program already attached");
9238 if (!offload && dev_xdp_prog(dev, other_mode)) {
9239 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9242 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9243 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9246 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9247 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9250 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9251 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9256 /* don't call drivers if the effective program didn't change */
9257 if (new_prog != cur_prog) {
9258 bpf_op = dev_xdp_bpf_op(dev, mode);
9260 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9264 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9270 dev_xdp_set_link(dev, mode, link);
9272 dev_xdp_set_prog(dev, mode, new_prog);
9274 bpf_prog_put(cur_prog);
9279 static int dev_xdp_attach_link(struct net_device *dev,
9280 struct netlink_ext_ack *extack,
9281 struct bpf_xdp_link *link)
9283 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9286 static int dev_xdp_detach_link(struct net_device *dev,
9287 struct netlink_ext_ack *extack,
9288 struct bpf_xdp_link *link)
9290 enum bpf_xdp_mode mode;
9295 mode = dev_xdp_mode(dev, link->flags);
9296 if (dev_xdp_link(dev, mode) != link)
9299 bpf_op = dev_xdp_bpf_op(dev, mode);
9300 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9301 dev_xdp_set_link(dev, mode, NULL);
9305 static void bpf_xdp_link_release(struct bpf_link *link)
9307 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9311 /* if racing with net_device's tear down, xdp_link->dev might be
9312 * already NULL, in which case link was already auto-detached
9314 if (xdp_link->dev) {
9315 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9316 xdp_link->dev = NULL;
9322 static int bpf_xdp_link_detach(struct bpf_link *link)
9324 bpf_xdp_link_release(link);
9328 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9330 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9335 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9336 struct seq_file *seq)
9338 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9343 ifindex = xdp_link->dev->ifindex;
9346 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9349 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9350 struct bpf_link_info *info)
9352 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9357 ifindex = xdp_link->dev->ifindex;
9360 info->xdp.ifindex = ifindex;
9364 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9365 struct bpf_prog *old_prog)
9367 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9368 enum bpf_xdp_mode mode;
9374 /* link might have been auto-released already, so fail */
9375 if (!xdp_link->dev) {
9380 if (old_prog && link->prog != old_prog) {
9384 old_prog = link->prog;
9385 if (old_prog->type != new_prog->type ||
9386 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9391 if (old_prog == new_prog) {
9392 /* no-op, don't disturb drivers */
9393 bpf_prog_put(new_prog);
9397 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9398 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9399 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9400 xdp_link->flags, new_prog);
9404 old_prog = xchg(&link->prog, new_prog);
9405 bpf_prog_put(old_prog);
9412 static const struct bpf_link_ops bpf_xdp_link_lops = {
9413 .release = bpf_xdp_link_release,
9414 .dealloc = bpf_xdp_link_dealloc,
9415 .detach = bpf_xdp_link_detach,
9416 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9417 .fill_link_info = bpf_xdp_link_fill_link_info,
9418 .update_prog = bpf_xdp_link_update,
9421 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9423 struct net *net = current->nsproxy->net_ns;
9424 struct bpf_link_primer link_primer;
9425 struct bpf_xdp_link *link;
9426 struct net_device *dev;
9430 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9436 link = kzalloc(sizeof(*link), GFP_USER);
9442 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9444 link->flags = attr->link_create.flags;
9446 err = bpf_link_prime(&link->link, &link_primer);
9452 err = dev_xdp_attach_link(dev, NULL, link);
9457 bpf_link_cleanup(&link_primer);
9461 fd = bpf_link_settle(&link_primer);
9462 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9475 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9477 * @extack: netlink extended ack
9478 * @fd: new program fd or negative value to clear
9479 * @expected_fd: old program fd that userspace expects to replace or clear
9480 * @flags: xdp-related flags
9482 * Set or clear a bpf program for a device
9484 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9485 int fd, int expected_fd, u32 flags)
9487 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9488 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9494 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9495 mode != XDP_MODE_SKB);
9496 if (IS_ERR(new_prog))
9497 return PTR_ERR(new_prog);
9500 if (expected_fd >= 0) {
9501 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9502 mode != XDP_MODE_SKB);
9503 if (IS_ERR(old_prog)) {
9504 err = PTR_ERR(old_prog);
9510 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9513 if (err && new_prog)
9514 bpf_prog_put(new_prog);
9516 bpf_prog_put(old_prog);
9521 * dev_new_index - allocate an ifindex
9522 * @net: the applicable net namespace
9524 * Returns a suitable unique value for a new device interface
9525 * number. The caller must hold the rtnl semaphore or the
9526 * dev_base_lock to be sure it remains unique.
9528 static int dev_new_index(struct net *net)
9530 int ifindex = net->ifindex;
9535 if (!__dev_get_by_index(net, ifindex))
9536 return net->ifindex = ifindex;
9540 /* Delayed registration/unregisteration */
9541 LIST_HEAD(net_todo_list);
9542 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9544 static void net_set_todo(struct net_device *dev)
9546 list_add_tail(&dev->todo_list, &net_todo_list);
9547 atomic_inc(&dev_net(dev)->dev_unreg_count);
9550 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9551 struct net_device *upper, netdev_features_t features)
9553 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9554 netdev_features_t feature;
9557 for_each_netdev_feature(upper_disables, feature_bit) {
9558 feature = __NETIF_F_BIT(feature_bit);
9559 if (!(upper->wanted_features & feature)
9560 && (features & feature)) {
9561 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9562 &feature, upper->name);
9563 features &= ~feature;
9570 static void netdev_sync_lower_features(struct net_device *upper,
9571 struct net_device *lower, netdev_features_t features)
9573 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9574 netdev_features_t feature;
9577 for_each_netdev_feature(upper_disables, feature_bit) {
9578 feature = __NETIF_F_BIT(feature_bit);
9579 if (!(features & feature) && (lower->features & feature)) {
9580 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9581 &feature, lower->name);
9582 lower->wanted_features &= ~feature;
9583 __netdev_update_features(lower);
9585 if (unlikely(lower->features & feature))
9586 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9587 &feature, lower->name);
9589 netdev_features_change(lower);
9594 static netdev_features_t netdev_fix_features(struct net_device *dev,
9595 netdev_features_t features)
9597 /* Fix illegal checksum combinations */
9598 if ((features & NETIF_F_HW_CSUM) &&
9599 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9600 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9601 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9604 /* TSO requires that SG is present as well. */
9605 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9606 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9607 features &= ~NETIF_F_ALL_TSO;
9610 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9611 !(features & NETIF_F_IP_CSUM)) {
9612 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9613 features &= ~NETIF_F_TSO;
9614 features &= ~NETIF_F_TSO_ECN;
9617 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9618 !(features & NETIF_F_IPV6_CSUM)) {
9619 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9620 features &= ~NETIF_F_TSO6;
9623 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9624 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9625 features &= ~NETIF_F_TSO_MANGLEID;
9627 /* TSO ECN requires that TSO is present as well. */
9628 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9629 features &= ~NETIF_F_TSO_ECN;
9631 /* Software GSO depends on SG. */
9632 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9633 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9634 features &= ~NETIF_F_GSO;
9637 /* GSO partial features require GSO partial be set */
9638 if ((features & dev->gso_partial_features) &&
9639 !(features & NETIF_F_GSO_PARTIAL)) {
9641 "Dropping partially supported GSO features since no GSO partial.\n");
9642 features &= ~dev->gso_partial_features;
9645 if (!(features & NETIF_F_RXCSUM)) {
9646 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9647 * successfully merged by hardware must also have the
9648 * checksum verified by hardware. If the user does not
9649 * want to enable RXCSUM, logically, we should disable GRO_HW.
