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(&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(&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(&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(&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(&dev_base_lock);
1192 netdev_name_node_del(dev->name_node);
1193 write_unlock(&dev_base_lock);
1197 write_lock(&dev_base_lock);
1198 netdev_name_node_add(net, dev->name_node);
1199 write_unlock(&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 /* Other cpus might concurrently change txq->xmit_lock_owner
4097 * to -1 or to their cpu id, but not to our id.
4099 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4100 if (dev_xmit_recursion())
4101 goto recursion_alert;
4103 skb = validate_xmit_skb(skb, dev, &again);
4107 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4108 HARD_TX_LOCK(dev, txq, cpu);
4110 if (!netif_xmit_stopped(txq)) {
4111 dev_xmit_recursion_inc();
4112 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4113 dev_xmit_recursion_dec();
4114 if (dev_xmit_complete(rc)) {
4115 HARD_TX_UNLOCK(dev, txq);
4119 HARD_TX_UNLOCK(dev, txq);
4120 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4123 /* Recursion is detected! It is possible,
4127 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4133 rcu_read_unlock_bh();
4135 atomic_long_inc(&dev->tx_dropped);
4136 kfree_skb_list(skb);
4139 rcu_read_unlock_bh();
4143 int dev_queue_xmit(struct sk_buff *skb)
4145 return __dev_queue_xmit(skb, NULL);
4147 EXPORT_SYMBOL(dev_queue_xmit);
4149 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4151 return __dev_queue_xmit(skb, sb_dev);
4153 EXPORT_SYMBOL(dev_queue_xmit_accel);
4155 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4157 struct net_device *dev = skb->dev;
4158 struct sk_buff *orig_skb = skb;
4159 struct netdev_queue *txq;
4160 int ret = NETDEV_TX_BUSY;
4163 if (unlikely(!netif_running(dev) ||
4164 !netif_carrier_ok(dev)))
4167 skb = validate_xmit_skb_list(skb, dev, &again);
4168 if (skb != orig_skb)
4171 skb_set_queue_mapping(skb, queue_id);
4172 txq = skb_get_tx_queue(dev, skb);
4173 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4177 dev_xmit_recursion_inc();
4178 HARD_TX_LOCK(dev, txq, smp_processor_id());
4179 if (!netif_xmit_frozen_or_drv_stopped(txq))
4180 ret = netdev_start_xmit(skb, dev, txq, false);
4181 HARD_TX_UNLOCK(dev, txq);
4182 dev_xmit_recursion_dec();
4187 atomic_long_inc(&dev->tx_dropped);
4188 kfree_skb_list(skb);
4189 return NET_XMIT_DROP;
4191 EXPORT_SYMBOL(__dev_direct_xmit);
4193 /*************************************************************************
4195 *************************************************************************/
4197 int netdev_max_backlog __read_mostly = 1000;
4198 EXPORT_SYMBOL(netdev_max_backlog);
4200 int netdev_tstamp_prequeue __read_mostly = 1;
4201 int netdev_budget __read_mostly = 300;
4202 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4203 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4204 int weight_p __read_mostly = 64; /* old backlog weight */
4205 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4206 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4207 int dev_rx_weight __read_mostly = 64;
4208 int dev_tx_weight __read_mostly = 64;
4210 /* Called with irq disabled */
4211 static inline void ____napi_schedule(struct softnet_data *sd,
4212 struct napi_struct *napi)
4214 struct task_struct *thread;
4216 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4217 /* Paired with smp_mb__before_atomic() in
4218 * napi_enable()/dev_set_threaded().
4219 * Use READ_ONCE() to guarantee a complete
4220 * read on napi->thread. Only call
4221 * wake_up_process() when it's not NULL.
4223 thread = READ_ONCE(napi->thread);
4225 /* Avoid doing set_bit() if the thread is in
4226 * INTERRUPTIBLE state, cause napi_thread_wait()
4227 * makes sure to proceed with napi polling
4228 * if the thread is explicitly woken from here.
4230 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4231 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4232 wake_up_process(thread);
4237 list_add_tail(&napi->poll_list, &sd->poll_list);
4238 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4243 /* One global table that all flow-based protocols share. */
4244 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4245 EXPORT_SYMBOL(rps_sock_flow_table);
4246 u32 rps_cpu_mask __read_mostly;
4247 EXPORT_SYMBOL(rps_cpu_mask);
4249 struct static_key_false rps_needed __read_mostly;
4250 EXPORT_SYMBOL(rps_needed);
4251 struct static_key_false rfs_needed __read_mostly;
4252 EXPORT_SYMBOL(rfs_needed);
4254 static struct rps_dev_flow *
4255 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4256 struct rps_dev_flow *rflow, u16 next_cpu)
4258 if (next_cpu < nr_cpu_ids) {
4259 #ifdef CONFIG_RFS_ACCEL
4260 struct netdev_rx_queue *rxqueue;
4261 struct rps_dev_flow_table *flow_table;
4262 struct rps_dev_flow *old_rflow;
4267 /* Should we steer this flow to a different hardware queue? */
4268 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4269 !(dev->features & NETIF_F_NTUPLE))
4271 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4272 if (rxq_index == skb_get_rx_queue(skb))
4275 rxqueue = dev->_rx + rxq_index;
4276 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4279 flow_id = skb_get_hash(skb) & flow_table->mask;
4280 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4281 rxq_index, flow_id);
4285 rflow = &flow_table->flows[flow_id];
4287 if (old_rflow->filter == rflow->filter)
4288 old_rflow->filter = RPS_NO_FILTER;
4292 per_cpu(softnet_data, next_cpu).input_queue_head;
4295 rflow->cpu = next_cpu;
4300 * get_rps_cpu is called from netif_receive_skb and returns the target
4301 * CPU from the RPS map of the receiving queue for a given skb.
4302 * rcu_read_lock must be held on entry.
4304 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4305 struct rps_dev_flow **rflowp)
4307 const struct rps_sock_flow_table *sock_flow_table;
4308 struct netdev_rx_queue *rxqueue = dev->_rx;
4309 struct rps_dev_flow_table *flow_table;
4310 struct rps_map *map;
4315 if (skb_rx_queue_recorded(skb)) {
4316 u16 index = skb_get_rx_queue(skb);
4318 if (unlikely(index >= dev->real_num_rx_queues)) {
4319 WARN_ONCE(dev->real_num_rx_queues > 1,
4320 "%s received packet on queue %u, but number "
4321 "of RX queues is %u\n",
4322 dev->name, index, dev->real_num_rx_queues);
4328 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4330 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4331 map = rcu_dereference(rxqueue->rps_map);
4332 if (!flow_table && !map)
4335 skb_reset_network_header(skb);
4336 hash = skb_get_hash(skb);
4340 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4341 if (flow_table && sock_flow_table) {
4342 struct rps_dev_flow *rflow;
4346 /* First check into global flow table if there is a match */
4347 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4348 if ((ident ^ hash) & ~rps_cpu_mask)
4351 next_cpu = ident & rps_cpu_mask;
4353 /* OK, now we know there is a match,
4354 * we can look at the local (per receive queue) flow table
4356 rflow = &flow_table->flows[hash & flow_table->mask];
4360 * If the desired CPU (where last recvmsg was done) is
4361 * different from current CPU (one in the rx-queue flow
4362 * table entry), switch if one of the following holds:
4363 * - Current CPU is unset (>= nr_cpu_ids).
4364 * - Current CPU is offline.
4365 * - The current CPU's queue tail has advanced beyond the
4366 * last packet that was enqueued using this table entry.
4367 * This guarantees that all previous packets for the flow
4368 * have been dequeued, thus preserving in order delivery.
4370 if (unlikely(tcpu != next_cpu) &&
4371 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4372 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4373 rflow->last_qtail)) >= 0)) {
4375 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4378 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4388 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4389 if (cpu_online(tcpu)) {
4399 #ifdef CONFIG_RFS_ACCEL
4402 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4403 * @dev: Device on which the filter was set
4404 * @rxq_index: RX queue index
4405 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4406 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4408 * Drivers that implement ndo_rx_flow_steer() should periodically call
4409 * this function for each installed filter and remove the filters for
4410 * which it returns %true.
4412 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4413 u32 flow_id, u16 filter_id)
4415 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4416 struct rps_dev_flow_table *flow_table;
4417 struct rps_dev_flow *rflow;
4422 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4423 if (flow_table && flow_id <= flow_table->mask) {
4424 rflow = &flow_table->flows[flow_id];
4425 cpu = READ_ONCE(rflow->cpu);
4426 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4427 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4428 rflow->last_qtail) <
4429 (int)(10 * flow_table->mask)))
4435 EXPORT_SYMBOL(rps_may_expire_flow);
4437 #endif /* CONFIG_RFS_ACCEL */
4439 /* Called from hardirq (IPI) context */
4440 static void rps_trigger_softirq(void *data)
4442 struct softnet_data *sd = data;
4444 ____napi_schedule(sd, &sd->backlog);
4448 #endif /* CONFIG_RPS */
4451 * Check if this softnet_data structure is another cpu one
4452 * If yes, queue it to our IPI list and return 1
4455 static int rps_ipi_queued(struct softnet_data *sd)
4458 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4461 sd->rps_ipi_next = mysd->rps_ipi_list;
4462 mysd->rps_ipi_list = sd;
4464 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4467 #endif /* CONFIG_RPS */
4471 #ifdef CONFIG_NET_FLOW_LIMIT
4472 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4475 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4477 #ifdef CONFIG_NET_FLOW_LIMIT
4478 struct sd_flow_limit *fl;
4479 struct softnet_data *sd;
4480 unsigned int old_flow, new_flow;
4482 if (qlen < (netdev_max_backlog >> 1))
4485 sd = this_cpu_ptr(&softnet_data);
4488 fl = rcu_dereference(sd->flow_limit);
4490 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4491 old_flow = fl->history[fl->history_head];
4492 fl->history[fl->history_head] = new_flow;
4495 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4497 if (likely(fl->buckets[old_flow]))
4498 fl->buckets[old_flow]--;
4500 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4512 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4513 * queue (may be a remote CPU queue).
4515 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4516 unsigned int *qtail)
4518 struct softnet_data *sd;
4519 unsigned long flags;
4522 sd = &per_cpu(softnet_data, cpu);
4524 local_irq_save(flags);
4527 if (!netif_running(skb->dev))
4529 qlen = skb_queue_len(&sd->input_pkt_queue);
4530 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4533 __skb_queue_tail(&sd->input_pkt_queue, skb);
4534 input_queue_tail_incr_save(sd, qtail);
4536 local_irq_restore(flags);
4537 return NET_RX_SUCCESS;
4540 /* Schedule NAPI for backlog device
4541 * We can use non atomic operation since we own the queue lock
4543 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4544 if (!rps_ipi_queued(sd))
4545 ____napi_schedule(sd, &sd->backlog);
4554 local_irq_restore(flags);
4556 atomic_long_inc(&skb->dev->rx_dropped);
4561 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4563 struct net_device *dev = skb->dev;
4564 struct netdev_rx_queue *rxqueue;
4568 if (skb_rx_queue_recorded(skb)) {
4569 u16 index = skb_get_rx_queue(skb);
4571 if (unlikely(index >= dev->real_num_rx_queues)) {
4572 WARN_ONCE(dev->real_num_rx_queues > 1,
4573 "%s received packet on queue %u, but number "
4574 "of RX queues is %u\n",
4575 dev->name, index, dev->real_num_rx_queues);
4577 return rxqueue; /* Return first rxqueue */
4584 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4585 struct bpf_prog *xdp_prog)
4587 void *orig_data, *orig_data_end, *hard_start;
4588 struct netdev_rx_queue *rxqueue;
4589 bool orig_bcast, orig_host;
4590 u32 mac_len, frame_sz;
4591 __be16 orig_eth_type;
4596 /* The XDP program wants to see the packet starting at the MAC
4599 mac_len = skb->data - skb_mac_header(skb);
4600 hard_start = skb->data - skb_headroom(skb);
4602 /* SKB "head" area always have tailroom for skb_shared_info */
4603 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4604 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4606 rxqueue = netif_get_rxqueue(skb);
4607 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4608 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4609 skb_headlen(skb) + mac_len, true);
4611 orig_data_end = xdp->data_end;
4612 orig_data = xdp->data;
4613 eth = (struct ethhdr *)xdp->data;
4614 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4615 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4616 orig_eth_type = eth->h_proto;
4618 act = bpf_prog_run_xdp(xdp_prog, xdp);
4620 /* check if bpf_xdp_adjust_head was used */
4621 off = xdp->data - orig_data;
4624 __skb_pull(skb, off);
4626 __skb_push(skb, -off);
4628 skb->mac_header += off;
4629 skb_reset_network_header(skb);
4632 /* check if bpf_xdp_adjust_tail was used */
4633 off = xdp->data_end - orig_data_end;
4635 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4636 skb->len += off; /* positive on grow, negative on shrink */
4639 /* check if XDP changed eth hdr such SKB needs update */
4640 eth = (struct ethhdr *)xdp->data;
4641 if ((orig_eth_type != eth->h_proto) ||
4642 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4643 skb->dev->dev_addr)) ||
4644 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4645 __skb_push(skb, ETH_HLEN);
4646 skb->pkt_type = PACKET_HOST;
4647 skb->protocol = eth_type_trans(skb, skb->dev);
4650 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4651 * before calling us again on redirect path. We do not call do_redirect
4652 * as we leave that up to the caller.
4654 * Caller is responsible for managing lifetime of skb (i.e. calling
4655 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4660 __skb_push(skb, mac_len);
4663 metalen = xdp->data - xdp->data_meta;
4665 skb_metadata_set(skb, metalen);
4672 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4673 struct xdp_buff *xdp,
4674 struct bpf_prog *xdp_prog)
4678 /* Reinjected packets coming from act_mirred or similar should
4679 * not get XDP generic processing.
4681 if (skb_is_redirected(skb))
4684 /* XDP packets must be linear and must have sufficient headroom
4685 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4686 * native XDP provides, thus we need to do it here as well.
4688 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4689 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4690 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4691 int troom = skb->tail + skb->data_len - skb->end;
4693 /* In case we have to go down the path and also linearize,
4694 * then lets do the pskb_expand_head() work just once here.
4696 if (pskb_expand_head(skb,
4697 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4698 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4700 if (skb_linearize(skb))
4704 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4711 bpf_warn_invalid_xdp_action(act);
4714 trace_xdp_exception(skb->dev, xdp_prog, act);
4725 /* When doing generic XDP we have to bypass the qdisc layer and the
4726 * network taps in order to match in-driver-XDP behavior.
4728 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4730 struct net_device *dev = skb->dev;
4731 struct netdev_queue *txq;
4732 bool free_skb = true;
4735 txq = netdev_core_pick_tx(dev, skb, NULL);
4736 cpu = smp_processor_id();
4737 HARD_TX_LOCK(dev, txq, cpu);
4738 if (!netif_xmit_stopped(txq)) {
4739 rc = netdev_start_xmit(skb, dev, txq, 0);
4740 if (dev_xmit_complete(rc))
4743 HARD_TX_UNLOCK(dev, txq);
4745 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4750 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4752 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4755 struct xdp_buff xdp;
4759 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4760 if (act != XDP_PASS) {
4763 err = xdp_do_generic_redirect(skb->dev, skb,
4769 generic_xdp_tx(skb, xdp_prog);
4780 EXPORT_SYMBOL_GPL(do_xdp_generic);
4782 static int netif_rx_internal(struct sk_buff *skb)
4786 net_timestamp_check(netdev_tstamp_prequeue, skb);
4788 trace_netif_rx(skb);
4791 if (static_branch_unlikely(&rps_needed)) {
4792 struct rps_dev_flow voidflow, *rflow = &voidflow;
4798 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4800 cpu = smp_processor_id();
4802 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4811 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4818 * netif_rx - post buffer to the network code
4819 * @skb: buffer to post
4821 * This function receives a packet from a device driver and queues it for
4822 * the upper (protocol) levels to process. It always succeeds. The buffer
4823 * may be dropped during processing for congestion control or by the
4827 * NET_RX_SUCCESS (no congestion)
4828 * NET_RX_DROP (packet was dropped)
4832 int netif_rx(struct sk_buff *skb)
4836 trace_netif_rx_entry(skb);
4838 ret = netif_rx_internal(skb);
4839 trace_netif_rx_exit(ret);
4843 EXPORT_SYMBOL(netif_rx);
4845 int netif_rx_ni(struct sk_buff *skb)
4849 trace_netif_rx_ni_entry(skb);
4852 err = netif_rx_internal(skb);
4853 if (local_softirq_pending())
4856 trace_netif_rx_ni_exit(err);
4860 EXPORT_SYMBOL(netif_rx_ni);
4862 int netif_rx_any_context(struct sk_buff *skb)
4865 * If invoked from contexts which do not invoke bottom half
4866 * processing either at return from interrupt or when softrqs are
4867 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4871 return netif_rx(skb);
4873 return netif_rx_ni(skb);
4875 EXPORT_SYMBOL(netif_rx_any_context);
4877 static __latent_entropy void net_tx_action(struct softirq_action *h)
4879 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4881 if (sd->completion_queue) {
4882 struct sk_buff *clist;
4884 local_irq_disable();
4885 clist = sd->completion_queue;
4886 sd->completion_queue = NULL;
4890 struct sk_buff *skb = clist;
4892 clist = clist->next;
4894 WARN_ON(refcount_read(&skb->users));
4895 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4896 trace_consume_skb(skb);
4898 trace_kfree_skb(skb, net_tx_action);
4900 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4903 __kfree_skb_defer(skb);
4907 if (sd->output_queue) {
4910 local_irq_disable();
4911 head = sd->output_queue;
4912 sd->output_queue = NULL;
4913 sd->output_queue_tailp = &sd->output_queue;
4919 struct Qdisc *q = head;
4920 spinlock_t *root_lock = NULL;
4922 head = head->next_sched;
4924 /* We need to make sure head->next_sched is read
4925 * before clearing __QDISC_STATE_SCHED
4927 smp_mb__before_atomic();
4929 if (!(q->flags & TCQ_F_NOLOCK)) {
4930 root_lock = qdisc_lock(q);
4931 spin_lock(root_lock);
4932 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
4934 /* There is a synchronize_net() between
4935 * STATE_DEACTIVATED flag being set and
4936 * qdisc_reset()/some_qdisc_is_busy() in
4937 * dev_deactivate(), so we can safely bail out
4938 * early here to avoid data race between
4939 * qdisc_deactivate() and some_qdisc_is_busy()
4940 * for lockless qdisc.
