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>
154 #include "net-sysfs.h"
157 static DEFINE_SPINLOCK(ptype_lock);
158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
159 struct list_head ptype_all __read_mostly; /* Taps */
161 static int netif_rx_internal(struct sk_buff *skb);
162 static int call_netdevice_notifiers_info(unsigned long val,
163 struct netdev_notifier_info *info);
164 static int call_netdevice_notifiers_extack(unsigned long val,
165 struct net_device *dev,
166 struct netlink_ext_ack *extack);
167 static struct napi_struct *napi_by_id(unsigned int napi_id);
170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
175 * Writers must hold the rtnl semaphore while they loop through the
176 * dev_base_head list, and hold dev_base_lock for writing when they do the
177 * actual updates. This allows pure readers to access the list even
178 * while a writer is preparing to update it.
180 * To put it another way, dev_base_lock is held for writing only to
181 * protect against pure readers; the rtnl semaphore provides the
182 * protection against other writers.
184 * See, for example usages, register_netdevice() and
185 * unregister_netdevice(), which must be called with the rtnl
188 DEFINE_RWLOCK(dev_base_lock);
189 EXPORT_SYMBOL(dev_base_lock);
191 static DEFINE_MUTEX(ifalias_mutex);
193 /* protects napi_hash addition/deletion and napi_gen_id */
194 static DEFINE_SPINLOCK(napi_hash_lock);
196 static unsigned int napi_gen_id = NR_CPUS;
197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
199 static DECLARE_RWSEM(devnet_rename_sem);
201 static inline void dev_base_seq_inc(struct net *net)
203 while (++net->dev_base_seq == 0)
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
219 static inline void rps_lock(struct softnet_data *sd)
222 spin_lock(&sd->input_pkt_queue.lock);
226 static inline void rps_unlock(struct softnet_data *sd)
229 spin_unlock(&sd->input_pkt_queue.lock);
233 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
236 struct netdev_name_node *name_node;
238 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
241 INIT_HLIST_NODE(&name_node->hlist);
242 name_node->dev = dev;
243 name_node->name = name;
247 static struct netdev_name_node *
248 netdev_name_node_head_alloc(struct net_device *dev)
250 struct netdev_name_node *name_node;
252 name_node = netdev_name_node_alloc(dev, dev->name);
255 INIT_LIST_HEAD(&name_node->list);
259 static void netdev_name_node_free(struct netdev_name_node *name_node)
264 static void netdev_name_node_add(struct net *net,
265 struct netdev_name_node *name_node)
267 hlist_add_head_rcu(&name_node->hlist,
268 dev_name_hash(net, name_node->name));
271 static void netdev_name_node_del(struct netdev_name_node *name_node)
273 hlist_del_rcu(&name_node->hlist);
276 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
279 struct hlist_head *head = dev_name_hash(net, name);
280 struct netdev_name_node *name_node;
282 hlist_for_each_entry(name_node, head, hlist)
283 if (!strcmp(name_node->name, name))
288 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
291 struct hlist_head *head = dev_name_hash(net, name);
292 struct netdev_name_node *name_node;
294 hlist_for_each_entry_rcu(name_node, head, hlist)
295 if (!strcmp(name_node->name, name))
300 bool netdev_name_in_use(struct net *net, const char *name)
302 return netdev_name_node_lookup(net, name);
304 EXPORT_SYMBOL(netdev_name_in_use);
306 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
308 struct netdev_name_node *name_node;
309 struct net *net = dev_net(dev);
311 name_node = netdev_name_node_lookup(net, name);
314 name_node = netdev_name_node_alloc(dev, name);
317 netdev_name_node_add(net, name_node);
318 /* The node that holds dev->name acts as a head of per-device list. */
319 list_add_tail(&name_node->list, &dev->name_node->list);
323 EXPORT_SYMBOL(netdev_name_node_alt_create);
325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
327 list_del(&name_node->list);
328 netdev_name_node_del(name_node);
329 kfree(name_node->name);
330 netdev_name_node_free(name_node);
333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
335 struct netdev_name_node *name_node;
336 struct net *net = dev_net(dev);
338 name_node = netdev_name_node_lookup(net, name);
341 /* lookup might have found our primary name or a name belonging
344 if (name_node == dev->name_node || name_node->dev != dev)
347 __netdev_name_node_alt_destroy(name_node);
351 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
353 static void netdev_name_node_alt_flush(struct net_device *dev)
355 struct netdev_name_node *name_node, *tmp;
357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
358 __netdev_name_node_alt_destroy(name_node);
361 /* Device list insertion */
362 static void list_netdevice(struct net_device *dev)
364 struct net *net = dev_net(dev);
368 write_lock_bh(&dev_base_lock);
369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
370 netdev_name_node_add(net, dev->name_node);
371 hlist_add_head_rcu(&dev->index_hlist,
372 dev_index_hash(net, dev->ifindex));
373 write_unlock_bh(&dev_base_lock);
375 dev_base_seq_inc(net);
378 /* Device list removal
379 * caller must respect a RCU grace period before freeing/reusing dev
381 static void unlist_netdevice(struct net_device *dev)
385 /* Unlink dev from the device chain */
386 write_lock_bh(&dev_base_lock);
387 list_del_rcu(&dev->dev_list);
388 netdev_name_node_del(dev->name_node);
389 hlist_del_rcu(&dev->index_hlist);
390 write_unlock_bh(&dev_base_lock);
392 dev_base_seq_inc(dev_net(dev));
399 static RAW_NOTIFIER_HEAD(netdev_chain);
402 * Device drivers call our routines to queue packets here. We empty the
403 * queue in the local softnet handler.
406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
407 EXPORT_PER_CPU_SYMBOL(softnet_data);
409 #ifdef CONFIG_LOCKDEP
411 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
412 * according to dev->type
414 static const unsigned short netdev_lock_type[] = {
415 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
416 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
417 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
418 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
419 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
420 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
421 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
422 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
423 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
424 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
425 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
426 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
427 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
428 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
429 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
431 static const char *const netdev_lock_name[] = {
432 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
433 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
434 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
435 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
436 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
437 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
438 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
439 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
440 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
441 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
442 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
443 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
444 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
445 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
446 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
451 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
455 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
456 if (netdev_lock_type[i] == dev_type)
458 /* the last key is used by default */
459 return ARRAY_SIZE(netdev_lock_type) - 1;
462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
463 unsigned short dev_type)
467 i = netdev_lock_pos(dev_type);
468 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
469 netdev_lock_name[i]);
472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
476 i = netdev_lock_pos(dev->type);
477 lockdep_set_class_and_name(&dev->addr_list_lock,
478 &netdev_addr_lock_key[i],
479 netdev_lock_name[i]);
482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
483 unsigned short dev_type)
487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
492 /*******************************************************************************
494 * Protocol management and registration routines
496 *******************************************************************************/
500 * Add a protocol ID to the list. Now that the input handler is
501 * smarter we can dispense with all the messy stuff that used to be
504 * BEWARE!!! Protocol handlers, mangling input packets,
505 * MUST BE last in hash buckets and checking protocol handlers
506 * MUST start from promiscuous ptype_all chain in net_bh.
507 * It is true now, do not change it.
508 * Explanation follows: if protocol handler, mangling packet, will
509 * be the first on list, it is not able to sense, that packet
510 * is cloned and should be copied-on-write, so that it will
511 * change it and subsequent readers will get broken packet.
515 static inline struct list_head *ptype_head(const struct packet_type *pt)
517 if (pt->type == htons(ETH_P_ALL))
518 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
520 return pt->dev ? &pt->dev->ptype_specific :
521 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
525 * dev_add_pack - add packet handler
526 * @pt: packet type declaration
528 * Add a protocol handler to the networking stack. The passed &packet_type
529 * is linked into kernel lists and may not be freed until it has been
530 * removed from the kernel lists.
532 * This call does not sleep therefore it can not
533 * guarantee all CPU's that are in middle of receiving packets
534 * will see the new packet type (until the next received packet).
537 void dev_add_pack(struct packet_type *pt)
539 struct list_head *head = ptype_head(pt);
541 spin_lock(&ptype_lock);
542 list_add_rcu(&pt->list, head);
543 spin_unlock(&ptype_lock);
545 EXPORT_SYMBOL(dev_add_pack);
548 * __dev_remove_pack - remove packet handler
549 * @pt: packet type declaration
551 * Remove a protocol handler that was previously added to the kernel
552 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
553 * from the kernel lists and can be freed or reused once this function
556 * The packet type might still be in use by receivers
557 * and must not be freed until after all the CPU's have gone
558 * through a quiescent state.
560 void __dev_remove_pack(struct packet_type *pt)
562 struct list_head *head = ptype_head(pt);
563 struct packet_type *pt1;
565 spin_lock(&ptype_lock);
567 list_for_each_entry(pt1, head, list) {
569 list_del_rcu(&pt->list);
574 pr_warn("dev_remove_pack: %p not found\n", pt);
576 spin_unlock(&ptype_lock);
578 EXPORT_SYMBOL(__dev_remove_pack);
581 * dev_remove_pack - remove packet handler
582 * @pt: packet type declaration
584 * Remove a protocol handler that was previously added to the kernel
585 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
586 * from the kernel lists and can be freed or reused once this function
589 * This call sleeps to guarantee that no CPU is looking at the packet
592 void dev_remove_pack(struct packet_type *pt)
594 __dev_remove_pack(pt);
598 EXPORT_SYMBOL(dev_remove_pack);
601 /*******************************************************************************
603 * Device Interface Subroutines
605 *******************************************************************************/
608 * dev_get_iflink - get 'iflink' value of a interface
609 * @dev: targeted interface
611 * Indicates the ifindex the interface is linked to.
612 * Physical interfaces have the same 'ifindex' and 'iflink' values.
615 int dev_get_iflink(const struct net_device *dev)
617 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
618 return dev->netdev_ops->ndo_get_iflink(dev);
622 EXPORT_SYMBOL(dev_get_iflink);
625 * dev_fill_metadata_dst - Retrieve tunnel egress information.
626 * @dev: targeted interface
629 * For better visibility of tunnel traffic OVS needs to retrieve
630 * egress tunnel information for a packet. Following API allows
631 * user to get this info.
633 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
635 struct ip_tunnel_info *info;
637 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
640 info = skb_tunnel_info_unclone(skb);
643 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
646 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
648 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
650 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
652 int k = stack->num_paths++;
654 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
657 return &stack->path[k];
660 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
661 struct net_device_path_stack *stack)
663 const struct net_device *last_dev;
664 struct net_device_path_ctx ctx = {
668 struct net_device_path *path;
671 stack->num_paths = 0;
672 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
674 path = dev_fwd_path(stack);
678 memset(path, 0, sizeof(struct net_device_path));
679 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
683 if (WARN_ON_ONCE(last_dev == ctx.dev))
686 path = dev_fwd_path(stack);
689 path->type = DEV_PATH_ETHERNET;
694 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
697 * __dev_get_by_name - find a device by its name
698 * @net: the applicable net namespace
699 * @name: name to find
701 * Find an interface by name. Must be called under RTNL semaphore
702 * or @dev_base_lock. If the name is found a pointer to the device
703 * is returned. If the name is not found then %NULL is returned. The
704 * reference counters are not incremented so the caller must be
705 * careful with locks.
708 struct net_device *__dev_get_by_name(struct net *net, const char *name)
710 struct netdev_name_node *node_name;
712 node_name = netdev_name_node_lookup(net, name);
713 return node_name ? node_name->dev : NULL;
715 EXPORT_SYMBOL(__dev_get_by_name);
718 * dev_get_by_name_rcu - find a device by its name
719 * @net: the applicable net namespace
720 * @name: name to find
722 * Find an interface by name.
723 * If the name is found a pointer to the device is returned.
724 * If the name is not found then %NULL is returned.
725 * The reference counters are not incremented so the caller must be
726 * careful with locks. The caller must hold RCU lock.
729 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
731 struct netdev_name_node *node_name;
733 node_name = netdev_name_node_lookup_rcu(net, name);
734 return node_name ? node_name->dev : NULL;
736 EXPORT_SYMBOL(dev_get_by_name_rcu);
739 * dev_get_by_name - find a device by its name
740 * @net: the applicable net namespace
741 * @name: name to find
743 * Find an interface by name. This can be called from any
744 * context and does its own locking. The returned handle has
745 * the usage count incremented and the caller must use dev_put() to
746 * release it when it is no longer needed. %NULL is returned if no
747 * matching device is found.
750 struct net_device *dev_get_by_name(struct net *net, const char *name)
752 struct net_device *dev;
755 dev = dev_get_by_name_rcu(net, name);
760 EXPORT_SYMBOL(dev_get_by_name);
763 * __dev_get_by_index - find a device by its ifindex
764 * @net: the applicable net namespace
765 * @ifindex: index of device
767 * Search for an interface by index. Returns %NULL if the device
768 * is not found or a pointer to the device. The device has not
769 * had its reference counter increased so the caller must be careful
770 * about locking. The caller must hold either the RTNL semaphore
774 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
776 struct net_device *dev;
777 struct hlist_head *head = dev_index_hash(net, ifindex);
779 hlist_for_each_entry(dev, head, index_hlist)
780 if (dev->ifindex == ifindex)
785 EXPORT_SYMBOL(__dev_get_by_index);
788 * dev_get_by_index_rcu - 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 RCU lock.
798 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
800 struct net_device *dev;
801 struct hlist_head *head = dev_index_hash(net, ifindex);
803 hlist_for_each_entry_rcu(dev, head, index_hlist)
804 if (dev->ifindex == ifindex)
809 EXPORT_SYMBOL(dev_get_by_index_rcu);
813 * dev_get_by_index - 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 returned has
819 * had a reference added and the pointer is safe until the user calls
820 * dev_put to indicate they have finished with it.
823 struct net_device *dev_get_by_index(struct net *net, int ifindex)
825 struct net_device *dev;
828 dev = dev_get_by_index_rcu(net, ifindex);
833 EXPORT_SYMBOL(dev_get_by_index);
836 * dev_get_by_napi_id - find a device by napi_id
837 * @napi_id: ID of the NAPI struct
839 * Search for an interface by NAPI ID. Returns %NULL if the device
840 * is not found or a pointer to the device. The device has not had
841 * its reference counter increased so the caller must be careful
842 * about locking. The caller must hold RCU lock.
845 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
847 struct napi_struct *napi;
849 WARN_ON_ONCE(!rcu_read_lock_held());
851 if (napi_id < MIN_NAPI_ID)
854 napi = napi_by_id(napi_id);
856 return napi ? napi->dev : NULL;
858 EXPORT_SYMBOL(dev_get_by_napi_id);
861 * netdev_get_name - get a netdevice name, knowing its ifindex.
862 * @net: network namespace
863 * @name: a pointer to the buffer where the name will be stored.
864 * @ifindex: the ifindex of the interface to get the name from.
866 int netdev_get_name(struct net *net, char *name, int ifindex)
868 struct net_device *dev;
871 down_read(&devnet_rename_sem);
874 dev = dev_get_by_index_rcu(net, ifindex);
880 strcpy(name, dev->name);
885 up_read(&devnet_rename_sem);
890 * dev_getbyhwaddr_rcu - find a device by its hardware address
891 * @net: the applicable net namespace
892 * @type: media type of device
893 * @ha: hardware address
895 * Search for an interface by MAC address. Returns NULL if the device
896 * is not found or a pointer to the device.
897 * The caller must hold RCU or RTNL.
898 * The returned device has not had its ref count increased
899 * and the caller must therefore be careful about locking
903 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
906 struct net_device *dev;
908 for_each_netdev_rcu(net, dev)
909 if (dev->type == type &&
910 !memcmp(dev->dev_addr, ha, dev->addr_len))
915 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
917 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
919 struct net_device *dev, *ret = NULL;
922 for_each_netdev_rcu(net, dev)
923 if (dev->type == type) {
931 EXPORT_SYMBOL(dev_getfirstbyhwtype);
934 * __dev_get_by_flags - find any device with given flags
935 * @net: the applicable net namespace
936 * @if_flags: IFF_* values
937 * @mask: bitmask of bits in if_flags to check
939 * Search for any interface with the given flags. Returns NULL if a device
940 * is not found or a pointer to the device. Must be called inside
941 * rtnl_lock(), and result refcount is unchanged.
944 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
947 struct net_device *dev, *ret;
952 for_each_netdev(net, dev) {
953 if (((dev->flags ^ if_flags) & mask) == 0) {
960 EXPORT_SYMBOL(__dev_get_by_flags);
963 * dev_valid_name - check if name is okay for network device
966 * Network device names need to be valid file names to
967 * allow sysfs to work. We also disallow any kind of
970 bool dev_valid_name(const char *name)
974 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
976 if (!strcmp(name, ".") || !strcmp(name, ".."))
980 if (*name == '/' || *name == ':' || isspace(*name))
986 EXPORT_SYMBOL(dev_valid_name);
989 * __dev_alloc_name - allocate a name for a device
990 * @net: network namespace to allocate the device name in
991 * @name: name format string
992 * @buf: scratch buffer and result name string
994 * Passed a format string - eg "lt%d" it will try and find a suitable
995 * id. It scans list of devices to build up a free map, then chooses
996 * the first empty slot. The caller must hold the dev_base or rtnl lock
997 * while allocating the name and adding the device in order to avoid
999 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1000 * Returns the number of the unit assigned or a negative errno code.
1003 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1007 const int max_netdevices = 8*PAGE_SIZE;
1008 unsigned long *inuse;
1009 struct net_device *d;
1011 if (!dev_valid_name(name))
1014 p = strchr(name, '%');
1017 * Verify the string as this thing may have come from
1018 * the user. There must be either one "%d" and no other "%"
1021 if (p[1] != 'd' || strchr(p + 2, '%'))
1024 /* Use one page as a bit array of possible slots */
1025 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1029 for_each_netdev(net, d) {
1030 struct netdev_name_node *name_node;
1031 list_for_each_entry(name_node, &d->name_node->list, list) {
1032 if (!sscanf(name_node->name, name, &i))
1034 if (i < 0 || i >= max_netdevices)
1037 /* avoid cases where sscanf is not exact inverse of printf */
1038 snprintf(buf, IFNAMSIZ, name, i);
1039 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1042 if (!sscanf(d->name, name, &i))
1044 if (i < 0 || i >= max_netdevices)
1047 /* avoid cases where sscanf is not exact inverse of printf */
1048 snprintf(buf, IFNAMSIZ, name, i);
1049 if (!strncmp(buf, d->name, IFNAMSIZ))
1053 i = find_first_zero_bit(inuse, max_netdevices);
1054 free_page((unsigned long) inuse);
1057 snprintf(buf, IFNAMSIZ, name, i);
1058 if (!netdev_name_in_use(net, buf))
1061 /* It is possible to run out of possible slots
1062 * when the name is long and there isn't enough space left
1063 * for the digits, or if all bits are used.
1068 static int dev_alloc_name_ns(struct net *net,
1069 struct net_device *dev,
1076 ret = __dev_alloc_name(net, name, buf);
1078 strlcpy(dev->name, buf, IFNAMSIZ);
1083 * dev_alloc_name - allocate a name for a device
1085 * @name: name format string
1087 * Passed a format string - eg "lt%d" it will try and find a suitable
1088 * id. It scans list of devices to build up a free map, then chooses
1089 * the first empty slot. The caller must hold the dev_base or rtnl lock
1090 * while allocating the name and adding the device in order to avoid
1092 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1093 * Returns the number of the unit assigned or a negative errno code.
1096 int dev_alloc_name(struct net_device *dev, const char *name)
1098 return dev_alloc_name_ns(dev_net(dev), dev, name);
1100 EXPORT_SYMBOL(dev_alloc_name);
1102 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1107 if (!dev_valid_name(name))
1110 if (strchr(name, '%'))
1111 return dev_alloc_name_ns(net, dev, name);
1112 else if (netdev_name_in_use(net, name))
1114 else if (dev->name != name)
1115 strlcpy(dev->name, name, IFNAMSIZ);
1121 * dev_change_name - change name of a device
1123 * @newname: name (or format string) must be at least IFNAMSIZ
1125 * Change name of a device, can pass format strings "eth%d".
1128 int dev_change_name(struct net_device *dev, const char *newname)
1130 unsigned char old_assign_type;
1131 char oldname[IFNAMSIZ];
1137 BUG_ON(!dev_net(dev));
1141 /* Some auto-enslaved devices e.g. failover slaves are
1142 * special, as userspace might rename the device after
1143 * the interface had been brought up and running since
1144 * the point kernel initiated auto-enslavement. Allow
1145 * live name change even when these slave devices are
1148 * Typically, users of these auto-enslaving devices
1149 * don't actually care about slave name change, as
1150 * they are supposed to operate on master interface
1153 if (dev->flags & IFF_UP &&
1154 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1157 down_write(&devnet_rename_sem);
1159 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1160 up_write(&devnet_rename_sem);
1164 memcpy(oldname, dev->name, IFNAMSIZ);
1166 err = dev_get_valid_name(net, dev, newname);
1168 up_write(&devnet_rename_sem);
1172 if (oldname[0] && !strchr(oldname, '%'))
1173 netdev_info(dev, "renamed from %s\n", oldname);
1175 old_assign_type = dev->name_assign_type;
1176 dev->name_assign_type = NET_NAME_RENAMED;
1179 ret = device_rename(&dev->dev, dev->name);
1181 memcpy(dev->name, oldname, IFNAMSIZ);
1182 dev->name_assign_type = old_assign_type;
1183 up_write(&devnet_rename_sem);
1187 up_write(&devnet_rename_sem);
1189 netdev_adjacent_rename_links(dev, oldname);
1191 write_lock_bh(&dev_base_lock);
1192 netdev_name_node_del(dev->name_node);
1193 write_unlock_bh(&dev_base_lock);
1197 write_lock_bh(&dev_base_lock);
1198 netdev_name_node_add(net, dev->name_node);
1199 write_unlock_bh(&dev_base_lock);
1201 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1202 ret = notifier_to_errno(ret);
1205 /* err >= 0 after dev_alloc_name() or stores the first errno */
1208 down_write(&devnet_rename_sem);
1209 memcpy(dev->name, oldname, IFNAMSIZ);
1210 memcpy(oldname, newname, IFNAMSIZ);
1211 dev->name_assign_type = old_assign_type;
1212 old_assign_type = NET_NAME_RENAMED;
1215 netdev_err(dev, "name change rollback failed: %d\n",
1224 * dev_set_alias - change ifalias of a device
1226 * @alias: name up to IFALIASZ
1227 * @len: limit of bytes to copy from info
1229 * Set ifalias for a device,
1231 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1233 struct dev_ifalias *new_alias = NULL;
1235 if (len >= IFALIASZ)
1239 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1243 memcpy(new_alias->ifalias, alias, len);
1244 new_alias->ifalias[len] = 0;
1247 mutex_lock(&ifalias_mutex);
1248 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1249 mutex_is_locked(&ifalias_mutex));
1250 mutex_unlock(&ifalias_mutex);
1253 kfree_rcu(new_alias, rcuhead);
1257 EXPORT_SYMBOL(dev_set_alias);
1260 * dev_get_alias - get ifalias of a device
1262 * @name: buffer to store name of ifalias
1263 * @len: size of buffer
1265 * get ifalias for a device. Caller must make sure dev cannot go
1266 * away, e.g. rcu read lock or own a reference count to device.
1268 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1270 const struct dev_ifalias *alias;
1274 alias = rcu_dereference(dev->ifalias);
1276 ret = snprintf(name, len, "%s", alias->ifalias);
1283 * netdev_features_change - device changes features
1284 * @dev: device to cause notification
1286 * Called to indicate a device has changed features.
1288 void netdev_features_change(struct net_device *dev)
1290 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1292 EXPORT_SYMBOL(netdev_features_change);
1295 * netdev_state_change - device changes state
1296 * @dev: device to cause notification
1298 * Called to indicate a device has changed state. This function calls
1299 * the notifier chains for netdev_chain and sends a NEWLINK message
1300 * to the routing socket.
1302 void netdev_state_change(struct net_device *dev)
1304 if (dev->flags & IFF_UP) {
1305 struct netdev_notifier_change_info change_info = {
1309 call_netdevice_notifiers_info(NETDEV_CHANGE,
1311 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1314 EXPORT_SYMBOL(netdev_state_change);
1317 * __netdev_notify_peers - notify network peers about existence of @dev,
1318 * to be called when rtnl lock is already held.
1319 * @dev: network device
1321 * Generate traffic such that interested network peers are aware of
1322 * @dev, such as by generating a gratuitous ARP. This may be used when
1323 * a device wants to inform the rest of the network about some sort of
1324 * reconfiguration such as a failover event or virtual machine
1327 void __netdev_notify_peers(struct net_device *dev)
1330 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1331 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1333 EXPORT_SYMBOL(__netdev_notify_peers);
1336 * netdev_notify_peers - notify network peers about existence of @dev
1337 * @dev: network device
1339 * Generate traffic such that interested network peers are aware of
1340 * @dev, such as by generating a gratuitous ARP. This may be used when
1341 * a device wants to inform the rest of the network about some sort of
1342 * reconfiguration such as a failover event or virtual machine
1345 void netdev_notify_peers(struct net_device *dev)
1348 __netdev_notify_peers(dev);
1351 EXPORT_SYMBOL(netdev_notify_peers);
1353 static int napi_threaded_poll(void *data);
1355 static int napi_kthread_create(struct napi_struct *n)
1359 /* Create and wake up the kthread once to put it in
1360 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1361 * warning and work with loadavg.
1363 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1364 n->dev->name, n->napi_id);
1365 if (IS_ERR(n->thread)) {
1366 err = PTR_ERR(n->thread);
1367 pr_err("kthread_run failed with err %d\n", err);
1374 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1376 const struct net_device_ops *ops = dev->netdev_ops;
1380 dev_addr_check(dev);
1382 if (!netif_device_present(dev)) {
1383 /* may be detached because parent is runtime-suspended */
1384 if (dev->dev.parent)
1385 pm_runtime_resume(dev->dev.parent);
1386 if (!netif_device_present(dev))
1390 /* Block netpoll from trying to do any rx path servicing.
1391 * If we don't do this there is a chance ndo_poll_controller
1392 * or ndo_poll may be running while we open the device
1394 netpoll_poll_disable(dev);
1396 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1397 ret = notifier_to_errno(ret);
1401 set_bit(__LINK_STATE_START, &dev->state);
1403 if (ops->ndo_validate_addr)
1404 ret = ops->ndo_validate_addr(dev);
1406 if (!ret && ops->ndo_open)
1407 ret = ops->ndo_open(dev);
1409 netpoll_poll_enable(dev);
1412 clear_bit(__LINK_STATE_START, &dev->state);
1414 dev->flags |= IFF_UP;
1415 dev_set_rx_mode(dev);
1417 add_device_randomness(dev->dev_addr, dev->addr_len);
1424 * dev_open - prepare an interface for use.
1425 * @dev: device to open
1426 * @extack: netlink extended ack
1428 * Takes a device from down to up state. The device's private open
1429 * function is invoked and then the multicast lists are loaded. Finally
1430 * the device is moved into the up state and a %NETDEV_UP message is
1431 * sent to the netdev notifier chain.
1433 * Calling this function on an active interface is a nop. On a failure
1434 * a negative errno code is returned.
1436 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1440 if (dev->flags & IFF_UP)
1443 ret = __dev_open(dev, extack);
1447 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1448 call_netdevice_notifiers(NETDEV_UP, dev);
1452 EXPORT_SYMBOL(dev_open);
1454 static void __dev_close_many(struct list_head *head)
1456 struct net_device *dev;
1461 list_for_each_entry(dev, head, close_list) {
1462 /* Temporarily disable netpoll until the interface is down */
1463 netpoll_poll_disable(dev);
1465 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1467 clear_bit(__LINK_STATE_START, &dev->state);
1469 /* Synchronize to scheduled poll. We cannot touch poll list, it
1470 * can be even on different cpu. So just clear netif_running().
1472 * dev->stop() will invoke napi_disable() on all of it's
1473 * napi_struct instances on this device.
1475 smp_mb__after_atomic(); /* Commit netif_running(). */
1478 dev_deactivate_many(head);
1480 list_for_each_entry(dev, head, close_list) {
1481 const struct net_device_ops *ops = dev->netdev_ops;
1484 * Call the device specific close. This cannot fail.