9651 if (features & NETIF_F_GRO_HW) {
9652 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9653 features &= ~NETIF_F_GRO_HW;
9657 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9658 if (features & NETIF_F_RXFCS) {
9659 if (features & NETIF_F_LRO) {
9660 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9661 features &= ~NETIF_F_LRO;
9664 if (features & NETIF_F_GRO_HW) {
9665 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9666 features &= ~NETIF_F_GRO_HW;
9670 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9671 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9672 features &= ~NETIF_F_LRO;
9675 if (features & NETIF_F_HW_TLS_TX) {
9676 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9677 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9678 bool hw_csum = features & NETIF_F_HW_CSUM;
9680 if (!ip_csum && !hw_csum) {
9681 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9682 features &= ~NETIF_F_HW_TLS_TX;
9686 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9687 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9688 features &= ~NETIF_F_HW_TLS_RX;
9694 int __netdev_update_features(struct net_device *dev)
9696 struct net_device *upper, *lower;
9697 netdev_features_t features;
9698 struct list_head *iter;
9703 features = netdev_get_wanted_features(dev);
9705 if (dev->netdev_ops->ndo_fix_features)
9706 features = dev->netdev_ops->ndo_fix_features(dev, features);
9708 /* driver might be less strict about feature dependencies */
9709 features = netdev_fix_features(dev, features);
9711 /* some features can't be enabled if they're off on an upper device */
9712 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9713 features = netdev_sync_upper_features(dev, upper, features);
9715 if (dev->features == features)
9718 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9719 &dev->features, &features);
9721 if (dev->netdev_ops->ndo_set_features)
9722 err = dev->netdev_ops->ndo_set_features(dev, features);
9726 if (unlikely(err < 0)) {
9728 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9729 err, &features, &dev->features);
9730 /* return non-0 since some features might have changed and
9731 * it's better to fire a spurious notification than miss it
9737 /* some features must be disabled on lower devices when disabled
9738 * on an upper device (think: bonding master or bridge)
9740 netdev_for_each_lower_dev(dev, lower, iter)
9741 netdev_sync_lower_features(dev, lower, features);
9744 netdev_features_t diff = features ^ dev->features;
9746 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9747 /* udp_tunnel_{get,drop}_rx_info both need
9748 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9749 * device, or they won't do anything.
9750 * Thus we need to update dev->features
9751 * *before* calling udp_tunnel_get_rx_info,
9752 * but *after* calling udp_tunnel_drop_rx_info.
9754 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9755 dev->features = features;
9756 udp_tunnel_get_rx_info(dev);
9758 udp_tunnel_drop_rx_info(dev);
9762 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9763 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9764 dev->features = features;
9765 err |= vlan_get_rx_ctag_filter_info(dev);
9767 vlan_drop_rx_ctag_filter_info(dev);
9771 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9772 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9773 dev->features = features;
9774 err |= vlan_get_rx_stag_filter_info(dev);
9776 vlan_drop_rx_stag_filter_info(dev);
9780 dev->features = features;
9783 return err < 0 ? 0 : 1;
9787 * netdev_update_features - recalculate device features
9788 * @dev: the device to check
9790 * Recalculate dev->features set and send notifications if it
9791 * has changed. Should be called after driver or hardware dependent
9792 * conditions might have changed that influence the features.
9794 void netdev_update_features(struct net_device *dev)
9796 if (__netdev_update_features(dev))
9797 netdev_features_change(dev);
9799 EXPORT_SYMBOL(netdev_update_features);
9802 * netdev_change_features - recalculate device features
9803 * @dev: the device to check
9805 * Recalculate dev->features set and send notifications even
9806 * if they have not changed. Should be called instead of
9807 * netdev_update_features() if also dev->vlan_features might
9808 * have changed to allow the changes to be propagated to stacked
9811 void netdev_change_features(struct net_device *dev)
9813 __netdev_update_features(dev);
9814 netdev_features_change(dev);
9816 EXPORT_SYMBOL(netdev_change_features);
9819 * netif_stacked_transfer_operstate - transfer operstate
9820 * @rootdev: the root or lower level device to transfer state from
9821 * @dev: the device to transfer operstate to
9823 * Transfer operational state from root to device. This is normally
9824 * called when a stacking relationship exists between the root
9825 * device and the device(a leaf device).