4942 clear_bit(__QDISC_STATE_SCHED, &q->state);
4946 clear_bit(__QDISC_STATE_SCHED, &q->state);
4949 spin_unlock(root_lock);
4955 xfrm_dev_backlog(sd);
4958 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4959 /* This hook is defined here for ATM LANE */
4960 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4961 unsigned char *addr) __read_mostly;
4962 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4965 static inline struct sk_buff *
4966 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4967 struct net_device *orig_dev, bool *another)
4969 #ifdef CONFIG_NET_CLS_ACT
4970 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4971 struct tcf_result cl_res;
4973 /* If there's at least one ingress present somewhere (so
4974 * we get here via enabled static key), remaining devices
4975 * that are not configured with an ingress qdisc will bail
4982 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4986 qdisc_skb_cb(skb)->pkt_len = skb->len;
4987 qdisc_skb_cb(skb)->mru = 0;
4988 qdisc_skb_cb(skb)->post_ct = false;
4989 skb->tc_at_ingress = 1;
4990 mini_qdisc_bstats_cpu_update(miniq, skb);
4992 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
4994 case TC_ACT_RECLASSIFY:
4995 skb->tc_index = TC_H_MIN(cl_res.classid);
4998 mini_qdisc_qstats_cpu_drop(miniq);
5006 case TC_ACT_REDIRECT:
5007 /* skb_mac_header check was done by cls/act_bpf, so
5008 * we can safely push the L2 header back before
5009 * redirecting to another netdev
5011 __skb_push(skb, skb->mac_len);
5012 if (skb_do_redirect(skb) == -EAGAIN) {
5013 __skb_pull(skb, skb->mac_len);
5018 case TC_ACT_CONSUMED:
5023 #endif /* CONFIG_NET_CLS_ACT */
5028 * netdev_is_rx_handler_busy - check if receive handler is registered
5029 * @dev: device to check
5031 * Check if a receive handler is already registered for a given device.
5032 * Return true if there one.
5034 * The caller must hold the rtnl_mutex.
5036 bool netdev_is_rx_handler_busy(struct net_device *dev)
5039 return dev && rtnl_dereference(dev->rx_handler);
5041 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5044 * netdev_rx_handler_register - register receive handler
5045 * @dev: device to register a handler for
5046 * @rx_handler: receive handler to register
5047 * @rx_handler_data: data pointer that is used by rx handler
5049 * Register a receive handler for a device. This handler will then be
5050 * called from __netif_receive_skb. A negative errno code is returned
5053 * The caller must hold the rtnl_mutex.
5055 * For a general description of rx_handler, see enum rx_handler_result.
5057 int netdev_rx_handler_register(struct net_device *dev,
5058 rx_handler_func_t *rx_handler,
5059 void *rx_handler_data)
5061 if (netdev_is_rx_handler_busy(dev))
5064 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5067 /* Note: rx_handler_data must be set before rx_handler */
5068 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5069 rcu_assign_pointer(dev->rx_handler, rx_handler);
5073 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5076 * netdev_rx_handler_unregister - unregister receive handler
5077 * @dev: device to unregister a handler from
5079 * Unregister a receive handler from a device.
5081 * The caller must hold the rtnl_mutex.
5083 void netdev_rx_handler_unregister(struct net_device *dev)
5087 RCU_INIT_POINTER(dev->rx_handler, NULL);
5088 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5089 * section has a guarantee to see a non NULL rx_handler_data
5093 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5095 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5098 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5099 * the special handling of PFMEMALLOC skbs.
5101 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5103 switch (skb->protocol) {
5104 case htons(ETH_P_ARP):
5105 case htons(ETH_P_IP):
5106 case htons(ETH_P_IPV6):
5107 case htons(ETH_P_8021Q):
5108 case htons(ETH_P_8021AD):
5115 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5116 int *ret, struct net_device *orig_dev)
5118 if (nf_hook_ingress_active(skb)) {
5122 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5127 ingress_retval = nf_hook_ingress(skb);
5129 return ingress_retval;
5134 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5135 struct packet_type **ppt_prev)
5137 struct packet_type *ptype, *pt_prev;
5138 rx_handler_func_t *rx_handler;
5139 struct sk_buff *skb = *pskb;
5140 struct net_device *orig_dev;
5141 bool deliver_exact = false;
5142 int ret = NET_RX_DROP;
5145 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5147 trace_netif_receive_skb(skb);
5149 orig_dev = skb->dev;
5151 skb_reset_network_header(skb);
5152 if (!skb_transport_header_was_set(skb))
5153 skb_reset_transport_header(skb);
5154 skb_reset_mac_len(skb);
5159 skb->skb_iif = skb->dev->ifindex;
5161 __this_cpu_inc(softnet_data.processed);
5163 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5167 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5170 if (ret2 != XDP_PASS) {
5176 if (eth_type_vlan(skb->protocol)) {
5177 skb = skb_vlan_untag(skb);
5182 if (skb_skip_tc_classify(skb))
5188 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5190 ret = deliver_skb(skb, pt_prev, orig_dev);
5194 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5196 ret = deliver_skb(skb, pt_prev, orig_dev);
5201 #ifdef CONFIG_NET_INGRESS
5202 if (static_branch_unlikely(&ingress_needed_key)) {
5203 bool another = false;
5205 nf_skip_egress(skb, true);
5206 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5213 nf_skip_egress(skb, false);
5214 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5218 skb_reset_redirect(skb);
5220 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5223 if (skb_vlan_tag_present(skb)) {
5225 ret = deliver_skb(skb, pt_prev, orig_dev);
5228 if (vlan_do_receive(&skb))
5230 else if (unlikely(!skb))
5234 rx_handler = rcu_dereference(skb->dev->rx_handler);
5237 ret = deliver_skb(skb, pt_prev, orig_dev);
5240 switch (rx_handler(&skb)) {
5241 case RX_HANDLER_CONSUMED:
5242 ret = NET_RX_SUCCESS;
5244 case RX_HANDLER_ANOTHER:
5246 case RX_HANDLER_EXACT:
5247 deliver_exact = true;
5249 case RX_HANDLER_PASS:
5256 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5258 if (skb_vlan_tag_get_id(skb)) {
5259 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5262 skb->pkt_type = PACKET_OTHERHOST;
5263 } else if (eth_type_vlan(skb->protocol)) {
5264 /* Outer header is 802.1P with vlan 0, inner header is
5265 * 802.1Q or 802.1AD and vlan_do_receive() above could
5266 * not find vlan dev for vlan id 0.
5268 __vlan_hwaccel_clear_tag(skb);
5269 skb = skb_vlan_untag(skb);
5272 if (vlan_do_receive(&skb))
5273 /* After stripping off 802.1P header with vlan 0
5274 * vlan dev is found for inner header.
5277 else if (unlikely(!skb))
5280 /* We have stripped outer 802.1P vlan 0 header.
5281 * But could not find vlan dev.
5282 * check again for vlan id to set OTHERHOST.
5286 /* Note: we might in the future use prio bits
5287 * and set skb->priority like in vlan_do_receive()
5288 * For the time being, just ignore Priority Code Point
5290 __vlan_hwaccel_clear_tag(skb);
5293 type = skb->protocol;
5295 /* deliver only exact match when indicated */
5296 if (likely(!deliver_exact)) {
5297 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5298 &ptype_base[ntohs(type) &
5302 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5303 &orig_dev->ptype_specific);
5305 if (unlikely(skb->dev != orig_dev)) {
5306 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5307 &skb->dev->ptype_specific);
5311 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5313 *ppt_prev = pt_prev;
5317 atomic_long_inc(&skb->dev->rx_dropped);
5319 atomic_long_inc(&skb->dev->rx_nohandler);
5321 /* Jamal, now you will not able to escape explaining
5322 * me how you were going to use this. :-)
5328 /* The invariant here is that if *ppt_prev is not NULL
5329 * then skb should also be non-NULL.
5331 * Apparently *ppt_prev assignment above holds this invariant due to
5332 * skb dereferencing near it.
5338 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5340 struct net_device *orig_dev = skb->dev;
5341 struct packet_type *pt_prev = NULL;
5344 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5346 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5347 skb->dev, pt_prev, orig_dev);
5352 * netif_receive_skb_core - special purpose version of netif_receive_skb
5353 * @skb: buffer to process
5355 * More direct receive version of netif_receive_skb(). It should
5356 * only be used by callers that have a need to skip RPS and Generic XDP.
5357 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5359 * This function may only be called from softirq context and interrupts
5360 * should be enabled.
5362 * Return values (usually ignored):
5363 * NET_RX_SUCCESS: no congestion
5364 * NET_RX_DROP: packet was dropped
5366 int netif_receive_skb_core(struct sk_buff *skb)
5371 ret = __netif_receive_skb_one_core(skb, false);
5376 EXPORT_SYMBOL(netif_receive_skb_core);
5378 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5379 struct packet_type *pt_prev,
5380 struct net_device *orig_dev)
5382 struct sk_buff *skb, *next;
5386 if (list_empty(head))
5388 if (pt_prev->list_func != NULL)
5389 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5390 ip_list_rcv, head, pt_prev, orig_dev);
5392 list_for_each_entry_safe(skb, next, head, list) {
5393 skb_list_del_init(skb);
5394 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5398 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5400 /* Fast-path assumptions:
5401 * - There is no RX handler.
5402 * - Only one packet_type matches.
5403 * If either of these fails, we will end up doing some per-packet
5404 * processing in-line, then handling the 'last ptype' for the whole
5405 * sublist. This can't cause out-of-order delivery to any single ptype,
5406 * because the 'last ptype' must be constant across the sublist, and all
5407 * other ptypes are handled per-packet.
5409 /* Current (common) ptype of sublist */
5410 struct packet_type *pt_curr = NULL;
5411 /* Current (common) orig_dev of sublist */
5412 struct net_device *od_curr = NULL;
5413 struct list_head sublist;
5414 struct sk_buff *skb, *next;
5416 INIT_LIST_HEAD(&sublist);
5417 list_for_each_entry_safe(skb, next, head, list) {
5418 struct net_device *orig_dev = skb->dev;
5419 struct packet_type *pt_prev = NULL;
5421 skb_list_del_init(skb);
5422 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5425 if (pt_curr != pt_prev || od_curr != orig_dev) {
5426 /* dispatch old sublist */
5427 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5428 /* start new sublist */
5429 INIT_LIST_HEAD(&sublist);
5433 list_add_tail(&skb->list, &sublist);
5436 /* dispatch final sublist */
5437 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5440 static int __netif_receive_skb(struct sk_buff *skb)
5444 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5445 unsigned int noreclaim_flag;
5448 * PFMEMALLOC skbs are special, they should
5449 * - be delivered to SOCK_MEMALLOC sockets only
5450 * - stay away from userspace
5451 * - have bounded memory usage
5453 * Use PF_MEMALLOC as this saves us from propagating the allocation
5454 * context down to all allocation sites.
5456 noreclaim_flag = memalloc_noreclaim_save();
5457 ret = __netif_receive_skb_one_core(skb, true);
5458 memalloc_noreclaim_restore(noreclaim_flag);
5460 ret = __netif_receive_skb_one_core(skb, false);
5465 static void __netif_receive_skb_list(struct list_head *head)
5467 unsigned long noreclaim_flag = 0;
5468 struct sk_buff *skb, *next;
5469 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5471 list_for_each_entry_safe(skb, next, head, list) {
5472 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5473 struct list_head sublist;
5475 /* Handle the previous sublist */
5476 list_cut_before(&sublist, head, &skb->list);
5477 if (!list_empty(&sublist))
5478 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5479 pfmemalloc = !pfmemalloc;
5480 /* See comments in __netif_receive_skb */
5482 noreclaim_flag = memalloc_noreclaim_save();
5484 memalloc_noreclaim_restore(noreclaim_flag);
5487 /* Handle the remaining sublist */
5488 if (!list_empty(head))
5489 __netif_receive_skb_list_core(head, pfmemalloc);
5490 /* Restore pflags */
5492 memalloc_noreclaim_restore(noreclaim_flag);
5495 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5497 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5498 struct bpf_prog *new = xdp->prog;
5501 switch (xdp->command) {
5502 case XDP_SETUP_PROG:
5503 rcu_assign_pointer(dev->xdp_prog, new);
5508 static_branch_dec(&generic_xdp_needed_key);
5509 } else if (new && !old) {
5510 static_branch_inc(&generic_xdp_needed_key);
5511 dev_disable_lro(dev);
5512 dev_disable_gro_hw(dev);
5524 static int netif_receive_skb_internal(struct sk_buff *skb)
5528 net_timestamp_check(netdev_tstamp_prequeue, skb);
5530 if (skb_defer_rx_timestamp(skb))
5531 return NET_RX_SUCCESS;
5535 if (static_branch_unlikely(&rps_needed)) {
5536 struct rps_dev_flow voidflow, *rflow = &voidflow;
5537 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5540 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5546 ret = __netif_receive_skb(skb);
5551 void netif_receive_skb_list_internal(struct list_head *head)
5553 struct sk_buff *skb, *next;
5554 struct list_head sublist;
5556 INIT_LIST_HEAD(&sublist);
5557 list_for_each_entry_safe(skb, next, head, list) {
5558 net_timestamp_check(netdev_tstamp_prequeue, skb);
5559 skb_list_del_init(skb);
5560 if (!skb_defer_rx_timestamp(skb))
5561 list_add_tail(&skb->list, &sublist);
5563 list_splice_init(&sublist, head);
5567 if (static_branch_unlikely(&rps_needed)) {
5568 list_for_each_entry_safe(skb, next, head, list) {
5569 struct rps_dev_flow voidflow, *rflow = &voidflow;
5570 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5573 /* Will be handled, remove from list */
5574 skb_list_del_init(skb);
5575 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5580 __netif_receive_skb_list(head);
5585 * netif_receive_skb - process receive buffer from network
5586 * @skb: buffer to process
5588 * netif_receive_skb() is the main receive data processing function.
5589 * It always succeeds. The buffer may be dropped during processing
5590 * for congestion control or by the protocol layers.
5592 * This function may only be called from softirq context and interrupts
5593 * should be enabled.
5595 * Return values (usually ignored):
5596 * NET_RX_SUCCESS: no congestion
5597 * NET_RX_DROP: packet was dropped
5599 int netif_receive_skb(struct sk_buff *skb)
5603 trace_netif_receive_skb_entry(skb);
5605 ret = netif_receive_skb_internal(skb);
5606 trace_netif_receive_skb_exit(ret);
5610 EXPORT_SYMBOL(netif_receive_skb);
5613 * netif_receive_skb_list - process many receive buffers from network
5614 * @head: list of skbs to process.
5616 * Since return value of netif_receive_skb() is normally ignored, and
5617 * wouldn't be meaningful for a list, this function returns void.
5619 * This function may only be called from softirq context and interrupts
5620 * should be enabled.
5622 void netif_receive_skb_list(struct list_head *head)
5624 struct sk_buff *skb;
5626 if (list_empty(head))
5628 if (trace_netif_receive_skb_list_entry_enabled()) {
5629 list_for_each_entry(skb, head, list)
5630 trace_netif_receive_skb_list_entry(skb);
5632 netif_receive_skb_list_internal(head);
5633 trace_netif_receive_skb_list_exit(0);
5635 EXPORT_SYMBOL(netif_receive_skb_list);
5637 static DEFINE_PER_CPU(struct work_struct, flush_works);
5639 /* Network device is going away, flush any packets still pending */
5640 static void flush_backlog(struct work_struct *work)
5642 struct sk_buff *skb, *tmp;
5643 struct softnet_data *sd;
5646 sd = this_cpu_ptr(&softnet_data);
5648 local_irq_disable();
5650 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5651 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5652 __skb_unlink(skb, &sd->input_pkt_queue);
5653 dev_kfree_skb_irq(skb);
5654 input_queue_head_incr(sd);
5660 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5661 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5662 __skb_unlink(skb, &sd->process_queue);
5664 input_queue_head_incr(sd);
5670 static bool flush_required(int cpu)
5672 #if IS_ENABLED(CONFIG_RPS)
5673 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5676 local_irq_disable();
5679 /* as insertion into process_queue happens with the rps lock held,
5680 * process_queue access may race only with dequeue
5682 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5683 !skb_queue_empty_lockless(&sd->process_queue);
5689 /* without RPS we can't safely check input_pkt_queue: during a
5690 * concurrent remote skb_queue_splice() we can detect as empty both
5691 * input_pkt_queue and process_queue even if the latter could end-up
5692 * containing a lot of packets.
5697 static void flush_all_backlogs(void)
5699 static cpumask_t flush_cpus;
5702 /* since we are under rtnl lock protection we can use static data
5703 * for the cpumask and avoid allocating on stack the possibly
5710 cpumask_clear(&flush_cpus);
5711 for_each_online_cpu(cpu) {
5712 if (flush_required(cpu)) {
5713 queue_work_on(cpu, system_highpri_wq,
5714 per_cpu_ptr(&flush_works, cpu));
5715 cpumask_set_cpu(cpu, &flush_cpus);
5719 /* we can have in flight packet[s] on the cpus we are not flushing,
5720 * synchronize_net() in unregister_netdevice_many() will take care of
5723 for_each_cpu(cpu, &flush_cpus)
5724 flush_work(per_cpu_ptr(&flush_works, cpu));
5729 static void net_rps_send_ipi(struct softnet_data *remsd)
5733 struct softnet_data *next = remsd->rps_ipi_next;
5735 if (cpu_online(remsd->cpu))
5736 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5743 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5744 * Note: called with local irq disabled, but exits with local irq enabled.