1485 * Only if device is UP
1487 * We allow it to be called even after a DETACH hot-plug
1493 dev->flags &= ~IFF_UP;
1494 netpoll_poll_enable(dev);
1498 static void __dev_close(struct net_device *dev)
1502 list_add(&dev->close_list, &single);
1503 __dev_close_many(&single);
1507 void dev_close_many(struct list_head *head, bool unlink)
1509 struct net_device *dev, *tmp;
1511 /* Remove the devices that don't need to be closed */
1512 list_for_each_entry_safe(dev, tmp, head, close_list)
1513 if (!(dev->flags & IFF_UP))
1514 list_del_init(&dev->close_list);
1516 __dev_close_many(head);
1518 list_for_each_entry_safe(dev, tmp, head, close_list) {
1519 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1520 call_netdevice_notifiers(NETDEV_DOWN, dev);
1522 list_del_init(&dev->close_list);
1525 EXPORT_SYMBOL(dev_close_many);
1528 * dev_close - shutdown an interface.
1529 * @dev: device to shutdown
1531 * This function moves an active device into down state. A
1532 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1533 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1536 void dev_close(struct net_device *dev)
1538 if (dev->flags & IFF_UP) {
1541 list_add(&dev->close_list, &single);
1542 dev_close_many(&single, true);
1546 EXPORT_SYMBOL(dev_close);
1550 * dev_disable_lro - disable Large Receive Offload on a device
1553 * Disable Large Receive Offload (LRO) on a net device. Must be
1554 * called under RTNL. This is needed if received packets may be
1555 * forwarded to another interface.
1557 void dev_disable_lro(struct net_device *dev)
1559 struct net_device *lower_dev;
1560 struct list_head *iter;
1562 dev->wanted_features &= ~NETIF_F_LRO;
1563 netdev_update_features(dev);
1565 if (unlikely(dev->features & NETIF_F_LRO))
1566 netdev_WARN(dev, "failed to disable LRO!\n");
1568 netdev_for_each_lower_dev(dev, lower_dev, iter)
1569 dev_disable_lro(lower_dev);
1571 EXPORT_SYMBOL(dev_disable_lro);
1574 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1577 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1578 * called under RTNL. This is needed if Generic XDP is installed on
1581 static void dev_disable_gro_hw(struct net_device *dev)
1583 dev->wanted_features &= ~NETIF_F_GRO_HW;
1584 netdev_update_features(dev);
1586 if (unlikely(dev->features & NETIF_F_GRO_HW))
1587 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1590 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1593 case NETDEV_##val: \
1594 return "NETDEV_" __stringify(val);
1596 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1597 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1598 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1599 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1600 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1601 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1602 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1603 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1604 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1608 return "UNKNOWN_NETDEV_EVENT";
1610 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1612 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1613 struct net_device *dev)
1615 struct netdev_notifier_info info = {
1619 return nb->notifier_call(nb, val, &info);
1622 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1623 struct net_device *dev)
1627 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1628 err = notifier_to_errno(err);
1632 if (!(dev->flags & IFF_UP))
1635 call_netdevice_notifier(nb, NETDEV_UP, dev);
1639 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1640 struct net_device *dev)
1642 if (dev->flags & IFF_UP) {
1643 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1645 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1647 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1650 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1653 struct net_device *dev;
1656 for_each_netdev(net, dev) {
1657 err = call_netdevice_register_notifiers(nb, dev);
1664 for_each_netdev_continue_reverse(net, dev)
1665 call_netdevice_unregister_notifiers(nb, dev);
1669 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1672 struct net_device *dev;
1674 for_each_netdev(net, dev)
1675 call_netdevice_unregister_notifiers(nb, dev);
1678 static int dev_boot_phase = 1;
1681 * register_netdevice_notifier - register a network notifier block
1684 * Register a notifier to be called when network device events occur.
1685 * The notifier passed is linked into the kernel structures and must
1686 * not be reused until it has been unregistered. A negative errno code
1687 * is returned on a failure.
1689 * When registered all registration and up events are replayed
1690 * to the new notifier to allow device to have a race free
1691 * view of the network device list.
1694 int register_netdevice_notifier(struct notifier_block *nb)
1699 /* Close race with setup_net() and cleanup_net() */
1700 down_write(&pernet_ops_rwsem);
1702 err = raw_notifier_chain_register(&netdev_chain, nb);
1708 err = call_netdevice_register_net_notifiers(nb, net);
1715 up_write(&pernet_ops_rwsem);
1719 for_each_net_continue_reverse(net)
1720 call_netdevice_unregister_net_notifiers(nb, net);
1722 raw_notifier_chain_unregister(&netdev_chain, nb);
1725 EXPORT_SYMBOL(register_netdevice_notifier);
1728 * unregister_netdevice_notifier - unregister a network notifier block
1731 * Unregister a notifier previously registered by
1732 * register_netdevice_notifier(). The notifier is unlinked into the
1733 * kernel structures and may then be reused. A negative errno code
1734 * is returned on a failure.
1736 * After unregistering unregister and down device events are synthesized
1737 * for all devices on the device list to the removed notifier to remove
1738 * the need for special case cleanup code.
1741 int unregister_netdevice_notifier(struct notifier_block *nb)
1746 /* Close race with setup_net() and cleanup_net() */
1747 down_write(&pernet_ops_rwsem);
1749 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1754 call_netdevice_unregister_net_notifiers(nb, net);
1758 up_write(&pernet_ops_rwsem);
1761 EXPORT_SYMBOL(unregister_netdevice_notifier);
1763 static int __register_netdevice_notifier_net(struct net *net,
1764 struct notifier_block *nb,
1765 bool ignore_call_fail)
1769 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1775 err = call_netdevice_register_net_notifiers(nb, net);
1776 if (err && !ignore_call_fail)
1777 goto chain_unregister;
1782 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1786 static int __unregister_netdevice_notifier_net(struct net *net,
1787 struct notifier_block *nb)
1791 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1795 call_netdevice_unregister_net_notifiers(nb, net);
1800 * register_netdevice_notifier_net - register a per-netns network notifier block
1801 * @net: network namespace
1804 * Register a notifier to be called when network device events occur.
1805 * The notifier passed is linked into the kernel structures and must
1806 * not be reused until it has been unregistered. A negative errno code
1807 * is returned on a failure.
1809 * When registered all registration and up events are replayed
1810 * to the new notifier to allow device to have a race free
1811 * view of the network device list.
1814 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1819 err = __register_netdevice_notifier_net(net, nb, false);
1823 EXPORT_SYMBOL(register_netdevice_notifier_net);
1826 * unregister_netdevice_notifier_net - unregister a per-netns
1827 * network notifier block
1828 * @net: network namespace
1831 * Unregister a notifier previously registered by
1832 * register_netdevice_notifier(). The notifier is unlinked into the
1833 * kernel structures and may then be reused. A negative errno code
1834 * is returned on a failure.
1836 * After unregistering unregister and down device events are synthesized
1837 * for all devices on the device list to the removed notifier to remove
1838 * the need for special case cleanup code.
1841 int unregister_netdevice_notifier_net(struct net *net,
1842 struct notifier_block *nb)
1847 err = __unregister_netdevice_notifier_net(net, nb);
1851 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1853 int register_netdevice_notifier_dev_net(struct net_device *dev,
1854 struct notifier_block *nb,
1855 struct netdev_net_notifier *nn)
1860 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1863 list_add(&nn->list, &dev->net_notifier_list);
1868 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1870 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1871 struct notifier_block *nb,
1872 struct netdev_net_notifier *nn)
1877 list_del(&nn->list);
1878 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1882 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1884 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1887 struct netdev_net_notifier *nn;
1889 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1890 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1891 __register_netdevice_notifier_net(net, nn->nb, true);
1896 * call_netdevice_notifiers_info - call all network notifier blocks
1897 * @val: value passed unmodified to notifier function
1898 * @info: notifier information data
1900 * Call all network notifier blocks. Parameters and return value
1901 * are as for raw_notifier_call_chain().
1904 static int call_netdevice_notifiers_info(unsigned long val,
1905 struct netdev_notifier_info *info)
1907 struct net *net = dev_net(info->dev);
1912 /* Run per-netns notifier block chain first, then run the global one.
1913 * Hopefully, one day, the global one is going to be removed after
1914 * all notifier block registrators get converted to be per-netns.
1916 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1917 if (ret & NOTIFY_STOP_MASK)
1919 return raw_notifier_call_chain(&netdev_chain, val, info);
1922 static int call_netdevice_notifiers_extack(unsigned long val,
1923 struct net_device *dev,
1924 struct netlink_ext_ack *extack)
1926 struct netdev_notifier_info info = {
1931 return call_netdevice_notifiers_info(val, &info);
1935 * call_netdevice_notifiers - call all network notifier blocks
1936 * @val: value passed unmodified to notifier function
1937 * @dev: net_device pointer passed unmodified to notifier function
1939 * Call all network notifier blocks. Parameters and return value
1940 * are as for raw_notifier_call_chain().
1943 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1945 return call_netdevice_notifiers_extack(val, dev, NULL);
1947 EXPORT_SYMBOL(call_netdevice_notifiers);
1950 * call_netdevice_notifiers_mtu - call all network notifier blocks
1951 * @val: value passed unmodified to notifier function
1952 * @dev: net_device pointer passed unmodified to notifier function
1953 * @arg: additional u32 argument passed to the notifier function
1955 * Call all network notifier blocks. Parameters and return value
1956 * are as for raw_notifier_call_chain().
1958 static int call_netdevice_notifiers_mtu(unsigned long val,
1959 struct net_device *dev, u32 arg)
1961 struct netdev_notifier_info_ext info = {
1966 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1968 return call_netdevice_notifiers_info(val, &info.info);
1971 #ifdef CONFIG_NET_INGRESS
1972 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1974 void net_inc_ingress_queue(void)
1976 static_branch_inc(&ingress_needed_key);
1978 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1980 void net_dec_ingress_queue(void)
1982 static_branch_dec(&ingress_needed_key);
1984 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1987 #ifdef CONFIG_NET_EGRESS
1988 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1990 void net_inc_egress_queue(void)
1992 static_branch_inc(&egress_needed_key);
1994 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1996 void net_dec_egress_queue(void)
1998 static_branch_dec(&egress_needed_key);
2000 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2003 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2004 #ifdef CONFIG_JUMP_LABEL
2005 static atomic_t netstamp_needed_deferred;
2006 static atomic_t netstamp_wanted;
2007 static void netstamp_clear(struct work_struct *work)
2009 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2012 wanted = atomic_add_return(deferred, &netstamp_wanted);
2014 static_branch_enable(&netstamp_needed_key);
2016 static_branch_disable(&netstamp_needed_key);
2018 static DECLARE_WORK(netstamp_work, netstamp_clear);
2021 void net_enable_timestamp(void)
2023 #ifdef CONFIG_JUMP_LABEL
2027 wanted = atomic_read(&netstamp_wanted);
2030 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2033 atomic_inc(&netstamp_needed_deferred);
2034 schedule_work(&netstamp_work);
2036 static_branch_inc(&netstamp_needed_key);
2039 EXPORT_SYMBOL(net_enable_timestamp);
2041 void net_disable_timestamp(void)
2043 #ifdef CONFIG_JUMP_LABEL
2047 wanted = atomic_read(&netstamp_wanted);
2050 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2053 atomic_dec(&netstamp_needed_deferred);
2054 schedule_work(&netstamp_work);
2056 static_branch_dec(&netstamp_needed_key);
2059 EXPORT_SYMBOL(net_disable_timestamp);
2061 static inline void net_timestamp_set(struct sk_buff *skb)
2064 if (static_branch_unlikely(&netstamp_needed_key))
2065 __net_timestamp(skb);
2068 #define net_timestamp_check(COND, SKB) \
2069 if (static_branch_unlikely(&netstamp_needed_key)) { \
2070 if ((COND) && !(SKB)->tstamp) \
2071 __net_timestamp(SKB); \
2074 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2076 return __is_skb_forwardable(dev, skb, true);
2078 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2080 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2083 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2086 skb->protocol = eth_type_trans(skb, dev);
2087 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2093 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2095 return __dev_forward_skb2(dev, skb, true);
2097 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2100 * dev_forward_skb - loopback an skb to another netif
2102 * @dev: destination network device
2103 * @skb: buffer to forward
2106 * NET_RX_SUCCESS (no congestion)
2107 * NET_RX_DROP (packet was dropped, but freed)
2109 * dev_forward_skb can be used for injecting an skb from the
2110 * start_xmit function of one device into the receive queue
2111 * of another device.
2113 * The receiving device may be in another namespace, so
2114 * we have to clear all information in the skb that could
2115 * impact namespace isolation.
2117 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2119 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2121 EXPORT_SYMBOL_GPL(dev_forward_skb);
2123 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2125 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2128 static inline int deliver_skb(struct sk_buff *skb,
2129 struct packet_type *pt_prev,
2130 struct net_device *orig_dev)
2132 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2134 refcount_inc(&skb->users);
2135 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2138 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2139 struct packet_type **pt,
2140 struct net_device *orig_dev,
2142 struct list_head *ptype_list)
2144 struct packet_type *ptype, *pt_prev = *pt;
2146 list_for_each_entry_rcu(ptype, ptype_list, list) {
2147 if (ptype->type != type)
2150 deliver_skb(skb, pt_prev, orig_dev);
2156 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2158 if (!ptype->af_packet_priv || !skb->sk)
2161 if (ptype->id_match)
2162 return ptype->id_match(ptype, skb->sk);
2163 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2170 * dev_nit_active - return true if any network interface taps are in use
2172 * @dev: network device to check for the presence of taps
2174 bool dev_nit_active(struct net_device *dev)
2176 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2178 EXPORT_SYMBOL_GPL(dev_nit_active);
2181 * Support routine. Sends outgoing frames to any network
2182 * taps currently in use.
2185 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2187 struct packet_type *ptype;
2188 struct sk_buff *skb2 = NULL;
2189 struct packet_type *pt_prev = NULL;
2190 struct list_head *ptype_list = &ptype_all;
2194 list_for_each_entry_rcu(ptype, ptype_list, list) {
2195 if (ptype->ignore_outgoing)
2198 /* Never send packets back to the socket
2199 * they originated from - MvS (miquels@drinkel.ow.org)
2201 if (skb_loop_sk(ptype, skb))
2205 deliver_skb(skb2, pt_prev, skb->dev);
2210 /* need to clone skb, done only once */
2211 skb2 = skb_clone(skb, GFP_ATOMIC);
2215 net_timestamp_set(skb2);
2217 /* skb->nh should be correctly
2218 * set by sender, so that the second statement is
2219 * just protection against buggy protocols.
2221 skb_reset_mac_header(skb2);
2223 if (skb_network_header(skb2) < skb2->data ||
2224 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2225 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2226 ntohs(skb2->protocol),
2228 skb_reset_network_header(skb2);
2231 skb2->transport_header = skb2->network_header;
2232 skb2->pkt_type = PACKET_OUTGOING;
2236 if (ptype_list == &ptype_all) {
2237 ptype_list = &dev->ptype_all;
2242 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2243 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2249 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2252 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2253 * @dev: Network device
2254 * @txq: number of queues available
2256 * If real_num_tx_queues is changed the tc mappings may no longer be
2257 * valid. To resolve this verify the tc mapping remains valid and if
2258 * not NULL the mapping. With no priorities mapping to this
2259 * offset/count pair it will no longer be used. In the worst case TC0
2260 * is invalid nothing can be done so disable priority mappings. If is
2261 * expected that drivers will fix this mapping if they can before
2262 * calling netif_set_real_num_tx_queues.
2264 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2267 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2269 /* If TC0 is invalidated disable TC mapping */
2270 if (tc->offset + tc->count > txq) {
2271 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2276 /* Invalidated prio to tc mappings set to TC0 */
2277 for (i = 1; i < TC_BITMASK + 1; i++) {
2278 int q = netdev_get_prio_tc_map(dev, i);
2280 tc = &dev->tc_to_txq[q];
2281 if (tc->offset + tc->count > txq) {
2282 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",
2284 netdev_set_prio_tc_map(dev, i, 0);
2289 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2292 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2295 /* walk through the TCs and see if it falls into any of them */
2296 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2297 if ((txq - tc->offset) < tc->count)
2301 /* didn't find it, just return -1 to indicate no match */
2307 EXPORT_SYMBOL(netdev_txq_to_tc);
2310 static struct static_key xps_needed __read_mostly;
2311 static struct static_key xps_rxqs_needed __read_mostly;
2312 static DEFINE_MUTEX(xps_map_mutex);
2313 #define xmap_dereference(P) \
2314 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2316 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2317 struct xps_dev_maps *old_maps, int tci, u16 index)
2319 struct xps_map *map = NULL;
2323 map = xmap_dereference(dev_maps->attr_map[tci]);
2327 for (pos = map->len; pos--;) {
2328 if (map->queues[pos] != index)
2332 map->queues[pos] = map->queues[--map->len];
2337 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2338 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2339 kfree_rcu(map, rcu);
2346 static bool remove_xps_queue_cpu(struct net_device *dev,
2347 struct xps_dev_maps *dev_maps,
2348 int cpu, u16 offset, u16 count)
2350 int num_tc = dev_maps->num_tc;
2351 bool active = false;
2354 for (tci = cpu * num_tc; num_tc--; tci++) {
2357 for (i = count, j = offset; i--; j++) {
2358 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2368 static void reset_xps_maps(struct net_device *dev,
2369 struct xps_dev_maps *dev_maps,
2370 enum xps_map_type type)
2372 static_key_slow_dec_cpuslocked(&xps_needed);
2373 if (type == XPS_RXQS)
2374 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2376 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2378 kfree_rcu(dev_maps, rcu);
2381 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2382 u16 offset, u16 count)
2384 struct xps_dev_maps *dev_maps;
2385 bool active = false;
2388 dev_maps = xmap_dereference(dev->xps_maps[type]);
2392 for (j = 0; j < dev_maps->nr_ids; j++)
2393 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2395 reset_xps_maps(dev, dev_maps, type);
2397 if (type == XPS_CPUS) {
2398 for (i = offset + (count - 1); count--; i--)
2399 netdev_queue_numa_node_write(
2400 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2404 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2407 if (!static_key_false(&xps_needed))
2411 mutex_lock(&xps_map_mutex);
2413 if (static_key_false(&xps_rxqs_needed))
2414 clean_xps_maps(dev, XPS_RXQS, offset, count);
2416 clean_xps_maps(dev, XPS_CPUS, offset, count);
2418 mutex_unlock(&xps_map_mutex);
2422 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2424 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2427 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2428 u16 index, bool is_rxqs_map)
2430 struct xps_map *new_map;
2431 int alloc_len = XPS_MIN_MAP_ALLOC;
2434 for (pos = 0; map && pos < map->len; pos++) {
2435 if (map->queues[pos] != index)
2440 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2442 if (pos < map->alloc_len)
2445 alloc_len = map->alloc_len * 2;
2448 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2452 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2454 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2455 cpu_to_node(attr_index));
2459 for (i = 0; i < pos; i++)
2460 new_map->queues[i] = map->queues[i];
2461 new_map->alloc_len = alloc_len;
2467 /* Copy xps maps at a given index */
2468 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2469 struct xps_dev_maps *new_dev_maps, int index,
2470 int tc, bool skip_tc)
2472 int i, tci = index * dev_maps->num_tc;
2473 struct xps_map *map;
2475 /* copy maps belonging to foreign traffic classes */
2476 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2477 if (i == tc && skip_tc)
2480 /* fill in the new device map from the old device map */
2481 map = xmap_dereference(dev_maps->attr_map[tci]);
2482 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2486 /* Must be called under cpus_read_lock */
2487 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2488 u16 index, enum xps_map_type type)
2490 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2491 const unsigned long *online_mask = NULL;
2492 bool active = false, copy = false;
2493 int i, j, tci, numa_node_id = -2;
2494 int maps_sz, num_tc = 1, tc = 0;
2495 struct xps_map *map, *new_map;
2496 unsigned int nr_ids;
2499 /* Do not allow XPS on subordinate device directly */
2500 num_tc = dev->num_tc;
2504 /* If queue belongs to subordinate dev use its map */
2505 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2507 tc = netdev_txq_to_tc(dev, index);
2512 mutex_lock(&xps_map_mutex);
2514 dev_maps = xmap_dereference(dev->xps_maps[type]);
2515 if (type == XPS_RXQS) {
2516 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2517 nr_ids = dev->num_rx_queues;
2519 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2520 if (num_possible_cpus() > 1)
2521 online_mask = cpumask_bits(cpu_online_mask);
2522 nr_ids = nr_cpu_ids;
2525 if (maps_sz < L1_CACHE_BYTES)
2526 maps_sz = L1_CACHE_BYTES;
2528 /* The old dev_maps could be larger or smaller than the one we're
2529 * setting up now, as dev->num_tc or nr_ids could have been updated in
2530 * between. We could try to be smart, but let's be safe instead and only
2531 * copy foreign traffic classes if the two map sizes match.
2534 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2537 /* allocate memory for queue storage */
2538 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2540 if (!new_dev_maps) {
2541 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2542 if (!new_dev_maps) {
2543 mutex_unlock(&xps_map_mutex);
2547 new_dev_maps->nr_ids = nr_ids;
2548 new_dev_maps->num_tc = num_tc;
2551 tci = j * num_tc + tc;
2552 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2554 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2558 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2562 goto out_no_new_maps;
2565 /* Increment static keys at most once per type */
2566 static_key_slow_inc_cpuslocked(&xps_needed);
2567 if (type == XPS_RXQS)
2568 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2571 for (j = 0; j < nr_ids; j++) {
2572 bool skip_tc = false;
2574 tci = j * num_tc + tc;
2575 if (netif_attr_test_mask(j, mask, nr_ids) &&
2576 netif_attr_test_online(j, online_mask, nr_ids)) {
2577 /* add tx-queue to CPU/rx-queue maps */
2582 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2583 while ((pos < map->len) && (map->queues[pos] != index))
2586 if (pos == map->len)
2587 map->queues[map->len++] = index;
2589 if (type == XPS_CPUS) {
2590 if (numa_node_id == -2)
2591 numa_node_id = cpu_to_node(j);
2592 else if (numa_node_id != cpu_to_node(j))
2599 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2603 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2605 /* Cleanup old maps */
2607 goto out_no_old_maps;
2609 for (j = 0; j < dev_maps->nr_ids; j++) {
2610 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2611 map = xmap_dereference(dev_maps->attr_map[tci]);
2616 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2621 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2622 kfree_rcu(map, rcu);
2626 old_dev_maps = dev_maps;
2629 dev_maps = new_dev_maps;
2633 if (type == XPS_CPUS)
2634 /* update Tx queue numa node */
2635 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2636 (numa_node_id >= 0) ?
2637 numa_node_id : NUMA_NO_NODE);
2642 /* removes tx-queue from unused CPUs/rx-queues */
2643 for (j = 0; j < dev_maps->nr_ids; j++) {
2644 tci = j * dev_maps->num_tc;
2646 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2648 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2649 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2652 active |= remove_xps_queue(dev_maps,
2653 copy ? old_dev_maps : NULL,
2659 kfree_rcu(old_dev_maps, rcu);
2661 /* free map if not active */
2663 reset_xps_maps(dev, dev_maps, type);
2666 mutex_unlock(&xps_map_mutex);
2670 /* remove any maps that we added */
2671 for (j = 0; j < nr_ids; j++) {
2672 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2673 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2675 xmap_dereference(dev_maps->attr_map[tci]) :
2677 if (new_map && new_map != map)
2682 mutex_unlock(&xps_map_mutex);
2684 kfree(new_dev_maps);
2687 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2689 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2695 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2700 EXPORT_SYMBOL(netif_set_xps_queue);
2703 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2705 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2707 /* Unbind any subordinate channels */
2708 while (txq-- != &dev->_tx[0]) {
2710 netdev_unbind_sb_channel(dev, txq->sb_dev);
2714 void netdev_reset_tc(struct net_device *dev)
2717 netif_reset_xps_queues_gt(dev, 0);
2719 netdev_unbind_all_sb_channels(dev);
2721 /* Reset TC configuration of device */
2723 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2724 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2726 EXPORT_SYMBOL(netdev_reset_tc);
2728 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2730 if (tc >= dev->num_tc)
2734 netif_reset_xps_queues(dev, offset, count);
2736 dev->tc_to_txq[tc].count = count;
2737 dev->tc_to_txq[tc].offset = offset;
2740 EXPORT_SYMBOL(netdev_set_tc_queue);
2742 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2744 if (num_tc > TC_MAX_QUEUE)
2748 netif_reset_xps_queues_gt(dev, 0);
2750 netdev_unbind_all_sb_channels(dev);
2752 dev->num_tc = num_tc;
2755 EXPORT_SYMBOL(netdev_set_num_tc);
2757 void netdev_unbind_sb_channel(struct net_device *dev,
2758 struct net_device *sb_dev)
2760 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2763 netif_reset_xps_queues_gt(sb_dev, 0);
2765 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2766 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2768 while (txq-- != &dev->_tx[0]) {
2769 if (txq->sb_dev == sb_dev)
2773 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2775 int netdev_bind_sb_channel_queue(struct net_device *dev,
2776 struct net_device *sb_dev,
2777 u8 tc, u16 count, u16 offset)
2779 /* Make certain the sb_dev and dev are already configured */
2780 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2783 /* We cannot hand out queues we don't have */
2784 if ((offset + count) > dev->real_num_tx_queues)
2787 /* Record the mapping */
2788 sb_dev->tc_to_txq[tc].count = count;
2789 sb_dev->tc_to_txq[tc].offset = offset;
2791 /* Provide a way for Tx queue to find the tc_to_txq map or
2792 * XPS map for itself.
2795 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2799 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2801 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2803 /* Do not use a multiqueue device to represent a subordinate channel */
2804 if (netif_is_multiqueue(dev))
2807 /* We allow channels 1 - 32767 to be used for subordinate channels.
2808 * Channel 0 is meant to be "native" mode and used only to represent
2809 * the main root device. We allow writing 0 to reset the device back
2810 * to normal mode after being used as a subordinate channel.
2812 if (channel > S16_MAX)
2815 dev->num_tc = -channel;
2819 EXPORT_SYMBOL(netdev_set_sb_channel);
2822 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2823 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2825 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2830 disabling = txq < dev->real_num_tx_queues;
2832 if (txq < 1 || txq > dev->num_tx_queues)
2835 if (dev->reg_state == NETREG_REGISTERED ||
2836 dev->reg_state == NETREG_UNREGISTERING) {
2839 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2845 netif_setup_tc(dev, txq);
2847 dev_qdisc_change_real_num_tx(dev, txq);
2849 dev->real_num_tx_queues = txq;
2853 qdisc_reset_all_tx_gt(dev, txq);
2855 netif_reset_xps_queues_gt(dev, txq);
2859 dev->real_num_tx_queues = txq;
2864 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2868 * netif_set_real_num_rx_queues - set actual number of RX queues used
2869 * @dev: Network device
2870 * @rxq: Actual number of RX queues
2872 * This must be called either with the rtnl_lock held or before
2873 * registration of the net device. Returns 0 on success, or a
2874 * negative error code. If called before registration, it always
2877 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2881 if (rxq < 1 || rxq > dev->num_rx_queues)
2884 if (dev->reg_state == NETREG_REGISTERED) {
2887 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2893 dev->real_num_rx_queues = rxq;
2896 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2900 * netif_set_real_num_queues - set actual number of RX and TX queues used
2901 * @dev: Network device
2902 * @txq: Actual number of TX queues
2903 * @rxq: Actual number of RX queues
2905 * Set the real number of both TX and RX queues.
2906 * Does nothing if the number of queues is already correct.
2908 int netif_set_real_num_queues(struct net_device *dev,
2909 unsigned int txq, unsigned int rxq)
2911 unsigned int old_rxq = dev->real_num_rx_queues;
2914 if (txq < 1 || txq > dev->num_tx_queues ||
2915 rxq < 1 || rxq > dev->num_rx_queues)
2918 /* Start from increases, so the error path only does decreases -
2919 * decreases can't fail.