9827 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9828 struct net_device *dev)
9830 if (rootdev->operstate == IF_OPER_DORMANT)
9831 netif_dormant_on(dev);
9833 netif_dormant_off(dev);
9835 if (rootdev->operstate == IF_OPER_TESTING)
9836 netif_testing_on(dev);
9838 netif_testing_off(dev);
9840 if (netif_carrier_ok(rootdev))
9841 netif_carrier_on(dev);
9843 netif_carrier_off(dev);
9845 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9847 static int netif_alloc_rx_queues(struct net_device *dev)
9849 unsigned int i, count = dev->num_rx_queues;
9850 struct netdev_rx_queue *rx;
9851 size_t sz = count * sizeof(*rx);
9856 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9862 for (i = 0; i < count; i++) {
9865 /* XDP RX-queue setup */
9866 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9873 /* Rollback successful reg's and free other resources */
9875 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9881 static void netif_free_rx_queues(struct net_device *dev)
9883 unsigned int i, count = dev->num_rx_queues;
9885 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9889 for (i = 0; i < count; i++)
9890 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9895 static void netdev_init_one_queue(struct net_device *dev,
9896 struct netdev_queue *queue, void *_unused)
9898 /* Initialize queue lock */
9899 spin_lock_init(&queue->_xmit_lock);
9900 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9901 queue->xmit_lock_owner = -1;
9902 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9905 dql_init(&queue->dql, HZ);
9909 static void netif_free_tx_queues(struct net_device *dev)
9914 static int netif_alloc_netdev_queues(struct net_device *dev)
9916 unsigned int count = dev->num_tx_queues;
9917 struct netdev_queue *tx;
9918 size_t sz = count * sizeof(*tx);
9920 if (count < 1 || count > 0xffff)
9923 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9929 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9930 spin_lock_init(&dev->tx_global_lock);
9935 void netif_tx_stop_all_queues(struct net_device *dev)
9939 for (i = 0; i < dev->num_tx_queues; i++) {
9940 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9942 netif_tx_stop_queue(txq);
9945 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9948 * register_netdevice() - register a network device
9949 * @dev: device to register
9951 * Take a prepared network device structure and make it externally accessible.
9952 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9953 * Callers must hold the rtnl lock - you may want register_netdev()
9956 int register_netdevice(struct net_device *dev)
9959 struct net *net = dev_net(dev);
9961 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9962 NETDEV_FEATURE_COUNT);
9963 BUG_ON(dev_boot_phase);
9968 /* When net_device's are persistent, this will be fatal. */
9969 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9972 ret = ethtool_check_ops(dev->ethtool_ops);
9976 spin_lock_init(&dev->addr_list_lock);
9977 netdev_set_addr_lockdep_class(dev);
9979 ret = dev_get_valid_name(net, dev, dev->name);
9984 dev->name_node = netdev_name_node_head_alloc(dev);
9985 if (!dev->name_node)
9988 /* Init, if this function is available */
9989 if (dev->netdev_ops->ndo_init) {
9990 ret = dev->netdev_ops->ndo_init(dev);
9998 if (((dev->hw_features | dev->features) &
9999 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10000 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10001 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10002 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10009 dev->ifindex = dev_new_index(net);
10010 else if (__dev_get_by_index(net, dev->ifindex))
10013 /* Transfer changeable features to wanted_features and enable
10014 * software offloads (GSO and GRO).
10016 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10017 dev->features |= NETIF_F_SOFT_FEATURES;
10019 if (dev->udp_tunnel_nic_info) {
10020 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10021 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10024 dev->wanted_features = dev->features & dev->hw_features;
10026 if (!(dev->flags & IFF_LOOPBACK))
10027 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10029 /* If IPv4 TCP segmentation offload is supported we should also
10030 * allow the device to enable segmenting the frame with the option
10031 * of ignoring a static IP ID value. This doesn't enable the
10032 * feature itself but allows the user to enable it later.
10034 if (dev->hw_features & NETIF_F_TSO)
10035 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10036 if (dev->vlan_features & NETIF_F_TSO)
10037 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10038 if (dev->mpls_features & NETIF_F_TSO)
10039 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10040 if (dev->hw_enc_features & NETIF_F_TSO)
10041 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10043 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10045 dev->vlan_features |= NETIF_F_HIGHDMA;
10047 /* Make NETIF_F_SG inheritable to tunnel devices.
10049 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10051 /* Make NETIF_F_SG inheritable to MPLS.
10053 dev->mpls_features |= NETIF_F_SG;
10055 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10056 ret = notifier_to_errno(ret);
10060 ret = netdev_register_kobject(dev);
10061 write_lock(&dev_base_lock);
10062 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10063 write_unlock(&dev_base_lock);
10067 __netdev_update_features(dev);
10070 * Default initial state at registry is that the
10071 * device is present.
10074 set_bit(__LINK_STATE_PRESENT, &dev->state);
10076 linkwatch_init_dev(dev);
10078 dev_init_scheduler(dev);
10080 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10081 list_netdevice(dev);
10083 add_device_randomness(dev->dev_addr, dev->addr_len);
10085 /* If the device has permanent device address, driver should
10086 * set dev_addr and also addr_assign_type should be set to
10087 * NET_ADDR_PERM (default value).
10089 if (dev->addr_assign_type == NET_ADDR_PERM)
10090 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10092 /* Notify protocols, that a new device appeared. */
10093 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10094 ret = notifier_to_errno(ret);
10096 /* Expect explicit free_netdev() on failure */
10097 dev->needs_free_netdev = false;
10098 unregister_netdevice_queue(dev, NULL);
10102 * Prevent userspace races by waiting until the network
10103 * device is fully setup before sending notifications.
10105 if (!dev->rtnl_link_ops ||
10106 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10107 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10113 if (dev->netdev_ops->ndo_uninit)
10114 dev->netdev_ops->ndo_uninit(dev);
10115 if (dev->priv_destructor)
10116 dev->priv_destructor(dev);
10118 netdev_name_node_free(dev->name_node);
10121 EXPORT_SYMBOL(register_netdevice);
10124 * init_dummy_netdev - init a dummy network device for NAPI
10125 * @dev: device to init
10127 * This takes a network device structure and initialize the minimum
10128 * amount of fields so it can be used to schedule NAPI polls without
10129 * registering a full blown interface. This is to be used by drivers
10130 * that need to tie several hardware interfaces to a single NAPI
10131 * poll scheduler due to HW limitations.