5746 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5749 struct softnet_data *remsd = sd->rps_ipi_list;
5752 sd->rps_ipi_list = NULL;
5756 /* Send pending IPI's to kick RPS processing on remote cpus. */
5757 net_rps_send_ipi(remsd);
5763 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5766 return sd->rps_ipi_list != NULL;
5772 static int process_backlog(struct napi_struct *napi, int quota)
5774 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5778 /* Check if we have pending ipi, its better to send them now,
5779 * not waiting net_rx_action() end.
5781 if (sd_has_rps_ipi_waiting(sd)) {
5782 local_irq_disable();
5783 net_rps_action_and_irq_enable(sd);
5786 napi->weight = dev_rx_weight;
5788 struct sk_buff *skb;
5790 while ((skb = __skb_dequeue(&sd->process_queue))) {
5792 __netif_receive_skb(skb);
5794 input_queue_head_incr(sd);
5795 if (++work >= quota)
5800 local_irq_disable();
5802 if (skb_queue_empty(&sd->input_pkt_queue)) {
5804 * Inline a custom version of __napi_complete().
5805 * only current cpu owns and manipulates this napi,
5806 * and NAPI_STATE_SCHED is the only possible flag set
5808 * We can use a plain write instead of clear_bit(),
5809 * and we dont need an smp_mb() memory barrier.
5814 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5815 &sd->process_queue);
5825 * __napi_schedule - schedule for receive
5826 * @n: entry to schedule
5828 * The entry's receive function will be scheduled to run.
5829 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5831 void __napi_schedule(struct napi_struct *n)
5833 unsigned long flags;
5835 local_irq_save(flags);
5836 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5837 local_irq_restore(flags);
5839 EXPORT_SYMBOL(__napi_schedule);
5842 * napi_schedule_prep - check if napi can be scheduled
5845 * Test if NAPI routine is already running, and if not mark
5846 * it as running. This is used as a condition variable to
5847 * insure only one NAPI poll instance runs. We also make
5848 * sure there is no pending NAPI disable.
5850 bool napi_schedule_prep(struct napi_struct *n)
5852 unsigned long val, new;
5855 val = READ_ONCE(n->state);
5856 if (unlikely(val & NAPIF_STATE_DISABLE))
5858 new = val | NAPIF_STATE_SCHED;
5860 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5861 * This was suggested by Alexander Duyck, as compiler
5862 * emits better code than :
5863 * if (val & NAPIF_STATE_SCHED)
5864 * new |= NAPIF_STATE_MISSED;
5866 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5868 } while (cmpxchg(&n->state, val, new) != val);
5870 return !(val & NAPIF_STATE_SCHED);
5872 EXPORT_SYMBOL(napi_schedule_prep);
5875 * __napi_schedule_irqoff - schedule for receive
5876 * @n: entry to schedule
5878 * Variant of __napi_schedule() assuming hard irqs are masked.
5880 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5881 * because the interrupt disabled assumption might not be true
5882 * due to force-threaded interrupts and spinlock substitution.
5884 void __napi_schedule_irqoff(struct napi_struct *n)
5886 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5887 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5891 EXPORT_SYMBOL(__napi_schedule_irqoff);
5893 bool napi_complete_done(struct napi_struct *n, int work_done)
5895 unsigned long flags, val, new, timeout = 0;
5899 * 1) Don't let napi dequeue from the cpu poll list
5900 * just in case its running on a different cpu.
5901 * 2) If we are busy polling, do nothing here, we have
5902 * the guarantee we will be called later.
5904 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5905 NAPIF_STATE_IN_BUSY_POLL)))
5910 timeout = READ_ONCE(n->dev->gro_flush_timeout);
5911 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
5913 if (n->defer_hard_irqs_count > 0) {
5914 n->defer_hard_irqs_count--;
5915 timeout = READ_ONCE(n->dev->gro_flush_timeout);
5919 if (n->gro_bitmask) {
5920 /* When the NAPI instance uses a timeout and keeps postponing
5921 * it, we need to bound somehow the time packets are kept in
5924 napi_gro_flush(n, !!timeout);
5929 if (unlikely(!list_empty(&n->poll_list))) {
5930 /* If n->poll_list is not empty, we need to mask irqs */
5931 local_irq_save(flags);
5932 list_del_init(&n->poll_list);
5933 local_irq_restore(flags);
5937 val = READ_ONCE(n->state);
5939 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5941 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
5942 NAPIF_STATE_SCHED_THREADED |
5943 NAPIF_STATE_PREFER_BUSY_POLL);
5945 /* If STATE_MISSED was set, leave STATE_SCHED set,
5946 * because we will call napi->poll() one more time.
5947 * This C code was suggested by Alexander Duyck to help gcc.
5949 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5951 } while (cmpxchg(&n->state, val, new) != val);
5953 if (unlikely(val & NAPIF_STATE_MISSED)) {
5959 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5960 HRTIMER_MODE_REL_PINNED);
5963 EXPORT_SYMBOL(napi_complete_done);
5965 /* must be called under rcu_read_lock(), as we dont take a reference */
5966 static struct napi_struct *napi_by_id(unsigned int napi_id)
5968 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5969 struct napi_struct *napi;
5971 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5972 if (napi->napi_id == napi_id)
5978 #if defined(CONFIG_NET_RX_BUSY_POLL)
5980 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
5982 if (!skip_schedule) {
5983 gro_normal_list(napi);
5984 __napi_schedule(napi);
5988 if (napi->gro_bitmask) {
5989 /* flush too old packets
5990 * If HZ < 1000, flush all packets.
5992 napi_gro_flush(napi, HZ >= 1000);
5995 gro_normal_list(napi);
5996 clear_bit(NAPI_STATE_SCHED, &napi->state);
5999 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6002 bool skip_schedule = false;
6003 unsigned long timeout;
6006 /* Busy polling means there is a high chance device driver hard irq
6007 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6008 * set in napi_schedule_prep().
6009 * Since we are about to call napi->poll() once more, we can safely
6010 * clear NAPI_STATE_MISSED.
6012 * Note: x86 could use a single "lock and ..." instruction
6013 * to perform these two clear_bit()
6015 clear_bit(NAPI_STATE_MISSED, &napi->state);
6016 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6020 if (prefer_busy_poll) {
6021 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6022 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6023 if (napi->defer_hard_irqs_count && timeout) {
6024 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6025 skip_schedule = true;
6029 /* All we really want here is to re-enable device interrupts.
6030 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6032 rc = napi->poll(napi, budget);
6033 /* We can't gro_normal_list() here, because napi->poll() might have
6034 * rearmed the napi (napi_complete_done()) in which case it could
6035 * already be running on another CPU.
6037 trace_napi_poll(napi, rc, budget);
6038 netpoll_poll_unlock(have_poll_lock);
6040 __busy_poll_stop(napi, skip_schedule);
6044 void napi_busy_loop(unsigned int napi_id,
6045 bool (*loop_end)(void *, unsigned long),
6046 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6048 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6049 int (*napi_poll)(struct napi_struct *napi, int budget);
6050 void *have_poll_lock = NULL;
6051 struct napi_struct *napi;
6058 napi = napi_by_id(napi_id);
6068 unsigned long val = READ_ONCE(napi->state);
6070 /* If multiple threads are competing for this napi,
6071 * we avoid dirtying napi->state as much as we can.
6073 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6074 NAPIF_STATE_IN_BUSY_POLL)) {
6075 if (prefer_busy_poll)
6076 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6079 if (cmpxchg(&napi->state, val,
6080 val | NAPIF_STATE_IN_BUSY_POLL |
6081 NAPIF_STATE_SCHED) != val) {
6082 if (prefer_busy_poll)
6083 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6086 have_poll_lock = netpoll_poll_lock(napi);
6087 napi_poll = napi->poll;
6089 work = napi_poll(napi, budget);
6090 trace_napi_poll(napi, work, budget);
6091 gro_normal_list(napi);
6094 __NET_ADD_STATS(dev_net(napi->dev),
6095 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6098 if (!loop_end || loop_end(loop_end_arg, start_time))
6101 if (unlikely(need_resched())) {
6103 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6107 if (loop_end(loop_end_arg, start_time))
6114 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6119 EXPORT_SYMBOL(napi_busy_loop);
6121 #endif /* CONFIG_NET_RX_BUSY_POLL */
6123 static void napi_hash_add(struct napi_struct *napi)
6125 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6128 spin_lock(&napi_hash_lock);
6130 /* 0..NR_CPUS range is reserved for sender_cpu use */
6132 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6133 napi_gen_id = MIN_NAPI_ID;
6134 } while (napi_by_id(napi_gen_id));
6135 napi->napi_id = napi_gen_id;
6137 hlist_add_head_rcu(&napi->napi_hash_node,
6138 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6140 spin_unlock(&napi_hash_lock);
6143 /* Warning : caller is responsible to make sure rcu grace period
6144 * is respected before freeing memory containing @napi
6146 static void napi_hash_del(struct napi_struct *napi)
6148 spin_lock(&napi_hash_lock);
6150 hlist_del_init_rcu(&napi->napi_hash_node);
6152 spin_unlock(&napi_hash_lock);
6155 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6157 struct napi_struct *napi;
6159 napi = container_of(timer, struct napi_struct, timer);
6161 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6162 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6164 if (!napi_disable_pending(napi) &&
6165 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6166 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6167 __napi_schedule_irqoff(napi);
6170 return HRTIMER_NORESTART;
6173 static void init_gro_hash(struct napi_struct *napi)
6177 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6178 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6179 napi->gro_hash[i].count = 0;
6181 napi->gro_bitmask = 0;
6184 int dev_set_threaded(struct net_device *dev, bool threaded)
6186 struct napi_struct *napi;
6189 if (dev->threaded == threaded)
6193 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6194 if (!napi->thread) {
6195 err = napi_kthread_create(napi);
6204 dev->threaded = threaded;
6206 /* Make sure kthread is created before THREADED bit
6209 smp_mb__before_atomic();
6211 /* Setting/unsetting threaded mode on a napi might not immediately
6212 * take effect, if the current napi instance is actively being
6213 * polled. In this case, the switch between threaded mode and
6214 * softirq mode will happen in the next round of napi_schedule().
6215 * This should not cause hiccups/stalls to the live traffic.
6217 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6219 set_bit(NAPI_STATE_THREADED, &napi->state);
6221 clear_bit(NAPI_STATE_THREADED, &napi->state);
6226 EXPORT_SYMBOL(dev_set_threaded);
6228 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6229 int (*poll)(struct napi_struct *, int), int weight)
6231 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6234 INIT_LIST_HEAD(&napi->poll_list);
6235 INIT_HLIST_NODE(&napi->napi_hash_node);
6236 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6237 napi->timer.function = napi_watchdog;
6238 init_gro_hash(napi);
6240 INIT_LIST_HEAD(&napi->rx_list);
6243 if (weight > NAPI_POLL_WEIGHT)
6244 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6246 napi->weight = weight;
6248 #ifdef CONFIG_NETPOLL
6249 napi->poll_owner = -1;
6251 set_bit(NAPI_STATE_SCHED, &napi->state);
6252 set_bit(NAPI_STATE_NPSVC, &napi->state);
6253 list_add_rcu(&napi->dev_list, &dev->napi_list);
6254 napi_hash_add(napi);
6255 /* Create kthread for this napi if dev->threaded is set.
6256 * Clear dev->threaded if kthread creation failed so that
6257 * threaded mode will not be enabled in napi_enable().
6259 if (dev->threaded && napi_kthread_create(napi))
6262 EXPORT_SYMBOL(netif_napi_add);
6264 void napi_disable(struct napi_struct *n)
6266 unsigned long val, new;
6269 set_bit(NAPI_STATE_DISABLE, &n->state);
6272 val = READ_ONCE(n->state);
6273 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6274 usleep_range(20, 200);
6278 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6279 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6281 if (cmpxchg(&n->state, val, new) == val)
6285 hrtimer_cancel(&n->timer);
6287 clear_bit(NAPI_STATE_DISABLE, &n->state);
6289 EXPORT_SYMBOL(napi_disable);
6292 * napi_enable - enable NAPI scheduling
6295 * Resume NAPI from being scheduled on this context.
6296 * Must be paired with napi_disable.
6298 void napi_enable(struct napi_struct *n)
6300 unsigned long val, new;
6303 val = READ_ONCE(n->state);
6304 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6306 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6307 if (n->dev->threaded && n->thread)
6308 new |= NAPIF_STATE_THREADED;
6309 } while (cmpxchg(&n->state, val, new) != val);
6311 EXPORT_SYMBOL(napi_enable);
6313 static void flush_gro_hash(struct napi_struct *napi)
6317 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6318 struct sk_buff *skb, *n;
6320 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6322 napi->gro_hash[i].count = 0;
6326 /* Must be called in process context */
6327 void __netif_napi_del(struct napi_struct *napi)
6329 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6332 napi_hash_del(napi);
6333 list_del_rcu(&napi->dev_list);
6334 napi_free_frags(napi);
6336 flush_gro_hash(napi);
6337 napi->gro_bitmask = 0;
6340 kthread_stop(napi->thread);
6341 napi->thread = NULL;
6344 EXPORT_SYMBOL(__netif_napi_del);
6346 static int __napi_poll(struct napi_struct *n, bool *repoll)
6352 /* This NAPI_STATE_SCHED test is for avoiding a race
6353 * with netpoll's poll_napi(). Only the entity which
6354 * obtains the lock and sees NAPI_STATE_SCHED set will
6355 * actually make the ->poll() call. Therefore we avoid
6356 * accidentally calling ->poll() when NAPI is not scheduled.
6359 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6360 work = n->poll(n, weight);
6361 trace_napi_poll(n, work, weight);
6364 if (unlikely(work > weight))
6365 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6366 n->poll, work, weight);
6368 if (likely(work < weight))
6371 /* Drivers must not modify the NAPI state if they
6372 * consume the entire weight. In such cases this code
6373 * still "owns" the NAPI instance and therefore can
6374 * move the instance around on the list at-will.
6376 if (unlikely(napi_disable_pending(n))) {
6381 /* The NAPI context has more processing work, but busy-polling
6382 * is preferred. Exit early.
6384 if (napi_prefer_busy_poll(n)) {
6385 if (napi_complete_done(n, work)) {
6386 /* If timeout is not set, we need to make sure
6387 * that the NAPI is re-scheduled.
6394 if (n->gro_bitmask) {
6395 /* flush too old packets
6396 * If HZ < 1000, flush all packets.
6398 napi_gro_flush(n, HZ >= 1000);
6403 /* Some drivers may have called napi_schedule
6404 * prior to exhausting their budget.
6406 if (unlikely(!list_empty(&n->poll_list))) {
6407 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6408 n->dev ? n->dev->name : "backlog");
6417 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6419 bool do_repoll = false;
6423 list_del_init(&n->poll_list);
6425 have = netpoll_poll_lock(n);
6427 work = __napi_poll(n, &do_repoll);
6430 list_add_tail(&n->poll_list, repoll);
6432 netpoll_poll_unlock(have);
6437 static int napi_thread_wait(struct napi_struct *napi)
6441 set_current_state(TASK_INTERRUPTIBLE);
6443 while (!kthread_should_stop()) {
6444 /* Testing SCHED_THREADED bit here to make sure the current
6445 * kthread owns this napi and could poll on this napi.
6446 * Testing SCHED bit is not enough because SCHED bit might be
6447 * set by some other busy poll thread or by napi_disable().
6449 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6450 WARN_ON(!list_empty(&napi->poll_list));
6451 __set_current_state(TASK_RUNNING);
6456 /* woken being true indicates this thread owns this napi. */
6458 set_current_state(TASK_INTERRUPTIBLE);
6460 __set_current_state(TASK_RUNNING);
6465 static int napi_threaded_poll(void *data)
6467 struct napi_struct *napi = data;
6470 while (!napi_thread_wait(napi)) {
6472 bool repoll = false;
6476 have = netpoll_poll_lock(napi);
6477 __napi_poll(napi, &repoll);
6478 netpoll_poll_unlock(have);
6491 static __latent_entropy void net_rx_action(struct softirq_action *h)
6493 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6494 unsigned long time_limit = jiffies +
6495 usecs_to_jiffies(netdev_budget_usecs);
6496 int budget = netdev_budget;
6500 local_irq_disable();
6501 list_splice_init(&sd->poll_list, &list);
6505 struct napi_struct *n;
6507 if (list_empty(&list)) {
6508 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6513 n = list_first_entry(&list, struct napi_struct, poll_list);
6514 budget -= napi_poll(n, &repoll);
6516 /* If softirq window is exhausted then punt.
6517 * Allow this to run for 2 jiffies since which will allow
6518 * an average latency of 1.5/HZ.
6520 if (unlikely(budget <= 0 ||
6521 time_after_eq(jiffies, time_limit))) {
6527 local_irq_disable();
6529 list_splice_tail_init(&sd->poll_list, &list);
6530 list_splice_tail(&repoll, &list);
6531 list_splice(&list, &sd->poll_list);
6532 if (!list_empty(&sd->poll_list))
6533 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6535 net_rps_action_and_irq_enable(sd);
6538 struct netdev_adjacent {
6539 struct net_device *dev;
6540 netdevice_tracker dev_tracker;
6542 /* upper master flag, there can only be one master device per list */
6545 /* lookup ignore flag */
6548 /* counter for the number of times this device was added to us */
6551 /* private field for the users */
6554 struct list_head list;
6555 struct rcu_head rcu;
6558 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6559 struct list_head *adj_list)
6561 struct netdev_adjacent *adj;
6563 list_for_each_entry(adj, adj_list, list) {
6564 if (adj->dev == adj_dev)
6570 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6571 struct netdev_nested_priv *priv)
6573 struct net_device *dev = (struct net_device *)priv->data;
6575 return upper_dev == dev;
6579 * netdev_has_upper_dev - Check if device is linked to an upper device
6581 * @upper_dev: upper device to check
6583 * Find out if a device is linked to specified upper device and return true
6584 * in case it is. Note that this checks only immediate upper device,
6585 * not through a complete stack of devices. The caller must hold the RTNL lock.
6587 bool netdev_has_upper_dev(struct net_device *dev,
6588 struct net_device *upper_dev)
6590 struct netdev_nested_priv priv = {
6591 .data = (void *)upper_dev,
6596 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6599 EXPORT_SYMBOL(netdev_has_upper_dev);
6602 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6604 * @upper_dev: upper device to check
6606 * Find out if a device is linked to specified upper device and return true
6607 * in case it is. Note that this checks the entire upper device chain.