2921 if (rxq > dev->real_num_rx_queues) {
2922 err = netif_set_real_num_rx_queues(dev, rxq);
2926 if (txq > dev->real_num_tx_queues) {
2927 err = netif_set_real_num_tx_queues(dev, txq);
2931 if (rxq < dev->real_num_rx_queues)
2932 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2933 if (txq < dev->real_num_tx_queues)
2934 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2938 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2941 EXPORT_SYMBOL(netif_set_real_num_queues);
2944 * netif_get_num_default_rss_queues - default number of RSS queues
2946 * This routine should set an upper limit on the number of RSS queues
2947 * used by default by multiqueue devices.
2949 int netif_get_num_default_rss_queues(void)
2951 return is_kdump_kernel() ?
2952 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2954 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2956 static void __netif_reschedule(struct Qdisc *q)
2958 struct softnet_data *sd;
2959 unsigned long flags;
2961 local_irq_save(flags);
2962 sd = this_cpu_ptr(&softnet_data);
2963 q->next_sched = NULL;
2964 *sd->output_queue_tailp = q;
2965 sd->output_queue_tailp = &q->next_sched;
2966 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2967 local_irq_restore(flags);
2970 void __netif_schedule(struct Qdisc *q)
2972 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2973 __netif_reschedule(q);
2975 EXPORT_SYMBOL(__netif_schedule);
2977 struct dev_kfree_skb_cb {
2978 enum skb_free_reason reason;
2981 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2983 return (struct dev_kfree_skb_cb *)skb->cb;
2986 void netif_schedule_queue(struct netdev_queue *txq)
2989 if (!netif_xmit_stopped(txq)) {
2990 struct Qdisc *q = rcu_dereference(txq->qdisc);
2992 __netif_schedule(q);
2996 EXPORT_SYMBOL(netif_schedule_queue);
2998 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3000 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3004 q = rcu_dereference(dev_queue->qdisc);
3005 __netif_schedule(q);
3009 EXPORT_SYMBOL(netif_tx_wake_queue);
3011 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3013 unsigned long flags;
3018 if (likely(refcount_read(&skb->users) == 1)) {
3020 refcount_set(&skb->users, 0);
3021 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3024 get_kfree_skb_cb(skb)->reason = reason;
3025 local_irq_save(flags);
3026 skb->next = __this_cpu_read(softnet_data.completion_queue);
3027 __this_cpu_write(softnet_data.completion_queue, skb);
3028 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3029 local_irq_restore(flags);
3031 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3033 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3035 if (in_hardirq() || irqs_disabled())
3036 __dev_kfree_skb_irq(skb, reason);
3040 EXPORT_SYMBOL(__dev_kfree_skb_any);
3044 * netif_device_detach - mark device as removed
3045 * @dev: network device
3047 * Mark device as removed from system and therefore no longer available.
3049 void netif_device_detach(struct net_device *dev)
3051 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3052 netif_running(dev)) {
3053 netif_tx_stop_all_queues(dev);
3056 EXPORT_SYMBOL(netif_device_detach);
3059 * netif_device_attach - mark device as attached
3060 * @dev: network device
3062 * Mark device as attached from system and restart if needed.
3064 void netif_device_attach(struct net_device *dev)
3066 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3067 netif_running(dev)) {
3068 netif_tx_wake_all_queues(dev);
3069 __netdev_watchdog_up(dev);
3072 EXPORT_SYMBOL(netif_device_attach);
3075 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3076 * to be used as a distribution range.
3078 static u16 skb_tx_hash(const struct net_device *dev,
3079 const struct net_device *sb_dev,
3080 struct sk_buff *skb)
3084 u16 qcount = dev->real_num_tx_queues;
3087 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3089 qoffset = sb_dev->tc_to_txq[tc].offset;
3090 qcount = sb_dev->tc_to_txq[tc].count;
3091 if (unlikely(!qcount)) {
3092 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3093 sb_dev->name, qoffset, tc);
3095 qcount = dev->real_num_tx_queues;
3099 if (skb_rx_queue_recorded(skb)) {
3100 hash = skb_get_rx_queue(skb);
3101 if (hash >= qoffset)
3103 while (unlikely(hash >= qcount))
3105 return hash + qoffset;
3108 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3111 static void skb_warn_bad_offload(const struct sk_buff *skb)
3113 static const netdev_features_t null_features;
3114 struct net_device *dev = skb->dev;
3115 const char *name = "";
3117 if (!net_ratelimit())
3121 if (dev->dev.parent)
3122 name = dev_driver_string(dev->dev.parent);
3124 name = netdev_name(dev);
3126 skb_dump(KERN_WARNING, skb, false);
3127 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3128 name, dev ? &dev->features : &null_features,
3129 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3133 * Invalidate hardware checksum when packet is to be mangled, and
3134 * complete checksum manually on outgoing path.
3136 int skb_checksum_help(struct sk_buff *skb)
3139 int ret = 0, offset;
3141 if (skb->ip_summed == CHECKSUM_COMPLETE)
3142 goto out_set_summed;
3144 if (unlikely(skb_is_gso(skb))) {
3145 skb_warn_bad_offload(skb);
3149 /* Before computing a checksum, we should make sure no frag could
3150 * be modified by an external entity : checksum could be wrong.
3152 if (skb_has_shared_frag(skb)) {
3153 ret = __skb_linearize(skb);
3158 offset = skb_checksum_start_offset(skb);
3159 BUG_ON(offset >= skb_headlen(skb));
3160 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3162 offset += skb->csum_offset;
3163 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3165 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3169 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3171 skb->ip_summed = CHECKSUM_NONE;
3175 EXPORT_SYMBOL(skb_checksum_help);
3177 int skb_crc32c_csum_help(struct sk_buff *skb)
3180 int ret = 0, offset, start;
3182 if (skb->ip_summed != CHECKSUM_PARTIAL)
3185 if (unlikely(skb_is_gso(skb)))
3188 /* Before computing a checksum, we should make sure no frag could
3189 * be modified by an external entity : checksum could be wrong.
3191 if (unlikely(skb_has_shared_frag(skb))) {
3192 ret = __skb_linearize(skb);
3196 start = skb_checksum_start_offset(skb);
3197 offset = start + offsetof(struct sctphdr, checksum);
3198 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3203 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3207 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3208 skb->len - start, ~(__u32)0,
3210 *(__le32 *)(skb->data + offset) = crc32c_csum;
3211 skb->ip_summed = CHECKSUM_NONE;
3212 skb->csum_not_inet = 0;
3217 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3219 __be16 type = skb->protocol;
3221 /* Tunnel gso handlers can set protocol to ethernet. */
3222 if (type == htons(ETH_P_TEB)) {
3225 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3228 eth = (struct ethhdr *)skb->data;
3229 type = eth->h_proto;
3232 return __vlan_get_protocol(skb, type, depth);
3235 /* openvswitch calls this on rx path, so we need a different check.
3237 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3240 return skb->ip_summed != CHECKSUM_PARTIAL &&
3241 skb->ip_summed != CHECKSUM_UNNECESSARY;
3243 return skb->ip_summed == CHECKSUM_NONE;
3247 * __skb_gso_segment - Perform segmentation on skb.
3248 * @skb: buffer to segment
3249 * @features: features for the output path (see dev->features)
3250 * @tx_path: whether it is called in TX path
3252 * This function segments the given skb and returns a list of segments.
3254 * It may return NULL if the skb requires no segmentation. This is
3255 * only possible when GSO is used for verifying header integrity.
3257 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3259 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3260 netdev_features_t features, bool tx_path)
3262 struct sk_buff *segs;
3264 if (unlikely(skb_needs_check(skb, tx_path))) {
3267 /* We're going to init ->check field in TCP or UDP header */
3268 err = skb_cow_head(skb, 0);
3270 return ERR_PTR(err);
3273 /* Only report GSO partial support if it will enable us to
3274 * support segmentation on this frame without needing additional
3277 if (features & NETIF_F_GSO_PARTIAL) {
3278 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3279 struct net_device *dev = skb->dev;
3281 partial_features |= dev->features & dev->gso_partial_features;
3282 if (!skb_gso_ok(skb, features | partial_features))
3283 features &= ~NETIF_F_GSO_PARTIAL;
3286 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3287 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3289 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3290 SKB_GSO_CB(skb)->encap_level = 0;
3292 skb_reset_mac_header(skb);
3293 skb_reset_mac_len(skb);
3295 segs = skb_mac_gso_segment(skb, features);
3297 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3298 skb_warn_bad_offload(skb);
3302 EXPORT_SYMBOL(__skb_gso_segment);
3304 /* Take action when hardware reception checksum errors are detected. */
3306 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3308 netdev_err(dev, "hw csum failure\n");
3309 skb_dump(KERN_ERR, skb, true);
3313 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3315 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3317 EXPORT_SYMBOL(netdev_rx_csum_fault);
3320 /* XXX: check that highmem exists at all on the given machine. */
3321 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3323 #ifdef CONFIG_HIGHMEM
3326 if (!(dev->features & NETIF_F_HIGHDMA)) {
3327 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3328 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3330 if (PageHighMem(skb_frag_page(frag)))
3338 /* If MPLS offload request, verify we are testing hardware MPLS features
3339 * instead of standard features for the netdev.
3341 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3342 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3343 netdev_features_t features,
3346 if (eth_p_mpls(type))
3347 features &= skb->dev->mpls_features;
3352 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3353 netdev_features_t features,
3360 static netdev_features_t harmonize_features(struct sk_buff *skb,
3361 netdev_features_t features)
3365 type = skb_network_protocol(skb, NULL);
3366 features = net_mpls_features(skb, features, type);
3368 if (skb->ip_summed != CHECKSUM_NONE &&
3369 !can_checksum_protocol(features, type)) {
3370 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3372 if (illegal_highdma(skb->dev, skb))
3373 features &= ~NETIF_F_SG;
3378 netdev_features_t passthru_features_check(struct sk_buff *skb,
3379 struct net_device *dev,
3380 netdev_features_t features)
3384 EXPORT_SYMBOL(passthru_features_check);
3386 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3387 struct net_device *dev,
3388 netdev_features_t features)
3390 return vlan_features_check(skb, features);
3393 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3394 struct net_device *dev,
3395 netdev_features_t features)
3397 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3399 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3400 return features & ~NETIF_F_GSO_MASK;
3402 if (!skb_shinfo(skb)->gso_type) {
3403 skb_warn_bad_offload(skb);
3404 return features & ~NETIF_F_GSO_MASK;
3407 /* Support for GSO partial features requires software
3408 * intervention before we can actually process the packets
3409 * so we need to strip support for any partial features now
3410 * and we can pull them back in after we have partially
3411 * segmented the frame.
3413 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3414 features &= ~dev->gso_partial_features;
3416 /* Make sure to clear the IPv4 ID mangling feature if the
3417 * IPv4 header has the potential to be fragmented.
3419 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3420 struct iphdr *iph = skb->encapsulation ?
3421 inner_ip_hdr(skb) : ip_hdr(skb);
3423 if (!(iph->frag_off & htons(IP_DF)))
3424 features &= ~NETIF_F_TSO_MANGLEID;
3430 netdev_features_t netif_skb_features(struct sk_buff *skb)
3432 struct net_device *dev = skb->dev;
3433 netdev_features_t features = dev->features;
3435 if (skb_is_gso(skb))
3436 features = gso_features_check(skb, dev, features);
3438 /* If encapsulation offload request, verify we are testing
3439 * hardware encapsulation features instead of standard
3440 * features for the netdev
3442 if (skb->encapsulation)
3443 features &= dev->hw_enc_features;
3445 if (skb_vlan_tagged(skb))
3446 features = netdev_intersect_features(features,
3447 dev->vlan_features |
3448 NETIF_F_HW_VLAN_CTAG_TX |
3449 NETIF_F_HW_VLAN_STAG_TX);
3451 if (dev->netdev_ops->ndo_features_check)
3452 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3455 features &= dflt_features_check(skb, dev, features);
3457 return harmonize_features(skb, features);
3459 EXPORT_SYMBOL(netif_skb_features);
3461 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3462 struct netdev_queue *txq, bool more)
3467 if (dev_nit_active(dev))
3468 dev_queue_xmit_nit(skb, dev);
3471 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3472 trace_net_dev_start_xmit(skb, dev);
3473 rc = netdev_start_xmit(skb, dev, txq, more);
3474 trace_net_dev_xmit(skb, rc, dev, len);
3479 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3480 struct netdev_queue *txq, int *ret)
3482 struct sk_buff *skb = first;
3483 int rc = NETDEV_TX_OK;
3486 struct sk_buff *next = skb->next;
3488 skb_mark_not_on_list(skb);
3489 rc = xmit_one(skb, dev, txq, next != NULL);
3490 if (unlikely(!dev_xmit_complete(rc))) {
3496 if (netif_tx_queue_stopped(txq) && skb) {
3497 rc = NETDEV_TX_BUSY;
3507 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3508 netdev_features_t features)
3510 if (skb_vlan_tag_present(skb) &&
3511 !vlan_hw_offload_capable(features, skb->vlan_proto))
3512 skb = __vlan_hwaccel_push_inside(skb);
3516 int skb_csum_hwoffload_help(struct sk_buff *skb,
3517 const netdev_features_t features)
3519 if (unlikely(skb_csum_is_sctp(skb)))
3520 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3521 skb_crc32c_csum_help(skb);
3523 if (features & NETIF_F_HW_CSUM)
3526 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3527 switch (skb->csum_offset) {
3528 case offsetof(struct tcphdr, check):
3529 case offsetof(struct udphdr, check):
3534 return skb_checksum_help(skb);
3536 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3538 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3540 netdev_features_t features;
3542 features = netif_skb_features(skb);
3543 skb = validate_xmit_vlan(skb, features);
3547 skb = sk_validate_xmit_skb(skb, dev);
3551 if (netif_needs_gso(skb, features)) {
3552 struct sk_buff *segs;
3554 segs = skb_gso_segment(skb, features);
3562 if (skb_needs_linearize(skb, features) &&
3563 __skb_linearize(skb))
3566 /* If packet is not checksummed and device does not
3567 * support checksumming for this protocol, complete
3568 * checksumming here.
3570 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3571 if (skb->encapsulation)
3572 skb_set_inner_transport_header(skb,
3573 skb_checksum_start_offset(skb));
3575 skb_set_transport_header(skb,
3576 skb_checksum_start_offset(skb));
3577 if (skb_csum_hwoffload_help(skb, features))
3582 skb = validate_xmit_xfrm(skb, features, again);
3589 atomic_long_inc(&dev->tx_dropped);
3593 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3595 struct sk_buff *next, *head = NULL, *tail;
3597 for (; skb != NULL; skb = next) {
3599 skb_mark_not_on_list(skb);
3601 /* in case skb wont be segmented, point to itself */
3604 skb = validate_xmit_skb(skb, dev, again);
3612 /* If skb was segmented, skb->prev points to
3613 * the last segment. If not, it still contains skb.
3619 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3621 static void qdisc_pkt_len_init(struct sk_buff *skb)
3623 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3625 qdisc_skb_cb(skb)->pkt_len = skb->len;
3627 /* To get more precise estimation of bytes sent on wire,
3628 * we add to pkt_len the headers size of all segments
3630 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3631 unsigned int hdr_len;
3632 u16 gso_segs = shinfo->gso_segs;
3634 /* mac layer + network layer */
3635 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3637 /* + transport layer */
3638 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3639 const struct tcphdr *th;
3640 struct tcphdr _tcphdr;
3642 th = skb_header_pointer(skb, skb_transport_offset(skb),
3643 sizeof(_tcphdr), &_tcphdr);
3645 hdr_len += __tcp_hdrlen(th);
3647 struct udphdr _udphdr;
3649 if (skb_header_pointer(skb, skb_transport_offset(skb),
3650 sizeof(_udphdr), &_udphdr))
3651 hdr_len += sizeof(struct udphdr);
3654 if (shinfo->gso_type & SKB_GSO_DODGY)
3655 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3658 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3662 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3663 struct sk_buff **to_free,
3664 struct netdev_queue *txq)
3668 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3669 if (rc == NET_XMIT_SUCCESS)
3670 trace_qdisc_enqueue(q, txq, skb);
3674 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3675 struct net_device *dev,
3676 struct netdev_queue *txq)
3678 spinlock_t *root_lock = qdisc_lock(q);
3679 struct sk_buff *to_free = NULL;
3683 qdisc_calculate_pkt_len(skb, q);
3685 if (q->flags & TCQ_F_NOLOCK) {
3686 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3687 qdisc_run_begin(q)) {
3688 /* Retest nolock_qdisc_is_empty() within the protection
3689 * of q->seqlock to protect from racing with requeuing.
3691 if (unlikely(!nolock_qdisc_is_empty(q))) {
3692 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3699 qdisc_bstats_cpu_update(q, skb);
3700 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3701 !nolock_qdisc_is_empty(q))
3705 return NET_XMIT_SUCCESS;
3708 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3712 if (unlikely(to_free))
3713 kfree_skb_list(to_free);
3718 * Heuristic to force contended enqueues to serialize on a
3719 * separate lock before trying to get qdisc main lock.
3720 * This permits qdisc->running owner to get the lock more
3721 * often and dequeue packets faster.
3723 contended = qdisc_is_running(q);
3724 if (unlikely(contended))
3725 spin_lock(&q->busylock);
3727 spin_lock(root_lock);
3728 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3729 __qdisc_drop(skb, &to_free);
3731 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3732 qdisc_run_begin(q)) {
3734 * This is a work-conserving queue; there are no old skbs
3735 * waiting to be sent out; and the qdisc is not running -
3736 * xmit the skb directly.
3739 qdisc_bstats_update(q, skb);
3741 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3742 if (unlikely(contended)) {
3743 spin_unlock(&q->busylock);
3750 rc = NET_XMIT_SUCCESS;
3752 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3753 if (qdisc_run_begin(q)) {
3754 if (unlikely(contended)) {
3755 spin_unlock(&q->busylock);
3762 spin_unlock(root_lock);
3763 if (unlikely(to_free))
3764 kfree_skb_list(to_free);
3765 if (unlikely(contended))
3766 spin_unlock(&q->busylock);
3770 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3771 static void skb_update_prio(struct sk_buff *skb)
3773 const struct netprio_map *map;
3774 const struct sock *sk;
3775 unsigned int prioidx;
3779 map = rcu_dereference_bh(skb->dev->priomap);
3782 sk = skb_to_full_sk(skb);
3786 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3788 if (prioidx < map->priomap_len)
3789 skb->priority = map->priomap[prioidx];
3792 #define skb_update_prio(skb)
3796 * dev_loopback_xmit - loop back @skb
3797 * @net: network namespace this loopback is happening in
3798 * @sk: sk needed to be a netfilter okfn
3799 * @skb: buffer to transmit
3801 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3803 skb_reset_mac_header(skb);
3804 __skb_pull(skb, skb_network_offset(skb));
3805 skb->pkt_type = PACKET_LOOPBACK;
3806 if (skb->ip_summed == CHECKSUM_NONE)
3807 skb->ip_summed = CHECKSUM_UNNECESSARY;
3808 WARN_ON(!skb_dst(skb));
3813 EXPORT_SYMBOL(dev_loopback_xmit);
3815 #ifdef CONFIG_NET_EGRESS
3816 static struct sk_buff *
3817 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3819 #ifdef CONFIG_NET_CLS_ACT
3820 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3821 struct tcf_result cl_res;
3826 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3827 qdisc_skb_cb(skb)->mru = 0;
3828 qdisc_skb_cb(skb)->post_ct = false;
3829 mini_qdisc_bstats_cpu_update(miniq, skb);
3831 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3833 case TC_ACT_RECLASSIFY:
3834 skb->tc_index = TC_H_MIN(cl_res.classid);
3837 mini_qdisc_qstats_cpu_drop(miniq);
3838 *ret = NET_XMIT_DROP;
3844 *ret = NET_XMIT_SUCCESS;
3847 case TC_ACT_REDIRECT:
3848 /* No need to push/pop skb's mac_header here on egress! */
3849 skb_do_redirect(skb);
3850 *ret = NET_XMIT_SUCCESS;
3855 #endif /* CONFIG_NET_CLS_ACT */
3859 #endif /* CONFIG_NET_EGRESS */
3862 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3863 struct xps_dev_maps *dev_maps, unsigned int tci)
3865 int tc = netdev_get_prio_tc_map(dev, skb->priority);
3866 struct xps_map *map;
3867 int queue_index = -1;
3869 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3872 tci *= dev_maps->num_tc;
3875 map = rcu_dereference(dev_maps->attr_map[tci]);
3878 queue_index = map->queues[0];
3880 queue_index = map->queues[reciprocal_scale(
3881 skb_get_hash(skb), map->len)];
3882 if (unlikely(queue_index >= dev->real_num_tx_queues))
3889 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3890 struct sk_buff *skb)
3893 struct xps_dev_maps *dev_maps;
3894 struct sock *sk = skb->sk;
3895 int queue_index = -1;
3897 if (!static_key_false(&xps_needed))
3901 if (!static_key_false(&xps_rxqs_needed))
3904 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
3906 int tci = sk_rx_queue_get(sk);
3909 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3914 if (queue_index < 0) {
3915 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
3917 unsigned int tci = skb->sender_cpu - 1;
3919 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3931 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3932 struct net_device *sb_dev)
3936 EXPORT_SYMBOL(dev_pick_tx_zero);
3938 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3939 struct net_device *sb_dev)
3941 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3943 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3945 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3946 struct net_device *sb_dev)
3948 struct sock *sk = skb->sk;
3949 int queue_index = sk_tx_queue_get(sk);
3951 sb_dev = sb_dev ? : dev;
3953 if (queue_index < 0 || skb->ooo_okay ||
3954 queue_index >= dev->real_num_tx_queues) {
3955 int new_index = get_xps_queue(dev, sb_dev, skb);
3958 new_index = skb_tx_hash(dev, sb_dev, skb);
3960 if (queue_index != new_index && sk &&
3962 rcu_access_pointer(sk->sk_dst_cache))
3963 sk_tx_queue_set(sk, new_index);
3965 queue_index = new_index;
3970 EXPORT_SYMBOL(netdev_pick_tx);
3972 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3973 struct sk_buff *skb,
3974 struct net_device *sb_dev)
3976 int queue_index = 0;
3979 u32 sender_cpu = skb->sender_cpu - 1;
3981 if (sender_cpu >= (u32)NR_CPUS)
3982 skb->sender_cpu = raw_smp_processor_id() + 1;
3985 if (dev->real_num_tx_queues != 1) {
3986 const struct net_device_ops *ops = dev->netdev_ops;
3988 if (ops->ndo_select_queue)
3989 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3991 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3993 queue_index = netdev_cap_txqueue(dev, queue_index);
3996 skb_set_queue_mapping(skb, queue_index);
3997 return netdev_get_tx_queue(dev, queue_index);
4001 * __dev_queue_xmit - transmit a buffer
4002 * @skb: buffer to transmit
4003 * @sb_dev: suboordinate device used for L2 forwarding offload
4005 * Queue a buffer for transmission to a network device. The caller must
4006 * have set the device and priority and built the buffer before calling
4007 * this function. The function can be called from an interrupt.
4009 * A negative errno code is returned on a failure. A success does not
4010 * guarantee the frame will be transmitted as it may be dropped due
4011 * to congestion or traffic shaping.
4013 * -----------------------------------------------------------------------------------
4014 * I notice this method can also return errors from the queue disciplines,
4015 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4018 * Regardless of the return value, the skb is consumed, so it is currently
4019 * difficult to retry a send to this method. (You can bump the ref count
4020 * before sending to hold a reference for retry if you are careful.)
4022 * When calling this method, interrupts MUST be enabled. This is because
4023 * the BH enable code must have IRQs enabled so that it will not deadlock.
4026 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4028 struct net_device *dev = skb->dev;
4029 struct netdev_queue *txq;
4034 skb_reset_mac_header(skb);
4036 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4037 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4039 /* Disable soft irqs for various locks below. Also
4040 * stops preemption for RCU.
4044 skb_update_prio(skb);
4046 qdisc_pkt_len_init(skb);
4047 #ifdef CONFIG_NET_CLS_ACT
4048 skb->tc_at_ingress = 0;
4050 #ifdef CONFIG_NET_EGRESS
4051 if (static_branch_unlikely(&egress_needed_key)) {
4052 if (nf_hook_egress_active()) {
4053 skb = nf_hook_egress(skb, &rc, dev);
4057 nf_skip_egress(skb, true);
4058 skb = sch_handle_egress(skb, &rc, dev);
4061 nf_skip_egress(skb, false);
4064 /* If device/qdisc don't need skb->dst, release it right now while
4065 * its hot in this cpu cache.
4067 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4072 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4073 q = rcu_dereference_bh(txq->qdisc);
4075 trace_net_dev_queue(skb);
4077 rc = __dev_xmit_skb(skb, q, dev, txq);
4081 /* The device has no queue. Common case for software devices:
4082 * loopback, all the sorts of tunnels...
4084 * Really, it is unlikely that netif_tx_lock protection is necessary
4085 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4087 * However, it is possible, that they rely on protection
4090 * Check this and shot the lock. It is not prone from deadlocks.
4091 *Either shot noqueue qdisc, it is even simpler 8)
4093 if (dev->flags & IFF_UP) {
4094 int cpu = smp_processor_id(); /* ok because BHs are off */
4096 if (txq->xmit_lock_owner != cpu) {
4097 if (dev_xmit_recursion())
4098 goto recursion_alert;
4100 skb = validate_xmit_skb(skb, dev, &again);
4104 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4105 HARD_TX_LOCK(dev, txq, cpu);
4107 if (!netif_xmit_stopped(txq)) {
4108 dev_xmit_recursion_inc();
4109 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4110 dev_xmit_recursion_dec();
4111 if (dev_xmit_complete(rc)) {
4112 HARD_TX_UNLOCK(dev, txq);
4116 HARD_TX_UNLOCK(dev, txq);
4117 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4120 /* Recursion is detected! It is possible,
4124 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4130 rcu_read_unlock_bh();
4132 atomic_long_inc(&dev->tx_dropped);
4133 kfree_skb_list(skb);
4136 rcu_read_unlock_bh();
4140 int dev_queue_xmit(struct sk_buff *skb)
4142 return __dev_queue_xmit(skb, NULL);
4144 EXPORT_SYMBOL(dev_queue_xmit);
4146 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4148 return __dev_queue_xmit(skb, sb_dev);
4150 EXPORT_SYMBOL(dev_queue_xmit_accel);
4152 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4154 struct net_device *dev = skb->dev;
4155 struct sk_buff *orig_skb = skb;
4156 struct netdev_queue *txq;
4157 int ret = NETDEV_TX_BUSY;
4160 if (unlikely(!netif_running(dev) ||
4161 !netif_carrier_ok(dev)))
4164 skb = validate_xmit_skb_list(skb, dev, &again);
4165 if (skb != orig_skb)
4168 skb_set_queue_mapping(skb, queue_id);
4169 txq = skb_get_tx_queue(dev, skb);
4170 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4174 dev_xmit_recursion_inc();
4175 HARD_TX_LOCK(dev, txq, smp_processor_id());
4176 if (!netif_xmit_frozen_or_drv_stopped(txq))
4177 ret = netdev_start_xmit(skb, dev, txq, false);
4178 HARD_TX_UNLOCK(dev, txq);
4179 dev_xmit_recursion_dec();
4184 atomic_long_inc(&dev->tx_dropped);
4185 kfree_skb_list(skb);
4186 return NET_XMIT_DROP;
4188 EXPORT_SYMBOL(__dev_direct_xmit);
4190 /*************************************************************************
4192 *************************************************************************/
4194 int netdev_max_backlog __read_mostly = 1000;
4195 EXPORT_SYMBOL(netdev_max_backlog);
4197 int netdev_tstamp_prequeue __read_mostly = 1;
4198 int netdev_budget __read_mostly = 300;
4199 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4200 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4201 int weight_p __read_mostly = 64; /* old backlog weight */
4202 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4203 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4204 int dev_rx_weight __read_mostly = 64;
4205 int dev_tx_weight __read_mostly = 64;
4207 /* Called with irq disabled */
4208 static inline void ____napi_schedule(struct softnet_data *sd,
4209 struct napi_struct *napi)
4211 struct task_struct *thread;
4213 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4214 /* Paired with smp_mb__before_atomic() in
4215 * napi_enable()/dev_set_threaded().
