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_ingress.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"
156 #define MAX_GRO_SKBS 8
158 /* This should be increased if a protocol with a bigger head is added. */
159 #define GRO_MAX_HEAD (MAX_HEADER + 128)
161 static DEFINE_SPINLOCK(ptype_lock);
162 static DEFINE_SPINLOCK(offload_lock);
163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
164 struct list_head ptype_all __read_mostly; /* Taps */
165 static struct list_head offload_base __read_mostly;
167 static int netif_rx_internal(struct sk_buff *skb);
168 static int call_netdevice_notifiers_info(unsigned long val,
169 struct netdev_notifier_info *info);
170 static int call_netdevice_notifiers_extack(unsigned long val,
171 struct net_device *dev,
172 struct netlink_ext_ack *extack);
173 static struct napi_struct *napi_by_id(unsigned int napi_id);
176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base_head list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 static DEFINE_MUTEX(ifalias_mutex);
199 /* protects napi_hash addition/deletion and napi_gen_id */
200 static DEFINE_SPINLOCK(napi_hash_lock);
202 static unsigned int napi_gen_id = NR_CPUS;
203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
205 static DECLARE_RWSEM(devnet_rename_sem);
207 static inline void dev_base_seq_inc(struct net *net)
209 while (++net->dev_base_seq == 0)
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
225 static inline void rps_lock(struct softnet_data *sd)
228 spin_lock(&sd->input_pkt_queue.lock);
232 static inline void rps_unlock(struct softnet_data *sd)
235 spin_unlock(&sd->input_pkt_queue.lock);
239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
242 struct netdev_name_node *name_node;
244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
247 INIT_HLIST_NODE(&name_node->hlist);
248 name_node->dev = dev;
249 name_node->name = name;
253 static struct netdev_name_node *
254 netdev_name_node_head_alloc(struct net_device *dev)
256 struct netdev_name_node *name_node;
258 name_node = netdev_name_node_alloc(dev, dev->name);
261 INIT_LIST_HEAD(&name_node->list);
265 static void netdev_name_node_free(struct netdev_name_node *name_node)
270 static void netdev_name_node_add(struct net *net,
271 struct netdev_name_node *name_node)
273 hlist_add_head_rcu(&name_node->hlist,
274 dev_name_hash(net, name_node->name));
277 static void netdev_name_node_del(struct netdev_name_node *name_node)
279 hlist_del_rcu(&name_node->hlist);
282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
285 struct hlist_head *head = dev_name_hash(net, name);
286 struct netdev_name_node *name_node;
288 hlist_for_each_entry(name_node, head, hlist)
289 if (!strcmp(name_node->name, name))
294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
297 struct hlist_head *head = dev_name_hash(net, name);
298 struct netdev_name_node *name_node;
300 hlist_for_each_entry_rcu(name_node, head, hlist)
301 if (!strcmp(name_node->name, name))
306 bool netdev_name_in_use(struct net *net, const char *name)
308 return netdev_name_node_lookup(net, name);
310 EXPORT_SYMBOL(netdev_name_in_use);
312 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
314 struct netdev_name_node *name_node;
315 struct net *net = dev_net(dev);
317 name_node = netdev_name_node_lookup(net, name);
320 name_node = netdev_name_node_alloc(dev, name);
323 netdev_name_node_add(net, name_node);
324 /* The node that holds dev->name acts as a head of per-device list. */
325 list_add_tail(&name_node->list, &dev->name_node->list);
329 EXPORT_SYMBOL(netdev_name_node_alt_create);
331 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
333 list_del(&name_node->list);
334 netdev_name_node_del(name_node);
335 kfree(name_node->name);
336 netdev_name_node_free(name_node);
339 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
341 struct netdev_name_node *name_node;
342 struct net *net = dev_net(dev);
344 name_node = netdev_name_node_lookup(net, name);
347 /* lookup might have found our primary name or a name belonging
350 if (name_node == dev->name_node || name_node->dev != dev)
353 __netdev_name_node_alt_destroy(name_node);
357 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
359 static void netdev_name_node_alt_flush(struct net_device *dev)
361 struct netdev_name_node *name_node, *tmp;
363 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
364 __netdev_name_node_alt_destroy(name_node);
367 /* Device list insertion */
368 static void list_netdevice(struct net_device *dev)
370 struct net *net = dev_net(dev);
374 write_lock_bh(&dev_base_lock);
375 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
376 netdev_name_node_add(net, dev->name_node);
377 hlist_add_head_rcu(&dev->index_hlist,
378 dev_index_hash(net, dev->ifindex));
379 write_unlock_bh(&dev_base_lock);
381 dev_base_seq_inc(net);
384 /* Device list removal
385 * caller must respect a RCU grace period before freeing/reusing dev
387 static void unlist_netdevice(struct net_device *dev)
391 /* Unlink dev from the device chain */
392 write_lock_bh(&dev_base_lock);
393 list_del_rcu(&dev->dev_list);
394 netdev_name_node_del(dev->name_node);
395 hlist_del_rcu(&dev->index_hlist);
396 write_unlock_bh(&dev_base_lock);
398 dev_base_seq_inc(dev_net(dev));
405 static RAW_NOTIFIER_HEAD(netdev_chain);
408 * Device drivers call our routines to queue packets here. We empty the
409 * queue in the local softnet handler.
412 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
413 EXPORT_PER_CPU_SYMBOL(softnet_data);
415 #ifdef CONFIG_LOCKDEP
417 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
418 * according to dev->type
420 static const unsigned short netdev_lock_type[] = {
421 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
422 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
423 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
424 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
425 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
426 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
427 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
428 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
429 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
430 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
431 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
432 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
433 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
434 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
435 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
437 static const char *const netdev_lock_name[] = {
438 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
439 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
440 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
441 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
442 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
443 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
444 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
445 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
446 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
447 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
448 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
449 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
450 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
451 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
452 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
454 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
455 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
457 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
461 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
462 if (netdev_lock_type[i] == dev_type)
464 /* the last key is used by default */
465 return ARRAY_SIZE(netdev_lock_type) - 1;
468 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
469 unsigned short dev_type)
473 i = netdev_lock_pos(dev_type);
474 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
475 netdev_lock_name[i]);
478 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
482 i = netdev_lock_pos(dev->type);
483 lockdep_set_class_and_name(&dev->addr_list_lock,
484 &netdev_addr_lock_key[i],
485 netdev_lock_name[i]);
488 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
489 unsigned short dev_type)
493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
498 /*******************************************************************************
500 * Protocol management and registration routines
502 *******************************************************************************/
506 * Add a protocol ID to the list. Now that the input handler is
507 * smarter we can dispense with all the messy stuff that used to be
510 * BEWARE!!! Protocol handlers, mangling input packets,
511 * MUST BE last in hash buckets and checking protocol handlers
512 * MUST start from promiscuous ptype_all chain in net_bh.
513 * It is true now, do not change it.
514 * Explanation follows: if protocol handler, mangling packet, will
515 * be the first on list, it is not able to sense, that packet
516 * is cloned and should be copied-on-write, so that it will
517 * change it and subsequent readers will get broken packet.
521 static inline struct list_head *ptype_head(const struct packet_type *pt)
523 if (pt->type == htons(ETH_P_ALL))
524 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
526 return pt->dev ? &pt->dev->ptype_specific :
527 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
531 * dev_add_pack - add packet handler
532 * @pt: packet type declaration
534 * Add a protocol handler to the networking stack. The passed &packet_type
535 * is linked into kernel lists and may not be freed until it has been
536 * removed from the kernel lists.
538 * This call does not sleep therefore it can not
539 * guarantee all CPU's that are in middle of receiving packets
540 * will see the new packet type (until the next received packet).
543 void dev_add_pack(struct packet_type *pt)
545 struct list_head *head = ptype_head(pt);
547 spin_lock(&ptype_lock);
548 list_add_rcu(&pt->list, head);
549 spin_unlock(&ptype_lock);
551 EXPORT_SYMBOL(dev_add_pack);
554 * __dev_remove_pack - remove packet handler
555 * @pt: packet type declaration
557 * Remove a protocol handler that was previously added to the kernel
558 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
559 * from the kernel lists and can be freed or reused once this function
562 * The packet type might still be in use by receivers
563 * and must not be freed until after all the CPU's have gone
564 * through a quiescent state.
566 void __dev_remove_pack(struct packet_type *pt)
568 struct list_head *head = ptype_head(pt);
569 struct packet_type *pt1;
571 spin_lock(&ptype_lock);
573 list_for_each_entry(pt1, head, list) {
575 list_del_rcu(&pt->list);
580 pr_warn("dev_remove_pack: %p not found\n", pt);
582 spin_unlock(&ptype_lock);
584 EXPORT_SYMBOL(__dev_remove_pack);
587 * dev_remove_pack - remove packet handler
588 * @pt: packet type declaration
590 * Remove a protocol handler that was previously added to the kernel
591 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
592 * from the kernel lists and can be freed or reused once this function
595 * This call sleeps to guarantee that no CPU is looking at the packet
598 void dev_remove_pack(struct packet_type *pt)
600 __dev_remove_pack(pt);
604 EXPORT_SYMBOL(dev_remove_pack);
608 * dev_add_offload - register offload handlers
609 * @po: protocol offload declaration
611 * Add protocol offload handlers to the networking stack. The passed
612 * &proto_offload is linked into kernel lists and may not be freed until
613 * it has been removed from the kernel lists.
615 * This call does not sleep therefore it can not
616 * guarantee all CPU's that are in middle of receiving packets
617 * will see the new offload handlers (until the next received packet).
619 void dev_add_offload(struct packet_offload *po)
621 struct packet_offload *elem;
623 spin_lock(&offload_lock);
624 list_for_each_entry(elem, &offload_base, list) {
625 if (po->priority < elem->priority)
628 list_add_rcu(&po->list, elem->list.prev);
629 spin_unlock(&offload_lock);
631 EXPORT_SYMBOL(dev_add_offload);
634 * __dev_remove_offload - remove offload handler
635 * @po: packet offload declaration
637 * Remove a protocol offload handler that was previously added to the
638 * kernel offload handlers by dev_add_offload(). The passed &offload_type
639 * is removed from the kernel lists and can be freed or reused once this
642 * The packet type might still be in use by receivers
643 * and must not be freed until after all the CPU's have gone
644 * through a quiescent state.
646 static void __dev_remove_offload(struct packet_offload *po)
648 struct list_head *head = &offload_base;
649 struct packet_offload *po1;
651 spin_lock(&offload_lock);
653 list_for_each_entry(po1, head, list) {
655 list_del_rcu(&po->list);
660 pr_warn("dev_remove_offload: %p not found\n", po);
662 spin_unlock(&offload_lock);
666 * dev_remove_offload - remove packet offload handler
667 * @po: packet offload declaration
669 * Remove a packet offload handler that was previously added to the kernel
670 * offload handlers by dev_add_offload(). The passed &offload_type is
671 * removed from the kernel lists and can be freed or reused once this
674 * This call sleeps to guarantee that no CPU is looking at the packet
677 void dev_remove_offload(struct packet_offload *po)
679 __dev_remove_offload(po);
683 EXPORT_SYMBOL(dev_remove_offload);
685 /*******************************************************************************
687 * Device Interface Subroutines
689 *******************************************************************************/
692 * dev_get_iflink - get 'iflink' value of a interface
693 * @dev: targeted interface
695 * Indicates the ifindex the interface is linked to.
696 * Physical interfaces have the same 'ifindex' and 'iflink' values.
699 int dev_get_iflink(const struct net_device *dev)
701 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
702 return dev->netdev_ops->ndo_get_iflink(dev);
706 EXPORT_SYMBOL(dev_get_iflink);
709 * dev_fill_metadata_dst - Retrieve tunnel egress information.
710 * @dev: targeted interface
713 * For better visibility of tunnel traffic OVS needs to retrieve
714 * egress tunnel information for a packet. Following API allows
715 * user to get this info.
717 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
719 struct ip_tunnel_info *info;
721 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
724 info = skb_tunnel_info_unclone(skb);
727 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
730 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
732 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
734 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
736 int k = stack->num_paths++;
738 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
741 return &stack->path[k];
744 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
745 struct net_device_path_stack *stack)
747 const struct net_device *last_dev;
748 struct net_device_path_ctx ctx = {
752 struct net_device_path *path;
755 stack->num_paths = 0;
756 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
758 path = dev_fwd_path(stack);
762 memset(path, 0, sizeof(struct net_device_path));
763 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
767 if (WARN_ON_ONCE(last_dev == ctx.dev))
770 path = dev_fwd_path(stack);
773 path->type = DEV_PATH_ETHERNET;
778 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
781 * __dev_get_by_name - find a device by its name
782 * @net: the applicable net namespace
783 * @name: name to find
785 * Find an interface by name. Must be called under RTNL semaphore
786 * or @dev_base_lock. If the name is found a pointer to the device
787 * is returned. If the name is not found then %NULL is returned. The
788 * reference counters are not incremented so the caller must be
789 * careful with locks.
792 struct net_device *__dev_get_by_name(struct net *net, const char *name)
794 struct netdev_name_node *node_name;
796 node_name = netdev_name_node_lookup(net, name);
797 return node_name ? node_name->dev : NULL;
799 EXPORT_SYMBOL(__dev_get_by_name);
802 * dev_get_by_name_rcu - find a device by its name
803 * @net: the applicable net namespace
804 * @name: name to find
806 * Find an interface by name.
807 * If the name is found a pointer to the device is returned.
808 * If the name is not found then %NULL is returned.
809 * The reference counters are not incremented so the caller must be
810 * careful with locks. The caller must hold RCU lock.
813 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
815 struct netdev_name_node *node_name;
817 node_name = netdev_name_node_lookup_rcu(net, name);
818 return node_name ? node_name->dev : NULL;
820 EXPORT_SYMBOL(dev_get_by_name_rcu);
823 * dev_get_by_name - find a device by its name
824 * @net: the applicable net namespace
825 * @name: name to find
827 * Find an interface by name. This can be called from any
828 * context and does its own locking. The returned handle has
829 * the usage count incremented and the caller must use dev_put() to
830 * release it when it is no longer needed. %NULL is returned if no
831 * matching device is found.
834 struct net_device *dev_get_by_name(struct net *net, const char *name)
836 struct net_device *dev;
839 dev = dev_get_by_name_rcu(net, name);
844 EXPORT_SYMBOL(dev_get_by_name);
847 * __dev_get_by_index - find a device by its ifindex
848 * @net: the applicable net namespace
849 * @ifindex: index of device
851 * Search for an interface by index. Returns %NULL if the device
852 * is not found or a pointer to the device. The device has not
853 * had its reference counter increased so the caller must be careful
854 * about locking. The caller must hold either the RTNL semaphore
858 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
860 struct net_device *dev;
861 struct hlist_head *head = dev_index_hash(net, ifindex);
863 hlist_for_each_entry(dev, head, index_hlist)
864 if (dev->ifindex == ifindex)
869 EXPORT_SYMBOL(__dev_get_by_index);
872 * dev_get_by_index_rcu - find a device by its ifindex
873 * @net: the applicable net namespace
874 * @ifindex: index of device
876 * Search for an interface by index. Returns %NULL if the device
877 * is not found or a pointer to the device. The device has not
878 * had its reference counter increased so the caller must be careful
879 * about locking. The caller must hold RCU lock.
882 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
884 struct net_device *dev;
885 struct hlist_head *head = dev_index_hash(net, ifindex);
887 hlist_for_each_entry_rcu(dev, head, index_hlist)
888 if (dev->ifindex == ifindex)
893 EXPORT_SYMBOL(dev_get_by_index_rcu);
897 * dev_get_by_index - find a device by its ifindex
898 * @net: the applicable net namespace
899 * @ifindex: index of device
901 * Search for an interface by index. Returns NULL if the device
902 * is not found or a pointer to the device. The device returned has
903 * had a reference added and the pointer is safe until the user calls
904 * dev_put to indicate they have finished with it.
907 struct net_device *dev_get_by_index(struct net *net, int ifindex)
909 struct net_device *dev;
912 dev = dev_get_by_index_rcu(net, ifindex);
917 EXPORT_SYMBOL(dev_get_by_index);
920 * dev_get_by_napi_id - find a device by napi_id
921 * @napi_id: ID of the NAPI struct
923 * Search for an interface by NAPI ID. Returns %NULL if the device
924 * is not found or a pointer to the device. The device has not had
925 * its reference counter increased so the caller must be careful
926 * about locking. The caller must hold RCU lock.
929 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
931 struct napi_struct *napi;
933 WARN_ON_ONCE(!rcu_read_lock_held());
935 if (napi_id < MIN_NAPI_ID)
938 napi = napi_by_id(napi_id);
940 return napi ? napi->dev : NULL;
942 EXPORT_SYMBOL(dev_get_by_napi_id);
945 * netdev_get_name - get a netdevice name, knowing its ifindex.
946 * @net: network namespace
947 * @name: a pointer to the buffer where the name will be stored.
948 * @ifindex: the ifindex of the interface to get the name from.
950 int netdev_get_name(struct net *net, char *name, int ifindex)
952 struct net_device *dev;
955 down_read(&devnet_rename_sem);
958 dev = dev_get_by_index_rcu(net, ifindex);
964 strcpy(name, dev->name);
969 up_read(&devnet_rename_sem);
974 * dev_getbyhwaddr_rcu - find a device by its hardware address
975 * @net: the applicable net namespace
976 * @type: media type of device
977 * @ha: hardware address
979 * Search for an interface by MAC address. Returns NULL if the device
980 * is not found or a pointer to the device.
981 * The caller must hold RCU or RTNL.
982 * The returned device has not had its ref count increased
983 * and the caller must therefore be careful about locking
987 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
990 struct net_device *dev;
992 for_each_netdev_rcu(net, dev)
993 if (dev->type == type &&
994 !memcmp(dev->dev_addr, ha, dev->addr_len))
999 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1001 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1003 struct net_device *dev, *ret = NULL;
1006 for_each_netdev_rcu(net, dev)
1007 if (dev->type == type) {
1015 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1018 * __dev_get_by_flags - find any device with given flags
1019 * @net: the applicable net namespace
1020 * @if_flags: IFF_* values
1021 * @mask: bitmask of bits in if_flags to check
1023 * Search for any interface with the given flags. Returns NULL if a device
1024 * is not found or a pointer to the device. Must be called inside
1025 * rtnl_lock(), and result refcount is unchanged.
1028 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1029 unsigned short mask)
1031 struct net_device *dev, *ret;
1036 for_each_netdev(net, dev) {
1037 if (((dev->flags ^ if_flags) & mask) == 0) {
1044 EXPORT_SYMBOL(__dev_get_by_flags);
1047 * dev_valid_name - check if name is okay for network device
1048 * @name: name string
1050 * Network device names need to be valid file names to
1051 * allow sysfs to work. We also disallow any kind of
1054 bool dev_valid_name(const char *name)
1058 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1060 if (!strcmp(name, ".") || !strcmp(name, ".."))
1064 if (*name == '/' || *name == ':' || isspace(*name))
1070 EXPORT_SYMBOL(dev_valid_name);
1073 * __dev_alloc_name - allocate a name for a device
1074 * @net: network namespace to allocate the device name in
1075 * @name: name format string
1076 * @buf: scratch buffer and result name string
1078 * Passed a format string - eg "lt%d" it will try and find a suitable
1079 * id. It scans list of devices to build up a free map, then chooses
1080 * the first empty slot. The caller must hold the dev_base or rtnl lock
1081 * while allocating the name and adding the device in order to avoid
1083 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1084 * Returns the number of the unit assigned or a negative errno code.
1087 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1091 const int max_netdevices = 8*PAGE_SIZE;
1092 unsigned long *inuse;
1093 struct net_device *d;
1095 if (!dev_valid_name(name))
1098 p = strchr(name, '%');
1101 * Verify the string as this thing may have come from
1102 * the user. There must be either one "%d" and no other "%"
1105 if (p[1] != 'd' || strchr(p + 2, '%'))
1108 /* Use one page as a bit array of possible slots */
1109 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1113 for_each_netdev(net, d) {
1114 struct netdev_name_node *name_node;
1115 list_for_each_entry(name_node, &d->name_node->list, list) {
1116 if (!sscanf(name_node->name, name, &i))
1118 if (i < 0 || i >= max_netdevices)
1121 /* avoid cases where sscanf is not exact inverse of printf */
1122 snprintf(buf, IFNAMSIZ, name, i);
1123 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1126 if (!sscanf(d->name, name, &i))
1128 if (i < 0 || i >= max_netdevices)
1131 /* avoid cases where sscanf is not exact inverse of printf */
1132 snprintf(buf, IFNAMSIZ, name, i);
1133 if (!strncmp(buf, d->name, IFNAMSIZ))
1137 i = find_first_zero_bit(inuse, max_netdevices);
1138 free_page((unsigned long) inuse);
1141 snprintf(buf, IFNAMSIZ, name, i);
1142 if (!netdev_name_in_use(net, buf))
1145 /* It is possible to run out of possible slots
1146 * when the name is long and there isn't enough space left
1147 * for the digits, or if all bits are used.
1152 static int dev_alloc_name_ns(struct net *net,
1153 struct net_device *dev,
1160 ret = __dev_alloc_name(net, name, buf);
1162 strlcpy(dev->name, buf, IFNAMSIZ);
1167 * dev_alloc_name - allocate a name for a device
1169 * @name: name format string
1171 * Passed a format string - eg "lt%d" it will try and find a suitable
1172 * id. It scans list of devices to build up a free map, then chooses
1173 * the first empty slot. The caller must hold the dev_base or rtnl lock
1174 * while allocating the name and adding the device in order to avoid
1176 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1177 * Returns the number of the unit assigned or a negative errno code.
1180 int dev_alloc_name(struct net_device *dev, const char *name)
1182 return dev_alloc_name_ns(dev_net(dev), dev, name);
1184 EXPORT_SYMBOL(dev_alloc_name);
1186 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1191 if (!dev_valid_name(name))
1194 if (strchr(name, '%'))
1195 return dev_alloc_name_ns(net, dev, name);
1196 else if (netdev_name_in_use(net, name))
1198 else if (dev->name != name)
1199 strlcpy(dev->name, name, IFNAMSIZ);
1205 * dev_change_name - change name of a device
1207 * @newname: name (or format string) must be at least IFNAMSIZ
1209 * Change name of a device, can pass format strings "eth%d".
1212 int dev_change_name(struct net_device *dev, const char *newname)
1214 unsigned char old_assign_type;
1215 char oldname[IFNAMSIZ];
1221 BUG_ON(!dev_net(dev));
1225 /* Some auto-enslaved devices e.g. failover slaves are
1226 * special, as userspace might rename the device after
1227 * the interface had been brought up and running since
1228 * the point kernel initiated auto-enslavement. Allow
1229 * live name change even when these slave devices are
1232 * Typically, users of these auto-enslaving devices
1233 * don't actually care about slave name change, as
1234 * they are supposed to operate on master interface
1237 if (dev->flags & IFF_UP &&
1238 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1241 down_write(&devnet_rename_sem);
1243 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1244 up_write(&devnet_rename_sem);
1248 memcpy(oldname, dev->name, IFNAMSIZ);
1250 err = dev_get_valid_name(net, dev, newname);
1252 up_write(&devnet_rename_sem);
1256 if (oldname[0] && !strchr(oldname, '%'))
1257 netdev_info(dev, "renamed from %s\n", oldname);
1259 old_assign_type = dev->name_assign_type;
1260 dev->name_assign_type = NET_NAME_RENAMED;
1263 ret = device_rename(&dev->dev, dev->name);
1265 memcpy(dev->name, oldname, IFNAMSIZ);
1266 dev->name_assign_type = old_assign_type;
1267 up_write(&devnet_rename_sem);
1271 up_write(&devnet_rename_sem);
1273 netdev_adjacent_rename_links(dev, oldname);
1275 write_lock_bh(&dev_base_lock);
1276 netdev_name_node_del(dev->name_node);
1277 write_unlock_bh(&dev_base_lock);
1281 write_lock_bh(&dev_base_lock);
1282 netdev_name_node_add(net, dev->name_node);
1283 write_unlock_bh(&dev_base_lock);
1285 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1286 ret = notifier_to_errno(ret);
1289 /* err >= 0 after dev_alloc_name() or stores the first errno */
1292 down_write(&devnet_rename_sem);
1293 memcpy(dev->name, oldname, IFNAMSIZ);
1294 memcpy(oldname, newname, IFNAMSIZ);
1295 dev->name_assign_type = old_assign_type;
1296 old_assign_type = NET_NAME_RENAMED;
1299 pr_err("%s: name change rollback failed: %d\n",
1308 * dev_set_alias - change ifalias of a device
1310 * @alias: name up to IFALIASZ
1311 * @len: limit of bytes to copy from info
1313 * Set ifalias for a device,
1315 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1317 struct dev_ifalias *new_alias = NULL;
1319 if (len >= IFALIASZ)
1323 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1327 memcpy(new_alias->ifalias, alias, len);
1328 new_alias->ifalias[len] = 0;
1331 mutex_lock(&ifalias_mutex);
1332 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1333 mutex_is_locked(&ifalias_mutex));
1334 mutex_unlock(&ifalias_mutex);
1337 kfree_rcu(new_alias, rcuhead);
1341 EXPORT_SYMBOL(dev_set_alias);
1344 * dev_get_alias - get ifalias of a device
1346 * @name: buffer to store name of ifalias
1347 * @len: size of buffer
1349 * get ifalias for a device. Caller must make sure dev cannot go
1350 * away, e.g. rcu read lock or own a reference count to device.
1352 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1354 const struct dev_ifalias *alias;
1358 alias = rcu_dereference(dev->ifalias);
1360 ret = snprintf(name, len, "%s", alias->ifalias);
1367 * netdev_features_change - device changes features
1368 * @dev: device to cause notification
1370 * Called to indicate a device has changed features.
1372 void netdev_features_change(struct net_device *dev)
1374 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1376 EXPORT_SYMBOL(netdev_features_change);
1379 * netdev_state_change - device changes state
1380 * @dev: device to cause notification
1382 * Called to indicate a device has changed state. This function calls
1383 * the notifier chains for netdev_chain and sends a NEWLINK message
1384 * to the routing socket.
1386 void netdev_state_change(struct net_device *dev)
1388 if (dev->flags & IFF_UP) {
1389 struct netdev_notifier_change_info change_info = {
1393 call_netdevice_notifiers_info(NETDEV_CHANGE,
1395 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1398 EXPORT_SYMBOL(netdev_state_change);
1401 * __netdev_notify_peers - notify network peers about existence of @dev,
1402 * to be called when rtnl lock is already held.
1403 * @dev: network device
1405 * Generate traffic such that interested network peers are aware of
1406 * @dev, such as by generating a gratuitous ARP. This may be used when
1407 * a device wants to inform the rest of the network about some sort of
1408 * reconfiguration such as a failover event or virtual machine
1411 void __netdev_notify_peers(struct net_device *dev)
1414 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1415 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1417 EXPORT_SYMBOL(__netdev_notify_peers);
1420 * netdev_notify_peers - notify network peers about existence of @dev
1421 * @dev: network device
1423 * Generate traffic such that interested network peers are aware of
1424 * @dev, such as by generating a gratuitous ARP. This may be used when
1425 * a device wants to inform the rest of the network about some sort of
1426 * reconfiguration such as a failover event or virtual machine
1429 void netdev_notify_peers(struct net_device *dev)
1432 __netdev_notify_peers(dev);
1435 EXPORT_SYMBOL(netdev_notify_peers);
1437 static int napi_threaded_poll(void *data);
1439 static int napi_kthread_create(struct napi_struct *n)
1443 /* Create and wake up the kthread once to put it in
1444 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1445 * warning and work with loadavg.
1447 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1448 n->dev->name, n->napi_id);
1449 if (IS_ERR(n->thread)) {
1450 err = PTR_ERR(n->thread);
1451 pr_err("kthread_run failed with err %d\n", err);
1458 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1460 const struct net_device_ops *ops = dev->netdev_ops;
1465 if (!netif_device_present(dev)) {
1466 /* may be detached because parent is runtime-suspended */
1467 if (dev->dev.parent)
1468 pm_runtime_resume(dev->dev.parent);
1469 if (!netif_device_present(dev))
1473 /* Block netpoll from trying to do any rx path servicing.
1474 * If we don't do this there is a chance ndo_poll_controller
1475 * or ndo_poll may be running while we open the device
1477 netpoll_poll_disable(dev);
1479 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1480 ret = notifier_to_errno(ret);
1484 set_bit(__LINK_STATE_START, &dev->state);
1486 if (ops->ndo_validate_addr)
1487 ret = ops->ndo_validate_addr(dev);
1489 if (!ret && ops->ndo_open)
1490 ret = ops->ndo_open(dev);
1492 netpoll_poll_enable(dev);
1495 clear_bit(__LINK_STATE_START, &dev->state);
1497 dev->flags |= IFF_UP;
1498 dev_set_rx_mode(dev);
1500 add_device_randomness(dev->dev_addr, dev->addr_len);
1507 * dev_open - prepare an interface for use.
1508 * @dev: device to open
1509 * @extack: netlink extended ack
1511 * Takes a device from down to up state. The device's private open
1512 * function is invoked and then the multicast lists are loaded. Finally
1513 * the device is moved into the up state and a %NETDEV_UP message is
1514 * sent to the netdev notifier chain.
1516 * Calling this function on an active interface is a nop. On a failure
1517 * a negative errno code is returned.
1519 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1523 if (dev->flags & IFF_UP)
1526 ret = __dev_open(dev, extack);
1530 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1531 call_netdevice_notifiers(NETDEV_UP, dev);
1535 EXPORT_SYMBOL(dev_open);
1537 static void __dev_close_many(struct list_head *head)
1539 struct net_device *dev;
1544 list_for_each_entry(dev, head, close_list) {
1545 /* Temporarily disable netpoll until the interface is down */
1546 netpoll_poll_disable(dev);
1548 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1550 clear_bit(__LINK_STATE_START, &dev->state);
1552 /* Synchronize to scheduled poll. We cannot touch poll list, it
1553 * can be even on different cpu. So just clear netif_running().
