1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_netdev.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
154 #include "net-sysfs.h"
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 netdev_err(dev, "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 netdev_warn(dev, "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 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
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;
3174 if (unlikely(!qcount)) {
3175 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3176 sb_dev->name, qoffset, tc);
3178 qcount = dev->real_num_tx_queues;
3182 if (skb_rx_queue_recorded(skb)) {
3183 hash = skb_get_rx_queue(skb);
3184 if (hash >= qoffset)
3186 while (unlikely(hash >= qcount))
3188 return hash + qoffset;
3191 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3194 static void skb_warn_bad_offload(const struct sk_buff *skb)
3196 static const netdev_features_t null_features;
3197 struct net_device *dev = skb->dev;
3198 const char *name = "";
3200 if (!net_ratelimit())
3204 if (dev->dev.parent)
3205 name = dev_driver_string(dev->dev.parent);
3207 name = netdev_name(dev);
3209 skb_dump(KERN_WARNING, skb, false);
3210 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3211 name, dev ? &dev->features : &null_features,
3212 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3216 * Invalidate hardware checksum when packet is to be mangled, and
3217 * complete checksum manually on outgoing path.
3219 int skb_checksum_help(struct sk_buff *skb)
3222 int ret = 0, offset;
3224 if (skb->ip_summed == CHECKSUM_COMPLETE)
3225 goto out_set_summed;
3227 if (unlikely(skb_is_gso(skb))) {
3228 skb_warn_bad_offload(skb);
3232 /* Before computing a checksum, we should make sure no frag could
3233 * be modified by an external entity : checksum could be wrong.
3235 if (skb_has_shared_frag(skb)) {
3236 ret = __skb_linearize(skb);
3241 offset = skb_checksum_start_offset(skb);
3242 BUG_ON(offset >= skb_headlen(skb));
3243 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3245 offset += skb->csum_offset;
3246 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3248 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3252 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3254 skb->ip_summed = CHECKSUM_NONE;
3258 EXPORT_SYMBOL(skb_checksum_help);
3260 int skb_crc32c_csum_help(struct sk_buff *skb)
3263 int ret = 0, offset, start;
3265 if (skb->ip_summed != CHECKSUM_PARTIAL)
3268 if (unlikely(skb_is_gso(skb)))
3271 /* Before computing a checksum, we should make sure no frag could
3272 * be modified by an external entity : checksum could be wrong.
3274 if (unlikely(skb_has_shared_frag(skb))) {
3275 ret = __skb_linearize(skb);
3279 start = skb_checksum_start_offset(skb);
3280 offset = start + offsetof(struct sctphdr, checksum);
3281 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3286 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3290 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3291 skb->len - start, ~(__u32)0,
3293 *(__le32 *)(skb->data + offset) = crc32c_csum;
3294 skb->ip_summed = CHECKSUM_NONE;
3295 skb->csum_not_inet = 0;
3300 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3302 __be16 type = skb->protocol;
3304 /* Tunnel gso handlers can set protocol to ethernet. */
3305 if (type == htons(ETH_P_TEB)) {
3308 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3311 eth = (struct ethhdr *)skb->data;
3312 type = eth->h_proto;
3315 return __vlan_get_protocol(skb, type, depth);
3319 * skb_mac_gso_segment - mac layer segmentation handler.
3320 * @skb: buffer to segment
3321 * @features: features for the output path (see dev->features)
3323 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3324 netdev_features_t features)
3326 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3327 struct packet_offload *ptype;
3328 int vlan_depth = skb->mac_len;
3329 __be16 type = skb_network_protocol(skb, &vlan_depth);
3331 if (unlikely(!type))
3332 return ERR_PTR(-EINVAL);
3334 __skb_pull(skb, vlan_depth);
3337 list_for_each_entry_rcu(ptype, &offload_base, list) {
3338 if (ptype->type == type && ptype->callbacks.gso_segment) {
3339 segs = ptype->callbacks.gso_segment(skb, features);
3345 __skb_push(skb, skb->data - skb_mac_header(skb));
3349 EXPORT_SYMBOL(skb_mac_gso_segment);
3352 /* openvswitch calls this on rx path, so we need a different check.
3354 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3357 return skb->ip_summed != CHECKSUM_PARTIAL &&
3358 skb->ip_summed != CHECKSUM_UNNECESSARY;
3360 return skb->ip_summed == CHECKSUM_NONE;
3364 * __skb_gso_segment - Perform segmentation on skb.
3365 * @skb: buffer to segment
3366 * @features: features for the output path (see dev->features)
3367 * @tx_path: whether it is called in TX path
3369 * This function segments the given skb and returns a list of segments.
3371 * It may return NULL if the skb requires no segmentation. This is
3372 * only possible when GSO is used for verifying header integrity.
3374 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3376 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3377 netdev_features_t features, bool tx_path)
3379 struct sk_buff *segs;
3381 if (unlikely(skb_needs_check(skb, tx_path))) {
3384 /* We're going to init ->check field in TCP or UDP header */
3385 err = skb_cow_head(skb, 0);
3387 return ERR_PTR(err);
3390 /* Only report GSO partial support if it will enable us to
3391 * support segmentation on this frame without needing additional
3394 if (features & NETIF_F_GSO_PARTIAL) {
3395 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3396 struct net_device *dev = skb->dev;
3398 partial_features |= dev->features & dev->gso_partial_features;
3399 if (!skb_gso_ok(skb, features | partial_features))
3400 features &= ~NETIF_F_GSO_PARTIAL;
3403 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3404 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3406 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3407 SKB_GSO_CB(skb)->encap_level = 0;
3409 skb_reset_mac_header(skb);
3410 skb_reset_mac_len(skb);
3412 segs = skb_mac_gso_segment(skb, features);
3414 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3415 skb_warn_bad_offload(skb);
3419 EXPORT_SYMBOL(__skb_gso_segment);
3421 /* Take action when hardware reception checksum errors are detected. */
3423 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3425 netdev_err(dev, "hw csum failure\n");
3426 skb_dump(KERN_ERR, skb, true);
3430 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3432 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3434 EXPORT_SYMBOL(netdev_rx_csum_fault);
3437 /* XXX: check that highmem exists at all on the given machine. */
3438 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3440 #ifdef CONFIG_HIGHMEM
3443 if (!(dev->features & NETIF_F_HIGHDMA)) {
3444 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3445 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3447 if (PageHighMem(skb_frag_page(frag)))
3455 /* If MPLS offload request, verify we are testing hardware MPLS features
3456 * instead of standard features for the netdev.
3458 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3459 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3460 netdev_features_t features,
3463 if (eth_p_mpls(type))
3464 features &= skb->dev->mpls_features;
3469 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3470 netdev_features_t features,
3477 static netdev_features_t harmonize_features(struct sk_buff *skb,
3478 netdev_features_t features)
3482 type = skb_network_protocol(skb, NULL);
3483 features = net_mpls_features(skb, features, type);
3485 if (skb->ip_summed != CHECKSUM_NONE &&
3486 !can_checksum_protocol(features, type)) {
3487 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3489 if (illegal_highdma(skb->dev, skb))
3490 features &= ~NETIF_F_SG;
3495 netdev_features_t passthru_features_check(struct sk_buff *skb,
3496 struct net_device *dev,
3497 netdev_features_t features)
3501 EXPORT_SYMBOL(passthru_features_check);
3503 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3504 struct net_device *dev,
3505 netdev_features_t features)
3507 return vlan_features_check(skb, features);
3510 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3511 struct net_device *dev,
3512 netdev_features_t features)
3514 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3516 if (gso_segs > dev->gso_max_segs)
3517 return features & ~NETIF_F_GSO_MASK;
3519 if (!skb_shinfo(skb)->gso_type) {
3520 skb_warn_bad_offload(skb);
3521 return features & ~NETIF_F_GSO_MASK;
3524 /* Support for GSO partial features requires software
3525 * intervention before we can actually process the packets
3526 * so we need to strip support for any partial features now
3527 * and we can pull them back in after we have partially
3528 * segmented the frame.
3530 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3531 features &= ~dev->gso_partial_features;
3533 /* Make sure to clear the IPv4 ID mangling feature if the
3534 * IPv4 header has the potential to be fragmented.
3536 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3537 struct iphdr *iph = skb->encapsulation ?
3538 inner_ip_hdr(skb) : ip_hdr(skb);
3540 if (!(iph->frag_off & htons(IP_DF)))
3541 features &= ~NETIF_F_TSO_MANGLEID;
3547 netdev_features_t netif_skb_features(struct sk_buff *skb)
3549 struct net_device *dev = skb->dev;
3550 netdev_features_t features = dev->features;
3552 if (skb_is_gso(skb))
3553 features = gso_features_check(skb, dev, features);
3555 /* If encapsulation offload request, verify we are testing
3556 * hardware encapsulation features instead of standard
3557 * features for the netdev
3559 if (skb->encapsulation)
3560 features &= dev->hw_enc_features;
3562 if (skb_vlan_tagged(skb))
3563 features = netdev_intersect_features(features,
3564 dev->vlan_features |
3565 NETIF_F_HW_VLAN_CTAG_TX |
3566 NETIF_F_HW_VLAN_STAG_TX);
3568 if (dev->netdev_ops->ndo_features_check)
3569 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3572 features &= dflt_features_check(skb, dev, features);
3574 return harmonize_features(skb, features);
3576 EXPORT_SYMBOL(netif_skb_features);
3578 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3579 struct netdev_queue *txq, bool more)
3584 if (dev_nit_active(dev))
3585 dev_queue_xmit_nit(skb, dev);
3588 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3589 trace_net_dev_start_xmit(skb, dev);
3590 rc = netdev_start_xmit(skb, dev, txq, more);
3591 trace_net_dev_xmit(skb, rc, dev, len);
3596 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3597 struct netdev_queue *txq, int *ret)
3599 struct sk_buff *skb = first;
3600 int rc = NETDEV_TX_OK;
3603 struct sk_buff *next = skb->next;
3605 skb_mark_not_on_list(skb);
3606 rc = xmit_one(skb, dev, txq, next != NULL);
3607 if (unlikely(!dev_xmit_complete(rc))) {
3613 if (netif_tx_queue_stopped(txq) && skb) {
3614 rc = NETDEV_TX_BUSY;
3624 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3625 netdev_features_t features)
3627 if (skb_vlan_tag_present(skb) &&
3628 !vlan_hw_offload_capable(features, skb->vlan_proto))
3629 skb = __vlan_hwaccel_push_inside(skb);
3633 int skb_csum_hwoffload_help(struct sk_buff *skb,
3634 const netdev_features_t features)
3636 if (unlikely(skb_csum_is_sctp(skb)))
3637 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3638 skb_crc32c_csum_help(skb);
3640 if (features & NETIF_F_HW_CSUM)
3643 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3644 switch (skb->csum_offset) {
3645 case offsetof(struct tcphdr, check):
3646 case offsetof(struct udphdr, check):
3651 return skb_checksum_help(skb);
3653 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3655 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3657 netdev_features_t features;
3659 features = netif_skb_features(skb);
3660 skb = validate_xmit_vlan(skb, features);
3664 skb = sk_validate_xmit_skb(skb, dev);
3668 if (netif_needs_gso(skb, features)) {
3669 struct sk_buff *segs;
3671 segs = skb_gso_segment(skb, features);
3679 if (skb_needs_linearize(skb, features) &&
3680 __skb_linearize(skb))
3683 /* If packet is not checksummed and device does not
3684 * support checksumming for this protocol, complete
3685 * checksumming here.
3687 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3688 if (skb->encapsulation)
3689 skb_set_inner_transport_header(skb,
3690 skb_checksum_start_offset(skb));
3692 skb_set_transport_header(skb,
3693 skb_checksum_start_offset(skb));
3694 if (skb_csum_hwoffload_help(skb, features))
3699 skb = validate_xmit_xfrm(skb, features, again);
3706 atomic_long_inc(&dev->tx_dropped);
3710 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3712 struct sk_buff *next, *head = NULL, *tail;
3714 for (; skb != NULL; skb = next) {
3716 skb_mark_not_on_list(skb);
3718 /* in case skb wont be segmented, point to itself */
3721 skb = validate_xmit_skb(skb, dev, again);
3729 /* If skb was segmented, skb->prev points to
3730 * the last segment. If not, it still contains skb.