4216 * Use READ_ONCE() to guarantee a complete
4217 * read on napi->thread. Only call
4218 * wake_up_process() when it's not NULL.
4220 thread = READ_ONCE(napi->thread);
4222 /* Avoid doing set_bit() if the thread is in
4223 * INTERRUPTIBLE state, cause napi_thread_wait()
4224 * makes sure to proceed with napi polling
4225 * if the thread is explicitly woken from here.
4227 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4228 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4229 wake_up_process(thread);
4234 list_add_tail(&napi->poll_list, &sd->poll_list);
4235 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4240 /* One global table that all flow-based protocols share. */
4241 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4242 EXPORT_SYMBOL(rps_sock_flow_table);
4243 u32 rps_cpu_mask __read_mostly;
4244 EXPORT_SYMBOL(rps_cpu_mask);
4246 struct static_key_false rps_needed __read_mostly;
4247 EXPORT_SYMBOL(rps_needed);
4248 struct static_key_false rfs_needed __read_mostly;
4249 EXPORT_SYMBOL(rfs_needed);
4251 static struct rps_dev_flow *
4252 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4253 struct rps_dev_flow *rflow, u16 next_cpu)
4255 if (next_cpu < nr_cpu_ids) {
4256 #ifdef CONFIG_RFS_ACCEL
4257 struct netdev_rx_queue *rxqueue;
4258 struct rps_dev_flow_table *flow_table;
4259 struct rps_dev_flow *old_rflow;
4264 /* Should we steer this flow to a different hardware queue? */
4265 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4266 !(dev->features & NETIF_F_NTUPLE))
4268 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4269 if (rxq_index == skb_get_rx_queue(skb))
4272 rxqueue = dev->_rx + rxq_index;
4273 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4276 flow_id = skb_get_hash(skb) & flow_table->mask;
4277 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4278 rxq_index, flow_id);
4282 rflow = &flow_table->flows[flow_id];
4284 if (old_rflow->filter == rflow->filter)
4285 old_rflow->filter = RPS_NO_FILTER;
4289 per_cpu(softnet_data, next_cpu).input_queue_head;
4292 rflow->cpu = next_cpu;
4297 * get_rps_cpu is called from netif_receive_skb and returns the target
4298 * CPU from the RPS map of the receiving queue for a given skb.
4299 * rcu_read_lock must be held on entry.
4301 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4302 struct rps_dev_flow **rflowp)
4304 const struct rps_sock_flow_table *sock_flow_table;
4305 struct netdev_rx_queue *rxqueue = dev->_rx;
4306 struct rps_dev_flow_table *flow_table;
4307 struct rps_map *map;
4312 if (skb_rx_queue_recorded(skb)) {
4313 u16 index = skb_get_rx_queue(skb);
4315 if (unlikely(index >= dev->real_num_rx_queues)) {
4316 WARN_ONCE(dev->real_num_rx_queues > 1,
4317 "%s received packet on queue %u, but number "
4318 "of RX queues is %u\n",
4319 dev->name, index, dev->real_num_rx_queues);
4325 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4327 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4328 map = rcu_dereference(rxqueue->rps_map);
4329 if (!flow_table && !map)
4332 skb_reset_network_header(skb);
4333 hash = skb_get_hash(skb);
4337 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4338 if (flow_table && sock_flow_table) {
4339 struct rps_dev_flow *rflow;
4343 /* First check into global flow table if there is a match */
4344 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4345 if ((ident ^ hash) & ~rps_cpu_mask)
4348 next_cpu = ident & rps_cpu_mask;
4350 /* OK, now we know there is a match,
4351 * we can look at the local (per receive queue) flow table
4353 rflow = &flow_table->flows[hash & flow_table->mask];
4357 * If the desired CPU (where last recvmsg was done) is
4358 * different from current CPU (one in the rx-queue flow
4359 * table entry), switch if one of the following holds:
4360 * - Current CPU is unset (>= nr_cpu_ids).
4361 * - Current CPU is offline.
4362 * - The current CPU's queue tail has advanced beyond the
4363 * last packet that was enqueued using this table entry.
4364 * This guarantees that all previous packets for the flow
4365 * have been dequeued, thus preserving in order delivery.
4367 if (unlikely(tcpu != next_cpu) &&
4368 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4369 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4370 rflow->last_qtail)) >= 0)) {
4372 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4375 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4385 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4386 if (cpu_online(tcpu)) {
4396 #ifdef CONFIG_RFS_ACCEL
4399 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4400 * @dev: Device on which the filter was set
4401 * @rxq_index: RX queue index
4402 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4403 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4405 * Drivers that implement ndo_rx_flow_steer() should periodically call
4406 * this function for each installed filter and remove the filters for
4407 * which it returns %true.
4409 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4410 u32 flow_id, u16 filter_id)
4412 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4413 struct rps_dev_flow_table *flow_table;
4414 struct rps_dev_flow *rflow;
4419 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4420 if (flow_table && flow_id <= flow_table->mask) {
4421 rflow = &flow_table->flows[flow_id];
4422 cpu = READ_ONCE(rflow->cpu);
4423 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4424 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4425 rflow->last_qtail) <
4426 (int)(10 * flow_table->mask)))
4432 EXPORT_SYMBOL(rps_may_expire_flow);
4434 #endif /* CONFIG_RFS_ACCEL */
4436 /* Called from hardirq (IPI) context */
4437 static void rps_trigger_softirq(void *data)
4439 struct softnet_data *sd = data;
4441 ____napi_schedule(sd, &sd->backlog);
4445 #endif /* CONFIG_RPS */
4448 * Check if this softnet_data structure is another cpu one
4449 * If yes, queue it to our IPI list and return 1
4452 static int rps_ipi_queued(struct softnet_data *sd)
4455 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4458 sd->rps_ipi_next = mysd->rps_ipi_list;
4459 mysd->rps_ipi_list = sd;
4461 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4464 #endif /* CONFIG_RPS */
4468 #ifdef CONFIG_NET_FLOW_LIMIT
4469 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4472 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4474 #ifdef CONFIG_NET_FLOW_LIMIT
4475 struct sd_flow_limit *fl;
4476 struct softnet_data *sd;
4477 unsigned int old_flow, new_flow;
4479 if (qlen < (netdev_max_backlog >> 1))
4482 sd = this_cpu_ptr(&softnet_data);
4485 fl = rcu_dereference(sd->flow_limit);
4487 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4488 old_flow = fl->history[fl->history_head];
4489 fl->history[fl->history_head] = new_flow;
4492 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4494 if (likely(fl->buckets[old_flow]))
4495 fl->buckets[old_flow]--;
4497 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4509 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4510 * queue (may be a remote CPU queue).
4512 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4513 unsigned int *qtail)
4515 struct softnet_data *sd;
4516 unsigned long flags;
4519 sd = &per_cpu(softnet_data, cpu);
4521 local_irq_save(flags);
4524 if (!netif_running(skb->dev))
4526 qlen = skb_queue_len(&sd->input_pkt_queue);
4527 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4530 __skb_queue_tail(&sd->input_pkt_queue, skb);
4531 input_queue_tail_incr_save(sd, qtail);
4533 local_irq_restore(flags);
4534 return NET_RX_SUCCESS;
4537 /* Schedule NAPI for backlog device
4538 * We can use non atomic operation since we own the queue lock
4540 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4541 if (!rps_ipi_queued(sd))
4542 ____napi_schedule(sd, &sd->backlog);
4551 local_irq_restore(flags);
4553 atomic_long_inc(&skb->dev->rx_dropped);
4558 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4560 struct net_device *dev = skb->dev;
4561 struct netdev_rx_queue *rxqueue;
4565 if (skb_rx_queue_recorded(skb)) {
4566 u16 index = skb_get_rx_queue(skb);
4568 if (unlikely(index >= dev->real_num_rx_queues)) {
4569 WARN_ONCE(dev->real_num_rx_queues > 1,
4570 "%s received packet on queue %u, but number "
4571 "of RX queues is %u\n",
4572 dev->name, index, dev->real_num_rx_queues);
4574 return rxqueue; /* Return first rxqueue */
4581 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4582 struct bpf_prog *xdp_prog)
4584 void *orig_data, *orig_data_end, *hard_start;
4585 struct netdev_rx_queue *rxqueue;
4586 bool orig_bcast, orig_host;
4587 u32 mac_len, frame_sz;
4588 __be16 orig_eth_type;
4593 /* The XDP program wants to see the packet starting at the MAC
4596 mac_len = skb->data - skb_mac_header(skb);
4597 hard_start = skb->data - skb_headroom(skb);
4599 /* SKB "head" area always have tailroom for skb_shared_info */
4600 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4601 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4603 rxqueue = netif_get_rxqueue(skb);
4604 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4605 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4606 skb_headlen(skb) + mac_len, true);
4608 orig_data_end = xdp->data_end;
4609 orig_data = xdp->data;
4610 eth = (struct ethhdr *)xdp->data;
4611 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4612 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4613 orig_eth_type = eth->h_proto;
4615 act = bpf_prog_run_xdp(xdp_prog, xdp);
4617 /* check if bpf_xdp_adjust_head was used */
4618 off = xdp->data - orig_data;
4621 __skb_pull(skb, off);
4623 __skb_push(skb, -off);
4625 skb->mac_header += off;
4626 skb_reset_network_header(skb);
4629 /* check if bpf_xdp_adjust_tail was used */
4630 off = xdp->data_end - orig_data_end;
4632 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4633 skb->len += off; /* positive on grow, negative on shrink */
4636 /* check if XDP changed eth hdr such SKB needs update */
4637 eth = (struct ethhdr *)xdp->data;
4638 if ((orig_eth_type != eth->h_proto) ||
4639 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4640 skb->dev->dev_addr)) ||
4641 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4642 __skb_push(skb, ETH_HLEN);
4643 skb->pkt_type = PACKET_HOST;
4644 skb->protocol = eth_type_trans(skb, skb->dev);
4647 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4648 * before calling us again on redirect path. We do not call do_redirect
4649 * as we leave that up to the caller.
4651 * Caller is responsible for managing lifetime of skb (i.e. calling
4652 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4657 __skb_push(skb, mac_len);
4660 metalen = xdp->data - xdp->data_meta;
4662 skb_metadata_set(skb, metalen);
4669 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4670 struct xdp_buff *xdp,
4671 struct bpf_prog *xdp_prog)
4675 /* Reinjected packets coming from act_mirred or similar should
4676 * not get XDP generic processing.
4678 if (skb_is_redirected(skb))
4681 /* XDP packets must be linear and must have sufficient headroom
4682 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4683 * native XDP provides, thus we need to do it here as well.
4685 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4686 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4687 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4688 int troom = skb->tail + skb->data_len - skb->end;
4690 /* In case we have to go down the path and also linearize,
4691 * then lets do the pskb_expand_head() work just once here.
4693 if (pskb_expand_head(skb,
4694 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4695 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4697 if (skb_linearize(skb))
4701 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4708 bpf_warn_invalid_xdp_action(act);
4711 trace_xdp_exception(skb->dev, xdp_prog, act);
4722 /* When doing generic XDP we have to bypass the qdisc layer and the
4723 * network taps in order to match in-driver-XDP behavior.
4725 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4727 struct net_device *dev = skb->dev;
4728 struct netdev_queue *txq;
4729 bool free_skb = true;
4732 txq = netdev_core_pick_tx(dev, skb, NULL);
4733 cpu = smp_processor_id();
4734 HARD_TX_LOCK(dev, txq, cpu);
4735 if (!netif_xmit_stopped(txq)) {
4736 rc = netdev_start_xmit(skb, dev, txq, 0);
4737 if (dev_xmit_complete(rc))
4740 HARD_TX_UNLOCK(dev, txq);
4742 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4747 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4749 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4752 struct xdp_buff xdp;
4756 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4757 if (act != XDP_PASS) {
4760 err = xdp_do_generic_redirect(skb->dev, skb,
4766 generic_xdp_tx(skb, xdp_prog);
4777 EXPORT_SYMBOL_GPL(do_xdp_generic);
4779 static int netif_rx_internal(struct sk_buff *skb)
4783 net_timestamp_check(netdev_tstamp_prequeue, skb);
4785 trace_netif_rx(skb);
4788 if (static_branch_unlikely(&rps_needed)) {
4789 struct rps_dev_flow voidflow, *rflow = &voidflow;
4795 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4797 cpu = smp_processor_id();
4799 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4808 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4815 * netif_rx - post buffer to the network code
4816 * @skb: buffer to post
4818 * This function receives a packet from a device driver and queues it for
4819 * the upper (protocol) levels to process. It always succeeds. The buffer
4820 * may be dropped during processing for congestion control or by the
4824 * NET_RX_SUCCESS (no congestion)
4825 * NET_RX_DROP (packet was dropped)
4829 int netif_rx(struct sk_buff *skb)
4833 trace_netif_rx_entry(skb);
4835 ret = netif_rx_internal(skb);
4836 trace_netif_rx_exit(ret);
4840 EXPORT_SYMBOL(netif_rx);
4842 int netif_rx_ni(struct sk_buff *skb)
4846 trace_netif_rx_ni_entry(skb);
4849 err = netif_rx_internal(skb);
4850 if (local_softirq_pending())
4853 trace_netif_rx_ni_exit(err);
4857 EXPORT_SYMBOL(netif_rx_ni);
4859 int netif_rx_any_context(struct sk_buff *skb)
4862 * If invoked from contexts which do not invoke bottom half
4863 * processing either at return from interrupt or when softrqs are
4864 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4868 return netif_rx(skb);
4870 return netif_rx_ni(skb);
4872 EXPORT_SYMBOL(netif_rx_any_context);
4874 static __latent_entropy void net_tx_action(struct softirq_action *h)
4876 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4878 if (sd->completion_queue) {
4879 struct sk_buff *clist;
4881 local_irq_disable();
4882 clist = sd->completion_queue;
4883 sd->completion_queue = NULL;
4887 struct sk_buff *skb = clist;
4889 clist = clist->next;
4891 WARN_ON(refcount_read(&skb->users));
4892 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4893 trace_consume_skb(skb);
4895 trace_kfree_skb(skb, net_tx_action);
4897 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4900 __kfree_skb_defer(skb);
4904 if (sd->output_queue) {
4907 local_irq_disable();
4908 head = sd->output_queue;
4909 sd->output_queue = NULL;
4910 sd->output_queue_tailp = &sd->output_queue;
4916 struct Qdisc *q = head;
4917 spinlock_t *root_lock = NULL;
4919 head = head->next_sched;
4921 /* We need to make sure head->next_sched is read
4922 * before clearing __QDISC_STATE_SCHED
4924 smp_mb__before_atomic();
4926 if (!(q->flags & TCQ_F_NOLOCK)) {
4927 root_lock = qdisc_lock(q);
4928 spin_lock(root_lock);
4929 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
4931 /* There is a synchronize_net() between
4932 * STATE_DEACTIVATED flag being set and
4933 * qdisc_reset()/some_qdisc_is_busy() in
4934 * dev_deactivate(), so we can safely bail out
4935 * early here to avoid data race between
4936 * qdisc_deactivate() and some_qdisc_is_busy()
4937 * for lockless qdisc.
4939 clear_bit(__QDISC_STATE_SCHED, &q->state);
4943 clear_bit(__QDISC_STATE_SCHED, &q->state);
4946 spin_unlock(root_lock);
4952 xfrm_dev_backlog(sd);
4955 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4956 /* This hook is defined here for ATM LANE */
4957 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4958 unsigned char *addr) __read_mostly;
4959 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4962 static inline struct sk_buff *
4963 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4964 struct net_device *orig_dev, bool *another)
4966 #ifdef CONFIG_NET_CLS_ACT
4967 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4968 struct tcf_result cl_res;
4970 /* If there's at least one ingress present somewhere (so
4971 * we get here via enabled static key), remaining devices
4972 * that are not configured with an ingress qdisc will bail
4979 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4983 qdisc_skb_cb(skb)->pkt_len = skb->len;
4984 qdisc_skb_cb(skb)->mru = 0;
4985 qdisc_skb_cb(skb)->post_ct = false;
4986 skb->tc_at_ingress = 1;
4987 mini_qdisc_bstats_cpu_update(miniq, skb);
4989 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
4991 case TC_ACT_RECLASSIFY:
4992 skb->tc_index = TC_H_MIN(cl_res.classid);
4995 mini_qdisc_qstats_cpu_drop(miniq);
5003 case TC_ACT_REDIRECT:
5004 /* skb_mac_header check was done by cls/act_bpf, so
5005 * we can safely push the L2 header back before
5006 * redirecting to another netdev
5008 __skb_push(skb, skb->mac_len);
5009 if (skb_do_redirect(skb) == -EAGAIN) {
5010 __skb_pull(skb, skb->mac_len);
5015 case TC_ACT_CONSUMED:
5020 #endif /* CONFIG_NET_CLS_ACT */
5025 * netdev_is_rx_handler_busy - check if receive handler is registered
5026 * @dev: device to check
5028 * Check if a receive handler is already registered for a given device.
5029 * Return true if there one.
5031 * The caller must hold the rtnl_mutex.
5033 bool netdev_is_rx_handler_busy(struct net_device *dev)
5036 return dev && rtnl_dereference(dev->rx_handler);
5038 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5041 * netdev_rx_handler_register - register receive handler
5042 * @dev: device to register a handler for
5043 * @rx_handler: receive handler to register
5044 * @rx_handler_data: data pointer that is used by rx handler
5046 * Register a receive handler for a device. This handler will then be
5047 * called from __netif_receive_skb. A negative errno code is returned
5050 * The caller must hold the rtnl_mutex.
5052 * For a general description of rx_handler, see enum rx_handler_result.
5054 int netdev_rx_handler_register(struct net_device *dev,
5055 rx_handler_func_t *rx_handler,
5056 void *rx_handler_data)
5058 if (netdev_is_rx_handler_busy(dev))
5061 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5064 /* Note: rx_handler_data must be set before rx_handler */
5065 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5066 rcu_assign_pointer(dev->rx_handler, rx_handler);
5070 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5073 * netdev_rx_handler_unregister - unregister receive handler
5074 * @dev: device to unregister a handler from
5076 * Unregister a receive handler from a device.
5078 * The caller must hold the rtnl_mutex.
5080 void netdev_rx_handler_unregister(struct net_device *dev)
5084 RCU_INIT_POINTER(dev->rx_handler, NULL);
5085 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5086 * section has a guarantee to see a non NULL rx_handler_data
5090 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5092 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5095 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5096 * the special handling of PFMEMALLOC skbs.
5098 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5100 switch (skb->protocol) {
5101 case htons(ETH_P_ARP):
5102 case htons(ETH_P_IP):
5103 case htons(ETH_P_IPV6):
5104 case htons(ETH_P_8021Q):
5105 case htons(ETH_P_8021AD):
5112 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5113 int *ret, struct net_device *orig_dev)
5115 if (nf_hook_ingress_active(skb)) {
5119 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5124 ingress_retval = nf_hook_ingress(skb);
5126 return ingress_retval;
5131 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5132 struct packet_type **ppt_prev)
5134 struct packet_type *ptype, *pt_prev;
5135 rx_handler_func_t *rx_handler;
5136 struct sk_buff *skb = *pskb;
5137 struct net_device *orig_dev;
5138 bool deliver_exact = false;
5139 int ret = NET_RX_DROP;
5142 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5144 trace_netif_receive_skb(skb);
5146 orig_dev = skb->dev;
5148 skb_reset_network_header(skb);
5149 if (!skb_transport_header_was_set(skb))
5150 skb_reset_transport_header(skb);
5151 skb_reset_mac_len(skb);
5156 skb->skb_iif = skb->dev->ifindex;
5158 __this_cpu_inc(softnet_data.processed);
5160 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5164 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5167 if (ret2 != XDP_PASS) {
5173 if (eth_type_vlan(skb->protocol)) {
5174 skb = skb_vlan_untag(skb);
5179 if (skb_skip_tc_classify(skb))
5185 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5187 ret = deliver_skb(skb, pt_prev, orig_dev);
5191 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5193 ret = deliver_skb(skb, pt_prev, orig_dev);
5198 #ifdef CONFIG_NET_INGRESS
5199 if (static_branch_unlikely(&ingress_needed_key)) {
5200 bool another = false;
5202 nf_skip_egress(skb, true);
5203 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5210 nf_skip_egress(skb, false);
5211 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5215 skb_reset_redirect(skb);
5217 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5220 if (skb_vlan_tag_present(skb)) {
5222 ret = deliver_skb(skb, pt_prev, orig_dev);
5225 if (vlan_do_receive(&skb))
5227 else if (unlikely(!skb))
5231 rx_handler = rcu_dereference(skb->dev->rx_handler);
5234 ret = deliver_skb(skb, pt_prev, orig_dev);
5237 switch (rx_handler(&skb)) {
5238 case RX_HANDLER_CONSUMED:
5239 ret = NET_RX_SUCCESS;
5241 case RX_HANDLER_ANOTHER:
5243 case RX_HANDLER_EXACT:
5244 deliver_exact = true;
5246 case RX_HANDLER_PASS:
5253 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5255 if (skb_vlan_tag_get_id(skb)) {
5256 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5259 skb->pkt_type = PACKET_OTHERHOST;
5260 } else if (eth_type_vlan(skb->protocol)) {
5261 /* Outer header is 802.1P with vlan 0, inner header is
5262 * 802.1Q or 802.1AD and vlan_do_receive() above could
5263 * not find vlan dev for vlan id 0.
5265 __vlan_hwaccel_clear_tag(skb);
5266 skb = skb_vlan_untag(skb);
5269 if (vlan_do_receive(&skb))
5270 /* After stripping off 802.1P header with vlan 0
5271 * vlan dev is found for inner header.
5274 else if (unlikely(!skb))
5277 /* We have stripped outer 802.1P vlan 0 header.
5278 * But could not find vlan dev.
5279 * check again for vlan id to set OTHERHOST.
5283 /* Note: we might in the future use prio bits
5284 * and set skb->priority like in vlan_do_receive()
5285 * For the time being, just ignore Priority Code Point
5287 __vlan_hwaccel_clear_tag(skb);
5290 type = skb->protocol;
5292 /* deliver only exact match when indicated */
5293 if (likely(!deliver_exact)) {
5294 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5295 &ptype_base[ntohs(type) &
5299 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5300 &orig_dev->ptype_specific);
5302 if (unlikely(skb->dev != orig_dev)) {
5303 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5304 &skb->dev->ptype_specific);
5308 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5310 *ppt_prev = pt_prev;
5314 atomic_long_inc(&skb->dev->rx_dropped);
5316 atomic_long_inc(&skb->dev->rx_nohandler);
5318 /* Jamal, now you will not able to escape explaining
5319 * me how you were going to use this. :-)
5325 /* The invariant here is that if *ppt_prev is not NULL
5326 * then skb should also be non-NULL.
5328 * Apparently *ppt_prev assignment above holds this invariant due to
5329 * skb dereferencing near it.
5335 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5337 struct net_device *orig_dev = skb->dev;
5338 struct packet_type *pt_prev = NULL;
5341 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5343 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5344 skb->dev, pt_prev, orig_dev);
5349 * netif_receive_skb_core - special purpose version of netif_receive_skb
5350 * @skb: buffer to process
5352 * More direct receive version of netif_receive_skb(). It should
5353 * only be used by callers that have a need to skip RPS and Generic XDP.
5354 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5356 * This function may only be called from softirq context and interrupts
5357 * should be enabled.
5359 * Return values (usually ignored):
5360 * NET_RX_SUCCESS: no congestion
5361 * NET_RX_DROP: packet was dropped
5363 int netif_receive_skb_core(struct sk_buff *skb)
5368 ret = __netif_receive_skb_one_core(skb, false);
5373 EXPORT_SYMBOL(netif_receive_skb_core);
5375 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5376 struct packet_type *pt_prev,
5377 struct net_device *orig_dev)
5379 struct sk_buff *skb, *next;
5383 if (list_empty(head))
5385 if (pt_prev->list_func != NULL)
5386 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5387 ip_list_rcv, head, pt_prev, orig_dev);
5389 list_for_each_entry_safe(skb, next, head, list) {
5390 skb_list_del_init(skb);
5391 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5395 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5397 /* Fast-path assumptions:
5398 * - There is no RX handler.
5399 * - Only one packet_type matches.
5400 * If either of these fails, we will end up doing some per-packet
5401 * processing in-line, then handling the 'last ptype' for the whole
5402 * sublist. This can't cause out-of-order delivery to any single ptype,
5403 * because the 'last ptype' must be constant across the sublist, and all
5404 * other ptypes are handled per-packet.
5406 /* Current (common) ptype of sublist */
5407 struct packet_type *pt_curr = NULL;
5408 /* Current (common) orig_dev of sublist */
5409 struct net_device *od_curr = NULL;
5410 struct list_head sublist;
5411 struct sk_buff *skb, *next;
5413 INIT_LIST_HEAD(&sublist);
5414 list_for_each_entry_safe(skb, next, head, list) {
5415 struct net_device *orig_dev = skb->dev;
5416 struct packet_type *pt_prev = NULL;
5418 skb_list_del_init(skb);
5419 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5422 if (pt_curr != pt_prev || od_curr != orig_dev) {
5423 /* dispatch old sublist */
5424 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5425 /* start new sublist */
5426 INIT_LIST_HEAD(&sublist);
5430 list_add_tail(&skb->list, &sublist);
5433 /* dispatch final sublist */
5434 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5437 static int __netif_receive_skb(struct sk_buff *skb)
5441 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5442 unsigned int noreclaim_flag;
5445 * PFMEMALLOC skbs are special, they should
5446 * - be delivered to SOCK_MEMALLOC sockets only
5447 * - stay away from userspace
5448 * - have bounded memory usage
5450 * Use PF_MEMALLOC as this saves us from propagating the allocation
5451 * context down to all allocation sites.
5453 noreclaim_flag = memalloc_noreclaim_save();
5454 ret = __netif_receive_skb_one_core(skb, true);
5455 memalloc_noreclaim_restore(noreclaim_flag);
5457 ret = __netif_receive_skb_one_core(skb, false);
5462 static void __netif_receive_skb_list(struct list_head *head)
5464 unsigned long noreclaim_flag = 0;
5465 struct sk_buff *skb, *next;
5466 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5468 list_for_each_entry_safe(skb, next, head, list) {
5469 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5470 struct list_head sublist;
5472 /* Handle the previous sublist */
5473 list_cut_before(&sublist, head, &skb->list);
5474 if (!list_empty(&sublist))
5475 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5476 pfmemalloc = !pfmemalloc;
5477 /* See comments in __netif_receive_skb */
5479 noreclaim_flag = memalloc_noreclaim_save();
5481 memalloc_noreclaim_restore(noreclaim_flag);
5484 /* Handle the remaining sublist */
5485 if (!list_empty(head))
5486 __netif_receive_skb_list_core(head, pfmemalloc);
5487 /* Restore pflags */
5489 memalloc_noreclaim_restore(noreclaim_flag);
5492 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5494 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5495 struct bpf_prog *new = xdp->prog;
5498 switch (xdp->command) {
5499 case XDP_SETUP_PROG:
5500 rcu_assign_pointer(dev->xdp_prog, new);
5505 static_branch_dec(&generic_xdp_needed_key);
5506 } else if (new && !old) {
5507 static_branch_inc(&generic_xdp_needed_key);
5508 dev_disable_lro(dev);
5509 dev_disable_gro_hw(dev);
5521 static int netif_receive_skb_internal(struct sk_buff *skb)
5525 net_timestamp_check(netdev_tstamp_prequeue, skb);
5527 if (skb_defer_rx_timestamp(skb))
5528 return NET_RX_SUCCESS;
5532 if (static_branch_unlikely(&rps_needed)) {
5533 struct rps_dev_flow voidflow, *rflow = &voidflow;
5534 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5537 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5543 ret = __netif_receive_skb(skb);
5548 void netif_receive_skb_list_internal(struct list_head *head)
5550 struct sk_buff *skb, *next;
5551 struct list_head sublist;
5553 INIT_LIST_HEAD(&sublist);
5554 list_for_each_entry_safe(skb, next, head, list) {
5555 net_timestamp_check(netdev_tstamp_prequeue, skb);
5556 skb_list_del_init(skb);
5557 if (!skb_defer_rx_timestamp(skb))
5558 list_add_tail(&skb->list, &sublist);
5560 list_splice_init(&sublist, head);
5564 if (static_branch_unlikely(&rps_needed)) {
5565 list_for_each_entry_safe(skb, next, head, list) {
5566 struct rps_dev_flow voidflow, *rflow = &voidflow;
5567 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5570 /* Will be handled, remove from list */
5571 skb_list_del_init(skb);
5572 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5577 __netif_receive_skb_list(head);
5582 * netif_receive_skb - process receive buffer from network
5583 * @skb: buffer to process
5585 * netif_receive_skb() is the main receive data processing function.