6608 * The caller must hold rcu lock.
6611 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6612 struct net_device *upper_dev)
6614 struct netdev_nested_priv priv = {
6615 .data = (void *)upper_dev,
6618 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6621 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6624 * netdev_has_any_upper_dev - Check if device is linked to some device
6627 * Find out if a device is linked to an upper device and return true in case
6628 * it is. The caller must hold the RTNL lock.
6630 bool netdev_has_any_upper_dev(struct net_device *dev)
6634 return !list_empty(&dev->adj_list.upper);
6636 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6639 * netdev_master_upper_dev_get - Get master upper device
6642 * Find a master upper device and return pointer to it or NULL in case
6643 * it's not there. The caller must hold the RTNL lock.
6645 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6647 struct netdev_adjacent *upper;
6651 if (list_empty(&dev->adj_list.upper))
6654 upper = list_first_entry(&dev->adj_list.upper,
6655 struct netdev_adjacent, list);
6656 if (likely(upper->master))
6660 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6662 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6664 struct netdev_adjacent *upper;
6668 if (list_empty(&dev->adj_list.upper))
6671 upper = list_first_entry(&dev->adj_list.upper,
6672 struct netdev_adjacent, list);
6673 if (likely(upper->master) && !upper->ignore)
6679 * netdev_has_any_lower_dev - Check if device is linked to some device
6682 * Find out if a device is linked to a lower device and return true in case
6683 * it is. The caller must hold the RTNL lock.
6685 static bool netdev_has_any_lower_dev(struct net_device *dev)
6689 return !list_empty(&dev->adj_list.lower);
6692 void *netdev_adjacent_get_private(struct list_head *adj_list)
6694 struct netdev_adjacent *adj;
6696 adj = list_entry(adj_list, struct netdev_adjacent, list);
6698 return adj->private;
6700 EXPORT_SYMBOL(netdev_adjacent_get_private);
6703 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6705 * @iter: list_head ** of the current position
6707 * Gets the next device from the dev's upper list, starting from iter
6708 * position. The caller must hold RCU read lock.
6710 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6711 struct list_head **iter)
6713 struct netdev_adjacent *upper;
6715 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6717 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6719 if (&upper->list == &dev->adj_list.upper)
6722 *iter = &upper->list;
6726 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6728 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6729 struct list_head **iter,
6732 struct netdev_adjacent *upper;
6734 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6736 if (&upper->list == &dev->adj_list.upper)
6739 *iter = &upper->list;
6740 *ignore = upper->ignore;
6745 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6746 struct list_head **iter)
6748 struct netdev_adjacent *upper;
6750 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6752 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6754 if (&upper->list == &dev->adj_list.upper)
6757 *iter = &upper->list;
6762 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6763 int (*fn)(struct net_device *dev,
6764 struct netdev_nested_priv *priv),
6765 struct netdev_nested_priv *priv)
6767 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6768 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6773 iter = &dev->adj_list.upper;
6777 ret = fn(now, priv);
6784 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6791 niter = &udev->adj_list.upper;
6792 dev_stack[cur] = now;
6793 iter_stack[cur++] = iter;
6800 next = dev_stack[--cur];
6801 niter = iter_stack[cur];
6811 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6812 int (*fn)(struct net_device *dev,
6813 struct netdev_nested_priv *priv),
6814 struct netdev_nested_priv *priv)
6816 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6817 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6821 iter = &dev->adj_list.upper;
6825 ret = fn(now, priv);
6832 udev = netdev_next_upper_dev_rcu(now, &iter);
6837 niter = &udev->adj_list.upper;
6838 dev_stack[cur] = now;
6839 iter_stack[cur++] = iter;
6846 next = dev_stack[--cur];
6847 niter = iter_stack[cur];
6856 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6858 static bool __netdev_has_upper_dev(struct net_device *dev,
6859 struct net_device *upper_dev)
6861 struct netdev_nested_priv priv = {
6863 .data = (void *)upper_dev,
6868 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6873 * netdev_lower_get_next_private - Get the next ->private from the
6874 * lower neighbour list
6876 * @iter: list_head ** of the current position
6878 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6879 * list, starting from iter position. The caller must hold either hold the
6880 * RTNL lock or its own locking that guarantees that the neighbour lower
6881 * list will remain unchanged.
6883 void *netdev_lower_get_next_private(struct net_device *dev,
6884 struct list_head **iter)
6886 struct netdev_adjacent *lower;
6888 lower = list_entry(*iter, struct netdev_adjacent, list);
6890 if (&lower->list == &dev->adj_list.lower)
6893 *iter = lower->list.next;
6895 return lower->private;
6897 EXPORT_SYMBOL(netdev_lower_get_next_private);
6900 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6901 * lower neighbour list, RCU
6904 * @iter: list_head ** of the current position
6906 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6907 * list, starting from iter position. The caller must hold RCU read lock.
6909 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6910 struct list_head **iter)
6912 struct netdev_adjacent *lower;
6914 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
6916 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6918 if (&lower->list == &dev->adj_list.lower)
6921 *iter = &lower->list;
6923 return lower->private;
6925 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6928 * netdev_lower_get_next - Get the next device from the lower neighbour
6931 * @iter: list_head ** of the current position
6933 * Gets the next netdev_adjacent from the dev's lower neighbour
6934 * list, starting from iter position. The caller must hold RTNL lock or
6935 * its own locking that guarantees that the neighbour lower
6936 * list will remain unchanged.
6938 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6940 struct netdev_adjacent *lower;
6942 lower = list_entry(*iter, struct netdev_adjacent, list);
6944 if (&lower->list == &dev->adj_list.lower)
6947 *iter = lower->list.next;
6951 EXPORT_SYMBOL(netdev_lower_get_next);
6953 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6954 struct list_head **iter)
6956 struct netdev_adjacent *lower;
6958 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6960 if (&lower->list == &dev->adj_list.lower)
6963 *iter = &lower->list;
6968 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
6969 struct list_head **iter,
6972 struct netdev_adjacent *lower;
6974 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6976 if (&lower->list == &dev->adj_list.lower)
6979 *iter = &lower->list;
6980 *ignore = lower->ignore;
6985 int netdev_walk_all_lower_dev(struct net_device *dev,
6986 int (*fn)(struct net_device *dev,
6987 struct netdev_nested_priv *priv),
6988 struct netdev_nested_priv *priv)
6990 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6991 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6995 iter = &dev->adj_list.lower;
6999 ret = fn(now, priv);
7006 ldev = netdev_next_lower_dev(now, &iter);
7011 niter = &ldev->adj_list.lower;
7012 dev_stack[cur] = now;
7013 iter_stack[cur++] = iter;
7020 next = dev_stack[--cur];
7021 niter = iter_stack[cur];
7030 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7032 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7033 int (*fn)(struct net_device *dev,
7034 struct netdev_nested_priv *priv),
7035 struct netdev_nested_priv *priv)
7037 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7038 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7043 iter = &dev->adj_list.lower;
7047 ret = fn(now, priv);
7054 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7061 niter = &ldev->adj_list.lower;
7062 dev_stack[cur] = now;
7063 iter_stack[cur++] = iter;
7070 next = dev_stack[--cur];
7071 niter = iter_stack[cur];
7081 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7082 struct list_head **iter)
7084 struct netdev_adjacent *lower;
7086 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7087 if (&lower->list == &dev->adj_list.lower)
7090 *iter = &lower->list;
7094 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7096 static u8 __netdev_upper_depth(struct net_device *dev)
7098 struct net_device *udev;
7099 struct list_head *iter;
7103 for (iter = &dev->adj_list.upper,
7104 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7106 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7109 if (max_depth < udev->upper_level)
7110 max_depth = udev->upper_level;
7116 static u8 __netdev_lower_depth(struct net_device *dev)
7118 struct net_device *ldev;
7119 struct list_head *iter;
7123 for (iter = &dev->adj_list.lower,
7124 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7126 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7129 if (max_depth < ldev->lower_level)
7130 max_depth = ldev->lower_level;
7136 static int __netdev_update_upper_level(struct net_device *dev,
7137 struct netdev_nested_priv *__unused)
7139 dev->upper_level = __netdev_upper_depth(dev) + 1;
7143 static int __netdev_update_lower_level(struct net_device *dev,
7144 struct netdev_nested_priv *priv)
7146 dev->lower_level = __netdev_lower_depth(dev) + 1;
7148 #ifdef CONFIG_LOCKDEP
7152 if (priv->flags & NESTED_SYNC_IMM)
7153 dev->nested_level = dev->lower_level - 1;
7154 if (priv->flags & NESTED_SYNC_TODO)
7155 net_unlink_todo(dev);
7160 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7161 int (*fn)(struct net_device *dev,
7162 struct netdev_nested_priv *priv),
7163 struct netdev_nested_priv *priv)
7165 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7166 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7170 iter = &dev->adj_list.lower;
7174 ret = fn(now, priv);
7181 ldev = netdev_next_lower_dev_rcu(now, &iter);
7186 niter = &ldev->adj_list.lower;
7187 dev_stack[cur] = now;
7188 iter_stack[cur++] = iter;
7195 next = dev_stack[--cur];
7196 niter = iter_stack[cur];
7205 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7208 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7209 * lower neighbour list, RCU
7213 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7214 * list. The caller must hold RCU read lock.
7216 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7218 struct netdev_adjacent *lower;
7220 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7221 struct netdev_adjacent, list);
7223 return lower->private;
7226 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7229 * netdev_master_upper_dev_get_rcu - Get master upper device
7232 * Find a master upper device and return pointer to it or NULL in case
7233 * it's not there. The caller must hold the RCU read lock.
7235 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7237 struct netdev_adjacent *upper;
7239 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7240 struct netdev_adjacent, list);
7241 if (upper && likely(upper->master))
7245 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7247 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7248 struct net_device *adj_dev,
7249 struct list_head *dev_list)
7251 char linkname[IFNAMSIZ+7];
7253 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7254 "upper_%s" : "lower_%s", adj_dev->name);
7255 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7258 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7260 struct list_head *dev_list)
7262 char linkname[IFNAMSIZ+7];
7264 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7265 "upper_%s" : "lower_%s", name);
7266 sysfs_remove_link(&(dev->dev.kobj), linkname);
7269 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7270 struct net_device *adj_dev,
7271 struct list_head *dev_list)
7273 return (dev_list == &dev->adj_list.upper ||
7274 dev_list == &dev->adj_list.lower) &&
7275 net_eq(dev_net(dev), dev_net(adj_dev));
7278 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7279 struct net_device *adj_dev,
7280 struct list_head *dev_list,
7281 void *private, bool master)
7283 struct netdev_adjacent *adj;
7286 adj = __netdev_find_adj(adj_dev, dev_list);
7290 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7291 dev->name, adj_dev->name, adj->ref_nr);
7296 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7301 adj->master = master;
7303 adj->private = private;
7304 adj->ignore = false;
7305 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7307 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7308 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7310 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7311 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7316 /* Ensure that master link is always the first item in list. */
7318 ret = sysfs_create_link(&(dev->dev.kobj),
7319 &(adj_dev->dev.kobj), "master");
7321 goto remove_symlinks;
7323 list_add_rcu(&adj->list, dev_list);
7325 list_add_tail_rcu(&adj->list, dev_list);
7331 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7332 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7334 dev_put_track(adj_dev, &adj->dev_tracker);
7340 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7341 struct net_device *adj_dev,
7343 struct list_head *dev_list)
7345 struct netdev_adjacent *adj;
7347 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7348 dev->name, adj_dev->name, ref_nr);
7350 adj = __netdev_find_adj(adj_dev, dev_list);
7353 pr_err("Adjacency does not exist for device %s from %s\n",
7354 dev->name, adj_dev->name);
7359 if (adj->ref_nr > ref_nr) {
7360 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7361 dev->name, adj_dev->name, ref_nr,
7362 adj->ref_nr - ref_nr);
7363 adj->ref_nr -= ref_nr;
7368 sysfs_remove_link(&(dev->dev.kobj), "master");
7370 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7371 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7373 list_del_rcu(&adj->list);
7374 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7375 adj_dev->name, dev->name, adj_dev->name);
7376 dev_put_track(adj_dev, &adj->dev_tracker);
7377 kfree_rcu(adj, rcu);
7380 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7381 struct net_device *upper_dev,
7382 struct list_head *up_list,
7383 struct list_head *down_list,
7384 void *private, bool master)
7388 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7393 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7396 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7403 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7404 struct net_device *upper_dev,
7406 struct list_head *up_list,
7407 struct list_head *down_list)
7409 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7410 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7413 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7414 struct net_device *upper_dev,
7415 void *private, bool master)
7417 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7418 &dev->adj_list.upper,
7419 &upper_dev->adj_list.lower,
7423 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7424 struct net_device *upper_dev)
7426 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7427 &dev->adj_list.upper,
7428 &upper_dev->adj_list.lower);
7431 static int __netdev_upper_dev_link(struct net_device *dev,
7432 struct net_device *upper_dev, bool master,
7433 void *upper_priv, void *upper_info,
7434 struct netdev_nested_priv *priv,
7435 struct netlink_ext_ack *extack)
7437 struct netdev_notifier_changeupper_info changeupper_info = {
7442 .upper_dev = upper_dev,
7445 .upper_info = upper_info,
7447 struct net_device *master_dev;
7452 if (dev == upper_dev)
7455 /* To prevent loops, check if dev is not upper device to upper_dev. */
7456 if (__netdev_has_upper_dev(upper_dev, dev))
7459 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7463 if (__netdev_has_upper_dev(dev, upper_dev))
7466 master_dev = __netdev_master_upper_dev_get(dev);
7468 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7471 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7472 &changeupper_info.info);
7473 ret = notifier_to_errno(ret);
7477 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7482 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7483 &changeupper_info.info);
7484 ret = notifier_to_errno(ret);
7488 __netdev_update_upper_level(dev, NULL);
7489 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7491 __netdev_update_lower_level(upper_dev, priv);
7492 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7498 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7504 * netdev_upper_dev_link - Add a link to the upper device
7506 * @upper_dev: new upper device
7507 * @extack: netlink extended ack
7509 * Adds a link to device which is upper to this one. The caller must hold
7510 * the RTNL lock. On a failure a negative errno code is returned.
7511 * On success the reference counts are adjusted and the function
7514 int netdev_upper_dev_link(struct net_device *dev,
7515 struct net_device *upper_dev,
7516 struct netlink_ext_ack *extack)
7518 struct netdev_nested_priv priv = {
7519 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7523 return __netdev_upper_dev_link(dev, upper_dev, false,
7524 NULL, NULL, &priv, extack);
7526 EXPORT_SYMBOL(netdev_upper_dev_link);
7529 * netdev_master_upper_dev_link - Add a master link to the upper device
7531 * @upper_dev: new upper device
7532 * @upper_priv: upper device private
7533 * @upper_info: upper info to be passed down via notifier
7534 * @extack: netlink extended ack
7536 * Adds a link to device which is upper to this one. In this case, only
7537 * one master upper device can be linked, although other non-master devices
7538 * might be linked as well. The caller must hold the RTNL lock.
7539 * On a failure a negative errno code is returned. On success the reference
7540 * counts are adjusted and the function returns zero.
7542 int netdev_master_upper_dev_link(struct net_device *dev,
7543 struct net_device *upper_dev,
7544 void *upper_priv, void *upper_info,
7545 struct netlink_ext_ack *extack)
7547 struct netdev_nested_priv priv = {
7548 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7552 return __netdev_upper_dev_link(dev, upper_dev, true,
7553 upper_priv, upper_info, &priv, extack);
7555 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7557 static void __netdev_upper_dev_unlink(struct net_device *dev,
7558 struct net_device *upper_dev,
7559 struct netdev_nested_priv *priv)
7561 struct netdev_notifier_changeupper_info changeupper_info = {
7565 .upper_dev = upper_dev,
7571 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7573 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7574 &changeupper_info.info);
7576 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7578 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7579 &changeupper_info.info);
7581 __netdev_update_upper_level(dev, NULL);
7582 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7584 __netdev_update_lower_level(upper_dev, priv);
7585 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7590 * netdev_upper_dev_unlink - Removes a link to upper device
7592 * @upper_dev: new upper device
7594 * Removes a link to device which is upper to this one. The caller must hold
7597 void netdev_upper_dev_unlink(struct net_device *dev,
7598 struct net_device *upper_dev)
7600 struct netdev_nested_priv priv = {
7601 .flags = NESTED_SYNC_TODO,
7605 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7607 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7609 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7610 struct net_device *lower_dev,
7613 struct netdev_adjacent *adj;
7615 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7619 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7624 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7625 struct net_device *lower_dev)
7627 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7630 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7631 struct net_device *lower_dev)
7633 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7636 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7637 struct net_device *new_dev,
7638 struct net_device *dev,
7639 struct netlink_ext_ack *extack)
7641 struct netdev_nested_priv priv = {
7650 if (old_dev && new_dev != old_dev)
7651 netdev_adjacent_dev_disable(dev, old_dev);
7652 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7655 if (old_dev && new_dev != old_dev)
7656 netdev_adjacent_dev_enable(dev, old_dev);
7662 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7664 void netdev_adjacent_change_commit(struct net_device *old_dev,
7665 struct net_device *new_dev,
7666 struct net_device *dev)
7668 struct netdev_nested_priv priv = {
7669 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7673 if (!new_dev || !old_dev)
7676 if (new_dev == old_dev)
7679 netdev_adjacent_dev_enable(dev, old_dev);
7680 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7682 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7684 void netdev_adjacent_change_abort(struct net_device *old_dev,
7685 struct net_device *new_dev,
7686 struct net_device *dev)
7688 struct netdev_nested_priv priv = {
7696 if (old_dev && new_dev != old_dev)
7697 netdev_adjacent_dev_enable(dev, old_dev);
7699 __netdev_upper_dev_unlink(new_dev, dev, &priv);
7701 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7704 * netdev_bonding_info_change - Dispatch event about slave change
7706 * @bonding_info: info to dispatch
7708 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7709 * The caller must hold the RTNL lock.