10133 int init_dummy_netdev(struct net_device *dev)
10135 /* Clear everything. Note we don't initialize spinlocks
10136 * are they aren't supposed to be taken by any of the
10137 * NAPI code and this dummy netdev is supposed to be
10138 * only ever used for NAPI polls
10140 memset(dev, 0, sizeof(struct net_device));
10142 /* make sure we BUG if trying to hit standard
10143 * register/unregister code path
10145 dev->reg_state = NETREG_DUMMY;
10147 /* NAPI wants this */
10148 INIT_LIST_HEAD(&dev->napi_list);
10150 /* a dummy interface is started by default */
10151 set_bit(__LINK_STATE_PRESENT, &dev->state);
10152 set_bit(__LINK_STATE_START, &dev->state);
10154 /* napi_busy_loop stats accounting wants this */
10155 dev_net_set(dev, &init_net);
10157 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10158 * because users of this 'device' dont need to change
10164 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10168 * register_netdev - register a network device
10169 * @dev: device to register
10171 * Take a completed network device structure and add it to the kernel
10172 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10173 * chain. 0 is returned on success. A negative errno code is returned
10174 * on a failure to set up the device, or if the name is a duplicate.
10176 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10177 * and expands the device name if you passed a format string to
10180 int register_netdev(struct net_device *dev)
10184 if (rtnl_lock_killable())
10186 err = register_netdevice(dev);
10190 EXPORT_SYMBOL(register_netdev);
10192 int netdev_refcnt_read(const struct net_device *dev)
10194 #ifdef CONFIG_PCPU_DEV_REFCNT
10197 for_each_possible_cpu(i)
10198 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10201 return refcount_read(&dev->dev_refcnt);
10204 EXPORT_SYMBOL(netdev_refcnt_read);
10206 int netdev_unregister_timeout_secs __read_mostly = 10;
10208 #define WAIT_REFS_MIN_MSECS 1
10209 #define WAIT_REFS_MAX_MSECS 250
10211 * netdev_wait_allrefs_any - wait until all references are gone.
10212 * @list: list of net_devices to wait on
10214 * This is called when unregistering network devices.
10216 * Any protocol or device that holds a reference should register
10217 * for netdevice notification, and cleanup and put back the
10218 * reference if they receive an UNREGISTER event.
10219 * We can get stuck here if buggy protocols don't correctly
10222 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10224 unsigned long rebroadcast_time, warning_time;
10225 struct net_device *dev;
10228 rebroadcast_time = warning_time = jiffies;
10230 list_for_each_entry(dev, list, todo_list)
10231 if (netdev_refcnt_read(dev) == 1)
10235 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10238 /* Rebroadcast unregister notification */
10239 list_for_each_entry(dev, list, todo_list)
10240 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10246 list_for_each_entry(dev, list, todo_list)
10247 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10249 /* We must not have linkwatch events
10250 * pending on unregister. If this
10251 * happens, we simply run the queue
10252 * unscheduled, resulting in a noop
10255 linkwatch_run_queue();
10261 rebroadcast_time = jiffies;
10266 wait = WAIT_REFS_MIN_MSECS;
10269 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10272 list_for_each_entry(dev, list, todo_list)
10273 if (netdev_refcnt_read(dev) == 1)
10276 if (time_after(jiffies, warning_time +
10277 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10278 list_for_each_entry(dev, list, todo_list) {
10279 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10280 dev->name, netdev_refcnt_read(dev));
10281 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10284 warning_time = jiffies;
10289 /* The sequence is:
10293 * register_netdevice(x1);
10294 * register_netdevice(x2);
10296 * unregister_netdevice(y1);
10297 * unregister_netdevice(y2);
10303 * We are invoked by rtnl_unlock().
10304 * This allows us to deal with problems:
10305 * 1) We can delete sysfs objects which invoke hotplug
10306 * without deadlocking with linkwatch via keventd.
10307 * 2) Since we run with the RTNL semaphore not held, we can sleep
10308 * safely in order to wait for the netdev refcnt to drop to zero.
10310 * We must not return until all unregister events added during
10311 * the interval the lock was held have been completed.
10313 void netdev_run_todo(void)
10315 struct net_device *dev, *tmp;
10316 struct list_head list;
10317 #ifdef CONFIG_LOCKDEP
10318 struct list_head unlink_list;
10320 list_replace_init(&net_unlink_list, &unlink_list);
10322 while (!list_empty(&unlink_list)) {
10323 struct net_device *dev = list_first_entry(&unlink_list,
10326 list_del_init(&dev->unlink_list);
10327 dev->nested_level = dev->lower_level - 1;
10331 /* Snapshot list, allow later requests */
10332 list_replace_init(&net_todo_list, &list);
10336 /* Wait for rcu callbacks to finish before next phase */
10337 if (!list_empty(&list))
10340 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10341 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10342 netdev_WARN(dev, "run_todo but not unregistering\n");
10343 list_del(&dev->todo_list);
10347 write_lock(&dev_base_lock);
10348 dev->reg_state = NETREG_UNREGISTERED;
10349 write_unlock(&dev_base_lock);
10350 linkwatch_forget_dev(dev);
10353 while (!list_empty(&list)) {
10354 dev = netdev_wait_allrefs_any(&list);
10355 list_del(&dev->todo_list);
10358 BUG_ON(netdev_refcnt_read(dev) != 1);
10359 BUG_ON(!list_empty(&dev->ptype_all));
10360 BUG_ON(!list_empty(&dev->ptype_specific));
10361 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10362 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10364 if (dev->priv_destructor)
10365 dev->priv_destructor(dev);
10366 if (dev->needs_free_netdev)
10369 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10370 wake_up(&netdev_unregistering_wq);
10372 /* Free network device */
10373 kobject_put(&dev->dev.kobj);
10377 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10378 * all the same fields in the same order as net_device_stats, with only
10379 * the type differing, but rtnl_link_stats64 may have additional fields
10380 * at the end for newer counters.
10382 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10383 const struct net_device_stats *netdev_stats)
10385 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10386 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10387 u64 *dst = (u64 *)stats64;
10389 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10390 for (i = 0; i < n; i++)
10391 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10392 /* zero out counters that only exist in rtnl_link_stats64 */
10393 memset((char *)stats64 + n * sizeof(u64), 0,
10394 sizeof(*stats64) - n * sizeof(u64));
10396 EXPORT_SYMBOL(netdev_stats_to_stats64);
10398 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10400 struct net_device_core_stats __percpu *p;
10402 p = alloc_percpu_gfp(struct net_device_core_stats,
10403 GFP_ATOMIC | __GFP_NOWARN);
10405 if (p && cmpxchg(&dev->core_stats, NULL, p))
10408 /* This READ_ONCE() pairs with the cmpxchg() above */
10409 return READ_ONCE(dev->core_stats);
10411 EXPORT_SYMBOL(netdev_core_stats_alloc);
10414 * dev_get_stats - get network device statistics
10415 * @dev: device to get statistics from
10416 * @storage: place to store stats
10418 * Get network statistics from device. Return @storage.