5586 * It always succeeds. The buffer may be dropped during processing
5587 * for congestion control or by the protocol layers.
5589 * This function may only be called from softirq context and interrupts
5590 * should be enabled.
5592 * Return values (usually ignored):
5593 * NET_RX_SUCCESS: no congestion
5594 * NET_RX_DROP: packet was dropped
5596 int netif_receive_skb(struct sk_buff *skb)
5600 trace_netif_receive_skb_entry(skb);
5602 ret = netif_receive_skb_internal(skb);
5603 trace_netif_receive_skb_exit(ret);
5607 EXPORT_SYMBOL(netif_receive_skb);
5610 * netif_receive_skb_list - process many receive buffers from network
5611 * @head: list of skbs to process.
5613 * Since return value of netif_receive_skb() is normally ignored, and
5614 * wouldn't be meaningful for a list, this function returns void.
5616 * This function may only be called from softirq context and interrupts
5617 * should be enabled.
5619 void netif_receive_skb_list(struct list_head *head)
5621 struct sk_buff *skb;
5623 if (list_empty(head))
5625 if (trace_netif_receive_skb_list_entry_enabled()) {
5626 list_for_each_entry(skb, head, list)
5627 trace_netif_receive_skb_list_entry(skb);
5629 netif_receive_skb_list_internal(head);
5630 trace_netif_receive_skb_list_exit(0);
5632 EXPORT_SYMBOL(netif_receive_skb_list);
5634 static DEFINE_PER_CPU(struct work_struct, flush_works);
5636 /* Network device is going away, flush any packets still pending */
5637 static void flush_backlog(struct work_struct *work)
5639 struct sk_buff *skb, *tmp;
5640 struct softnet_data *sd;
5643 sd = this_cpu_ptr(&softnet_data);
5645 local_irq_disable();
5647 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5648 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5649 __skb_unlink(skb, &sd->input_pkt_queue);
5650 dev_kfree_skb_irq(skb);
5651 input_queue_head_incr(sd);
5657 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5658 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5659 __skb_unlink(skb, &sd->process_queue);
5661 input_queue_head_incr(sd);
5667 static bool flush_required(int cpu)
5669 #if IS_ENABLED(CONFIG_RPS)
5670 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5673 local_irq_disable();
5676 /* as insertion into process_queue happens with the rps lock held,
5677 * process_queue access may race only with dequeue
5679 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5680 !skb_queue_empty_lockless(&sd->process_queue);
5686 /* without RPS we can't safely check input_pkt_queue: during a
5687 * concurrent remote skb_queue_splice() we can detect as empty both
5688 * input_pkt_queue and process_queue even if the latter could end-up
5689 * containing a lot of packets.
5694 static void flush_all_backlogs(void)
5696 static cpumask_t flush_cpus;
5699 /* since we are under rtnl lock protection we can use static data
5700 * for the cpumask and avoid allocating on stack the possibly
5707 cpumask_clear(&flush_cpus);
5708 for_each_online_cpu(cpu) {
5709 if (flush_required(cpu)) {
5710 queue_work_on(cpu, system_highpri_wq,
5711 per_cpu_ptr(&flush_works, cpu));
5712 cpumask_set_cpu(cpu, &flush_cpus);
5716 /* we can have in flight packet[s] on the cpus we are not flushing,
5717 * synchronize_net() in unregister_netdevice_many() will take care of
5720 for_each_cpu(cpu, &flush_cpus)
5721 flush_work(per_cpu_ptr(&flush_works, cpu));
5726 static void net_rps_send_ipi(struct softnet_data *remsd)
5730 struct softnet_data *next = remsd->rps_ipi_next;
5732 if (cpu_online(remsd->cpu))
5733 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5740 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5741 * Note: called with local irq disabled, but exits with local irq enabled.
5743 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5746 struct softnet_data *remsd = sd->rps_ipi_list;
5749 sd->rps_ipi_list = NULL;
5753 /* Send pending IPI's to kick RPS processing on remote cpus. */
5754 net_rps_send_ipi(remsd);
5760 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5763 return sd->rps_ipi_list != NULL;
5769 static int process_backlog(struct napi_struct *napi, int quota)
5771 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5775 /* Check if we have pending ipi, its better to send them now,
5776 * not waiting net_rx_action() end.
5778 if (sd_has_rps_ipi_waiting(sd)) {
5779 local_irq_disable();
5780 net_rps_action_and_irq_enable(sd);
5783 napi->weight = dev_rx_weight;
5785 struct sk_buff *skb;
5787 while ((skb = __skb_dequeue(&sd->process_queue))) {
5789 __netif_receive_skb(skb);
5791 input_queue_head_incr(sd);
5792 if (++work >= quota)
5797 local_irq_disable();
5799 if (skb_queue_empty(&sd->input_pkt_queue)) {
5801 * Inline a custom version of __napi_complete().
5802 * only current cpu owns and manipulates this napi,
5803 * and NAPI_STATE_SCHED is the only possible flag set
5805 * We can use a plain write instead of clear_bit(),
5806 * and we dont need an smp_mb() memory barrier.
5811 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5812 &sd->process_queue);
5822 * __napi_schedule - schedule for receive
5823 * @n: entry to schedule
5825 * The entry's receive function will be scheduled to run.
5826 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5828 void __napi_schedule(struct napi_struct *n)
5830 unsigned long flags;
5832 local_irq_save(flags);
5833 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5834 local_irq_restore(flags);
5836 EXPORT_SYMBOL(__napi_schedule);
5839 * napi_schedule_prep - check if napi can be scheduled
5842 * Test if NAPI routine is already running, and if not mark
5843 * it as running. This is used as a condition variable to
5844 * insure only one NAPI poll instance runs. We also make
5845 * sure there is no pending NAPI disable.
5847 bool napi_schedule_prep(struct napi_struct *n)
5849 unsigned long val, new;
5852 val = READ_ONCE(n->state);
5853 if (unlikely(val & NAPIF_STATE_DISABLE))
5855 new = val | NAPIF_STATE_SCHED;
5857 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5858 * This was suggested by Alexander Duyck, as compiler
5859 * emits better code than :
5860 * if (val & NAPIF_STATE_SCHED)
5861 * new |= NAPIF_STATE_MISSED;
5863 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5865 } while (cmpxchg(&n->state, val, new) != val);
5867 return !(val & NAPIF_STATE_SCHED);
5869 EXPORT_SYMBOL(napi_schedule_prep);
5872 * __napi_schedule_irqoff - schedule for receive
5873 * @n: entry to schedule
5875 * Variant of __napi_schedule() assuming hard irqs are masked.
5877 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5878 * because the interrupt disabled assumption might not be true
5879 * due to force-threaded interrupts and spinlock substitution.
5881 void __napi_schedule_irqoff(struct napi_struct *n)
5883 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5884 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5888 EXPORT_SYMBOL(__napi_schedule_irqoff);
5890 bool napi_complete_done(struct napi_struct *n, int work_done)
5892 unsigned long flags, val, new, timeout = 0;
5896 * 1) Don't let napi dequeue from the cpu poll list
5897 * just in case its running on a different cpu.
5898 * 2) If we are busy polling, do nothing here, we have
5899 * the guarantee we will be called later.
5901 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5902 NAPIF_STATE_IN_BUSY_POLL)))
5907 timeout = READ_ONCE(n->dev->gro_flush_timeout);
5908 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
5910 if (n->defer_hard_irqs_count > 0) {
5911 n->defer_hard_irqs_count--;
5912 timeout = READ_ONCE(n->dev->gro_flush_timeout);
5916 if (n->gro_bitmask) {
5917 /* When the NAPI instance uses a timeout and keeps postponing
5918 * it, we need to bound somehow the time packets are kept in
5921 napi_gro_flush(n, !!timeout);
5926 if (unlikely(!list_empty(&n->poll_list))) {
5927 /* If n->poll_list is not empty, we need to mask irqs */
5928 local_irq_save(flags);
5929 list_del_init(&n->poll_list);
5930 local_irq_restore(flags);
5934 val = READ_ONCE(n->state);
5936 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5938 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
5939 NAPIF_STATE_SCHED_THREADED |
5940 NAPIF_STATE_PREFER_BUSY_POLL);
5942 /* If STATE_MISSED was set, leave STATE_SCHED set,
5943 * because we will call napi->poll() one more time.
5944 * This C code was suggested by Alexander Duyck to help gcc.
5946 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5948 } while (cmpxchg(&n->state, val, new) != val);
5950 if (unlikely(val & NAPIF_STATE_MISSED)) {
5956 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5957 HRTIMER_MODE_REL_PINNED);
5960 EXPORT_SYMBOL(napi_complete_done);
5962 /* must be called under rcu_read_lock(), as we dont take a reference */
5963 static struct napi_struct *napi_by_id(unsigned int napi_id)
5965 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5966 struct napi_struct *napi;
5968 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5969 if (napi->napi_id == napi_id)
5975 #if defined(CONFIG_NET_RX_BUSY_POLL)
5977 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
5979 if (!skip_schedule) {
5980 gro_normal_list(napi);
5981 __napi_schedule(napi);
5985 if (napi->gro_bitmask) {
5986 /* flush too old packets
5987 * If HZ < 1000, flush all packets.
5989 napi_gro_flush(napi, HZ >= 1000);
5992 gro_normal_list(napi);
5993 clear_bit(NAPI_STATE_SCHED, &napi->state);
5996 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
5999 bool skip_schedule = false;
6000 unsigned long timeout;
6003 /* Busy polling means there is a high chance device driver hard irq
6004 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6005 * set in napi_schedule_prep().
6006 * Since we are about to call napi->poll() once more, we can safely
6007 * clear NAPI_STATE_MISSED.
6009 * Note: x86 could use a single "lock and ..." instruction
6010 * to perform these two clear_bit()
6012 clear_bit(NAPI_STATE_MISSED, &napi->state);
6013 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6017 if (prefer_busy_poll) {
6018 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6019 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6020 if (napi->defer_hard_irqs_count && timeout) {
6021 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6022 skip_schedule = true;
6026 /* All we really want here is to re-enable device interrupts.
6027 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6029 rc = napi->poll(napi, budget);
6030 /* We can't gro_normal_list() here, because napi->poll() might have
6031 * rearmed the napi (napi_complete_done()) in which case it could
6032 * already be running on another CPU.
6034 trace_napi_poll(napi, rc, budget);
6035 netpoll_poll_unlock(have_poll_lock);
6037 __busy_poll_stop(napi, skip_schedule);
6041 void napi_busy_loop(unsigned int napi_id,
6042 bool (*loop_end)(void *, unsigned long),
6043 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6045 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6046 int (*napi_poll)(struct napi_struct *napi, int budget);
6047 void *have_poll_lock = NULL;
6048 struct napi_struct *napi;
6055 napi = napi_by_id(napi_id);
6065 unsigned long val = READ_ONCE(napi->state);
6067 /* If multiple threads are competing for this napi,
6068 * we avoid dirtying napi->state as much as we can.
6070 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6071 NAPIF_STATE_IN_BUSY_POLL)) {
6072 if (prefer_busy_poll)
6073 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6076 if (cmpxchg(&napi->state, val,
6077 val | NAPIF_STATE_IN_BUSY_POLL |
6078 NAPIF_STATE_SCHED) != val) {
6079 if (prefer_busy_poll)
6080 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6083 have_poll_lock = netpoll_poll_lock(napi);
6084 napi_poll = napi->poll;
6086 work = napi_poll(napi, budget);
6087 trace_napi_poll(napi, work, budget);
6088 gro_normal_list(napi);
6091 __NET_ADD_STATS(dev_net(napi->dev),
6092 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6095 if (!loop_end || loop_end(loop_end_arg, start_time))
6098 if (unlikely(need_resched())) {
6100 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6104 if (loop_end(loop_end_arg, start_time))
6111 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6116 EXPORT_SYMBOL(napi_busy_loop);
6118 #endif /* CONFIG_NET_RX_BUSY_POLL */
6120 static void napi_hash_add(struct napi_struct *napi)
6122 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6125 spin_lock(&napi_hash_lock);
6127 /* 0..NR_CPUS range is reserved for sender_cpu use */
6129 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6130 napi_gen_id = MIN_NAPI_ID;
6131 } while (napi_by_id(napi_gen_id));
6132 napi->napi_id = napi_gen_id;
6134 hlist_add_head_rcu(&napi->napi_hash_node,
6135 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6137 spin_unlock(&napi_hash_lock);
6140 /* Warning : caller is responsible to make sure rcu grace period
6141 * is respected before freeing memory containing @napi
6143 static void napi_hash_del(struct napi_struct *napi)
6145 spin_lock(&napi_hash_lock);
6147 hlist_del_init_rcu(&napi->napi_hash_node);
6149 spin_unlock(&napi_hash_lock);
6152 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6154 struct napi_struct *napi;
6156 napi = container_of(timer, struct napi_struct, timer);
6158 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6159 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6161 if (!napi_disable_pending(napi) &&
6162 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6163 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6164 __napi_schedule_irqoff(napi);
6167 return HRTIMER_NORESTART;
6170 static void init_gro_hash(struct napi_struct *napi)
6174 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6175 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6176 napi->gro_hash[i].count = 0;
6178 napi->gro_bitmask = 0;
6181 int dev_set_threaded(struct net_device *dev, bool threaded)
6183 struct napi_struct *napi;
6186 if (dev->threaded == threaded)
6190 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6191 if (!napi->thread) {
6192 err = napi_kthread_create(napi);
6201 dev->threaded = threaded;
6203 /* Make sure kthread is created before THREADED bit
6206 smp_mb__before_atomic();
6208 /* Setting/unsetting threaded mode on a napi might not immediately
6209 * take effect, if the current napi instance is actively being
6210 * polled. In this case, the switch between threaded mode and
6211 * softirq mode will happen in the next round of napi_schedule().
6212 * This should not cause hiccups/stalls to the live traffic.
6214 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6216 set_bit(NAPI_STATE_THREADED, &napi->state);
6218 clear_bit(NAPI_STATE_THREADED, &napi->state);
6223 EXPORT_SYMBOL(dev_set_threaded);
6225 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6226 int (*poll)(struct napi_struct *, int), int weight)
6228 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6231 INIT_LIST_HEAD(&napi->poll_list);
6232 INIT_HLIST_NODE(&napi->napi_hash_node);
6233 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6234 napi->timer.function = napi_watchdog;
6235 init_gro_hash(napi);
6237 INIT_LIST_HEAD(&napi->rx_list);
6240 if (weight > NAPI_POLL_WEIGHT)
6241 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6243 napi->weight = weight;
6245 #ifdef CONFIG_NETPOLL
6246 napi->poll_owner = -1;
6248 set_bit(NAPI_STATE_SCHED, &napi->state);
6249 set_bit(NAPI_STATE_NPSVC, &napi->state);
6250 list_add_rcu(&napi->dev_list, &dev->napi_list);
6251 napi_hash_add(napi);
6252 /* Create kthread for this napi if dev->threaded is set.
6253 * Clear dev->threaded if kthread creation failed so that
6254 * threaded mode will not be enabled in napi_enable().
6256 if (dev->threaded && napi_kthread_create(napi))
6259 EXPORT_SYMBOL(netif_napi_add);
6261 void napi_disable(struct napi_struct *n)
6263 unsigned long val, new;
6266 set_bit(NAPI_STATE_DISABLE, &n->state);
6269 val = READ_ONCE(n->state);
6270 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6271 usleep_range(20, 200);
6275 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6276 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6278 if (cmpxchg(&n->state, val, new) == val)
6282 hrtimer_cancel(&n->timer);
6284 clear_bit(NAPI_STATE_DISABLE, &n->state);
6286 EXPORT_SYMBOL(napi_disable);
6289 * napi_enable - enable NAPI scheduling
6292 * Resume NAPI from being scheduled on this context.
6293 * Must be paired with napi_disable.
6295 void napi_enable(struct napi_struct *n)
6297 unsigned long val, new;
6300 val = READ_ONCE(n->state);
6301 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6303 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6304 if (n->dev->threaded && n->thread)
6305 new |= NAPIF_STATE_THREADED;
6306 } while (cmpxchg(&n->state, val, new) != val);
6308 EXPORT_SYMBOL(napi_enable);
6310 static void flush_gro_hash(struct napi_struct *napi)
6314 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6315 struct sk_buff *skb, *n;
6317 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6319 napi->gro_hash[i].count = 0;
6323 /* Must be called in process context */
6324 void __netif_napi_del(struct napi_struct *napi)
6326 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6329 napi_hash_del(napi);
6330 list_del_rcu(&napi->dev_list);
6331 napi_free_frags(napi);
6333 flush_gro_hash(napi);
6334 napi->gro_bitmask = 0;
6337 kthread_stop(napi->thread);
6338 napi->thread = NULL;
6341 EXPORT_SYMBOL(__netif_napi_del);
6343 static int __napi_poll(struct napi_struct *n, bool *repoll)
6349 /* This NAPI_STATE_SCHED test is for avoiding a race
6350 * with netpoll's poll_napi(). Only the entity which
6351 * obtains the lock and sees NAPI_STATE_SCHED set will
6352 * actually make the ->poll() call. Therefore we avoid
6353 * accidentally calling ->poll() when NAPI is not scheduled.
6356 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6357 work = n->poll(n, weight);
6358 trace_napi_poll(n, work, weight);
6361 if (unlikely(work > weight))
6362 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6363 n->poll, work, weight);
6365 if (likely(work < weight))
6368 /* Drivers must not modify the NAPI state if they
6369 * consume the entire weight. In such cases this code
6370 * still "owns" the NAPI instance and therefore can
6371 * move the instance around on the list at-will.
6373 if (unlikely(napi_disable_pending(n))) {
6378 /* The NAPI context has more processing work, but busy-polling
6379 * is preferred. Exit early.
6381 if (napi_prefer_busy_poll(n)) {
6382 if (napi_complete_done(n, work)) {
6383 /* If timeout is not set, we need to make sure
6384 * that the NAPI is re-scheduled.
6391 if (n->gro_bitmask) {
6392 /* flush too old packets
6393 * If HZ < 1000, flush all packets.
6395 napi_gro_flush(n, HZ >= 1000);
6400 /* Some drivers may have called napi_schedule
6401 * prior to exhausting their budget.
6403 if (unlikely(!list_empty(&n->poll_list))) {
6404 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6405 n->dev ? n->dev->name : "backlog");
6414 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6416 bool do_repoll = false;
6420 list_del_init(&n->poll_list);
6422 have = netpoll_poll_lock(n);
6424 work = __napi_poll(n, &do_repoll);
6427 list_add_tail(&n->poll_list, repoll);
6429 netpoll_poll_unlock(have);
6434 static int napi_thread_wait(struct napi_struct *napi)
6438 set_current_state(TASK_INTERRUPTIBLE);
6440 while (!kthread_should_stop()) {
6441 /* Testing SCHED_THREADED bit here to make sure the current
6442 * kthread owns this napi and could poll on this napi.
6443 * Testing SCHED bit is not enough because SCHED bit might be
6444 * set by some other busy poll thread or by napi_disable().
6446 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6447 WARN_ON(!list_empty(&napi->poll_list));
6448 __set_current_state(TASK_RUNNING);
6453 /* woken being true indicates this thread owns this napi. */
6455 set_current_state(TASK_INTERRUPTIBLE);
6457 __set_current_state(TASK_RUNNING);
6462 static int napi_threaded_poll(void *data)
6464 struct napi_struct *napi = data;
6467 while (!napi_thread_wait(napi)) {
6469 bool repoll = false;
6473 have = netpoll_poll_lock(napi);
6474 __napi_poll(napi, &repoll);
6475 netpoll_poll_unlock(have);
6488 static __latent_entropy void net_rx_action(struct softirq_action *h)
6490 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6491 unsigned long time_limit = jiffies +
6492 usecs_to_jiffies(netdev_budget_usecs);
6493 int budget = netdev_budget;
6497 local_irq_disable();
6498 list_splice_init(&sd->poll_list, &list);
6502 struct napi_struct *n;
6504 if (list_empty(&list)) {
6505 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6510 n = list_first_entry(&list, struct napi_struct, poll_list);
6511 budget -= napi_poll(n, &repoll);
6513 /* If softirq window is exhausted then punt.
6514 * Allow this to run for 2 jiffies since which will allow
6515 * an average latency of 1.5/HZ.
6517 if (unlikely(budget <= 0 ||
6518 time_after_eq(jiffies, time_limit))) {
6524 local_irq_disable();
6526 list_splice_tail_init(&sd->poll_list, &list);
6527 list_splice_tail(&repoll, &list);
6528 list_splice(&list, &sd->poll_list);
6529 if (!list_empty(&sd->poll_list))
6530 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6532 net_rps_action_and_irq_enable(sd);
6535 struct netdev_adjacent {
6536 struct net_device *dev;
6538 /* upper master flag, there can only be one master device per list */
6541 /* lookup ignore flag */
6544 /* counter for the number of times this device was added to us */
6547 /* private field for the users */
6550 struct list_head list;
6551 struct rcu_head rcu;
6554 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6555 struct list_head *adj_list)
6557 struct netdev_adjacent *adj;
6559 list_for_each_entry(adj, adj_list, list) {
6560 if (adj->dev == adj_dev)
6566 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6567 struct netdev_nested_priv *priv)
6569 struct net_device *dev = (struct net_device *)priv->data;
6571 return upper_dev == dev;
6575 * netdev_has_upper_dev - Check if device is linked to an upper device
6577 * @upper_dev: upper device to check
6579 * Find out if a device is linked to specified upper device and return true
6580 * in case it is. Note that this checks only immediate upper device,
6581 * not through a complete stack of devices. The caller must hold the RTNL lock.
6583 bool netdev_has_upper_dev(struct net_device *dev,
6584 struct net_device *upper_dev)
6586 struct netdev_nested_priv priv = {
6587 .data = (void *)upper_dev,
6592 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6595 EXPORT_SYMBOL(netdev_has_upper_dev);
6598 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6600 * @upper_dev: upper device to check
6602 * Find out if a device is linked to specified upper device and return true
6603 * in case it is. Note that this checks the entire upper device chain.
6604 * The caller must hold rcu lock.
6607 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6608 struct net_device *upper_dev)
6610 struct netdev_nested_priv priv = {
6611 .data = (void *)upper_dev,
6614 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6617 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6620 * netdev_has_any_upper_dev - Check if device is linked to some device
6623 * Find out if a device is linked to an upper device and return true in case
6624 * it is. The caller must hold the RTNL lock.
6626 bool netdev_has_any_upper_dev(struct net_device *dev)
6630 return !list_empty(&dev->adj_list.upper);
6632 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6635 * netdev_master_upper_dev_get - Get master upper device
6638 * Find a master upper device and return pointer to it or NULL in case
6639 * it's not there. The caller must hold the RTNL lock.
6641 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6643 struct netdev_adjacent *upper;
6647 if (list_empty(&dev->adj_list.upper))
6650 upper = list_first_entry(&dev->adj_list.upper,
6651 struct netdev_adjacent, list);
6652 if (likely(upper->master))
6656 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6658 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6660 struct netdev_adjacent *upper;
6664 if (list_empty(&dev->adj_list.upper))
6667 upper = list_first_entry(&dev->adj_list.upper,
6668 struct netdev_adjacent, list);
6669 if (likely(upper->master) && !upper->ignore)
6675 * netdev_has_any_lower_dev - Check if device is linked to some device
6678 * Find out if a device is linked to a lower device and return true in case
6679 * it is. The caller must hold the RTNL lock.
6681 static bool netdev_has_any_lower_dev(struct net_device *dev)
6685 return !list_empty(&dev->adj_list.lower);
6688 void *netdev_adjacent_get_private(struct list_head *adj_list)
6690 struct netdev_adjacent *adj;
6692 adj = list_entry(adj_list, struct netdev_adjacent, list);
6694 return adj->private;
6696 EXPORT_SYMBOL(netdev_adjacent_get_private);
6699 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6701 * @iter: list_head ** of the current position
6703 * Gets the next device from the dev's upper list, starting from iter
6704 * position. The caller must hold RCU read lock.
6706 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6707 struct list_head **iter)
6709 struct netdev_adjacent *upper;
6711 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6713 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6715 if (&upper->list == &dev->adj_list.upper)
6718 *iter = &upper->list;
6722 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6724 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6725 struct list_head **iter,
6728 struct netdev_adjacent *upper;
6730 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6732 if (&upper->list == &dev->adj_list.upper)
6735 *iter = &upper->list;
6736 *ignore = upper->ignore;
6741 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6742 struct list_head **iter)
6744 struct netdev_adjacent *upper;
6746 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6748 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6750 if (&upper->list == &dev->adj_list.upper)
6753 *iter = &upper->list;
6758 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6759 int (*fn)(struct net_device *dev,
6760 struct netdev_nested_priv *priv),
6761 struct netdev_nested_priv *priv)
6763 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6764 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6769 iter = &dev->adj_list.upper;
6773 ret = fn(now, priv);
6780 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6787 niter = &udev->adj_list.upper;
6788 dev_stack[cur] = now;
6789 iter_stack[cur++] = iter;
6796 next = dev_stack[--cur];
6797 niter = iter_stack[cur];
6807 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6808 int (*fn)(struct net_device *dev,
6809 struct netdev_nested_priv *priv),
6810 struct netdev_nested_priv *priv)
6812 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6813 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6817 iter = &dev->adj_list.upper;
6821 ret = fn(now, priv);
6828 udev = netdev_next_upper_dev_rcu(now, &iter);
6833 niter = &udev->adj_list.upper;
6834 dev_stack[cur] = now;
6835 iter_stack[cur++] = iter;
6842 next = dev_stack[--cur];
6843 niter = iter_stack[cur];
6852 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6854 static bool __netdev_has_upper_dev(struct net_device *dev,
6855 struct net_device *upper_dev)
6857 struct netdev_nested_priv priv = {
6859 .data = (void *)upper_dev,
6864 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6869 * netdev_lower_get_next_private - Get the next ->private from the
6870 * lower neighbour list
6872 * @iter: list_head ** of the current position
6874 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6875 * list, starting from iter position. The caller must hold either hold the
6876 * RTNL lock or its own locking that guarantees that the neighbour lower
6877 * list will remain unchanged.
6879 void *netdev_lower_get_next_private(struct net_device *dev,
6880 struct list_head **iter)
6882 struct netdev_adjacent *lower;
6884 lower = list_entry(*iter, struct netdev_adjacent, list);
6886 if (&lower->list == &dev->adj_list.lower)
6889 *iter = lower->list.next;
6891 return lower->private;
6893 EXPORT_SYMBOL(netdev_lower_get_next_private);
6896 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6897 * lower neighbour list, RCU
6900 * @iter: list_head ** of the current position
6902 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6903 * list, starting from iter position. The caller must hold RCU read lock.
6905 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6906 struct list_head **iter)
6908 struct netdev_adjacent *lower;
6910 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
6912 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6914 if (&lower->list == &dev->adj_list.lower)
6917 *iter = &lower->list;
6919 return lower->private;
6921 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6924 * netdev_lower_get_next - Get the next device from the lower neighbour
6927 * @iter: list_head ** of the current position
6929 * Gets the next netdev_adjacent from the dev's lower neighbour
6930 * list, starting from iter position. The caller must hold RTNL lock or
6931 * its own locking that guarantees that the neighbour lower
6932 * list will remain unchanged.