7711 void netdev_bonding_info_change(struct net_device *dev,
7712 struct netdev_bonding_info *bonding_info)
7714 struct netdev_notifier_bonding_info info = {
7718 memcpy(&info.bonding_info, bonding_info,
7719 sizeof(struct netdev_bonding_info));
7720 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7723 EXPORT_SYMBOL(netdev_bonding_info_change);
7726 * netdev_get_xmit_slave - Get the xmit slave of master device
7729 * @all_slaves: assume all the slaves are active
7731 * The reference counters are not incremented so the caller must be
7732 * careful with locks. The caller must hold RCU lock.
7733 * %NULL is returned if no slave is found.
7736 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
7737 struct sk_buff *skb,
7740 const struct net_device_ops *ops = dev->netdev_ops;
7742 if (!ops->ndo_get_xmit_slave)
7744 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
7746 EXPORT_SYMBOL(netdev_get_xmit_slave);
7748 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
7751 const struct net_device_ops *ops = dev->netdev_ops;
7753 if (!ops->ndo_sk_get_lower_dev)
7755 return ops->ndo_sk_get_lower_dev(dev, sk);
7759 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
7763 * %NULL is returned if no lower device is found.
7766 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
7769 struct net_device *lower;
7771 lower = netdev_sk_get_lower_dev(dev, sk);
7774 lower = netdev_sk_get_lower_dev(dev, sk);
7779 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
7781 static void netdev_adjacent_add_links(struct net_device *dev)
7783 struct netdev_adjacent *iter;
7785 struct net *net = dev_net(dev);
7787 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7788 if (!net_eq(net, dev_net(iter->dev)))
7790 netdev_adjacent_sysfs_add(iter->dev, dev,
7791 &iter->dev->adj_list.lower);
7792 netdev_adjacent_sysfs_add(dev, iter->dev,
7793 &dev->adj_list.upper);
7796 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7797 if (!net_eq(net, dev_net(iter->dev)))
7799 netdev_adjacent_sysfs_add(iter->dev, dev,
7800 &iter->dev->adj_list.upper);
7801 netdev_adjacent_sysfs_add(dev, iter->dev,
7802 &dev->adj_list.lower);
7806 static void netdev_adjacent_del_links(struct net_device *dev)
7808 struct netdev_adjacent *iter;
7810 struct net *net = dev_net(dev);
7812 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7813 if (!net_eq(net, dev_net(iter->dev)))
7815 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7816 &iter->dev->adj_list.lower);
7817 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7818 &dev->adj_list.upper);
7821 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7822 if (!net_eq(net, dev_net(iter->dev)))
7824 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7825 &iter->dev->adj_list.upper);
7826 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7827 &dev->adj_list.lower);
7831 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7833 struct netdev_adjacent *iter;
7835 struct net *net = dev_net(dev);
7837 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7838 if (!net_eq(net, dev_net(iter->dev)))
7840 netdev_adjacent_sysfs_del(iter->dev, oldname,
7841 &iter->dev->adj_list.lower);
7842 netdev_adjacent_sysfs_add(iter->dev, dev,
7843 &iter->dev->adj_list.lower);
7846 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7847 if (!net_eq(net, dev_net(iter->dev)))
7849 netdev_adjacent_sysfs_del(iter->dev, oldname,
7850 &iter->dev->adj_list.upper);
7851 netdev_adjacent_sysfs_add(iter->dev, dev,
7852 &iter->dev->adj_list.upper);
7856 void *netdev_lower_dev_get_private(struct net_device *dev,
7857 struct net_device *lower_dev)
7859 struct netdev_adjacent *lower;
7863 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7867 return lower->private;
7869 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7873 * netdev_lower_state_changed - Dispatch event about lower device state change
7874 * @lower_dev: device
7875 * @lower_state_info: state to dispatch
7877 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7878 * The caller must hold the RTNL lock.
7880 void netdev_lower_state_changed(struct net_device *lower_dev,
7881 void *lower_state_info)
7883 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7884 .info.dev = lower_dev,
7888 changelowerstate_info.lower_state_info = lower_state_info;
7889 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7890 &changelowerstate_info.info);
7892 EXPORT_SYMBOL(netdev_lower_state_changed);
7894 static void dev_change_rx_flags(struct net_device *dev, int flags)
7896 const struct net_device_ops *ops = dev->netdev_ops;
7898 if (ops->ndo_change_rx_flags)
7899 ops->ndo_change_rx_flags(dev, flags);
7902 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7904 unsigned int old_flags = dev->flags;
7910 dev->flags |= IFF_PROMISC;
7911 dev->promiscuity += inc;
7912 if (dev->promiscuity == 0) {
7915 * If inc causes overflow, untouch promisc and return error.
7918 dev->flags &= ~IFF_PROMISC;
7920 dev->promiscuity -= inc;
7921 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
7925 if (dev->flags != old_flags) {
7926 pr_info("device %s %s promiscuous mode\n",
7928 dev->flags & IFF_PROMISC ? "entered" : "left");
7929 if (audit_enabled) {
7930 current_uid_gid(&uid, &gid);
7931 audit_log(audit_context(), GFP_ATOMIC,
7932 AUDIT_ANOM_PROMISCUOUS,
7933 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7934 dev->name, (dev->flags & IFF_PROMISC),
7935 (old_flags & IFF_PROMISC),
7936 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7937 from_kuid(&init_user_ns, uid),
7938 from_kgid(&init_user_ns, gid),
7939 audit_get_sessionid(current));
7942 dev_change_rx_flags(dev, IFF_PROMISC);
7945 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7950 * dev_set_promiscuity - update promiscuity count on a device
7954 * Add or remove promiscuity from a device. While the count in the device
7955 * remains above zero the interface remains promiscuous. Once it hits zero
7956 * the device reverts back to normal filtering operation. A negative inc
7957 * value is used to drop promiscuity on the device.
7958 * Return 0 if successful or a negative errno code on error.
7960 int dev_set_promiscuity(struct net_device *dev, int inc)
7962 unsigned int old_flags = dev->flags;
7965 err = __dev_set_promiscuity(dev, inc, true);
7968 if (dev->flags != old_flags)
7969 dev_set_rx_mode(dev);
7972 EXPORT_SYMBOL(dev_set_promiscuity);
7974 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7976 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7980 dev->flags |= IFF_ALLMULTI;
7981 dev->allmulti += inc;
7982 if (dev->allmulti == 0) {
7985 * If inc causes overflow, untouch allmulti and return error.
7988 dev->flags &= ~IFF_ALLMULTI;
7990 dev->allmulti -= inc;
7991 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
7995 if (dev->flags ^ old_flags) {
7996 dev_change_rx_flags(dev, IFF_ALLMULTI);
7997 dev_set_rx_mode(dev);
7999 __dev_notify_flags(dev, old_flags,
8000 dev->gflags ^ old_gflags);
8006 * dev_set_allmulti - update allmulti count on a device
8010 * Add or remove reception of all multicast frames to a device. While the
8011 * count in the device remains above zero the interface remains listening
8012 * to all interfaces. Once it hits zero the device reverts back to normal
8013 * filtering operation. A negative @inc value is used to drop the counter
8014 * when releasing a resource needing all multicasts.
8015 * Return 0 if successful or a negative errno code on error.
8018 int dev_set_allmulti(struct net_device *dev, int inc)
8020 return __dev_set_allmulti(dev, inc, true);
8022 EXPORT_SYMBOL(dev_set_allmulti);
8025 * Upload unicast and multicast address lists to device and
8026 * configure RX filtering. When the device doesn't support unicast
8027 * filtering it is put in promiscuous mode while unicast addresses
8030 void __dev_set_rx_mode(struct net_device *dev)
8032 const struct net_device_ops *ops = dev->netdev_ops;
8034 /* dev_open will call this function so the list will stay sane. */
8035 if (!(dev->flags&IFF_UP))
8038 if (!netif_device_present(dev))
8041 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8042 /* Unicast addresses changes may only happen under the rtnl,
8043 * therefore calling __dev_set_promiscuity here is safe.
8045 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8046 __dev_set_promiscuity(dev, 1, false);
8047 dev->uc_promisc = true;
8048 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8049 __dev_set_promiscuity(dev, -1, false);
8050 dev->uc_promisc = false;
8054 if (ops->ndo_set_rx_mode)
8055 ops->ndo_set_rx_mode(dev);
8058 void dev_set_rx_mode(struct net_device *dev)
8060 netif_addr_lock_bh(dev);
8061 __dev_set_rx_mode(dev);
8062 netif_addr_unlock_bh(dev);
8066 * dev_get_flags - get flags reported to userspace
8069 * Get the combination of flag bits exported through APIs to userspace.
8071 unsigned int dev_get_flags(const struct net_device *dev)
8075 flags = (dev->flags & ~(IFF_PROMISC |
8080 (dev->gflags & (IFF_PROMISC |
8083 if (netif_running(dev)) {
8084 if (netif_oper_up(dev))
8085 flags |= IFF_RUNNING;
8086 if (netif_carrier_ok(dev))
8087 flags |= IFF_LOWER_UP;
8088 if (netif_dormant(dev))
8089 flags |= IFF_DORMANT;
8094 EXPORT_SYMBOL(dev_get_flags);
8096 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8097 struct netlink_ext_ack *extack)
8099 unsigned int old_flags = dev->flags;
8105 * Set the flags on our device.
8108 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8109 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8111 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8115 * Load in the correct multicast list now the flags have changed.
8118 if ((old_flags ^ flags) & IFF_MULTICAST)
8119 dev_change_rx_flags(dev, IFF_MULTICAST);
8121 dev_set_rx_mode(dev);
8124 * Have we downed the interface. We handle IFF_UP ourselves
8125 * according to user attempts to set it, rather than blindly
8130 if ((old_flags ^ flags) & IFF_UP) {
8131 if (old_flags & IFF_UP)
8134 ret = __dev_open(dev, extack);
8137 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8138 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8139 unsigned int old_flags = dev->flags;
8141 dev->gflags ^= IFF_PROMISC;
8143 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8144 if (dev->flags != old_flags)
8145 dev_set_rx_mode(dev);
8148 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8149 * is important. Some (broken) drivers set IFF_PROMISC, when
8150 * IFF_ALLMULTI is requested not asking us and not reporting.
8152 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8153 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8155 dev->gflags ^= IFF_ALLMULTI;
8156 __dev_set_allmulti(dev, inc, false);
8162 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8163 unsigned int gchanges)
8165 unsigned int changes = dev->flags ^ old_flags;
8168 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8170 if (changes & IFF_UP) {
8171 if (dev->flags & IFF_UP)
8172 call_netdevice_notifiers(NETDEV_UP, dev);
8174 call_netdevice_notifiers(NETDEV_DOWN, dev);
8177 if (dev->flags & IFF_UP &&
8178 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8179 struct netdev_notifier_change_info change_info = {
8183 .flags_changed = changes,
8186 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8191 * dev_change_flags - change device settings
8193 * @flags: device state flags
8194 * @extack: netlink extended ack
8196 * Change settings on device based state flags. The flags are
8197 * in the userspace exported format.
8199 int dev_change_flags(struct net_device *dev, unsigned int flags,
8200 struct netlink_ext_ack *extack)
8203 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8205 ret = __dev_change_flags(dev, flags, extack);
8209 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8210 __dev_notify_flags(dev, old_flags, changes);
8213 EXPORT_SYMBOL(dev_change_flags);
8215 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8217 const struct net_device_ops *ops = dev->netdev_ops;
8219 if (ops->ndo_change_mtu)
8220 return ops->ndo_change_mtu(dev, new_mtu);
8222 /* Pairs with all the lockless reads of dev->mtu in the stack */
8223 WRITE_ONCE(dev->mtu, new_mtu);
8226 EXPORT_SYMBOL(__dev_set_mtu);
8228 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8229 struct netlink_ext_ack *extack)
8231 /* MTU must be positive, and in range */
8232 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8233 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8237 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8238 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8245 * dev_set_mtu_ext - Change maximum transfer unit
8247 * @new_mtu: new transfer unit
8248 * @extack: netlink extended ack
8250 * Change the maximum transfer size of the network device.
8252 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8253 struct netlink_ext_ack *extack)
8257 if (new_mtu == dev->mtu)
8260 err = dev_validate_mtu(dev, new_mtu, extack);
8264 if (!netif_device_present(dev))
8267 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8268 err = notifier_to_errno(err);
8272 orig_mtu = dev->mtu;
8273 err = __dev_set_mtu(dev, new_mtu);
8276 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8278 err = notifier_to_errno(err);
8280 /* setting mtu back and notifying everyone again,
8281 * so that they have a chance to revert changes.
8283 __dev_set_mtu(dev, orig_mtu);
8284 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8291 int dev_set_mtu(struct net_device *dev, int new_mtu)
8293 struct netlink_ext_ack extack;
8296 memset(&extack, 0, sizeof(extack));
8297 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8298 if (err && extack._msg)
8299 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8302 EXPORT_SYMBOL(dev_set_mtu);
8305 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8307 * @new_len: new tx queue length
8309 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8311 unsigned int orig_len = dev->tx_queue_len;
8314 if (new_len != (unsigned int)new_len)
8317 if (new_len != orig_len) {
8318 dev->tx_queue_len = new_len;
8319 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8320 res = notifier_to_errno(res);
8323 res = dev_qdisc_change_tx_queue_len(dev);
8331 netdev_err(dev, "refused to change device tx_queue_len\n");
8332 dev->tx_queue_len = orig_len;
8337 * dev_set_group - Change group this device belongs to
8339 * @new_group: group this device should belong to
8341 void dev_set_group(struct net_device *dev, int new_group)
8343 dev->group = new_group;
8345 EXPORT_SYMBOL(dev_set_group);
8348 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8350 * @addr: new address
8351 * @extack: netlink extended ack
8353 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8354 struct netlink_ext_ack *extack)
8356 struct netdev_notifier_pre_changeaddr_info info = {
8358 .info.extack = extack,
8363 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8364 return notifier_to_errno(rc);
8366 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8369 * dev_set_mac_address - Change Media Access Control Address
8372 * @extack: netlink extended ack
8374 * Change the hardware (MAC) address of the device
8376 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8377 struct netlink_ext_ack *extack)
8379 const struct net_device_ops *ops = dev->netdev_ops;
8382 if (!ops->ndo_set_mac_address)
8384 if (sa->sa_family != dev->type)
8386 if (!netif_device_present(dev))
8388 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8391 err = ops->ndo_set_mac_address(dev, sa);
8394 dev->addr_assign_type = NET_ADDR_SET;
8395 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8396 add_device_randomness(dev->dev_addr, dev->addr_len);
8399 EXPORT_SYMBOL(dev_set_mac_address);
8401 static DECLARE_RWSEM(dev_addr_sem);
8403 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8404 struct netlink_ext_ack *extack)
8408 down_write(&dev_addr_sem);
8409 ret = dev_set_mac_address(dev, sa, extack);
8410 up_write(&dev_addr_sem);
8413 EXPORT_SYMBOL(dev_set_mac_address_user);
8415 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8417 size_t size = sizeof(sa->sa_data);
8418 struct net_device *dev;
8421 down_read(&dev_addr_sem);
8424 dev = dev_get_by_name_rcu(net, dev_name);
8430 memset(sa->sa_data, 0, size);
8432 memcpy(sa->sa_data, dev->dev_addr,
8433 min_t(size_t, size, dev->addr_len));
8434 sa->sa_family = dev->type;
8438 up_read(&dev_addr_sem);
8441 EXPORT_SYMBOL(dev_get_mac_address);
8444 * dev_change_carrier - Change device carrier
8446 * @new_carrier: new value
8448 * Change device carrier
8450 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8452 const struct net_device_ops *ops = dev->netdev_ops;
8454 if (!ops->ndo_change_carrier)
8456 if (!netif_device_present(dev))
8458 return ops->ndo_change_carrier(dev, new_carrier);
8460 EXPORT_SYMBOL(dev_change_carrier);
8463 * dev_get_phys_port_id - Get device physical port ID
8467 * Get device physical port ID
8469 int dev_get_phys_port_id(struct net_device *dev,
8470 struct netdev_phys_item_id *ppid)
8472 const struct net_device_ops *ops = dev->netdev_ops;
8474 if (!ops->ndo_get_phys_port_id)
8476 return ops->ndo_get_phys_port_id(dev, ppid);
8478 EXPORT_SYMBOL(dev_get_phys_port_id);
8481 * dev_get_phys_port_name - Get device physical port name
8484 * @len: limit of bytes to copy to name
8486 * Get device physical port name
8488 int dev_get_phys_port_name(struct net_device *dev,
8489 char *name, size_t len)
8491 const struct net_device_ops *ops = dev->netdev_ops;
8494 if (ops->ndo_get_phys_port_name) {
8495 err = ops->ndo_get_phys_port_name(dev, name, len);
8496 if (err != -EOPNOTSUPP)
8499 return devlink_compat_phys_port_name_get(dev, name, len);
8501 EXPORT_SYMBOL(dev_get_phys_port_name);
8504 * dev_get_port_parent_id - Get the device's port parent identifier
8505 * @dev: network device
8506 * @ppid: pointer to a storage for the port's parent identifier
8507 * @recurse: allow/disallow recursion to lower devices
8509 * Get the devices's port parent identifier
8511 int dev_get_port_parent_id(struct net_device *dev,
8512 struct netdev_phys_item_id *ppid,
8515 const struct net_device_ops *ops = dev->netdev_ops;
8516 struct netdev_phys_item_id first = { };
8517 struct net_device *lower_dev;
8518 struct list_head *iter;
8521 if (ops->ndo_get_port_parent_id) {
8522 err = ops->ndo_get_port_parent_id(dev, ppid);
8523 if (err != -EOPNOTSUPP)
8527 err = devlink_compat_switch_id_get(dev, ppid);
8528 if (!recurse || err != -EOPNOTSUPP)
8531 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8532 err = dev_get_port_parent_id(lower_dev, ppid, true);
8537 else if (memcmp(&first, ppid, sizeof(*ppid)))
8543 EXPORT_SYMBOL(dev_get_port_parent_id);
8546 * netdev_port_same_parent_id - Indicate if two network devices have
8547 * the same port parent identifier
8548 * @a: first network device
8549 * @b: second network device
8551 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8553 struct netdev_phys_item_id a_id = { };
8554 struct netdev_phys_item_id b_id = { };
8556 if (dev_get_port_parent_id(a, &a_id, true) ||
8557 dev_get_port_parent_id(b, &b_id, true))
8560 return netdev_phys_item_id_same(&a_id, &b_id);
8562 EXPORT_SYMBOL(netdev_port_same_parent_id);
8565 * dev_change_proto_down - set carrier according to proto_down.