3736 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3738 static void qdisc_pkt_len_init(struct sk_buff *skb)
3740 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3742 qdisc_skb_cb(skb)->pkt_len = skb->len;
3744 /* To get more precise estimation of bytes sent on wire,
3745 * we add to pkt_len the headers size of all segments
3747 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3748 unsigned int hdr_len;
3749 u16 gso_segs = shinfo->gso_segs;
3751 /* mac layer + network layer */
3752 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3754 /* + transport layer */
3755 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3756 const struct tcphdr *th;
3757 struct tcphdr _tcphdr;
3759 th = skb_header_pointer(skb, skb_transport_offset(skb),
3760 sizeof(_tcphdr), &_tcphdr);
3762 hdr_len += __tcp_hdrlen(th);
3764 struct udphdr _udphdr;
3766 if (skb_header_pointer(skb, skb_transport_offset(skb),
3767 sizeof(_udphdr), &_udphdr))
3768 hdr_len += sizeof(struct udphdr);
3771 if (shinfo->gso_type & SKB_GSO_DODGY)
3772 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3775 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3779 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3780 struct sk_buff **to_free,
3781 struct netdev_queue *txq)
3785 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3786 if (rc == NET_XMIT_SUCCESS)
3787 trace_qdisc_enqueue(q, txq, skb);
3791 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3792 struct net_device *dev,
3793 struct netdev_queue *txq)
3795 spinlock_t *root_lock = qdisc_lock(q);
3796 struct sk_buff *to_free = NULL;
3800 qdisc_calculate_pkt_len(skb, q);
3802 if (q->flags & TCQ_F_NOLOCK) {
3803 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3804 qdisc_run_begin(q)) {
3805 /* Retest nolock_qdisc_is_empty() within the protection
3806 * of q->seqlock to protect from racing with requeuing.
3808 if (unlikely(!nolock_qdisc_is_empty(q))) {
3809 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3816 qdisc_bstats_cpu_update(q, skb);
3817 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3818 !nolock_qdisc_is_empty(q))
3822 return NET_XMIT_SUCCESS;
3825 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3829 if (unlikely(to_free))
3830 kfree_skb_list(to_free);
3835 * Heuristic to force contended enqueues to serialize on a
3836 * separate lock before trying to get qdisc main lock.
3837 * This permits qdisc->running owner to get the lock more
3838 * often and dequeue packets faster.
3840 contended = qdisc_is_running(q);
3841 if (unlikely(contended))
3842 spin_lock(&q->busylock);
3844 spin_lock(root_lock);
3845 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3846 __qdisc_drop(skb, &to_free);
3848 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3849 qdisc_run_begin(q)) {
3851 * This is a work-conserving queue; there are no old skbs
3852 * waiting to be sent out; and the qdisc is not running -
3853 * xmit the skb directly.
3856 qdisc_bstats_update(q, skb);
3858 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3859 if (unlikely(contended)) {
3860 spin_unlock(&q->busylock);
3867 rc = NET_XMIT_SUCCESS;
3869 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3870 if (qdisc_run_begin(q)) {
3871 if (unlikely(contended)) {
3872 spin_unlock(&q->busylock);
3879 spin_unlock(root_lock);
3880 if (unlikely(to_free))
3881 kfree_skb_list(to_free);
3882 if (unlikely(contended))
3883 spin_unlock(&q->busylock);
3887 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3888 static void skb_update_prio(struct sk_buff *skb)
3890 const struct netprio_map *map;
3891 const struct sock *sk;
3892 unsigned int prioidx;
3896 map = rcu_dereference_bh(skb->dev->priomap);
3899 sk = skb_to_full_sk(skb);
3903 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3905 if (prioidx < map->priomap_len)
3906 skb->priority = map->priomap[prioidx];
3909 #define skb_update_prio(skb)
3913 * dev_loopback_xmit - loop back @skb
3914 * @net: network namespace this loopback is happening in
3915 * @sk: sk needed to be a netfilter okfn
3916 * @skb: buffer to transmit
3918 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3920 skb_reset_mac_header(skb);
3921 __skb_pull(skb, skb_network_offset(skb));
3922 skb->pkt_type = PACKET_LOOPBACK;
3923 if (skb->ip_summed == CHECKSUM_NONE)
3924 skb->ip_summed = CHECKSUM_UNNECESSARY;
3925 WARN_ON(!skb_dst(skb));
3930 EXPORT_SYMBOL(dev_loopback_xmit);
3932 #ifdef CONFIG_NET_EGRESS
3933 static struct sk_buff *
3934 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3936 #ifdef CONFIG_NET_CLS_ACT
3937 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3938 struct tcf_result cl_res;
3943 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3944 qdisc_skb_cb(skb)->mru = 0;
3945 qdisc_skb_cb(skb)->post_ct = false;
3946 mini_qdisc_bstats_cpu_update(miniq, skb);
3948 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3950 case TC_ACT_RECLASSIFY:
3951 skb->tc_index = TC_H_MIN(cl_res.classid);
3954 mini_qdisc_qstats_cpu_drop(miniq);
3955 *ret = NET_XMIT_DROP;
3961 *ret = NET_XMIT_SUCCESS;
3964 case TC_ACT_REDIRECT:
3965 /* No need to push/pop skb's mac_header here on egress! */
3966 skb_do_redirect(skb);
3967 *ret = NET_XMIT_SUCCESS;
3972 #endif /* CONFIG_NET_CLS_ACT */
3976 #endif /* CONFIG_NET_EGRESS */
3979 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3980 struct xps_dev_maps *dev_maps, unsigned int tci)
3982 int tc = netdev_get_prio_tc_map(dev, skb->priority);
3983 struct xps_map *map;
3984 int queue_index = -1;
3986 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3989 tci *= dev_maps->num_tc;
3992 map = rcu_dereference(dev_maps->attr_map[tci]);
3995 queue_index = map->queues[0];
3997 queue_index = map->queues[reciprocal_scale(
3998 skb_get_hash(skb), map->len)];
3999 if (unlikely(queue_index >= dev->real_num_tx_queues))
4006 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4007 struct sk_buff *skb)
4010 struct xps_dev_maps *dev_maps;
4011 struct sock *sk = skb->sk;
4012 int queue_index = -1;
4014 if (!static_key_false(&xps_needed))
4018 if (!static_key_false(&xps_rxqs_needed))
4021 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4023 int tci = sk_rx_queue_get(sk);
4026 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4031 if (queue_index < 0) {
4032 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4034 unsigned int tci = skb->sender_cpu - 1;
4036 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4048 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4049 struct net_device *sb_dev)
4053 EXPORT_SYMBOL(dev_pick_tx_zero);
4055 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4056 struct net_device *sb_dev)
4058 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4060 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4062 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4063 struct net_device *sb_dev)
4065 struct sock *sk = skb->sk;
4066 int queue_index = sk_tx_queue_get(sk);
4068 sb_dev = sb_dev ? : dev;
4070 if (queue_index < 0 || skb->ooo_okay ||
4071 queue_index >= dev->real_num_tx_queues) {
4072 int new_index = get_xps_queue(dev, sb_dev, skb);
4075 new_index = skb_tx_hash(dev, sb_dev, skb);
4077 if (queue_index != new_index && sk &&
4079 rcu_access_pointer(sk->sk_dst_cache))
4080 sk_tx_queue_set(sk, new_index);
4082 queue_index = new_index;
4087 EXPORT_SYMBOL(netdev_pick_tx);
4089 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4090 struct sk_buff *skb,
4091 struct net_device *sb_dev)
4093 int queue_index = 0;
4096 u32 sender_cpu = skb->sender_cpu - 1;
4098 if (sender_cpu >= (u32)NR_CPUS)
4099 skb->sender_cpu = raw_smp_processor_id() + 1;
4102 if (dev->real_num_tx_queues != 1) {
4103 const struct net_device_ops *ops = dev->netdev_ops;
4105 if (ops->ndo_select_queue)
4106 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4108 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4110 queue_index = netdev_cap_txqueue(dev, queue_index);
4113 skb_set_queue_mapping(skb, queue_index);
4114 return netdev_get_tx_queue(dev, queue_index);
4118 * __dev_queue_xmit - transmit a buffer
4119 * @skb: buffer to transmit
4120 * @sb_dev: suboordinate device used for L2 forwarding offload
4122 * Queue a buffer for transmission to a network device. The caller must
4123 * have set the device and priority and built the buffer before calling
4124 * this function. The function can be called from an interrupt.
4126 * A negative errno code is returned on a failure. A success does not
4127 * guarantee the frame will be transmitted as it may be dropped due
4128 * to congestion or traffic shaping.
4130 * -----------------------------------------------------------------------------------
4131 * I notice this method can also return errors from the queue disciplines,
4132 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4135 * Regardless of the return value, the skb is consumed, so it is currently
4136 * difficult to retry a send to this method. (You can bump the ref count
4137 * before sending to hold a reference for retry if you are careful.)
4139 * When calling this method, interrupts MUST be enabled. This is because
4140 * the BH enable code must have IRQs enabled so that it will not deadlock.
4143 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4145 struct net_device *dev = skb->dev;
4146 struct netdev_queue *txq;
4151 skb_reset_mac_header(skb);
4153 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4154 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4156 /* Disable soft irqs for various locks below. Also
4157 * stops preemption for RCU.
4161 skb_update_prio(skb);
4163 qdisc_pkt_len_init(skb);
4164 #ifdef CONFIG_NET_CLS_ACT
4165 skb->tc_at_ingress = 0;
4167 #ifdef CONFIG_NET_EGRESS
4168 if (static_branch_unlikely(&egress_needed_key)) {
4169 if (nf_hook_egress_active()) {
4170 skb = nf_hook_egress(skb, &rc, dev);
4174 nf_skip_egress(skb, true);
4175 skb = sch_handle_egress(skb, &rc, dev);
4178 nf_skip_egress(skb, false);
4181 /* If device/qdisc don't need skb->dst, release it right now while
4182 * its hot in this cpu cache.
4184 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4189 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4190 q = rcu_dereference_bh(txq->qdisc);
4192 trace_net_dev_queue(skb);
4194 rc = __dev_xmit_skb(skb, q, dev, txq);
4198 /* The device has no queue. Common case for software devices:
4199 * loopback, all the sorts of tunnels...
4201 * Really, it is unlikely that netif_tx_lock protection is necessary
4202 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4204 * However, it is possible, that they rely on protection
4207 * Check this and shot the lock. It is not prone from deadlocks.
4208 *Either shot noqueue qdisc, it is even simpler 8)
4210 if (dev->flags & IFF_UP) {
4211 int cpu = smp_processor_id(); /* ok because BHs are off */
4213 /* Other cpus might concurrently change txq->xmit_lock_owner
4214 * to -1 or to their cpu id, but not to our id.
4216 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4217 if (dev_xmit_recursion())
4218 goto recursion_alert;
4220 skb = validate_xmit_skb(skb, dev, &again);
4224 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4225 HARD_TX_LOCK(dev, txq, cpu);
4227 if (!netif_xmit_stopped(txq)) {
4228 dev_xmit_recursion_inc();
4229 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4230 dev_xmit_recursion_dec();
4231 if (dev_xmit_complete(rc)) {
4232 HARD_TX_UNLOCK(dev, txq);
4236 HARD_TX_UNLOCK(dev, txq);
4237 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4240 /* Recursion is detected! It is possible,
4244 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4250 rcu_read_unlock_bh();
4252 atomic_long_inc(&dev->tx_dropped);
4253 kfree_skb_list(skb);
4256 rcu_read_unlock_bh();
4260 int dev_queue_xmit(struct sk_buff *skb)
4262 return __dev_queue_xmit(skb, NULL);
4264 EXPORT_SYMBOL(dev_queue_xmit);
4266 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4268 return __dev_queue_xmit(skb, sb_dev);
4270 EXPORT_SYMBOL(dev_queue_xmit_accel);
4272 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4274 struct net_device *dev = skb->dev;
4275 struct sk_buff *orig_skb = skb;
4276 struct netdev_queue *txq;
4277 int ret = NETDEV_TX_BUSY;
4280 if (unlikely(!netif_running(dev) ||
4281 !netif_carrier_ok(dev)))
4284 skb = validate_xmit_skb_list(skb, dev, &again);
4285 if (skb != orig_skb)
4288 skb_set_queue_mapping(skb, queue_id);
4289 txq = skb_get_tx_queue(dev, skb);
4290 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4294 dev_xmit_recursion_inc();
4295 HARD_TX_LOCK(dev, txq, smp_processor_id());
4296 if (!netif_xmit_frozen_or_drv_stopped(txq))
4297 ret = netdev_start_xmit(skb, dev, txq, false);
4298 HARD_TX_UNLOCK(dev, txq);
4299 dev_xmit_recursion_dec();
4304 atomic_long_inc(&dev->tx_dropped);
4305 kfree_skb_list(skb);
4306 return NET_XMIT_DROP;
4308 EXPORT_SYMBOL(__dev_direct_xmit);
4310 /*************************************************************************
4312 *************************************************************************/
4314 int netdev_max_backlog __read_mostly = 1000;
4315 EXPORT_SYMBOL(netdev_max_backlog);
4317 int netdev_tstamp_prequeue __read_mostly = 1;
4318 int netdev_budget __read_mostly = 300;
4319 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4320 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4321 int weight_p __read_mostly = 64; /* old backlog weight */
4322 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4323 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4324 int dev_rx_weight __read_mostly = 64;
4325 int dev_tx_weight __read_mostly = 64;
4326 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4327 int gro_normal_batch __read_mostly = 8;
4329 /* Called with irq disabled */
4330 static inline void ____napi_schedule(struct softnet_data *sd,
4331 struct napi_struct *napi)
4333 struct task_struct *thread;
4335 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4336 /* Paired with smp_mb__before_atomic() in
4337 * napi_enable()/dev_set_threaded().
4338 * Use READ_ONCE() to guarantee a complete
4339 * read on napi->thread. Only call
4340 * wake_up_process() when it's not NULL.