6934 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6936 struct netdev_adjacent *lower;
6938 lower = list_entry(*iter, struct netdev_adjacent, list);
6940 if (&lower->list == &dev->adj_list.lower)
6943 *iter = lower->list.next;
6947 EXPORT_SYMBOL(netdev_lower_get_next);
6949 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6950 struct list_head **iter)
6952 struct netdev_adjacent *lower;
6954 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6956 if (&lower->list == &dev->adj_list.lower)
6959 *iter = &lower->list;
6964 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
6965 struct list_head **iter,
6968 struct netdev_adjacent *lower;
6970 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6972 if (&lower->list == &dev->adj_list.lower)
6975 *iter = &lower->list;
6976 *ignore = lower->ignore;
6981 int netdev_walk_all_lower_dev(struct net_device *dev,
6982 int (*fn)(struct net_device *dev,
6983 struct netdev_nested_priv *priv),
6984 struct netdev_nested_priv *priv)
6986 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6987 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6991 iter = &dev->adj_list.lower;
6995 ret = fn(now, priv);
7002 ldev = netdev_next_lower_dev(now, &iter);
7007 niter = &ldev->adj_list.lower;
7008 dev_stack[cur] = now;
7009 iter_stack[cur++] = iter;
7016 next = dev_stack[--cur];
7017 niter = iter_stack[cur];
7026 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7028 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7029 int (*fn)(struct net_device *dev,
7030 struct netdev_nested_priv *priv),
7031 struct netdev_nested_priv *priv)
7033 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7034 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7039 iter = &dev->adj_list.lower;
7043 ret = fn(now, priv);
7050 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7057 niter = &ldev->adj_list.lower;
7058 dev_stack[cur] = now;
7059 iter_stack[cur++] = iter;
7066 next = dev_stack[--cur];
7067 niter = iter_stack[cur];
7077 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7078 struct list_head **iter)
7080 struct netdev_adjacent *lower;
7082 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7083 if (&lower->list == &dev->adj_list.lower)
7086 *iter = &lower->list;
7090 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7092 static u8 __netdev_upper_depth(struct net_device *dev)
7094 struct net_device *udev;
7095 struct list_head *iter;
7099 for (iter = &dev->adj_list.upper,
7100 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7102 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7105 if (max_depth < udev->upper_level)
7106 max_depth = udev->upper_level;
7112 static u8 __netdev_lower_depth(struct net_device *dev)
7114 struct net_device *ldev;
7115 struct list_head *iter;
7119 for (iter = &dev->adj_list.lower,
7120 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7122 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7125 if (max_depth < ldev->lower_level)
7126 max_depth = ldev->lower_level;
7132 static int __netdev_update_upper_level(struct net_device *dev,
7133 struct netdev_nested_priv *__unused)
7135 dev->upper_level = __netdev_upper_depth(dev) + 1;
7139 static int __netdev_update_lower_level(struct net_device *dev,
7140 struct netdev_nested_priv *priv)
7142 dev->lower_level = __netdev_lower_depth(dev) + 1;
7144 #ifdef CONFIG_LOCKDEP
7148 if (priv->flags & NESTED_SYNC_IMM)
7149 dev->nested_level = dev->lower_level - 1;
7150 if (priv->flags & NESTED_SYNC_TODO)
7151 net_unlink_todo(dev);
7156 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7157 int (*fn)(struct net_device *dev,
7158 struct netdev_nested_priv *priv),
7159 struct netdev_nested_priv *priv)
7161 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7162 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7166 iter = &dev->adj_list.lower;
7170 ret = fn(now, priv);
7177 ldev = netdev_next_lower_dev_rcu(now, &iter);
7182 niter = &ldev->adj_list.lower;
7183 dev_stack[cur] = now;
7184 iter_stack[cur++] = iter;
7191 next = dev_stack[--cur];
7192 niter = iter_stack[cur];
7201 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7204 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7205 * lower neighbour list, RCU
7209 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7210 * list. The caller must hold RCU read lock.
7212 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7214 struct netdev_adjacent *lower;
7216 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7217 struct netdev_adjacent, list);
7219 return lower->private;
7222 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7225 * netdev_master_upper_dev_get_rcu - Get master upper device
7228 * Find a master upper device and return pointer to it or NULL in case
7229 * it's not there. The caller must hold the RCU read lock.
7231 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7233 struct netdev_adjacent *upper;
7235 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7236 struct netdev_adjacent, list);
7237 if (upper && likely(upper->master))
7241 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7243 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7244 struct net_device *adj_dev,
7245 struct list_head *dev_list)
7247 char linkname[IFNAMSIZ+7];
7249 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7250 "upper_%s" : "lower_%s", adj_dev->name);
7251 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7254 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7256 struct list_head *dev_list)
7258 char linkname[IFNAMSIZ+7];
7260 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7261 "upper_%s" : "lower_%s", name);
7262 sysfs_remove_link(&(dev->dev.kobj), linkname);
7265 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7266 struct net_device *adj_dev,
7267 struct list_head *dev_list)
7269 return (dev_list == &dev->adj_list.upper ||
7270 dev_list == &dev->adj_list.lower) &&
7271 net_eq(dev_net(dev), dev_net(adj_dev));
7274 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7275 struct net_device *adj_dev,
7276 struct list_head *dev_list,
7277 void *private, bool master)
7279 struct netdev_adjacent *adj;
7282 adj = __netdev_find_adj(adj_dev, dev_list);
7286 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7287 dev->name, adj_dev->name, adj->ref_nr);
7292 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7297 adj->master = master;
7299 adj->private = private;
7300 adj->ignore = false;
7303 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7304 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7306 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7307 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7312 /* Ensure that master link is always the first item in list. */
7314 ret = sysfs_create_link(&(dev->dev.kobj),
7315 &(adj_dev->dev.kobj), "master");
7317 goto remove_symlinks;
7319 list_add_rcu(&adj->list, dev_list);
7321 list_add_tail_rcu(&adj->list, dev_list);
7327 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7328 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7336 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7337 struct net_device *adj_dev,
7339 struct list_head *dev_list)
7341 struct netdev_adjacent *adj;
7343 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7344 dev->name, adj_dev->name, ref_nr);
7346 adj = __netdev_find_adj(adj_dev, dev_list);
7349 pr_err("Adjacency does not exist for device %s from %s\n",
7350 dev->name, adj_dev->name);
7355 if (adj->ref_nr > ref_nr) {
7356 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7357 dev->name, adj_dev->name, ref_nr,
7358 adj->ref_nr - ref_nr);
7359 adj->ref_nr -= ref_nr;
7364 sysfs_remove_link(&(dev->dev.kobj), "master");
7366 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7367 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7369 list_del_rcu(&adj->list);
7370 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7371 adj_dev->name, dev->name, adj_dev->name);
7373 kfree_rcu(adj, rcu);
7376 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7377 struct net_device *upper_dev,
7378 struct list_head *up_list,
7379 struct list_head *down_list,
7380 void *private, bool master)
7384 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7389 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7392 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7399 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7400 struct net_device *upper_dev,
7402 struct list_head *up_list,
7403 struct list_head *down_list)
7405 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7406 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7409 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7410 struct net_device *upper_dev,
7411 void *private, bool master)
7413 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7414 &dev->adj_list.upper,
7415 &upper_dev->adj_list.lower,
7419 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7420 struct net_device *upper_dev)
7422 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7423 &dev->adj_list.upper,
7424 &upper_dev->adj_list.lower);
7427 static int __netdev_upper_dev_link(struct net_device *dev,
7428 struct net_device *upper_dev, bool master,
7429 void *upper_priv, void *upper_info,
7430 struct netdev_nested_priv *priv,
7431 struct netlink_ext_ack *extack)
7433 struct netdev_notifier_changeupper_info changeupper_info = {
7438 .upper_dev = upper_dev,
7441 .upper_info = upper_info,
7443 struct net_device *master_dev;
7448 if (dev == upper_dev)
7451 /* To prevent loops, check if dev is not upper device to upper_dev. */
7452 if (__netdev_has_upper_dev(upper_dev, dev))
7455 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7459 if (__netdev_has_upper_dev(dev, upper_dev))
7462 master_dev = __netdev_master_upper_dev_get(dev);
7464 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7467 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7468 &changeupper_info.info);
7469 ret = notifier_to_errno(ret);
7473 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7478 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7479 &changeupper_info.info);
7480 ret = notifier_to_errno(ret);
7484 __netdev_update_upper_level(dev, NULL);
7485 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7487 __netdev_update_lower_level(upper_dev, priv);
7488 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7494 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7500 * netdev_upper_dev_link - Add a link to the upper device
7502 * @upper_dev: new upper device
7503 * @extack: netlink extended ack
7505 * Adds a link to device which is upper to this one. The caller must hold
7506 * the RTNL lock. On a failure a negative errno code is returned.
7507 * On success the reference counts are adjusted and the function
7510 int netdev_upper_dev_link(struct net_device *dev,
7511 struct net_device *upper_dev,
7512 struct netlink_ext_ack *extack)
7514 struct netdev_nested_priv priv = {
7515 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7519 return __netdev_upper_dev_link(dev, upper_dev, false,
7520 NULL, NULL, &priv, extack);
7522 EXPORT_SYMBOL(netdev_upper_dev_link);
7525 * netdev_master_upper_dev_link - Add a master link to the upper device
7527 * @upper_dev: new upper device
7528 * @upper_priv: upper device private
7529 * @upper_info: upper info to be passed down via notifier
7530 * @extack: netlink extended ack
7532 * Adds a link to device which is upper to this one. In this case, only
7533 * one master upper device can be linked, although other non-master devices
7534 * might be linked as well. The caller must hold the RTNL lock.
7535 * On a failure a negative errno code is returned. On success the reference
7536 * counts are adjusted and the function returns zero.
7538 int netdev_master_upper_dev_link(struct net_device *dev,
7539 struct net_device *upper_dev,
7540 void *upper_priv, void *upper_info,
7541 struct netlink_ext_ack *extack)
7543 struct netdev_nested_priv priv = {
7544 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7548 return __netdev_upper_dev_link(dev, upper_dev, true,
7549 upper_priv, upper_info, &priv, extack);
7551 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7553 static void __netdev_upper_dev_unlink(struct net_device *dev,
7554 struct net_device *upper_dev,
7555 struct netdev_nested_priv *priv)
7557 struct netdev_notifier_changeupper_info changeupper_info = {
7561 .upper_dev = upper_dev,
7567 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7569 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7570 &changeupper_info.info);
7572 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7574 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7575 &changeupper_info.info);
7577 __netdev_update_upper_level(dev, NULL);
7578 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7580 __netdev_update_lower_level(upper_dev, priv);
7581 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7586 * netdev_upper_dev_unlink - Removes a link to upper device
7588 * @upper_dev: new upper device
7590 * Removes a link to device which is upper to this one. The caller must hold
7593 void netdev_upper_dev_unlink(struct net_device *dev,
7594 struct net_device *upper_dev)
7596 struct netdev_nested_priv priv = {
7597 .flags = NESTED_SYNC_TODO,
7601 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7603 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7605 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7606 struct net_device *lower_dev,
7609 struct netdev_adjacent *adj;
7611 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7615 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7620 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7621 struct net_device *lower_dev)
7623 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7626 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7627 struct net_device *lower_dev)
7629 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7632 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7633 struct net_device *new_dev,
7634 struct net_device *dev,
7635 struct netlink_ext_ack *extack)
7637 struct netdev_nested_priv priv = {
7646 if (old_dev && new_dev != old_dev)
7647 netdev_adjacent_dev_disable(dev, old_dev);
7648 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7651 if (old_dev && new_dev != old_dev)
7652 netdev_adjacent_dev_enable(dev, old_dev);
7658 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7660 void netdev_adjacent_change_commit(struct net_device *old_dev,
7661 struct net_device *new_dev,
7662 struct net_device *dev)
7664 struct netdev_nested_priv priv = {
7665 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7669 if (!new_dev || !old_dev)
7672 if (new_dev == old_dev)
7675 netdev_adjacent_dev_enable(dev, old_dev);
7676 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7678 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7680 void netdev_adjacent_change_abort(struct net_device *old_dev,
7681 struct net_device *new_dev,
7682 struct net_device *dev)
7684 struct netdev_nested_priv priv = {
7692 if (old_dev && new_dev != old_dev)
7693 netdev_adjacent_dev_enable(dev, old_dev);
7695 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7697 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7700 * netdev_bonding_info_change - Dispatch event about slave change
7702 * @bonding_info: info to dispatch
7704 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7705 * The caller must hold the RTNL lock.
7707 void netdev_bonding_info_change(struct net_device *dev,
7708 struct netdev_bonding_info *bonding_info)
7710 struct netdev_notifier_bonding_info info = {
7714 memcpy(&info.bonding_info, bonding_info,
7715 sizeof(struct netdev_bonding_info));
7716 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7719 EXPORT_SYMBOL(netdev_bonding_info_change);
7722 * netdev_get_xmit_slave - Get the xmit slave of master device
7725 * @all_slaves: assume all the slaves are active
7727 * The reference counters are not incremented so the caller must be
7728 * careful with locks. The caller must hold RCU lock.
7729 * %NULL is returned if no slave is found.
7732 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
7733 struct sk_buff *skb,
7736 const struct net_device_ops *ops = dev->netdev_ops;
7738 if (!ops->ndo_get_xmit_slave)
7740 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
7742 EXPORT_SYMBOL(netdev_get_xmit_slave);
7744 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
7747 const struct net_device_ops *ops = dev->netdev_ops;
7749 if (!ops->ndo_sk_get_lower_dev)
7751 return ops->ndo_sk_get_lower_dev(dev, sk);
7755 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
7759 * %NULL is returned if no lower device is found.
7762 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
7765 struct net_device *lower;
7767 lower = netdev_sk_get_lower_dev(dev, sk);
7770 lower = netdev_sk_get_lower_dev(dev, sk);
7775 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
7777 static void netdev_adjacent_add_links(struct net_device *dev)
7779 struct netdev_adjacent *iter;
7781 struct net *net = dev_net(dev);
7783 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7784 if (!net_eq(net, dev_net(iter->dev)))
7786 netdev_adjacent_sysfs_add(iter->dev, dev,
7787 &iter->dev->adj_list.lower);
7788 netdev_adjacent_sysfs_add(dev, iter->dev,
7789 &dev->adj_list.upper);
7792 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7793 if (!net_eq(net, dev_net(iter->dev)))
7795 netdev_adjacent_sysfs_add(iter->dev, dev,
7796 &iter->dev->adj_list.upper);
7797 netdev_adjacent_sysfs_add(dev, iter->dev,
7798 &dev->adj_list.lower);
7802 static void netdev_adjacent_del_links(struct net_device *dev)
7804 struct netdev_adjacent *iter;
7806 struct net *net = dev_net(dev);
7808 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7809 if (!net_eq(net, dev_net(iter->dev)))
7811 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7812 &iter->dev->adj_list.lower);
7813 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7814 &dev->adj_list.upper);
7817 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7818 if (!net_eq(net, dev_net(iter->dev)))
7820 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7821 &iter->dev->adj_list.upper);
7822 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7823 &dev->adj_list.lower);
7827 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7829 struct netdev_adjacent *iter;
7831 struct net *net = dev_net(dev);
7833 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7834 if (!net_eq(net, dev_net(iter->dev)))
7836 netdev_adjacent_sysfs_del(iter->dev, oldname,
7837 &iter->dev->adj_list.lower);
7838 netdev_adjacent_sysfs_add(iter->dev, dev,
7839 &iter->dev->adj_list.lower);
7842 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7843 if (!net_eq(net, dev_net(iter->dev)))
7845 netdev_adjacent_sysfs_del(iter->dev, oldname,
7846 &iter->dev->adj_list.upper);
7847 netdev_adjacent_sysfs_add(iter->dev, dev,
7848 &iter->dev->adj_list.upper);
7852 void *netdev_lower_dev_get_private(struct net_device *dev,
7853 struct net_device *lower_dev)
7855 struct netdev_adjacent *lower;
7859 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7863 return lower->private;
7865 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7869 * netdev_lower_state_changed - Dispatch event about lower device state change
7870 * @lower_dev: device
7871 * @lower_state_info: state to dispatch
7873 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7874 * The caller must hold the RTNL lock.
7876 void netdev_lower_state_changed(struct net_device *lower_dev,
7877 void *lower_state_info)
7879 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7880 .info.dev = lower_dev,
7884 changelowerstate_info.lower_state_info = lower_state_info;
7885 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7886 &changelowerstate_info.info);
7888 EXPORT_SYMBOL(netdev_lower_state_changed);
7890 static void dev_change_rx_flags(struct net_device *dev, int flags)
7892 const struct net_device_ops *ops = dev->netdev_ops;
7894 if (ops->ndo_change_rx_flags)
7895 ops->ndo_change_rx_flags(dev, flags);
7898 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7900 unsigned int old_flags = dev->flags;
7906 dev->flags |= IFF_PROMISC;
7907 dev->promiscuity += inc;
7908 if (dev->promiscuity == 0) {
7911 * If inc causes overflow, untouch promisc and return error.
7914 dev->flags &= ~IFF_PROMISC;
7916 dev->promiscuity -= inc;
7917 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
7921 if (dev->flags != old_flags) {
7922 pr_info("device %s %s promiscuous mode\n",
7924 dev->flags & IFF_PROMISC ? "entered" : "left");
7925 if (audit_enabled) {
7926 current_uid_gid(&uid, &gid);
7927 audit_log(audit_context(), GFP_ATOMIC,
7928 AUDIT_ANOM_PROMISCUOUS,
7929 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7930 dev->name, (dev->flags & IFF_PROMISC),
7931 (old_flags & IFF_PROMISC),
7932 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7933 from_kuid(&init_user_ns, uid),
7934 from_kgid(&init_user_ns, gid),
7935 audit_get_sessionid(current));
7938 dev_change_rx_flags(dev, IFF_PROMISC);
7941 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7946 * dev_set_promiscuity - update promiscuity count on a device
7950 * Add or remove promiscuity from a device. While the count in the device
7951 * remains above zero the interface remains promiscuous. Once it hits zero
7952 * the device reverts back to normal filtering operation. A negative inc
7953 * value is used to drop promiscuity on the device.
7954 * Return 0 if successful or a negative errno code on error.
7956 int dev_set_promiscuity(struct net_device *dev, int inc)
7958 unsigned int old_flags = dev->flags;
7961 err = __dev_set_promiscuity(dev, inc, true);
7964 if (dev->flags != old_flags)
7965 dev_set_rx_mode(dev);
7968 EXPORT_SYMBOL(dev_set_promiscuity);
7970 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7972 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7976 dev->flags |= IFF_ALLMULTI;
7977 dev->allmulti += inc;
7978 if (dev->allmulti == 0) {
7981 * If inc causes overflow, untouch allmulti and return error.
7984 dev->flags &= ~IFF_ALLMULTI;
7986 dev->allmulti -= inc;
7987 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
7991 if (dev->flags ^ old_flags) {
7992 dev_change_rx_flags(dev, IFF_ALLMULTI);
7993 dev_set_rx_mode(dev);
7995 __dev_notify_flags(dev, old_flags,
7996 dev->gflags ^ old_gflags);
8002 * dev_set_allmulti - update allmulti count on a device
8006 * Add or remove reception of all multicast frames to a device. While the
8007 * count in the device remains above zero the interface remains listening
8008 * to all interfaces. Once it hits zero the device reverts back to normal
8009 * filtering operation. A negative @inc value is used to drop the counter
8010 * when releasing a resource needing all multicasts.
8011 * Return 0 if successful or a negative errno code on error.
8014 int dev_set_allmulti(struct net_device *dev, int inc)
8016 return __dev_set_allmulti(dev, inc, true);
8018 EXPORT_SYMBOL(dev_set_allmulti);
8021 * Upload unicast and multicast address lists to device and
8022 * configure RX filtering. When the device doesn't support unicast
8023 * filtering it is put in promiscuous mode while unicast addresses
8026 void __dev_set_rx_mode(struct net_device *dev)
8028 const struct net_device_ops *ops = dev->netdev_ops;
8030 /* dev_open will call this function so the list will stay sane. */
8031 if (!(dev->flags&IFF_UP))
8034 if (!netif_device_present(dev))
8037 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8038 /* Unicast addresses changes may only happen under the rtnl,
8039 * therefore calling __dev_set_promiscuity here is safe.
8041 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8042 __dev_set_promiscuity(dev, 1, false);
8043 dev->uc_promisc = true;
8044 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8045 __dev_set_promiscuity(dev, -1, false);
8046 dev->uc_promisc = false;
8050 if (ops->ndo_set_rx_mode)
8051 ops->ndo_set_rx_mode(dev);
8054 void dev_set_rx_mode(struct net_device *dev)
8056 netif_addr_lock_bh(dev);
8057 __dev_set_rx_mode(dev);
8058 netif_addr_unlock_bh(dev);
8062 * dev_get_flags - get flags reported to userspace
8065 * Get the combination of flag bits exported through APIs to userspace.
8067 unsigned int dev_get_flags(const struct net_device *dev)
8071 flags = (dev->flags & ~(IFF_PROMISC |
8076 (dev->gflags & (IFF_PROMISC |
8079 if (netif_running(dev)) {
8080 if (netif_oper_up(dev))
8081 flags |= IFF_RUNNING;
8082 if (netif_carrier_ok(dev))
8083 flags |= IFF_LOWER_UP;
8084 if (netif_dormant(dev))
8085 flags |= IFF_DORMANT;
8090 EXPORT_SYMBOL(dev_get_flags);
8092 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8093 struct netlink_ext_ack *extack)
8095 unsigned int old_flags = dev->flags;
8101 * Set the flags on our device.
8104 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8105 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8107 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8111 * Load in the correct multicast list now the flags have changed.
8114 if ((old_flags ^ flags) & IFF_MULTICAST)
8115 dev_change_rx_flags(dev, IFF_MULTICAST);
8117 dev_set_rx_mode(dev);
8120 * Have we downed the interface. We handle IFF_UP ourselves
8121 * according to user attempts to set it, rather than blindly
8126 if ((old_flags ^ flags) & IFF_UP) {
8127 if (old_flags & IFF_UP)
8130 ret = __dev_open(dev, extack);
8133 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8134 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8135 unsigned int old_flags = dev->flags;
8137 dev->gflags ^= IFF_PROMISC;
8139 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8140 if (dev->flags != old_flags)
8141 dev_set_rx_mode(dev);
8144 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8145 * is important. Some (broken) drivers set IFF_PROMISC, when
8146 * IFF_ALLMULTI is requested not asking us and not reporting.
8148 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8149 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8151 dev->gflags ^= IFF_ALLMULTI;
8152 __dev_set_allmulti(dev, inc, false);
8158 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8159 unsigned int gchanges)
8161 unsigned int changes = dev->flags ^ old_flags;
8164 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8166 if (changes & IFF_UP) {
8167 if (dev->flags & IFF_UP)
8168 call_netdevice_notifiers(NETDEV_UP, dev);
8170 call_netdevice_notifiers(NETDEV_DOWN, dev);
8173 if (dev->flags & IFF_UP &&
8174 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8175 struct netdev_notifier_change_info change_info = {
8179 .flags_changed = changes,
8182 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8187 * dev_change_flags - change device settings
8189 * @flags: device state flags
8190 * @extack: netlink extended ack
8192 * Change settings on device based state flags. The flags are
8193 * in the userspace exported format.
8195 int dev_change_flags(struct net_device *dev, unsigned int flags,
8196 struct netlink_ext_ack *extack)
8199 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8201 ret = __dev_change_flags(dev, flags, extack);
8205 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8206 __dev_notify_flags(dev, old_flags, changes);
8209 EXPORT_SYMBOL(dev_change_flags);
8211 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8213 const struct net_device_ops *ops = dev->netdev_ops;
8215 if (ops->ndo_change_mtu)
8216 return ops->ndo_change_mtu(dev, new_mtu);
8218 /* Pairs with all the lockless reads of dev->mtu in the stack */
8219 WRITE_ONCE(dev->mtu, new_mtu);
8222 EXPORT_SYMBOL(__dev_set_mtu);
8224 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8225 struct netlink_ext_ack *extack)
8227 /* MTU must be positive, and in range */
8228 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8229 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8233 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8234 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8241 * dev_set_mtu_ext - Change maximum transfer unit
8243 * @new_mtu: new transfer unit
8244 * @extack: netlink extended ack
8246 * Change the maximum transfer size of the network device.
8248 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8249 struct netlink_ext_ack *extack)
8253 if (new_mtu == dev->mtu)
8256 err = dev_validate_mtu(dev, new_mtu, extack);
8260 if (!netif_device_present(dev))
8263 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8264 err = notifier_to_errno(err);
8268 orig_mtu = dev->mtu;
8269 err = __dev_set_mtu(dev, new_mtu);
8272 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8274 err = notifier_to_errno(err);
8276 /* setting mtu back and notifying everyone again,
8277 * so that they have a chance to revert changes.