1555 * dev->stop() will invoke napi_disable() on all of it's
1556 * napi_struct instances on this device.
1558 smp_mb__after_atomic(); /* Commit netif_running(). */
1561 dev_deactivate_many(head);
1563 list_for_each_entry(dev, head, close_list) {
1564 const struct net_device_ops *ops = dev->netdev_ops;
1567 * Call the device specific close. This cannot fail.
1568 * Only if device is UP
1570 * We allow it to be called even after a DETACH hot-plug
1576 dev->flags &= ~IFF_UP;
1577 netpoll_poll_enable(dev);
1581 static void __dev_close(struct net_device *dev)
1585 list_add(&dev->close_list, &single);
1586 __dev_close_many(&single);
1590 void dev_close_many(struct list_head *head, bool unlink)
1592 struct net_device *dev, *tmp;
1594 /* Remove the devices that don't need to be closed */
1595 list_for_each_entry_safe(dev, tmp, head, close_list)
1596 if (!(dev->flags & IFF_UP))
1597 list_del_init(&dev->close_list);
1599 __dev_close_many(head);
1601 list_for_each_entry_safe(dev, tmp, head, close_list) {
1602 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1603 call_netdevice_notifiers(NETDEV_DOWN, dev);
1605 list_del_init(&dev->close_list);
1608 EXPORT_SYMBOL(dev_close_many);
1611 * dev_close - shutdown an interface.
1612 * @dev: device to shutdown
1614 * This function moves an active device into down state. A
1615 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1616 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1619 void dev_close(struct net_device *dev)
1621 if (dev->flags & IFF_UP) {
1624 list_add(&dev->close_list, &single);
1625 dev_close_many(&single, true);
1629 EXPORT_SYMBOL(dev_close);
1633 * dev_disable_lro - disable Large Receive Offload on a device
1636 * Disable Large Receive Offload (LRO) on a net device. Must be
1637 * called under RTNL. This is needed if received packets may be
1638 * forwarded to another interface.
1640 void dev_disable_lro(struct net_device *dev)
1642 struct net_device *lower_dev;
1643 struct list_head *iter;
1645 dev->wanted_features &= ~NETIF_F_LRO;
1646 netdev_update_features(dev);
1648 if (unlikely(dev->features & NETIF_F_LRO))
1649 netdev_WARN(dev, "failed to disable LRO!\n");
1651 netdev_for_each_lower_dev(dev, lower_dev, iter)
1652 dev_disable_lro(lower_dev);
1654 EXPORT_SYMBOL(dev_disable_lro);
1657 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1660 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1661 * called under RTNL. This is needed if Generic XDP is installed on
1664 static void dev_disable_gro_hw(struct net_device *dev)
1666 dev->wanted_features &= ~NETIF_F_GRO_HW;
1667 netdev_update_features(dev);
1669 if (unlikely(dev->features & NETIF_F_GRO_HW))
1670 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1673 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1676 case NETDEV_##val: \
1677 return "NETDEV_" __stringify(val);
1679 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1680 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1681 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1682 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1683 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1684 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1685 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1686 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1687 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1691 return "UNKNOWN_NETDEV_EVENT";
1693 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1695 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1696 struct net_device *dev)
1698 struct netdev_notifier_info info = {
1702 return nb->notifier_call(nb, val, &info);
1705 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1706 struct net_device *dev)
1710 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1711 err = notifier_to_errno(err);
1715 if (!(dev->flags & IFF_UP))
1718 call_netdevice_notifier(nb, NETDEV_UP, dev);
1722 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1723 struct net_device *dev)
1725 if (dev->flags & IFF_UP) {
1726 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1728 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1730 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1733 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1736 struct net_device *dev;
1739 for_each_netdev(net, dev) {
1740 err = call_netdevice_register_notifiers(nb, dev);
1747 for_each_netdev_continue_reverse(net, dev)
1748 call_netdevice_unregister_notifiers(nb, dev);
1752 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1755 struct net_device *dev;
1757 for_each_netdev(net, dev)
1758 call_netdevice_unregister_notifiers(nb, dev);
1761 static int dev_boot_phase = 1;
1764 * register_netdevice_notifier - register a network notifier block
1767 * Register a notifier to be called when network device events occur.
1768 * The notifier passed is linked into the kernel structures and must
1769 * not be reused until it has been unregistered. A negative errno code
1770 * is returned on a failure.
1772 * When registered all registration and up events are replayed
1773 * to the new notifier to allow device to have a race free
1774 * view of the network device list.
1777 int register_netdevice_notifier(struct notifier_block *nb)
1782 /* Close race with setup_net() and cleanup_net() */
1783 down_write(&pernet_ops_rwsem);
1785 err = raw_notifier_chain_register(&netdev_chain, nb);
1791 err = call_netdevice_register_net_notifiers(nb, net);
1798 up_write(&pernet_ops_rwsem);
1802 for_each_net_continue_reverse(net)
1803 call_netdevice_unregister_net_notifiers(nb, net);
1805 raw_notifier_chain_unregister(&netdev_chain, nb);
1808 EXPORT_SYMBOL(register_netdevice_notifier);
1811 * unregister_netdevice_notifier - unregister a network notifier block
1814 * Unregister a notifier previously registered by
1815 * register_netdevice_notifier(). The notifier is unlinked into the
1816 * kernel structures and may then be reused. A negative errno code
1817 * is returned on a failure.
1819 * After unregistering unregister and down device events are synthesized
1820 * for all devices on the device list to the removed notifier to remove
1821 * the need for special case cleanup code.
1824 int unregister_netdevice_notifier(struct notifier_block *nb)
1829 /* Close race with setup_net() and cleanup_net() */
1830 down_write(&pernet_ops_rwsem);
1832 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1837 call_netdevice_unregister_net_notifiers(nb, net);
1841 up_write(&pernet_ops_rwsem);
1844 EXPORT_SYMBOL(unregister_netdevice_notifier);
1846 static int __register_netdevice_notifier_net(struct net *net,
1847 struct notifier_block *nb,
1848 bool ignore_call_fail)
1852 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1858 err = call_netdevice_register_net_notifiers(nb, net);
1859 if (err && !ignore_call_fail)
1860 goto chain_unregister;
1865 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1869 static int __unregister_netdevice_notifier_net(struct net *net,
1870 struct notifier_block *nb)
1874 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1878 call_netdevice_unregister_net_notifiers(nb, net);
1883 * register_netdevice_notifier_net - register a per-netns network notifier block
1884 * @net: network namespace
1887 * Register a notifier to be called when network device events occur.
1888 * The notifier passed is linked into the kernel structures and must
1889 * not be reused until it has been unregistered. A negative errno code
1890 * is returned on a failure.
1892 * When registered all registration and up events are replayed
1893 * to the new notifier to allow device to have a race free
1894 * view of the network device list.
1897 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1902 err = __register_netdevice_notifier_net(net, nb, false);
1906 EXPORT_SYMBOL(register_netdevice_notifier_net);
1909 * unregister_netdevice_notifier_net - unregister a per-netns
1910 * network notifier block
1911 * @net: network namespace
1914 * Unregister a notifier previously registered by
1915 * register_netdevice_notifier(). The notifier is unlinked into the
1916 * kernel structures and may then be reused. A negative errno code
1917 * is returned on a failure.
1919 * After unregistering unregister and down device events are synthesized
1920 * for all devices on the device list to the removed notifier to remove
1921 * the need for special case cleanup code.
1924 int unregister_netdevice_notifier_net(struct net *net,
1925 struct notifier_block *nb)
1930 err = __unregister_netdevice_notifier_net(net, nb);
1934 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1936 int register_netdevice_notifier_dev_net(struct net_device *dev,
1937 struct notifier_block *nb,
1938 struct netdev_net_notifier *nn)
1943 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1946 list_add(&nn->list, &dev->net_notifier_list);
1951 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1953 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1954 struct notifier_block *nb,
1955 struct netdev_net_notifier *nn)
1960 list_del(&nn->list);
1961 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1965 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1967 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1970 struct netdev_net_notifier *nn;
1972 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1973 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1974 __register_netdevice_notifier_net(net, nn->nb, true);
1979 * call_netdevice_notifiers_info - call all network notifier blocks
1980 * @val: value passed unmodified to notifier function
1981 * @info: notifier information data
1983 * Call all network notifier blocks. Parameters and return value
1984 * are as for raw_notifier_call_chain().
1987 static int call_netdevice_notifiers_info(unsigned long val,
1988 struct netdev_notifier_info *info)
1990 struct net *net = dev_net(info->dev);
1995 /* Run per-netns notifier block chain first, then run the global one.
1996 * Hopefully, one day, the global one is going to be removed after
1997 * all notifier block registrators get converted to be per-netns.
1999 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2000 if (ret & NOTIFY_STOP_MASK)
2002 return raw_notifier_call_chain(&netdev_chain, val, info);
2005 static int call_netdevice_notifiers_extack(unsigned long val,
2006 struct net_device *dev,
2007 struct netlink_ext_ack *extack)
2009 struct netdev_notifier_info info = {
2014 return call_netdevice_notifiers_info(val, &info);
2018 * call_netdevice_notifiers - call all network notifier blocks
2019 * @val: value passed unmodified to notifier function
2020 * @dev: net_device pointer passed unmodified to notifier function
2022 * Call all network notifier blocks. Parameters and return value
2023 * are as for raw_notifier_call_chain().
2026 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2028 return call_netdevice_notifiers_extack(val, dev, NULL);
2030 EXPORT_SYMBOL(call_netdevice_notifiers);
2033 * call_netdevice_notifiers_mtu - call all network notifier blocks
2034 * @val: value passed unmodified to notifier function
2035 * @dev: net_device pointer passed unmodified to notifier function
2036 * @arg: additional u32 argument passed to the notifier function
2038 * Call all network notifier blocks. Parameters and return value
2039 * are as for raw_notifier_call_chain().
2041 static int call_netdevice_notifiers_mtu(unsigned long val,
2042 struct net_device *dev, u32 arg)
2044 struct netdev_notifier_info_ext info = {
2049 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2051 return call_netdevice_notifiers_info(val, &info.info);
2054 #ifdef CONFIG_NET_INGRESS
2055 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2057 void net_inc_ingress_queue(void)
2059 static_branch_inc(&ingress_needed_key);
2061 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2063 void net_dec_ingress_queue(void)
2065 static_branch_dec(&ingress_needed_key);
2067 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2070 #ifdef CONFIG_NET_EGRESS
2071 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2073 void net_inc_egress_queue(void)
2075 static_branch_inc(&egress_needed_key);
2077 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2079 void net_dec_egress_queue(void)
2081 static_branch_dec(&egress_needed_key);
2083 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2086 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2087 #ifdef CONFIG_JUMP_LABEL
2088 static atomic_t netstamp_needed_deferred;
2089 static atomic_t netstamp_wanted;
2090 static void netstamp_clear(struct work_struct *work)
2092 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2095 wanted = atomic_add_return(deferred, &netstamp_wanted);
2097 static_branch_enable(&netstamp_needed_key);
2099 static_branch_disable(&netstamp_needed_key);
2101 static DECLARE_WORK(netstamp_work, netstamp_clear);
2104 void net_enable_timestamp(void)
2106 #ifdef CONFIG_JUMP_LABEL
2110 wanted = atomic_read(&netstamp_wanted);
2113 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2116 atomic_inc(&netstamp_needed_deferred);
2117 schedule_work(&netstamp_work);
2119 static_branch_inc(&netstamp_needed_key);
2122 EXPORT_SYMBOL(net_enable_timestamp);
2124 void net_disable_timestamp(void)
2126 #ifdef CONFIG_JUMP_LABEL
2130 wanted = atomic_read(&netstamp_wanted);
2133 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2136 atomic_dec(&netstamp_needed_deferred);
2137 schedule_work(&netstamp_work);
2139 static_branch_dec(&netstamp_needed_key);
2142 EXPORT_SYMBOL(net_disable_timestamp);
2144 static inline void net_timestamp_set(struct sk_buff *skb)
2147 if (static_branch_unlikely(&netstamp_needed_key))
2148 __net_timestamp(skb);
2151 #define net_timestamp_check(COND, SKB) \
2152 if (static_branch_unlikely(&netstamp_needed_key)) { \
2153 if ((COND) && !(SKB)->tstamp) \
2154 __net_timestamp(SKB); \
2157 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2159 return __is_skb_forwardable(dev, skb, true);
2161 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2163 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2166 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2169 skb->protocol = eth_type_trans(skb, dev);
2170 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2176 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2178 return __dev_forward_skb2(dev, skb, true);
2180 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2183 * dev_forward_skb - loopback an skb to another netif
2185 * @dev: destination network device
2186 * @skb: buffer to forward
2189 * NET_RX_SUCCESS (no congestion)
2190 * NET_RX_DROP (packet was dropped, but freed)
2192 * dev_forward_skb can be used for injecting an skb from the
2193 * start_xmit function of one device into the receive queue
2194 * of another device.
2196 * The receiving device may be in another namespace, so
2197 * we have to clear all information in the skb that could
2198 * impact namespace isolation.
2200 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2202 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2204 EXPORT_SYMBOL_GPL(dev_forward_skb);
2206 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2208 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2211 static inline int deliver_skb(struct sk_buff *skb,
2212 struct packet_type *pt_prev,
2213 struct net_device *orig_dev)
2215 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2217 refcount_inc(&skb->users);
2218 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2221 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2222 struct packet_type **pt,
2223 struct net_device *orig_dev,
2225 struct list_head *ptype_list)
2227 struct packet_type *ptype, *pt_prev = *pt;
2229 list_for_each_entry_rcu(ptype, ptype_list, list) {
2230 if (ptype->type != type)
2233 deliver_skb(skb, pt_prev, orig_dev);
2239 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2241 if (!ptype->af_packet_priv || !skb->sk)
2244 if (ptype->id_match)
2245 return ptype->id_match(ptype, skb->sk);
2246 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2253 * dev_nit_active - return true if any network interface taps are in use
2255 * @dev: network device to check for the presence of taps
2257 bool dev_nit_active(struct net_device *dev)
2259 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2261 EXPORT_SYMBOL_GPL(dev_nit_active);
2264 * Support routine. Sends outgoing frames to any network
2265 * taps currently in use.
2268 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2270 struct packet_type *ptype;
2271 struct sk_buff *skb2 = NULL;
2272 struct packet_type *pt_prev = NULL;
2273 struct list_head *ptype_list = &ptype_all;
2277 list_for_each_entry_rcu(ptype, ptype_list, list) {
2278 if (ptype->ignore_outgoing)
2281 /* Never send packets back to the socket
2282 * they originated from - MvS (miquels@drinkel.ow.org)
2284 if (skb_loop_sk(ptype, skb))
2288 deliver_skb(skb2, pt_prev, skb->dev);
2293 /* need to clone skb, done only once */
2294 skb2 = skb_clone(skb, GFP_ATOMIC);
2298 net_timestamp_set(skb2);
2300 /* skb->nh should be correctly
2301 * set by sender, so that the second statement is
2302 * just protection against buggy protocols.
2304 skb_reset_mac_header(skb2);
2306 if (skb_network_header(skb2) < skb2->data ||
2307 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2308 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2309 ntohs(skb2->protocol),
2311 skb_reset_network_header(skb2);
2314 skb2->transport_header = skb2->network_header;
2315 skb2->pkt_type = PACKET_OUTGOING;
2319 if (ptype_list == &ptype_all) {
2320 ptype_list = &dev->ptype_all;
2325 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2326 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2332 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2335 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2336 * @dev: Network device
2337 * @txq: number of queues available
2339 * If real_num_tx_queues is changed the tc mappings may no longer be
2340 * valid. To resolve this verify the tc mapping remains valid and if
2341 * not NULL the mapping. With no priorities mapping to this
2342 * offset/count pair it will no longer be used. In the worst case TC0
2343 * is invalid nothing can be done so disable priority mappings. If is
2344 * expected that drivers will fix this mapping if they can before
2345 * calling netif_set_real_num_tx_queues.
2347 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2350 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2352 /* If TC0 is invalidated disable TC mapping */
2353 if (tc->offset + tc->count > txq) {
2354 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2359 /* Invalidated prio to tc mappings set to TC0 */
2360 for (i = 1; i < TC_BITMASK + 1; i++) {
2361 int q = netdev_get_prio_tc_map(dev, i);
2363 tc = &dev->tc_to_txq[q];
2364 if (tc->offset + tc->count > txq) {
2365 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2367 netdev_set_prio_tc_map(dev, i, 0);
2372 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2375 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2378 /* walk through the TCs and see if it falls into any of them */
2379 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2380 if ((txq - tc->offset) < tc->count)
2384 /* didn't find it, just return -1 to indicate no match */
2390 EXPORT_SYMBOL(netdev_txq_to_tc);
2393 static struct static_key xps_needed __read_mostly;
2394 static struct static_key xps_rxqs_needed __read_mostly;
2395 static DEFINE_MUTEX(xps_map_mutex);
2396 #define xmap_dereference(P) \
2397 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2399 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2400 struct xps_dev_maps *old_maps, int tci, u16 index)
2402 struct xps_map *map = NULL;
2406 map = xmap_dereference(dev_maps->attr_map[tci]);
2410 for (pos = map->len; pos--;) {
2411 if (map->queues[pos] != index)
2415 map->queues[pos] = map->queues[--map->len];
2420 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2421 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2422 kfree_rcu(map, rcu);
2429 static bool remove_xps_queue_cpu(struct net_device *dev,
2430 struct xps_dev_maps *dev_maps,
2431 int cpu, u16 offset, u16 count)
2433 int num_tc = dev_maps->num_tc;
2434 bool active = false;
2437 for (tci = cpu * num_tc; num_tc--; tci++) {
2440 for (i = count, j = offset; i--; j++) {
2441 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2451 static void reset_xps_maps(struct net_device *dev,
2452 struct xps_dev_maps *dev_maps,
2453 enum xps_map_type type)
2455 static_key_slow_dec_cpuslocked(&xps_needed);
2456 if (type == XPS_RXQS)
2457 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2459 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2461 kfree_rcu(dev_maps, rcu);
2464 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2465 u16 offset, u16 count)
2467 struct xps_dev_maps *dev_maps;
2468 bool active = false;
2471 dev_maps = xmap_dereference(dev->xps_maps[type]);
2475 for (j = 0; j < dev_maps->nr_ids; j++)
2476 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2478 reset_xps_maps(dev, dev_maps, type);
2480 if (type == XPS_CPUS) {
2481 for (i = offset + (count - 1); count--; i--)
2482 netdev_queue_numa_node_write(
2483 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2487 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2490 if (!static_key_false(&xps_needed))
2494 mutex_lock(&xps_map_mutex);
2496 if (static_key_false(&xps_rxqs_needed))
2497 clean_xps_maps(dev, XPS_RXQS, offset, count);
2499 clean_xps_maps(dev, XPS_CPUS, offset, count);
2501 mutex_unlock(&xps_map_mutex);
2505 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2507 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2510 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2511 u16 index, bool is_rxqs_map)
2513 struct xps_map *new_map;
2514 int alloc_len = XPS_MIN_MAP_ALLOC;
2517 for (pos = 0; map && pos < map->len; pos++) {
2518 if (map->queues[pos] != index)
2523 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2525 if (pos < map->alloc_len)
2528 alloc_len = map->alloc_len * 2;
2531 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2535 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2537 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2538 cpu_to_node(attr_index));
2542 for (i = 0; i < pos; i++)
2543 new_map->queues[i] = map->queues[i];
2544 new_map->alloc_len = alloc_len;
2550 /* Copy xps maps at a given index */
2551 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2552 struct xps_dev_maps *new_dev_maps, int index,
2553 int tc, bool skip_tc)
2555 int i, tci = index * dev_maps->num_tc;
2556 struct xps_map *map;
2558 /* copy maps belonging to foreign traffic classes */
2559 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2560 if (i == tc && skip_tc)
2563 /* fill in the new device map from the old device map */
2564 map = xmap_dereference(dev_maps->attr_map[tci]);
2565 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2569 /* Must be called under cpus_read_lock */
2570 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2571 u16 index, enum xps_map_type type)
2573 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2574 const unsigned long *online_mask = NULL;
2575 bool active = false, copy = false;
2576 int i, j, tci, numa_node_id = -2;
2577 int maps_sz, num_tc = 1, tc = 0;
2578 struct xps_map *map, *new_map;
2579 unsigned int nr_ids;
2582 /* Do not allow XPS on subordinate device directly */
2583 num_tc = dev->num_tc;
2587 /* If queue belongs to subordinate dev use its map */
2588 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2590 tc = netdev_txq_to_tc(dev, index);
2595 mutex_lock(&xps_map_mutex);
2597 dev_maps = xmap_dereference(dev->xps_maps[type]);
2598 if (type == XPS_RXQS) {
2599 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2600 nr_ids = dev->num_rx_queues;
2602 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2603 if (num_possible_cpus() > 1)
2604 online_mask = cpumask_bits(cpu_online_mask);
2605 nr_ids = nr_cpu_ids;
2608 if (maps_sz < L1_CACHE_BYTES)
2609 maps_sz = L1_CACHE_BYTES;
2611 /* The old dev_maps could be larger or smaller than the one we're
2612 * setting up now, as dev->num_tc or nr_ids could have been updated in
2613 * between. We could try to be smart, but let's be safe instead and only
2614 * copy foreign traffic classes if the two map sizes match.
2617 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2620 /* allocate memory for queue storage */
2621 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2623 if (!new_dev_maps) {
2624 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2625 if (!new_dev_maps) {
2626 mutex_unlock(&xps_map_mutex);
2630 new_dev_maps->nr_ids = nr_ids;
2631 new_dev_maps->num_tc = num_tc;
2634 tci = j * num_tc + tc;
2635 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2637 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2641 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2645 goto out_no_new_maps;
2648 /* Increment static keys at most once per type */
2649 static_key_slow_inc_cpuslocked(&xps_needed);
2650 if (type == XPS_RXQS)
2651 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2654 for (j = 0; j < nr_ids; j++) {
2655 bool skip_tc = false;
2657 tci = j * num_tc + tc;
2658 if (netif_attr_test_mask(j, mask, nr_ids) &&
2659 netif_attr_test_online(j, online_mask, nr_ids)) {
2660 /* add tx-queue to CPU/rx-queue maps */
2665 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2666 while ((pos < map->len) && (map->queues[pos] != index))
2669 if (pos == map->len)
2670 map->queues[map->len++] = index;
2672 if (type == XPS_CPUS) {
2673 if (numa_node_id == -2)
2674 numa_node_id = cpu_to_node(j);
2675 else if (numa_node_id != cpu_to_node(j))
2682 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2686 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2688 /* Cleanup old maps */
2690 goto out_no_old_maps;
2692 for (j = 0; j < dev_maps->nr_ids; j++) {
2693 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2694 map = xmap_dereference(dev_maps->attr_map[tci]);
2699 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2704 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2705 kfree_rcu(map, rcu);
2709 old_dev_maps = dev_maps;
2712 dev_maps = new_dev_maps;
2716 if (type == XPS_CPUS)
2717 /* update Tx queue numa node */
2718 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2719 (numa_node_id >= 0) ?
2720 numa_node_id : NUMA_NO_NODE);
2725 /* removes tx-queue from unused CPUs/rx-queues */
2726 for (j = 0; j < dev_maps->nr_ids; j++) {
2727 tci = j * dev_maps->num_tc;
2729 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2731 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2732 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2735 active |= remove_xps_queue(dev_maps,
2736 copy ? old_dev_maps : NULL,
2742 kfree_rcu(old_dev_maps, rcu);
2744 /* free map if not active */
2746 reset_xps_maps(dev, dev_maps, type);
2749 mutex_unlock(&xps_map_mutex);
2753 /* remove any maps that we added */
2754 for (j = 0; j < nr_ids; j++) {
2755 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2756 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2758 xmap_dereference(dev_maps->attr_map[tci]) :
2760 if (new_map && new_map != map)
2765 mutex_unlock(&xps_map_mutex);
2767 kfree(new_dev_maps);
2770 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2772 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2778 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2783 EXPORT_SYMBOL(netif_set_xps_queue);
2786 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2788 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2790 /* Unbind any subordinate channels */
2791 while (txq-- != &dev->_tx[0]) {
2793 netdev_unbind_sb_channel(dev, txq->sb_dev);
2797 void netdev_reset_tc(struct net_device *dev)
2800 netif_reset_xps_queues_gt(dev, 0);
2802 netdev_unbind_all_sb_channels(dev);
2804 /* Reset TC configuration of device */
2806 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2807 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2809 EXPORT_SYMBOL(netdev_reset_tc);
2811 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2813 if (tc >= dev->num_tc)
2817 netif_reset_xps_queues(dev, offset, count);
2819 dev->tc_to_txq[tc].count = count;
2820 dev->tc_to_txq[tc].offset = offset;
2823 EXPORT_SYMBOL(netdev_set_tc_queue);
2825 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2827 if (num_tc > TC_MAX_QUEUE)
2831 netif_reset_xps_queues_gt(dev, 0);
2833 netdev_unbind_all_sb_channels(dev);
2835 dev->num_tc = num_tc;
2838 EXPORT_SYMBOL(netdev_set_num_tc);
2840 void netdev_unbind_sb_channel(struct net_device *dev,
2841 struct net_device *sb_dev)
2843 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2846 netif_reset_xps_queues_gt(sb_dev, 0);
2848 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2849 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2851 while (txq-- != &dev->_tx[0]) {
2852 if (txq->sb_dev == sb_dev)
2856 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2858 int netdev_bind_sb_channel_queue(struct net_device *dev,
2859 struct net_device *sb_dev,
2860 u8 tc, u16 count, u16 offset)
2862 /* Make certain the sb_dev and dev are already configured */
2863 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2866 /* We cannot hand out queues we don't have */
2867 if ((offset + count) > dev->real_num_tx_queues)
2870 /* Record the mapping */
2871 sb_dev->tc_to_txq[tc].count = count;
2872 sb_dev->tc_to_txq[tc].offset = offset;
2874 /* Provide a way for Tx queue to find the tc_to_txq map or
2875 * XPS map for itself.
2878 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2882 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2884 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2886 /* Do not use a multiqueue device to represent a subordinate channel */
2887 if (netif_is_multiqueue(dev))
2890 /* We allow channels 1 - 32767 to be used for subordinate channels.
2891 * Channel 0 is meant to be "native" mode and used only to represent
2892 * the main root device. We allow writing 0 to reset the device back
2893 * to normal mode after being used as a subordinate channel.
2895 if (channel > S16_MAX)
2898 dev->num_tc = -channel;
2902 EXPORT_SYMBOL(netdev_set_sb_channel);
2905 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2906 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2908 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2913 disabling = txq < dev->real_num_tx_queues;
2915 if (txq < 1 || txq > dev->num_tx_queues)
2918 if (dev->reg_state == NETREG_REGISTERED ||
2919 dev->reg_state == NETREG_UNREGISTERING) {
2922 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2928 netif_setup_tc(dev, txq);
2930 dev_qdisc_change_real_num_tx(dev, txq);
2932 dev->real_num_tx_queues = txq;
2936 qdisc_reset_all_tx_gt(dev, txq);
2938 netif_reset_xps_queues_gt(dev, txq);
2942 dev->real_num_tx_queues = txq;
2947 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2951 * netif_set_real_num_rx_queues - set actual number of RX queues used
2952 * @dev: Network device
2953 * @rxq: Actual number of RX queues
2955 * This must be called either with the rtnl_lock held or before
2956 * registration of the net device. Returns 0 on success, or a
2957 * negative error code. If called before registration, it always
2960 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2964 if (rxq < 1 || rxq > dev->num_rx_queues)
2967 if (dev->reg_state == NETREG_REGISTERED) {
2970 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2976 dev->real_num_rx_queues = rxq;
2979 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2983 * netif_set_real_num_queues - set actual number of RX and TX queues used
2984 * @dev: Network device
2985 * @txq: Actual number of TX queues
2986 * @rxq: Actual number of RX queues
2988 * Set the real number of both TX and RX queues.
2989 * Does nothing if the number of queues is already correct.
2991 int netif_set_real_num_queues(struct net_device *dev,
2992 unsigned int txq, unsigned int rxq)
2994 unsigned int old_rxq = dev->real_num_rx_queues;
2997 if (txq < 1 || txq > dev->num_tx_queues ||
2998 rxq < 1 || rxq > dev->num_rx_queues)
3001 /* Start from increases, so the error path only does decreases -
3002 * decreases can't fail.
3004 if (rxq > dev->real_num_rx_queues) {
3005 err = netif_set_real_num_rx_queues(dev, rxq);
3009 if (txq > dev->real_num_tx_queues) {
3010 err = netif_set_real_num_tx_queues(dev, txq);
3014 if (rxq < dev->real_num_rx_queues)
3015 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3016 if (txq < dev->real_num_tx_queues)
3017 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3021 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3024 EXPORT_SYMBOL(netif_set_real_num_queues);
3027 * netif_get_num_default_rss_queues - default number of RSS queues
3029 * This routine should set an upper limit on the number of RSS queues
3030 * used by default by multiqueue devices.
3032 int netif_get_num_default_rss_queues(void)
3034 return is_kdump_kernel() ?