4342 thread = READ_ONCE(napi->thread);
4344 /* Avoid doing set_bit() if the thread is in
4345 * INTERRUPTIBLE state, cause napi_thread_wait()
4346 * makes sure to proceed with napi polling
4347 * if the thread is explicitly woken from here.
4349 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4350 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4351 wake_up_process(thread);
4356 list_add_tail(&napi->poll_list, &sd->poll_list);
4357 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4362 /* One global table that all flow-based protocols share. */
4363 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4364 EXPORT_SYMBOL(rps_sock_flow_table);
4365 u32 rps_cpu_mask __read_mostly;
4366 EXPORT_SYMBOL(rps_cpu_mask);
4368 struct static_key_false rps_needed __read_mostly;
4369 EXPORT_SYMBOL(rps_needed);
4370 struct static_key_false rfs_needed __read_mostly;
4371 EXPORT_SYMBOL(rfs_needed);
4373 static struct rps_dev_flow *
4374 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4375 struct rps_dev_flow *rflow, u16 next_cpu)
4377 if (next_cpu < nr_cpu_ids) {
4378 #ifdef CONFIG_RFS_ACCEL
4379 struct netdev_rx_queue *rxqueue;
4380 struct rps_dev_flow_table *flow_table;
4381 struct rps_dev_flow *old_rflow;
4386 /* Should we steer this flow to a different hardware queue? */
4387 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4388 !(dev->features & NETIF_F_NTUPLE))
4390 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4391 if (rxq_index == skb_get_rx_queue(skb))
4394 rxqueue = dev->_rx + rxq_index;
4395 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4398 flow_id = skb_get_hash(skb) & flow_table->mask;
4399 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4400 rxq_index, flow_id);
4404 rflow = &flow_table->flows[flow_id];
4406 if (old_rflow->filter == rflow->filter)
4407 old_rflow->filter = RPS_NO_FILTER;
4411 per_cpu(softnet_data, next_cpu).input_queue_head;
4414 rflow->cpu = next_cpu;
4419 * get_rps_cpu is called from netif_receive_skb and returns the target
4420 * CPU from the RPS map of the receiving queue for a given skb.
4421 * rcu_read_lock must be held on entry.
4423 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4424 struct rps_dev_flow **rflowp)
4426 const struct rps_sock_flow_table *sock_flow_table;
4427 struct netdev_rx_queue *rxqueue = dev->_rx;
4428 struct rps_dev_flow_table *flow_table;
4429 struct rps_map *map;
4434 if (skb_rx_queue_recorded(skb)) {
4435 u16 index = skb_get_rx_queue(skb);
4437 if (unlikely(index >= dev->real_num_rx_queues)) {
4438 WARN_ONCE(dev->real_num_rx_queues > 1,
4439 "%s received packet on queue %u, but number "
4440 "of RX queues is %u\n",
4441 dev->name, index, dev->real_num_rx_queues);
4447 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4449 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4450 map = rcu_dereference(rxqueue->rps_map);
4451 if (!flow_table && !map)
4454 skb_reset_network_header(skb);
4455 hash = skb_get_hash(skb);
4459 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4460 if (flow_table && sock_flow_table) {
4461 struct rps_dev_flow *rflow;
4465 /* First check into global flow table if there is a match */
4466 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4467 if ((ident ^ hash) & ~rps_cpu_mask)
4470 next_cpu = ident & rps_cpu_mask;
4472 /* OK, now we know there is a match,
4473 * we can look at the local (per receive queue) flow table
4475 rflow = &flow_table->flows[hash & flow_table->mask];
4479 * If the desired CPU (where last recvmsg was done) is
4480 * different from current CPU (one in the rx-queue flow
4481 * table entry), switch if one of the following holds:
4482 * - Current CPU is unset (>= nr_cpu_ids).
4483 * - Current CPU is offline.
4484 * - The current CPU's queue tail has advanced beyond the
4485 * last packet that was enqueued using this table entry.
4486 * This guarantees that all previous packets for the flow
4487 * have been dequeued, thus preserving in order delivery.
4489 if (unlikely(tcpu != next_cpu) &&
4490 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4491 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4492 rflow->last_qtail)) >= 0)) {
4494 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4497 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4507 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4508 if (cpu_online(tcpu)) {
4518 #ifdef CONFIG_RFS_ACCEL
4521 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4522 * @dev: Device on which the filter was set
4523 * @rxq_index: RX queue index
4524 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4525 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4527 * Drivers that implement ndo_rx_flow_steer() should periodically call
4528 * this function for each installed filter and remove the filters for
4529 * which it returns %true.
4531 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4532 u32 flow_id, u16 filter_id)
4534 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4535 struct rps_dev_flow_table *flow_table;
4536 struct rps_dev_flow *rflow;
4541 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4542 if (flow_table && flow_id <= flow_table->mask) {
4543 rflow = &flow_table->flows[flow_id];
4544 cpu = READ_ONCE(rflow->cpu);
4545 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4546 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4547 rflow->last_qtail) <
4548 (int)(10 * flow_table->mask)))
4554 EXPORT_SYMBOL(rps_may_expire_flow);
4556 #endif /* CONFIG_RFS_ACCEL */
4558 /* Called from hardirq (IPI) context */
4559 static void rps_trigger_softirq(void *data)
4561 struct softnet_data *sd = data;
4563 ____napi_schedule(sd, &sd->backlog);
4567 #endif /* CONFIG_RPS */
4570 * Check if this softnet_data structure is another cpu one
4571 * If yes, queue it to our IPI list and return 1
4574 static int rps_ipi_queued(struct softnet_data *sd)
4577 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4580 sd->rps_ipi_next = mysd->rps_ipi_list;
4581 mysd->rps_ipi_list = sd;
4583 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4586 #endif /* CONFIG_RPS */
4590 #ifdef CONFIG_NET_FLOW_LIMIT
4591 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4594 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4596 #ifdef CONFIG_NET_FLOW_LIMIT
4597 struct sd_flow_limit *fl;
4598 struct softnet_data *sd;
4599 unsigned int old_flow, new_flow;
4601 if (qlen < (netdev_max_backlog >> 1))
4604 sd = this_cpu_ptr(&softnet_data);
4607 fl = rcu_dereference(sd->flow_limit);
4609 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4610 old_flow = fl->history[fl->history_head];
4611 fl->history[fl->history_head] = new_flow;
4614 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4616 if (likely(fl->buckets[old_flow]))
4617 fl->buckets[old_flow]--;
4619 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4631 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4632 * queue (may be a remote CPU queue).
4634 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4635 unsigned int *qtail)
4637 struct softnet_data *sd;
4638 unsigned long flags;
4641 sd = &per_cpu(softnet_data, cpu);
4643 local_irq_save(flags);
4646 if (!netif_running(skb->dev))
4648 qlen = skb_queue_len(&sd->input_pkt_queue);
4649 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4652 __skb_queue_tail(&sd->input_pkt_queue, skb);
4653 input_queue_tail_incr_save(sd, qtail);
4655 local_irq_restore(flags);
4656 return NET_RX_SUCCESS;
4659 /* Schedule NAPI for backlog device
4660 * We can use non atomic operation since we own the queue lock
4662 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4663 if (!rps_ipi_queued(sd))
4664 ____napi_schedule(sd, &sd->backlog);
4673 local_irq_restore(flags);
4675 atomic_long_inc(&skb->dev->rx_dropped);
4680 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4682 struct net_device *dev = skb->dev;
4683 struct netdev_rx_queue *rxqueue;
4687 if (skb_rx_queue_recorded(skb)) {
4688 u16 index = skb_get_rx_queue(skb);
4690 if (unlikely(index >= dev->real_num_rx_queues)) {
4691 WARN_ONCE(dev->real_num_rx_queues > 1,
4692 "%s received packet on queue %u, but number "
4693 "of RX queues is %u\n",
4694 dev->name, index, dev->real_num_rx_queues);
4696 return rxqueue; /* Return first rxqueue */
4703 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4704 struct bpf_prog *xdp_prog)
4706 void *orig_data, *orig_data_end, *hard_start;
4707 struct netdev_rx_queue *rxqueue;
4708 bool orig_bcast, orig_host;
4709 u32 mac_len, frame_sz;
4710 __be16 orig_eth_type;
4715 /* The XDP program wants to see the packet starting at the MAC
4718 mac_len = skb->data - skb_mac_header(skb);
4719 hard_start = skb->data - skb_headroom(skb);
4721 /* SKB "head" area always have tailroom for skb_shared_info */
4722 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4723 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4725 rxqueue = netif_get_rxqueue(skb);
4726 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4727 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4728 skb_headlen(skb) + mac_len, true);
4730 orig_data_end = xdp->data_end;
4731 orig_data = xdp->data;
4732 eth = (struct ethhdr *)xdp->data;
4733 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4734 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4735 orig_eth_type = eth->h_proto;
4737 act = bpf_prog_run_xdp(xdp_prog, xdp);
4739 /* check if bpf_xdp_adjust_head was used */
4740 off = xdp->data - orig_data;
4743 __skb_pull(skb, off);
4745 __skb_push(skb, -off);
4747 skb->mac_header += off;
4748 skb_reset_network_header(skb);
4751 /* check if bpf_xdp_adjust_tail was used */
4752 off = xdp->data_end - orig_data_end;
4754 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4755 skb->len += off; /* positive on grow, negative on shrink */
4758 /* check if XDP changed eth hdr such SKB needs update */
4759 eth = (struct ethhdr *)xdp->data;
4760 if ((orig_eth_type != eth->h_proto) ||
4761 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4762 skb->dev->dev_addr)) ||
4763 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4764 __skb_push(skb, ETH_HLEN);
4765 skb->pkt_type = PACKET_HOST;
4766 skb->protocol = eth_type_trans(skb, skb->dev);
4769 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4770 * before calling us again on redirect path. We do not call do_redirect
4771 * as we leave that up to the caller.
4773 * Caller is responsible for managing lifetime of skb (i.e. calling
4774 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4779 __skb_push(skb, mac_len);
4782 metalen = xdp->data - xdp->data_meta;
4784 skb_metadata_set(skb, metalen);
4791 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4792 struct xdp_buff *xdp,
4793 struct bpf_prog *xdp_prog)
4797 /* Reinjected packets coming from act_mirred or similar should
4798 * not get XDP generic processing.
4800 if (skb_is_redirected(skb))
4803 /* XDP packets must be linear and must have sufficient headroom
4804 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4805 * native XDP provides, thus we need to do it here as well.
4807 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4808 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4809 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4810 int troom = skb->tail + skb->data_len - skb->end;
4812 /* In case we have to go down the path and also linearize,
4813 * then lets do the pskb_expand_head() work just once here.
4815 if (pskb_expand_head(skb,
4816 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4817 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4819 if (skb_linearize(skb))
4823 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4830 bpf_warn_invalid_xdp_action(act);
4833 trace_xdp_exception(skb->dev, xdp_prog, act);
4844 /* When doing generic XDP we have to bypass the qdisc layer and the
4845 * network taps in order to match in-driver-XDP behavior.
4847 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4849 struct net_device *dev = skb->dev;
4850 struct netdev_queue *txq;
4851 bool free_skb = true;
4854 txq = netdev_core_pick_tx(dev, skb, NULL);
4855 cpu = smp_processor_id();
4856 HARD_TX_LOCK(dev, txq, cpu);
4857 if (!netif_xmit_stopped(txq)) {
4858 rc = netdev_start_xmit(skb, dev, txq, 0);
4859 if (dev_xmit_complete(rc))
4862 HARD_TX_UNLOCK(dev, txq);
4864 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4869 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4871 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4874 struct xdp_buff xdp;
4878 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4879 if (act != XDP_PASS) {
4882 err = xdp_do_generic_redirect(skb->dev, skb,
4888 generic_xdp_tx(skb, xdp_prog);
4899 EXPORT_SYMBOL_GPL(do_xdp_generic);
4901 static int netif_rx_internal(struct sk_buff *skb)
4905 net_timestamp_check(netdev_tstamp_prequeue, skb);
4907 trace_netif_rx(skb);
4910 if (static_branch_unlikely(&rps_needed)) {
4911 struct rps_dev_flow voidflow, *rflow = &voidflow;
4917 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4919 cpu = smp_processor_id();
4921 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4930 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4937 * netif_rx - post buffer to the network code
4938 * @skb: buffer to post
4940 * This function receives a packet from a device driver and queues it for
4941 * the upper (protocol) levels to process. It always succeeds. The buffer
4942 * may be dropped during processing for congestion control or by the
4946 * NET_RX_SUCCESS (no congestion)
4947 * NET_RX_DROP (packet was dropped)
4951 int netif_rx(struct sk_buff *skb)
4955 trace_netif_rx_entry(skb);
4957 ret = netif_rx_internal(skb);
4958 trace_netif_rx_exit(ret);
4962 EXPORT_SYMBOL(netif_rx);
4964 int netif_rx_ni(struct sk_buff *skb)
4968 trace_netif_rx_ni_entry(skb);
4971 err = netif_rx_internal(skb);
4972 if (local_softirq_pending())
4975 trace_netif_rx_ni_exit(err);
4979 EXPORT_SYMBOL(netif_rx_ni);
4981 int netif_rx_any_context(struct sk_buff *skb)
4984 * If invoked from contexts which do not invoke bottom half
4985 * processing either at return from interrupt or when softrqs are
4986 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4990 return netif_rx(skb);
4992 return netif_rx_ni(skb);
4994 EXPORT_SYMBOL(netif_rx_any_context);
4996 static __latent_entropy void net_tx_action(struct softirq_action *h)
4998 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5000 if (sd->completion_queue) {
5001 struct sk_buff *clist;
5003 local_irq_disable();
5004 clist = sd->completion_queue;
5005 sd->completion_queue = NULL;
5009 struct sk_buff *skb = clist;
5011 clist = clist->next;
5013 WARN_ON(refcount_read(&skb->users));
5014 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5015 trace_consume_skb(skb);
5017 trace_kfree_skb(skb, net_tx_action);
5019 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5022 __kfree_skb_defer(skb);
5026 if (sd->output_queue) {
5029 local_irq_disable();
5030 head = sd->output_queue;
5031 sd->output_queue = NULL;
5032 sd->output_queue_tailp = &sd->output_queue;
5038 struct Qdisc *q = head;
5039 spinlock_t *root_lock = NULL;
5041 head = head->next_sched;
5043 /* We need to make sure head->next_sched is read
5044 * before clearing __QDISC_STATE_SCHED
5046 smp_mb__before_atomic();
5048 if (!(q->flags & TCQ_F_NOLOCK)) {
5049 root_lock = qdisc_lock(q);
5050 spin_lock(root_lock);
5051 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5053 /* There is a synchronize_net() between
5054 * STATE_DEACTIVATED flag being set and
5055 * qdisc_reset()/some_qdisc_is_busy() in
5056 * dev_deactivate(), so we can safely bail out
5057 * early here to avoid data race between
5058 * qdisc_deactivate() and some_qdisc_is_busy()
5059 * for lockless qdisc.