10419 * The device driver may provide its own method by setting
10420 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10421 * otherwise the internal statistics structure is used.
10423 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10424 struct rtnl_link_stats64 *storage)
10426 const struct net_device_ops *ops = dev->netdev_ops;
10427 const struct net_device_core_stats __percpu *p;
10429 if (ops->ndo_get_stats64) {
10430 memset(storage, 0, sizeof(*storage));
10431 ops->ndo_get_stats64(dev, storage);
10432 } else if (ops->ndo_get_stats) {
10433 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10435 netdev_stats_to_stats64(storage, &dev->stats);
10438 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10439 p = READ_ONCE(dev->core_stats);
10441 const struct net_device_core_stats *core_stats;
10444 for_each_possible_cpu(i) {
10445 core_stats = per_cpu_ptr(p, i);
10446 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10447 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10448 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10449 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10454 EXPORT_SYMBOL(dev_get_stats);
10457 * dev_fetch_sw_netstats - get per-cpu network device statistics
10458 * @s: place to store stats
10459 * @netstats: per-cpu network stats to read from
10461 * Read per-cpu network statistics and populate the related fields in @s.
10463 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10464 const struct pcpu_sw_netstats __percpu *netstats)
10468 for_each_possible_cpu(cpu) {
10469 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10470 const struct pcpu_sw_netstats *stats;
10471 unsigned int start;
10473 stats = per_cpu_ptr(netstats, cpu);
10475 start = u64_stats_fetch_begin_irq(&stats->syncp);
10476 rx_packets = u64_stats_read(&stats->rx_packets);
10477 rx_bytes = u64_stats_read(&stats->rx_bytes);
10478 tx_packets = u64_stats_read(&stats->tx_packets);
10479 tx_bytes = u64_stats_read(&stats->tx_bytes);
10480 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10482 s->rx_packets += rx_packets;
10483 s->rx_bytes += rx_bytes;
10484 s->tx_packets += tx_packets;
10485 s->tx_bytes += tx_bytes;
10488 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10491 * dev_get_tstats64 - ndo_get_stats64 implementation
10492 * @dev: device to get statistics from
10493 * @s: place to store stats
10495 * Populate @s from dev->stats and dev->tstats. Can be used as
10496 * ndo_get_stats64() callback.
10498 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10500 netdev_stats_to_stats64(s, &dev->stats);
10501 dev_fetch_sw_netstats(s, dev->tstats);
10503 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10505 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10507 struct netdev_queue *queue = dev_ingress_queue(dev);
10509 #ifdef CONFIG_NET_CLS_ACT
10512 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10515 netdev_init_one_queue(dev, queue, NULL);
10516 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10517 queue->qdisc_sleeping = &noop_qdisc;
10518 rcu_assign_pointer(dev->ingress_queue, queue);
10523 static const struct ethtool_ops default_ethtool_ops;
10525 void netdev_set_default_ethtool_ops(struct net_device *dev,
10526 const struct ethtool_ops *ops)
10528 if (dev->ethtool_ops == &default_ethtool_ops)
10529 dev->ethtool_ops = ops;
10531 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10533 void netdev_freemem(struct net_device *dev)
10535 char *addr = (char *)dev - dev->padded;
10541 * alloc_netdev_mqs - allocate network device
10542 * @sizeof_priv: size of private data to allocate space for
10543 * @name: device name format string
10544 * @name_assign_type: origin of device name
10545 * @setup: callback to initialize device
10546 * @txqs: the number of TX subqueues to allocate
10547 * @rxqs: the number of RX subqueues to allocate
10549 * Allocates a struct net_device with private data area for driver use
10550 * and performs basic initialization. Also allocates subqueue structs
10551 * for each queue on the device.
10553 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10554 unsigned char name_assign_type,
10555 void (*setup)(struct net_device *),
10556 unsigned int txqs, unsigned int rxqs)
10558 struct net_device *dev;
10559 unsigned int alloc_size;
10560 struct net_device *p;
10562 BUG_ON(strlen(name) >= sizeof(dev->name));
10565 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10570 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10574 alloc_size = sizeof(struct net_device);
10576 /* ensure 32-byte alignment of private area */
10577 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10578 alloc_size += sizeof_priv;
10580 /* ensure 32-byte alignment of whole construct */
10581 alloc_size += NETDEV_ALIGN - 1;
10583 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10587 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10588 dev->padded = (char *)dev - (char *)p;
10590 ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10591 #ifdef CONFIG_PCPU_DEV_REFCNT
10592 dev->pcpu_refcnt = alloc_percpu(int);
10593 if (!dev->pcpu_refcnt)
10597 refcount_set(&dev->dev_refcnt, 1);
10600 if (dev_addr_init(dev))
10606 dev_net_set(dev, &init_net);
10608 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10609 dev->gso_max_segs = GSO_MAX_SEGS;
10610 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10611 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10612 dev->tso_max_segs = TSO_MAX_SEGS;
10613 dev->upper_level = 1;
10614 dev->lower_level = 1;
10615 #ifdef CONFIG_LOCKDEP
10616 dev->nested_level = 0;
10617 INIT_LIST_HEAD(&dev->unlink_list);
10620 INIT_LIST_HEAD(&dev->napi_list);
10621 INIT_LIST_HEAD(&dev->unreg_list);
10622 INIT_LIST_HEAD(&dev->close_list);
10623 INIT_LIST_HEAD(&dev->link_watch_list);
10624 INIT_LIST_HEAD(&dev->adj_list.upper);
10625 INIT_LIST_HEAD(&dev->adj_list.lower);
10626 INIT_LIST_HEAD(&dev->ptype_all);
10627 INIT_LIST_HEAD(&dev->ptype_specific);
10628 INIT_LIST_HEAD(&dev->net_notifier_list);
10629 #ifdef CONFIG_NET_SCHED
10630 hash_init(dev->qdisc_hash);
10632 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10635 if (!dev->tx_queue_len) {
10636 dev->priv_flags |= IFF_NO_QUEUE;
10637 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10640 dev->num_tx_queues = txqs;
10641 dev->real_num_tx_queues = txqs;
10642 if (netif_alloc_netdev_queues(dev))
10645 dev->num_rx_queues = rxqs;
10646 dev->real_num_rx_queues = rxqs;
10647 if (netif_alloc_rx_queues(dev))
10650 strcpy(dev->name, name);
10651 dev->name_assign_type = name_assign_type;
10652 dev->group = INIT_NETDEV_GROUP;
10653 if (!dev->ethtool_ops)
10654 dev->ethtool_ops = &default_ethtool_ops;
10656 nf_hook_netdev_init(dev);
10665 #ifdef CONFIG_PCPU_DEV_REFCNT
10666 free_percpu(dev->pcpu_refcnt);
10669 netdev_freemem(dev);
10672 EXPORT_SYMBOL(alloc_netdev_mqs);
10675 * free_netdev - free network device
10678 * This function does the last stage of destroying an allocated device
10679 * interface. The reference to the device object is released. If this
10680 * is the last reference then it will be freed.Must be called in process
10683 void free_netdev(struct net_device *dev)
10685 struct napi_struct *p, *n;
10689 /* When called immediately after register_netdevice() failed the unwind
10690 * handling may still be dismantling the device. Handle that case by
10691 * deferring the free.