8568 * @proto_down: new value
8570 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8572 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8574 if (!netif_device_present(dev))
8577 netif_carrier_off(dev);
8579 netif_carrier_on(dev);
8580 dev->proto_down = proto_down;
8583 EXPORT_SYMBOL(dev_change_proto_down);
8586 * dev_change_proto_down_reason - proto down reason
8589 * @mask: proto down mask
8590 * @value: proto down value
8592 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8598 dev->proto_down_reason = value;
8600 for_each_set_bit(b, &mask, 32) {
8601 if (value & (1 << b))
8602 dev->proto_down_reason |= BIT(b);
8604 dev->proto_down_reason &= ~BIT(b);
8608 EXPORT_SYMBOL(dev_change_proto_down_reason);
8610 struct bpf_xdp_link {
8611 struct bpf_link link;
8612 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
8616 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
8618 if (flags & XDP_FLAGS_HW_MODE)
8620 if (flags & XDP_FLAGS_DRV_MODE)
8621 return XDP_MODE_DRV;
8622 if (flags & XDP_FLAGS_SKB_MODE)
8623 return XDP_MODE_SKB;
8624 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
8627 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
8631 return generic_xdp_install;
8634 return dev->netdev_ops->ndo_bpf;
8640 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
8641 enum bpf_xdp_mode mode)
8643 return dev->xdp_state[mode].link;
8646 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
8647 enum bpf_xdp_mode mode)
8649 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
8652 return link->link.prog;
8653 return dev->xdp_state[mode].prog;
8656 u8 dev_xdp_prog_count(struct net_device *dev)
8661 for (i = 0; i < __MAX_XDP_MODE; i++)
8662 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
8666 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
8668 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
8670 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
8672 return prog ? prog->aux->id : 0;
8675 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
8676 struct bpf_xdp_link *link)
8678 dev->xdp_state[mode].link = link;
8679 dev->xdp_state[mode].prog = NULL;
8682 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
8683 struct bpf_prog *prog)
8685 dev->xdp_state[mode].link = NULL;
8686 dev->xdp_state[mode].prog = prog;
8689 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
8690 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
8691 u32 flags, struct bpf_prog *prog)
8693 struct netdev_bpf xdp;
8696 memset(&xdp, 0, sizeof(xdp));
8697 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
8698 xdp.extack = extack;
8702 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
8703 * "moved" into driver), so they don't increment it on their own, but
8704 * they do decrement refcnt when program is detached or replaced.
8705 * Given net_device also owns link/prog, we need to bump refcnt here
8706 * to prevent drivers from underflowing it.
8710 err = bpf_op(dev, &xdp);
8717 if (mode != XDP_MODE_HW)
8718 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
8723 static void dev_xdp_uninstall(struct net_device *dev)
8725 struct bpf_xdp_link *link;
8726 struct bpf_prog *prog;
8727 enum bpf_xdp_mode mode;
8732 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
8733 prog = dev_xdp_prog(dev, mode);
8737 bpf_op = dev_xdp_bpf_op(dev, mode);
8741 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
8743 /* auto-detach link from net device */
8744 link = dev_xdp_link(dev, mode);
8750 dev_xdp_set_link(dev, mode, NULL);
8754 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
8755 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
8756 struct bpf_prog *old_prog, u32 flags)
8758 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
8759 struct bpf_prog *cur_prog;
8760 struct net_device *upper;
8761 struct list_head *iter;
8762 enum bpf_xdp_mode mode;
8768 /* either link or prog attachment, never both */
8769 if (link && (new_prog || old_prog))
8771 /* link supports only XDP mode flags */
8772 if (link && (flags & ~XDP_FLAGS_MODES)) {
8773 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
8776 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
8777 if (num_modes > 1) {
8778 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
8781 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
8782 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
8783 NL_SET_ERR_MSG(extack,
8784 "More than one program loaded, unset mode is ambiguous");
8787 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
8788 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
8789 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
8793 mode = dev_xdp_mode(dev, flags);
8794 /* can't replace attached link */
8795 if (dev_xdp_link(dev, mode)) {
8796 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
8800 /* don't allow if an upper device already has a program */
8801 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
8802 if (dev_xdp_prog_count(upper) > 0) {
8803 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
8808 cur_prog = dev_xdp_prog(dev, mode);
8809 /* can't replace attached prog with link */
8810 if (link && cur_prog) {
8811 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
8814 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
8815 NL_SET_ERR_MSG(extack, "Active program does not match expected");
8819 /* put effective new program into new_prog */
8821 new_prog = link->link.prog;
8824 bool offload = mode == XDP_MODE_HW;
8825 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
8826 ? XDP_MODE_DRV : XDP_MODE_SKB;
8828 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
8829 NL_SET_ERR_MSG(extack, "XDP program already attached");
8832 if (!offload && dev_xdp_prog(dev, other_mode)) {
8833 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
8836 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
8837 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
8840 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
8841 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
8844 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
8845 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
8850 /* don't call drivers if the effective program didn't change */
8851 if (new_prog != cur_prog) {
8852 bpf_op = dev_xdp_bpf_op(dev, mode);
8854 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
8858 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
8864 dev_xdp_set_link(dev, mode, link);
8866 dev_xdp_set_prog(dev, mode, new_prog);
8868 bpf_prog_put(cur_prog);
8873 static int dev_xdp_attach_link(struct net_device *dev,
8874 struct netlink_ext_ack *extack,
8875 struct bpf_xdp_link *link)
8877 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
8880 static int dev_xdp_detach_link(struct net_device *dev,
8881 struct netlink_ext_ack *extack,
8882 struct bpf_xdp_link *link)
8884 enum bpf_xdp_mode mode;
8889 mode = dev_xdp_mode(dev, link->flags);
8890 if (dev_xdp_link(dev, mode) != link)
8893 bpf_op = dev_xdp_bpf_op(dev, mode);
8894 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
8895 dev_xdp_set_link(dev, mode, NULL);
8899 static void bpf_xdp_link_release(struct bpf_link *link)
8901 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8905 /* if racing with net_device's tear down, xdp_link->dev might be
8906 * already NULL, in which case link was already auto-detached
8908 if (xdp_link->dev) {
8909 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
8910 xdp_link->dev = NULL;
8916 static int bpf_xdp_link_detach(struct bpf_link *link)
8918 bpf_xdp_link_release(link);
8922 static void bpf_xdp_link_dealloc(struct bpf_link *link)
8924 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8929 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
8930 struct seq_file *seq)
8932 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8937 ifindex = xdp_link->dev->ifindex;
8940 seq_printf(seq, "ifindex:\t%u\n", ifindex);
8943 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
8944 struct bpf_link_info *info)
8946 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8951 ifindex = xdp_link->dev->ifindex;
8954 info->xdp.ifindex = ifindex;
8958 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
8959 struct bpf_prog *old_prog)
8961 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
8962 enum bpf_xdp_mode mode;
8968 /* link might have been auto-released already, so fail */
8969 if (!xdp_link->dev) {
8974 if (old_prog && link->prog != old_prog) {
8978 old_prog = link->prog;
8979 if (old_prog == new_prog) {
8980 /* no-op, don't disturb drivers */
8981 bpf_prog_put(new_prog);
8985 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
8986 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
8987 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
8988 xdp_link->flags, new_prog);
8992 old_prog = xchg(&link->prog, new_prog);
8993 bpf_prog_put(old_prog);
9000 static const struct bpf_link_ops bpf_xdp_link_lops = {
9001 .release = bpf_xdp_link_release,
9002 .dealloc = bpf_xdp_link_dealloc,
9003 .detach = bpf_xdp_link_detach,
9004 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9005 .fill_link_info = bpf_xdp_link_fill_link_info,
9006 .update_prog = bpf_xdp_link_update,
9009 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9011 struct net *net = current->nsproxy->net_ns;
9012 struct bpf_link_primer link_primer;
9013 struct bpf_xdp_link *link;
9014 struct net_device *dev;
9018 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9024 link = kzalloc(sizeof(*link), GFP_USER);
9030 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9032 link->flags = attr->link_create.flags;
9034 err = bpf_link_prime(&link->link, &link_primer);
9040 err = dev_xdp_attach_link(dev, NULL, link);
9045 bpf_link_cleanup(&link_primer);
9049 fd = bpf_link_settle(&link_primer);
9050 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9063 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9065 * @extack: netlink extended ack
9066 * @fd: new program fd or negative value to clear
9067 * @expected_fd: old program fd that userspace expects to replace or clear
9068 * @flags: xdp-related flags
9070 * Set or clear a bpf program for a device
9072 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9073 int fd, int expected_fd, u32 flags)
9075 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9076 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9082 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9083 mode != XDP_MODE_SKB);
9084 if (IS_ERR(new_prog))
9085 return PTR_ERR(new_prog);
9088 if (expected_fd >= 0) {
9089 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9090 mode != XDP_MODE_SKB);
9091 if (IS_ERR(old_prog)) {
9092 err = PTR_ERR(old_prog);
9098 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9101 if (err && new_prog)
9102 bpf_prog_put(new_prog);
9104 bpf_prog_put(old_prog);
9109 * dev_new_index - allocate an ifindex
9110 * @net: the applicable net namespace
9112 * Returns a suitable unique value for a new device interface
9113 * number. The caller must hold the rtnl semaphore or the
9114 * dev_base_lock to be sure it remains unique.
9116 static int dev_new_index(struct net *net)
9118 int ifindex = net->ifindex;
9123 if (!__dev_get_by_index(net, ifindex))
9124 return net->ifindex = ifindex;
9128 /* Delayed registration/unregisteration */
9129 static LIST_HEAD(net_todo_list);
9130 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9132 static void net_set_todo(struct net_device *dev)
9134 list_add_tail(&dev->todo_list, &net_todo_list);
9135 dev_net(dev)->dev_unreg_count++;
9138 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9139 struct net_device *upper, netdev_features_t features)
9141 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9142 netdev_features_t feature;
9145 for_each_netdev_feature(upper_disables, feature_bit) {
9146 feature = __NETIF_F_BIT(feature_bit);
9147 if (!(upper->wanted_features & feature)
9148 && (features & feature)) {
9149 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9150 &feature, upper->name);
9151 features &= ~feature;
9158 static void netdev_sync_lower_features(struct net_device *upper,
9159 struct net_device *lower, netdev_features_t features)
9161 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9162 netdev_features_t feature;
9165 for_each_netdev_feature(upper_disables, feature_bit) {
9166 feature = __NETIF_F_BIT(feature_bit);
9167 if (!(features & feature) && (lower->features & feature)) {
9168 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9169 &feature, lower->name);
9170 lower->wanted_features &= ~feature;
9171 __netdev_update_features(lower);
9173 if (unlikely(lower->features & feature))
9174 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9175 &feature, lower->name);
9177 netdev_features_change(lower);
9182 static netdev_features_t netdev_fix_features(struct net_device *dev,
9183 netdev_features_t features)
9185 /* Fix illegal checksum combinations */
9186 if ((features & NETIF_F_HW_CSUM) &&
9187 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9188 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9189 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9192 /* TSO requires that SG is present as well. */
9193 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9194 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9195 features &= ~NETIF_F_ALL_TSO;
9198 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9199 !(features & NETIF_F_IP_CSUM)) {
9200 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9201 features &= ~NETIF_F_TSO;
9202 features &= ~NETIF_F_TSO_ECN;
9205 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9206 !(features & NETIF_F_IPV6_CSUM)) {
9207 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9208 features &= ~NETIF_F_TSO6;
9211 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9212 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9213 features &= ~NETIF_F_TSO_MANGLEID;
9215 /* TSO ECN requires that TSO is present as well. */
9216 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9217 features &= ~NETIF_F_TSO_ECN;
9219 /* Software GSO depends on SG. */
9220 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9221 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9222 features &= ~NETIF_F_GSO;
9225 /* GSO partial features require GSO partial be set */
9226 if ((features & dev->gso_partial_features) &&
9227 !(features & NETIF_F_GSO_PARTIAL)) {
9229 "Dropping partially supported GSO features since no GSO partial.\n");
9230 features &= ~dev->gso_partial_features;
9233 if (!(features & NETIF_F_RXCSUM)) {
9234 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9235 * successfully merged by hardware must also have the
9236 * checksum verified by hardware. If the user does not
9237 * want to enable RXCSUM, logically, we should disable GRO_HW.
9239 if (features & NETIF_F_GRO_HW) {
9240 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9241 features &= ~NETIF_F_GRO_HW;
9245 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9246 if (features & NETIF_F_RXFCS) {
9247 if (features & NETIF_F_LRO) {
9248 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9249 features &= ~NETIF_F_LRO;
9252 if (features & NETIF_F_GRO_HW) {
9253 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9254 features &= ~NETIF_F_GRO_HW;
9258 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9259 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9260 features &= ~NETIF_F_LRO;
9263 if (features & NETIF_F_HW_TLS_TX) {
9264 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9265 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9266 bool hw_csum = features & NETIF_F_HW_CSUM;
9268 if (!ip_csum && !hw_csum) {
9269 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9270 features &= ~NETIF_F_HW_TLS_TX;
9274 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9275 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9276 features &= ~NETIF_F_HW_TLS_RX;
9282 int __netdev_update_features(struct net_device *dev)
9284 struct net_device *upper, *lower;
9285 netdev_features_t features;
9286 struct list_head *iter;
9291 features = netdev_get_wanted_features(dev);
9293 if (dev->netdev_ops->ndo_fix_features)
9294 features = dev->netdev_ops->ndo_fix_features(dev, features);
9296 /* driver might be less strict about feature dependencies */
9297 features = netdev_fix_features(dev, features);
9299 /* some features can't be enabled if they're off on an upper device */
9300 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9301 features = netdev_sync_upper_features(dev, upper, features);
9303 if (dev->features == features)
9306 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9307 &dev->features, &features);
9309 if (dev->netdev_ops->ndo_set_features)
9310 err = dev->netdev_ops->ndo_set_features(dev, features);
9314 if (unlikely(err < 0)) {
9316 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9317 err, &features, &dev->features);
9318 /* return non-0 since some features might have changed and
9319 * it's better to fire a spurious notification than miss it
9325 /* some features must be disabled on lower devices when disabled
9326 * on an upper device (think: bonding master or bridge)
9328 netdev_for_each_lower_dev(dev, lower, iter)
9329 netdev_sync_lower_features(dev, lower, features);
9332 netdev_features_t diff = features ^ dev->features;
9334 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9335 /* udp_tunnel_{get,drop}_rx_info both need
9336 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9337 * device, or they won't do anything.
9338 * Thus we need to update dev->features
9339 * *before* calling udp_tunnel_get_rx_info,
9340 * but *after* calling udp_tunnel_drop_rx_info.
9342 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9343 dev->features = features;
9344 udp_tunnel_get_rx_info(dev);
9346 udp_tunnel_drop_rx_info(dev);
9350 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9351 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9352 dev->features = features;
9353 err |= vlan_get_rx_ctag_filter_info(dev);
9355 vlan_drop_rx_ctag_filter_info(dev);
9359 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9360 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9361 dev->features = features;
9362 err |= vlan_get_rx_stag_filter_info(dev);
9364 vlan_drop_rx_stag_filter_info(dev);
9368 dev->features = features;
9371 return err < 0 ? 0 : 1;
9375 * netdev_update_features - recalculate device features
9376 * @dev: the device to check
9378 * Recalculate dev->features set and send notifications if it
9379 * has changed. Should be called after driver or hardware dependent
9380 * conditions might have changed that influence the features.
9382 void netdev_update_features(struct net_device *dev)
9384 if (__netdev_update_features(dev))
9385 netdev_features_change(dev);
9387 EXPORT_SYMBOL(netdev_update_features);
9390 * netdev_change_features - recalculate device features
9391 * @dev: the device to check
9393 * Recalculate dev->features set and send notifications even
9394 * if they have not changed. Should be called instead of
9395 * netdev_update_features() if also dev->vlan_features might
9396 * have changed to allow the changes to be propagated to stacked
9399 void netdev_change_features(struct net_device *dev)
9401 __netdev_update_features(dev);
9402 netdev_features_change(dev);
9404 EXPORT_SYMBOL(netdev_change_features);
9407 * netif_stacked_transfer_operstate - transfer operstate
9408 * @rootdev: the root or lower level device to transfer state from
9409 * @dev: the device to transfer operstate to
9411 * Transfer operational state from root to device. This is normally
9412 * called when a stacking relationship exists between the root
9413 * device and the device(a leaf device).
9415 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9416 struct net_device *dev)
9418 if (rootdev->operstate == IF_OPER_DORMANT)
9419 netif_dormant_on(dev);
9421 netif_dormant_off(dev);
9423 if (rootdev->operstate == IF_OPER_TESTING)
9424 netif_testing_on(dev);
9426 netif_testing_off(dev);
9428 if (netif_carrier_ok(rootdev))
9429 netif_carrier_on(dev);
9431 netif_carrier_off(dev);
9433 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9435 static int netif_alloc_rx_queues(struct net_device *dev)
9437 unsigned int i, count = dev->num_rx_queues;
9438 struct netdev_rx_queue *rx;
9439 size_t sz = count * sizeof(*rx);
9444 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9450 for (i = 0; i < count; i++) {
9453 /* XDP RX-queue setup */
9454 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9461 /* Rollback successful reg's and free other resources */
9463 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9469 static void netif_free_rx_queues(struct net_device *dev)
9471 unsigned int i, count = dev->num_rx_queues;
9473 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9477 for (i = 0; i < count; i++)
9478 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9483 static void netdev_init_one_queue(struct net_device *dev,
9484 struct netdev_queue *queue, void *_unused)
9486 /* Initialize queue lock */
9487 spin_lock_init(&queue->_xmit_lock);
9488 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9489 queue->xmit_lock_owner = -1;
9490 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9493 dql_init(&queue->dql, HZ);
9497 static void netif_free_tx_queues(struct net_device *dev)
9502 static int netif_alloc_netdev_queues(struct net_device *dev)
9504 unsigned int count = dev->num_tx_queues;
9505 struct netdev_queue *tx;
9506 size_t sz = count * sizeof(*tx);
9508 if (count < 1 || count > 0xffff)
9511 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9517 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9518 spin_lock_init(&dev->tx_global_lock);
9523 void netif_tx_stop_all_queues(struct net_device *dev)
9527 for (i = 0; i < dev->num_tx_queues; i++) {
9528 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9530 netif_tx_stop_queue(txq);
9533 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9536 * register_netdevice - register a network device
9537 * @dev: device to register
9539 * Take a completed network device structure and add it to the kernel
9540 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9541 * chain. 0 is returned on success. A negative errno code is returned
9542 * on a failure to set up the device, or if the name is a duplicate.