5061 clear_bit(__QDISC_STATE_SCHED, &q->state);
5065 clear_bit(__QDISC_STATE_SCHED, &q->state);
5068 spin_unlock(root_lock);
5074 xfrm_dev_backlog(sd);
5077 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5078 /* This hook is defined here for ATM LANE */
5079 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5080 unsigned char *addr) __read_mostly;
5081 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5084 static inline struct sk_buff *
5085 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5086 struct net_device *orig_dev, bool *another)
5088 #ifdef CONFIG_NET_CLS_ACT
5089 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5090 struct tcf_result cl_res;
5092 /* If there's at least one ingress present somewhere (so
5093 * we get here via enabled static key), remaining devices
5094 * that are not configured with an ingress qdisc will bail
5101 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5105 qdisc_skb_cb(skb)->pkt_len = skb->len;
5106 qdisc_skb_cb(skb)->mru = 0;
5107 qdisc_skb_cb(skb)->post_ct = false;
5108 skb->tc_at_ingress = 1;
5109 mini_qdisc_bstats_cpu_update(miniq, skb);
5111 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5113 case TC_ACT_RECLASSIFY:
5114 skb->tc_index = TC_H_MIN(cl_res.classid);
5117 mini_qdisc_qstats_cpu_drop(miniq);
5125 case TC_ACT_REDIRECT:
5126 /* skb_mac_header check was done by cls/act_bpf, so
5127 * we can safely push the L2 header back before
5128 * redirecting to another netdev
5130 __skb_push(skb, skb->mac_len);
5131 if (skb_do_redirect(skb) == -EAGAIN) {
5132 __skb_pull(skb, skb->mac_len);
5137 case TC_ACT_CONSUMED:
5142 #endif /* CONFIG_NET_CLS_ACT */
5147 * netdev_is_rx_handler_busy - check if receive handler is registered
5148 * @dev: device to check
5150 * Check if a receive handler is already registered for a given device.
5151 * Return true if there one.
5153 * The caller must hold the rtnl_mutex.
5155 bool netdev_is_rx_handler_busy(struct net_device *dev)
5158 return dev && rtnl_dereference(dev->rx_handler);
5160 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5163 * netdev_rx_handler_register - register receive handler
5164 * @dev: device to register a handler for
5165 * @rx_handler: receive handler to register
5166 * @rx_handler_data: data pointer that is used by rx handler
5168 * Register a receive handler for a device. This handler will then be
5169 * called from __netif_receive_skb. A negative errno code is returned
5172 * The caller must hold the rtnl_mutex.
5174 * For a general description of rx_handler, see enum rx_handler_result.
5176 int netdev_rx_handler_register(struct net_device *dev,
5177 rx_handler_func_t *rx_handler,
5178 void *rx_handler_data)
5180 if (netdev_is_rx_handler_busy(dev))
5183 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5186 /* Note: rx_handler_data must be set before rx_handler */
5187 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5188 rcu_assign_pointer(dev->rx_handler, rx_handler);
5192 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5195 * netdev_rx_handler_unregister - unregister receive handler
5196 * @dev: device to unregister a handler from
5198 * Unregister a receive handler from a device.
5200 * The caller must hold the rtnl_mutex.
5202 void netdev_rx_handler_unregister(struct net_device *dev)
5206 RCU_INIT_POINTER(dev->rx_handler, NULL);
5207 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5208 * section has a guarantee to see a non NULL rx_handler_data
5212 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5214 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5217 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5218 * the special handling of PFMEMALLOC skbs.
5220 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5222 switch (skb->protocol) {
5223 case htons(ETH_P_ARP):
5224 case htons(ETH_P_IP):
5225 case htons(ETH_P_IPV6):
5226 case htons(ETH_P_8021Q):
5227 case htons(ETH_P_8021AD):
5234 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5235 int *ret, struct net_device *orig_dev)
5237 if (nf_hook_ingress_active(skb)) {
5241 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5246 ingress_retval = nf_hook_ingress(skb);
5248 return ingress_retval;
5253 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5254 struct packet_type **ppt_prev)
5256 struct packet_type *ptype, *pt_prev;
5257 rx_handler_func_t *rx_handler;
5258 struct sk_buff *skb = *pskb;
5259 struct net_device *orig_dev;
5260 bool deliver_exact = false;
5261 int ret = NET_RX_DROP;
5264 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5266 trace_netif_receive_skb(skb);
5268 orig_dev = skb->dev;
5270 skb_reset_network_header(skb);
5271 if (!skb_transport_header_was_set(skb))
5272 skb_reset_transport_header(skb);
5273 skb_reset_mac_len(skb);
5278 skb->skb_iif = skb->dev->ifindex;
5280 __this_cpu_inc(softnet_data.processed);
5282 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5286 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5289 if (ret2 != XDP_PASS) {
5295 if (eth_type_vlan(skb->protocol)) {
5296 skb = skb_vlan_untag(skb);
5301 if (skb_skip_tc_classify(skb))
5307 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5309 ret = deliver_skb(skb, pt_prev, orig_dev);
5313 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5315 ret = deliver_skb(skb, pt_prev, orig_dev);
5320 #ifdef CONFIG_NET_INGRESS
5321 if (static_branch_unlikely(&ingress_needed_key)) {
5322 bool another = false;
5324 nf_skip_egress(skb, true);
5325 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5332 nf_skip_egress(skb, false);
5333 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5337 skb_reset_redirect(skb);
5339 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5342 if (skb_vlan_tag_present(skb)) {
5344 ret = deliver_skb(skb, pt_prev, orig_dev);
5347 if (vlan_do_receive(&skb))
5349 else if (unlikely(!skb))
5353 rx_handler = rcu_dereference(skb->dev->rx_handler);
5356 ret = deliver_skb(skb, pt_prev, orig_dev);
5359 switch (rx_handler(&skb)) {
5360 case RX_HANDLER_CONSUMED:
5361 ret = NET_RX_SUCCESS;
5363 case RX_HANDLER_ANOTHER:
5365 case RX_HANDLER_EXACT:
5366 deliver_exact = true;
5368 case RX_HANDLER_PASS:
5375 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5377 if (skb_vlan_tag_get_id(skb)) {
5378 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5381 skb->pkt_type = PACKET_OTHERHOST;
5382 } else if (eth_type_vlan(skb->protocol)) {
5383 /* Outer header is 802.1P with vlan 0, inner header is
5384 * 802.1Q or 802.1AD and vlan_do_receive() above could
5385 * not find vlan dev for vlan id 0.
5387 __vlan_hwaccel_clear_tag(skb);
5388 skb = skb_vlan_untag(skb);
5391 if (vlan_do_receive(&skb))
5392 /* After stripping off 802.1P header with vlan 0
5393 * vlan dev is found for inner header.
5396 else if (unlikely(!skb))
5399 /* We have stripped outer 802.1P vlan 0 header.
5400 * But could not find vlan dev.
5401 * check again for vlan id to set OTHERHOST.
5405 /* Note: we might in the future use prio bits
5406 * and set skb->priority like in vlan_do_receive()
5407 * For the time being, just ignore Priority Code Point
5409 __vlan_hwaccel_clear_tag(skb);
5412 type = skb->protocol;
5414 /* deliver only exact match when indicated */
5415 if (likely(!deliver_exact)) {
5416 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5417 &ptype_base[ntohs(type) &
5421 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5422 &orig_dev->ptype_specific);
5424 if (unlikely(skb->dev != orig_dev)) {
5425 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5426 &skb->dev->ptype_specific);
5430 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5432 *ppt_prev = pt_prev;
5436 atomic_long_inc(&skb->dev->rx_dropped);
5438 atomic_long_inc(&skb->dev->rx_nohandler);
5440 /* Jamal, now you will not able to escape explaining
5441 * me how you were going to use this. :-)
5447 /* The invariant here is that if *ppt_prev is not NULL
5448 * then skb should also be non-NULL.
5450 * Apparently *ppt_prev assignment above holds this invariant due to
5451 * skb dereferencing near it.
5457 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5459 struct net_device *orig_dev = skb->dev;
5460 struct packet_type *pt_prev = NULL;
5463 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5465 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5466 skb->dev, pt_prev, orig_dev);
5471 * netif_receive_skb_core - special purpose version of netif_receive_skb
5472 * @skb: buffer to process
5474 * More direct receive version of netif_receive_skb(). It should
5475 * only be used by callers that have a need to skip RPS and Generic XDP.
5476 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5478 * This function may only be called from softirq context and interrupts
5479 * should be enabled.
5481 * Return values (usually ignored):
5482 * NET_RX_SUCCESS: no congestion
5483 * NET_RX_DROP: packet was dropped
5485 int netif_receive_skb_core(struct sk_buff *skb)
5490 ret = __netif_receive_skb_one_core(skb, false);
5495 EXPORT_SYMBOL(netif_receive_skb_core);
5497 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5498 struct packet_type *pt_prev,
5499 struct net_device *orig_dev)
5501 struct sk_buff *skb, *next;
5505 if (list_empty(head))
5507 if (pt_prev->list_func != NULL)
5508 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5509 ip_list_rcv, head, pt_prev, orig_dev);
5511 list_for_each_entry_safe(skb, next, head, list) {
5512 skb_list_del_init(skb);
5513 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5517 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5519 /* Fast-path assumptions:
5520 * - There is no RX handler.
5521 * - Only one packet_type matches.
5522 * If either of these fails, we will end up doing some per-packet
5523 * processing in-line, then handling the 'last ptype' for the whole
5524 * sublist. This can't cause out-of-order delivery to any single ptype,
5525 * because the 'last ptype' must be constant across the sublist, and all
5526 * other ptypes are handled per-packet.
5528 /* Current (common) ptype of sublist */
5529 struct packet_type *pt_curr = NULL;
5530 /* Current (common) orig_dev of sublist */
5531 struct net_device *od_curr = NULL;
5532 struct list_head sublist;
5533 struct sk_buff *skb, *next;
5535 INIT_LIST_HEAD(&sublist);
5536 list_for_each_entry_safe(skb, next, head, list) {
5537 struct net_device *orig_dev = skb->dev;
5538 struct packet_type *pt_prev = NULL;
5540 skb_list_del_init(skb);
5541 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5544 if (pt_curr != pt_prev || od_curr != orig_dev) {
5545 /* dispatch old sublist */
5546 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5547 /* start new sublist */
5548 INIT_LIST_HEAD(&sublist);
5552 list_add_tail(&skb->list, &sublist);
5555 /* dispatch final sublist */
5556 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5559 static int __netif_receive_skb(struct sk_buff *skb)
5563 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5564 unsigned int noreclaim_flag;
5567 * PFMEMALLOC skbs are special, they should
5568 * - be delivered to SOCK_MEMALLOC sockets only
5569 * - stay away from userspace
5570 * - have bounded memory usage
5572 * Use PF_MEMALLOC as this saves us from propagating the allocation
5573 * context down to all allocation sites.