3035 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3037 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3039 static void __netif_reschedule(struct Qdisc *q)
3041 struct softnet_data *sd;
3042 unsigned long flags;
3044 local_irq_save(flags);
3045 sd = this_cpu_ptr(&softnet_data);
3046 q->next_sched = NULL;
3047 *sd->output_queue_tailp = q;
3048 sd->output_queue_tailp = &q->next_sched;
3049 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3050 local_irq_restore(flags);
3053 void __netif_schedule(struct Qdisc *q)
3055 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3056 __netif_reschedule(q);
3058 EXPORT_SYMBOL(__netif_schedule);
3060 struct dev_kfree_skb_cb {
3061 enum skb_free_reason reason;
3064 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3066 return (struct dev_kfree_skb_cb *)skb->cb;
3069 void netif_schedule_queue(struct netdev_queue *txq)
3072 if (!netif_xmit_stopped(txq)) {
3073 struct Qdisc *q = rcu_dereference(txq->qdisc);
3075 __netif_schedule(q);
3079 EXPORT_SYMBOL(netif_schedule_queue);
3081 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3083 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3087 q = rcu_dereference(dev_queue->qdisc);
3088 __netif_schedule(q);
3092 EXPORT_SYMBOL(netif_tx_wake_queue);
3094 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3096 unsigned long flags;
3101 if (likely(refcount_read(&skb->users) == 1)) {
3103 refcount_set(&skb->users, 0);
3104 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3107 get_kfree_skb_cb(skb)->reason = reason;
3108 local_irq_save(flags);
3109 skb->next = __this_cpu_read(softnet_data.completion_queue);
3110 __this_cpu_write(softnet_data.completion_queue, skb);
3111 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3112 local_irq_restore(flags);
3114 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3116 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3118 if (in_hardirq() || irqs_disabled())
3119 __dev_kfree_skb_irq(skb, reason);
3123 EXPORT_SYMBOL(__dev_kfree_skb_any);
3127 * netif_device_detach - mark device as removed
3128 * @dev: network device
3130 * Mark device as removed from system and therefore no longer available.
3132 void netif_device_detach(struct net_device *dev)
3134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3135 netif_running(dev)) {
3136 netif_tx_stop_all_queues(dev);
3139 EXPORT_SYMBOL(netif_device_detach);
3142 * netif_device_attach - mark device as attached
3143 * @dev: network device
3145 * Mark device as attached from system and restart if needed.
3147 void netif_device_attach(struct net_device *dev)
3149 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3150 netif_running(dev)) {
3151 netif_tx_wake_all_queues(dev);
3152 __netdev_watchdog_up(dev);
3155 EXPORT_SYMBOL(netif_device_attach);
3158 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3159 * to be used as a distribution range.
3161 static u16 skb_tx_hash(const struct net_device *dev,
3162 const struct net_device *sb_dev,
3163 struct sk_buff *skb)
3167 u16 qcount = dev->real_num_tx_queues;
3170 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3172 qoffset = sb_dev->tc_to_txq[tc].offset;
3173 qcount = sb_dev->tc_to_txq[tc].count;
3176 if (skb_rx_queue_recorded(skb)) {
3177 hash = skb_get_rx_queue(skb);
3178 if (hash >= qoffset)
3180 while (unlikely(hash >= qcount))
3182 return hash + qoffset;
3185 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3188 static void skb_warn_bad_offload(const struct sk_buff *skb)
3190 static const netdev_features_t null_features;
3191 struct net_device *dev = skb->dev;
3192 const char *name = "";
3194 if (!net_ratelimit())
3198 if (dev->dev.parent)
3199 name = dev_driver_string(dev->dev.parent);
3201 name = netdev_name(dev);
3203 skb_dump(KERN_WARNING, skb, false);
3204 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3205 name, dev ? &dev->features : &null_features,
3206 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3210 * Invalidate hardware checksum when packet is to be mangled, and
3211 * complete checksum manually on outgoing path.
3213 int skb_checksum_help(struct sk_buff *skb)
3216 int ret = 0, offset;
3218 if (skb->ip_summed == CHECKSUM_COMPLETE)
3219 goto out_set_summed;
3221 if (unlikely(skb_is_gso(skb))) {
3222 skb_warn_bad_offload(skb);
3226 /* Before computing a checksum, we should make sure no frag could
3227 * be modified by an external entity : checksum could be wrong.
3229 if (skb_has_shared_frag(skb)) {
3230 ret = __skb_linearize(skb);
3235 offset = skb_checksum_start_offset(skb);
3236 BUG_ON(offset >= skb_headlen(skb));
3237 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3239 offset += skb->csum_offset;
3240 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3242 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3246 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3248 skb->ip_summed = CHECKSUM_NONE;
3252 EXPORT_SYMBOL(skb_checksum_help);
3254 int skb_crc32c_csum_help(struct sk_buff *skb)
3257 int ret = 0, offset, start;
3259 if (skb->ip_summed != CHECKSUM_PARTIAL)
3262 if (unlikely(skb_is_gso(skb)))
3265 /* Before computing a checksum, we should make sure no frag could
3266 * be modified by an external entity : checksum could be wrong.
3268 if (unlikely(skb_has_shared_frag(skb))) {
3269 ret = __skb_linearize(skb);
3273 start = skb_checksum_start_offset(skb);
3274 offset = start + offsetof(struct sctphdr, checksum);
3275 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3280 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3284 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3285 skb->len - start, ~(__u32)0,
3287 *(__le32 *)(skb->data + offset) = crc32c_csum;
3288 skb->ip_summed = CHECKSUM_NONE;
3289 skb->csum_not_inet = 0;
3294 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3296 __be16 type = skb->protocol;
3298 /* Tunnel gso handlers can set protocol to ethernet. */
3299 if (type == htons(ETH_P_TEB)) {
3302 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3305 eth = (struct ethhdr *)skb->data;
3306 type = eth->h_proto;
3309 return __vlan_get_protocol(skb, type, depth);
3313 * skb_mac_gso_segment - mac layer segmentation handler.
3314 * @skb: buffer to segment
3315 * @features: features for the output path (see dev->features)
3317 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3318 netdev_features_t features)
3320 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3321 struct packet_offload *ptype;
3322 int vlan_depth = skb->mac_len;
3323 __be16 type = skb_network_protocol(skb, &vlan_depth);
3325 if (unlikely(!type))
3326 return ERR_PTR(-EINVAL);
3328 __skb_pull(skb, vlan_depth);
3331 list_for_each_entry_rcu(ptype, &offload_base, list) {
3332 if (ptype->type == type && ptype->callbacks.gso_segment) {
3333 segs = ptype->callbacks.gso_segment(skb, features);
3339 __skb_push(skb, skb->data - skb_mac_header(skb));
3343 EXPORT_SYMBOL(skb_mac_gso_segment);
3346 /* openvswitch calls this on rx path, so we need a different check.
3348 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3351 return skb->ip_summed != CHECKSUM_PARTIAL &&
3352 skb->ip_summed != CHECKSUM_UNNECESSARY;
3354 return skb->ip_summed == CHECKSUM_NONE;
3358 * __skb_gso_segment - Perform segmentation on skb.
3359 * @skb: buffer to segment
3360 * @features: features for the output path (see dev->features)
3361 * @tx_path: whether it is called in TX path
3363 * This function segments the given skb and returns a list of segments.
3365 * It may return NULL if the skb requires no segmentation. This is
3366 * only possible when GSO is used for verifying header integrity.
3368 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3370 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3371 netdev_features_t features, bool tx_path)
3373 struct sk_buff *segs;
3375 if (unlikely(skb_needs_check(skb, tx_path))) {
3378 /* We're going to init ->check field in TCP or UDP header */
3379 err = skb_cow_head(skb, 0);
3381 return ERR_PTR(err);
3384 /* Only report GSO partial support if it will enable us to
3385 * support segmentation on this frame without needing additional
3388 if (features & NETIF_F_GSO_PARTIAL) {
3389 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3390 struct net_device *dev = skb->dev;
3392 partial_features |= dev->features & dev->gso_partial_features;
3393 if (!skb_gso_ok(skb, features | partial_features))
3394 features &= ~NETIF_F_GSO_PARTIAL;
3397 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3398 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3400 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3401 SKB_GSO_CB(skb)->encap_level = 0;
3403 skb_reset_mac_header(skb);
3404 skb_reset_mac_len(skb);
3406 segs = skb_mac_gso_segment(skb, features);
3408 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3409 skb_warn_bad_offload(skb);
3413 EXPORT_SYMBOL(__skb_gso_segment);
3415 /* Take action when hardware reception checksum errors are detected. */
3417 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3419 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3420 skb_dump(KERN_ERR, skb, true);
3424 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3426 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3428 EXPORT_SYMBOL(netdev_rx_csum_fault);
3431 /* XXX: check that highmem exists at all on the given machine. */
3432 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3434 #ifdef CONFIG_HIGHMEM
3437 if (!(dev->features & NETIF_F_HIGHDMA)) {
3438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3441 if (PageHighMem(skb_frag_page(frag)))
3449 /* If MPLS offload request, verify we are testing hardware MPLS features
3450 * instead of standard features for the netdev.
3452 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3453 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3454 netdev_features_t features,
3457 if (eth_p_mpls(type))
3458 features &= skb->dev->mpls_features;
3463 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3464 netdev_features_t features,
3471 static netdev_features_t harmonize_features(struct sk_buff *skb,
3472 netdev_features_t features)
3476 type = skb_network_protocol(skb, NULL);
3477 features = net_mpls_features(skb, features, type);
3479 if (skb->ip_summed != CHECKSUM_NONE &&
3480 !can_checksum_protocol(features, type)) {
3481 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3483 if (illegal_highdma(skb->dev, skb))
3484 features &= ~NETIF_F_SG;
3489 netdev_features_t passthru_features_check(struct sk_buff *skb,
3490 struct net_device *dev,
3491 netdev_features_t features)
3495 EXPORT_SYMBOL(passthru_features_check);
3497 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3498 struct net_device *dev,
3499 netdev_features_t features)
3501 return vlan_features_check(skb, features);
3504 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3505 struct net_device *dev,
3506 netdev_features_t features)
3508 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3510 if (gso_segs > dev->gso_max_segs)
3511 return features & ~NETIF_F_GSO_MASK;
3513 if (!skb_shinfo(skb)->gso_type) {
3514 skb_warn_bad_offload(skb);
3515 return features & ~NETIF_F_GSO_MASK;
3518 /* Support for GSO partial features requires software
3519 * intervention before we can actually process the packets
3520 * so we need to strip support for any partial features now
3521 * and we can pull them back in after we have partially
3522 * segmented the frame.
3524 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3525 features &= ~dev->gso_partial_features;
3527 /* Make sure to clear the IPv4 ID mangling feature if the
3528 * IPv4 header has the potential to be fragmented.
3530 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3531 struct iphdr *iph = skb->encapsulation ?
3532 inner_ip_hdr(skb) : ip_hdr(skb);
3534 if (!(iph->frag_off & htons(IP_DF)))
3535 features &= ~NETIF_F_TSO_MANGLEID;
3541 netdev_features_t netif_skb_features(struct sk_buff *skb)
3543 struct net_device *dev = skb->dev;
3544 netdev_features_t features = dev->features;
3546 if (skb_is_gso(skb))
3547 features = gso_features_check(skb, dev, features);
3549 /* If encapsulation offload request, verify we are testing
3550 * hardware encapsulation features instead of standard
3551 * features for the netdev
3553 if (skb->encapsulation)
3554 features &= dev->hw_enc_features;
3556 if (skb_vlan_tagged(skb))
3557 features = netdev_intersect_features(features,
3558 dev->vlan_features |
3559 NETIF_F_HW_VLAN_CTAG_TX |
3560 NETIF_F_HW_VLAN_STAG_TX);
3562 if (dev->netdev_ops->ndo_features_check)
3563 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3566 features &= dflt_features_check(skb, dev, features);
3568 return harmonize_features(skb, features);
3570 EXPORT_SYMBOL(netif_skb_features);
3572 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3573 struct netdev_queue *txq, bool more)
3578 if (dev_nit_active(dev))
3579 dev_queue_xmit_nit(skb, dev);
3582 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3583 trace_net_dev_start_xmit(skb, dev);
3584 rc = netdev_start_xmit(skb, dev, txq, more);
3585 trace_net_dev_xmit(skb, rc, dev, len);
3590 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3591 struct netdev_queue *txq, int *ret)
3593 struct sk_buff *skb = first;
3594 int rc = NETDEV_TX_OK;
3597 struct sk_buff *next = skb->next;
3599 skb_mark_not_on_list(skb);
3600 rc = xmit_one(skb, dev, txq, next != NULL);
3601 if (unlikely(!dev_xmit_complete(rc))) {
3607 if (netif_tx_queue_stopped(txq) && skb) {
3608 rc = NETDEV_TX_BUSY;
3618 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3619 netdev_features_t features)
3621 if (skb_vlan_tag_present(skb) &&
3622 !vlan_hw_offload_capable(features, skb->vlan_proto))
3623 skb = __vlan_hwaccel_push_inside(skb);
3627 int skb_csum_hwoffload_help(struct sk_buff *skb,
3628 const netdev_features_t features)
3630 if (unlikely(skb_csum_is_sctp(skb)))
3631 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3632 skb_crc32c_csum_help(skb);
3634 if (features & NETIF_F_HW_CSUM)
3637 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3638 switch (skb->csum_offset) {
3639 case offsetof(struct tcphdr, check):
3640 case offsetof(struct udphdr, check):
3645 return skb_checksum_help(skb);
3647 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3649 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3651 netdev_features_t features;
3653 features = netif_skb_features(skb);
3654 skb = validate_xmit_vlan(skb, features);
3658 skb = sk_validate_xmit_skb(skb, dev);
3662 if (netif_needs_gso(skb, features)) {
3663 struct sk_buff *segs;
3665 segs = skb_gso_segment(skb, features);
3673 if (skb_needs_linearize(skb, features) &&
3674 __skb_linearize(skb))
3677 /* If packet is not checksummed and device does not
3678 * support checksumming for this protocol, complete
3679 * checksumming here.
3681 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3682 if (skb->encapsulation)
3683 skb_set_inner_transport_header(skb,
3684 skb_checksum_start_offset(skb));
3686 skb_set_transport_header(skb,
3687 skb_checksum_start_offset(skb));
3688 if (skb_csum_hwoffload_help(skb, features))
3693 skb = validate_xmit_xfrm(skb, features, again);
3700 atomic_long_inc(&dev->tx_dropped);
3704 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3706 struct sk_buff *next, *head = NULL, *tail;
3708 for (; skb != NULL; skb = next) {
3710 skb_mark_not_on_list(skb);
3712 /* in case skb wont be segmented, point to itself */
3715 skb = validate_xmit_skb(skb, dev, again);
3723 /* If skb was segmented, skb->prev points to
3724 * the last segment. If not, it still contains skb.
3730 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3732 static void qdisc_pkt_len_init(struct sk_buff *skb)
3734 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3736 qdisc_skb_cb(skb)->pkt_len = skb->len;
3738 /* To get more precise estimation of bytes sent on wire,
3739 * we add to pkt_len the headers size of all segments
3741 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3742 unsigned int hdr_len;
3743 u16 gso_segs = shinfo->gso_segs;
3745 /* mac layer + network layer */
3746 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3748 /* + transport layer */
3749 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3750 const struct tcphdr *th;
3751 struct tcphdr _tcphdr;
3753 th = skb_header_pointer(skb, skb_transport_offset(skb),
3754 sizeof(_tcphdr), &_tcphdr);
3756 hdr_len += __tcp_hdrlen(th);
3758 struct udphdr _udphdr;
3760 if (skb_header_pointer(skb, skb_transport_offset(skb),
3761 sizeof(_udphdr), &_udphdr))
3762 hdr_len += sizeof(struct udphdr);
3765 if (shinfo->gso_type & SKB_GSO_DODGY)
3766 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3769 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3773 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3774 struct sk_buff **to_free,
3775 struct netdev_queue *txq)
3779 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3780 if (rc == NET_XMIT_SUCCESS)
3781 trace_qdisc_enqueue(q, txq, skb);
3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3786 struct net_device *dev,
3787 struct netdev_queue *txq)
3789 spinlock_t *root_lock = qdisc_lock(q);
3790 struct sk_buff *to_free = NULL;
3794 qdisc_calculate_pkt_len(skb, q);
3796 if (q->flags & TCQ_F_NOLOCK) {
3797 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3798 qdisc_run_begin(q)) {
3799 /* Retest nolock_qdisc_is_empty() within the protection
3800 * of q->seqlock to protect from racing with requeuing.
3802 if (unlikely(!nolock_qdisc_is_empty(q))) {
3803 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3810 qdisc_bstats_cpu_update(q, skb);
3811 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3812 !nolock_qdisc_is_empty(q))
3816 return NET_XMIT_SUCCESS;
3819 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3823 if (unlikely(to_free))
3824 kfree_skb_list(to_free);
3829 * Heuristic to force contended enqueues to serialize on a
3830 * separate lock before trying to get qdisc main lock.
3831 * This permits qdisc->running owner to get the lock more
3832 * often and dequeue packets faster.
3834 contended = qdisc_is_running(q);
3835 if (unlikely(contended))
3836 spin_lock(&q->busylock);
3838 spin_lock(root_lock);
3839 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3840 __qdisc_drop(skb, &to_free);
3842 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3843 qdisc_run_begin(q)) {
3845 * This is a work-conserving queue; there are no old skbs
3846 * waiting to be sent out; and the qdisc is not running -
3847 * xmit the skb directly.
3850 qdisc_bstats_update(q, skb);
3852 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3853 if (unlikely(contended)) {
3854 spin_unlock(&q->busylock);
3861 rc = NET_XMIT_SUCCESS;
3863 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3864 if (qdisc_run_begin(q)) {
3865 if (unlikely(contended)) {
3866 spin_unlock(&q->busylock);
3873 spin_unlock(root_lock);
3874 if (unlikely(to_free))
3875 kfree_skb_list(to_free);
3876 if (unlikely(contended))
3877 spin_unlock(&q->busylock);
3881 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3882 static void skb_update_prio(struct sk_buff *skb)
3884 const struct netprio_map *map;
3885 const struct sock *sk;
3886 unsigned int prioidx;
3890 map = rcu_dereference_bh(skb->dev->priomap);
3893 sk = skb_to_full_sk(skb);
3897 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3899 if (prioidx < map->priomap_len)
3900 skb->priority = map->priomap[prioidx];
3903 #define skb_update_prio(skb)
3907 * dev_loopback_xmit - loop back @skb
3908 * @net: network namespace this loopback is happening in
3909 * @sk: sk needed to be a netfilter okfn
3910 * @skb: buffer to transmit
3912 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3914 skb_reset_mac_header(skb);
3915 __skb_pull(skb, skb_network_offset(skb));
3916 skb->pkt_type = PACKET_LOOPBACK;
3917 skb->ip_summed = CHECKSUM_UNNECESSARY;
3918 WARN_ON(!skb_dst(skb));
3923 EXPORT_SYMBOL(dev_loopback_xmit);
3925 #ifdef CONFIG_NET_EGRESS
3926 static struct sk_buff *
3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3929 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3930 struct tcf_result cl_res;
3935 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3936 qdisc_skb_cb(skb)->mru = 0;
3937 qdisc_skb_cb(skb)->post_ct = false;
3938 mini_qdisc_bstats_cpu_update(miniq, skb);
3940 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3942 case TC_ACT_RECLASSIFY:
3943 skb->tc_index = TC_H_MIN(cl_res.classid);
3946 mini_qdisc_qstats_cpu_drop(miniq);
3947 *ret = NET_XMIT_DROP;
3953 *ret = NET_XMIT_SUCCESS;
3956 case TC_ACT_REDIRECT:
3957 /* No need to push/pop skb's mac_header here on egress! */
3958 skb_do_redirect(skb);
3959 *ret = NET_XMIT_SUCCESS;
3967 #endif /* CONFIG_NET_EGRESS */
3970 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3971 struct xps_dev_maps *dev_maps, unsigned int tci)
3973 int tc = netdev_get_prio_tc_map(dev, skb->priority);
3974 struct xps_map *map;
3975 int queue_index = -1;
3977 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3980 tci *= dev_maps->num_tc;
3983 map = rcu_dereference(dev_maps->attr_map[tci]);
3986 queue_index = map->queues[0];
3988 queue_index = map->queues[reciprocal_scale(
3989 skb_get_hash(skb), map->len)];
3990 if (unlikely(queue_index >= dev->real_num_tx_queues))
3997 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3998 struct sk_buff *skb)
4001 struct xps_dev_maps *dev_maps;
4002 struct sock *sk = skb->sk;
4003 int queue_index = -1;
4005 if (!static_key_false(&xps_needed))
4009 if (!static_key_false(&xps_rxqs_needed))
4012 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4014 int tci = sk_rx_queue_get(sk);
4017 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4022 if (queue_index < 0) {
4023 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4025 unsigned int tci = skb->sender_cpu - 1;
4027 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4039 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4040 struct net_device *sb_dev)
4044 EXPORT_SYMBOL(dev_pick_tx_zero);
4046 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4047 struct net_device *sb_dev)
4049 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4051 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4053 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4054 struct net_device *sb_dev)
4056 struct sock *sk = skb->sk;
4057 int queue_index = sk_tx_queue_get(sk);
4059 sb_dev = sb_dev ? : dev;
4061 if (queue_index < 0 || skb->ooo_okay ||
4062 queue_index >= dev->real_num_tx_queues) {
4063 int new_index = get_xps_queue(dev, sb_dev, skb);
4066 new_index = skb_tx_hash(dev, sb_dev, skb);
4068 if (queue_index != new_index && sk &&
4070 rcu_access_pointer(sk->sk_dst_cache))
4071 sk_tx_queue_set(sk, new_index);
4073 queue_index = new_index;
4078 EXPORT_SYMBOL(netdev_pick_tx);
4080 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4081 struct sk_buff *skb,
4082 struct net_device *sb_dev)
4084 int queue_index = 0;
4087 u32 sender_cpu = skb->sender_cpu - 1;
4089 if (sender_cpu >= (u32)NR_CPUS)
4090 skb->sender_cpu = raw_smp_processor_id() + 1;
4093 if (dev->real_num_tx_queues != 1) {
4094 const struct net_device_ops *ops = dev->netdev_ops;
4096 if (ops->ndo_select_queue)
4097 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4099 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4101 queue_index = netdev_cap_txqueue(dev, queue_index);
4104 skb_set_queue_mapping(skb, queue_index);
4105 return netdev_get_tx_queue(dev, queue_index);
4109 * __dev_queue_xmit - transmit a buffer
4110 * @skb: buffer to transmit
4111 * @sb_dev: suboordinate device used for L2 forwarding offload
4113 * Queue a buffer for transmission to a network device. The caller must
4114 * have set the device and priority and built the buffer before calling
4115 * this function. The function can be called from an interrupt.
4117 * A negative errno code is returned on a failure. A success does not
4118 * guarantee the frame will be transmitted as it may be dropped due
4119 * to congestion or traffic shaping.
4121 * -----------------------------------------------------------------------------------
4122 * I notice this method can also return errors from the queue disciplines,
4123 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4126 * Regardless of the return value, the skb is consumed, so it is currently
4127 * difficult to retry a send to this method. (You can bump the ref count
4128 * before sending to hold a reference for retry if you are careful.)
4130 * When calling this method, interrupts MUST be enabled. This is because
4131 * the BH enable code must have IRQs enabled so that it will not deadlock.
4134 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4136 struct net_device *dev = skb->dev;
4137 struct netdev_queue *txq;
4142 skb_reset_mac_header(skb);
4144 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4145 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4147 /* Disable soft irqs for various locks below. Also
4148 * stops preemption for RCU.
4152 skb_update_prio(skb);
4154 qdisc_pkt_len_init(skb);
4155 #ifdef CONFIG_NET_CLS_ACT
4156 skb->tc_at_ingress = 0;
4157 # ifdef CONFIG_NET_EGRESS
4158 if (static_branch_unlikely(&egress_needed_key)) {
4159 skb = sch_handle_egress(skb, &rc, dev);
4165 /* If device/qdisc don't need skb->dst, release it right now while
4166 * its hot in this cpu cache.
4168 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4173 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4174 q = rcu_dereference_bh(txq->qdisc);
4176 trace_net_dev_queue(skb);
4178 rc = __dev_xmit_skb(skb, q, dev, txq);
4182 /* The device has no queue. Common case for software devices:
4183 * loopback, all the sorts of tunnels...
4185 * Really, it is unlikely that netif_tx_lock protection is necessary
4186 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4188 * However, it is possible, that they rely on protection
4191 * Check this and shot the lock. It is not prone from deadlocks.
4192 *Either shot noqueue qdisc, it is even simpler 8)
4194 if (dev->flags & IFF_UP) {
4195 int cpu = smp_processor_id(); /* ok because BHs are off */
4197 if (txq->xmit_lock_owner != cpu) {
4198 if (dev_xmit_recursion())
4199 goto recursion_alert;
4201 skb = validate_xmit_skb(skb, dev, &again);
4205 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4206 HARD_TX_LOCK(dev, txq, cpu);
4208 if (!netif_xmit_stopped(txq)) {
4209 dev_xmit_recursion_inc();
4210 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4211 dev_xmit_recursion_dec();
4212 if (dev_xmit_complete(rc)) {
4213 HARD_TX_UNLOCK(dev, txq);
4217 HARD_TX_UNLOCK(dev, txq);
4218 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4221 /* Recursion is detected! It is possible,
4225 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4231 rcu_read_unlock_bh();
4233 atomic_long_inc(&dev->tx_dropped);
4234 kfree_skb_list(skb);
4237 rcu_read_unlock_bh();
4241 int dev_queue_xmit(struct sk_buff *skb)
4243 return __dev_queue_xmit(skb, NULL);
4245 EXPORT_SYMBOL(dev_queue_xmit);
4247 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4249 return __dev_queue_xmit(skb, sb_dev);
4251 EXPORT_SYMBOL(dev_queue_xmit_accel);
4253 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4255 struct net_device *dev = skb->dev;
4256 struct sk_buff *orig_skb = skb;
4257 struct netdev_queue *txq;
4258 int ret = NETDEV_TX_BUSY;
4261 if (unlikely(!netif_running(dev) ||
4262 !netif_carrier_ok(dev)))
4265 skb = validate_xmit_skb_list(skb, dev, &again);
4266 if (skb != orig_skb)
4269 skb_set_queue_mapping(skb, queue_id);
4270 txq = skb_get_tx_queue(dev, skb);
4271 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4275 dev_xmit_recursion_inc();
4276 HARD_TX_LOCK(dev, txq, smp_processor_id());
4277 if (!netif_xmit_frozen_or_drv_stopped(txq))
4278 ret = netdev_start_xmit(skb, dev, txq, false);
4279 HARD_TX_UNLOCK(dev, txq);
4280 dev_xmit_recursion_dec();
4285 atomic_long_inc(&dev->tx_dropped);
4286 kfree_skb_list(skb);
4287 return NET_XMIT_DROP;
4289 EXPORT_SYMBOL(__dev_direct_xmit);
4291 /*************************************************************************
4293 *************************************************************************/
4295 int netdev_max_backlog __read_mostly = 1000;
4296 EXPORT_SYMBOL(netdev_max_backlog);
4298 int netdev_tstamp_prequeue __read_mostly = 1;
4299 int netdev_budget __read_mostly = 300;
4300 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4301 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4302 int weight_p __read_mostly = 64; /* old backlog weight */
4303 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4304 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4305 int dev_rx_weight __read_mostly = 64;
4306 int dev_tx_weight __read_mostly = 64;
4307 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4308 int gro_normal_batch __read_mostly = 8;
4310 /* Called with irq disabled */
4311 static inline void ____napi_schedule(struct softnet_data *sd,
4312 struct napi_struct *napi)
4314 struct task_struct *thread;
4316 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4317 /* Paired with smp_mb__before_atomic() in
4318 * napi_enable()/dev_set_threaded().
4319 * Use READ_ONCE() to guarantee a complete
4320 * read on napi->thread. Only call
4321 * wake_up_process() when it's not NULL.
4323 thread = READ_ONCE(napi->thread);
4325 /* Avoid doing set_bit() if the thread is in
4326 * INTERRUPTIBLE state, cause napi_thread_wait()
4327 * makes sure to proceed with napi polling
4328 * if the thread is explicitly woken from here.
4330 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4331 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4332 wake_up_process(thread);
4337 list_add_tail(&napi->poll_list, &sd->poll_list);
4338 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4343 /* One global table that all flow-based protocols share. */
4344 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4345 EXPORT_SYMBOL(rps_sock_flow_table);
4346 u32 rps_cpu_mask __read_mostly;
4347 EXPORT_SYMBOL(rps_cpu_mask);
4349 struct static_key_false rps_needed __read_mostly;
4350 EXPORT_SYMBOL(rps_needed);
4351 struct static_key_false rfs_needed __read_mostly;
4352 EXPORT_SYMBOL(rfs_needed);
4354 static struct rps_dev_flow *
4355 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4356 struct rps_dev_flow *rflow, u16 next_cpu)
4358 if (next_cpu < nr_cpu_ids) {
4359 #ifdef CONFIG_RFS_ACCEL
4360 struct netdev_rx_queue *rxqueue;
4361 struct rps_dev_flow_table *flow_table;
4362 struct rps_dev_flow *old_rflow;
4367 /* Should we steer this flow to a different hardware queue? */
4368 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4369 !(dev->features & NETIF_F_NTUPLE))
4371 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4372 if (rxq_index == skb_get_rx_queue(skb))
4375 rxqueue = dev->_rx + rxq_index;
4376 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4379 flow_id = skb_get_hash(skb) & flow_table->mask;
4380 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4381 rxq_index, flow_id);
4385 rflow = &flow_table->flows[flow_id];
4387 if (old_rflow->filter == rflow->filter)
4388 old_rflow->filter = RPS_NO_FILTER;
4392 per_cpu(softnet_data, next_cpu).input_queue_head;
4395 rflow->cpu = next_cpu;
4400 * get_rps_cpu is called from netif_receive_skb and returns the target
4401 * CPU from the RPS map of the receiving queue for a given skb.
4402 * rcu_read_lock must be held on entry.