9544 * Callers must hold the rtnl semaphore. You may want
9545 * register_netdev() instead of this.
9548 * The locking appears insufficient to guarantee two parallel registers
9549 * will not get the same name.
9552 int register_netdevice(struct net_device *dev)
9555 struct net *net = dev_net(dev);
9557 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9558 NETDEV_FEATURE_COUNT);
9559 BUG_ON(dev_boot_phase);
9564 /* When net_device's are persistent, this will be fatal. */
9565 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9568 ret = ethtool_check_ops(dev->ethtool_ops);
9572 spin_lock_init(&dev->addr_list_lock);
9573 netdev_set_addr_lockdep_class(dev);
9575 ret = dev_get_valid_name(net, dev, dev->name);
9580 dev->name_node = netdev_name_node_head_alloc(dev);
9581 if (!dev->name_node)
9584 /* Init, if this function is available */
9585 if (dev->netdev_ops->ndo_init) {
9586 ret = dev->netdev_ops->ndo_init(dev);
9594 if (((dev->hw_features | dev->features) &
9595 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9596 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9597 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9598 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9605 dev->ifindex = dev_new_index(net);
9606 else if (__dev_get_by_index(net, dev->ifindex))
9609 /* Transfer changeable features to wanted_features and enable
9610 * software offloads (GSO and GRO).
9612 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9613 dev->features |= NETIF_F_SOFT_FEATURES;
9615 if (dev->udp_tunnel_nic_info) {
9616 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9617 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9620 dev->wanted_features = dev->features & dev->hw_features;
9622 if (!(dev->flags & IFF_LOOPBACK))
9623 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9625 /* If IPv4 TCP segmentation offload is supported we should also
9626 * allow the device to enable segmenting the frame with the option
9627 * of ignoring a static IP ID value. This doesn't enable the
9628 * feature itself but allows the user to enable it later.
9630 if (dev->hw_features & NETIF_F_TSO)
9631 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9632 if (dev->vlan_features & NETIF_F_TSO)
9633 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9634 if (dev->mpls_features & NETIF_F_TSO)
9635 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9636 if (dev->hw_enc_features & NETIF_F_TSO)
9637 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9639 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9641 dev->vlan_features |= NETIF_F_HIGHDMA;
9643 /* Make NETIF_F_SG inheritable to tunnel devices.
9645 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9647 /* Make NETIF_F_SG inheritable to MPLS.
9649 dev->mpls_features |= NETIF_F_SG;
9651 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9652 ret = notifier_to_errno(ret);
9656 ret = netdev_register_kobject(dev);
9658 dev->reg_state = NETREG_UNREGISTERED;
9661 dev->reg_state = NETREG_REGISTERED;
9663 __netdev_update_features(dev);
9666 * Default initial state at registry is that the
9667 * device is present.
9670 set_bit(__LINK_STATE_PRESENT, &dev->state);
9672 linkwatch_init_dev(dev);
9674 dev_init_scheduler(dev);
9676 list_netdevice(dev);
9677 add_device_randomness(dev->dev_addr, dev->addr_len);
9679 /* If the device has permanent device address, driver should
9680 * set dev_addr and also addr_assign_type should be set to
9681 * NET_ADDR_PERM (default value).
9683 if (dev->addr_assign_type == NET_ADDR_PERM)
9684 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9686 /* Notify protocols, that a new device appeared. */
9687 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9688 ret = notifier_to_errno(ret);
9690 /* Expect explicit free_netdev() on failure */
9691 dev->needs_free_netdev = false;
9692 unregister_netdevice_queue(dev, NULL);
9696 * Prevent userspace races by waiting until the network
9697 * device is fully setup before sending notifications.
9699 if (!dev->rtnl_link_ops ||
9700 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9701 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9707 if (dev->netdev_ops->ndo_uninit)
9708 dev->netdev_ops->ndo_uninit(dev);
9709 if (dev->priv_destructor)
9710 dev->priv_destructor(dev);
9712 netdev_name_node_free(dev->name_node);
9715 EXPORT_SYMBOL(register_netdevice);
9718 * init_dummy_netdev - init a dummy network device for NAPI
9719 * @dev: device to init
9721 * This takes a network device structure and initialize the minimum
9722 * amount of fields so it can be used to schedule NAPI polls without
9723 * registering a full blown interface. This is to be used by drivers
9724 * that need to tie several hardware interfaces to a single NAPI
9725 * poll scheduler due to HW limitations.
9727 int init_dummy_netdev(struct net_device *dev)
9729 /* Clear everything. Note we don't initialize spinlocks
9730 * are they aren't supposed to be taken by any of the
9731 * NAPI code and this dummy netdev is supposed to be
9732 * only ever used for NAPI polls
9734 memset(dev, 0, sizeof(struct net_device));
9736 /* make sure we BUG if trying to hit standard
9737 * register/unregister code path
9739 dev->reg_state = NETREG_DUMMY;
9741 /* NAPI wants this */
9742 INIT_LIST_HEAD(&dev->napi_list);
9744 /* a dummy interface is started by default */
9745 set_bit(__LINK_STATE_PRESENT, &dev->state);
9746 set_bit(__LINK_STATE_START, &dev->state);
9748 /* napi_busy_loop stats accounting wants this */
9749 dev_net_set(dev, &init_net);
9751 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9752 * because users of this 'device' dont need to change
9758 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9762 * register_netdev - register a network device
9763 * @dev: device to register
9765 * Take a completed network device structure and add it to the kernel
9766 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9767 * chain. 0 is returned on success. A negative errno code is returned
9768 * on a failure to set up the device, or if the name is a duplicate.
9770 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9771 * and expands the device name if you passed a format string to
9774 int register_netdev(struct net_device *dev)
9778 if (rtnl_lock_killable())
9780 err = register_netdevice(dev);
9784 EXPORT_SYMBOL(register_netdev);
9786 int netdev_refcnt_read(const struct net_device *dev)
9788 #ifdef CONFIG_PCPU_DEV_REFCNT
9791 for_each_possible_cpu(i)
9792 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9795 return refcount_read(&dev->dev_refcnt);
9798 EXPORT_SYMBOL(netdev_refcnt_read);
9800 int netdev_unregister_timeout_secs __read_mostly = 10;
9802 #define WAIT_REFS_MIN_MSECS 1
9803 #define WAIT_REFS_MAX_MSECS 250
9805 * netdev_wait_allrefs - wait until all references are gone.
9806 * @dev: target net_device
9808 * This is called when unregistering network devices.
9810 * Any protocol or device that holds a reference should register
9811 * for netdevice notification, and cleanup and put back the
9812 * reference if they receive an UNREGISTER event.
9813 * We can get stuck here if buggy protocols don't correctly
9816 static void netdev_wait_allrefs(struct net_device *dev)
9818 unsigned long rebroadcast_time, warning_time;
9819 int wait = 0, refcnt;
9821 linkwatch_forget_dev(dev);
9823 rebroadcast_time = warning_time = jiffies;
9824 refcnt = netdev_refcnt_read(dev);
9826 while (refcnt != 1) {
9827 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9830 /* Rebroadcast unregister notification */
9831 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9837 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9839 /* We must not have linkwatch events
9840 * pending on unregister. If this
9841 * happens, we simply run the queue
9842 * unscheduled, resulting in a noop
9845 linkwatch_run_queue();
9850 rebroadcast_time = jiffies;
9855 wait = WAIT_REFS_MIN_MSECS;
9858 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
9861 refcnt = netdev_refcnt_read(dev);
9864 time_after(jiffies, warning_time +
9865 netdev_unregister_timeout_secs * HZ)) {
9866 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9868 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
9869 warning_time = jiffies;
9878 * register_netdevice(x1);
9879 * register_netdevice(x2);
9881 * unregister_netdevice(y1);
9882 * unregister_netdevice(y2);
9888 * We are invoked by rtnl_unlock().
9889 * This allows us to deal with problems:
9890 * 1) We can delete sysfs objects which invoke hotplug
9891 * without deadlocking with linkwatch via keventd.
9892 * 2) Since we run with the RTNL semaphore not held, we can sleep
9893 * safely in order to wait for the netdev refcnt to drop to zero.
9895 * We must not return until all unregister events added during
9896 * the interval the lock was held have been completed.
9898 void netdev_run_todo(void)
9900 struct list_head list;
9901 #ifdef CONFIG_LOCKDEP
9902 struct list_head unlink_list;
9904 list_replace_init(&net_unlink_list, &unlink_list);
9906 while (!list_empty(&unlink_list)) {
9907 struct net_device *dev = list_first_entry(&unlink_list,
9910 list_del_init(&dev->unlink_list);
9911 dev->nested_level = dev->lower_level - 1;
9915 /* Snapshot list, allow later requests */
9916 list_replace_init(&net_todo_list, &list);
9921 /* Wait for rcu callbacks to finish before next phase */
9922 if (!list_empty(&list))
9925 while (!list_empty(&list)) {
9926 struct net_device *dev
9927 = list_first_entry(&list, struct net_device, todo_list);
9928 list_del(&dev->todo_list);
9930 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9931 pr_err("network todo '%s' but state %d\n",
9932 dev->name, dev->reg_state);
9937 dev->reg_state = NETREG_UNREGISTERED;
9939 netdev_wait_allrefs(dev);
9942 BUG_ON(netdev_refcnt_read(dev) != 1);
9943 BUG_ON(!list_empty(&dev->ptype_all));
9944 BUG_ON(!list_empty(&dev->ptype_specific));
9945 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9946 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9947 #if IS_ENABLED(CONFIG_DECNET)
9948 WARN_ON(dev->dn_ptr);
9950 if (dev->priv_destructor)
9951 dev->priv_destructor(dev);
9952 if (dev->needs_free_netdev)
9955 /* Report a network device has been unregistered */
9957 dev_net(dev)->dev_unreg_count--;
9959 wake_up(&netdev_unregistering_wq);
9961 /* Free network device */
9962 kobject_put(&dev->dev.kobj);
9966 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9967 * all the same fields in the same order as net_device_stats, with only
9968 * the type differing, but rtnl_link_stats64 may have additional fields
9969 * at the end for newer counters.
9971 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9972 const struct net_device_stats *netdev_stats)
9974 #if BITS_PER_LONG == 64
9975 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9976 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9977 /* zero out counters that only exist in rtnl_link_stats64 */
9978 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9979 sizeof(*stats64) - sizeof(*netdev_stats));
9981 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9982 const unsigned long *src = (const unsigned long *)netdev_stats;
9983 u64 *dst = (u64 *)stats64;
9985 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9986 for (i = 0; i < n; i++)
9988 /* zero out counters that only exist in rtnl_link_stats64 */
9989 memset((char *)stats64 + n * sizeof(u64), 0,
9990 sizeof(*stats64) - n * sizeof(u64));
9993 EXPORT_SYMBOL(netdev_stats_to_stats64);
9996 * dev_get_stats - get network device statistics
9997 * @dev: device to get statistics from
9998 * @storage: place to store stats
10000 * Get network statistics from device. Return @storage.
10001 * The device driver may provide its own method by setting
10002 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10003 * otherwise the internal statistics structure is used.
10005 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10006 struct rtnl_link_stats64 *storage)
10008 const struct net_device_ops *ops = dev->netdev_ops;
10010 if (ops->ndo_get_stats64) {
10011 memset(storage, 0, sizeof(*storage));
10012 ops->ndo_get_stats64(dev, storage);
10013 } else if (ops->ndo_get_stats) {
10014 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10016 netdev_stats_to_stats64(storage, &dev->stats);
10018 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10019 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10020 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10023 EXPORT_SYMBOL(dev_get_stats);
10026 * dev_fetch_sw_netstats - get per-cpu network device statistics
10027 * @s: place to store stats
10028 * @netstats: per-cpu network stats to read from
10030 * Read per-cpu network statistics and populate the related fields in @s.
10032 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10033 const struct pcpu_sw_netstats __percpu *netstats)
10037 for_each_possible_cpu(cpu) {
10038 const struct pcpu_sw_netstats *stats;
10039 struct pcpu_sw_netstats tmp;
10040 unsigned int start;
10042 stats = per_cpu_ptr(netstats, cpu);
10044 start = u64_stats_fetch_begin_irq(&stats->syncp);
10045 tmp.rx_packets = stats->rx_packets;
10046 tmp.rx_bytes = stats->rx_bytes;
10047 tmp.tx_packets = stats->tx_packets;
10048 tmp.tx_bytes = stats->tx_bytes;
10049 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10051 s->rx_packets += tmp.rx_packets;
10052 s->rx_bytes += tmp.rx_bytes;
10053 s->tx_packets += tmp.tx_packets;
10054 s->tx_bytes += tmp.tx_bytes;
10057 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10060 * dev_get_tstats64 - ndo_get_stats64 implementation
10061 * @dev: device to get statistics from
10062 * @s: place to store stats
10064 * Populate @s from dev->stats and dev->tstats. Can be used as
10065 * ndo_get_stats64() callback.
10067 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10069 netdev_stats_to_stats64(s, &dev->stats);
10070 dev_fetch_sw_netstats(s, dev->tstats);
10072 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10074 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10076 struct netdev_queue *queue = dev_ingress_queue(dev);
10078 #ifdef CONFIG_NET_CLS_ACT
10081 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10084 netdev_init_one_queue(dev, queue, NULL);
10085 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10086 queue->qdisc_sleeping = &noop_qdisc;
10087 rcu_assign_pointer(dev->ingress_queue, queue);
10092 static const struct ethtool_ops default_ethtool_ops;
10094 void netdev_set_default_ethtool_ops(struct net_device *dev,
10095 const struct ethtool_ops *ops)
10097 if (dev->ethtool_ops == &default_ethtool_ops)
10098 dev->ethtool_ops = ops;
10100 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10102 void netdev_freemem(struct net_device *dev)
10104 char *addr = (char *)dev - dev->padded;
10110 * alloc_netdev_mqs - allocate network device
10111 * @sizeof_priv: size of private data to allocate space for
10112 * @name: device name format string
10113 * @name_assign_type: origin of device name
10114 * @setup: callback to initialize device
10115 * @txqs: the number of TX subqueues to allocate
10116 * @rxqs: the number of RX subqueues to allocate
10118 * Allocates a struct net_device with private data area for driver use
10119 * and performs basic initialization. Also allocates subqueue structs
10120 * for each queue on the device.
10122 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10123 unsigned char name_assign_type,
10124 void (*setup)(struct net_device *),
10125 unsigned int txqs, unsigned int rxqs)
10127 struct net_device *dev;
10128 unsigned int alloc_size;
10129 struct net_device *p;
10131 BUG_ON(strlen(name) >= sizeof(dev->name));
10134 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10139 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10143 alloc_size = sizeof(struct net_device);
10145 /* ensure 32-byte alignment of private area */
10146 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10147 alloc_size += sizeof_priv;
10149 /* ensure 32-byte alignment of whole construct */
10150 alloc_size += NETDEV_ALIGN - 1;
10152 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10156 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10157 dev->padded = (char *)dev - (char *)p;
10159 ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10160 #ifdef CONFIG_PCPU_DEV_REFCNT
10161 dev->pcpu_refcnt = alloc_percpu(int);
10162 if (!dev->pcpu_refcnt)
10166 refcount_set(&dev->dev_refcnt, 1);
10169 if (dev_addr_init(dev))
10175 dev_net_set(dev, &init_net);
10177 dev->gso_max_size = GSO_MAX_SIZE;
10178 dev->gso_max_segs = GSO_MAX_SEGS;
10179 dev->upper_level = 1;
10180 dev->lower_level = 1;
10181 #ifdef CONFIG_LOCKDEP
10182 dev->nested_level = 0;
10183 INIT_LIST_HEAD(&dev->unlink_list);
10186 INIT_LIST_HEAD(&dev->napi_list);
10187 INIT_LIST_HEAD(&dev->unreg_list);
10188 INIT_LIST_HEAD(&dev->close_list);
10189 INIT_LIST_HEAD(&dev->link_watch_list);
10190 INIT_LIST_HEAD(&dev->adj_list.upper);
10191 INIT_LIST_HEAD(&dev->adj_list.lower);
10192 INIT_LIST_HEAD(&dev->ptype_all);
10193 INIT_LIST_HEAD(&dev->ptype_specific);
10194 INIT_LIST_HEAD(&dev->net_notifier_list);
10195 #ifdef CONFIG_NET_SCHED
10196 hash_init(dev->qdisc_hash);
10198 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10201 if (!dev->tx_queue_len) {
10202 dev->priv_flags |= IFF_NO_QUEUE;
10203 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10206 dev->num_tx_queues = txqs;
10207 dev->real_num_tx_queues = txqs;
10208 if (netif_alloc_netdev_queues(dev))
10211 dev->num_rx_queues = rxqs;
10212 dev->real_num_rx_queues = rxqs;
10213 if (netif_alloc_rx_queues(dev))
10216 strcpy(dev->name, name);
10217 dev->name_assign_type = name_assign_type;
10218 dev->group = INIT_NETDEV_GROUP;
10219 if (!dev->ethtool_ops)
10220 dev->ethtool_ops = &default_ethtool_ops;
10222 nf_hook_netdev_init(dev);
10231 #ifdef CONFIG_PCPU_DEV_REFCNT
10232 free_percpu(dev->pcpu_refcnt);
10235 netdev_freemem(dev);
10238 EXPORT_SYMBOL(alloc_netdev_mqs);
10241 * free_netdev - free network device
10244 * This function does the last stage of destroying an allocated device
10245 * interface. The reference to the device object is released. If this
10246 * is the last reference then it will be freed.Must be called in process
10249 void free_netdev(struct net_device *dev)
10251 struct napi_struct *p, *n;
10255 /* When called immediately after register_netdevice() failed the unwind
10256 * handling may still be dismantling the device. Handle that case by
10257 * deferring the free.