5575 noreclaim_flag = memalloc_noreclaim_save();
5576 ret = __netif_receive_skb_one_core(skb, true);
5577 memalloc_noreclaim_restore(noreclaim_flag);
5579 ret = __netif_receive_skb_one_core(skb, false);
5584 static void __netif_receive_skb_list(struct list_head *head)
5586 unsigned long noreclaim_flag = 0;
5587 struct sk_buff *skb, *next;
5588 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5590 list_for_each_entry_safe(skb, next, head, list) {
5591 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5592 struct list_head sublist;
5594 /* Handle the previous sublist */
5595 list_cut_before(&sublist, head, &skb->list);
5596 if (!list_empty(&sublist))
5597 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5598 pfmemalloc = !pfmemalloc;
5599 /* See comments in __netif_receive_skb */
5601 noreclaim_flag = memalloc_noreclaim_save();
5603 memalloc_noreclaim_restore(noreclaim_flag);
5606 /* Handle the remaining sublist */
5607 if (!list_empty(head))
5608 __netif_receive_skb_list_core(head, pfmemalloc);
5609 /* Restore pflags */
5611 memalloc_noreclaim_restore(noreclaim_flag);
5614 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5616 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5617 struct bpf_prog *new = xdp->prog;
5620 switch (xdp->command) {
5621 case XDP_SETUP_PROG:
5622 rcu_assign_pointer(dev->xdp_prog, new);
5627 static_branch_dec(&generic_xdp_needed_key);
5628 } else if (new && !old) {
5629 static_branch_inc(&generic_xdp_needed_key);
5630 dev_disable_lro(dev);
5631 dev_disable_gro_hw(dev);
5643 static int netif_receive_skb_internal(struct sk_buff *skb)
5647 net_timestamp_check(netdev_tstamp_prequeue, skb);
5649 if (skb_defer_rx_timestamp(skb))
5650 return NET_RX_SUCCESS;
5654 if (static_branch_unlikely(&rps_needed)) {
5655 struct rps_dev_flow voidflow, *rflow = &voidflow;
5656 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5659 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5665 ret = __netif_receive_skb(skb);
5670 static void netif_receive_skb_list_internal(struct list_head *head)
5672 struct sk_buff *skb, *next;
5673 struct list_head sublist;
5675 INIT_LIST_HEAD(&sublist);
5676 list_for_each_entry_safe(skb, next, head, list) {
5677 net_timestamp_check(netdev_tstamp_prequeue, skb);
5678 skb_list_del_init(skb);
5679 if (!skb_defer_rx_timestamp(skb))
5680 list_add_tail(&skb->list, &sublist);
5682 list_splice_init(&sublist, head);
5686 if (static_branch_unlikely(&rps_needed)) {
5687 list_for_each_entry_safe(skb, next, head, list) {
5688 struct rps_dev_flow voidflow, *rflow = &voidflow;
5689 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5692 /* Will be handled, remove from list */
5693 skb_list_del_init(skb);
5694 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5699 __netif_receive_skb_list(head);
5704 * netif_receive_skb - process receive buffer from network
5705 * @skb: buffer to process
5707 * netif_receive_skb() is the main receive data processing function.
5708 * It always succeeds. The buffer may be dropped during processing
5709 * for congestion control or by the protocol layers.
5711 * This function may only be called from softirq context and interrupts
5712 * should be enabled.
5714 * Return values (usually ignored):
5715 * NET_RX_SUCCESS: no congestion
5716 * NET_RX_DROP: packet was dropped
5718 int netif_receive_skb(struct sk_buff *skb)
5722 trace_netif_receive_skb_entry(skb);
5724 ret = netif_receive_skb_internal(skb);
5725 trace_netif_receive_skb_exit(ret);
5729 EXPORT_SYMBOL(netif_receive_skb);
5732 * netif_receive_skb_list - process many receive buffers from network
5733 * @head: list of skbs to process.
5735 * Since return value of netif_receive_skb() is normally ignored, and
5736 * wouldn't be meaningful for a list, this function returns void.
5738 * This function may only be called from softirq context and interrupts
5739 * should be enabled.
5741 void netif_receive_skb_list(struct list_head *head)
5743 struct sk_buff *skb;
5745 if (list_empty(head))
5747 if (trace_netif_receive_skb_list_entry_enabled()) {
5748 list_for_each_entry(skb, head, list)
5749 trace_netif_receive_skb_list_entry(skb);
5751 netif_receive_skb_list_internal(head);
5752 trace_netif_receive_skb_list_exit(0);
5754 EXPORT_SYMBOL(netif_receive_skb_list);
5756 static DEFINE_PER_CPU(struct work_struct, flush_works);
5758 /* Network device is going away, flush any packets still pending */
5759 static void flush_backlog(struct work_struct *work)
5761 struct sk_buff *skb, *tmp;
5762 struct softnet_data *sd;
5765 sd = this_cpu_ptr(&softnet_data);
5767 local_irq_disable();
5769 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5770 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5771 __skb_unlink(skb, &sd->input_pkt_queue);
5772 dev_kfree_skb_irq(skb);
5773 input_queue_head_incr(sd);
5779 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5780 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5781 __skb_unlink(skb, &sd->process_queue);
5783 input_queue_head_incr(sd);
5789 static bool flush_required(int cpu)
5791 #if IS_ENABLED(CONFIG_RPS)
5792 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5795 local_irq_disable();
5798 /* as insertion into process_queue happens with the rps lock held,
5799 * process_queue access may race only with dequeue
5801 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5802 !skb_queue_empty_lockless(&sd->process_queue);
5808 /* without RPS we can't safely check input_pkt_queue: during a
5809 * concurrent remote skb_queue_splice() we can detect as empty both
5810 * input_pkt_queue and process_queue even if the latter could end-up
5811 * containing a lot of packets.
5816 static void flush_all_backlogs(void)
5818 static cpumask_t flush_cpus;
5821 /* since we are under rtnl lock protection we can use static data
5822 * for the cpumask and avoid allocating on stack the possibly
5829 cpumask_clear(&flush_cpus);
5830 for_each_online_cpu(cpu) {
5831 if (flush_required(cpu)) {
5832 queue_work_on(cpu, system_highpri_wq,
5833 per_cpu_ptr(&flush_works, cpu));
5834 cpumask_set_cpu(cpu, &flush_cpus);
5838 /* we can have in flight packet[s] on the cpus we are not flushing,
5839 * synchronize_net() in unregister_netdevice_many() will take care of
5842 for_each_cpu(cpu, &flush_cpus)
5843 flush_work(per_cpu_ptr(&flush_works, cpu));
5848 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5849 static void gro_normal_list(struct napi_struct *napi)
5851 if (!napi->rx_count)
5853 netif_receive_skb_list_internal(&napi->rx_list);
5854 INIT_LIST_HEAD(&napi->rx_list);
5858 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5859 * pass the whole batch up to the stack.
5861 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5863 list_add_tail(&skb->list, &napi->rx_list);
5864 napi->rx_count += segs;
5865 if (napi->rx_count >= gro_normal_batch)
5866 gro_normal_list(napi);
5869 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5871 struct packet_offload *ptype;
5872 __be16 type = skb->protocol;
5873 struct list_head *head = &offload_base;
5876 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5878 if (NAPI_GRO_CB(skb)->count == 1) {
5879 skb_shinfo(skb)->gso_size = 0;
5884 list_for_each_entry_rcu(ptype, head, list) {
5885 if (ptype->type != type || !ptype->callbacks.gro_complete)
5888 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5889 ipv6_gro_complete, inet_gro_complete,
5896 WARN_ON(&ptype->list == head);
5902 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5905 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5908 struct list_head *head = &napi->gro_hash[index].list;
5909 struct sk_buff *skb, *p;
5911 list_for_each_entry_safe_reverse(skb, p, head, list) {
5912 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5914 skb_list_del_init(skb);
5915 napi_gro_complete(napi, skb);
5916 napi->gro_hash[index].count--;
5919 if (!napi->gro_hash[index].count)
5920 __clear_bit(index, &napi->gro_bitmask);
5923 /* napi->gro_hash[].list contains packets ordered by age.
5924 * youngest packets at the head of it.
5925 * Complete skbs in reverse order to reduce latencies.
5927 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5929 unsigned long bitmask = napi->gro_bitmask;
5930 unsigned int i, base = ~0U;
5932 while ((i = ffs(bitmask)) != 0) {
5935 __napi_gro_flush_chain(napi, base, flush_old);
5938 EXPORT_SYMBOL(napi_gro_flush);
5940 static void gro_list_prepare(const struct list_head *head,
5941 const struct sk_buff *skb)
5943 unsigned int maclen = skb->dev->hard_header_len;
5944 u32 hash = skb_get_hash_raw(skb);
5947 list_for_each_entry(p, head, list) {
5948 unsigned long diffs;
5950 NAPI_GRO_CB(p)->flush = 0;
5952 if (hash != skb_get_hash_raw(p)) {
5953 NAPI_GRO_CB(p)->same_flow = 0;
5957 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5958 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5959 if (skb_vlan_tag_present(p))
5960 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5961 diffs |= skb_metadata_differs(p, skb);
5962 if (maclen == ETH_HLEN)
5963 diffs |= compare_ether_header(skb_mac_header(p),
5964 skb_mac_header(skb));
5966 diffs = memcmp(skb_mac_header(p),
5967 skb_mac_header(skb),
5970 /* in most common scenarions 'slow_gro' is 0
5971 * otherwise we are already on some slower paths
5972 * either skip all the infrequent tests altogether or
5973 * avoid trying too hard to skip each of them individually
5975 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
5976 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5977 struct tc_skb_ext *skb_ext;
5978 struct tc_skb_ext *p_ext;
5981 diffs |= p->sk != skb->sk;
5982 diffs |= skb_metadata_dst_cmp(p, skb);
5983 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
5985 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5986 skb_ext = skb_ext_find(skb, TC_SKB_EXT);
5987 p_ext = skb_ext_find(p, TC_SKB_EXT);
5989 diffs |= (!!p_ext) ^ (!!skb_ext);
5990 if (!diffs && unlikely(skb_ext))
5991 diffs |= p_ext->chain ^ skb_ext->chain;
5995 NAPI_GRO_CB(p)->same_flow = !diffs;
5999 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
6001 const struct skb_shared_info *pinfo = skb_shinfo(skb);
6002 const skb_frag_t *frag0 = &pinfo->frags[0];
6004 NAPI_GRO_CB(skb)->data_offset = 0;
6005 NAPI_GRO_CB(skb)->frag0 = NULL;
6006 NAPI_GRO_CB(skb)->frag0_len = 0;
6008 if (!skb_headlen(skb) && pinfo->nr_frags &&
6009 !PageHighMem(skb_frag_page(frag0)) &&
6010 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
6011 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
6012 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
6013 skb_frag_size(frag0),
6014 skb->end - skb->tail);
6018 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
6020 struct skb_shared_info *pinfo = skb_shinfo(skb);
6022 BUG_ON(skb->end - skb->tail < grow);
6024 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6026 skb->data_len -= grow;
6029 skb_frag_off_add(&pinfo->frags[0], grow);
6030 skb_frag_size_sub(&pinfo->frags[0], grow);
6032 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6033 skb_frag_unref(skb, 0);
6034 memmove(pinfo->frags, pinfo->frags + 1,
6035 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
6039 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6041 struct sk_buff *oldest;
6043 oldest = list_last_entry(head, struct sk_buff, list);
6045 /* We are called with head length >= MAX_GRO_SKBS, so this is
6048 if (WARN_ON_ONCE(!oldest))
6051 /* Do not adjust napi->gro_hash[].count, caller is adding a new
6054 skb_list_del_init(oldest);
6055 napi_gro_complete(napi, oldest);
6058 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6060 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6061 struct gro_list *gro_list = &napi->gro_hash[bucket];
6062 struct list_head *head = &offload_base;
6063 struct packet_offload *ptype;
6064 __be16 type = skb->protocol;
6065 struct sk_buff *pp = NULL;
6066 enum gro_result ret;
6070 if (netif_elide_gro(skb->dev))
6073 gro_list_prepare(&gro_list->list, skb);
6076 list_for_each_entry_rcu(ptype, head, list) {
6077 if (ptype->type != type || !ptype->callbacks.gro_receive)
6080 skb_set_network_header(skb, skb_gro_offset(skb));
6081 skb_reset_mac_len(skb);
6082 NAPI_GRO_CB(skb)->same_flow = 0;
6083 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6084 NAPI_GRO_CB(skb)->free = 0;
6085 NAPI_GRO_CB(skb)->encap_mark = 0;
6086 NAPI_GRO_CB(skb)->recursion_counter = 0;
6087 NAPI_GRO_CB(skb)->is_fou = 0;
6088 NAPI_GRO_CB(skb)->is_atomic = 1;
6089 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6091 /* Setup for GRO checksum validation */
6092 switch (skb->ip_summed) {
6093 case CHECKSUM_COMPLETE:
6094 NAPI_GRO_CB(skb)->csum = skb->csum;
6095 NAPI_GRO_CB(skb)->csum_valid = 1;
6096 NAPI_GRO_CB(skb)->csum_cnt = 0;
6098 case CHECKSUM_UNNECESSARY:
6099 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6100 NAPI_GRO_CB(skb)->csum_valid = 0;
6103 NAPI_GRO_CB(skb)->csum_cnt = 0;
6104 NAPI_GRO_CB(skb)->csum_valid = 0;
6107 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6108 ipv6_gro_receive, inet_gro_receive,
6109 &gro_list->list, skb);
6114 if (&ptype->list == head)
6117 if (PTR_ERR(pp) == -EINPROGRESS) {
6122 same_flow = NAPI_GRO_CB(skb)->same_flow;
6123 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6126 skb_list_del_init(pp);
6127 napi_gro_complete(napi, pp);
6134 if (NAPI_GRO_CB(skb)->flush)
6137 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6138 gro_flush_oldest(napi, &gro_list->list);
6142 NAPI_GRO_CB(skb)->count = 1;
6143 NAPI_GRO_CB(skb)->age = jiffies;
6144 NAPI_GRO_CB(skb)->last = skb;
6145 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6146 list_add(&skb->list, &gro_list->list);
6150 grow = skb_gro_offset(skb) - skb_headlen(skb);
6152 gro_pull_from_frag0(skb, grow);
6154 if (gro_list->count) {
6155 if (!test_bit(bucket, &napi->gro_bitmask))
6156 __set_bit(bucket, &napi->gro_bitmask);
6157 } else if (test_bit(bucket, &napi->gro_bitmask)) {
6158 __clear_bit(bucket, &napi->gro_bitmask);
6168 struct packet_offload *gro_find_receive_by_type(__be16 type)
6170 struct list_head *offload_head = &offload_base;
6171 struct packet_offload *ptype;
6173 list_for_each_entry_rcu(ptype, offload_head, list) {
6174 if (ptype->type != type || !ptype->callbacks.gro_receive)
6180 EXPORT_SYMBOL(gro_find_receive_by_type);
6182 struct packet_offload *gro_find_complete_by_type(__be16 type)
6184 struct list_head *offload_head = &offload_base;
6185 struct packet_offload *ptype;
6187 list_for_each_entry_rcu(ptype, offload_head, list) {
6188 if (ptype->type != type || !ptype->callbacks.gro_complete)
6194 EXPORT_SYMBOL(gro_find_complete_by_type);
6196 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6197 struct sk_buff *skb,
6202 gro_normal_one(napi, skb, 1);
6205 case GRO_MERGED_FREE:
6206 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6207 napi_skb_free_stolen_head(skb);
6208 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6211 __kfree_skb_defer(skb);
6223 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6227 skb_mark_napi_id(skb, napi);
6228 trace_napi_gro_receive_entry(skb);
6230 skb_gro_reset_offset(skb, 0);
6232 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6233 trace_napi_gro_receive_exit(ret);
6237 EXPORT_SYMBOL(napi_gro_receive);
6239 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6241 if (unlikely(skb->pfmemalloc)) {
6245 __skb_pull(skb, skb_headlen(skb));
6246 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6247 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6248 __vlan_hwaccel_clear_tag(skb);
6249 skb->dev = napi->dev;
6252 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6253 skb->pkt_type = PACKET_HOST;
6255 skb->encapsulation = 0;
6256 skb_shinfo(skb)->gso_type = 0;
6257 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6258 if (unlikely(skb->slow_gro)) {
6268 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6270 struct sk_buff *skb = napi->skb;
6273 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6276 skb_mark_napi_id(skb, napi);
6281 EXPORT_SYMBOL(napi_get_frags);
6283 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6284 struct sk_buff *skb,
6290 __skb_push(skb, ETH_HLEN);
6291 skb->protocol = eth_type_trans(skb, skb->dev);
6292 if (ret == GRO_NORMAL)
6293 gro_normal_one(napi, skb, 1);
6296 case GRO_MERGED_FREE:
6297 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6298 napi_skb_free_stolen_head(skb);
6300 napi_reuse_skb(napi, skb);
6311 /* Upper GRO stack assumes network header starts at gro_offset=0
6312 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6313 * We copy ethernet header into skb->data to have a common layout.