10693 if (dev->reg_state == NETREG_UNREGISTERING) {
10695 dev->needs_free_netdev = true;
10699 netif_free_tx_queues(dev);
10700 netif_free_rx_queues(dev);
10702 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10704 /* Flush device addresses */
10705 dev_addr_flush(dev);
10707 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10710 ref_tracker_dir_exit(&dev->refcnt_tracker);
10711 #ifdef CONFIG_PCPU_DEV_REFCNT
10712 free_percpu(dev->pcpu_refcnt);
10713 dev->pcpu_refcnt = NULL;
10715 free_percpu(dev->core_stats);
10716 dev->core_stats = NULL;
10717 free_percpu(dev->xdp_bulkq);
10718 dev->xdp_bulkq = NULL;
10720 /* Compatibility with error handling in drivers */
10721 if (dev->reg_state == NETREG_UNINITIALIZED) {
10722 netdev_freemem(dev);
10726 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10727 dev->reg_state = NETREG_RELEASED;
10729 /* will free via device release */
10730 put_device(&dev->dev);
10732 EXPORT_SYMBOL(free_netdev);
10735 * synchronize_net - Synchronize with packet receive processing
10737 * Wait for packets currently being received to be done.
10738 * Does not block later packets from starting.
10740 void synchronize_net(void)
10743 if (rtnl_is_locked())
10744 synchronize_rcu_expedited();
10748 EXPORT_SYMBOL(synchronize_net);
10751 * unregister_netdevice_queue - remove device from the kernel
10755 * This function shuts down a device interface and removes it
10756 * from the kernel tables.
10757 * If head not NULL, device is queued to be unregistered later.
10759 * Callers must hold the rtnl semaphore. You may want
10760 * unregister_netdev() instead of this.
10763 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10768 list_move_tail(&dev->unreg_list, head);
10772 list_add(&dev->unreg_list, &single);
10773 unregister_netdevice_many(&single);
10776 EXPORT_SYMBOL(unregister_netdevice_queue);
10779 * unregister_netdevice_many - unregister many devices
10780 * @head: list of devices
10782 * Note: As most callers use a stack allocated list_head,
10783 * we force a list_del() to make sure stack wont be corrupted later.
10785 void unregister_netdevice_many(struct list_head *head)
10787 struct net_device *dev, *tmp;
10788 LIST_HEAD(close_head);
10790 BUG_ON(dev_boot_phase);
10793 if (list_empty(head))
10796 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10797 /* Some devices call without registering
10798 * for initialization unwind. Remove those
10799 * devices and proceed with the remaining.
10801 if (dev->reg_state == NETREG_UNINITIALIZED) {
10802 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10806 list_del(&dev->unreg_list);
10809 dev->dismantle = true;
10810 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10813 /* If device is running, close it first. */
10814 list_for_each_entry(dev, head, unreg_list)
10815 list_add_tail(&dev->close_list, &close_head);
10816 dev_close_many(&close_head, true);
10818 list_for_each_entry(dev, head, unreg_list) {
10819 /* And unlink it from device chain. */
10820 write_lock(&dev_base_lock);
10821 unlist_netdevice(dev, false);
10822 dev->reg_state = NETREG_UNREGISTERING;
10823 write_unlock(&dev_base_lock);
10825 flush_all_backlogs();
10829 list_for_each_entry(dev, head, unreg_list) {
10830 struct sk_buff *skb = NULL;
10832 /* Shutdown queueing discipline. */
10835 dev_xdp_uninstall(dev);
10837 netdev_offload_xstats_disable_all(dev);
10839 /* Notify protocols, that we are about to destroy
10840 * this device. They should clean all the things.
10842 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10844 if (!dev->rtnl_link_ops ||
10845 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10846 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10847 GFP_KERNEL, NULL, 0);
10850 * Flush the unicast and multicast chains
10855 netdev_name_node_alt_flush(dev);
10856 netdev_name_node_free(dev->name_node);
10858 if (dev->netdev_ops->ndo_uninit)
10859 dev->netdev_ops->ndo_uninit(dev);
10862 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10864 /* Notifier chain MUST detach us all upper devices. */
10865 WARN_ON(netdev_has_any_upper_dev(dev));
10866 WARN_ON(netdev_has_any_lower_dev(dev));
10868 /* Remove entries from kobject tree */
10869 netdev_unregister_kobject(dev);
10871 /* Remove XPS queueing entries */
10872 netif_reset_xps_queues_gt(dev, 0);
10878 list_for_each_entry(dev, head, unreg_list) {
10879 netdev_put(dev, &dev->dev_registered_tracker);
10885 EXPORT_SYMBOL(unregister_netdevice_many);
10888 * unregister_netdev - remove device from the kernel
10891 * This function shuts down a device interface and removes it
10892 * from the kernel tables.
10894 * This is just a wrapper for unregister_netdevice that takes
10895 * the rtnl semaphore. In general you want to use this and not
10896 * unregister_netdevice.