10259 if (dev->reg_state == NETREG_UNREGISTERING) {
10261 dev->needs_free_netdev = true;
10265 netif_free_tx_queues(dev);
10266 netif_free_rx_queues(dev);
10268 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10270 /* Flush device addresses */
10271 dev_addr_flush(dev);
10273 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10276 ref_tracker_dir_exit(&dev->refcnt_tracker);
10277 #ifdef CONFIG_PCPU_DEV_REFCNT
10278 free_percpu(dev->pcpu_refcnt);
10279 dev->pcpu_refcnt = NULL;
10281 free_percpu(dev->xdp_bulkq);
10282 dev->xdp_bulkq = NULL;
10284 /* Compatibility with error handling in drivers */
10285 if (dev->reg_state == NETREG_UNINITIALIZED) {
10286 netdev_freemem(dev);
10290 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10291 dev->reg_state = NETREG_RELEASED;
10293 /* will free via device release */
10294 put_device(&dev->dev);
10296 EXPORT_SYMBOL(free_netdev);
10299 * synchronize_net - Synchronize with packet receive processing
10301 * Wait for packets currently being received to be done.
10302 * Does not block later packets from starting.
10304 void synchronize_net(void)
10307 if (rtnl_is_locked())
10308 synchronize_rcu_expedited();
10312 EXPORT_SYMBOL(synchronize_net);
10315 * unregister_netdevice_queue - remove device from the kernel
10319 * This function shuts down a device interface and removes it
10320 * from the kernel tables.
10321 * If head not NULL, device is queued to be unregistered later.
10323 * Callers must hold the rtnl semaphore. You may want
10324 * unregister_netdev() instead of this.
10327 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10332 list_move_tail(&dev->unreg_list, head);
10336 list_add(&dev->unreg_list, &single);
10337 unregister_netdevice_many(&single);
10340 EXPORT_SYMBOL(unregister_netdevice_queue);
10343 * unregister_netdevice_many - unregister many devices
10344 * @head: list of devices
10346 * Note: As most callers use a stack allocated list_head,
10347 * we force a list_del() to make sure stack wont be corrupted later.
10349 void unregister_netdevice_many(struct list_head *head)
10351 struct net_device *dev, *tmp;
10352 LIST_HEAD(close_head);
10354 BUG_ON(dev_boot_phase);
10357 if (list_empty(head))
10360 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10361 /* Some devices call without registering
10362 * for initialization unwind. Remove those
10363 * devices and proceed with the remaining.
10365 if (dev->reg_state == NETREG_UNINITIALIZED) {
10366 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10370 list_del(&dev->unreg_list);
10373 dev->dismantle = true;
10374 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10377 /* If device is running, close it first. */
10378 list_for_each_entry(dev, head, unreg_list)
10379 list_add_tail(&dev->close_list, &close_head);
10380 dev_close_many(&close_head, true);
10382 list_for_each_entry(dev, head, unreg_list) {
10383 /* And unlink it from device chain. */
10384 unlist_netdevice(dev);
10386 dev->reg_state = NETREG_UNREGISTERING;
10388 flush_all_backlogs();
10392 list_for_each_entry(dev, head, unreg_list) {
10393 struct sk_buff *skb = NULL;
10395 /* Shutdown queueing discipline. */
10398 dev_xdp_uninstall(dev);
10400 /* Notify protocols, that we are about to destroy
10401 * this device. They should clean all the things.
10403 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10405 if (!dev->rtnl_link_ops ||
10406 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10407 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10408 GFP_KERNEL, NULL, 0);
10411 * Flush the unicast and multicast chains
10416 netdev_name_node_alt_flush(dev);
10417 netdev_name_node_free(dev->name_node);
10419 if (dev->netdev_ops->ndo_uninit)
10420 dev->netdev_ops->ndo_uninit(dev);
10423 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10425 /* Notifier chain MUST detach us all upper devices. */
10426 WARN_ON(netdev_has_any_upper_dev(dev));
10427 WARN_ON(netdev_has_any_lower_dev(dev));
10429 /* Remove entries from kobject tree */
10430 netdev_unregister_kobject(dev);
10432 /* Remove XPS queueing entries */
10433 netif_reset_xps_queues_gt(dev, 0);
10439 list_for_each_entry(dev, head, unreg_list) {
10446 EXPORT_SYMBOL(unregister_netdevice_many);
10449 * unregister_netdev - remove device from the kernel
10452 * This function shuts down a device interface and removes it
10453 * from the kernel tables.
10455 * This is just a wrapper for unregister_netdevice that takes
10456 * the rtnl semaphore. In general you want to use this and not
10457 * unregister_netdevice.
10459 void unregister_netdev(struct net_device *dev)
10462 unregister_netdevice(dev);
10465 EXPORT_SYMBOL(unregister_netdev);
10468 * __dev_change_net_namespace - move device to different nethost namespace
10470 * @net: network namespace
10471 * @pat: If not NULL name pattern to try if the current device name
10472 * is already taken in the destination network namespace.
10473 * @new_ifindex: If not zero, specifies device index in the target
10476 * This function shuts down a device interface and moves it
10477 * to a new network namespace. On success 0 is returned, on
10478 * a failure a netagive errno code is returned.
10480 * Callers must hold the rtnl semaphore.
10483 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10484 const char *pat, int new_ifindex)
10486 struct net *net_old = dev_net(dev);
10491 /* Don't allow namespace local devices to be moved. */
10493 if (dev->features & NETIF_F_NETNS_LOCAL)
10496 /* Ensure the device has been registrered */
10497 if (dev->reg_state != NETREG_REGISTERED)
10500 /* Get out if there is nothing todo */
10502 if (net_eq(net_old, net))
10505 /* Pick the destination device name, and ensure
10506 * we can use it in the destination network namespace.
10509 if (netdev_name_in_use(net, dev->name)) {
10510 /* We get here if we can't use the current device name */
10513 err = dev_get_valid_name(net, dev, pat);
10518 /* Check that new_ifindex isn't used yet. */
10520 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10524 * And now a mini version of register_netdevice unregister_netdevice.
10527 /* If device is running close it first. */
10530 /* And unlink it from device chain */
10531 unlist_netdevice(dev);
10535 /* Shutdown queueing discipline. */
10538 /* Notify protocols, that we are about to destroy
10539 * this device. They should clean all the things.
10541 * Note that dev->reg_state stays at NETREG_REGISTERED.
10542 * This is wanted because this way 8021q and macvlan know
10543 * the device is just moving and can keep their slaves up.
10545 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10548 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10549 /* If there is an ifindex conflict assign a new one */
10550 if (!new_ifindex) {
10551 if (__dev_get_by_index(net, dev->ifindex))
10552 new_ifindex = dev_new_index(net);
10554 new_ifindex = dev->ifindex;
10557 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10561 * Flush the unicast and multicast chains
10566 /* Send a netdev-removed uevent to the old namespace */
10567 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10568 netdev_adjacent_del_links(dev);
10570 /* Move per-net netdevice notifiers that are following the netdevice */
10571 move_netdevice_notifiers_dev_net(dev, net);
10573 /* Actually switch the network namespace */
10574 dev_net_set(dev, net);
10575 dev->ifindex = new_ifindex;
10577 /* Send a netdev-add uevent to the new namespace */
10578 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10579 netdev_adjacent_add_links(dev);
10581 /* Fixup kobjects */
10582 err = device_rename(&dev->dev, dev->name);
10585 /* Adapt owner in case owning user namespace of target network
10586 * namespace is different from the original one.
10588 err = netdev_change_owner(dev, net_old, net);
10591 /* Add the device back in the hashes */
10592 list_netdevice(dev);
10594 /* Notify protocols, that a new device appeared. */
10595 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10598 * Prevent userspace races by waiting until the network
10599 * device is fully setup before sending notifications.
10601 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10608 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
10610 static int dev_cpu_dead(unsigned int oldcpu)
10612 struct sk_buff **list_skb;
10613 struct sk_buff *skb;
10615 struct softnet_data *sd, *oldsd, *remsd = NULL;
10617 local_irq_disable();
10618 cpu = smp_processor_id();
10619 sd = &per_cpu(softnet_data, cpu);
10620 oldsd = &per_cpu(softnet_data, oldcpu);
10622 /* Find end of our completion_queue. */
10623 list_skb = &sd->completion_queue;
10625 list_skb = &(*list_skb)->next;
10626 /* Append completion queue from offline CPU. */
10627 *list_skb = oldsd->completion_queue;
10628 oldsd->completion_queue = NULL;
10630 /* Append output queue from offline CPU. */
10631 if (oldsd->output_queue) {
10632 *sd->output_queue_tailp = oldsd->output_queue;
10633 sd->output_queue_tailp = oldsd->output_queue_tailp;
10634 oldsd->output_queue = NULL;
10635 oldsd->output_queue_tailp = &oldsd->output_queue;
10637 /* Append NAPI poll list from offline CPU, with one exception :
10638 * process_backlog() must be called by cpu owning percpu backlog.
10639 * We properly handle process_queue & input_pkt_queue later.
10641 while (!list_empty(&oldsd->poll_list)) {
10642 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10643 struct napi_struct,
10646 list_del_init(&napi->poll_list);
10647 if (napi->poll == process_backlog)
10650 ____napi_schedule(sd, napi);
10653 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10654 local_irq_enable();
10657 remsd = oldsd->rps_ipi_list;
10658 oldsd->rps_ipi_list = NULL;
10660 /* send out pending IPI's on offline CPU */
10661 net_rps_send_ipi(remsd);
10663 /* Process offline CPU's input_pkt_queue */
10664 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10666 input_queue_head_incr(oldsd);
10668 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10670 input_queue_head_incr(oldsd);
10677 * netdev_increment_features - increment feature set by one
10678 * @all: current feature set
10679 * @one: new feature set
10680 * @mask: mask feature set
10682 * Computes a new feature set after adding a device with feature set
10683 * @one to the master device with current feature set @all. Will not
10684 * enable anything that is off in @mask. Returns the new feature set.
10686 netdev_features_t netdev_increment_features(netdev_features_t all,
10687 netdev_features_t one, netdev_features_t mask)
10689 if (mask & NETIF_F_HW_CSUM)
10690 mask |= NETIF_F_CSUM_MASK;
10691 mask |= NETIF_F_VLAN_CHALLENGED;
10693 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10694 all &= one | ~NETIF_F_ALL_FOR_ALL;
10696 /* If one device supports hw checksumming, set for all. */
10697 if (all & NETIF_F_HW_CSUM)
10698 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10702 EXPORT_SYMBOL(netdev_increment_features);
10704 static struct hlist_head * __net_init netdev_create_hash(void)
10707 struct hlist_head *hash;
10709 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10711 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10712 INIT_HLIST_HEAD(&hash[i]);
10717 /* Initialize per network namespace state */
10718 static int __net_init netdev_init(struct net *net)
10720 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10721 8 * sizeof_field(struct napi_struct, gro_bitmask));
10723 if (net != &init_net)
10724 INIT_LIST_HEAD(&net->dev_base_head);
10726 net->dev_name_head = netdev_create_hash();
10727 if (net->dev_name_head == NULL)
10730 net->dev_index_head = netdev_create_hash();
10731 if (net->dev_index_head == NULL)
10734 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10739 kfree(net->dev_name_head);
10745 * netdev_drivername - network driver for the device
10746 * @dev: network device
10748 * Determine network driver for device.
10750 const char *netdev_drivername(const struct net_device *dev)
10752 const struct device_driver *driver;
10753 const struct device *parent;
10754 const char *empty = "";
10756 parent = dev->dev.parent;
10760 driver = parent->driver;
10761 if (driver && driver->name)
10762 return driver->name;
10766 static void __netdev_printk(const char *level, const struct net_device *dev,
10767 struct va_format *vaf)
10769 if (dev && dev->dev.parent) {
10770 dev_printk_emit(level[1] - '0',
10773 dev_driver_string(dev->dev.parent),
10774 dev_name(dev->dev.parent),
10775 netdev_name(dev), netdev_reg_state(dev),
10778 printk("%s%s%s: %pV",
10779 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10781 printk("%s(NULL net_device): %pV", level, vaf);
10785 void netdev_printk(const char *level, const struct net_device *dev,
10786 const char *format, ...)
10788 struct va_format vaf;
10791 va_start(args, format);
10796 __netdev_printk(level, dev, &vaf);
10800 EXPORT_SYMBOL(netdev_printk);
10802 #define define_netdev_printk_level(func, level) \
10803 void func(const struct net_device *dev, const char *fmt, ...) \
10805 struct va_format vaf; \
10808 va_start(args, fmt); \
10813 __netdev_printk(level, dev, &vaf); \
10817 EXPORT_SYMBOL(func);
10819 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10820 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10821 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10822 define_netdev_printk_level(netdev_err, KERN_ERR);
10823 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10824 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10825 define_netdev_printk_level(netdev_info, KERN_INFO);
10827 static void __net_exit netdev_exit(struct net *net)
10829 kfree(net->dev_name_head);
10830 kfree(net->dev_index_head);
10831 if (net != &init_net)
10832 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10835 static struct pernet_operations __net_initdata netdev_net_ops = {
10836 .init = netdev_init,
10837 .exit = netdev_exit,
10840 static void __net_exit default_device_exit(struct net *net)
10842 struct net_device *dev, *aux;
10844 * Push all migratable network devices back to the
10845 * initial network namespace
10848 for_each_netdev_safe(net, dev, aux) {
10850 char fb_name[IFNAMSIZ];
10852 /* Ignore unmoveable devices (i.e. loopback) */
10853 if (dev->features & NETIF_F_NETNS_LOCAL)
10856 /* Leave virtual devices for the generic cleanup */
10857 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
10860 /* Push remaining network devices to init_net */
10861 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10862 if (netdev_name_in_use(&init_net, fb_name))
10863 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10864 err = dev_change_net_namespace(dev, &init_net, fb_name);
10866 pr_emerg("%s: failed to move %s to init_net: %d\n",
10867 __func__, dev->name, err);
10874 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10876 /* Return with the rtnl_lock held when there are no network
10877 * devices unregistering in any network namespace in net_list.
10880 bool unregistering;
10881 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10883 add_wait_queue(&netdev_unregistering_wq, &wait);
10885 unregistering = false;
10887 list_for_each_entry(net, net_list, exit_list) {
10888 if (net->dev_unreg_count > 0) {
10889 unregistering = true;
10893 if (!unregistering)
10897 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10899 remove_wait_queue(&netdev_unregistering_wq, &wait);
10902 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10904 /* At exit all network devices most be removed from a network
10905 * namespace. Do this in the reverse order of registration.
10906 * Do this across as many network namespaces as possible to
10907 * improve batching efficiency.
10909 struct net_device *dev;
10911 LIST_HEAD(dev_kill_list);
10913 /* To prevent network device cleanup code from dereferencing
10914 * loopback devices or network devices that have been freed
10915 * wait here for all pending unregistrations to complete,
10916 * before unregistring the loopback device and allowing the
10917 * network namespace be freed.
10919 * The netdev todo list containing all network devices
10920 * unregistrations that happen in default_device_exit_batch
10921 * will run in the rtnl_unlock() at the end of
10922 * default_device_exit_batch.
10924 rtnl_lock_unregistering(net_list);
10925 list_for_each_entry(net, net_list, exit_list) {
10926 for_each_netdev_reverse(net, dev) {
10927 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10928 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10930 unregister_netdevice_queue(dev, &dev_kill_list);
10933 unregister_netdevice_many(&dev_kill_list);
10937 static struct pernet_operations __net_initdata default_device_ops = {
10938 .exit = default_device_exit,
10939 .exit_batch = default_device_exit_batch,
10943 * Initialize the DEV module. At boot time this walks the device list and
10944 * unhooks any devices that fail to initialise (normally hardware not
10945 * present) and leaves us with a valid list of present and active devices.
10950 * This is called single threaded during boot, so no need
10951 * to take the rtnl semaphore.
10953 static int __init net_dev_init(void)
10955 int i, rc = -ENOMEM;
10957 BUG_ON(!dev_boot_phase);
10959 if (dev_proc_init())
10962 if (netdev_kobject_init())
10965 INIT_LIST_HEAD(&ptype_all);
10966 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10967 INIT_LIST_HEAD(&ptype_base[i]);
10969 if (register_pernet_subsys(&netdev_net_ops))
10973 * Initialise the packet receive queues.
10976 for_each_possible_cpu(i) {
10977 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10978 struct softnet_data *sd = &per_cpu(softnet_data, i);
10980 INIT_WORK(flush, flush_backlog);
10982 skb_queue_head_init(&sd->input_pkt_queue);
10983 skb_queue_head_init(&sd->process_queue);
10984 #ifdef CONFIG_XFRM_OFFLOAD
10985 skb_queue_head_init(&sd->xfrm_backlog);
10987 INIT_LIST_HEAD(&sd->poll_list);
10988 sd->output_queue_tailp = &sd->output_queue;
10990 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
10994 init_gro_hash(&sd->backlog);
10995 sd->backlog.poll = process_backlog;
10996 sd->backlog.weight = weight_p;
10999 dev_boot_phase = 0;
11001 /* The loopback device is special if any other network devices
11002 * is present in a network namespace the loopback device must
11003 * be present. Since we now dynamically allocate and free the
11004 * loopback device ensure this invariant is maintained by
11005 * keeping the loopback device as the first device on the
11006 * list of network devices. Ensuring the loopback devices
11007 * is the first device that appears and the last network device
11010 if (register_pernet_device(&loopback_net_ops))
11013 if (register_pernet_device(&default_device_ops))
11016 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11017 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11019 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11020 NULL, dev_cpu_dead);
11027 subsys_initcall(net_dev_init);