6315 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6317 struct sk_buff *skb = napi->skb;
6318 const struct ethhdr *eth;
6319 unsigned int hlen = sizeof(*eth);
6323 skb_reset_mac_header(skb);
6324 skb_gro_reset_offset(skb, hlen);
6326 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6327 eth = skb_gro_header_slow(skb, hlen, 0);
6328 if (unlikely(!eth)) {
6329 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6330 __func__, napi->dev->name);
6331 napi_reuse_skb(napi, skb);
6335 eth = (const struct ethhdr *)skb->data;
6336 gro_pull_from_frag0(skb, hlen);
6337 NAPI_GRO_CB(skb)->frag0 += hlen;
6338 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6340 __skb_pull(skb, hlen);
6343 * This works because the only protocols we care about don't require
6345 * We'll fix it up properly in napi_frags_finish()
6347 skb->protocol = eth->h_proto;
6352 gro_result_t napi_gro_frags(struct napi_struct *napi)
6355 struct sk_buff *skb = napi_frags_skb(napi);
6357 trace_napi_gro_frags_entry(skb);
6359 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6360 trace_napi_gro_frags_exit(ret);
6364 EXPORT_SYMBOL(napi_gro_frags);
6366 /* Compute the checksum from gro_offset and return the folded value
6367 * after adding in any pseudo checksum.
6369 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6374 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6376 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6377 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6378 /* See comments in __skb_checksum_complete(). */
6380 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6381 !skb->csum_complete_sw)
6382 netdev_rx_csum_fault(skb->dev, skb);
6385 NAPI_GRO_CB(skb)->csum = wsum;
6386 NAPI_GRO_CB(skb)->csum_valid = 1;
6390 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6392 static void net_rps_send_ipi(struct softnet_data *remsd)
6396 struct softnet_data *next = remsd->rps_ipi_next;
6398 if (cpu_online(remsd->cpu))
6399 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6406 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6407 * Note: called with local irq disabled, but exits with local irq enabled.
6409 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6412 struct softnet_data *remsd = sd->rps_ipi_list;
6415 sd->rps_ipi_list = NULL;
6419 /* Send pending IPI's to kick RPS processing on remote cpus. */
6420 net_rps_send_ipi(remsd);
6426 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6429 return sd->rps_ipi_list != NULL;
6435 static int process_backlog(struct napi_struct *napi, int quota)
6437 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6441 /* Check if we have pending ipi, its better to send them now,
6442 * not waiting net_rx_action() end.
6444 if (sd_has_rps_ipi_waiting(sd)) {
6445 local_irq_disable();
6446 net_rps_action_and_irq_enable(sd);
6449 napi->weight = dev_rx_weight;
6451 struct sk_buff *skb;
6453 while ((skb = __skb_dequeue(&sd->process_queue))) {
6455 __netif_receive_skb(skb);
6457 input_queue_head_incr(sd);
6458 if (++work >= quota)
6463 local_irq_disable();
6465 if (skb_queue_empty(&sd->input_pkt_queue)) {
6467 * Inline a custom version of __napi_complete().
6468 * only current cpu owns and manipulates this napi,
6469 * and NAPI_STATE_SCHED is the only possible flag set
6471 * We can use a plain write instead of clear_bit(),
6472 * and we dont need an smp_mb() memory barrier.
6477 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6478 &sd->process_queue);
6488 * __napi_schedule - schedule for receive
6489 * @n: entry to schedule
6491 * The entry's receive function will be scheduled to run.
6492 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6494 void __napi_schedule(struct napi_struct *n)
6496 unsigned long flags;
6498 local_irq_save(flags);
6499 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6500 local_irq_restore(flags);
6502 EXPORT_SYMBOL(__napi_schedule);
6505 * napi_schedule_prep - check if napi can be scheduled
6508 * Test if NAPI routine is already running, and if not mark
6509 * it as running. This is used as a condition variable to
6510 * insure only one NAPI poll instance runs. We also make
6511 * sure there is no pending NAPI disable.
6513 bool napi_schedule_prep(struct napi_struct *n)
6515 unsigned long val, new;
6518 val = READ_ONCE(n->state);
6519 if (unlikely(val & NAPIF_STATE_DISABLE))
6521 new = val | NAPIF_STATE_SCHED;
6523 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6524 * This was suggested by Alexander Duyck, as compiler
6525 * emits better code than :
6526 * if (val & NAPIF_STATE_SCHED)
6527 * new |= NAPIF_STATE_MISSED;
6529 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6531 } while (cmpxchg(&n->state, val, new) != val);
6533 return !(val & NAPIF_STATE_SCHED);
6535 EXPORT_SYMBOL(napi_schedule_prep);
6538 * __napi_schedule_irqoff - schedule for receive
6539 * @n: entry to schedule
6541 * Variant of __napi_schedule() assuming hard irqs are masked.
6543 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6544 * because the interrupt disabled assumption might not be true
6545 * due to force-threaded interrupts and spinlock substitution.
6547 void __napi_schedule_irqoff(struct napi_struct *n)
6549 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6550 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6554 EXPORT_SYMBOL(__napi_schedule_irqoff);
6556 bool napi_complete_done(struct napi_struct *n, int work_done)
6558 unsigned long flags, val, new, timeout = 0;
6562 * 1) Don't let napi dequeue from the cpu poll list
6563 * just in case its running on a different cpu.
6564 * 2) If we are busy polling, do nothing here, we have
6565 * the guarantee we will be called later.
6567 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6568 NAPIF_STATE_IN_BUSY_POLL)))
6573 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6574 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6576 if (n->defer_hard_irqs_count > 0) {
6577 n->defer_hard_irqs_count--;
6578 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6582 if (n->gro_bitmask) {
6583 /* When the NAPI instance uses a timeout and keeps postponing
6584 * it, we need to bound somehow the time packets are kept in
6587 napi_gro_flush(n, !!timeout);
6592 if (unlikely(!list_empty(&n->poll_list))) {
6593 /* If n->poll_list is not empty, we need to mask irqs */
6594 local_irq_save(flags);
6595 list_del_init(&n->poll_list);
6596 local_irq_restore(flags);
6600 val = READ_ONCE(n->state);
6602 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6604 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6605 NAPIF_STATE_SCHED_THREADED |
6606 NAPIF_STATE_PREFER_BUSY_POLL);
6608 /* If STATE_MISSED was set, leave STATE_SCHED set,
6609 * because we will call napi->poll() one more time.
6610 * This C code was suggested by Alexander Duyck to help gcc.
6612 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6614 } while (cmpxchg(&n->state, val, new) != val);
6616 if (unlikely(val & NAPIF_STATE_MISSED)) {
6622 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6623 HRTIMER_MODE_REL_PINNED);
6626 EXPORT_SYMBOL(napi_complete_done);
6628 /* must be called under rcu_read_lock(), as we dont take a reference */
6629 static struct napi_struct *napi_by_id(unsigned int napi_id)
6631 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6632 struct napi_struct *napi;
6634 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6635 if (napi->napi_id == napi_id)
6641 #if defined(CONFIG_NET_RX_BUSY_POLL)
6643 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6645 if (!skip_schedule) {
6646 gro_normal_list(napi);
6647 __napi_schedule(napi);
6651 if (napi->gro_bitmask) {
6652 /* flush too old packets
6653 * If HZ < 1000, flush all packets.
6655 napi_gro_flush(napi, HZ >= 1000);
6658 gro_normal_list(napi);
6659 clear_bit(NAPI_STATE_SCHED, &napi->state);
6662 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6665 bool skip_schedule = false;
6666 unsigned long timeout;
6669 /* Busy polling means there is a high chance device driver hard irq
6670 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6671 * set in napi_schedule_prep().
6672 * Since we are about to call napi->poll() once more, we can safely
6673 * clear NAPI_STATE_MISSED.
6675 * Note: x86 could use a single "lock and ..." instruction
6676 * to perform these two clear_bit()
6678 clear_bit(NAPI_STATE_MISSED, &napi->state);
6679 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6683 if (prefer_busy_poll) {
6684 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6685 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6686 if (napi->defer_hard_irqs_count && timeout) {
6687 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6688 skip_schedule = true;
6692 /* All we really want here is to re-enable device interrupts.
6693 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6695 rc = napi->poll(napi, budget);
6696 /* We can't gro_normal_list() here, because napi->poll() might have
6697 * rearmed the napi (napi_complete_done()) in which case it could
6698 * already be running on another CPU.
6700 trace_napi_poll(napi, rc, budget);
6701 netpoll_poll_unlock(have_poll_lock);
6703 __busy_poll_stop(napi, skip_schedule);
6707 void napi_busy_loop(unsigned int napi_id,
6708 bool (*loop_end)(void *, unsigned long),
6709 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6711 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6712 int (*napi_poll)(struct napi_struct *napi, int budget);
6713 void *have_poll_lock = NULL;
6714 struct napi_struct *napi;
6721 napi = napi_by_id(napi_id);
6731 unsigned long val = READ_ONCE(napi->state);
6733 /* If multiple threads are competing for this napi,
6734 * we avoid dirtying napi->state as much as we can.