4404 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4405 struct rps_dev_flow **rflowp)
4407 const struct rps_sock_flow_table *sock_flow_table;
4408 struct netdev_rx_queue *rxqueue = dev->_rx;
4409 struct rps_dev_flow_table *flow_table;
4410 struct rps_map *map;
4415 if (skb_rx_queue_recorded(skb)) {
4416 u16 index = skb_get_rx_queue(skb);
4418 if (unlikely(index >= dev->real_num_rx_queues)) {
4419 WARN_ONCE(dev->real_num_rx_queues > 1,
4420 "%s received packet on queue %u, but number "
4421 "of RX queues is %u\n",
4422 dev->name, index, dev->real_num_rx_queues);
4428 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4430 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4431 map = rcu_dereference(rxqueue->rps_map);
4432 if (!flow_table && !map)
4435 skb_reset_network_header(skb);
4436 hash = skb_get_hash(skb);
4440 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4441 if (flow_table && sock_flow_table) {
4442 struct rps_dev_flow *rflow;
4446 /* First check into global flow table if there is a match */
4447 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4448 if ((ident ^ hash) & ~rps_cpu_mask)
4451 next_cpu = ident & rps_cpu_mask;
4453 /* OK, now we know there is a match,
4454 * we can look at the local (per receive queue) flow table
4456 rflow = &flow_table->flows[hash & flow_table->mask];
4460 * If the desired CPU (where last recvmsg was done) is
4461 * different from current CPU (one in the rx-queue flow
4462 * table entry), switch if one of the following holds:
4463 * - Current CPU is unset (>= nr_cpu_ids).
4464 * - Current CPU is offline.
4465 * - The current CPU's queue tail has advanced beyond the
4466 * last packet that was enqueued using this table entry.
4467 * This guarantees that all previous packets for the flow
4468 * have been dequeued, thus preserving in order delivery.
4470 if (unlikely(tcpu != next_cpu) &&
4471 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4472 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4473 rflow->last_qtail)) >= 0)) {
4475 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4478 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4488 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4489 if (cpu_online(tcpu)) {
4499 #ifdef CONFIG_RFS_ACCEL
4502 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4503 * @dev: Device on which the filter was set
4504 * @rxq_index: RX queue index
4505 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4506 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4508 * Drivers that implement ndo_rx_flow_steer() should periodically call
4509 * this function for each installed filter and remove the filters for
4510 * which it returns %true.
4512 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4513 u32 flow_id, u16 filter_id)
4515 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4516 struct rps_dev_flow_table *flow_table;
4517 struct rps_dev_flow *rflow;
4522 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4523 if (flow_table && flow_id <= flow_table->mask) {
4524 rflow = &flow_table->flows[flow_id];
4525 cpu = READ_ONCE(rflow->cpu);
4526 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4527 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4528 rflow->last_qtail) <
4529 (int)(10 * flow_table->mask)))
4535 EXPORT_SYMBOL(rps_may_expire_flow);
4537 #endif /* CONFIG_RFS_ACCEL */
4539 /* Called from hardirq (IPI) context */
4540 static void rps_trigger_softirq(void *data)
4542 struct softnet_data *sd = data;
4544 ____napi_schedule(sd, &sd->backlog);
4548 #endif /* CONFIG_RPS */
4551 * Check if this softnet_data structure is another cpu one
4552 * If yes, queue it to our IPI list and return 1
4555 static int rps_ipi_queued(struct softnet_data *sd)
4558 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4561 sd->rps_ipi_next = mysd->rps_ipi_list;
4562 mysd->rps_ipi_list = sd;
4564 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4567 #endif /* CONFIG_RPS */
4571 #ifdef CONFIG_NET_FLOW_LIMIT
4572 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4575 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4577 #ifdef CONFIG_NET_FLOW_LIMIT
4578 struct sd_flow_limit *fl;
4579 struct softnet_data *sd;
4580 unsigned int old_flow, new_flow;
4582 if (qlen < (netdev_max_backlog >> 1))
4585 sd = this_cpu_ptr(&softnet_data);
4588 fl = rcu_dereference(sd->flow_limit);
4590 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4591 old_flow = fl->history[fl->history_head];
4592 fl->history[fl->history_head] = new_flow;
4595 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4597 if (likely(fl->buckets[old_flow]))
4598 fl->buckets[old_flow]--;
4600 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4612 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4613 * queue (may be a remote CPU queue).
4615 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4616 unsigned int *qtail)
4618 struct softnet_data *sd;
4619 unsigned long flags;
4622 sd = &per_cpu(softnet_data, cpu);
4624 local_irq_save(flags);
4627 if (!netif_running(skb->dev))
4629 qlen = skb_queue_len(&sd->input_pkt_queue);
4630 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4633 __skb_queue_tail(&sd->input_pkt_queue, skb);
4634 input_queue_tail_incr_save(sd, qtail);
4636 local_irq_restore(flags);
4637 return NET_RX_SUCCESS;
4640 /* Schedule NAPI for backlog device
4641 * We can use non atomic operation since we own the queue lock
4643 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4644 if (!rps_ipi_queued(sd))
4645 ____napi_schedule(sd, &sd->backlog);
4654 local_irq_restore(flags);
4656 atomic_long_inc(&skb->dev->rx_dropped);
4661 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4663 struct net_device *dev = skb->dev;
4664 struct netdev_rx_queue *rxqueue;
4668 if (skb_rx_queue_recorded(skb)) {
4669 u16 index = skb_get_rx_queue(skb);
4671 if (unlikely(index >= dev->real_num_rx_queues)) {
4672 WARN_ONCE(dev->real_num_rx_queues > 1,
4673 "%s received packet on queue %u, but number "
4674 "of RX queues is %u\n",
4675 dev->name, index, dev->real_num_rx_queues);
4677 return rxqueue; /* Return first rxqueue */
4684 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4685 struct bpf_prog *xdp_prog)
4687 void *orig_data, *orig_data_end, *hard_start;
4688 struct netdev_rx_queue *rxqueue;
4689 bool orig_bcast, orig_host;
4690 u32 mac_len, frame_sz;
4691 __be16 orig_eth_type;
4696 /* The XDP program wants to see the packet starting at the MAC
4699 mac_len = skb->data - skb_mac_header(skb);
4700 hard_start = skb->data - skb_headroom(skb);
4702 /* SKB "head" area always have tailroom for skb_shared_info */
4703 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4704 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4706 rxqueue = netif_get_rxqueue(skb);
4707 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4708 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4709 skb_headlen(skb) + mac_len, true);
4711 orig_data_end = xdp->data_end;
4712 orig_data = xdp->data;
4713 eth = (struct ethhdr *)xdp->data;
4714 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4715 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4716 orig_eth_type = eth->h_proto;
4718 act = bpf_prog_run_xdp(xdp_prog, xdp);
4720 /* check if bpf_xdp_adjust_head was used */
4721 off = xdp->data - orig_data;
4724 __skb_pull(skb, off);
4726 __skb_push(skb, -off);
4728 skb->mac_header += off;
4729 skb_reset_network_header(skb);
4732 /* check if bpf_xdp_adjust_tail was used */
4733 off = xdp->data_end - orig_data_end;
4735 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4736 skb->len += off; /* positive on grow, negative on shrink */
4739 /* check if XDP changed eth hdr such SKB needs update */
4740 eth = (struct ethhdr *)xdp->data;
4741 if ((orig_eth_type != eth->h_proto) ||
4742 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4743 skb->dev->dev_addr)) ||
4744 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4745 __skb_push(skb, ETH_HLEN);
4746 skb->pkt_type = PACKET_HOST;
4747 skb->protocol = eth_type_trans(skb, skb->dev);
4750 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4751 * before calling us again on redirect path. We do not call do_redirect
4752 * as we leave that up to the caller.
4754 * Caller is responsible for managing lifetime of skb (i.e. calling
4755 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4760 __skb_push(skb, mac_len);
4763 metalen = xdp->data - xdp->data_meta;
4765 skb_metadata_set(skb, metalen);
4772 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4773 struct xdp_buff *xdp,
4774 struct bpf_prog *xdp_prog)
4778 /* Reinjected packets coming from act_mirred or similar should
4779 * not get XDP generic processing.
4781 if (skb_is_redirected(skb))
4784 /* XDP packets must be linear and must have sufficient headroom
4785 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4786 * native XDP provides, thus we need to do it here as well.
4788 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4789 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4790 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4791 int troom = skb->tail + skb->data_len - skb->end;
4793 /* In case we have to go down the path and also linearize,
4794 * then lets do the pskb_expand_head() work just once here.
4796 if (pskb_expand_head(skb,
4797 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4798 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4800 if (skb_linearize(skb))
4804 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4811 bpf_warn_invalid_xdp_action(act);
4814 trace_xdp_exception(skb->dev, xdp_prog, act);
4825 /* When doing generic XDP we have to bypass the qdisc layer and the
4826 * network taps in order to match in-driver-XDP behavior.
4828 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4830 struct net_device *dev = skb->dev;
4831 struct netdev_queue *txq;
4832 bool free_skb = true;
4835 txq = netdev_core_pick_tx(dev, skb, NULL);
4836 cpu = smp_processor_id();
4837 HARD_TX_LOCK(dev, txq, cpu);
4838 if (!netif_xmit_stopped(txq)) {
4839 rc = netdev_start_xmit(skb, dev, txq, 0);
4840 if (dev_xmit_complete(rc))
4843 HARD_TX_UNLOCK(dev, txq);
4845 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4850 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4852 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4855 struct xdp_buff xdp;
4859 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4860 if (act != XDP_PASS) {
4863 err = xdp_do_generic_redirect(skb->dev, skb,
4869 generic_xdp_tx(skb, xdp_prog);
4880 EXPORT_SYMBOL_GPL(do_xdp_generic);
4882 static int netif_rx_internal(struct sk_buff *skb)
4886 net_timestamp_check(netdev_tstamp_prequeue, skb);
4888 trace_netif_rx(skb);
4891 if (static_branch_unlikely(&rps_needed)) {
4892 struct rps_dev_flow voidflow, *rflow = &voidflow;
4898 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4900 cpu = smp_processor_id();
4902 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4911 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4918 * netif_rx - post buffer to the network code
4919 * @skb: buffer to post
4921 * This function receives a packet from a device driver and queues it for
4922 * the upper (protocol) levels to process. It always succeeds. The buffer
4923 * may be dropped during processing for congestion control or by the
4927 * NET_RX_SUCCESS (no congestion)
4928 * NET_RX_DROP (packet was dropped)
4932 int netif_rx(struct sk_buff *skb)
4936 trace_netif_rx_entry(skb);
4938 ret = netif_rx_internal(skb);
4939 trace_netif_rx_exit(ret);
4943 EXPORT_SYMBOL(netif_rx);
4945 int netif_rx_ni(struct sk_buff *skb)
4949 trace_netif_rx_ni_entry(skb);
4952 err = netif_rx_internal(skb);
4953 if (local_softirq_pending())
4956 trace_netif_rx_ni_exit(err);
4960 EXPORT_SYMBOL(netif_rx_ni);
4962 int netif_rx_any_context(struct sk_buff *skb)
4965 * If invoked from contexts which do not invoke bottom half
4966 * processing either at return from interrupt or when softrqs are
4967 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4971 return netif_rx(skb);
4973 return netif_rx_ni(skb);
4975 EXPORT_SYMBOL(netif_rx_any_context);
4977 static __latent_entropy void net_tx_action(struct softirq_action *h)
4979 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4981 if (sd->completion_queue) {
4982 struct sk_buff *clist;
4984 local_irq_disable();
4985 clist = sd->completion_queue;
4986 sd->completion_queue = NULL;
4990 struct sk_buff *skb = clist;
4992 clist = clist->next;
4994 WARN_ON(refcount_read(&skb->users));
4995 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4996 trace_consume_skb(skb);
4998 trace_kfree_skb(skb, net_tx_action);
5000 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5003 __kfree_skb_defer(skb);
5007 if (sd->output_queue) {
5010 local_irq_disable();
5011 head = sd->output_queue;
5012 sd->output_queue = NULL;
5013 sd->output_queue_tailp = &sd->output_queue;
5019 struct Qdisc *q = head;
5020 spinlock_t *root_lock = NULL;
5022 head = head->next_sched;
5024 /* We need to make sure head->next_sched is read
5025 * before clearing __QDISC_STATE_SCHED
5027 smp_mb__before_atomic();
5029 if (!(q->flags & TCQ_F_NOLOCK)) {
5030 root_lock = qdisc_lock(q);
5031 spin_lock(root_lock);
5032 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5034 /* There is a synchronize_net() between
5035 * STATE_DEACTIVATED flag being set and
5036 * qdisc_reset()/some_qdisc_is_busy() in
5037 * dev_deactivate(), so we can safely bail out
5038 * early here to avoid data race between
5039 * qdisc_deactivate() and some_qdisc_is_busy()
5040 * for lockless qdisc.
5042 clear_bit(__QDISC_STATE_SCHED, &q->state);
5046 clear_bit(__QDISC_STATE_SCHED, &q->state);
5049 spin_unlock(root_lock);
5055 xfrm_dev_backlog(sd);
5058 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5059 /* This hook is defined here for ATM LANE */
5060 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5061 unsigned char *addr) __read_mostly;
5062 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5065 static inline struct sk_buff *
5066 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5067 struct net_device *orig_dev, bool *another)
5069 #ifdef CONFIG_NET_CLS_ACT
5070 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5071 struct tcf_result cl_res;
5073 /* If there's at least one ingress present somewhere (so
5074 * we get here via enabled static key), remaining devices
5075 * that are not configured with an ingress qdisc will bail
5082 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5086 qdisc_skb_cb(skb)->pkt_len = skb->len;
5087 qdisc_skb_cb(skb)->mru = 0;
5088 qdisc_skb_cb(skb)->post_ct = false;
5089 skb->tc_at_ingress = 1;
5090 mini_qdisc_bstats_cpu_update(miniq, skb);
5092 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5094 case TC_ACT_RECLASSIFY:
5095 skb->tc_index = TC_H_MIN(cl_res.classid);
5098 mini_qdisc_qstats_cpu_drop(miniq);
5106 case TC_ACT_REDIRECT:
5107 /* skb_mac_header check was done by cls/act_bpf, so
5108 * we can safely push the L2 header back before
5109 * redirecting to another netdev
5111 __skb_push(skb, skb->mac_len);
5112 if (skb_do_redirect(skb) == -EAGAIN) {
5113 __skb_pull(skb, skb->mac_len);
5118 case TC_ACT_CONSUMED:
5123 #endif /* CONFIG_NET_CLS_ACT */
5128 * netdev_is_rx_handler_busy - check if receive handler is registered
5129 * @dev: device to check
5131 * Check if a receive handler is already registered for a given device.
5132 * Return true if there one.
5134 * The caller must hold the rtnl_mutex.
5136 bool netdev_is_rx_handler_busy(struct net_device *dev)
5139 return dev && rtnl_dereference(dev->rx_handler);
5141 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5144 * netdev_rx_handler_register - register receive handler
5145 * @dev: device to register a handler for
5146 * @rx_handler: receive handler to register
5147 * @rx_handler_data: data pointer that is used by rx handler
5149 * Register a receive handler for a device. This handler will then be
5150 * called from __netif_receive_skb. A negative errno code is returned
5153 * The caller must hold the rtnl_mutex.
5155 * For a general description of rx_handler, see enum rx_handler_result.
5157 int netdev_rx_handler_register(struct net_device *dev,
5158 rx_handler_func_t *rx_handler,
5159 void *rx_handler_data)
5161 if (netdev_is_rx_handler_busy(dev))
5164 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5167 /* Note: rx_handler_data must be set before rx_handler */
5168 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5169 rcu_assign_pointer(dev->rx_handler, rx_handler);
5173 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5176 * netdev_rx_handler_unregister - unregister receive handler
5177 * @dev: device to unregister a handler from
5179 * Unregister a receive handler from a device.
5181 * The caller must hold the rtnl_mutex.
5183 void netdev_rx_handler_unregister(struct net_device *dev)
5187 RCU_INIT_POINTER(dev->rx_handler, NULL);
5188 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5189 * section has a guarantee to see a non NULL rx_handler_data
5193 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5195 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5198 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5199 * the special handling of PFMEMALLOC skbs.
5201 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5203 switch (skb->protocol) {
5204 case htons(ETH_P_ARP):
5205 case htons(ETH_P_IP):
5206 case htons(ETH_P_IPV6):
5207 case htons(ETH_P_8021Q):
5208 case htons(ETH_P_8021AD):
5215 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5216 int *ret, struct net_device *orig_dev)
5218 if (nf_hook_ingress_active(skb)) {
5222 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5227 ingress_retval = nf_hook_ingress(skb);
5229 return ingress_retval;
5234 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5235 struct packet_type **ppt_prev)
5237 struct packet_type *ptype, *pt_prev;
5238 rx_handler_func_t *rx_handler;
5239 struct sk_buff *skb = *pskb;
5240 struct net_device *orig_dev;
5241 bool deliver_exact = false;
5242 int ret = NET_RX_DROP;
5245 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5247 trace_netif_receive_skb(skb);
5249 orig_dev = skb->dev;
5251 skb_reset_network_header(skb);
5252 if (!skb_transport_header_was_set(skb))
5253 skb_reset_transport_header(skb);
5254 skb_reset_mac_len(skb);
5259 skb->skb_iif = skb->dev->ifindex;
5261 __this_cpu_inc(softnet_data.processed);
5263 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5267 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5270 if (ret2 != XDP_PASS) {
5276 if (eth_type_vlan(skb->protocol)) {
5277 skb = skb_vlan_untag(skb);
5282 if (skb_skip_tc_classify(skb))
5288 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5290 ret = deliver_skb(skb, pt_prev, orig_dev);
5294 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5296 ret = deliver_skb(skb, pt_prev, orig_dev);
5301 #ifdef CONFIG_NET_INGRESS
5302 if (static_branch_unlikely(&ingress_needed_key)) {
5303 bool another = false;
5305 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5312 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5316 skb_reset_redirect(skb);
5318 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5321 if (skb_vlan_tag_present(skb)) {
5323 ret = deliver_skb(skb, pt_prev, orig_dev);
5326 if (vlan_do_receive(&skb))
5328 else if (unlikely(!skb))
5332 rx_handler = rcu_dereference(skb->dev->rx_handler);
5335 ret = deliver_skb(skb, pt_prev, orig_dev);
5338 switch (rx_handler(&skb)) {
5339 case RX_HANDLER_CONSUMED:
5340 ret = NET_RX_SUCCESS;
5342 case RX_HANDLER_ANOTHER:
5344 case RX_HANDLER_EXACT:
5345 deliver_exact = true;
5347 case RX_HANDLER_PASS:
5354 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5356 if (skb_vlan_tag_get_id(skb)) {
5357 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5360 skb->pkt_type = PACKET_OTHERHOST;
5361 } else if (eth_type_vlan(skb->protocol)) {
5362 /* Outer header is 802.1P with vlan 0, inner header is
5363 * 802.1Q or 802.1AD and vlan_do_receive() above could
5364 * not find vlan dev for vlan id 0.
5366 __vlan_hwaccel_clear_tag(skb);
5367 skb = skb_vlan_untag(skb);
5370 if (vlan_do_receive(&skb))
5371 /* After stripping off 802.1P header with vlan 0
5372 * vlan dev is found for inner header.
5375 else if (unlikely(!skb))
5378 /* We have stripped outer 802.1P vlan 0 header.
5379 * But could not find vlan dev.
5380 * check again for vlan id to set OTHERHOST.
5384 /* Note: we might in the future use prio bits
5385 * and set skb->priority like in vlan_do_receive()
5386 * For the time being, just ignore Priority Code Point
5388 __vlan_hwaccel_clear_tag(skb);
5391 type = skb->protocol;
5393 /* deliver only exact match when indicated */
5394 if (likely(!deliver_exact)) {
5395 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5396 &ptype_base[ntohs(type) &
5400 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5401 &orig_dev->ptype_specific);
5403 if (unlikely(skb->dev != orig_dev)) {
5404 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5405 &skb->dev->ptype_specific);
5409 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5411 *ppt_prev = pt_prev;
5415 atomic_long_inc(&skb->dev->rx_dropped);
5417 atomic_long_inc(&skb->dev->rx_nohandler);
5419 /* Jamal, now you will not able to escape explaining
5420 * me how you were going to use this. :-)
5426 /* The invariant here is that if *ppt_prev is not NULL
5427 * then skb should also be non-NULL.
5429 * Apparently *ppt_prev assignment above holds this invariant due to
5430 * skb dereferencing near it.
5436 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5438 struct net_device *orig_dev = skb->dev;
5439 struct packet_type *pt_prev = NULL;
5442 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5444 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5445 skb->dev, pt_prev, orig_dev);
5450 * netif_receive_skb_core - special purpose version of netif_receive_skb
5451 * @skb: buffer to process
5453 * More direct receive version of netif_receive_skb(). It should
5454 * only be used by callers that have a need to skip RPS and Generic XDP.
5455 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5457 * This function may only be called from softirq context and interrupts
5458 * should be enabled.
5460 * Return values (usually ignored):
5461 * NET_RX_SUCCESS: no congestion
5462 * NET_RX_DROP: packet was dropped
5464 int netif_receive_skb_core(struct sk_buff *skb)
5469 ret = __netif_receive_skb_one_core(skb, false);
5474 EXPORT_SYMBOL(netif_receive_skb_core);
5476 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5477 struct packet_type *pt_prev,
5478 struct net_device *orig_dev)
5480 struct sk_buff *skb, *next;
5484 if (list_empty(head))
5486 if (pt_prev->list_func != NULL)
5487 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5488 ip_list_rcv, head, pt_prev, orig_dev);
5490 list_for_each_entry_safe(skb, next, head, list) {
5491 skb_list_del_init(skb);
5492 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5496 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5498 /* Fast-path assumptions:
5499 * - There is no RX handler.
5500 * - Only one packet_type matches.
5501 * If either of these fails, we will end up doing some per-packet
5502 * processing in-line, then handling the 'last ptype' for the whole
5503 * sublist. This can't cause out-of-order delivery to any single ptype,
5504 * because the 'last ptype' must be constant across the sublist, and all
5505 * other ptypes are handled per-packet.
5507 /* Current (common) ptype of sublist */
5508 struct packet_type *pt_curr = NULL;
5509 /* Current (common) orig_dev of sublist */
5510 struct net_device *od_curr = NULL;
5511 struct list_head sublist;
5512 struct sk_buff *skb, *next;
5514 INIT_LIST_HEAD(&sublist);
5515 list_for_each_entry_safe(skb, next, head, list) {
5516 struct net_device *orig_dev = skb->dev;
5517 struct packet_type *pt_prev = NULL;
5519 skb_list_del_init(skb);
5520 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5523 if (pt_curr != pt_prev || od_curr != orig_dev) {
5524 /* dispatch old sublist */
5525 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5526 /* start new sublist */
5527 INIT_LIST_HEAD(&sublist);
5531 list_add_tail(&skb->list, &sublist);
5534 /* dispatch final sublist */
5535 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5538 static int __netif_receive_skb(struct sk_buff *skb)
5542 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5543 unsigned int noreclaim_flag;
5546 * PFMEMALLOC skbs are special, they should
5547 * - be delivered to SOCK_MEMALLOC sockets only
5548 * - stay away from userspace
5549 * - have bounded memory usage
5551 * Use PF_MEMALLOC as this saves us from propagating the allocation
5552 * context down to all allocation sites.
5554 noreclaim_flag = memalloc_noreclaim_save();
5555 ret = __netif_receive_skb_one_core(skb, true);
5556 memalloc_noreclaim_restore(noreclaim_flag);
5558 ret = __netif_receive_skb_one_core(skb, false);
5563 static void __netif_receive_skb_list(struct list_head *head)
5565 unsigned long noreclaim_flag = 0;
5566 struct sk_buff *skb, *next;
5567 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5569 list_for_each_entry_safe(skb, next, head, list) {
5570 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5571 struct list_head sublist;
5573 /* Handle the previous sublist */
5574 list_cut_before(&sublist, head, &skb->list);
5575 if (!list_empty(&sublist))
5576 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5577 pfmemalloc = !pfmemalloc;
5578 /* See comments in __netif_receive_skb */
5580 noreclaim_flag = memalloc_noreclaim_save();
5582 memalloc_noreclaim_restore(noreclaim_flag);
5585 /* Handle the remaining sublist */
5586 if (!list_empty(head))
5587 __netif_receive_skb_list_core(head, pfmemalloc);
5588 /* Restore pflags */
5590 memalloc_noreclaim_restore(noreclaim_flag);
5593 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5595 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5596 struct bpf_prog *new = xdp->prog;
5599 switch (xdp->command) {
5600 case XDP_SETUP_PROG:
5601 rcu_assign_pointer(dev->xdp_prog, new);
5606 static_branch_dec(&generic_xdp_needed_key);
5607 } else if (new && !old) {
5608 static_branch_inc(&generic_xdp_needed_key);
5609 dev_disable_lro(dev);
5610 dev_disable_gro_hw(dev);
5622 static int netif_receive_skb_internal(struct sk_buff *skb)
5626 net_timestamp_check(netdev_tstamp_prequeue, skb);
5628 if (skb_defer_rx_timestamp(skb))
5629 return NET_RX_SUCCESS;
5633 if (static_branch_unlikely(&rps_needed)) {
5634 struct rps_dev_flow voidflow, *rflow = &voidflow;
5635 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5638 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5644 ret = __netif_receive_skb(skb);
5649 static void netif_receive_skb_list_internal(struct list_head *head)
5651 struct sk_buff *skb, *next;
5652 struct list_head sublist;
5654 INIT_LIST_HEAD(&sublist);
5655 list_for_each_entry_safe(skb, next, head, list) {
5656 net_timestamp_check(netdev_tstamp_prequeue, skb);
5657 skb_list_del_init(skb);
5658 if (!skb_defer_rx_timestamp(skb))
5659 list_add_tail(&skb->list, &sublist);
5661 list_splice_init(&sublist, head);
5665 if (static_branch_unlikely(&rps_needed)) {
5666 list_for_each_entry_safe(skb, next, head, list) {
5667 struct rps_dev_flow voidflow, *rflow = &voidflow;
5668 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5671 /* Will be handled, remove from list */
5672 skb_list_del_init(skb);
5673 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5678 __netif_receive_skb_list(head);
5683 * netif_receive_skb - process receive buffer from network
5684 * @skb: buffer to process
5686 * netif_receive_skb() is the main receive data processing function.
5687 * It always succeeds. The buffer may be dropped during processing
5688 * for congestion control or by the protocol layers.
5690 * This function may only be called from softirq context and interrupts
5691 * should be enabled.
5693 * Return values (usually ignored):
5694 * NET_RX_SUCCESS: no congestion
5695 * NET_RX_DROP: packet was dropped
5697 int netif_receive_skb(struct sk_buff *skb)
5701 trace_netif_receive_skb_entry(skb);
5703 ret = netif_receive_skb_internal(skb);
5704 trace_netif_receive_skb_exit(ret);
5708 EXPORT_SYMBOL(netif_receive_skb);
5711 * netif_receive_skb_list - process many receive buffers from network
5712 * @head: list of skbs to process.
5714 * Since return value of netif_receive_skb() is normally ignored, and
5715 * wouldn't be meaningful for a list, this function returns void.
5717 * This function may only be called from softirq context and interrupts
5718 * should be enabled.
5720 void netif_receive_skb_list(struct list_head *head)
5722 struct sk_buff *skb;
5724 if (list_empty(head))
5726 if (trace_netif_receive_skb_list_entry_enabled()) {
5727 list_for_each_entry(skb, head, list)
5728 trace_netif_receive_skb_list_entry(skb);
5730 netif_receive_skb_list_internal(head);
5731 trace_netif_receive_skb_list_exit(0);
5733 EXPORT_SYMBOL(netif_receive_skb_list);
5735 static DEFINE_PER_CPU(struct work_struct, flush_works);
5737 /* Network device is going away, flush any packets still pending */
5738 static void flush_backlog(struct work_struct *work)
5740 struct sk_buff *skb, *tmp;
5741 struct softnet_data *sd;
5744 sd = this_cpu_ptr(&softnet_data);
5746 local_irq_disable();
5748 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5749 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5750 __skb_unlink(skb, &sd->input_pkt_queue);
5751 dev_kfree_skb_irq(skb);
5752 input_queue_head_incr(sd);
5758 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5759 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5760 __skb_unlink(skb, &sd->process_queue);
5762 input_queue_head_incr(sd);
5768 static bool flush_required(int cpu)
5770 #if IS_ENABLED(CONFIG_RPS)
5771 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5774 local_irq_disable();
5777 /* as insertion into process_queue happens with the rps lock held,
5778 * process_queue access may race only with dequeue
5780 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5781 !skb_queue_empty_lockless(&sd->process_queue);
5787 /* without RPS we can't safely check input_pkt_queue: during a
5788 * concurrent remote skb_queue_splice() we can detect as empty both
5789 * input_pkt_queue and process_queue even if the latter could end-up
5790 * containing a lot of packets.
5795 static void flush_all_backlogs(void)
5797 static cpumask_t flush_cpus;
5800 /* since we are under rtnl lock protection we can use static data
5801 * for the cpumask and avoid allocating on stack the possibly
5808 cpumask_clear(&flush_cpus);
5809 for_each_online_cpu(cpu) {
5810 if (flush_required(cpu)) {
5811 queue_work_on(cpu, system_highpri_wq,
5812 per_cpu_ptr(&flush_works, cpu));
5813 cpumask_set_cpu(cpu, &flush_cpus);
5817 /* we can have in flight packet[s] on the cpus we are not flushing,
5818 * synchronize_net() in unregister_netdevice_many() will take care of
5821 for_each_cpu(cpu, &flush_cpus)
5822 flush_work(per_cpu_ptr(&flush_works, cpu));
5827 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5828 static void gro_normal_list(struct napi_struct *napi)
5830 if (!napi->rx_count)
5832 netif_receive_skb_list_internal(&napi->rx_list);
5833 INIT_LIST_HEAD(&napi->rx_list);
5837 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5838 * pass the whole batch up to the stack.