8279 __dev_set_mtu(dev, orig_mtu);
8280 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8287 int dev_set_mtu(struct net_device *dev, int new_mtu)
8289 struct netlink_ext_ack extack;
8292 memset(&extack, 0, sizeof(extack));
8293 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8294 if (err && extack._msg)
8295 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8298 EXPORT_SYMBOL(dev_set_mtu);
8301 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8303 * @new_len: new tx queue length
8305 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8307 unsigned int orig_len = dev->tx_queue_len;
8310 if (new_len != (unsigned int)new_len)
8313 if (new_len != orig_len) {
8314 dev->tx_queue_len = new_len;
8315 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8316 res = notifier_to_errno(res);
8319 res = dev_qdisc_change_tx_queue_len(dev);
8327 netdev_err(dev, "refused to change device tx_queue_len\n");
8328 dev->tx_queue_len = orig_len;
8333 * dev_set_group - Change group this device belongs to
8335 * @new_group: group this device should belong to
8337 void dev_set_group(struct net_device *dev, int new_group)
8339 dev->group = new_group;
8341 EXPORT_SYMBOL(dev_set_group);
8344 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8346 * @addr: new address
8347 * @extack: netlink extended ack
8349 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8350 struct netlink_ext_ack *extack)
8352 struct netdev_notifier_pre_changeaddr_info info = {
8354 .info.extack = extack,
8359 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8360 return notifier_to_errno(rc);
8362 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8365 * dev_set_mac_address - Change Media Access Control Address
8368 * @extack: netlink extended ack
8370 * Change the hardware (MAC) address of the device
8372 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8373 struct netlink_ext_ack *extack)
8375 const struct net_device_ops *ops = dev->netdev_ops;
8378 if (!ops->ndo_set_mac_address)
8380 if (sa->sa_family != dev->type)
8382 if (!netif_device_present(dev))
8384 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8387 err = ops->ndo_set_mac_address(dev, sa);
8390 dev->addr_assign_type = NET_ADDR_SET;
8391 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8392 add_device_randomness(dev->dev_addr, dev->addr_len);
8395 EXPORT_SYMBOL(dev_set_mac_address);
8397 static DECLARE_RWSEM(dev_addr_sem);
8399 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8400 struct netlink_ext_ack *extack)
8404 down_write(&dev_addr_sem);
8405 ret = dev_set_mac_address(dev, sa, extack);
8406 up_write(&dev_addr_sem);
8409 EXPORT_SYMBOL(dev_set_mac_address_user);
8411 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8413 size_t size = sizeof(sa->sa_data);
8414 struct net_device *dev;
8417 down_read(&dev_addr_sem);
8420 dev = dev_get_by_name_rcu(net, dev_name);
8426 memset(sa->sa_data, 0, size);
8428 memcpy(sa->sa_data, dev->dev_addr,
8429 min_t(size_t, size, dev->addr_len));
8430 sa->sa_family = dev->type;
8434 up_read(&dev_addr_sem);
8437 EXPORT_SYMBOL(dev_get_mac_address);
8440 * dev_change_carrier - Change device carrier
8442 * @new_carrier: new value
8444 * Change device carrier
8446 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8448 const struct net_device_ops *ops = dev->netdev_ops;
8450 if (!ops->ndo_change_carrier)
8452 if (!netif_device_present(dev))
8454 return ops->ndo_change_carrier(dev, new_carrier);
8456 EXPORT_SYMBOL(dev_change_carrier);
8459 * dev_get_phys_port_id - Get device physical port ID
8463 * Get device physical port ID
8465 int dev_get_phys_port_id(struct net_device *dev,
8466 struct netdev_phys_item_id *ppid)
8468 const struct net_device_ops *ops = dev->netdev_ops;
8470 if (!ops->ndo_get_phys_port_id)
8472 return ops->ndo_get_phys_port_id(dev, ppid);
8474 EXPORT_SYMBOL(dev_get_phys_port_id);
8477 * dev_get_phys_port_name - Get device physical port name
8480 * @len: limit of bytes to copy to name
8482 * Get device physical port name
8484 int dev_get_phys_port_name(struct net_device *dev,
8485 char *name, size_t len)
8487 const struct net_device_ops *ops = dev->netdev_ops;
8490 if (ops->ndo_get_phys_port_name) {
8491 err = ops->ndo_get_phys_port_name(dev, name, len);
8492 if (err != -EOPNOTSUPP)
8495 return devlink_compat_phys_port_name_get(dev, name, len);
8497 EXPORT_SYMBOL(dev_get_phys_port_name);
8500 * dev_get_port_parent_id - Get the device's port parent identifier
8501 * @dev: network device
8502 * @ppid: pointer to a storage for the port's parent identifier
8503 * @recurse: allow/disallow recursion to lower devices
8505 * Get the devices's port parent identifier
8507 int dev_get_port_parent_id(struct net_device *dev,
8508 struct netdev_phys_item_id *ppid,
8511 const struct net_device_ops *ops = dev->netdev_ops;
8512 struct netdev_phys_item_id first = { };
8513 struct net_device *lower_dev;
8514 struct list_head *iter;
8517 if (ops->ndo_get_port_parent_id) {
8518 err = ops->ndo_get_port_parent_id(dev, ppid);
8519 if (err != -EOPNOTSUPP)
8523 err = devlink_compat_switch_id_get(dev, ppid);
8524 if (!recurse || err != -EOPNOTSUPP)
8527 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8528 err = dev_get_port_parent_id(lower_dev, ppid, true);
8533 else if (memcmp(&first, ppid, sizeof(*ppid)))
8539 EXPORT_SYMBOL(dev_get_port_parent_id);
8542 * netdev_port_same_parent_id - Indicate if two network devices have
8543 * the same port parent identifier
8544 * @a: first network device
8545 * @b: second network device
8547 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8549 struct netdev_phys_item_id a_id = { };
8550 struct netdev_phys_item_id b_id = { };
8552 if (dev_get_port_parent_id(a, &a_id, true) ||
8553 dev_get_port_parent_id(b, &b_id, true))
8556 return netdev_phys_item_id_same(&a_id, &b_id);
8558 EXPORT_SYMBOL(netdev_port_same_parent_id);
8561 * dev_change_proto_down - update protocol port state information
8563 * @proto_down: new value
8565 * This info can be used by switch drivers to set the phys state of the
8568 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8570 const struct net_device_ops *ops = dev->netdev_ops;
8572 if (!ops->ndo_change_proto_down)
8574 if (!netif_device_present(dev))
8576 return ops->ndo_change_proto_down(dev, proto_down);
8578 EXPORT_SYMBOL(dev_change_proto_down);
8581 * dev_change_proto_down_generic - generic implementation for
8582 * ndo_change_proto_down that sets carrier according to
8586 * @proto_down: new value
8588 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8591 netif_carrier_off(dev);
8593 netif_carrier_on(dev);
8594 dev->proto_down = proto_down;
8597 EXPORT_SYMBOL(dev_change_proto_down_generic);
8600 * dev_change_proto_down_reason - proto down reason
8603 * @mask: proto down mask
8604 * @value: proto down value
8606 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8612 dev->proto_down_reason = value;
8614 for_each_set_bit(b, &mask, 32) {
8615 if (value & (1 << b))
8616 dev->proto_down_reason |= BIT(b);
8618 dev->proto_down_reason &= ~BIT(b);
8622 EXPORT_SYMBOL(dev_change_proto_down_reason);
8624 struct bpf_xdp_link {
8625 struct bpf_link link;
8626 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
8630 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
8632 if (flags & XDP_FLAGS_HW_MODE)
8634 if (flags & XDP_FLAGS_DRV_MODE)
8635 return XDP_MODE_DRV;
8636 if (flags & XDP_FLAGS_SKB_MODE)
8637 return XDP_MODE_SKB;
8638 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
8641 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
8645 return generic_xdp_install;
8648 return dev->netdev_ops->ndo_bpf;
8654 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
8655 enum bpf_xdp_mode mode)
8657 return dev->xdp_state[mode].link;
8660 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
8661 enum bpf_xdp_mode mode)
8663 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
8666 return link->link.prog;
8667 return dev->xdp_state[mode].prog;
8670 u8 dev_xdp_prog_count(struct net_device *dev)
8675 for (i = 0; i < __MAX_XDP_MODE; i++)
8676 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
8680 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
8682 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
8684 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
8686 return prog ? prog->aux->id : 0;
8689 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
8690 struct bpf_xdp_link *link)
8692 dev->xdp_state[mode].link = link;
8693 dev->xdp_state[mode].prog = NULL;
8696 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
8697 struct bpf_prog *prog)
8699 dev->xdp_state[mode].link = NULL;
8700 dev->xdp_state[mode].prog = prog;
8703 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
8704 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
8705 u32 flags, struct bpf_prog *prog)
8707 struct netdev_bpf xdp;
8710 memset(&xdp, 0, sizeof(xdp));
8711 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
8712 xdp.extack = extack;
8716 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
8717 * "moved" into driver), so they don't increment it on their own, but
8718 * they do decrement refcnt when program is detached or replaced.
8719 * Given net_device also owns link/prog, we need to bump refcnt here
8720 * to prevent drivers from underflowing it.
8724 err = bpf_op(dev, &xdp);
8731 if (mode != XDP_MODE_HW)
8732 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
8737 static void dev_xdp_uninstall(struct net_device *dev)
8739 struct bpf_xdp_link *link;
8740 struct bpf_prog *prog;
8741 enum bpf_xdp_mode mode;
8746 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
8747 prog = dev_xdp_prog(dev, mode);
8751 bpf_op = dev_xdp_bpf_op(dev, mode);
8755 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
8757 /* auto-detach link from net device */
8758 link = dev_xdp_link(dev, mode);
8764 dev_xdp_set_link(dev, mode, NULL);
8768 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
8769 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
8770 struct bpf_prog *old_prog, u32 flags)
8772 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
8773 struct bpf_prog *cur_prog;
8774 struct net_device *upper;
8775 struct list_head *iter;
8776 enum bpf_xdp_mode mode;
8782 /* either link or prog attachment, never both */
8783 if (link && (new_prog || old_prog))
8785 /* link supports only XDP mode flags */
8786 if (link && (flags & ~XDP_FLAGS_MODES)) {
8787 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
8790 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
8791 if (num_modes > 1) {
8792 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
8795 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
8796 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
8797 NL_SET_ERR_MSG(extack,
8798 "More than one program loaded, unset mode is ambiguous");
8801 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
8802 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
8803 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
8807 mode = dev_xdp_mode(dev, flags);
8808 /* can't replace attached link */
8809 if (dev_xdp_link(dev, mode)) {
8810 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
8814 /* don't allow if an upper device already has a program */
8815 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
8816 if (dev_xdp_prog_count(upper) > 0) {
8817 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
8822 cur_prog = dev_xdp_prog(dev, mode);
8823 /* can't replace attached prog with link */
8824 if (link && cur_prog) {
8825 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
8828 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
8829 NL_SET_ERR_MSG(extack, "Active program does not match expected");
8833 /* put effective new program into new_prog */
8835 new_prog = link->link.prog;
8838 bool offload = mode == XDP_MODE_HW;
8839 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
8840 ? XDP_MODE_DRV : XDP_MODE_SKB;
8842 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
8843 NL_SET_ERR_MSG(extack, "XDP program already attached");
8846 if (!offload && dev_xdp_prog(dev, other_mode)) {
8847 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
8850 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
8851 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
8854 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
8855 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
8858 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
8859 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
8864 /* don't call drivers if the effective program didn't change */
8865 if (new_prog != cur_prog) {
8866 bpf_op = dev_xdp_bpf_op(dev, mode);
8868 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
8872 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
8878 dev_xdp_set_link(dev, mode, link);
8880 dev_xdp_set_prog(dev, mode, new_prog);
8882 bpf_prog_put(cur_prog);
8887 static int dev_xdp_attach_link(struct net_device *dev,
8888 struct netlink_ext_ack *extack,
8889 struct bpf_xdp_link *link)
8891 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
8894 static int dev_xdp_detach_link(struct net_device *dev,
8895 struct netlink_ext_ack *extack,
8896 struct bpf_xdp_link *link)
8898 enum bpf_xdp_mode mode;
8903 mode = dev_xdp_mode(dev, link->flags);
8904 if (dev_xdp_link(dev, mode) != link)
8907 bpf_op = dev_xdp_bpf_op(dev, mode);
8908 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
8909 dev_xdp_set_link(dev, mode, NULL);
8913 static void bpf_xdp_link_release(struct bpf_link *link)
8915 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8919 /* if racing with net_device's tear down, xdp_link->dev might be
8920 * already NULL, in which case link was already auto-detached
8922 if (xdp_link->dev) {
8923 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
8924 xdp_link->dev = NULL;
8930 static int bpf_xdp_link_detach(struct bpf_link *link)
8932 bpf_xdp_link_release(link);
8936 static void bpf_xdp_link_dealloc(struct bpf_link *link)
8938 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8943 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
8944 struct seq_file *seq)
8946 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8951 ifindex = xdp_link->dev->ifindex;
8954 seq_printf(seq, "ifindex:\t%u\n", ifindex);
8957 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
8958 struct bpf_link_info *info)
8960 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8965 ifindex = xdp_link->dev->ifindex;
8968 info->xdp.ifindex = ifindex;
8972 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
8973 struct bpf_prog *old_prog)
8975 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8976 enum bpf_xdp_mode mode;
8982 /* link might have been auto-released already, so fail */
8983 if (!xdp_link->dev) {
8988 if (old_prog && link->prog != old_prog) {
8992 old_prog = link->prog;
8993 if (old_prog == new_prog) {
8994 /* no-op, don't disturb drivers */
8995 bpf_prog_put(new_prog);
8999 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9000 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9001 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9002 xdp_link->flags, new_prog);
9006 old_prog = xchg(&link->prog, new_prog);
9007 bpf_prog_put(old_prog);
9014 static const struct bpf_link_ops bpf_xdp_link_lops = {
9015 .release = bpf_xdp_link_release,
9016 .dealloc = bpf_xdp_link_dealloc,
9017 .detach = bpf_xdp_link_detach,
9018 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9019 .fill_link_info = bpf_xdp_link_fill_link_info,
9020 .update_prog = bpf_xdp_link_update,
9023 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9025 struct net *net = current->nsproxy->net_ns;
9026 struct bpf_link_primer link_primer;
9027 struct bpf_xdp_link *link;
9028 struct net_device *dev;
9032 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9038 link = kzalloc(sizeof(*link), GFP_USER);
9044 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9046 link->flags = attr->link_create.flags;
9048 err = bpf_link_prime(&link->link, &link_primer);
9054 err = dev_xdp_attach_link(dev, NULL, link);
9059 bpf_link_cleanup(&link_primer);
9063 fd = bpf_link_settle(&link_primer);
9064 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9077 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9079 * @extack: netlink extended ack
9080 * @fd: new program fd or negative value to clear
9081 * @expected_fd: old program fd that userspace expects to replace or clear
9082 * @flags: xdp-related flags
9084 * Set or clear a bpf program for a device
9086 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9087 int fd, int expected_fd, u32 flags)
9089 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9090 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9096 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9097 mode != XDP_MODE_SKB);
9098 if (IS_ERR(new_prog))
9099 return PTR_ERR(new_prog);
9102 if (expected_fd >= 0) {
9103 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9104 mode != XDP_MODE_SKB);
9105 if (IS_ERR(old_prog)) {
9106 err = PTR_ERR(old_prog);
9112 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9115 if (err && new_prog)
9116 bpf_prog_put(new_prog);
9118 bpf_prog_put(old_prog);
9123 * dev_new_index - allocate an ifindex
9124 * @net: the applicable net namespace
9126 * Returns a suitable unique value for a new device interface
9127 * number. The caller must hold the rtnl semaphore or the
9128 * dev_base_lock to be sure it remains unique.
9130 static int dev_new_index(struct net *net)
9132 int ifindex = net->ifindex;
9137 if (!__dev_get_by_index(net, ifindex))
9138 return net->ifindex = ifindex;
9142 /* Delayed registration/unregisteration */
9143 static LIST_HEAD(net_todo_list);
9144 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9146 static void net_set_todo(struct net_device *dev)
9148 list_add_tail(&dev->todo_list, &net_todo_list);
9149 dev_net(dev)->dev_unreg_count++;
9152 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9153 struct net_device *upper, netdev_features_t features)
9155 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9156 netdev_features_t feature;
9159 for_each_netdev_feature(upper_disables, feature_bit) {
9160 feature = __NETIF_F_BIT(feature_bit);
9161 if (!(upper->wanted_features & feature)
9162 && (features & feature)) {
9163 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9164 &feature, upper->name);
9165 features &= ~feature;
9172 static void netdev_sync_lower_features(struct net_device *upper,
9173 struct net_device *lower, netdev_features_t features)
9175 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9176 netdev_features_t feature;
9179 for_each_netdev_feature(upper_disables, feature_bit) {
9180 feature = __NETIF_F_BIT(feature_bit);
9181 if (!(features & feature) && (lower->features & feature)) {
9182 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9183 &feature, lower->name);
9184 lower->wanted_features &= ~feature;
9185 __netdev_update_features(lower);
9187 if (unlikely(lower->features & feature))
9188 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9189 &feature, lower->name);
9191 netdev_features_change(lower);
9196 static netdev_features_t netdev_fix_features(struct net_device *dev,
9197 netdev_features_t features)
9199 /* Fix illegal checksum combinations */
9200 if ((features & NETIF_F_HW_CSUM) &&
9201 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9202 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9203 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9206 /* TSO requires that SG is present as well. */
9207 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9208 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9209 features &= ~NETIF_F_ALL_TSO;
9212 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9213 !(features & NETIF_F_IP_CSUM)) {
9214 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9215 features &= ~NETIF_F_TSO;
9216 features &= ~NETIF_F_TSO_ECN;
9219 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9220 !(features & NETIF_F_IPV6_CSUM)) {
9221 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9222 features &= ~NETIF_F_TSO6;
9225 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9226 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9227 features &= ~NETIF_F_TSO_MANGLEID;
9229 /* TSO ECN requires that TSO is present as well. */
9230 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9231 features &= ~NETIF_F_TSO_ECN;
9233 /* Software GSO depends on SG. */
9234 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9235 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9236 features &= ~NETIF_F_GSO;
9239 /* GSO partial features require GSO partial be set */
9240 if ((features & dev->gso_partial_features) &&
9241 !(features & NETIF_F_GSO_PARTIAL)) {
9243 "Dropping partially supported GSO features since no GSO partial.\n");
9244 features &= ~dev->gso_partial_features;
9247 if (!(features & NETIF_F_RXCSUM)) {
9248 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9249 * successfully merged by hardware must also have the
9250 * checksum verified by hardware. If the user does not
9251 * want to enable RXCSUM, logically, we should disable GRO_HW.
9253 if (features & NETIF_F_GRO_HW) {
9254 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9255 features &= ~NETIF_F_GRO_HW;
9259 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9260 if (features & NETIF_F_RXFCS) {
9261 if (features & NETIF_F_LRO) {
9262 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9263 features &= ~NETIF_F_LRO;
9266 if (features & NETIF_F_GRO_HW) {
9267 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9268 features &= ~NETIF_F_GRO_HW;
9272 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9273 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9274 features &= ~NETIF_F_LRO;
9277 if (features & NETIF_F_HW_TLS_TX) {
9278 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9279 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9280 bool hw_csum = features & NETIF_F_HW_CSUM;
9282 if (!ip_csum && !hw_csum) {
9283 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9284 features &= ~NETIF_F_HW_TLS_TX;
9288 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9289 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9290 features &= ~NETIF_F_HW_TLS_RX;
9296 int __netdev_update_features(struct net_device *dev)
9298 struct net_device *upper, *lower;
9299 netdev_features_t features;
9300 struct list_head *iter;
9305 features = netdev_get_wanted_features(dev);
9307 if (dev->netdev_ops->ndo_fix_features)
9308 features = dev->netdev_ops->ndo_fix_features(dev, features);
9310 /* driver might be less strict about feature dependencies */
9311 features = netdev_fix_features(dev, features);
9313 /* some features can't be enabled if they're off on an upper device */
9314 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9315 features = netdev_sync_upper_features(dev, upper, features);
9317 if (dev->features == features)
9320 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9321 &dev->features, &features);
9323 if (dev->netdev_ops->ndo_set_features)
9324 err = dev->netdev_ops->ndo_set_features(dev, features);
9328 if (unlikely(err < 0)) {
9330 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9331 err, &features, &dev->features);
9332 /* return non-0 since some features might have changed and
9333 * it's better to fire a spurious notification than miss it
9339 /* some features must be disabled on lower devices when disabled
9340 * on an upper device (think: bonding master or bridge)
9342 netdev_for_each_lower_dev(dev, lower, iter)
9343 netdev_sync_lower_features(dev, lower, features);
9346 netdev_features_t diff = features ^ dev->features;
9348 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9349 /* udp_tunnel_{get,drop}_rx_info both need
9350 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9351 * device, or they won't do anything.
9352 * Thus we need to update dev->features
9353 * *before* calling udp_tunnel_get_rx_info,
9354 * but *after* calling udp_tunnel_drop_rx_info.
9356 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9357 dev->features = features;
9358 udp_tunnel_get_rx_info(dev);
9360 udp_tunnel_drop_rx_info(dev);
9364 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9365 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9366 dev->features = features;
9367 err |= vlan_get_rx_ctag_filter_info(dev);
9369 vlan_drop_rx_ctag_filter_info(dev);
9373 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9374 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9375 dev->features = features;
9376 err |= vlan_get_rx_stag_filter_info(dev);
9378 vlan_drop_rx_stag_filter_info(dev);
9382 dev->features = features;
9385 return err < 0 ? 0 : 1;
9389 * netdev_update_features - recalculate device features
9390 * @dev: the device to check
9392 * Recalculate dev->features set and send notifications if it
9393 * has changed. Should be called after driver or hardware dependent
9394 * conditions might have changed that influence the features.
9396 void netdev_update_features(struct net_device *dev)
9398 if (__netdev_update_features(dev))
9399 netdev_features_change(dev);
9401 EXPORT_SYMBOL(netdev_update_features);
9404 * netdev_change_features - recalculate device features
9405 * @dev: the device to check
9407 * Recalculate dev->features set and send notifications even
9408 * if they have not changed. Should be called instead of
9409 * netdev_update_features() if also dev->vlan_features might
9410 * have changed to allow the changes to be propagated to stacked
9413 void netdev_change_features(struct net_device *dev)
9415 __netdev_update_features(dev);
9416 netdev_features_change(dev);
9418 EXPORT_SYMBOL(netdev_change_features);
9421 * netif_stacked_transfer_operstate - transfer operstate
9422 * @rootdev: the root or lower level device to transfer state from
9423 * @dev: the device to transfer operstate to
9425 * Transfer operational state from root to device. This is normally
9426 * called when a stacking relationship exists between the root
9427 * device and the device(a leaf device).
9429 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9430 struct net_device *dev)
9432 if (rootdev->operstate == IF_OPER_DORMANT)
9433 netif_dormant_on(dev);
9435 netif_dormant_off(dev);
9437 if (rootdev->operstate == IF_OPER_TESTING)
9438 netif_testing_on(dev);
9440 netif_testing_off(dev);
9442 if (netif_carrier_ok(rootdev))
9443 netif_carrier_on(dev);
9445 netif_carrier_off(dev);
9447 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9449 static int netif_alloc_rx_queues(struct net_device *dev)
9451 unsigned int i, count = dev->num_rx_queues;
9452 struct netdev_rx_queue *rx;
9453 size_t sz = count * sizeof(*rx);
9458 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9464 for (i = 0; i < count; i++) {
9467 /* XDP RX-queue setup */
9468 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9475 /* Rollback successful reg's and free other resources */
9477 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9483 static void netif_free_rx_queues(struct net_device *dev)
9485 unsigned int i, count = dev->num_rx_queues;
9487 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9491 for (i = 0; i < count; i++)
9492 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9497 static void netdev_init_one_queue(struct net_device *dev,
9498 struct netdev_queue *queue, void *_unused)
9500 /* Initialize queue lock */
9501 spin_lock_init(&queue->_xmit_lock);
9502 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9503 queue->xmit_lock_owner = -1;
9504 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9507 dql_init(&queue->dql, HZ);
9511 static void netif_free_tx_queues(struct net_device *dev)
9516 static int netif_alloc_netdev_queues(struct net_device *dev)
9518 unsigned int count = dev->num_tx_queues;
9519 struct netdev_queue *tx;
9520 size_t sz = count * sizeof(*tx);
9522 if (count < 1 || count > 0xffff)
9525 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9531 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9532 spin_lock_init(&dev->tx_global_lock);
9537 void netif_tx_stop_all_queues(struct net_device *dev)
9541 for (i = 0; i < dev->num_tx_queues; i++) {
9542 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9544 netif_tx_stop_queue(txq);
9547 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9550 * register_netdevice - register a network device
9551 * @dev: device to register
9553 * Take a completed network device structure and add it to the kernel
9554 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9555 * chain. 0 is returned on success. A negative errno code is returned
9556 * on a failure to set up the device, or if the name is a duplicate.
9558 * Callers must hold the rtnl semaphore. You may want
9559 * register_netdev() instead of this.
9562 * The locking appears insufficient to guarantee two parallel registers
9563 * will not get the same name.
9566 int register_netdevice(struct net_device *dev)
9569 struct net *net = dev_net(dev);
9571 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9572 NETDEV_FEATURE_COUNT);
9573 BUG_ON(dev_boot_phase);
9578 /* When net_device's are persistent, this will be fatal. */
9579 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9582 ret = ethtool_check_ops(dev->ethtool_ops);
9586 spin_lock_init(&dev->addr_list_lock);
9587 netdev_set_addr_lockdep_class(dev);
9589 ret = dev_get_valid_name(net, dev, dev->name);
9594 dev->name_node = netdev_name_node_head_alloc(dev);
9595 if (!dev->name_node)
9598 /* Init, if this function is available */
9599 if (dev->netdev_ops->ndo_init) {
9600 ret = dev->netdev_ops->ndo_init(dev);
9608 if (((dev->hw_features | dev->features) &
9609 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9610 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9611 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9612 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9619 dev->ifindex = dev_new_index(net);
9620 else if (__dev_get_by_index(net, dev->ifindex))
9623 /* Transfer changeable features to wanted_features and enable
9624 * software offloads (GSO and GRO).
9626 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9627 dev->features |= NETIF_F_SOFT_FEATURES;
9629 if (dev->udp_tunnel_nic_info) {
9630 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9631 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9634 dev->wanted_features = dev->features & dev->hw_features;
9636 if (!(dev->flags & IFF_LOOPBACK))
9637 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9639 /* If IPv4 TCP segmentation offload is supported we should also
9640 * allow the device to enable segmenting the frame with the option
9641 * of ignoring a static IP ID value. This doesn't enable the
9642 * feature itself but allows the user to enable it later.
9644 if (dev->hw_features & NETIF_F_TSO)
9645 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9646 if (dev->vlan_features & NETIF_F_TSO)
9647 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9648 if (dev->mpls_features & NETIF_F_TSO)
9649 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9650 if (dev->hw_enc_features & NETIF_F_TSO)
9651 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9653 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9655 dev->vlan_features |= NETIF_F_HIGHDMA;
9657 /* Make NETIF_F_SG inheritable to tunnel devices.
9659 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9661 /* Make NETIF_F_SG inheritable to MPLS.
9663 dev->mpls_features |= NETIF_F_SG;
9665 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9666 ret = notifier_to_errno(ret);
9670 ret = netdev_register_kobject(dev);
9672 dev->reg_state = NETREG_UNREGISTERED;
9675 dev->reg_state = NETREG_REGISTERED;
9677 __netdev_update_features(dev);
9680 * Default initial state at registry is that the
9681 * device is present.
9684 set_bit(__LINK_STATE_PRESENT, &dev->state);
9686 linkwatch_init_dev(dev);
9688 dev_init_scheduler(dev);
9690 list_netdevice(dev);
9691 add_device_randomness(dev->dev_addr, dev->addr_len);
9693 /* If the device has permanent device address, driver should
9694 * set dev_addr and also addr_assign_type should be set to
9695 * NET_ADDR_PERM (default value).
9697 if (dev->addr_assign_type == NET_ADDR_PERM)
9698 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9700 /* Notify protocols, that a new device appeared. */
9701 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9702 ret = notifier_to_errno(ret);
9704 /* Expect explicit free_netdev() on failure */
9705 dev->needs_free_netdev = false;
9706 unregister_netdevice_queue(dev, NULL);
9710 * Prevent userspace races by waiting until the network
9711 * device is fully setup before sending notifications.
9713 if (!dev->rtnl_link_ops ||
9714 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9715 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9721 if (dev->netdev_ops->ndo_uninit)
9722 dev->netdev_ops->ndo_uninit(dev);
9723 if (dev->priv_destructor)
9724 dev->priv_destructor(dev);
9726 netdev_name_node_free(dev->name_node);
9729 EXPORT_SYMBOL(register_netdevice);
9732 * init_dummy_netdev - init a dummy network device for NAPI
9733 * @dev: device to init
9735 * This takes a network device structure and initialize the minimum
9736 * amount of fields so it can be used to schedule NAPI polls without
9737 * registering a full blown interface. This is to be used by drivers
9738 * that need to tie several hardware interfaces to a single NAPI
9739 * poll scheduler due to HW limitations.
9741 int init_dummy_netdev(struct net_device *dev)
9743 /* Clear everything. Note we don't initialize spinlocks
9744 * are they aren't supposed to be taken by any of the
9745 * NAPI code and this dummy netdev is supposed to be
9746 * only ever used for NAPI polls
9748 memset(dev, 0, sizeof(struct net_device));
9750 /* make sure we BUG if trying to hit standard
9751 * register/unregister code path
9753 dev->reg_state = NETREG_DUMMY;
9755 /* NAPI wants this */
9756 INIT_LIST_HEAD(&dev->napi_list);
9758 /* a dummy interface is started by default */
9759 set_bit(__LINK_STATE_PRESENT, &dev->state);
9760 set_bit(__LINK_STATE_START, &dev->state);
9762 /* napi_busy_loop stats accounting wants this */
9763 dev_net_set(dev, &init_net);
9765 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9766 * because users of this 'device' dont need to change
9772 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9776 * register_netdev - register a network device
9777 * @dev: device to register
9779 * Take a completed network device structure and add it to the kernel
9780 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9781 * chain. 0 is returned on success. A negative errno code is returned
9782 * on a failure to set up the device, or if the name is a duplicate.
9784 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9785 * and expands the device name if you passed a format string to
9788 int register_netdev(struct net_device *dev)
9792 if (rtnl_lock_killable())
9794 err = register_netdevice(dev);
9798 EXPORT_SYMBOL(register_netdev);
9800 int netdev_refcnt_read(const struct net_device *dev)
9802 #ifdef CONFIG_PCPU_DEV_REFCNT
9805 for_each_possible_cpu(i)
9806 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9809 return refcount_read(&dev->dev_refcnt);
9812 EXPORT_SYMBOL(netdev_refcnt_read);
9814 int netdev_unregister_timeout_secs __read_mostly = 10;
9816 #define WAIT_REFS_MIN_MSECS 1
9817 #define WAIT_REFS_MAX_MSECS 250
9819 * netdev_wait_allrefs - wait until all references are gone.
9820 * @dev: target net_device
9822 * This is called when unregistering network devices.
9824 * Any protocol or device that holds a reference should register
9825 * for netdevice notification, and cleanup and put back the
9826 * reference if they receive an UNREGISTER event.