6736 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6737 NAPIF_STATE_IN_BUSY_POLL)) {
6738 if (prefer_busy_poll)
6739 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6742 if (cmpxchg(&napi->state, val,
6743 val | NAPIF_STATE_IN_BUSY_POLL |
6744 NAPIF_STATE_SCHED) != val) {
6745 if (prefer_busy_poll)
6746 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6749 have_poll_lock = netpoll_poll_lock(napi);
6750 napi_poll = napi->poll;
6752 work = napi_poll(napi, budget);
6753 trace_napi_poll(napi, work, budget);
6754 gro_normal_list(napi);
6757 __NET_ADD_STATS(dev_net(napi->dev),
6758 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6761 if (!loop_end || loop_end(loop_end_arg, start_time))
6764 if (unlikely(need_resched())) {
6766 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6770 if (loop_end(loop_end_arg, start_time))
6777 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6782 EXPORT_SYMBOL(napi_busy_loop);
6784 #endif /* CONFIG_NET_RX_BUSY_POLL */
6786 static void napi_hash_add(struct napi_struct *napi)
6788 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6791 spin_lock(&napi_hash_lock);
6793 /* 0..NR_CPUS range is reserved for sender_cpu use */
6795 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6796 napi_gen_id = MIN_NAPI_ID;
6797 } while (napi_by_id(napi_gen_id));
6798 napi->napi_id = napi_gen_id;
6800 hlist_add_head_rcu(&napi->napi_hash_node,
6801 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6803 spin_unlock(&napi_hash_lock);
6806 /* Warning : caller is responsible to make sure rcu grace period
6807 * is respected before freeing memory containing @napi
6809 static void napi_hash_del(struct napi_struct *napi)
6811 spin_lock(&napi_hash_lock);
6813 hlist_del_init_rcu(&napi->napi_hash_node);
6815 spin_unlock(&napi_hash_lock);
6818 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6820 struct napi_struct *napi;
6822 napi = container_of(timer, struct napi_struct, timer);
6824 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6825 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6827 if (!napi_disable_pending(napi) &&
6828 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6829 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6830 __napi_schedule_irqoff(napi);
6833 return HRTIMER_NORESTART;
6836 static void init_gro_hash(struct napi_struct *napi)
6840 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6841 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6842 napi->gro_hash[i].count = 0;
6844 napi->gro_bitmask = 0;
6847 int dev_set_threaded(struct net_device *dev, bool threaded)
6849 struct napi_struct *napi;
6852 if (dev->threaded == threaded)
6856 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6857 if (!napi->thread) {
6858 err = napi_kthread_create(napi);
6867 dev->threaded = threaded;
6869 /* Make sure kthread is created before THREADED bit
6872 smp_mb__before_atomic();
6874 /* Setting/unsetting threaded mode on a napi might not immediately
6875 * take effect, if the current napi instance is actively being
6876 * polled. In this case, the switch between threaded mode and
6877 * softirq mode will happen in the next round of napi_schedule().
6878 * This should not cause hiccups/stalls to the live traffic.
6880 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6882 set_bit(NAPI_STATE_THREADED, &napi->state);
6884 clear_bit(NAPI_STATE_THREADED, &napi->state);
6889 EXPORT_SYMBOL(dev_set_threaded);
6891 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6892 int (*poll)(struct napi_struct *, int), int weight)
6894 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6897 INIT_LIST_HEAD(&napi->poll_list);
6898 INIT_HLIST_NODE(&napi->napi_hash_node);
6899 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6900 napi->timer.function = napi_watchdog;
6901 init_gro_hash(napi);
6903 INIT_LIST_HEAD(&napi->rx_list);
6906 if (weight > NAPI_POLL_WEIGHT)
6907 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6909 napi->weight = weight;
6911 #ifdef CONFIG_NETPOLL
6912 napi->poll_owner = -1;
6914 set_bit(NAPI_STATE_SCHED, &napi->state);
6915 set_bit(NAPI_STATE_NPSVC, &napi->state);
6916 list_add_rcu(&napi->dev_list, &dev->napi_list);
6917 napi_hash_add(napi);
6918 /* Create kthread for this napi if dev->threaded is set.
6919 * Clear dev->threaded if kthread creation failed so that
6920 * threaded mode will not be enabled in napi_enable().
6922 if (dev->threaded && napi_kthread_create(napi))
6925 EXPORT_SYMBOL(netif_napi_add);
6927 void napi_disable(struct napi_struct *n)
6929 unsigned long val, new;
6932 set_bit(NAPI_STATE_DISABLE, &n->state);
6935 val = READ_ONCE(n->state);
6936 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6937 usleep_range(20, 200);
6941 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6942 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6944 if (cmpxchg(&n->state, val, new) == val)
6948 hrtimer_cancel(&n->timer);
6950 clear_bit(NAPI_STATE_DISABLE, &n->state);
6952 EXPORT_SYMBOL(napi_disable);
6955 * napi_enable - enable NAPI scheduling
6958 * Resume NAPI from being scheduled on this context.
6959 * Must be paired with napi_disable.
6961 void napi_enable(struct napi_struct *n)
6963 unsigned long val, new;
6966 val = READ_ONCE(n->state);
6967 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6969 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6970 if (n->dev->threaded && n->thread)
6971 new |= NAPIF_STATE_THREADED;
6972 } while (cmpxchg(&n->state, val, new) != val);
6974 EXPORT_SYMBOL(napi_enable);
6976 static void flush_gro_hash(struct napi_struct *napi)
6980 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6981 struct sk_buff *skb, *n;
6983 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6985 napi->gro_hash[i].count = 0;
6989 /* Must be called in process context */
6990 void __netif_napi_del(struct napi_struct *napi)
6992 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6995 napi_hash_del(napi);
6996 list_del_rcu(&napi->dev_list);
6997 napi_free_frags(napi);
6999 flush_gro_hash(napi);
7000 napi->gro_bitmask = 0;
7003 kthread_stop(napi->thread);
7004 napi->thread = NULL;
7007 EXPORT_SYMBOL(__netif_napi_del);
7009 static int __napi_poll(struct napi_struct *n, bool *repoll)
7015 /* This NAPI_STATE_SCHED test is for avoiding a race
7016 * with netpoll's poll_napi(). Only the entity which
7017 * obtains the lock and sees NAPI_STATE_SCHED set will
7018 * actually make the ->poll() call. Therefore we avoid
7019 * accidentally calling ->poll() when NAPI is not scheduled.
7022 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
7023 work = n->poll(n, weight);
7024 trace_napi_poll(n, work, weight);
7027 if (unlikely(work > weight))
7028 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7029 n->poll, work, weight);
7031 if (likely(work < weight))
7034 /* Drivers must not modify the NAPI state if they
7035 * consume the entire weight. In such cases this code
7036 * still "owns" the NAPI instance and therefore can
7037 * move the instance around on the list at-will.
7039 if (unlikely(napi_disable_pending(n))) {
7044 /* The NAPI context has more processing work, but busy-polling
7045 * is preferred. Exit early.
7047 if (napi_prefer_busy_poll(n)) {
7048 if (napi_complete_done(n, work)) {
7049 /* If timeout is not set, we need to make sure
7050 * that the NAPI is re-scheduled.
7057 if (n->gro_bitmask) {
7058 /* flush too old packets
7059 * If HZ < 1000, flush all packets.
7061 napi_gro_flush(n, HZ >= 1000);
7066 /* Some drivers may have called napi_schedule
7067 * prior to exhausting their budget.
7069 if (unlikely(!list_empty(&n->poll_list))) {
7070 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7071 n->dev ? n->dev->name : "backlog");
7080 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7082 bool do_repoll = false;
7086 list_del_init(&n->poll_list);
7088 have = netpoll_poll_lock(n);
7090 work = __napi_poll(n, &do_repoll);
7093 list_add_tail(&n->poll_list, repoll);
7095 netpoll_poll_unlock(have);
7100 static int napi_thread_wait(struct napi_struct *napi)
7104 set_current_state(TASK_INTERRUPTIBLE);
7106 while (!kthread_should_stop()) {
7107 /* Testing SCHED_THREADED bit here to make sure the current
7108 * kthread owns this napi and could poll on this napi.
7109 * Testing SCHED bit is not enough because SCHED bit might be
7110 * set by some other busy poll thread or by napi_disable().
7112 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7113 WARN_ON(!list_empty(&napi->poll_list));
7114 __set_current_state(TASK_RUNNING);
7119 /* woken being true indicates this thread owns this napi. */
7121 set_current_state(TASK_INTERRUPTIBLE);
7123 __set_current_state(TASK_RUNNING);
7128 static int napi_threaded_poll(void *data)
7130 struct napi_struct *napi = data;
7133 while (!napi_thread_wait(napi)) {
7135 bool repoll = false;
7139 have = netpoll_poll_lock(napi);
7140 __napi_poll(napi, &repoll);
7141 netpoll_poll_unlock(have);
7154 static __latent_entropy void net_rx_action(struct softirq_action *h)
7156 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7157 unsigned long time_limit = jiffies +
7158 usecs_to_jiffies(netdev_budget_usecs);
7159 int budget = netdev_budget;
7163 local_irq_disable();
7164 list_splice_init(&sd->poll_list, &list);
7168 struct napi_struct *n;
7170 if (list_empty(&list)) {
7171 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7176 n = list_first_entry(&list, struct napi_struct, poll_list);
7177 budget -= napi_poll(n, &repoll);
7179 /* If softirq window is exhausted then punt.
7180 * Allow this to run for 2 jiffies since which will allow
7181 * an average latency of 1.5/HZ.
7183 if (unlikely(budget <= 0 ||
7184 time_after_eq(jiffies, time_limit))) {
7190 local_irq_disable();
7192 list_splice_tail_init(&sd->poll_list, &list);
7193 list_splice_tail(&repoll, &list);
7194 list_splice(&list, &sd->poll_list);
7195 if (!list_empty(&sd->poll_list))
7196 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7198 net_rps_action_and_irq_enable(sd);
7201 struct netdev_adjacent {
7202 struct net_device *dev;
7204 /* upper master flag, there can only be one master device per list */
7207 /* lookup ignore flag */
7210 /* counter for the number of times this device was added to us */
7213 /* private field for the users */
7216 struct list_head list;
7217 struct rcu_head rcu;
7220 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7221 struct list_head *adj_list)
7223 struct netdev_adjacent *adj;
7225 list_for_each_entry(adj, adj_list, list) {
7226 if (adj->dev == adj_dev)
7232 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7233 struct netdev_nested_priv *priv)
7235 struct net_device *dev = (struct net_device *)priv->data;
7237 return upper_dev == dev;
7241 * netdev_has_upper_dev - Check if device is linked to an upper device
7243 * @upper_dev: upper device to check
7245 * Find out if a device is linked to specified upper device and return true
7246 * in case it is. Note that this checks only immediate upper device,
7247 * not through a complete stack of devices. The caller must hold the RTNL lock.
7249 bool netdev_has_upper_dev(struct net_device *dev,
7250 struct net_device *upper_dev)
7252 struct netdev_nested_priv priv = {
7253 .data = (void *)upper_dev,
7258 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7261 EXPORT_SYMBOL(netdev_has_upper_dev);
7264 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7266 * @upper_dev: upper device to check
7268 * Find out if a device is linked to specified upper device and return true
7269 * in case it is. Note that this checks the entire upper device chain.
7270 * The caller must hold rcu lock.
7273 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7274 struct net_device *upper_dev)
7276 struct netdev_nested_priv priv = {
7277 .data = (void *)upper_dev,
7280 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7283 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7286 * netdev_has_any_upper_dev - Check if device is linked to some device
7289 * Find out if a device is linked to an upper device and return true in case
7290 * it is. The caller must hold the RTNL lock.
7292 bool netdev_has_any_upper_dev(struct net_device *dev)
7296 return !list_empty(&dev->adj_list.upper);
7298 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7301 * netdev_master_upper_dev_get - Get master upper device
7304 * Find a master upper device and return pointer to it or NULL in case
7305 * it's not there. The caller must hold the RTNL lock.
7307 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7309 struct netdev_adjacent *upper;
7313 if (list_empty(&dev->adj_list.upper))
7316 upper = list_first_entry(&dev->adj_list.upper,
7317 struct netdev_adjacent, list);
7318 if (likely(upper->master))
7322 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7324 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7326 struct netdev_adjacent *upper;
7330 if (list_empty(&dev->adj_list.upper))
7333 upper = list_first_entry(&dev->adj_list.upper,
7334 struct netdev_adjacent, list);
7335 if (likely(upper->master) && !upper->ignore)
7341 * netdev_has_any_lower_dev - Check if device is linked to some device
7344 * Find out if a device is linked to a lower device and return true in case
7345 * it is. The caller must hold the RTNL lock.
7347 static bool netdev_has_any_lower_dev(struct net_device *dev)
7351 return !list_empty(&dev->adj_list.lower);
7354 void *netdev_adjacent_get_private(struct list_head *adj_list)
7356 struct netdev_adjacent *adj;
7358 adj = list_entry(adj_list, struct netdev_adjacent, list);
7360 return adj->private;
7362 EXPORT_SYMBOL(netdev_adjacent_get_private);
7365 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7367 * @iter: list_head ** of the current position
7369 * Gets the next device from the dev's upper list, starting from iter
7370 * position. The caller must hold RCU read lock.