5840 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5842 list_add_tail(&skb->list, &napi->rx_list);
5843 napi->rx_count += segs;
5844 if (napi->rx_count >= gro_normal_batch)
5845 gro_normal_list(napi);
5848 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5850 struct packet_offload *ptype;
5851 __be16 type = skb->protocol;
5852 struct list_head *head = &offload_base;
5855 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5857 if (NAPI_GRO_CB(skb)->count == 1) {
5858 skb_shinfo(skb)->gso_size = 0;
5863 list_for_each_entry_rcu(ptype, head, list) {
5864 if (ptype->type != type || !ptype->callbacks.gro_complete)
5867 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5868 ipv6_gro_complete, inet_gro_complete,
5875 WARN_ON(&ptype->list == head);
5881 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5884 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5887 struct list_head *head = &napi->gro_hash[index].list;
5888 struct sk_buff *skb, *p;
5890 list_for_each_entry_safe_reverse(skb, p, head, list) {
5891 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5893 skb_list_del_init(skb);
5894 napi_gro_complete(napi, skb);
5895 napi->gro_hash[index].count--;
5898 if (!napi->gro_hash[index].count)
5899 __clear_bit(index, &napi->gro_bitmask);
5902 /* napi->gro_hash[].list contains packets ordered by age.
5903 * youngest packets at the head of it.
5904 * Complete skbs in reverse order to reduce latencies.
5906 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5908 unsigned long bitmask = napi->gro_bitmask;
5909 unsigned int i, base = ~0U;
5911 while ((i = ffs(bitmask)) != 0) {
5914 __napi_gro_flush_chain(napi, base, flush_old);
5917 EXPORT_SYMBOL(napi_gro_flush);
5919 static void gro_list_prepare(const struct list_head *head,
5920 const struct sk_buff *skb)
5922 unsigned int maclen = skb->dev->hard_header_len;
5923 u32 hash = skb_get_hash_raw(skb);
5926 list_for_each_entry(p, head, list) {
5927 unsigned long diffs;
5929 NAPI_GRO_CB(p)->flush = 0;
5931 if (hash != skb_get_hash_raw(p)) {
5932 NAPI_GRO_CB(p)->same_flow = 0;
5936 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5937 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5938 if (skb_vlan_tag_present(p))
5939 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5940 diffs |= skb_metadata_differs(p, skb);
5941 if (maclen == ETH_HLEN)
5942 diffs |= compare_ether_header(skb_mac_header(p),
5943 skb_mac_header(skb));
5945 diffs = memcmp(skb_mac_header(p),
5946 skb_mac_header(skb),
5949 /* in most common scenarions 'slow_gro' is 0
5950 * otherwise we are already on some slower paths
5951 * either skip all the infrequent tests altogether or
5952 * avoid trying too hard to skip each of them individually
5954 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
5955 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5956 struct tc_skb_ext *skb_ext;
5957 struct tc_skb_ext *p_ext;
5960 diffs |= p->sk != skb->sk;
5961 diffs |= skb_metadata_dst_cmp(p, skb);
5962 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
5964 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5965 skb_ext = skb_ext_find(skb, TC_SKB_EXT);
5966 p_ext = skb_ext_find(p, TC_SKB_EXT);
5968 diffs |= (!!p_ext) ^ (!!skb_ext);
5969 if (!diffs && unlikely(skb_ext))
5970 diffs |= p_ext->chain ^ skb_ext->chain;
5974 NAPI_GRO_CB(p)->same_flow = !diffs;
5978 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
5980 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5981 const skb_frag_t *frag0 = &pinfo->frags[0];
5983 NAPI_GRO_CB(skb)->data_offset = 0;
5984 NAPI_GRO_CB(skb)->frag0 = NULL;
5985 NAPI_GRO_CB(skb)->frag0_len = 0;
5987 if (!skb_headlen(skb) && pinfo->nr_frags &&
5988 !PageHighMem(skb_frag_page(frag0)) &&
5989 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
5990 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5991 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5992 skb_frag_size(frag0),
5993 skb->end - skb->tail);
5997 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5999 struct skb_shared_info *pinfo = skb_shinfo(skb);
6001 BUG_ON(skb->end - skb->tail < grow);
6003 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6005 skb->data_len -= grow;
6008 skb_frag_off_add(&pinfo->frags[0], grow);
6009 skb_frag_size_sub(&pinfo->frags[0], grow);
6011 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6012 skb_frag_unref(skb, 0);
6013 memmove(pinfo->frags, pinfo->frags + 1,
6014 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
6018 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6020 struct sk_buff *oldest;
6022 oldest = list_last_entry(head, struct sk_buff, list);
6024 /* We are called with head length >= MAX_GRO_SKBS, so this is
6027 if (WARN_ON_ONCE(!oldest))
6030 /* Do not adjust napi->gro_hash[].count, caller is adding a new
6033 skb_list_del_init(oldest);
6034 napi_gro_complete(napi, oldest);
6037 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6039 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6040 struct gro_list *gro_list = &napi->gro_hash[bucket];
6041 struct list_head *head = &offload_base;
6042 struct packet_offload *ptype;
6043 __be16 type = skb->protocol;
6044 struct sk_buff *pp = NULL;
6045 enum gro_result ret;
6049 if (netif_elide_gro(skb->dev))
6052 gro_list_prepare(&gro_list->list, skb);
6055 list_for_each_entry_rcu(ptype, head, list) {
6056 if (ptype->type != type || !ptype->callbacks.gro_receive)
6059 skb_set_network_header(skb, skb_gro_offset(skb));
6060 skb_reset_mac_len(skb);
6061 NAPI_GRO_CB(skb)->same_flow = 0;
6062 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6063 NAPI_GRO_CB(skb)->free = 0;
6064 NAPI_GRO_CB(skb)->encap_mark = 0;
6065 NAPI_GRO_CB(skb)->recursion_counter = 0;
6066 NAPI_GRO_CB(skb)->is_fou = 0;
6067 NAPI_GRO_CB(skb)->is_atomic = 1;
6068 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6070 /* Setup for GRO checksum validation */
6071 switch (skb->ip_summed) {
6072 case CHECKSUM_COMPLETE:
6073 NAPI_GRO_CB(skb)->csum = skb->csum;
6074 NAPI_GRO_CB(skb)->csum_valid = 1;
6075 NAPI_GRO_CB(skb)->csum_cnt = 0;
6077 case CHECKSUM_UNNECESSARY:
6078 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6079 NAPI_GRO_CB(skb)->csum_valid = 0;
6082 NAPI_GRO_CB(skb)->csum_cnt = 0;
6083 NAPI_GRO_CB(skb)->csum_valid = 0;
6086 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6087 ipv6_gro_receive, inet_gro_receive,
6088 &gro_list->list, skb);
6093 if (&ptype->list == head)
6096 if (PTR_ERR(pp) == -EINPROGRESS) {
6101 same_flow = NAPI_GRO_CB(skb)->same_flow;
6102 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6105 skb_list_del_init(pp);
6106 napi_gro_complete(napi, pp);
6113 if (NAPI_GRO_CB(skb)->flush)
6116 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6117 gro_flush_oldest(napi, &gro_list->list);
6121 NAPI_GRO_CB(skb)->count = 1;
6122 NAPI_GRO_CB(skb)->age = jiffies;
6123 NAPI_GRO_CB(skb)->last = skb;
6124 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6125 list_add(&skb->list, &gro_list->list);
6129 grow = skb_gro_offset(skb) - skb_headlen(skb);
6131 gro_pull_from_frag0(skb, grow);
6133 if (gro_list->count) {
6134 if (!test_bit(bucket, &napi->gro_bitmask))
6135 __set_bit(bucket, &napi->gro_bitmask);
6136 } else if (test_bit(bucket, &napi->gro_bitmask)) {
6137 __clear_bit(bucket, &napi->gro_bitmask);
6147 struct packet_offload *gro_find_receive_by_type(__be16 type)
6149 struct list_head *offload_head = &offload_base;
6150 struct packet_offload *ptype;
6152 list_for_each_entry_rcu(ptype, offload_head, list) {
6153 if (ptype->type != type || !ptype->callbacks.gro_receive)
6159 EXPORT_SYMBOL(gro_find_receive_by_type);
6161 struct packet_offload *gro_find_complete_by_type(__be16 type)
6163 struct list_head *offload_head = &offload_base;
6164 struct packet_offload *ptype;
6166 list_for_each_entry_rcu(ptype, offload_head, list) {
6167 if (ptype->type != type || !ptype->callbacks.gro_complete)
6173 EXPORT_SYMBOL(gro_find_complete_by_type);
6175 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6176 struct sk_buff *skb,
6181 gro_normal_one(napi, skb, 1);
6184 case GRO_MERGED_FREE:
6185 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6186 napi_skb_free_stolen_head(skb);
6187 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6190 __kfree_skb_defer(skb);
6202 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6206 skb_mark_napi_id(skb, napi);
6207 trace_napi_gro_receive_entry(skb);
6209 skb_gro_reset_offset(skb, 0);
6211 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6212 trace_napi_gro_receive_exit(ret);
6216 EXPORT_SYMBOL(napi_gro_receive);
6218 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6220 if (unlikely(skb->pfmemalloc)) {
6224 __skb_pull(skb, skb_headlen(skb));
6225 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6226 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6227 __vlan_hwaccel_clear_tag(skb);
6228 skb->dev = napi->dev;
6231 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6232 skb->pkt_type = PACKET_HOST;
6234 skb->encapsulation = 0;
6235 skb_shinfo(skb)->gso_type = 0;
6236 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6237 if (unlikely(skb->slow_gro)) {
6247 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6249 struct sk_buff *skb = napi->skb;
6252 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6255 skb_mark_napi_id(skb, napi);
6260 EXPORT_SYMBOL(napi_get_frags);
6262 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6263 struct sk_buff *skb,
6269 __skb_push(skb, ETH_HLEN);
6270 skb->protocol = eth_type_trans(skb, skb->dev);
6271 if (ret == GRO_NORMAL)
6272 gro_normal_one(napi, skb, 1);
6275 case GRO_MERGED_FREE:
6276 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6277 napi_skb_free_stolen_head(skb);
6279 napi_reuse_skb(napi, skb);
6290 /* Upper GRO stack assumes network header starts at gro_offset=0
6291 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6292 * We copy ethernet header into skb->data to have a common layout.
6294 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6296 struct sk_buff *skb = napi->skb;
6297 const struct ethhdr *eth;
6298 unsigned int hlen = sizeof(*eth);
6302 skb_reset_mac_header(skb);
6303 skb_gro_reset_offset(skb, hlen);
6305 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6306 eth = skb_gro_header_slow(skb, hlen, 0);
6307 if (unlikely(!eth)) {
6308 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6309 __func__, napi->dev->name);
6310 napi_reuse_skb(napi, skb);
6314 eth = (const struct ethhdr *)skb->data;
6315 gro_pull_from_frag0(skb, hlen);
6316 NAPI_GRO_CB(skb)->frag0 += hlen;
6317 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6319 __skb_pull(skb, hlen);
6322 * This works because the only protocols we care about don't require
6324 * We'll fix it up properly in napi_frags_finish()
6326 skb->protocol = eth->h_proto;
6331 gro_result_t napi_gro_frags(struct napi_struct *napi)
6334 struct sk_buff *skb = napi_frags_skb(napi);
6336 trace_napi_gro_frags_entry(skb);
6338 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6339 trace_napi_gro_frags_exit(ret);
6343 EXPORT_SYMBOL(napi_gro_frags);
6345 /* Compute the checksum from gro_offset and return the folded value
6346 * after adding in any pseudo checksum.
6348 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6353 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6355 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6356 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6357 /* See comments in __skb_checksum_complete(). */
6359 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6360 !skb->csum_complete_sw)
6361 netdev_rx_csum_fault(skb->dev, skb);
6364 NAPI_GRO_CB(skb)->csum = wsum;
6365 NAPI_GRO_CB(skb)->csum_valid = 1;
6369 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6371 static void net_rps_send_ipi(struct softnet_data *remsd)
6375 struct softnet_data *next = remsd->rps_ipi_next;
6377 if (cpu_online(remsd->cpu))
6378 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6385 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6386 * Note: called with local irq disabled, but exits with local irq enabled.
6388 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6391 struct softnet_data *remsd = sd->rps_ipi_list;
6394 sd->rps_ipi_list = NULL;
6398 /* Send pending IPI's to kick RPS processing on remote cpus. */
6399 net_rps_send_ipi(remsd);
6405 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6408 return sd->rps_ipi_list != NULL;
6414 static int process_backlog(struct napi_struct *napi, int quota)
6416 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6420 /* Check if we have pending ipi, its better to send them now,
6421 * not waiting net_rx_action() end.
6423 if (sd_has_rps_ipi_waiting(sd)) {
6424 local_irq_disable();
6425 net_rps_action_and_irq_enable(sd);
6428 napi->weight = dev_rx_weight;
6430 struct sk_buff *skb;
6432 while ((skb = __skb_dequeue(&sd->process_queue))) {
6434 __netif_receive_skb(skb);
6436 input_queue_head_incr(sd);
6437 if (++work >= quota)
6442 local_irq_disable();
6444 if (skb_queue_empty(&sd->input_pkt_queue)) {
6446 * Inline a custom version of __napi_complete().
6447 * only current cpu owns and manipulates this napi,
6448 * and NAPI_STATE_SCHED is the only possible flag set
6450 * We can use a plain write instead of clear_bit(),
6451 * and we dont need an smp_mb() memory barrier.
6456 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6457 &sd->process_queue);
6467 * __napi_schedule - schedule for receive
6468 * @n: entry to schedule
6470 * The entry's receive function will be scheduled to run.
6471 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6473 void __napi_schedule(struct napi_struct *n)
6475 unsigned long flags;
6477 local_irq_save(flags);
6478 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6479 local_irq_restore(flags);
6481 EXPORT_SYMBOL(__napi_schedule);
6484 * napi_schedule_prep - check if napi can be scheduled
6487 * Test if NAPI routine is already running, and if not mark
6488 * it as running. This is used as a condition variable to
6489 * insure only one NAPI poll instance runs. We also make
6490 * sure there is no pending NAPI disable.
6492 bool napi_schedule_prep(struct napi_struct *n)
6494 unsigned long val, new;
6497 val = READ_ONCE(n->state);
6498 if (unlikely(val & NAPIF_STATE_DISABLE))
6500 new = val | NAPIF_STATE_SCHED;
6502 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6503 * This was suggested by Alexander Duyck, as compiler
6504 * emits better code than :
6505 * if (val & NAPIF_STATE_SCHED)
6506 * new |= NAPIF_STATE_MISSED;
6508 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6510 } while (cmpxchg(&n->state, val, new) != val);
6512 return !(val & NAPIF_STATE_SCHED);
6514 EXPORT_SYMBOL(napi_schedule_prep);
6517 * __napi_schedule_irqoff - schedule for receive
6518 * @n: entry to schedule
6520 * Variant of __napi_schedule() assuming hard irqs are masked.
6522 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6523 * because the interrupt disabled assumption might not be true
6524 * due to force-threaded interrupts and spinlock substitution.
6526 void __napi_schedule_irqoff(struct napi_struct *n)
6528 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6529 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6533 EXPORT_SYMBOL(__napi_schedule_irqoff);
6535 bool napi_complete_done(struct napi_struct *n, int work_done)
6537 unsigned long flags, val, new, timeout = 0;
6541 * 1) Don't let napi dequeue from the cpu poll list
6542 * just in case its running on a different cpu.
6543 * 2) If we are busy polling, do nothing here, we have
6544 * the guarantee we will be called later.
6546 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6547 NAPIF_STATE_IN_BUSY_POLL)))
6552 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6553 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6555 if (n->defer_hard_irqs_count > 0) {
6556 n->defer_hard_irqs_count--;
6557 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6561 if (n->gro_bitmask) {
6562 /* When the NAPI instance uses a timeout and keeps postponing
6563 * it, we need to bound somehow the time packets are kept in
6566 napi_gro_flush(n, !!timeout);
6571 if (unlikely(!list_empty(&n->poll_list))) {
6572 /* If n->poll_list is not empty, we need to mask irqs */
6573 local_irq_save(flags);
6574 list_del_init(&n->poll_list);
6575 local_irq_restore(flags);
6579 val = READ_ONCE(n->state);
6581 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6583 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6584 NAPIF_STATE_SCHED_THREADED |
6585 NAPIF_STATE_PREFER_BUSY_POLL);
6587 /* If STATE_MISSED was set, leave STATE_SCHED set,
6588 * because we will call napi->poll() one more time.
6589 * This C code was suggested by Alexander Duyck to help gcc.
6591 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6593 } while (cmpxchg(&n->state, val, new) != val);
6595 if (unlikely(val & NAPIF_STATE_MISSED)) {
6601 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6602 HRTIMER_MODE_REL_PINNED);
6605 EXPORT_SYMBOL(napi_complete_done);
6607 /* must be called under rcu_read_lock(), as we dont take a reference */
6608 static struct napi_struct *napi_by_id(unsigned int napi_id)
6610 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6611 struct napi_struct *napi;
6613 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6614 if (napi->napi_id == napi_id)
6620 #if defined(CONFIG_NET_RX_BUSY_POLL)
6622 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6624 if (!skip_schedule) {
6625 gro_normal_list(napi);
6626 __napi_schedule(napi);
6630 if (napi->gro_bitmask) {
6631 /* flush too old packets
6632 * If HZ < 1000, flush all packets.
6634 napi_gro_flush(napi, HZ >= 1000);
6637 gro_normal_list(napi);
6638 clear_bit(NAPI_STATE_SCHED, &napi->state);
6641 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6644 bool skip_schedule = false;
6645 unsigned long timeout;
6648 /* Busy polling means there is a high chance device driver hard irq
6649 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6650 * set in napi_schedule_prep().
6651 * Since we are about to call napi->poll() once more, we can safely
6652 * clear NAPI_STATE_MISSED.
6654 * Note: x86 could use a single "lock and ..." instruction
6655 * to perform these two clear_bit()
6657 clear_bit(NAPI_STATE_MISSED, &napi->state);
6658 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6662 if (prefer_busy_poll) {
6663 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6664 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6665 if (napi->defer_hard_irqs_count && timeout) {
6666 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6667 skip_schedule = true;
6671 /* All we really want here is to re-enable device interrupts.
6672 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6674 rc = napi->poll(napi, budget);
6675 /* We can't gro_normal_list() here, because napi->poll() might have
6676 * rearmed the napi (napi_complete_done()) in which case it could
6677 * already be running on another CPU.
6679 trace_napi_poll(napi, rc, budget);
6680 netpoll_poll_unlock(have_poll_lock);
6682 __busy_poll_stop(napi, skip_schedule);
6686 void napi_busy_loop(unsigned int napi_id,
6687 bool (*loop_end)(void *, unsigned long),
6688 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6690 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6691 int (*napi_poll)(struct napi_struct *napi, int budget);
6692 void *have_poll_lock = NULL;
6693 struct napi_struct *napi;
6700 napi = napi_by_id(napi_id);
6710 unsigned long val = READ_ONCE(napi->state);
6712 /* If multiple threads are competing for this napi,
6713 * we avoid dirtying napi->state as much as we can.
6715 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6716 NAPIF_STATE_IN_BUSY_POLL)) {
6717 if (prefer_busy_poll)
6718 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6721 if (cmpxchg(&napi->state, val,
6722 val | NAPIF_STATE_IN_BUSY_POLL |
6723 NAPIF_STATE_SCHED) != val) {
6724 if (prefer_busy_poll)
6725 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6728 have_poll_lock = netpoll_poll_lock(napi);
6729 napi_poll = napi->poll;
6731 work = napi_poll(napi, budget);
6732 trace_napi_poll(napi, work, budget);
6733 gro_normal_list(napi);
6736 __NET_ADD_STATS(dev_net(napi->dev),
6737 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6740 if (!loop_end || loop_end(loop_end_arg, start_time))
6743 if (unlikely(need_resched())) {
6745 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6749 if (loop_end(loop_end_arg, start_time))
6756 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6761 EXPORT_SYMBOL(napi_busy_loop);
6763 #endif /* CONFIG_NET_RX_BUSY_POLL */
6765 static void napi_hash_add(struct napi_struct *napi)
6767 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6770 spin_lock(&napi_hash_lock);
6772 /* 0..NR_CPUS range is reserved for sender_cpu use */
6774 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6775 napi_gen_id = MIN_NAPI_ID;
6776 } while (napi_by_id(napi_gen_id));
6777 napi->napi_id = napi_gen_id;
6779 hlist_add_head_rcu(&napi->napi_hash_node,
6780 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6782 spin_unlock(&napi_hash_lock);
6785 /* Warning : caller is responsible to make sure rcu grace period
6786 * is respected before freeing memory containing @napi
6788 static void napi_hash_del(struct napi_struct *napi)
6790 spin_lock(&napi_hash_lock);
6792 hlist_del_init_rcu(&napi->napi_hash_node);
6794 spin_unlock(&napi_hash_lock);
6797 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6799 struct napi_struct *napi;
6801 napi = container_of(timer, struct napi_struct, timer);
6803 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6804 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6806 if (!napi_disable_pending(napi) &&
6807 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6808 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6809 __napi_schedule_irqoff(napi);
6812 return HRTIMER_NORESTART;
6815 static void init_gro_hash(struct napi_struct *napi)
6819 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6820 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6821 napi->gro_hash[i].count = 0;
6823 napi->gro_bitmask = 0;
6826 int dev_set_threaded(struct net_device *dev, bool threaded)
6828 struct napi_struct *napi;
6831 if (dev->threaded == threaded)
6835 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6836 if (!napi->thread) {
6837 err = napi_kthread_create(napi);
6846 dev->threaded = threaded;
6848 /* Make sure kthread is created before THREADED bit
6851 smp_mb__before_atomic();
6853 /* Setting/unsetting threaded mode on a napi might not immediately
6854 * take effect, if the current napi instance is actively being
6855 * polled. In this case, the switch between threaded mode and
6856 * softirq mode will happen in the next round of napi_schedule().
6857 * This should not cause hiccups/stalls to the live traffic.
6859 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6861 set_bit(NAPI_STATE_THREADED, &napi->state);
6863 clear_bit(NAPI_STATE_THREADED, &napi->state);
6868 EXPORT_SYMBOL(dev_set_threaded);
6870 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6871 int (*poll)(struct napi_struct *, int), int weight)
6873 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6876 INIT_LIST_HEAD(&napi->poll_list);
6877 INIT_HLIST_NODE(&napi->napi_hash_node);
6878 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6879 napi->timer.function = napi_watchdog;
6880 init_gro_hash(napi);
6882 INIT_LIST_HEAD(&napi->rx_list);
6885 if (weight > NAPI_POLL_WEIGHT)
6886 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6888 napi->weight = weight;
6890 #ifdef CONFIG_NETPOLL
6891 napi->poll_owner = -1;
6893 set_bit(NAPI_STATE_SCHED, &napi->state);
6894 set_bit(NAPI_STATE_NPSVC, &napi->state);
6895 list_add_rcu(&napi->dev_list, &dev->napi_list);
6896 napi_hash_add(napi);
6897 /* Create kthread for this napi if dev->threaded is set.
6898 * Clear dev->threaded if kthread creation failed so that
6899 * threaded mode will not be enabled in napi_enable().
6901 if (dev->threaded && napi_kthread_create(napi))
6904 EXPORT_SYMBOL(netif_napi_add);
6906 void napi_disable(struct napi_struct *n)
6908 unsigned long val, new;
6911 set_bit(NAPI_STATE_DISABLE, &n->state);
6914 val = READ_ONCE(n->state);
6915 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6916 usleep_range(20, 200);
6920 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6921 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6922 } while (cmpxchg(&n->state, val, new) != val);
6924 hrtimer_cancel(&n->timer);
6926 clear_bit(NAPI_STATE_DISABLE, &n->state);
6928 EXPORT_SYMBOL(napi_disable);
6931 * napi_enable - enable NAPI scheduling
6934 * Resume NAPI from being scheduled on this context.
6935 * Must be paired with napi_disable.
6937 void napi_enable(struct napi_struct *n)
6939 unsigned long val, new;
6942 val = READ_ONCE(n->state);
6943 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6945 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6946 if (n->dev->threaded && n->thread)
6947 new |= NAPIF_STATE_THREADED;
6948 } while (cmpxchg(&n->state, val, new) != val);
6950 EXPORT_SYMBOL(napi_enable);
6952 static void flush_gro_hash(struct napi_struct *napi)
6956 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6957 struct sk_buff *skb, *n;
6959 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6961 napi->gro_hash[i].count = 0;
6965 /* Must be called in process context */
6966 void __netif_napi_del(struct napi_struct *napi)
6968 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6971 napi_hash_del(napi);
6972 list_del_rcu(&napi->dev_list);
6973 napi_free_frags(napi);
6975 flush_gro_hash(napi);
6976 napi->gro_bitmask = 0;
6979 kthread_stop(napi->thread);
6980 napi->thread = NULL;
6983 EXPORT_SYMBOL(__netif_napi_del);
6985 static int __napi_poll(struct napi_struct *n, bool *repoll)
6991 /* This NAPI_STATE_SCHED test is for avoiding a race
6992 * with netpoll's poll_napi(). Only the entity which
6993 * obtains the lock and sees NAPI_STATE_SCHED set will
6994 * actually make the ->poll() call. Therefore we avoid
6995 * accidentally calling ->poll() when NAPI is not scheduled.
6998 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6999 work = n->poll(n, weight);
7000 trace_napi_poll(n, work, weight);
7003 if (unlikely(work > weight))
7004 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7005 n->poll, work, weight);
7007 if (likely(work < weight))
7010 /* Drivers must not modify the NAPI state if they
7011 * consume the entire weight. In such cases this code
7012 * still "owns" the NAPI instance and therefore can
7013 * move the instance around on the list at-will.
7015 if (unlikely(napi_disable_pending(n))) {
7020 /* The NAPI context has more processing work, but busy-polling
7021 * is preferred. Exit early.
7023 if (napi_prefer_busy_poll(n)) {
7024 if (napi_complete_done(n, work)) {
7025 /* If timeout is not set, we need to make sure
7026 * that the NAPI is re-scheduled.
7033 if (n->gro_bitmask) {
7034 /* flush too old packets
7035 * If HZ < 1000, flush all packets.
7037 napi_gro_flush(n, HZ >= 1000);
7042 /* Some drivers may have called napi_schedule
7043 * prior to exhausting their budget.
7045 if (unlikely(!list_empty(&n->poll_list))) {
7046 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7047 n->dev ? n->dev->name : "backlog");
7056 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7058 bool do_repoll = false;
7062 list_del_init(&n->poll_list);
7064 have = netpoll_poll_lock(n);
7066 work = __napi_poll(n, &do_repoll);
7069 list_add_tail(&n->poll_list, repoll);
7071 netpoll_poll_unlock(have);
7076 static int napi_thread_wait(struct napi_struct *napi)
7080 set_current_state(TASK_INTERRUPTIBLE);
7082 while (!kthread_should_stop()) {
7083 /* Testing SCHED_THREADED bit here to make sure the current
7084 * kthread owns this napi and could poll on this napi.
7085 * Testing SCHED bit is not enough because SCHED bit might be
7086 * set by some other busy poll thread or by napi_disable().
7088 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7089 WARN_ON(!list_empty(&napi->poll_list));
7090 __set_current_state(TASK_RUNNING);
7095 /* woken being true indicates this thread owns this napi. */
7097 set_current_state(TASK_INTERRUPTIBLE);
7099 __set_current_state(TASK_RUNNING);
7104 static int napi_threaded_poll(void *data)
7106 struct napi_struct *napi = data;
7109 while (!napi_thread_wait(napi)) {
7111 bool repoll = false;
7115 have = netpoll_poll_lock(napi);
7116 __napi_poll(napi, &repoll);
7117 netpoll_poll_unlock(have);
7130 static __latent_entropy void net_rx_action(struct softirq_action *h)
7132 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7133 unsigned long time_limit = jiffies +
7134 usecs_to_jiffies(netdev_budget_usecs);
7135 int budget = netdev_budget;
7139 local_irq_disable();
7140 list_splice_init(&sd->poll_list, &list);
7144 struct napi_struct *n;
7146 if (list_empty(&list)) {
7147 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7152 n = list_first_entry(&list, struct napi_struct, poll_list);
7153 budget -= napi_poll(n, &repoll);
7155 /* If softirq window is exhausted then punt.
7156 * Allow this to run for 2 jiffies since which will allow
7157 * an average latency of 1.5/HZ.
7159 if (unlikely(budget <= 0 ||
7160 time_after_eq(jiffies, time_limit))) {
7166 local_irq_disable();
7168 list_splice_tail_init(&sd->poll_list, &list);
7169 list_splice_tail(&repoll, &list);
7170 list_splice(&list, &sd->poll_list);
7171 if (!list_empty(&sd->poll_list))
7172 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7174 net_rps_action_and_irq_enable(sd);
7177 struct netdev_adjacent {
7178 struct net_device *dev;
7180 /* upper master flag, there can only be one master device per list */
7183 /* lookup ignore flag */
7186 /* counter for the number of times this device was added to us */
7189 /* private field for the users */
7192 struct list_head list;
7193 struct rcu_head rcu;
7196 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7197 struct list_head *adj_list)
7199 struct netdev_adjacent *adj;
7201 list_for_each_entry(adj, adj_list, list) {
7202 if (adj->dev == adj_dev)
7208 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7209 struct netdev_nested_priv *priv)
7211 struct net_device *dev = (struct net_device *)priv->data;
7213 return upper_dev == dev;
7217 * netdev_has_upper_dev - Check if device is linked to an upper device
7219 * @upper_dev: upper device to check
7221 * Find out if a device is linked to specified upper device and return true
7222 * in case it is. Note that this checks only immediate upper device,
7223 * not through a complete stack of devices. The caller must hold the RTNL lock.
7225 bool netdev_has_upper_dev(struct net_device *dev,
7226 struct net_device *upper_dev)
7228 struct netdev_nested_priv priv = {
7229 .data = (void *)upper_dev,
7234 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7237 EXPORT_SYMBOL(netdev_has_upper_dev);
7240 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7242 * @upper_dev: upper device to check
7244 * Find out if a device is linked to specified upper device and return true
7245 * in case it is. Note that this checks the entire upper device chain.
7246 * The caller must hold rcu lock.