10898 void unregister_netdev(struct net_device *dev)
10901 unregister_netdevice(dev);
10904 EXPORT_SYMBOL(unregister_netdev);
10907 * __dev_change_net_namespace - move device to different nethost namespace
10909 * @net: network namespace
10910 * @pat: If not NULL name pattern to try if the current device name
10911 * is already taken in the destination network namespace.
10912 * @new_ifindex: If not zero, specifies device index in the target
10915 * This function shuts down a device interface and moves it
10916 * to a new network namespace. On success 0 is returned, on
10917 * a failure a netagive errno code is returned.
10919 * Callers must hold the rtnl semaphore.
10922 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10923 const char *pat, int new_ifindex)
10925 struct net *net_old = dev_net(dev);
10930 /* Don't allow namespace local devices to be moved. */
10932 if (dev->features & NETIF_F_NETNS_LOCAL)
10935 /* Ensure the device has been registrered */
10936 if (dev->reg_state != NETREG_REGISTERED)
10939 /* Get out if there is nothing todo */
10941 if (net_eq(net_old, net))
10944 /* Pick the destination device name, and ensure
10945 * we can use it in the destination network namespace.
10948 if (netdev_name_in_use(net, dev->name)) {
10949 /* We get here if we can't use the current device name */
10952 err = dev_get_valid_name(net, dev, pat);
10957 /* Check that new_ifindex isn't used yet. */
10959 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10963 * And now a mini version of register_netdevice unregister_netdevice.
10966 /* If device is running close it first. */
10969 /* And unlink it from device chain */
10970 unlist_netdevice(dev, true);
10974 /* Shutdown queueing discipline. */
10977 /* Notify protocols, that we are about to destroy
10978 * this device. They should clean all the things.
10980 * Note that dev->reg_state stays at NETREG_REGISTERED.
10981 * This is wanted because this way 8021q and macvlan know
10982 * the device is just moving and can keep their slaves up.
10984 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10987 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10988 /* If there is an ifindex conflict assign a new one */
10989 if (!new_ifindex) {
10990 if (__dev_get_by_index(net, dev->ifindex))
10991 new_ifindex = dev_new_index(net);
10993 new_ifindex = dev->ifindex;
10996 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11000 * Flush the unicast and multicast chains
11005 /* Send a netdev-removed uevent to the old namespace */
11006 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11007 netdev_adjacent_del_links(dev);
11009 /* Move per-net netdevice notifiers that are following the netdevice */
11010 move_netdevice_notifiers_dev_net(dev, net);
11012 /* Actually switch the network namespace */
11013 dev_net_set(dev, net);
11014 dev->ifindex = new_ifindex;
11016 /* Send a netdev-add uevent to the new namespace */
11017 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11018 netdev_adjacent_add_links(dev);
11020 /* Fixup kobjects */
11021 err = device_rename(&dev->dev, dev->name);
11024 /* Adapt owner in case owning user namespace of target network
11025 * namespace is different from the original one.
11027 err = netdev_change_owner(dev, net_old, net);
11030 /* Add the device back in the hashes */
11031 list_netdevice(dev);
11033 /* Notify protocols, that a new device appeared. */
11034 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11037 * Prevent userspace races by waiting until the network
11038 * device is fully setup before sending notifications.
11040 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11047 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11049 static int dev_cpu_dead(unsigned int oldcpu)
11051 struct sk_buff **list_skb;
11052 struct sk_buff *skb;
11054 struct softnet_data *sd, *oldsd, *remsd = NULL;
11056 local_irq_disable();
11057 cpu = smp_processor_id();
11058 sd = &per_cpu(softnet_data, cpu);
11059 oldsd = &per_cpu(softnet_data, oldcpu);
11061 /* Find end of our completion_queue. */
11062 list_skb = &sd->completion_queue;
11064 list_skb = &(*list_skb)->next;
11065 /* Append completion queue from offline CPU. */
11066 *list_skb = oldsd->completion_queue;
11067 oldsd->completion_queue = NULL;
11069 /* Append output queue from offline CPU. */
11070 if (oldsd->output_queue) {
11071 *sd->output_queue_tailp = oldsd->output_queue;
11072 sd->output_queue_tailp = oldsd->output_queue_tailp;
11073 oldsd->output_queue = NULL;
11074 oldsd->output_queue_tailp = &oldsd->output_queue;
11076 /* Append NAPI poll list from offline CPU, with one exception :
11077 * process_backlog() must be called by cpu owning percpu backlog.
11078 * We properly handle process_queue & input_pkt_queue later.
11080 while (!list_empty(&oldsd->poll_list)) {
11081 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11082 struct napi_struct,
11085 list_del_init(&napi->poll_list);
11086 if (napi->poll == process_backlog)
11089 ____napi_schedule(sd, napi);
11092 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11093 local_irq_enable();
11096 remsd = oldsd->rps_ipi_list;
11097 oldsd->rps_ipi_list = NULL;
11099 /* send out pending IPI's on offline CPU */
11100 net_rps_send_ipi(remsd);
11102 /* Process offline CPU's input_pkt_queue */
11103 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11105 input_queue_head_incr(oldsd);
11107 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11109 input_queue_head_incr(oldsd);
11116 * netdev_increment_features - increment feature set by one
11117 * @all: current feature set
11118 * @one: new feature set
11119 * @mask: mask feature set
11121 * Computes a new feature set after adding a device with feature set
11122 * @one to the master device with current feature set @all. Will not
11123 * enable anything that is off in @mask. Returns the new feature set.
11125 netdev_features_t netdev_increment_features(netdev_features_t all,
11126 netdev_features_t one, netdev_features_t mask)
11128 if (mask & NETIF_F_HW_CSUM)
11129 mask |= NETIF_F_CSUM_MASK;
11130 mask |= NETIF_F_VLAN_CHALLENGED;
11132 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11133 all &= one | ~NETIF_F_ALL_FOR_ALL;
11135 /* If one device supports hw checksumming, set for all. */
11136 if (all & NETIF_F_HW_CSUM)
11137 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11141 EXPORT_SYMBOL(netdev_increment_features);
11143 static struct hlist_head * __net_init netdev_create_hash(void)
11146 struct hlist_head *hash;
11148 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11150 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11151 INIT_HLIST_HEAD(&hash[i]);
11156 /* Initialize per network namespace state */
11157 static int __net_init netdev_init(struct net *net)
11159 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11160 8 * sizeof_field(struct napi_struct, gro_bitmask));
11162 INIT_LIST_HEAD(&net->dev_base_head);
11164 net->dev_name_head = netdev_create_hash();
11165 if (net->dev_name_head == NULL)
11168 net->dev_index_head = netdev_create_hash();
11169 if (net->dev_index_head == NULL)
11172 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11177 kfree(net->dev_name_head);
11183 * netdev_drivername - network driver for the device
11184 * @dev: network device
11186 * Determine network driver for device.