9827 * We can get stuck here if buggy protocols don't correctly
9830 static void netdev_wait_allrefs(struct net_device *dev)
9832 unsigned long rebroadcast_time, warning_time;
9833 int wait = 0, refcnt;
9835 linkwatch_forget_dev(dev);
9837 rebroadcast_time = warning_time = jiffies;
9838 refcnt = netdev_refcnt_read(dev);
9840 while (refcnt != 1) {
9841 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9844 /* Rebroadcast unregister notification */
9845 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9851 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9853 /* We must not have linkwatch events
9854 * pending on unregister. If this
9855 * happens, we simply run the queue
9856 * unscheduled, resulting in a noop
9859 linkwatch_run_queue();
9864 rebroadcast_time = jiffies;
9869 wait = WAIT_REFS_MIN_MSECS;
9872 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
9875 refcnt = netdev_refcnt_read(dev);
9878 time_after(jiffies, warning_time +
9879 netdev_unregister_timeout_secs * HZ)) {
9880 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9882 warning_time = jiffies;
9891 * register_netdevice(x1);
9892 * register_netdevice(x2);
9894 * unregister_netdevice(y1);
9895 * unregister_netdevice(y2);
9901 * We are invoked by rtnl_unlock().
9902 * This allows us to deal with problems:
9903 * 1) We can delete sysfs objects which invoke hotplug
9904 * without deadlocking with linkwatch via keventd.
9905 * 2) Since we run with the RTNL semaphore not held, we can sleep
9906 * safely in order to wait for the netdev refcnt to drop to zero.
9908 * We must not return until all unregister events added during
9909 * the interval the lock was held have been completed.
9911 void netdev_run_todo(void)
9913 struct list_head list;
9914 #ifdef CONFIG_LOCKDEP
9915 struct list_head unlink_list;
9917 list_replace_init(&net_unlink_list, &unlink_list);
9919 while (!list_empty(&unlink_list)) {
9920 struct net_device *dev = list_first_entry(&unlink_list,
9923 list_del_init(&dev->unlink_list);
9924 dev->nested_level = dev->lower_level - 1;
9928 /* Snapshot list, allow later requests */
9929 list_replace_init(&net_todo_list, &list);
9934 /* Wait for rcu callbacks to finish before next phase */
9935 if (!list_empty(&list))
9938 while (!list_empty(&list)) {
9939 struct net_device *dev
9940 = list_first_entry(&list, struct net_device, todo_list);
9941 list_del(&dev->todo_list);
9943 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9944 pr_err("network todo '%s' but state %d\n",
9945 dev->name, dev->reg_state);
9950 dev->reg_state = NETREG_UNREGISTERED;
9952 netdev_wait_allrefs(dev);
9955 BUG_ON(netdev_refcnt_read(dev) != 1);
9956 BUG_ON(!list_empty(&dev->ptype_all));
9957 BUG_ON(!list_empty(&dev->ptype_specific));
9958 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9959 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9960 #if IS_ENABLED(CONFIG_DECNET)
9961 WARN_ON(dev->dn_ptr);
9963 if (dev->priv_destructor)
9964 dev->priv_destructor(dev);
9965 if (dev->needs_free_netdev)
9968 /* Report a network device has been unregistered */
9970 dev_net(dev)->dev_unreg_count--;
9972 wake_up(&netdev_unregistering_wq);
9974 /* Free network device */
9975 kobject_put(&dev->dev.kobj);
9979 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9980 * all the same fields in the same order as net_device_stats, with only
9981 * the type differing, but rtnl_link_stats64 may have additional fields
9982 * at the end for newer counters.
9984 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9985 const struct net_device_stats *netdev_stats)
9987 #if BITS_PER_LONG == 64
9988 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9989 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9990 /* zero out counters that only exist in rtnl_link_stats64 */
9991 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9992 sizeof(*stats64) - sizeof(*netdev_stats));
9994 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9995 const unsigned long *src = (const unsigned long *)netdev_stats;
9996 u64 *dst = (u64 *)stats64;
9998 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9999 for (i = 0; i < n; i++)
10001 /* zero out counters that only exist in rtnl_link_stats64 */
10002 memset((char *)stats64 + n * sizeof(u64), 0,
10003 sizeof(*stats64) - n * sizeof(u64));
10006 EXPORT_SYMBOL(netdev_stats_to_stats64);
10009 * dev_get_stats - get network device statistics
10010 * @dev: device to get statistics from
10011 * @storage: place to store stats
10013 * Get network statistics from device. Return @storage.
10014 * The device driver may provide its own method by setting
10015 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10016 * otherwise the internal statistics structure is used.
10018 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10019 struct rtnl_link_stats64 *storage)
10021 const struct net_device_ops *ops = dev->netdev_ops;
10023 if (ops->ndo_get_stats64) {
10024 memset(storage, 0, sizeof(*storage));
10025 ops->ndo_get_stats64(dev, storage);
10026 } else if (ops->ndo_get_stats) {
10027 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10029 netdev_stats_to_stats64(storage, &dev->stats);
10031 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10032 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10033 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10036 EXPORT_SYMBOL(dev_get_stats);
10039 * dev_fetch_sw_netstats - get per-cpu network device statistics
10040 * @s: place to store stats
10041 * @netstats: per-cpu network stats to read from
10043 * Read per-cpu network statistics and populate the related fields in @s.
10045 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10046 const struct pcpu_sw_netstats __percpu *netstats)
10050 for_each_possible_cpu(cpu) {
10051 const struct pcpu_sw_netstats *stats;
10052 struct pcpu_sw_netstats tmp;
10053 unsigned int start;
10055 stats = per_cpu_ptr(netstats, cpu);
10057 start = u64_stats_fetch_begin_irq(&stats->syncp);
10058 tmp.rx_packets = stats->rx_packets;
10059 tmp.rx_bytes = stats->rx_bytes;
10060 tmp.tx_packets = stats->tx_packets;
10061 tmp.tx_bytes = stats->tx_bytes;
10062 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10064 s->rx_packets += tmp.rx_packets;
10065 s->rx_bytes += tmp.rx_bytes;
10066 s->tx_packets += tmp.tx_packets;
10067 s->tx_bytes += tmp.tx_bytes;
10070 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10073 * dev_get_tstats64 - ndo_get_stats64 implementation
10074 * @dev: device to get statistics from
10075 * @s: place to store stats
10077 * Populate @s from dev->stats and dev->tstats. Can be used as
10078 * ndo_get_stats64() callback.
10080 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10082 netdev_stats_to_stats64(s, &dev->stats);
10083 dev_fetch_sw_netstats(s, dev->tstats);
10085 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10087 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10089 struct netdev_queue *queue = dev_ingress_queue(dev);
10091 #ifdef CONFIG_NET_CLS_ACT
10094 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10097 netdev_init_one_queue(dev, queue, NULL);
10098 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10099 queue->qdisc_sleeping = &noop_qdisc;
10100 rcu_assign_pointer(dev->ingress_queue, queue);
10105 static const struct ethtool_ops default_ethtool_ops;
10107 void netdev_set_default_ethtool_ops(struct net_device *dev,
10108 const struct ethtool_ops *ops)
10110 if (dev->ethtool_ops == &default_ethtool_ops)
10111 dev->ethtool_ops = ops;
10113 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10115 void netdev_freemem(struct net_device *dev)
10117 char *addr = (char *)dev - dev->padded;
10123 * alloc_netdev_mqs - allocate network device
10124 * @sizeof_priv: size of private data to allocate space for
10125 * @name: device name format string
10126 * @name_assign_type: origin of device name
10127 * @setup: callback to initialize device
10128 * @txqs: the number of TX subqueues to allocate
10129 * @rxqs: the number of RX subqueues to allocate
10131 * Allocates a struct net_device with private data area for driver use
10132 * and performs basic initialization. Also allocates subqueue structs
10133 * for each queue on the device.
10135 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10136 unsigned char name_assign_type,
10137 void (*setup)(struct net_device *),
10138 unsigned int txqs, unsigned int rxqs)
10140 struct net_device *dev;
10141 unsigned int alloc_size;
10142 struct net_device *p;
10144 BUG_ON(strlen(name) >= sizeof(dev->name));
10147 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10152 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10156 alloc_size = sizeof(struct net_device);
10158 /* ensure 32-byte alignment of private area */
10159 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10160 alloc_size += sizeof_priv;
10162 /* ensure 32-byte alignment of whole construct */
10163 alloc_size += NETDEV_ALIGN - 1;
10165 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10169 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10170 dev->padded = (char *)dev - (char *)p;
10172 #ifdef CONFIG_PCPU_DEV_REFCNT
10173 dev->pcpu_refcnt = alloc_percpu(int);
10174 if (!dev->pcpu_refcnt)
10178 refcount_set(&dev->dev_refcnt, 1);
10181 if (dev_addr_init(dev))
10187 dev_net_set(dev, &init_net);
10189 dev->gso_max_size = GSO_MAX_SIZE;
10190 dev->gso_max_segs = GSO_MAX_SEGS;
10191 dev->upper_level = 1;
10192 dev->lower_level = 1;
10193 #ifdef CONFIG_LOCKDEP
10194 dev->nested_level = 0;
10195 INIT_LIST_HEAD(&dev->unlink_list);
10198 INIT_LIST_HEAD(&dev->napi_list);
10199 INIT_LIST_HEAD(&dev->unreg_list);
10200 INIT_LIST_HEAD(&dev->close_list);
10201 INIT_LIST_HEAD(&dev->link_watch_list);
10202 INIT_LIST_HEAD(&dev->adj_list.upper);
10203 INIT_LIST_HEAD(&dev->adj_list.lower);
10204 INIT_LIST_HEAD(&dev->ptype_all);
10205 INIT_LIST_HEAD(&dev->ptype_specific);
10206 INIT_LIST_HEAD(&dev->net_notifier_list);
10207 #ifdef CONFIG_NET_SCHED
10208 hash_init(dev->qdisc_hash);
10210 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10213 if (!dev->tx_queue_len) {
10214 dev->priv_flags |= IFF_NO_QUEUE;
10215 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10218 dev->num_tx_queues = txqs;
10219 dev->real_num_tx_queues = txqs;
10220 if (netif_alloc_netdev_queues(dev))
10223 dev->num_rx_queues = rxqs;
10224 dev->real_num_rx_queues = rxqs;
10225 if (netif_alloc_rx_queues(dev))
10228 strcpy(dev->name, name);
10229 dev->name_assign_type = name_assign_type;
10230 dev->group = INIT_NETDEV_GROUP;
10231 if (!dev->ethtool_ops)
10232 dev->ethtool_ops = &default_ethtool_ops;
10234 nf_hook_netdev_init(dev);
10243 #ifdef CONFIG_PCPU_DEV_REFCNT
10244 free_percpu(dev->pcpu_refcnt);
10247 netdev_freemem(dev);
10250 EXPORT_SYMBOL(alloc_netdev_mqs);
10253 * free_netdev - free network device
10256 * This function does the last stage of destroying an allocated device
10257 * interface. The reference to the device object is released. If this
10258 * is the last reference then it will be freed.Must be called in process
10261 void free_netdev(struct net_device *dev)
10263 struct napi_struct *p, *n;
10267 /* When called immediately after register_netdevice() failed the unwind
10268 * handling may still be dismantling the device. Handle that case by
10269 * deferring the free.
10271 if (dev->reg_state == NETREG_UNREGISTERING) {
10273 dev->needs_free_netdev = true;
10277 netif_free_tx_queues(dev);
10278 netif_free_rx_queues(dev);
10280 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10282 /* Flush device addresses */
10283 dev_addr_flush(dev);
10285 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10288 #ifdef CONFIG_PCPU_DEV_REFCNT
10289 free_percpu(dev->pcpu_refcnt);
10290 dev->pcpu_refcnt = NULL;
10292 free_percpu(dev->xdp_bulkq);
10293 dev->xdp_bulkq = NULL;
10295 /* Compatibility with error handling in drivers */
10296 if (dev->reg_state == NETREG_UNINITIALIZED) {
10297 netdev_freemem(dev);
10301 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10302 dev->reg_state = NETREG_RELEASED;
10304 /* will free via device release */
10305 put_device(&dev->dev);
10307 EXPORT_SYMBOL(free_netdev);
10310 * synchronize_net - Synchronize with packet receive processing
10312 * Wait for packets currently being received to be done.
10313 * Does not block later packets from starting.
10315 void synchronize_net(void)
10318 if (rtnl_is_locked())
10319 synchronize_rcu_expedited();
10323 EXPORT_SYMBOL(synchronize_net);
10326 * unregister_netdevice_queue - remove device from the kernel
10330 * This function shuts down a device interface and removes it
10331 * from the kernel tables.
10332 * If head not NULL, device is queued to be unregistered later.
10334 * Callers must hold the rtnl semaphore. You may want
10335 * unregister_netdev() instead of this.
10338 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10343 list_move_tail(&dev->unreg_list, head);
10347 list_add(&dev->unreg_list, &single);
10348 unregister_netdevice_many(&single);
10351 EXPORT_SYMBOL(unregister_netdevice_queue);
10354 * unregister_netdevice_many - unregister many devices
10355 * @head: list of devices
10357 * Note: As most callers use a stack allocated list_head,
10358 * we force a list_del() to make sure stack wont be corrupted later.
10360 void unregister_netdevice_many(struct list_head *head)
10362 struct net_device *dev, *tmp;
10363 LIST_HEAD(close_head);
10365 BUG_ON(dev_boot_phase);
10368 if (list_empty(head))
10371 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10372 /* Some devices call without registering
10373 * for initialization unwind. Remove those
10374 * devices and proceed with the remaining.
10376 if (dev->reg_state == NETREG_UNINITIALIZED) {
10377 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10381 list_del(&dev->unreg_list);
10384 dev->dismantle = true;
10385 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10388 /* If device is running, close it first. */
10389 list_for_each_entry(dev, head, unreg_list)
10390 list_add_tail(&dev->close_list, &close_head);
10391 dev_close_many(&close_head, true);
10393 list_for_each_entry(dev, head, unreg_list) {
10394 /* And unlink it from device chain. */
10395 unlist_netdevice(dev);
10397 dev->reg_state = NETREG_UNREGISTERING;
10399 flush_all_backlogs();
10403 list_for_each_entry(dev, head, unreg_list) {
10404 struct sk_buff *skb = NULL;
10406 /* Shutdown queueing discipline. */
10409 dev_xdp_uninstall(dev);
10411 /* Notify protocols, that we are about to destroy
10412 * this device. They should clean all the things.
10414 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10416 if (!dev->rtnl_link_ops ||
10417 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10418 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10419 GFP_KERNEL, NULL, 0);
10422 * Flush the unicast and multicast chains
10427 netdev_name_node_alt_flush(dev);
10428 netdev_name_node_free(dev->name_node);
10430 if (dev->netdev_ops->ndo_uninit)
10431 dev->netdev_ops->ndo_uninit(dev);
10434 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10436 /* Notifier chain MUST detach us all upper devices. */
10437 WARN_ON(netdev_has_any_upper_dev(dev));
10438 WARN_ON(netdev_has_any_lower_dev(dev));
10440 /* Remove entries from kobject tree */
10441 netdev_unregister_kobject(dev);
10443 /* Remove XPS queueing entries */
10444 netif_reset_xps_queues_gt(dev, 0);
10450 list_for_each_entry(dev, head, unreg_list) {
10457 EXPORT_SYMBOL(unregister_netdevice_many);
10460 * unregister_netdev - remove device from the kernel
10463 * This function shuts down a device interface and removes it
10464 * from the kernel tables.
10466 * This is just a wrapper for unregister_netdevice that takes
10467 * the rtnl semaphore. In general you want to use this and not
10468 * unregister_netdevice.
10470 void unregister_netdev(struct net_device *dev)
10473 unregister_netdevice(dev);
10476 EXPORT_SYMBOL(unregister_netdev);
10479 * __dev_change_net_namespace - move device to different nethost namespace
10481 * @net: network namespace
10482 * @pat: If not NULL name pattern to try if the current device name
10483 * is already taken in the destination network namespace.
10484 * @new_ifindex: If not zero, specifies device index in the target
10487 * This function shuts down a device interface and moves it
10488 * to a new network namespace. On success 0 is returned, on
10489 * a failure a netagive errno code is returned.
10491 * Callers must hold the rtnl semaphore.
10494 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10495 const char *pat, int new_ifindex)
10497 struct net *net_old = dev_net(dev);
10502 /* Don't allow namespace local devices to be moved. */
10504 if (dev->features & NETIF_F_NETNS_LOCAL)
10507 /* Ensure the device has been registrered */
10508 if (dev->reg_state != NETREG_REGISTERED)
10511 /* Get out if there is nothing todo */
10513 if (net_eq(net_old, net))
10516 /* Pick the destination device name, and ensure
10517 * we can use it in the destination network namespace.
10520 if (netdev_name_in_use(net, dev->name)) {
10521 /* We get here if we can't use the current device name */
10524 err = dev_get_valid_name(net, dev, pat);
10529 /* Check that new_ifindex isn't used yet. */
10531 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10535 * And now a mini version of register_netdevice unregister_netdevice.
10538 /* If device is running close it first. */
10541 /* And unlink it from device chain */
10542 unlist_netdevice(dev);
10546 /* Shutdown queueing discipline. */
10549 /* Notify protocols, that we are about to destroy
10550 * this device. They should clean all the things.
10552 * Note that dev->reg_state stays at NETREG_REGISTERED.
10553 * This is wanted because this way 8021q and macvlan know
10554 * the device is just moving and can keep their slaves up.
10556 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10559 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10560 /* If there is an ifindex conflict assign a new one */
10561 if (!new_ifindex) {
10562 if (__dev_get_by_index(net, dev->ifindex))
10563 new_ifindex = dev_new_index(net);
10565 new_ifindex = dev->ifindex;
10568 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10572 * Flush the unicast and multicast chains
10577 /* Send a netdev-removed uevent to the old namespace */
10578 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10579 netdev_adjacent_del_links(dev);
10581 /* Move per-net netdevice notifiers that are following the netdevice */
10582 move_netdevice_notifiers_dev_net(dev, net);
10584 /* Actually switch the network namespace */
10585 dev_net_set(dev, net);
10586 dev->ifindex = new_ifindex;
10588 /* Send a netdev-add uevent to the new namespace */
10589 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10590 netdev_adjacent_add_links(dev);
10592 /* Fixup kobjects */
10593 err = device_rename(&dev->dev, dev->name);
10596 /* Adapt owner in case owning user namespace of target network
10597 * namespace is different from the original one.
10599 err = netdev_change_owner(dev, net_old, net);
10602 /* Add the device back in the hashes */
10603 list_netdevice(dev);
10605 /* Notify protocols, that a new device appeared. */
10606 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10609 * Prevent userspace races by waiting until the network
10610 * device is fully setup before sending notifications.
10612 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10619 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
10621 static int dev_cpu_dead(unsigned int oldcpu)
10623 struct sk_buff **list_skb;
10624 struct sk_buff *skb;
10626 struct softnet_data *sd, *oldsd, *remsd = NULL;
10628 local_irq_disable();
10629 cpu = smp_processor_id();
10630 sd = &per_cpu(softnet_data, cpu);
10631 oldsd = &per_cpu(softnet_data, oldcpu);
10633 /* Find end of our completion_queue. */
10634 list_skb = &sd->completion_queue;
10636 list_skb = &(*list_skb)->next;
10637 /* Append completion queue from offline CPU. */
10638 *list_skb = oldsd->completion_queue;
10639 oldsd->completion_queue = NULL;
10641 /* Append output queue from offline CPU. */
10642 if (oldsd->output_queue) {
10643 *sd->output_queue_tailp = oldsd->output_queue;
10644 sd->output_queue_tailp = oldsd->output_queue_tailp;
10645 oldsd->output_queue = NULL;
10646 oldsd->output_queue_tailp = &oldsd->output_queue;
10648 /* Append NAPI poll list from offline CPU, with one exception :
10649 * process_backlog() must be called by cpu owning percpu backlog.
10650 * We properly handle process_queue & input_pkt_queue later.
10652 while (!list_empty(&oldsd->poll_list)) {
10653 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10654 struct napi_struct,
10657 list_del_init(&napi->poll_list);
10658 if (napi->poll == process_backlog)
10661 ____napi_schedule(sd, napi);
10664 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10665 local_irq_enable();
10668 remsd = oldsd->rps_ipi_list;
10669 oldsd->rps_ipi_list = NULL;
10671 /* send out pending IPI's on offline CPU */
10672 net_rps_send_ipi(remsd);
10674 /* Process offline CPU's input_pkt_queue */
10675 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10677 input_queue_head_incr(oldsd);
10679 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10681 input_queue_head_incr(oldsd);
10688 * netdev_increment_features - increment feature set by one
10689 * @all: current feature set
10690 * @one: new feature set
10691 * @mask: mask feature set
10693 * Computes a new feature set after adding a device with feature set
10694 * @one to the master device with current feature set @all. Will not
10695 * enable anything that is off in @mask. Returns the new feature set.
10697 netdev_features_t netdev_increment_features(netdev_features_t all,
10698 netdev_features_t one, netdev_features_t mask)
10700 if (mask & NETIF_F_HW_CSUM)
10701 mask |= NETIF_F_CSUM_MASK;
10702 mask |= NETIF_F_VLAN_CHALLENGED;
10704 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10705 all &= one | ~NETIF_F_ALL_FOR_ALL;
10707 /* If one device supports hw checksumming, set for all. */
10708 if (all & NETIF_F_HW_CSUM)
10709 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10713 EXPORT_SYMBOL(netdev_increment_features);
10715 static struct hlist_head * __net_init netdev_create_hash(void)
10718 struct hlist_head *hash;
10720 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10722 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10723 INIT_HLIST_HEAD(&hash[i]);
10728 /* Initialize per network namespace state */
10729 static int __net_init netdev_init(struct net *net)
10731 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10732 8 * sizeof_field(struct napi_struct, gro_bitmask));
10734 if (net != &init_net)
10735 INIT_LIST_HEAD(&net->dev_base_head);
10737 net->dev_name_head = netdev_create_hash();
10738 if (net->dev_name_head == NULL)
10741 net->dev_index_head = netdev_create_hash();
10742 if (net->dev_index_head == NULL)
10745 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10750 kfree(net->dev_name_head);
10756 * netdev_drivername - network driver for the device
10757 * @dev: network device
10759 * Determine network driver for device.
10761 const char *netdev_drivername(const struct net_device *dev)
10763 const struct device_driver *driver;
10764 const struct device *parent;
10765 const char *empty = "";
10767 parent = dev->dev.parent;
10771 driver = parent->driver;
10772 if (driver && driver->name)
10773 return driver->name;
10777 static void __netdev_printk(const char *level, const struct net_device *dev,
10778 struct va_format *vaf)
10780 if (dev && dev->dev.parent) {
10781 dev_printk_emit(level[1] - '0',
10784 dev_driver_string(dev->dev.parent),
10785 dev_name(dev->dev.parent),
10786 netdev_name(dev), netdev_reg_state(dev),
10789 printk("%s%s%s: %pV",
10790 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10792 printk("%s(NULL net_device): %pV", level, vaf);
10796 void netdev_printk(const char *level, const struct net_device *dev,
10797 const char *format, ...)
10799 struct va_format vaf;
10802 va_start(args, format);
10807 __netdev_printk(level, dev, &vaf);
10811 EXPORT_SYMBOL(netdev_printk);
10813 #define define_netdev_printk_level(func, level) \
10814 void func(const struct net_device *dev, const char *fmt, ...) \
10816 struct va_format vaf; \
10819 va_start(args, fmt); \
10824 __netdev_printk(level, dev, &vaf); \
10828 EXPORT_SYMBOL(func);
10830 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10831 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10832 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10833 define_netdev_printk_level(netdev_err, KERN_ERR);
10834 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10835 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10836 define_netdev_printk_level(netdev_info, KERN_INFO);
10838 static void __net_exit netdev_exit(struct net *net)
10840 kfree(net->dev_name_head);
10841 kfree(net->dev_index_head);
10842 if (net != &init_net)
10843 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10846 static struct pernet_operations __net_initdata netdev_net_ops = {
10847 .init = netdev_init,
10848 .exit = netdev_exit,
10851 static void __net_exit default_device_exit(struct net *net)
10853 struct net_device *dev, *aux;
10855 * Push all migratable network devices back to the
10856 * initial network namespace
10859 for_each_netdev_safe(net, dev, aux) {
10861 char fb_name[IFNAMSIZ];
10863 /* Ignore unmoveable devices (i.e. loopback) */
10864 if (dev->features & NETIF_F_NETNS_LOCAL)
10867 /* Leave virtual devices for the generic cleanup */
10868 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
10871 /* Push remaining network devices to init_net */
10872 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10873 if (netdev_name_in_use(&init_net, fb_name))
10874 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10875 err = dev_change_net_namespace(dev, &init_net, fb_name);
10877 pr_emerg("%s: failed to move %s to init_net: %d\n",
10878 __func__, dev->name, err);
10885 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10887 /* Return with the rtnl_lock held when there are no network
10888 * devices unregistering in any network namespace in net_list.
10891 bool unregistering;
10892 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10894 add_wait_queue(&netdev_unregistering_wq, &wait);
10896 unregistering = false;
10898 list_for_each_entry(net, net_list, exit_list) {
10899 if (net->dev_unreg_count > 0) {
10900 unregistering = true;
10904 if (!unregistering)
10908 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10910 remove_wait_queue(&netdev_unregistering_wq, &wait);
10913 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10915 /* At exit all network devices most be removed from a network
10916 * namespace. Do this in the reverse order of registration.
10917 * Do this across as many network namespaces as possible to
10918 * improve batching efficiency.
10920 struct net_device *dev;
10922 LIST_HEAD(dev_kill_list);
10924 /* To prevent network device cleanup code from dereferencing
10925 * loopback devices or network devices that have been freed
10926 * wait here for all pending unregistrations to complete,
10927 * before unregistring the loopback device and allowing the
10928 * network namespace be freed.
10930 * The netdev todo list containing all network devices
10931 * unregistrations that happen in default_device_exit_batch
10932 * will run in the rtnl_unlock() at the end of
10933 * default_device_exit_batch.
10935 rtnl_lock_unregistering(net_list);
10936 list_for_each_entry(net, net_list, exit_list) {
10937 for_each_netdev_reverse(net, dev) {
10938 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10939 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10941 unregister_netdevice_queue(dev, &dev_kill_list);
10944 unregister_netdevice_many(&dev_kill_list);
10948 static struct pernet_operations __net_initdata default_device_ops = {
10949 .exit = default_device_exit,
10950 .exit_batch = default_device_exit_batch,
10954 * Initialize the DEV module. At boot time this walks the device list and
10955 * unhooks any devices that fail to initialise (normally hardware not
10956 * present) and leaves us with a valid list of present and active devices.
10961 * This is called single threaded during boot, so no need
10962 * to take the rtnl semaphore.
10964 static int __init net_dev_init(void)
10966 int i, rc = -ENOMEM;
10968 BUG_ON(!dev_boot_phase);
10970 if (dev_proc_init())
10973 if (netdev_kobject_init())
10976 INIT_LIST_HEAD(&ptype_all);
10977 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10978 INIT_LIST_HEAD(&ptype_base[i]);
10980 if (register_pernet_subsys(&netdev_net_ops))
10984 * Initialise the packet receive queues.
10987 for_each_possible_cpu(i) {
10988 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10989 struct softnet_data *sd = &per_cpu(softnet_data, i);
10991 INIT_WORK(flush, flush_backlog);
10993 skb_queue_head_init(&sd->input_pkt_queue);
10994 skb_queue_head_init(&sd->process_queue);
10995 #ifdef CONFIG_XFRM_OFFLOAD
10996 skb_queue_head_init(&sd->xfrm_backlog);
10998 INIT_LIST_HEAD(&sd->poll_list);
10999 sd->output_queue_tailp = &sd->output_queue;
11001 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11005 init_gro_hash(&sd->backlog);
11006 sd->backlog.poll = process_backlog;
11007 sd->backlog.weight = weight_p;
11010 dev_boot_phase = 0;
11012 /* The loopback device is special if any other network devices
11013 * is present in a network namespace the loopback device must
11014 * be present. Since we now dynamically allocate and free the
11015 * loopback device ensure this invariant is maintained by
11016 * keeping the loopback device as the first device on the
11017 * list of network devices. Ensuring the loopback devices
11018 * is the first device that appears and the last network device
11021 if (register_pernet_device(&loopback_net_ops))
11024 if (register_pernet_device(&default_device_ops))
11027 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11028 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11030 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11031 NULL, dev_cpu_dead);
11038 subsys_initcall(net_dev_init);