7249 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7250 struct net_device *upper_dev)
7252 struct netdev_nested_priv priv = {
7253 .data = (void *)upper_dev,
7256 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7259 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7262 * netdev_has_any_upper_dev - Check if device is linked to some device
7265 * Find out if a device is linked to an upper device and return true in case
7266 * it is. The caller must hold the RTNL lock.
7268 bool netdev_has_any_upper_dev(struct net_device *dev)
7272 return !list_empty(&dev->adj_list.upper);
7274 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7277 * netdev_master_upper_dev_get - Get master upper device
7280 * Find a master upper device and return pointer to it or NULL in case
7281 * it's not there. The caller must hold the RTNL lock.
7283 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7285 struct netdev_adjacent *upper;
7289 if (list_empty(&dev->adj_list.upper))
7292 upper = list_first_entry(&dev->adj_list.upper,
7293 struct netdev_adjacent, list);
7294 if (likely(upper->master))
7298 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7300 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7302 struct netdev_adjacent *upper;
7306 if (list_empty(&dev->adj_list.upper))
7309 upper = list_first_entry(&dev->adj_list.upper,
7310 struct netdev_adjacent, list);
7311 if (likely(upper->master) && !upper->ignore)
7317 * netdev_has_any_lower_dev - Check if device is linked to some device
7320 * Find out if a device is linked to a lower device and return true in case
7321 * it is. The caller must hold the RTNL lock.
7323 static bool netdev_has_any_lower_dev(struct net_device *dev)
7327 return !list_empty(&dev->adj_list.lower);
7330 void *netdev_adjacent_get_private(struct list_head *adj_list)
7332 struct netdev_adjacent *adj;
7334 adj = list_entry(adj_list, struct netdev_adjacent, list);
7336 return adj->private;
7338 EXPORT_SYMBOL(netdev_adjacent_get_private);
7341 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7343 * @iter: list_head ** of the current position
7345 * Gets the next device from the dev's upper list, starting from iter
7346 * position. The caller must hold RCU read lock.
7348 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7349 struct list_head **iter)
7351 struct netdev_adjacent *upper;
7353 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7355 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7357 if (&upper->list == &dev->adj_list.upper)
7360 *iter = &upper->list;
7364 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7366 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7367 struct list_head **iter,
7370 struct netdev_adjacent *upper;
7372 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7374 if (&upper->list == &dev->adj_list.upper)
7377 *iter = &upper->list;
7378 *ignore = upper->ignore;
7383 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7384 struct list_head **iter)
7386 struct netdev_adjacent *upper;
7388 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7390 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7392 if (&upper->list == &dev->adj_list.upper)
7395 *iter = &upper->list;
7400 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7401 int (*fn)(struct net_device *dev,
7402 struct netdev_nested_priv *priv),
7403 struct netdev_nested_priv *priv)
7405 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7406 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7411 iter = &dev->adj_list.upper;
7415 ret = fn(now, priv);
7422 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7429 niter = &udev->adj_list.upper;
7430 dev_stack[cur] = now;
7431 iter_stack[cur++] = iter;
7438 next = dev_stack[--cur];
7439 niter = iter_stack[cur];
7449 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7450 int (*fn)(struct net_device *dev,
7451 struct netdev_nested_priv *priv),
7452 struct netdev_nested_priv *priv)
7454 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7455 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7459 iter = &dev->adj_list.upper;
7463 ret = fn(now, priv);
7470 udev = netdev_next_upper_dev_rcu(now, &iter);
7475 niter = &udev->adj_list.upper;
7476 dev_stack[cur] = now;
7477 iter_stack[cur++] = iter;
7484 next = dev_stack[--cur];
7485 niter = iter_stack[cur];
7494 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7496 static bool __netdev_has_upper_dev(struct net_device *dev,
7497 struct net_device *upper_dev)
7499 struct netdev_nested_priv priv = {
7501 .data = (void *)upper_dev,
7506 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7511 * netdev_lower_get_next_private - Get the next ->private from the
7512 * lower neighbour list
7514 * @iter: list_head ** of the current position
7516 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7517 * list, starting from iter position. The caller must hold either hold the
7518 * RTNL lock or its own locking that guarantees that the neighbour lower
7519 * list will remain unchanged.
7521 void *netdev_lower_get_next_private(struct net_device *dev,
7522 struct list_head **iter)
7524 struct netdev_adjacent *lower;
7526 lower = list_entry(*iter, struct netdev_adjacent, list);
7528 if (&lower->list == &dev->adj_list.lower)
7531 *iter = lower->list.next;
7533 return lower->private;
7535 EXPORT_SYMBOL(netdev_lower_get_next_private);
7538 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7539 * lower neighbour list, RCU
7542 * @iter: list_head ** of the current position
7544 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7545 * list, starting from iter position. The caller must hold RCU read lock.
7547 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7548 struct list_head **iter)
7550 struct netdev_adjacent *lower;
7552 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7554 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7556 if (&lower->list == &dev->adj_list.lower)
7559 *iter = &lower->list;
7561 return lower->private;
7563 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7566 * netdev_lower_get_next - Get the next device from the lower neighbour
7569 * @iter: list_head ** of the current position
7571 * Gets the next netdev_adjacent from the dev's lower neighbour
7572 * list, starting from iter position. The caller must hold RTNL lock or
7573 * its own locking that guarantees that the neighbour lower
7574 * list will remain unchanged.
7576 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7578 struct netdev_adjacent *lower;
7580 lower = list_entry(*iter, struct netdev_adjacent, list);
7582 if (&lower->list == &dev->adj_list.lower)
7585 *iter = lower->list.next;
7589 EXPORT_SYMBOL(netdev_lower_get_next);
7591 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7592 struct list_head **iter)
7594 struct netdev_adjacent *lower;
7596 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7598 if (&lower->list == &dev->adj_list.lower)
7601 *iter = &lower->list;
7606 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7607 struct list_head **iter,
7610 struct netdev_adjacent *lower;
7612 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7614 if (&lower->list == &dev->adj_list.lower)
7617 *iter = &lower->list;
7618 *ignore = lower->ignore;
7623 int netdev_walk_all_lower_dev(struct net_device *dev,
7624 int (*fn)(struct net_device *dev,
7625 struct netdev_nested_priv *priv),
7626 struct netdev_nested_priv *priv)
7628 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7629 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7633 iter = &dev->adj_list.lower;
7637 ret = fn(now, priv);
7644 ldev = netdev_next_lower_dev(now, &iter);
7649 niter = &ldev->adj_list.lower;
7650 dev_stack[cur] = now;
7651 iter_stack[cur++] = iter;
7658 next = dev_stack[--cur];
7659 niter = iter_stack[cur];
7668 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7670 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7671 int (*fn)(struct net_device *dev,
7672 struct netdev_nested_priv *priv),
7673 struct netdev_nested_priv *priv)
7675 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7676 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7681 iter = &dev->adj_list.lower;
7685 ret = fn(now, priv);
7692 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7699 niter = &ldev->adj_list.lower;
7700 dev_stack[cur] = now;
7701 iter_stack[cur++] = iter;
7708 next = dev_stack[--cur];
7709 niter = iter_stack[cur];
7719 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7720 struct list_head **iter)
7722 struct netdev_adjacent *lower;
7724 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7725 if (&lower->list == &dev->adj_list.lower)
7728 *iter = &lower->list;
7732 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7734 static u8 __netdev_upper_depth(struct net_device *dev)
7736 struct net_device *udev;
7737 struct list_head *iter;
7741 for (iter = &dev->adj_list.upper,
7742 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7744 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7747 if (max_depth < udev->upper_level)
7748 max_depth = udev->upper_level;
7754 static u8 __netdev_lower_depth(struct net_device *dev)
7756 struct net_device *ldev;
7757 struct list_head *iter;
7761 for (iter = &dev->adj_list.lower,
7762 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7764 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7767 if (max_depth < ldev->lower_level)
7768 max_depth = ldev->lower_level;
7774 static int __netdev_update_upper_level(struct net_device *dev,
7775 struct netdev_nested_priv *__unused)
7777 dev->upper_level = __netdev_upper_depth(dev) + 1;
7781 static int __netdev_update_lower_level(struct net_device *dev,
7782 struct netdev_nested_priv *priv)
7784 dev->lower_level = __netdev_lower_depth(dev) + 1;
7786 #ifdef CONFIG_LOCKDEP
7790 if (priv->flags & NESTED_SYNC_IMM)
7791 dev->nested_level = dev->lower_level - 1;
7792 if (priv->flags & NESTED_SYNC_TODO)
7793 net_unlink_todo(dev);
7798 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7799 int (*fn)(struct net_device *dev,
7800 struct netdev_nested_priv *priv),
7801 struct netdev_nested_priv *priv)
7803 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7804 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7808 iter = &dev->adj_list.lower;
7812 ret = fn(now, priv);
7819 ldev = netdev_next_lower_dev_rcu(now, &iter);
7824 niter = &ldev->adj_list.lower;
7825 dev_stack[cur] = now;
7826 iter_stack[cur++] = iter;
7833 next = dev_stack[--cur];
7834 niter = iter_stack[cur];
7843 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7846 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7847 * lower neighbour list, RCU
7851 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7852 * list. The caller must hold RCU read lock.
7854 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7856 struct netdev_adjacent *lower;
7858 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7859 struct netdev_adjacent, list);
7861 return lower->private;
7864 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7867 * netdev_master_upper_dev_get_rcu - Get master upper device
7870 * Find a master upper device and return pointer to it or NULL in case
7871 * it's not there. The caller must hold the RCU read lock.
7873 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7875 struct netdev_adjacent *upper;
7877 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7878 struct netdev_adjacent, list);
7879 if (upper && likely(upper->master))
7883 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7885 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7886 struct net_device *adj_dev,
7887 struct list_head *dev_list)
7889 char linkname[IFNAMSIZ+7];
7891 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7892 "upper_%s" : "lower_%s", adj_dev->name);
7893 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7896 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7898 struct list_head *dev_list)
7900 char linkname[IFNAMSIZ+7];
7902 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7903 "upper_%s" : "lower_%s", name);
7904 sysfs_remove_link(&(dev->dev.kobj), linkname);
7907 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7908 struct net_device *adj_dev,
7909 struct list_head *dev_list)
7911 return (dev_list == &dev->adj_list.upper ||
7912 dev_list == &dev->adj_list.lower) &&
7913 net_eq(dev_net(dev), dev_net(adj_dev));
7916 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7917 struct net_device *adj_dev,
7918 struct list_head *dev_list,
7919 void *private, bool master)
7921 struct netdev_adjacent *adj;
7924 adj = __netdev_find_adj(adj_dev, dev_list);
7928 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7929 dev->name, adj_dev->name, adj->ref_nr);
7934 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7939 adj->master = master;
7941 adj->private = private;
7942 adj->ignore = false;
7945 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7946 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7948 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7949 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7954 /* Ensure that master link is always the first item in list. */
7956 ret = sysfs_create_link(&(dev->dev.kobj),
7957 &(adj_dev->dev.kobj), "master");
7959 goto remove_symlinks;
7961 list_add_rcu(&adj->list, dev_list);
7963 list_add_tail_rcu(&adj->list, dev_list);
7969 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7970 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7978 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7979 struct net_device *adj_dev,
7981 struct list_head *dev_list)
7983 struct netdev_adjacent *adj;
7985 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7986 dev->name, adj_dev->name, ref_nr);
7988 adj = __netdev_find_adj(adj_dev, dev_list);
7991 pr_err("Adjacency does not exist for device %s from %s\n",
7992 dev->name, adj_dev->name);
7997 if (adj->ref_nr > ref_nr) {
7998 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7999 dev->name, adj_dev->name, ref_nr,
8000 adj->ref_nr - ref_nr);
8001 adj->ref_nr -= ref_nr;
8006 sysfs_remove_link(&(dev->dev.kobj), "master");
8008 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8009 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8011 list_del_rcu(&adj->list);
8012 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8013 adj_dev->name, dev->name, adj_dev->name);
8015 kfree_rcu(adj, rcu);
8018 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8019 struct net_device *upper_dev,
8020 struct list_head *up_list,
8021 struct list_head *down_list,
8022 void *private, bool master)
8026 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8031 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8034 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8041 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8042 struct net_device *upper_dev,
8044 struct list_head *up_list,
8045 struct list_head *down_list)
8047 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8048 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8051 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8052 struct net_device *upper_dev,
8053 void *private, bool master)
8055 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8056 &dev->adj_list.upper,
8057 &upper_dev->adj_list.lower,
8061 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8062 struct net_device *upper_dev)
8064 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8065 &dev->adj_list.upper,
8066 &upper_dev->adj_list.lower);
8069 static int __netdev_upper_dev_link(struct net_device *dev,
8070 struct net_device *upper_dev, bool master,
8071 void *upper_priv, void *upper_info,
8072 struct netdev_nested_priv *priv,
8073 struct netlink_ext_ack *extack)
8075 struct netdev_notifier_changeupper_info changeupper_info = {
8080 .upper_dev = upper_dev,
8083 .upper_info = upper_info,
8085 struct net_device *master_dev;
8090 if (dev == upper_dev)
8093 /* To prevent loops, check if dev is not upper device to upper_dev. */
8094 if (__netdev_has_upper_dev(upper_dev, dev))
8097 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8101 if (__netdev_has_upper_dev(dev, upper_dev))
8104 master_dev = __netdev_master_upper_dev_get(dev);
8106 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8109 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8110 &changeupper_info.info);
8111 ret = notifier_to_errno(ret);
8115 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8120 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8121 &changeupper_info.info);
8122 ret = notifier_to_errno(ret);
8126 __netdev_update_upper_level(dev, NULL);
8127 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8129 __netdev_update_lower_level(upper_dev, priv);
8130 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8136 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8142 * netdev_upper_dev_link - Add a link to the upper device
8144 * @upper_dev: new upper device
8145 * @extack: netlink extended ack
8147 * Adds a link to device which is upper to this one. The caller must hold
8148 * the RTNL lock. On a failure a negative errno code is returned.
8149 * On success the reference counts are adjusted and the function
8152 int netdev_upper_dev_link(struct net_device *dev,
8153 struct net_device *upper_dev,
8154 struct netlink_ext_ack *extack)
8156 struct netdev_nested_priv priv = {
8157 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8161 return __netdev_upper_dev_link(dev, upper_dev, false,
8162 NULL, NULL, &priv, extack);
8164 EXPORT_SYMBOL(netdev_upper_dev_link);
8167 * netdev_master_upper_dev_link - Add a master link to the upper device
8169 * @upper_dev: new upper device
8170 * @upper_priv: upper device private
8171 * @upper_info: upper info to be passed down via notifier
8172 * @extack: netlink extended ack
8174 * Adds a link to device which is upper to this one. In this case, only
8175 * one master upper device can be linked, although other non-master devices
8176 * might be linked as well. The caller must hold the RTNL lock.
8177 * On a failure a negative errno code is returned. On success the reference
8178 * counts are adjusted and the function returns zero.
8180 int netdev_master_upper_dev_link(struct net_device *dev,
8181 struct net_device *upper_dev,
8182 void *upper_priv, void *upper_info,
8183 struct netlink_ext_ack *extack)
8185 struct netdev_nested_priv priv = {
8186 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8190 return __netdev_upper_dev_link(dev, upper_dev, true,
8191 upper_priv, upper_info, &priv, extack);
8193 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8195 static void __netdev_upper_dev_unlink(struct net_device *dev,
8196 struct net_device *upper_dev,
8197 struct netdev_nested_priv *priv)
8199 struct netdev_notifier_changeupper_info changeupper_info = {
8203 .upper_dev = upper_dev,
8209 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8211 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8212 &changeupper_info.info);
8214 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8216 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8217 &changeupper_info.info);
8219 __netdev_update_upper_level(dev, NULL);
8220 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8222 __netdev_update_lower_level(upper_dev, priv);
8223 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8228 * netdev_upper_dev_unlink - Removes a link to upper device
8230 * @upper_dev: new upper device
8232 * Removes a link to device which is upper to this one. The caller must hold
8235 void netdev_upper_dev_unlink(struct net_device *dev,
8236 struct net_device *upper_dev)
8238 struct netdev_nested_priv priv = {
8239 .flags = NESTED_SYNC_TODO,
8243 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8245 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8247 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8248 struct net_device *lower_dev,
8251 struct netdev_adjacent *adj;
8253 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8257 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8262 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8263 struct net_device *lower_dev)
8265 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8268 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8269 struct net_device *lower_dev)
8271 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8274 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8275 struct net_device *new_dev,
8276 struct net_device *dev,
8277 struct netlink_ext_ack *extack)
8279 struct netdev_nested_priv priv = {
8288 if (old_dev && new_dev != old_dev)
8289 netdev_adjacent_dev_disable(dev, old_dev);
8290 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8293 if (old_dev && new_dev != old_dev)
8294 netdev_adjacent_dev_enable(dev, old_dev);
8300 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8302 void netdev_adjacent_change_commit(struct net_device *old_dev,
8303 struct net_device *new_dev,
8304 struct net_device *dev)
8306 struct netdev_nested_priv priv = {
8307 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8311 if (!new_dev || !old_dev)
8314 if (new_dev == old_dev)
8317 netdev_adjacent_dev_enable(dev, old_dev);
8318 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8320 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8322 void netdev_adjacent_change_abort(struct net_device *old_dev,
8323 struct net_device *new_dev,
8324 struct net_device *dev)
8326 struct netdev_nested_priv priv = {
8334 if (old_dev && new_dev != old_dev)
8335 netdev_adjacent_dev_enable(dev, old_dev);
8337 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8339 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8342 * netdev_bonding_info_change - Dispatch event about slave change
8344 * @bonding_info: info to dispatch
8346 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8347 * The caller must hold the RTNL lock.
8349 void netdev_bonding_info_change(struct net_device *dev,
8350 struct netdev_bonding_info *bonding_info)
8352 struct netdev_notifier_bonding_info info = {
8356 memcpy(&info.bonding_info, bonding_info,
8357 sizeof(struct netdev_bonding_info));
8358 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8361 EXPORT_SYMBOL(netdev_bonding_info_change);
8364 * netdev_get_xmit_slave - Get the xmit slave of master device
8367 * @all_slaves: assume all the slaves are active
8369 * The reference counters are not incremented so the caller must be
8370 * careful with locks. The caller must hold RCU lock.
8371 * %NULL is returned if no slave is found.
8374 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8375 struct sk_buff *skb,
8378 const struct net_device_ops *ops = dev->netdev_ops;
8380 if (!ops->ndo_get_xmit_slave)
8382 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8384 EXPORT_SYMBOL(netdev_get_xmit_slave);
8386 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8389 const struct net_device_ops *ops = dev->netdev_ops;
8391 if (!ops->ndo_sk_get_lower_dev)
8393 return ops->ndo_sk_get_lower_dev(dev, sk);
8397 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8401 * %NULL is returned if no lower device is found.
8404 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8407 struct net_device *lower;
8409 lower = netdev_sk_get_lower_dev(dev, sk);
8412 lower = netdev_sk_get_lower_dev(dev, sk);
8417 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8419 static void netdev_adjacent_add_links(struct net_device *dev)
8421 struct netdev_adjacent *iter;
8423 struct net *net = dev_net(dev);
8425 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8426 if (!net_eq(net, dev_net(iter->dev)))
8428 netdev_adjacent_sysfs_add(iter->dev, dev,
8429 &iter->dev->adj_list.lower);
8430 netdev_adjacent_sysfs_add(dev, iter->dev,
8431 &dev->adj_list.upper);
8434 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8435 if (!net_eq(net, dev_net(iter->dev)))
8437 netdev_adjacent_sysfs_add(iter->dev, dev,
8438 &iter->dev->adj_list.upper);
8439 netdev_adjacent_sysfs_add(dev, iter->dev,
8440 &dev->adj_list.lower);
8444 static void netdev_adjacent_del_links(struct net_device *dev)
8446 struct netdev_adjacent *iter;
8448 struct net *net = dev_net(dev);
8450 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8451 if (!net_eq(net, dev_net(iter->dev)))
8453 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8454 &iter->dev->adj_list.lower);
8455 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8456 &dev->adj_list.upper);
8459 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8460 if (!net_eq(net, dev_net(iter->dev)))
8462 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8463 &iter->dev->adj_list.upper);
8464 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8465 &dev->adj_list.lower);
8469 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8471 struct netdev_adjacent *iter;
8473 struct net *net = dev_net(dev);
8475 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8476 if (!net_eq(net, dev_net(iter->dev)))
8478 netdev_adjacent_sysfs_del(iter->dev, oldname,
8479 &iter->dev->adj_list.lower);
8480 netdev_adjacent_sysfs_add(iter->dev, dev,
8481 &iter->dev->adj_list.lower);
8484 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8485 if (!net_eq(net, dev_net(iter->dev)))
8487 netdev_adjacent_sysfs_del(iter->dev, oldname,
8488 &iter->dev->adj_list.upper);
8489 netdev_adjacent_sysfs_add(iter->dev, dev,
8490 &iter->dev->adj_list.upper);
8494 void *netdev_lower_dev_get_private(struct net_device *dev,
8495 struct net_device *lower_dev)
8497 struct netdev_adjacent *lower;
8501 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8505 return lower->private;
8507 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8511 * netdev_lower_state_changed - Dispatch event about lower device state change
8512 * @lower_dev: device
8513 * @lower_state_info: state to dispatch
8515 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8516 * The caller must hold the RTNL lock.
8518 void netdev_lower_state_changed(struct net_device *lower_dev,
8519 void *lower_state_info)
8521 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8522 .info.dev = lower_dev,
8526 changelowerstate_info.lower_state_info = lower_state_info;
8527 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8528 &changelowerstate_info.info);
8530 EXPORT_SYMBOL(netdev_lower_state_changed);
8532 static void dev_change_rx_flags(struct net_device *dev, int flags)
8534 const struct net_device_ops *ops = dev->netdev_ops;
8536 if (ops->ndo_change_rx_flags)
8537 ops->ndo_change_rx_flags(dev, flags);
8540 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8542 unsigned int old_flags = dev->flags;
8548 dev->flags |= IFF_PROMISC;
8549 dev->promiscuity += inc;
8550 if (dev->promiscuity == 0) {
8553 * If inc causes overflow, untouch promisc and return error.
8556 dev->flags &= ~IFF_PROMISC;
8558 dev->promiscuity -= inc;
8559 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8564 if (dev->flags != old_flags) {
8565 pr_info("device %s %s promiscuous mode\n",
8567 dev->flags & IFF_PROMISC ? "entered" : "left");
8568 if (audit_enabled) {
8569 current_uid_gid(&uid, &gid);
8570 audit_log(audit_context(), GFP_ATOMIC,
8571 AUDIT_ANOM_PROMISCUOUS,
8572 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8573 dev->name, (dev->flags & IFF_PROMISC),
8574 (old_flags & IFF_PROMISC),
8575 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8576 from_kuid(&init_user_ns, uid),
8577 from_kgid(&init_user_ns, gid),
8578 audit_get_sessionid(current));
8581 dev_change_rx_flags(dev, IFF_PROMISC);
8584 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8589 * dev_set_promiscuity - update promiscuity count on a device
8593 * Add or remove promiscuity from a device. While the count in the device
8594 * remains above zero the interface remains promiscuous. Once it hits zero
8595 * the device reverts back to normal filtering operation. A negative inc
8596 * value is used to drop promiscuity on the device.
8597 * Return 0 if successful or a negative errno code on error.
8599 int dev_set_promiscuity(struct net_device *dev, int inc)
8601 unsigned int old_flags = dev->flags;
8604 err = __dev_set_promiscuity(dev, inc, true);
8607 if (dev->flags != old_flags)
8608 dev_set_rx_mode(dev);
8611 EXPORT_SYMBOL(dev_set_promiscuity);
8613 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8615 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8619 dev->flags |= IFF_ALLMULTI;
8620 dev->allmulti += inc;
8621 if (dev->allmulti == 0) {
8624 * If inc causes overflow, untouch allmulti and return error.
8627 dev->flags &= ~IFF_ALLMULTI;
8629 dev->allmulti -= inc;
8630 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8635 if (dev->flags ^ old_flags) {
8636 dev_change_rx_flags(dev, IFF_ALLMULTI);
8637 dev_set_rx_mode(dev);
8639 __dev_notify_flags(dev, old_flags,
8640 dev->gflags ^ old_gflags);
8646 * dev_set_allmulti - update allmulti count on a device
8650 * Add or remove reception of all multicast frames to a device. While the
8651 * count in the device remains above zero the interface remains listening
8652 * to all interfaces. Once it hits zero the device reverts back to normal
8653 * filtering operation. A negative @inc value is used to drop the counter
8654 * when releasing a resource needing all multicasts.
8655 * Return 0 if successful or a negative errno code on error.
8658 int dev_set_allmulti(struct net_device *dev, int inc)
8660 return __dev_set_allmulti(dev, inc, true);
8662 EXPORT_SYMBOL(dev_set_allmulti);
8665 * Upload unicast and multicast address lists to device and
8666 * configure RX filtering. When the device doesn't support unicast
8667 * filtering it is put in promiscuous mode while unicast addresses
8670 void __dev_set_rx_mode(struct net_device *dev)
8672 const struct net_device_ops *ops = dev->netdev_ops;
8674 /* dev_open will call this function so the list will stay sane. */
8675 if (!(dev->flags&IFF_UP))
8678 if (!netif_device_present(dev))
8681 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8682 /* Unicast addresses changes may only happen under the rtnl,
8683 * therefore calling __dev_set_promiscuity here is safe.
8685 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8686 __dev_set_promiscuity(dev, 1, false);
8687 dev->uc_promisc = true;
8688 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8689 __dev_set_promiscuity(dev, -1, false);
8690 dev->uc_promisc = false;
8694 if (ops->ndo_set_rx_mode)
8695 ops->ndo_set_rx_mode(dev);
8698 void dev_set_rx_mode(struct net_device *dev)
8700 netif_addr_lock_bh(dev);
8701 __dev_set_rx_mode(dev);
8702 netif_addr_unlock_bh(dev);
8706 * dev_get_flags - get flags reported to userspace
8709 * Get the combination of flag bits exported through APIs to userspace.
8711 unsigned int dev_get_flags(const struct net_device *dev)
8715 flags = (dev->flags & ~(IFF_PROMISC |
8720 (dev->gflags & (IFF_PROMISC |
8723 if (netif_running(dev)) {
8724 if (netif_oper_up(dev))
8725 flags |= IFF_RUNNING;
8726 if (netif_carrier_ok(dev))
8727 flags |= IFF_LOWER_UP;
8728 if (netif_dormant(dev))
8729 flags |= IFF_DORMANT;
8734 EXPORT_SYMBOL(dev_get_flags);
8736 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8737 struct netlink_ext_ack *extack)
8739 unsigned int old_flags = dev->flags;
8745 * Set the flags on our device.
8748 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8749 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8751 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8755 * Load in the correct multicast list now the flags have changed.
8758 if ((old_flags ^ flags) & IFF_MULTICAST)
8759 dev_change_rx_flags(dev, IFF_MULTICAST);
8761 dev_set_rx_mode(dev);
8764 * Have we downed the interface. We handle IFF_UP ourselves
8765 * according to user attempts to set it, rather than blindly
8770 if ((old_flags ^ flags) & IFF_UP) {
8771 if (old_flags & IFF_UP)
8774 ret = __dev_open(dev, extack);
8777 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8778 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8779 unsigned int old_flags = dev->flags;
8781 dev->gflags ^= IFF_PROMISC;
8783 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8784 if (dev->flags != old_flags)
8785 dev_set_rx_mode(dev);
8788 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8789 * is important. Some (broken) drivers set IFF_PROMISC, when
8790 * IFF_ALLMULTI is requested not asking us and not reporting.
8792 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8793 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8795 dev->gflags ^= IFF_ALLMULTI;
8796 __dev_set_allmulti(dev, inc, false);
8802 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8803 unsigned int gchanges)
8805 unsigned int changes = dev->flags ^ old_flags;
8808 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8810 if (changes & IFF_UP) {
8811 if (dev->flags & IFF_UP)
8812 call_netdevice_notifiers(NETDEV_UP, dev);
8814 call_netdevice_notifiers(NETDEV_DOWN, dev);
8817 if (dev->flags & IFF_UP &&
8818 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8819 struct netdev_notifier_change_info change_info = {
8823 .flags_changed = changes,
8826 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8831 * dev_change_flags - change device settings
8833 * @flags: device state flags
8834 * @extack: netlink extended ack
8836 * Change settings on device based state flags. The flags are
8837 * in the userspace exported format.
8839 int dev_change_flags(struct net_device *dev, unsigned int flags,
8840 struct netlink_ext_ack *extack)
8843 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8845 ret = __dev_change_flags(dev, flags, extack);
8849 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8850 __dev_notify_flags(dev, old_flags, changes);
8853 EXPORT_SYMBOL(dev_change_flags);
8855 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8857 const struct net_device_ops *ops = dev->netdev_ops;
8859 if (ops->ndo_change_mtu)
8860 return ops->ndo_change_mtu(dev, new_mtu);
8862 /* Pairs with all the lockless reads of dev->mtu in the stack */
8863 WRITE_ONCE(dev->mtu, new_mtu);
8866 EXPORT_SYMBOL(__dev_set_mtu);
8868 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8869 struct netlink_ext_ack *extack)
8871 /* MTU must be positive, and in range */
8872 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8873 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8877 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8878 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8885 * dev_set_mtu_ext - Change maximum transfer unit
8887 * @new_mtu: new transfer unit
8888 * @extack: netlink extended ack
8890 * Change the maximum transfer size of the network device.
8892 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8893 struct netlink_ext_ack *extack)
8897 if (new_mtu == dev->mtu)
8900 err = dev_validate_mtu(dev, new_mtu, extack);
8904 if (!netif_device_present(dev))
8907 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8908 err = notifier_to_errno(err);
8912 orig_mtu = dev->mtu;
8913 err = __dev_set_mtu(dev, new_mtu);
8916 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8918 err = notifier_to_errno(err);
8920 /* setting mtu back and notifying everyone again,
8921 * so that they have a chance to revert changes.