11188 const char *netdev_drivername(const struct net_device *dev)
11190 const struct device_driver *driver;
11191 const struct device *parent;
11192 const char *empty = "";
11194 parent = dev->dev.parent;
11198 driver = parent->driver;
11199 if (driver && driver->name)
11200 return driver->name;
11204 static void __netdev_printk(const char *level, const struct net_device *dev,
11205 struct va_format *vaf)
11207 if (dev && dev->dev.parent) {
11208 dev_printk_emit(level[1] - '0',
11211 dev_driver_string(dev->dev.parent),
11212 dev_name(dev->dev.parent),
11213 netdev_name(dev), netdev_reg_state(dev),
11216 printk("%s%s%s: %pV",
11217 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11219 printk("%s(NULL net_device): %pV", level, vaf);
11223 void netdev_printk(const char *level, const struct net_device *dev,
11224 const char *format, ...)
11226 struct va_format vaf;
11229 va_start(args, format);
11234 __netdev_printk(level, dev, &vaf);
11238 EXPORT_SYMBOL(netdev_printk);
11240 #define define_netdev_printk_level(func, level) \
11241 void func(const struct net_device *dev, const char *fmt, ...) \
11243 struct va_format vaf; \
11246 va_start(args, fmt); \
11251 __netdev_printk(level, dev, &vaf); \
11255 EXPORT_SYMBOL(func);
11257 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11258 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11259 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11260 define_netdev_printk_level(netdev_err, KERN_ERR);
11261 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11262 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11263 define_netdev_printk_level(netdev_info, KERN_INFO);
11265 static void __net_exit netdev_exit(struct net *net)
11267 kfree(net->dev_name_head);
11268 kfree(net->dev_index_head);
11269 if (net != &init_net)
11270 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11273 static struct pernet_operations __net_initdata netdev_net_ops = {
11274 .init = netdev_init,
11275 .exit = netdev_exit,
11278 static void __net_exit default_device_exit_net(struct net *net)
11280 struct net_device *dev, *aux;
11282 * Push all migratable network devices back to the
11283 * initial network namespace
11286 for_each_netdev_safe(net, dev, aux) {
11288 char fb_name[IFNAMSIZ];
11290 /* Ignore unmoveable devices (i.e. loopback) */
11291 if (dev->features & NETIF_F_NETNS_LOCAL)
11294 /* Leave virtual devices for the generic cleanup */
11295 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11298 /* Push remaining network devices to init_net */
11299 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11300 if (netdev_name_in_use(&init_net, fb_name))
11301 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11302 err = dev_change_net_namespace(dev, &init_net, fb_name);
11304 pr_emerg("%s: failed to move %s to init_net: %d\n",
11305 __func__, dev->name, err);
11311 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11313 /* At exit all network devices most be removed from a network
11314 * namespace. Do this in the reverse order of registration.
11315 * Do this across as many network namespaces as possible to
11316 * improve batching efficiency.
11318 struct net_device *dev;
11320 LIST_HEAD(dev_kill_list);
11323 list_for_each_entry(net, net_list, exit_list) {
11324 default_device_exit_net(net);
11328 list_for_each_entry(net, net_list, exit_list) {
11329 for_each_netdev_reverse(net, dev) {
11330 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11331 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11333 unregister_netdevice_queue(dev, &dev_kill_list);
11336 unregister_netdevice_many(&dev_kill_list);
11340 static struct pernet_operations __net_initdata default_device_ops = {
11341 .exit_batch = default_device_exit_batch,
11345 * Initialize the DEV module. At boot time this walks the device list and
11346 * unhooks any devices that fail to initialise (normally hardware not
11347 * present) and leaves us with a valid list of present and active devices.
11352 * This is called single threaded during boot, so no need
11353 * to take the rtnl semaphore.
11355 static int __init net_dev_init(void)
11357 int i, rc = -ENOMEM;
11359 BUG_ON(!dev_boot_phase);
11361 if (dev_proc_init())
11364 if (netdev_kobject_init())
11367 INIT_LIST_HEAD(&ptype_all);
11368 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11369 INIT_LIST_HEAD(&ptype_base[i]);
11371 if (register_pernet_subsys(&netdev_net_ops))
11375 * Initialise the packet receive queues.
11378 for_each_possible_cpu(i) {
11379 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11380 struct softnet_data *sd = &per_cpu(softnet_data, i);
11382 INIT_WORK(flush, flush_backlog);
11384 skb_queue_head_init(&sd->input_pkt_queue);
11385 skb_queue_head_init(&sd->process_queue);
11386 #ifdef CONFIG_XFRM_OFFLOAD
11387 skb_queue_head_init(&sd->xfrm_backlog);
11389 INIT_LIST_HEAD(&sd->poll_list);
11390 sd->output_queue_tailp = &sd->output_queue;
11392 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11395 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11396 spin_lock_init(&sd->defer_lock);
11398 init_gro_hash(&sd->backlog);
11399 sd->backlog.poll = process_backlog;
11400 sd->backlog.weight = weight_p;
11403 dev_boot_phase = 0;
11405 /* The loopback device is special if any other network devices
11406 * is present in a network namespace the loopback device must
11407 * be present. Since we now dynamically allocate and free the
11408 * loopback device ensure this invariant is maintained by
11409 * keeping the loopback device as the first device on the
11410 * list of network devices. Ensuring the loopback devices
11411 * is the first device that appears and the last network device
11414 if (register_pernet_device(&loopback_net_ops))
11417 if (register_pernet_device(&default_device_ops))
11420 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11421 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11423 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11424 NULL, dev_cpu_dead);
11431 subsys_initcall(net_dev_init);