8923 __dev_set_mtu(dev, orig_mtu);
8924 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8931 int dev_set_mtu(struct net_device *dev, int new_mtu)
8933 struct netlink_ext_ack extack;
8936 memset(&extack, 0, sizeof(extack));
8937 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8938 if (err && extack._msg)
8939 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8942 EXPORT_SYMBOL(dev_set_mtu);
8945 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8947 * @new_len: new tx queue length
8949 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8951 unsigned int orig_len = dev->tx_queue_len;
8954 if (new_len != (unsigned int)new_len)
8957 if (new_len != orig_len) {
8958 dev->tx_queue_len = new_len;
8959 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8960 res = notifier_to_errno(res);
8963 res = dev_qdisc_change_tx_queue_len(dev);
8971 netdev_err(dev, "refused to change device tx_queue_len\n");
8972 dev->tx_queue_len = orig_len;
8977 * dev_set_group - Change group this device belongs to
8979 * @new_group: group this device should belong to
8981 void dev_set_group(struct net_device *dev, int new_group)
8983 dev->group = new_group;
8985 EXPORT_SYMBOL(dev_set_group);
8988 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8990 * @addr: new address
8991 * @extack: netlink extended ack
8993 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8994 struct netlink_ext_ack *extack)
8996 struct netdev_notifier_pre_changeaddr_info info = {
8998 .info.extack = extack,
9003 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9004 return notifier_to_errno(rc);
9006 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9009 * dev_set_mac_address - Change Media Access Control Address
9012 * @extack: netlink extended ack
9014 * Change the hardware (MAC) address of the device
9016 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9017 struct netlink_ext_ack *extack)
9019 const struct net_device_ops *ops = dev->netdev_ops;
9022 if (!ops->ndo_set_mac_address)
9024 if (sa->sa_family != dev->type)
9026 if (!netif_device_present(dev))
9028 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9031 err = ops->ndo_set_mac_address(dev, sa);
9034 dev->addr_assign_type = NET_ADDR_SET;
9035 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9036 add_device_randomness(dev->dev_addr, dev->addr_len);
9039 EXPORT_SYMBOL(dev_set_mac_address);
9041 static DECLARE_RWSEM(dev_addr_sem);
9043 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9044 struct netlink_ext_ack *extack)
9048 down_write(&dev_addr_sem);
9049 ret = dev_set_mac_address(dev, sa, extack);
9050 up_write(&dev_addr_sem);
9053 EXPORT_SYMBOL(dev_set_mac_address_user);
9055 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9057 size_t size = sizeof(sa->sa_data);
9058 struct net_device *dev;
9061 down_read(&dev_addr_sem);
9064 dev = dev_get_by_name_rcu(net, dev_name);
9070 memset(sa->sa_data, 0, size);
9072 memcpy(sa->sa_data, dev->dev_addr,
9073 min_t(size_t, size, dev->addr_len));
9074 sa->sa_family = dev->type;
9078 up_read(&dev_addr_sem);
9081 EXPORT_SYMBOL(dev_get_mac_address);
9084 * dev_change_carrier - Change device carrier
9086 * @new_carrier: new value
9088 * Change device carrier
9090 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9092 const struct net_device_ops *ops = dev->netdev_ops;
9094 if (!ops->ndo_change_carrier)
9096 if (!netif_device_present(dev))
9098 return ops->ndo_change_carrier(dev, new_carrier);
9100 EXPORT_SYMBOL(dev_change_carrier);
9103 * dev_get_phys_port_id - Get device physical port ID
9107 * Get device physical port ID
9109 int dev_get_phys_port_id(struct net_device *dev,
9110 struct netdev_phys_item_id *ppid)
9112 const struct net_device_ops *ops = dev->netdev_ops;
9114 if (!ops->ndo_get_phys_port_id)
9116 return ops->ndo_get_phys_port_id(dev, ppid);
9118 EXPORT_SYMBOL(dev_get_phys_port_id);
9121 * dev_get_phys_port_name - Get device physical port name
9124 * @len: limit of bytes to copy to name
9126 * Get device physical port name
9128 int dev_get_phys_port_name(struct net_device *dev,
9129 char *name, size_t len)
9131 const struct net_device_ops *ops = dev->netdev_ops;
9134 if (ops->ndo_get_phys_port_name) {
9135 err = ops->ndo_get_phys_port_name(dev, name, len);
9136 if (err != -EOPNOTSUPP)
9139 return devlink_compat_phys_port_name_get(dev, name, len);
9141 EXPORT_SYMBOL(dev_get_phys_port_name);
9144 * dev_get_port_parent_id - Get the device's port parent identifier
9145 * @dev: network device
9146 * @ppid: pointer to a storage for the port's parent identifier
9147 * @recurse: allow/disallow recursion to lower devices
9149 * Get the devices's port parent identifier
9151 int dev_get_port_parent_id(struct net_device *dev,
9152 struct netdev_phys_item_id *ppid,
9155 const struct net_device_ops *ops = dev->netdev_ops;
9156 struct netdev_phys_item_id first = { };
9157 struct net_device *lower_dev;
9158 struct list_head *iter;
9161 if (ops->ndo_get_port_parent_id) {
9162 err = ops->ndo_get_port_parent_id(dev, ppid);
9163 if (err != -EOPNOTSUPP)
9167 err = devlink_compat_switch_id_get(dev, ppid);
9168 if (!err || err != -EOPNOTSUPP)
9174 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9175 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9180 else if (memcmp(&first, ppid, sizeof(*ppid)))
9186 EXPORT_SYMBOL(dev_get_port_parent_id);
9189 * netdev_port_same_parent_id - Indicate if two network devices have
9190 * the same port parent identifier
9191 * @a: first network device
9192 * @b: second network device
9194 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9196 struct netdev_phys_item_id a_id = { };
9197 struct netdev_phys_item_id b_id = { };
9199 if (dev_get_port_parent_id(a, &a_id, true) ||
9200 dev_get_port_parent_id(b, &b_id, true))
9203 return netdev_phys_item_id_same(&a_id, &b_id);
9205 EXPORT_SYMBOL(netdev_port_same_parent_id);
9208 * dev_change_proto_down - update protocol port state information
9210 * @proto_down: new value
9212 * This info can be used by switch drivers to set the phys state of the
9215 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9217 const struct net_device_ops *ops = dev->netdev_ops;
9219 if (!ops->ndo_change_proto_down)
9221 if (!netif_device_present(dev))
9223 return ops->ndo_change_proto_down(dev, proto_down);
9225 EXPORT_SYMBOL(dev_change_proto_down);
9228 * dev_change_proto_down_generic - generic implementation for
9229 * ndo_change_proto_down that sets carrier according to
9233 * @proto_down: new value
9235 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9238 netif_carrier_off(dev);
9240 netif_carrier_on(dev);
9241 dev->proto_down = proto_down;
9244 EXPORT_SYMBOL(dev_change_proto_down_generic);
9247 * dev_change_proto_down_reason - proto down reason
9250 * @mask: proto down mask
9251 * @value: proto down value
9253 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9259 dev->proto_down_reason = value;
9261 for_each_set_bit(b, &mask, 32) {
9262 if (value & (1 << b))
9263 dev->proto_down_reason |= BIT(b);
9265 dev->proto_down_reason &= ~BIT(b);
9269 EXPORT_SYMBOL(dev_change_proto_down_reason);
9271 struct bpf_xdp_link {
9272 struct bpf_link link;
9273 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9277 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9279 if (flags & XDP_FLAGS_HW_MODE)
9281 if (flags & XDP_FLAGS_DRV_MODE)
9282 return XDP_MODE_DRV;
9283 if (flags & XDP_FLAGS_SKB_MODE)
9284 return XDP_MODE_SKB;
9285 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9288 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9292 return generic_xdp_install;
9295 return dev->netdev_ops->ndo_bpf;
9301 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9302 enum bpf_xdp_mode mode)
9304 return dev->xdp_state[mode].link;
9307 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9308 enum bpf_xdp_mode mode)
9310 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9313 return link->link.prog;
9314 return dev->xdp_state[mode].prog;
9317 u8 dev_xdp_prog_count(struct net_device *dev)
9322 for (i = 0; i < __MAX_XDP_MODE; i++)
9323 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9327 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9329 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9331 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9333 return prog ? prog->aux->id : 0;
9336 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9337 struct bpf_xdp_link *link)
9339 dev->xdp_state[mode].link = link;
9340 dev->xdp_state[mode].prog = NULL;
9343 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9344 struct bpf_prog *prog)
9346 dev->xdp_state[mode].link = NULL;
9347 dev->xdp_state[mode].prog = prog;
9350 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9351 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9352 u32 flags, struct bpf_prog *prog)
9354 struct netdev_bpf xdp;
9357 memset(&xdp, 0, sizeof(xdp));
9358 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9359 xdp.extack = extack;
9363 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9364 * "moved" into driver), so they don't increment it on their own, but
9365 * they do decrement refcnt when program is detached or replaced.
9366 * Given net_device also owns link/prog, we need to bump refcnt here
9367 * to prevent drivers from underflowing it.
9371 err = bpf_op(dev, &xdp);
9378 if (mode != XDP_MODE_HW)
9379 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9384 static void dev_xdp_uninstall(struct net_device *dev)
9386 struct bpf_xdp_link *link;
9387 struct bpf_prog *prog;
9388 enum bpf_xdp_mode mode;
9393 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9394 prog = dev_xdp_prog(dev, mode);
9398 bpf_op = dev_xdp_bpf_op(dev, mode);
9402 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9404 /* auto-detach link from net device */
9405 link = dev_xdp_link(dev, mode);
9411 dev_xdp_set_link(dev, mode, NULL);
9415 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9416 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9417 struct bpf_prog *old_prog, u32 flags)
9419 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9420 struct bpf_prog *cur_prog;
9421 struct net_device *upper;
9422 struct list_head *iter;
9423 enum bpf_xdp_mode mode;
9429 /* either link or prog attachment, never both */
9430 if (link && (new_prog || old_prog))
9432 /* link supports only XDP mode flags */
9433 if (link && (flags & ~XDP_FLAGS_MODES)) {
9434 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9437 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9438 if (num_modes > 1) {
9439 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9442 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9443 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9444 NL_SET_ERR_MSG(extack,
9445 "More than one program loaded, unset mode is ambiguous");
9448 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9449 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9450 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9454 mode = dev_xdp_mode(dev, flags);
9455 /* can't replace attached link */
9456 if (dev_xdp_link(dev, mode)) {
9457 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9461 /* don't allow if an upper device already has a program */
9462 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9463 if (dev_xdp_prog_count(upper) > 0) {
9464 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9469 cur_prog = dev_xdp_prog(dev, mode);
9470 /* can't replace attached prog with link */
9471 if (link && cur_prog) {
9472 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9475 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9476 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9480 /* put effective new program into new_prog */
9482 new_prog = link->link.prog;
9485 bool offload = mode == XDP_MODE_HW;
9486 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9487 ? XDP_MODE_DRV : XDP_MODE_SKB;
9489 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9490 NL_SET_ERR_MSG(extack, "XDP program already attached");
9493 if (!offload && dev_xdp_prog(dev, other_mode)) {
9494 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9497 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9498 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9501 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9502 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9505 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9506 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9511 /* don't call drivers if the effective program didn't change */
9512 if (new_prog != cur_prog) {
9513 bpf_op = dev_xdp_bpf_op(dev, mode);
9515 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9519 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9525 dev_xdp_set_link(dev, mode, link);
9527 dev_xdp_set_prog(dev, mode, new_prog);
9529 bpf_prog_put(cur_prog);
9534 static int dev_xdp_attach_link(struct net_device *dev,
9535 struct netlink_ext_ack *extack,
9536 struct bpf_xdp_link *link)
9538 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9541 static int dev_xdp_detach_link(struct net_device *dev,
9542 struct netlink_ext_ack *extack,
9543 struct bpf_xdp_link *link)
9545 enum bpf_xdp_mode mode;
9550 mode = dev_xdp_mode(dev, link->flags);
9551 if (dev_xdp_link(dev, mode) != link)
9554 bpf_op = dev_xdp_bpf_op(dev, mode);
9555 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9556 dev_xdp_set_link(dev, mode, NULL);
9560 static void bpf_xdp_link_release(struct bpf_link *link)
9562 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9566 /* if racing with net_device's tear down, xdp_link->dev might be
9567 * already NULL, in which case link was already auto-detached
9569 if (xdp_link->dev) {
9570 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9571 xdp_link->dev = NULL;
9577 static int bpf_xdp_link_detach(struct bpf_link *link)
9579 bpf_xdp_link_release(link);
9583 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9585 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9590 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9591 struct seq_file *seq)
9593 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9598 ifindex = xdp_link->dev->ifindex;
9601 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9604 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9605 struct bpf_link_info *info)
9607 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9612 ifindex = xdp_link->dev->ifindex;
9615 info->xdp.ifindex = ifindex;
9619 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9620 struct bpf_prog *old_prog)
9622 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9623 enum bpf_xdp_mode mode;
9629 /* link might have been auto-released already, so fail */
9630 if (!xdp_link->dev) {
9635 if (old_prog && link->prog != old_prog) {
9639 old_prog = link->prog;
9640 if (old_prog == new_prog) {
9641 /* no-op, don't disturb drivers */
9642 bpf_prog_put(new_prog);
9646 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9647 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9648 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9649 xdp_link->flags, new_prog);
9653 old_prog = xchg(&link->prog, new_prog);
9654 bpf_prog_put(old_prog);
9661 static const struct bpf_link_ops bpf_xdp_link_lops = {
9662 .release = bpf_xdp_link_release,
9663 .dealloc = bpf_xdp_link_dealloc,
9664 .detach = bpf_xdp_link_detach,
9665 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9666 .fill_link_info = bpf_xdp_link_fill_link_info,
9667 .update_prog = bpf_xdp_link_update,
9670 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9672 struct net *net = current->nsproxy->net_ns;
9673 struct bpf_link_primer link_primer;
9674 struct bpf_xdp_link *link;
9675 struct net_device *dev;
9679 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9685 link = kzalloc(sizeof(*link), GFP_USER);
9691 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9693 link->flags = attr->link_create.flags;
9695 err = bpf_link_prime(&link->link, &link_primer);
9701 err = dev_xdp_attach_link(dev, NULL, link);
9706 bpf_link_cleanup(&link_primer);
9710 fd = bpf_link_settle(&link_primer);
9711 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9724 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9726 * @extack: netlink extended ack
9727 * @fd: new program fd or negative value to clear
9728 * @expected_fd: old program fd that userspace expects to replace or clear
9729 * @flags: xdp-related flags
9731 * Set or clear a bpf program for a device
9733 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9734 int fd, int expected_fd, u32 flags)
9736 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9737 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9743 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9744 mode != XDP_MODE_SKB);
9745 if (IS_ERR(new_prog))
9746 return PTR_ERR(new_prog);
9749 if (expected_fd >= 0) {
9750 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9751 mode != XDP_MODE_SKB);
9752 if (IS_ERR(old_prog)) {
9753 err = PTR_ERR(old_prog);
9759 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9762 if (err && new_prog)
9763 bpf_prog_put(new_prog);
9765 bpf_prog_put(old_prog);
9770 * dev_new_index - allocate an ifindex
9771 * @net: the applicable net namespace
9773 * Returns a suitable unique value for a new device interface
9774 * number. The caller must hold the rtnl semaphore or the
9775 * dev_base_lock to be sure it remains unique.
9777 static int dev_new_index(struct net *net)
9779 int ifindex = net->ifindex;
9784 if (!__dev_get_by_index(net, ifindex))
9785 return net->ifindex = ifindex;
9789 /* Delayed registration/unregisteration */
9790 static LIST_HEAD(net_todo_list);
9791 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9793 static void net_set_todo(struct net_device *dev)
9795 list_add_tail(&dev->todo_list, &net_todo_list);
9796 dev_net(dev)->dev_unreg_count++;
9799 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9800 struct net_device *upper, netdev_features_t features)
9802 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9803 netdev_features_t feature;
9806 for_each_netdev_feature(upper_disables, feature_bit) {
9807 feature = __NETIF_F_BIT(feature_bit);
9808 if (!(upper->wanted_features & feature)
9809 && (features & feature)) {
9810 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9811 &feature, upper->name);
9812 features &= ~feature;
9819 static void netdev_sync_lower_features(struct net_device *upper,
9820 struct net_device *lower, netdev_features_t features)
9822 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9823 netdev_features_t feature;
9826 for_each_netdev_feature(upper_disables, feature_bit) {
9827 feature = __NETIF_F_BIT(feature_bit);
9828 if (!(features & feature) && (lower->features & feature)) {
9829 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9830 &feature, lower->name);
9831 lower->wanted_features &= ~feature;
9832 __netdev_update_features(lower);
9834 if (unlikely(lower->features & feature))
9835 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9836 &feature, lower->name);
9838 netdev_features_change(lower);
9843 static netdev_features_t netdev_fix_features(struct net_device *dev,
9844 netdev_features_t features)
9846 /* Fix illegal checksum combinations */
9847 if ((features & NETIF_F_HW_CSUM) &&
9848 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9849 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9850 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9853 /* TSO requires that SG is present as well. */
9854 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9855 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9856 features &= ~NETIF_F_ALL_TSO;
9859 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9860 !(features & NETIF_F_IP_CSUM)) {
9861 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9862 features &= ~NETIF_F_TSO;
9863 features &= ~NETIF_F_TSO_ECN;
9866 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9867 !(features & NETIF_F_IPV6_CSUM)) {
9868 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9869 features &= ~NETIF_F_TSO6;
9872 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9873 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9874 features &= ~NETIF_F_TSO_MANGLEID;
9876 /* TSO ECN requires that TSO is present as well. */
9877 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9878 features &= ~NETIF_F_TSO_ECN;
9880 /* Software GSO depends on SG. */
9881 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9882 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9883 features &= ~NETIF_F_GSO;
9886 /* GSO partial features require GSO partial be set */
9887 if ((features & dev->gso_partial_features) &&
9888 !(features & NETIF_F_GSO_PARTIAL)) {
9890 "Dropping partially supported GSO features since no GSO partial.\n");
9891 features &= ~dev->gso_partial_features;
9894 if (!(features & NETIF_F_RXCSUM)) {
9895 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9896 * successfully merged by hardware must also have the
9897 * checksum verified by hardware. If the user does not
9898 * want to enable RXCSUM, logically, we should disable GRO_HW.
9900 if (features & NETIF_F_GRO_HW) {
9901 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9902 features &= ~NETIF_F_GRO_HW;
9906 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9907 if (features & NETIF_F_RXFCS) {
9908 if (features & NETIF_F_LRO) {
9909 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9910 features &= ~NETIF_F_LRO;
9913 if (features & NETIF_F_GRO_HW) {
9914 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9915 features &= ~NETIF_F_GRO_HW;
9919 if (features & NETIF_F_HW_TLS_TX) {
9920 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9921 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9922 bool hw_csum = features & NETIF_F_HW_CSUM;
9924 if (!ip_csum && !hw_csum) {
9925 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9926 features &= ~NETIF_F_HW_TLS_TX;
9930 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9931 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9932 features &= ~NETIF_F_HW_TLS_RX;
9938 int __netdev_update_features(struct net_device *dev)
9940 struct net_device *upper, *lower;
9941 netdev_features_t features;
9942 struct list_head *iter;
9947 features = netdev_get_wanted_features(dev);
9949 if (dev->netdev_ops->ndo_fix_features)
9950 features = dev->netdev_ops->ndo_fix_features(dev, features);
9952 /* driver might be less strict about feature dependencies */
9953 features = netdev_fix_features(dev, features);
9955 /* some features can't be enabled if they're off on an upper device */
9956 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9957 features = netdev_sync_upper_features(dev, upper, features);
9959 if (dev->features == features)
9962 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9963 &dev->features, &features);
9965 if (dev->netdev_ops->ndo_set_features)
9966 err = dev->netdev_ops->ndo_set_features(dev, features);
9970 if (unlikely(err < 0)) {
9972 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9973 err, &features, &dev->features);
9974 /* return non-0 since some features might have changed and
9975 * it's better to fire a spurious notification than miss it
9981 /* some features must be disabled on lower devices when disabled
9982 * on an upper device (think: bonding master or bridge)
9984 netdev_for_each_lower_dev(dev, lower, iter)
9985 netdev_sync_lower_features(dev, lower, features);
9988 netdev_features_t diff = features ^ dev->features;
9990 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9991 /* udp_tunnel_{get,drop}_rx_info both need
9992 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9993 * device, or they won't do anything.
9994 * Thus we need to update dev->features
9995 * *before* calling udp_tunnel_get_rx_info,
9996 * but *after* calling udp_tunnel_drop_rx_info.
9998 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9999 dev->features = features;
10000 udp_tunnel_get_rx_info(dev);
10002 udp_tunnel_drop_rx_info(dev);
10006 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10007 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10008 dev->features = features;
10009 err |= vlan_get_rx_ctag_filter_info(dev);
10011 vlan_drop_rx_ctag_filter_info(dev);
10015 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10016 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10017 dev->features = features;
10018 err |= vlan_get_rx_stag_filter_info(dev);
10020 vlan_drop_rx_stag_filter_info(dev);
10024 dev->features = features;
10027 return err < 0 ? 0 : 1;
10031 * netdev_update_features - recalculate device features
10032 * @dev: the device to check
10034 * Recalculate dev->features set and send notifications if it
10035 * has changed. Should be called after driver or hardware dependent
10036 * conditions might have changed that influence the features.
10038 void netdev_update_features(struct net_device *dev)
10040 if (__netdev_update_features(dev))
10041 netdev_features_change(dev);
10043 EXPORT_SYMBOL(netdev_update_features);
10046 * netdev_change_features - recalculate device features
10047 * @dev: the device to check
10049 * Recalculate dev->features set and send notifications even
10050 * if they have not changed. Should be called instead of
10051 * netdev_update_features() if also dev->vlan_features might
10052 * have changed to allow the changes to be propagated to stacked
10055 void netdev_change_features(struct net_device *dev)
10057 __netdev_update_features(dev);
10058 netdev_features_change(dev);
10060 EXPORT_SYMBOL(netdev_change_features);
10063 * netif_stacked_transfer_operstate - transfer operstate
10064 * @rootdev: the root or lower level device to transfer state from
10065 * @dev: the device to transfer operstate to
10067 * Transfer operational state from root to device. This is normally
10068 * called when a stacking relationship exists between the root
10069 * device and the device(a leaf device).
10071 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10072 struct net_device *dev)
10074 if (rootdev->operstate == IF_OPER_DORMANT)
10075 netif_dormant_on(dev);
10077 netif_dormant_off(dev);
10079 if (rootdev->operstate == IF_OPER_TESTING)
10080 netif_testing_on(dev);
10082 netif_testing_off(dev);
10084 if (netif_carrier_ok(rootdev))
10085 netif_carrier_on(dev);
10087 netif_carrier_off(dev);
10089 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10091 static int netif_alloc_rx_queues(struct net_device *dev)
10093 unsigned int i, count = dev->num_rx_queues;
10094 struct netdev_rx_queue *rx;
10095 size_t sz = count * sizeof(*rx);
10100 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10106 for (i = 0; i < count; i++) {
10109 /* XDP RX-queue setup */
10110 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10117 /* Rollback successful reg's and free other resources */
10119 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10125 static void netif_free_rx_queues(struct net_device *dev)
10127 unsigned int i, count = dev->num_rx_queues;
10129 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10133 for (i = 0; i < count; i++)
10134 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10139 static void netdev_init_one_queue(struct net_device *dev,
10140 struct netdev_queue *queue, void *_unused)
10142 /* Initialize queue lock */
10143 spin_lock_init(&queue->_xmit_lock);
10144 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10145 queue->xmit_lock_owner = -1;
10146 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10149 dql_init(&queue->dql, HZ);
10153 static void netif_free_tx_queues(struct net_device *dev)
10158 static int netif_alloc_netdev_queues(struct net_device *dev)
10160 unsigned int count = dev->num_tx_queues;
10161 struct netdev_queue *tx;
10162 size_t sz = count * sizeof(*tx);
10164 if (count < 1 || count > 0xffff)
10167 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10173 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10174 spin_lock_init(&dev->tx_global_lock);
10179 void netif_tx_stop_all_queues(struct net_device *dev)
10183 for (i = 0; i < dev->num_tx_queues; i++) {
10184 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10186 netif_tx_stop_queue(txq);
10189 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10192 * register_netdevice - register a network device
10193 * @dev: device to register
10195 * Take a completed network device structure and add it to the kernel
10196 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10197 * chain. 0 is returned on success. A negative errno code is returned
10198 * on a failure to set up the device, or if the name is a duplicate.
10200 * Callers must hold the rtnl semaphore. You may want
10201 * register_netdev() instead of this.
10204 * The locking appears insufficient to guarantee two parallel registers
10205 * will not get the same name.
10208 int register_netdevice(struct net_device *dev)
10211 struct net *net = dev_net(dev);
10213 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10214 NETDEV_FEATURE_COUNT);
10215 BUG_ON(dev_boot_phase);
10220 /* When net_device's are persistent, this will be fatal. */
10221 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10224 ret = ethtool_check_ops(dev->ethtool_ops);
10228 spin_lock_init(&dev->addr_list_lock);
10229 netdev_set_addr_lockdep_class(dev);
10231 ret = dev_get_valid_name(net, dev, dev->name);
10236 dev->name_node = netdev_name_node_head_alloc(dev);
10237 if (!dev->name_node)
10240 /* Init, if this function is available */
10241 if (dev->netdev_ops->ndo_init) {
10242 ret = dev->netdev_ops->ndo_init(dev);
10246 goto err_free_name;
10250 if (((dev->hw_features | dev->features) &
10251 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10252 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10253 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10254 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10261 dev->ifindex = dev_new_index(net);
10262 else if (__dev_get_by_index(net, dev->ifindex))
10265 /* Transfer changeable features to wanted_features and enable
10266 * software offloads (GSO and GRO).
10268 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10269 dev->features |= NETIF_F_SOFT_FEATURES;
10271 if (dev->udp_tunnel_nic_info) {
10272 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10273 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10276 dev->wanted_features = dev->features & dev->hw_features;
10278 if (!(dev->flags & IFF_LOOPBACK))
10279 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10281 /* If IPv4 TCP segmentation offload is supported we should also
10282 * allow the device to enable segmenting the frame with the option
10283 * of ignoring a static IP ID value. This doesn't enable the
10284 * feature itself but allows the user to enable it later.
10286 if (dev->hw_features & NETIF_F_TSO)
10287 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10288 if (dev->vlan_features & NETIF_F_TSO)
10289 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10290 if (dev->mpls_features & NETIF_F_TSO)
10291 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10292 if (dev->hw_enc_features & NETIF_F_TSO)
10293 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10295 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10297 dev->vlan_features |= NETIF_F_HIGHDMA;
10299 /* Make NETIF_F_SG inheritable to tunnel devices.
10301 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10303 /* Make NETIF_F_SG inheritable to MPLS.
10305 dev->mpls_features |= NETIF_F_SG;
10307 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10308 ret = notifier_to_errno(ret);
10312 ret = netdev_register_kobject(dev);
10314 dev->reg_state = NETREG_UNREGISTERED;
10317 dev->reg_state = NETREG_REGISTERED;
10319 __netdev_update_features(dev);
10322 * Default initial state at registry is that the
10323 * device is present.
10326 set_bit(__LINK_STATE_PRESENT, &dev->state);
10328 linkwatch_init_dev(dev);
10330 dev_init_scheduler(dev);
10332 list_netdevice(dev);
10333 add_device_randomness(dev->dev_addr, dev->addr_len);
10335 /* If the device has permanent device address, driver should
10336 * set dev_addr and also addr_assign_type should be set to
10337 * NET_ADDR_PERM (default value).
10339 if (dev->addr_assign_type == NET_ADDR_PERM)
10340 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10342 /* Notify protocols, that a new device appeared. */
10343 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10344 ret = notifier_to_errno(ret);
10346 /* Expect explicit free_netdev() on failure */
10347 dev->needs_free_netdev = false;
10348 unregister_netdevice_queue(dev, NULL);
10352 * Prevent userspace races by waiting until the network
10353 * device is fully setup before sending notifications.
10355 if (!dev->rtnl_link_ops ||
10356 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10357 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10363 if (dev->netdev_ops->ndo_uninit)
10364 dev->netdev_ops->ndo_uninit(dev);
10365 if (dev->priv_destructor)
10366 dev->priv_destructor(dev);
10368 netdev_name_node_free(dev->name_node);
10371 EXPORT_SYMBOL(register_netdevice);
10374 * init_dummy_netdev - init a dummy network device for NAPI
10375 * @dev: device to init
10377 * This takes a network device structure and initialize the minimum
10378 * amount of fields so it can be used to schedule NAPI polls without
10379 * registering a full blown interface. This is to be used by drivers
10380 * that need to tie several hardware interfaces to a single NAPI
10381 * poll scheduler due to HW limitations.
10383 int init_dummy_netdev(struct net_device *dev)
10385 /* Clear everything. Note we don't initialize spinlocks
10386 * are they aren't supposed to be taken by any of the
10387 * NAPI code and this dummy netdev is supposed to be
10388 * only ever used for NAPI polls
10390 memset(dev, 0, sizeof(struct net_device));
10392 /* make sure we BUG if trying to hit standard
10393 * register/unregister code path
10395 dev->reg_state = NETREG_DUMMY;
10397 /* NAPI wants this */
10398 INIT_LIST_HEAD(&dev->napi_list);
10400 /* a dummy interface is started by default */
10401 set_bit(__LINK_STATE_PRESENT, &dev->state);
10402 set_bit(__LINK_STATE_START, &dev->state);
10404 /* napi_busy_loop stats accounting wants this */
10405 dev_net_set(dev, &init_net);
10407 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10408 * because users of this 'device' dont need to change
10414 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10418 * register_netdev - register a network device
10419 * @dev: device to register
10421 * Take a completed network device structure and add it to the kernel
10422 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10423 * chain. 0 is returned on success. A negative errno code is returned
10424 * on a failure to set up the device, or if the name is a duplicate.
10426 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10427 * and expands the device name if you passed a format string to
10430 int register_netdev(struct net_device *dev)
10434 if (rtnl_lock_killable())
10436 err = register_netdevice(dev);
10440 EXPORT_SYMBOL(register_netdev);
10442 int netdev_refcnt_read(const struct net_device *dev)
10444 #ifdef CONFIG_PCPU_DEV_REFCNT
10447 for_each_possible_cpu(i)
10448 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10451 return refcount_read(&dev->dev_refcnt);
10454 EXPORT_SYMBOL(netdev_refcnt_read);
10456 int netdev_unregister_timeout_secs __read_mostly = 10;
10458 #define WAIT_REFS_MIN_MSECS 1
10459 #define WAIT_REFS_MAX_MSECS 250
10461 * netdev_wait_allrefs - wait until all references are gone.
10462 * @dev: target net_device
10464 * This is called when unregistering network devices.
10466 * Any protocol or device that holds a reference should register
10467 * for netdevice notification, and cleanup and put back the
10468 * reference if they receive an UNREGISTER event.
10469 * We can get stuck here if buggy protocols don't correctly
10472 static void netdev_wait_allrefs(struct net_device *dev)
10474 unsigned long rebroadcast_time, warning_time;
10475 int wait = 0, refcnt;
10477 linkwatch_forget_dev(dev);
10479 rebroadcast_time = warning_time = jiffies;
10480 refcnt = netdev_refcnt_read(dev);
10482 while (refcnt != 1) {
10483 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10486 /* Rebroadcast unregister notification */
10487 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10493 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10495 /* We must not have linkwatch events
10496 * pending on unregister. If this
10497 * happens, we simply run the queue
10498 * unscheduled, resulting in a noop
10501 linkwatch_run_queue();
10506 rebroadcast_time = jiffies;
10511 wait = WAIT_REFS_MIN_MSECS;
10514 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10517 refcnt = netdev_refcnt_read(dev);
10520 time_after(jiffies, warning_time +
10521 netdev_unregister_timeout_secs * HZ)) {
10522 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10523 dev->name, refcnt);
10524 warning_time = jiffies;
10529 /* The sequence is:
10533 * register_netdevice(x1);
10534 * register_netdevice(x2);
10536 * unregister_netdevice(y1);
10537 * unregister_netdevice(y2);
10543 * We are invoked by rtnl_unlock().
10544 * This allows us to deal with problems:
10545 * 1) We can delete sysfs objects which invoke hotplug
10546 * without deadlocking with linkwatch via keventd.
10547 * 2) Since we run with the RTNL semaphore not held, we can sleep
10548 * safely in order to wait for the netdev refcnt to drop to zero.
10550 * We must not return until all unregister events added during
10551 * the interval the lock was held have been completed.
10553 void netdev_run_todo(void)
10555 struct list_head list;
10556 #ifdef CONFIG_LOCKDEP
10557 struct list_head unlink_list;
10559 list_replace_init(&net_unlink_list, &unlink_list);
10561 while (!list_empty(&unlink_list)) {
10562 struct net_device *dev = list_first_entry(&unlink_list,
10565 list_del_init(&dev->unlink_list);
10566 dev->nested_level = dev->lower_level - 1;
10570 /* Snapshot list, allow later requests */
10571 list_replace_init(&net_todo_list, &list);
10576 /* Wait for rcu callbacks to finish before next phase */
10577 if (!list_empty(&list))
10580 while (!list_empty(&list)) {
10581 struct net_device *dev
10582 = list_first_entry(&list, struct net_device, todo_list);
10583 list_del(&dev->todo_list);
10585 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10586 pr_err("network todo '%s' but state %d\n",
10587 dev->name, dev->reg_state);
10592 dev->reg_state = NETREG_UNREGISTERED;
10594 netdev_wait_allrefs(dev);
10597 BUG_ON(netdev_refcnt_read(dev) != 1);
10598 BUG_ON(!list_empty(&dev->ptype_all));
10599 BUG_ON(!list_empty(&dev->ptype_specific));
10600 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10601 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10602 #if IS_ENABLED(CONFIG_DECNET)
10603 WARN_ON(dev->dn_ptr);
10605 if (dev->priv_destructor)
10606 dev->priv_destructor(dev);
10607 if (dev->needs_free_netdev)
10610 /* Report a network device has been unregistered */
10612 dev_net(dev)->dev_unreg_count--;
10614 wake_up(&netdev_unregistering_wq);
10616 /* Free network device */
10617 kobject_put(&dev->dev.kobj);
10621 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10622 * all the same fields in the same order as net_device_stats, with only
10623 * the type differing, but rtnl_link_stats64 may have additional fields
10624 * at the end for newer counters.
10626 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10627 const struct net_device_stats *netdev_stats)
10629 #if BITS_PER_LONG == 64
10630 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10631 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10632 /* zero out counters that only exist in rtnl_link_stats64 */
10633 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10634 sizeof(*stats64) - sizeof(*netdev_stats));
10636 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10637 const unsigned long *src = (const unsigned long *)netdev_stats;
10638 u64 *dst = (u64 *)stats64;
10640 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10641 for (i = 0; i < n; i++)
10643 /* zero out counters that only exist in rtnl_link_stats64 */
10644 memset((char *)stats64 + n * sizeof(u64), 0,
10645 sizeof(*stats64) - n * sizeof(u64));
10648 EXPORT_SYMBOL(netdev_stats_to_stats64);
10651 * dev_get_stats - get network device statistics
10652 * @dev: device to get statistics from
10653 * @storage: place to store stats
10655 * Get network statistics from device. Return @storage.
10656 * The device driver may provide its own method by setting
10657 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10658 * otherwise the internal statistics structure is used.
10660 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10661 struct rtnl_link_stats64 *storage)
10663 const struct net_device_ops *ops = dev->netdev_ops;
10665 if (ops->ndo_get_stats64) {
10666 memset(storage, 0, sizeof(*storage));
10667 ops->ndo_get_stats64(dev, storage);
10668 } else if (ops->ndo_get_stats) {
10669 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10671 netdev_stats_to_stats64(storage, &dev->stats);
10673 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10674 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10675 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10678 EXPORT_SYMBOL(dev_get_stats);
10681 * dev_fetch_sw_netstats - get per-cpu network device statistics
10682 * @s: place to store stats
10683 * @netstats: per-cpu network stats to read from
10685 * Read per-cpu network statistics and populate the related fields in @s.
10687 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10688 const struct pcpu_sw_netstats __percpu *netstats)
10692 for_each_possible_cpu(cpu) {
10693 const struct pcpu_sw_netstats *stats;
10694 struct pcpu_sw_netstats tmp;
10695 unsigned int start;
10697 stats = per_cpu_ptr(netstats, cpu);
10699 start = u64_stats_fetch_begin_irq(&stats->syncp);
10700 tmp.rx_packets = stats->rx_packets;
10701 tmp.rx_bytes = stats->rx_bytes;
10702 tmp.tx_packets = stats->tx_packets;
10703 tmp.tx_bytes = stats->tx_bytes;
10704 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10706 s->rx_packets += tmp.rx_packets;
10707 s->rx_bytes += tmp.rx_bytes;
10708 s->tx_packets += tmp.tx_packets;
10709 s->tx_bytes += tmp.tx_bytes;
10712 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10715 * dev_get_tstats64 - ndo_get_stats64 implementation
10716 * @dev: device to get statistics from
10717 * @s: place to store stats
10719 * Populate @s from dev->stats and dev->tstats. Can be used as
10720 * ndo_get_stats64() callback.
10722 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10724 netdev_stats_to_stats64(s, &dev->stats);
10725 dev_fetch_sw_netstats(s, dev->tstats);
10727 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10729 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10731 struct netdev_queue *queue = dev_ingress_queue(dev);
10733 #ifdef CONFIG_NET_CLS_ACT
10736 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10739 netdev_init_one_queue(dev, queue, NULL);
10740 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10741 queue->qdisc_sleeping = &noop_qdisc;
10742 rcu_assign_pointer(dev->ingress_queue, queue);
10747 static const struct ethtool_ops default_ethtool_ops;
10749 void netdev_set_default_ethtool_ops(struct net_device *dev,
10750 const struct ethtool_ops *ops)
10752 if (dev->ethtool_ops == &default_ethtool_ops)
10753 dev->ethtool_ops = ops;
10755 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10757 void netdev_freemem(struct net_device *dev)
10759 char *addr = (char *)dev - dev->padded;
10765 * alloc_netdev_mqs - allocate network device
10766 * @sizeof_priv: size of private data to allocate space for
10767 * @name: device name format string
10768 * @name_assign_type: origin of device name
10769 * @setup: callback to initialize device
10770 * @txqs: the number of TX subqueues to allocate
10771 * @rxqs: the number of RX subqueues to allocate
10773 * Allocates a struct net_device with private data area for driver use
10774 * and performs basic initialization. Also allocates subqueue structs
10775 * for each queue on the device.
10777 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10778 unsigned char name_assign_type,
10779 void (*setup)(struct net_device *),
10780 unsigned int txqs, unsigned int rxqs)
10782 struct net_device *dev;
10783 unsigned int alloc_size;
10784 struct net_device *p;
10786 BUG_ON(strlen(name) >= sizeof(dev->name));
10789 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10794 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10798 alloc_size = sizeof(struct net_device);
10800 /* ensure 32-byte alignment of private area */
10801 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10802 alloc_size += sizeof_priv;
10804 /* ensure 32-byte alignment of whole construct */
10805 alloc_size += NETDEV_ALIGN - 1;
10807 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10811 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10812 dev->padded = (char *)dev - (char *)p;
10814 #ifdef CONFIG_PCPU_DEV_REFCNT
10815 dev->pcpu_refcnt = alloc_percpu(int);
10816 if (!dev->pcpu_refcnt)
10820 refcount_set(&dev->dev_refcnt, 1);
10823 if (dev_addr_init(dev))
10829 dev_net_set(dev, &init_net);
10831 dev->gso_max_size = GSO_MAX_SIZE;
10832 dev->gso_max_segs = GSO_MAX_SEGS;
10833 dev->upper_level = 1;
10834 dev->lower_level = 1;
10835 #ifdef CONFIG_LOCKDEP
10836 dev->nested_level = 0;
10837 INIT_LIST_HEAD(&dev->unlink_list);
10840 INIT_LIST_HEAD(&dev->napi_list);
10841 INIT_LIST_HEAD(&dev->unreg_list);
10842 INIT_LIST_HEAD(&dev->close_list);
10843 INIT_LIST_HEAD(&dev->link_watch_list);
10844 INIT_LIST_HEAD(&dev->adj_list.upper);
10845 INIT_LIST_HEAD(&dev->adj_list.lower);
10846 INIT_LIST_HEAD(&dev->ptype_all);
10847 INIT_LIST_HEAD(&dev->ptype_specific);
10848 INIT_LIST_HEAD(&dev->net_notifier_list);
10849 #ifdef CONFIG_NET_SCHED
10850 hash_init(dev->qdisc_hash);
10852 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10855 if (!dev->tx_queue_len) {
10856 dev->priv_flags |= IFF_NO_QUEUE;
10857 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10860 dev->num_tx_queues = txqs;
10861 dev->real_num_tx_queues = txqs;
10862 if (netif_alloc_netdev_queues(dev))
10865 dev->num_rx_queues = rxqs;
10866 dev->real_num_rx_queues = rxqs;
10867 if (netif_alloc_rx_queues(dev))
10870 strcpy(dev->name, name);
10871 dev->name_assign_type = name_assign_type;
10872 dev->group = INIT_NETDEV_GROUP;
10873 if (!dev->ethtool_ops)
10874 dev->ethtool_ops = &default_ethtool_ops;
10876 nf_hook_ingress_init(dev);
10885 #ifdef CONFIG_PCPU_DEV_REFCNT
10886 free_percpu(dev->pcpu_refcnt);
10889 netdev_freemem(dev);
10892 EXPORT_SYMBOL(alloc_netdev_mqs);
10895 * free_netdev - free network device
10898 * This function does the last stage of destroying an allocated device
10899 * interface. The reference to the device object is released. If this
10900 * is the last reference then it will be freed.Must be called in process
10903 void free_netdev(struct net_device *dev)
10905 struct napi_struct *p, *n;
10909 /* When called immediately after register_netdevice() failed the unwind
10910 * handling may still be dismantling the device. Handle that case by
10911 * deferring the free.
10913 if (dev->reg_state == NETREG_UNREGISTERING) {
10915 dev->needs_free_netdev = true;
10919 netif_free_tx_queues(dev);
10920 netif_free_rx_queues(dev);
10922 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10924 /* Flush device addresses */
10925 dev_addr_flush(dev);
10927 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10930 #ifdef CONFIG_PCPU_DEV_REFCNT
10931 free_percpu(dev->pcpu_refcnt);
10932 dev->pcpu_refcnt = NULL;
10934 free_percpu(dev->xdp_bulkq);
10935 dev->xdp_bulkq = NULL;
10937 /* Compatibility with error handling in drivers */
10938 if (dev->reg_state == NETREG_UNINITIALIZED) {
10939 netdev_freemem(dev);
10943 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10944 dev->reg_state = NETREG_RELEASED;
10946 /* will free via device release */
10947 put_device(&dev->dev);
10949 EXPORT_SYMBOL(free_netdev);
10952 * synchronize_net - Synchronize with packet receive processing
10954 * Wait for packets currently being received to be done.
10955 * Does not block later packets from starting.
10957 void synchronize_net(void)
10960 if (rtnl_is_locked())
10961 synchronize_rcu_expedited();
10965 EXPORT_SYMBOL(synchronize_net);
10968 * unregister_netdevice_queue - remove device from the kernel
10972 * This function shuts down a device interface and removes it
10973 * from the kernel tables.
10974 * If head not NULL, device is queued to be unregistered later.
10976 * Callers must hold the rtnl semaphore. You may want
10977 * unregister_netdev() instead of this.
10980 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10985 list_move_tail(&dev->unreg_list, head);
10989 list_add(&dev->unreg_list, &single);
10990 unregister_netdevice_many(&single);
10993 EXPORT_SYMBOL(unregister_netdevice_queue);
10996 * unregister_netdevice_many - unregister many devices
10997 * @head: list of devices
10999 * Note: As most callers use a stack allocated list_head,
11000 * we force a list_del() to make sure stack wont be corrupted later.
11002 void unregister_netdevice_many(struct list_head *head)
11004 struct net_device *dev, *tmp;
11005 LIST_HEAD(close_head);
11007 BUG_ON(dev_boot_phase);
11010 if (list_empty(head))
11013 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11014 /* Some devices call without registering
11015 * for initialization unwind. Remove those
11016 * devices and proceed with the remaining.
11018 if (dev->reg_state == NETREG_UNINITIALIZED) {
11019 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11023 list_del(&dev->unreg_list);
11026 dev->dismantle = true;
11027 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11030 /* If device is running, close it first. */
11031 list_for_each_entry(dev, head, unreg_list)
11032 list_add_tail(&dev->close_list, &close_head);
11033 dev_close_many(&close_head, true);
11035 list_for_each_entry(dev, head, unreg_list) {
11036 /* And unlink it from device chain. */
11037 unlist_netdevice(dev);
11039 dev->reg_state = NETREG_UNREGISTERING;
11041 flush_all_backlogs();
11045 list_for_each_entry(dev, head, unreg_list) {
11046 struct sk_buff *skb = NULL;
11048 /* Shutdown queueing discipline. */
11051 dev_xdp_uninstall(dev);
11053 /* Notify protocols, that we are about to destroy
11054 * this device. They should clean all the things.
11056 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11058 if (!dev->rtnl_link_ops ||
11059 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11060 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11061 GFP_KERNEL, NULL, 0);
11064 * Flush the unicast and multicast chains
11069 netdev_name_node_alt_flush(dev);
11070 netdev_name_node_free(dev->name_node);
11072 if (dev->netdev_ops->ndo_uninit)
11073 dev->netdev_ops->ndo_uninit(dev);
11076 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11078 /* Notifier chain MUST detach us all upper devices. */
11079 WARN_ON(netdev_has_any_upper_dev(dev));
11080 WARN_ON(netdev_has_any_lower_dev(dev));
11082 /* Remove entries from kobject tree */
11083 netdev_unregister_kobject(dev);
11085 /* Remove XPS queueing entries */
11086 netif_reset_xps_queues_gt(dev, 0);
11092 list_for_each_entry(dev, head, unreg_list) {
11099 EXPORT_SYMBOL(unregister_netdevice_many);
11102 * unregister_netdev - remove device from the kernel
11105 * This function shuts down a device interface and removes it
11106 * from the kernel tables.
11108 * This is just a wrapper for unregister_netdevice that takes
11109 * the rtnl semaphore. In general you want to use this and not
11110 * unregister_netdevice.
11112 void unregister_netdev(struct net_device *dev)
11115 unregister_netdevice(dev);
11118 EXPORT_SYMBOL(unregister_netdev);
11121 * __dev_change_net_namespace - move device to different nethost namespace
11123 * @net: network namespace
11124 * @pat: If not NULL name pattern to try if the current device name
11125 * is already taken in the destination network namespace.
11126 * @new_ifindex: If not zero, specifies device index in the target
11129 * This function shuts down a device interface and moves it
11130 * to a new network namespace. On success 0 is returned, on
11131 * a failure a netagive errno code is returned.
11133 * Callers must hold the rtnl semaphore.
11136 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11137 const char *pat, int new_ifindex)
11139 struct net *net_old = dev_net(dev);
11144 /* Don't allow namespace local devices to be moved. */
11146 if (dev->features & NETIF_F_NETNS_LOCAL)
11149 /* Ensure the device has been registrered */
11150 if (dev->reg_state != NETREG_REGISTERED)
11153 /* Get out if there is nothing todo */
11155 if (net_eq(net_old, net))
11158 /* Pick the destination device name, and ensure
11159 * we can use it in the destination network namespace.
11162 if (netdev_name_in_use(net, dev->name)) {
11163 /* We get here if we can't use the current device name */
11166 err = dev_get_valid_name(net, dev, pat);
11171 /* Check that new_ifindex isn't used yet. */
11173 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11177 * And now a mini version of register_netdevice unregister_netdevice.
11180 /* If device is running close it first. */
11183 /* And unlink it from device chain */
11184 unlist_netdevice(dev);
11188 /* Shutdown queueing discipline. */
11191 /* Notify protocols, that we are about to destroy
11192 * this device. They should clean all the things.
11194 * Note that dev->reg_state stays at NETREG_REGISTERED.
11195 * This is wanted because this way 8021q and macvlan know
11196 * the device is just moving and can keep their slaves up.
11198 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11201 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11202 /* If there is an ifindex conflict assign a new one */
11203 if (!new_ifindex) {
11204 if (__dev_get_by_index(net, dev->ifindex))
11205 new_ifindex = dev_new_index(net);
11207 new_ifindex = dev->ifindex;
11210 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11214 * Flush the unicast and multicast chains
11219 /* Send a netdev-removed uevent to the old namespace */
11220 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11221 netdev_adjacent_del_links(dev);
11223 /* Move per-net netdevice notifiers that are following the netdevice */
11224 move_netdevice_notifiers_dev_net(dev, net);
11226 /* Actually switch the network namespace */
11227 dev_net_set(dev, net);
11228 dev->ifindex = new_ifindex;
11230 /* Send a netdev-add uevent to the new namespace */
11231 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11232 netdev_adjacent_add_links(dev);
11234 /* Fixup kobjects */
11235 err = device_rename(&dev->dev, dev->name);
11238 /* Adapt owner in case owning user namespace of target network
11239 * namespace is different from the original one.
11241 err = netdev_change_owner(dev, net_old, net);
11244 /* Add the device back in the hashes */
11245 list_netdevice(dev);
11247 /* Notify protocols, that a new device appeared. */
11248 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11251 * Prevent userspace races by waiting until the network
11252 * device is fully setup before sending notifications.
11254 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11261 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11263 static int dev_cpu_dead(unsigned int oldcpu)
11265 struct sk_buff **list_skb;
11266 struct sk_buff *skb;
11268 struct softnet_data *sd, *oldsd, *remsd = NULL;
11270 local_irq_disable();
11271 cpu = smp_processor_id();
11272 sd = &per_cpu(softnet_data, cpu);
11273 oldsd = &per_cpu(softnet_data, oldcpu);
11275 /* Find end of our completion_queue. */
11276 list_skb = &sd->completion_queue;
11278 list_skb = &(*list_skb)->next;
11279 /* Append completion queue from offline CPU. */
11280 *list_skb = oldsd->completion_queue;
11281 oldsd->completion_queue = NULL;
11283 /* Append output queue from offline CPU. */
11284 if (oldsd->output_queue) {
11285 *sd->output_queue_tailp = oldsd->output_queue;
11286 sd->output_queue_tailp = oldsd->output_queue_tailp;
11287 oldsd->output_queue = NULL;
11288 oldsd->output_queue_tailp = &oldsd->output_queue;
11290 /* Append NAPI poll list from offline CPU, with one exception :
11291 * process_backlog() must be called by cpu owning percpu backlog.
11292 * We properly handle process_queue & input_pkt_queue later.
11294 while (!list_empty(&oldsd->poll_list)) {
11295 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11296 struct napi_struct,
11299 list_del_init(&napi->poll_list);
11300 if (napi->poll == process_backlog)
11303 ____napi_schedule(sd, napi);
11306 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11307 local_irq_enable();
11310 remsd = oldsd->rps_ipi_list;
11311 oldsd->rps_ipi_list = NULL;
11313 /* send out pending IPI's on offline CPU */
11314 net_rps_send_ipi(remsd);
11316 /* Process offline CPU's input_pkt_queue */
11317 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11319 input_queue_head_incr(oldsd);
11321 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11323 input_queue_head_incr(oldsd);
11330 * netdev_increment_features - increment feature set by one
11331 * @all: current feature set
11332 * @one: new feature set
11333 * @mask: mask feature set
11335 * Computes a new feature set after adding a device with feature set
11336 * @one to the master device with current feature set @all. Will not
11337 * enable anything that is off in @mask. Returns the new feature set.
11339 netdev_features_t netdev_increment_features(netdev_features_t all,
11340 netdev_features_t one, netdev_features_t mask)
11342 if (mask & NETIF_F_HW_CSUM)
11343 mask |= NETIF_F_CSUM_MASK;
11344 mask |= NETIF_F_VLAN_CHALLENGED;
11346 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11347 all &= one | ~NETIF_F_ALL_FOR_ALL;
11349 /* If one device supports hw checksumming, set for all. */
11350 if (all & NETIF_F_HW_CSUM)
11351 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11355 EXPORT_SYMBOL(netdev_increment_features);
11357 static struct hlist_head * __net_init netdev_create_hash(void)
11360 struct hlist_head *hash;
11362 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11364 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11365 INIT_HLIST_HEAD(&hash[i]);
11370 /* Initialize per network namespace state */
11371 static int __net_init netdev_init(struct net *net)
11373 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11374 8 * sizeof_field(struct napi_struct, gro_bitmask));
11376 if (net != &init_net)
11377 INIT_LIST_HEAD(&net->dev_base_head);
11379 net->dev_name_head = netdev_create_hash();
11380 if (net->dev_name_head == NULL)
11383 net->dev_index_head = netdev_create_hash();
11384 if (net->dev_index_head == NULL)
11387 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11392 kfree(net->dev_name_head);
11398 * netdev_drivername - network driver for the device
11399 * @dev: network device
11401 * Determine network driver for device.
11403 const char *netdev_drivername(const struct net_device *dev)
11405 const struct device_driver *driver;
11406 const struct device *parent;
11407 const char *empty = "";
11409 parent = dev->dev.parent;
11413 driver = parent->driver;
11414 if (driver && driver->name)
11415 return driver->name;
11419 static void __netdev_printk(const char *level, const struct net_device *dev,
11420 struct va_format *vaf)
11422 if (dev && dev->dev.parent) {
11423 dev_printk_emit(level[1] - '0',
11426 dev_driver_string(dev->dev.parent),
11427 dev_name(dev->dev.parent),
11428 netdev_name(dev), netdev_reg_state(dev),
11431 printk("%s%s%s: %pV",
11432 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11434 printk("%s(NULL net_device): %pV", level, vaf);
11438 void netdev_printk(const char *level, const struct net_device *dev,
11439 const char *format, ...)
11441 struct va_format vaf;
11444 va_start(args, format);
11449 __netdev_printk(level, dev, &vaf);
11453 EXPORT_SYMBOL(netdev_printk);
11455 #define define_netdev_printk_level(func, level) \
11456 void func(const struct net_device *dev, const char *fmt, ...) \
11458 struct va_format vaf; \
11461 va_start(args, fmt); \
11466 __netdev_printk(level, dev, &vaf); \
11470 EXPORT_SYMBOL(func);
11472 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11473 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11474 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11475 define_netdev_printk_level(netdev_err, KERN_ERR);
11476 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11477 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11478 define_netdev_printk_level(netdev_info, KERN_INFO);
11480 static void __net_exit netdev_exit(struct net *net)
11482 kfree(net->dev_name_head);
11483 kfree(net->dev_index_head);
11484 if (net != &init_net)
11485 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11488 static struct pernet_operations __net_initdata netdev_net_ops = {
11489 .init = netdev_init,
11490 .exit = netdev_exit,
11493 static void __net_exit default_device_exit(struct net *net)
11495 struct net_device *dev, *aux;
11497 * Push all migratable network devices back to the
11498 * initial network namespace
11501 for_each_netdev_safe(net, dev, aux) {
11503 char fb_name[IFNAMSIZ];
11505 /* Ignore unmoveable devices (i.e. loopback) */
11506 if (dev->features & NETIF_F_NETNS_LOCAL)
11509 /* Leave virtual devices for the generic cleanup */
11510 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11513 /* Push remaining network devices to init_net */
11514 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11515 if (netdev_name_in_use(&init_net, fb_name))
11516 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11517 err = dev_change_net_namespace(dev, &init_net, fb_name);
11519 pr_emerg("%s: failed to move %s to init_net: %d\n",
11520 __func__, dev->name, err);
11527 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11529 /* Return with the rtnl_lock held when there are no network
11530 * devices unregistering in any network namespace in net_list.
11533 bool unregistering;
11534 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11536 add_wait_queue(&netdev_unregistering_wq, &wait);
11538 unregistering = false;
11540 list_for_each_entry(net, net_list, exit_list) {
11541 if (net->dev_unreg_count > 0) {
11542 unregistering = true;
11546 if (!unregistering)
11550 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11552 remove_wait_queue(&netdev_unregistering_wq, &wait);
11555 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11557 /* At exit all network devices most be removed from a network
11558 * namespace. Do this in the reverse order of registration.
11559 * Do this across as many network namespaces as possible to
11560 * improve batching efficiency.
11562 struct net_device *dev;
11564 LIST_HEAD(dev_kill_list);
11566 /* To prevent network device cleanup code from dereferencing
11567 * loopback devices or network devices that have been freed
11568 * wait here for all pending unregistrations to complete,
11569 * before unregistring the loopback device and allowing the
11570 * network namespace be freed.
11572 * The netdev todo list containing all network devices
11573 * unregistrations that happen in default_device_exit_batch
11574 * will run in the rtnl_unlock() at the end of
11575 * default_device_exit_batch.
11577 rtnl_lock_unregistering(net_list);
11578 list_for_each_entry(net, net_list, exit_list) {
11579 for_each_netdev_reverse(net, dev) {
11580 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11581 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11583 unregister_netdevice_queue(dev, &dev_kill_list);
11586 unregister_netdevice_many(&dev_kill_list);
11590 static struct pernet_operations __net_initdata default_device_ops = {
11591 .exit = default_device_exit,
11592 .exit_batch = default_device_exit_batch,
11596 * Initialize the DEV module. At boot time this walks the device list and
11597 * unhooks any devices that fail to initialise (normally hardware not
11598 * present) and leaves us with a valid list of present and active devices.
11603 * This is called single threaded during boot, so no need
11604 * to take the rtnl semaphore.
11606 static int __init net_dev_init(void)
11608 int i, rc = -ENOMEM;
11610 BUG_ON(!dev_boot_phase);
11612 if (dev_proc_init())
11615 if (netdev_kobject_init())
11618 INIT_LIST_HEAD(&ptype_all);
11619 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11620 INIT_LIST_HEAD(&ptype_base[i]);
11622 INIT_LIST_HEAD(&offload_base);
11624 if (register_pernet_subsys(&netdev_net_ops))
11628 * Initialise the packet receive queues.
11631 for_each_possible_cpu(i) {
11632 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11633 struct softnet_data *sd = &per_cpu(softnet_data, i);
11635 INIT_WORK(flush, flush_backlog);
11637 skb_queue_head_init(&sd->input_pkt_queue);
11638 skb_queue_head_init(&sd->process_queue);
11639 #ifdef CONFIG_XFRM_OFFLOAD
11640 skb_queue_head_init(&sd->xfrm_backlog);
11642 INIT_LIST_HEAD(&sd->poll_list);
11643 sd->output_queue_tailp = &sd->output_queue;
11645 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11649 init_gro_hash(&sd->backlog);
11650 sd->backlog.poll = process_backlog;
11651 sd->backlog.weight = weight_p;
11654 dev_boot_phase = 0;
11656 /* The loopback device is special if any other network devices
11657 * is present in a network namespace the loopback device must
11658 * be present. Since we now dynamically allocate and free the
11659 * loopback device ensure this invariant is maintained by
11660 * keeping the loopback device as the first device on the
11661 * list of network devices. Ensuring the loopback devices
11662 * is the first device that appears and the last network device
11665 if (register_pernet_device(&loopback_net_ops))
11668 if (register_pernet_device(&default_device_ops))
11671 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11672 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11674 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11675 NULL, dev_cpu_dead);
11682 subsys_initcall(net_dev_init);