7372 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7373 struct list_head **iter)
7375 struct netdev_adjacent *upper;
7377 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7379 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7381 if (&upper->list == &dev->adj_list.upper)
7384 *iter = &upper->list;
7388 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7390 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7391 struct list_head **iter,
7394 struct netdev_adjacent *upper;
7396 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7398 if (&upper->list == &dev->adj_list.upper)
7401 *iter = &upper->list;
7402 *ignore = upper->ignore;
7407 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7408 struct list_head **iter)
7410 struct netdev_adjacent *upper;
7412 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7414 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7416 if (&upper->list == &dev->adj_list.upper)
7419 *iter = &upper->list;
7424 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7425 int (*fn)(struct net_device *dev,
7426 struct netdev_nested_priv *priv),
7427 struct netdev_nested_priv *priv)
7429 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7430 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7435 iter = &dev->adj_list.upper;
7439 ret = fn(now, priv);
7446 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7453 niter = &udev->adj_list.upper;
7454 dev_stack[cur] = now;
7455 iter_stack[cur++] = iter;
7462 next = dev_stack[--cur];
7463 niter = iter_stack[cur];
7473 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7474 int (*fn)(struct net_device *dev,
7475 struct netdev_nested_priv *priv),
7476 struct netdev_nested_priv *priv)
7478 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7479 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7483 iter = &dev->adj_list.upper;
7487 ret = fn(now, priv);
7494 udev = netdev_next_upper_dev_rcu(now, &iter);
7499 niter = &udev->adj_list.upper;
7500 dev_stack[cur] = now;
7501 iter_stack[cur++] = iter;
7508 next = dev_stack[--cur];
7509 niter = iter_stack[cur];
7518 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7520 static bool __netdev_has_upper_dev(struct net_device *dev,
7521 struct net_device *upper_dev)
7523 struct netdev_nested_priv priv = {
7525 .data = (void *)upper_dev,
7530 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7535 * netdev_lower_get_next_private - Get the next ->private from the
7536 * lower neighbour list
7538 * @iter: list_head ** of the current position
7540 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7541 * list, starting from iter position. The caller must hold either hold the
7542 * RTNL lock or its own locking that guarantees that the neighbour lower
7543 * list will remain unchanged.
7545 void *netdev_lower_get_next_private(struct net_device *dev,
7546 struct list_head **iter)
7548 struct netdev_adjacent *lower;
7550 lower = list_entry(*iter, struct netdev_adjacent, list);
7552 if (&lower->list == &dev->adj_list.lower)
7555 *iter = lower->list.next;
7557 return lower->private;
7559 EXPORT_SYMBOL(netdev_lower_get_next_private);
7562 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7563 * lower neighbour list, RCU
7566 * @iter: list_head ** of the current position
7568 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7569 * list, starting from iter position. The caller must hold RCU read lock.
7571 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7572 struct list_head **iter)
7574 struct netdev_adjacent *lower;
7576 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7578 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7580 if (&lower->list == &dev->adj_list.lower)
7583 *iter = &lower->list;
7585 return lower->private;
7587 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7590 * netdev_lower_get_next - Get the next device from the lower neighbour
7593 * @iter: list_head ** of the current position
7595 * Gets the next netdev_adjacent from the dev's lower neighbour
7596 * list, starting from iter position. The caller must hold RTNL lock or
7597 * its own locking that guarantees that the neighbour lower
7598 * list will remain unchanged.
7600 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7602 struct netdev_adjacent *lower;
7604 lower = list_entry(*iter, struct netdev_adjacent, list);
7606 if (&lower->list == &dev->adj_list.lower)
7609 *iter = lower->list.next;
7613 EXPORT_SYMBOL(netdev_lower_get_next);
7615 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7616 struct list_head **iter)
7618 struct netdev_adjacent *lower;
7620 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7622 if (&lower->list == &dev->adj_list.lower)
7625 *iter = &lower->list;
7630 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7631 struct list_head **iter,
7634 struct netdev_adjacent *lower;
7636 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7638 if (&lower->list == &dev->adj_list.lower)
7641 *iter = &lower->list;
7642 *ignore = lower->ignore;
7647 int netdev_walk_all_lower_dev(struct net_device *dev,
7648 int (*fn)(struct net_device *dev,
7649 struct netdev_nested_priv *priv),
7650 struct netdev_nested_priv *priv)
7652 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7653 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7657 iter = &dev->adj_list.lower;
7661 ret = fn(now, priv);
7668 ldev = netdev_next_lower_dev(now, &iter);
7673 niter = &ldev->adj_list.lower;
7674 dev_stack[cur] = now;
7675 iter_stack[cur++] = iter;
7682 next = dev_stack[--cur];
7683 niter = iter_stack[cur];
7692 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7694 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7695 int (*fn)(struct net_device *dev,
7696 struct netdev_nested_priv *priv),
7697 struct netdev_nested_priv *priv)
7699 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7700 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7705 iter = &dev->adj_list.lower;
7709 ret = fn(now, priv);
7716 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7723 niter = &ldev->adj_list.lower;
7724 dev_stack[cur] = now;
7725 iter_stack[cur++] = iter;
7732 next = dev_stack[--cur];
7733 niter = iter_stack[cur];
7743 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7744 struct list_head **iter)
7746 struct netdev_adjacent *lower;
7748 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7749 if (&lower->list == &dev->adj_list.lower)
7752 *iter = &lower->list;
7756 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7758 static u8 __netdev_upper_depth(struct net_device *dev)
7760 struct net_device *udev;
7761 struct list_head *iter;
7765 for (iter = &dev->adj_list.upper,
7766 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7768 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7771 if (max_depth < udev->upper_level)
7772 max_depth = udev->upper_level;
7778 static u8 __netdev_lower_depth(struct net_device *dev)
7780 struct net_device *ldev;
7781 struct list_head *iter;
7785 for (iter = &dev->adj_list.lower,
7786 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7788 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7791 if (max_depth < ldev->lower_level)
7792 max_depth = ldev->lower_level;
7798 static int __netdev_update_upper_level(struct net_device *dev,
7799 struct netdev_nested_priv *__unused)
7801 dev->upper_level = __netdev_upper_depth(dev) + 1;
7805 static int __netdev_update_lower_level(struct net_device *dev,
7806 struct netdev_nested_priv *priv)
7808 dev->lower_level = __netdev_lower_depth(dev) + 1;
7810 #ifdef CONFIG_LOCKDEP
7814 if (priv->flags & NESTED_SYNC_IMM)
7815 dev->nested_level = dev->lower_level - 1;
7816 if (priv->flags & NESTED_SYNC_TODO)
7817 net_unlink_todo(dev);
7822 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7823 int (*fn)(struct net_device *dev,
7824 struct netdev_nested_priv *priv),
7825 struct netdev_nested_priv *priv)
7827 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7828 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7832 iter = &dev->adj_list.lower;
7836 ret = fn(now, priv);
7843 ldev = netdev_next_lower_dev_rcu(now, &iter);
7848 niter = &ldev->adj_list.lower;
7849 dev_stack[cur] = now;
7850 iter_stack[cur++] = iter;
7857 next = dev_stack[--cur];
7858 niter = iter_stack[cur];
7867 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7870 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7871 * lower neighbour list, RCU
7875 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7876 * list. The caller must hold RCU read lock.
7878 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7880 struct netdev_adjacent *lower;
7882 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7883 struct netdev_adjacent, list);
7885 return lower->private;
7888 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7891 * netdev_master_upper_dev_get_rcu - Get master upper device
7894 * Find a master upper device and return pointer to it or NULL in case
7895 * it's not there. The caller must hold the RCU read lock.
7897 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7899 struct netdev_adjacent *upper;
7901 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7902 struct netdev_adjacent, list);
7903 if (upper && likely(upper->master))
7907 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7909 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7910 struct net_device *adj_dev,
7911 struct list_head *dev_list)
7913 char linkname[IFNAMSIZ+7];
7915 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7916 "upper_%s" : "lower_%s", adj_dev->name);
7917 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7920 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7922 struct list_head *dev_list)
7924 char linkname[IFNAMSIZ+7];
7926 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7927 "upper_%s" : "lower_%s", name);
7928 sysfs_remove_link(&(dev->dev.kobj), linkname);
7931 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7932 struct net_device *adj_dev,
7933 struct list_head *dev_list)
7935 return (dev_list == &dev->adj_list.upper ||
7936 dev_list == &dev->adj_list.lower) &&
7937 net_eq(dev_net(dev), dev_net(adj_dev));
7940 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7941 struct net_device *adj_dev,
7942 struct list_head *dev_list,
7943 void *private, bool master)
7945 struct netdev_adjacent *adj;
7948 adj = __netdev_find_adj(adj_dev, dev_list);
7952 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7953 dev->name, adj_dev->name, adj->ref_nr);
7958 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7963 adj->master = master;
7965 adj->private = private;
7966 adj->ignore = false;
7969 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7970 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7972 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7973 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7978 /* Ensure that master link is always the first item in list. */
7980 ret = sysfs_create_link(&(dev->dev.kobj),
7981 &(adj_dev->dev.kobj), "master");
7983 goto remove_symlinks;
7985 list_add_rcu(&adj->list, dev_list);
7987 list_add_tail_rcu(&adj->list, dev_list);
7993 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7994 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8002 static void __netdev_adjacent_dev_remove(struct net_device *dev,
8003 struct net_device *adj_dev,
8005 struct list_head *dev_list)
8007 struct netdev_adjacent *adj;
8009 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
8010 dev->name, adj_dev->name, ref_nr);
8012 adj = __netdev_find_adj(adj_dev, dev_list);
8015 pr_err("Adjacency does not exist for device %s from %s\n",
8016 dev->name, adj_dev->name);
8021 if (adj->ref_nr > ref_nr) {
8022 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8023 dev->name, adj_dev->name, ref_nr,
8024 adj->ref_nr - ref_nr);
8025 adj->ref_nr -= ref_nr;
8030 sysfs_remove_link(&(dev->dev.kobj), "master");
8032 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8033 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8035 list_del_rcu(&adj->list);
8036 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8037 adj_dev->name, dev->name, adj_dev->name);
8039 kfree_rcu(adj, rcu);
8042 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8043 struct net_device *upper_dev,
8044 struct list_head *up_list,
8045 struct list_head *down_list,
8046 void *private, bool master)
8050 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8055 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8058 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8065 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8066 struct net_device *upper_dev,
8068 struct list_head *up_list,
8069 struct list_head *down_list)
8071 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8072 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8075 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8076 struct net_device *upper_dev,
8077 void *private, bool master)
8079 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8080 &dev->adj_list.upper,
8081 &upper_dev->adj_list.lower,
8085 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8086 struct net_device *upper_dev)
8088 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8089 &dev->adj_list.upper,
8090 &upper_dev->adj_list.lower);
8093 static int __netdev_upper_dev_link(struct net_device *dev,
8094 struct net_device *upper_dev, bool master,
8095 void *upper_priv, void *upper_info,
8096 struct netdev_nested_priv *priv,
8097 struct netlink_ext_ack *extack)
8099 struct netdev_notifier_changeupper_info changeupper_info = {
8104 .upper_dev = upper_dev,
8107 .upper_info = upper_info,
8109 struct net_device *master_dev;
8114 if (dev == upper_dev)
8117 /* To prevent loops, check if dev is not upper device to upper_dev. */
8118 if (__netdev_has_upper_dev(upper_dev, dev))
8121 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8125 if (__netdev_has_upper_dev(dev, upper_dev))
8128 master_dev = __netdev_master_upper_dev_get(dev);
8130 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8133 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8134 &changeupper_info.info);
8135 ret = notifier_to_errno(ret);
8139 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8144 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8145 &changeupper_info.info);
8146 ret = notifier_to_errno(ret);
8150 __netdev_update_upper_level(dev, NULL);
8151 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8153 __netdev_update_lower_level(upper_dev, priv);
8154 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8160 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8166 * netdev_upper_dev_link - Add a link to the upper device
8168 * @upper_dev: new upper device
8169 * @extack: netlink extended ack
8171 * Adds a link to device which is upper to this one. The caller must hold
8172 * the RTNL lock. On a failure a negative errno code is returned.
8173 * On success the reference counts are adjusted and the function
8176 int netdev_upper_dev_link(struct net_device *dev,
8177 struct net_device *upper_dev,
8178 struct netlink_ext_ack *extack)
8180 struct netdev_nested_priv priv = {
8181 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8185 return __netdev_upper_dev_link(dev, upper_dev, false,
8186 NULL, NULL, &priv, extack);
8188 EXPORT_SYMBOL(netdev_upper_dev_link);
8191 * netdev_master_upper_dev_link - Add a master link to the upper device
8193 * @upper_dev: new upper device
8194 * @upper_priv: upper device private
8195 * @upper_info: upper info to be passed down via notifier
8196 * @extack: netlink extended ack
8198 * Adds a link to device which is upper to this one. In this case, only
8199 * one master upper device can be linked, although other non-master devices
8200 * might be linked as well. The caller must hold the RTNL lock.
8201 * On a failure a negative errno code is returned. On success the reference
8202 * counts are adjusted and the function returns zero.
8204 int netdev_master_upper_dev_link(struct net_device *dev,
8205 struct net_device *upper_dev,
8206 void *upper_priv, void *upper_info,
8207 struct netlink_ext_ack *extack)
8209 struct netdev_nested_priv priv = {
8210 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8214 return __netdev_upper_dev_link(dev, upper_dev, true,
8215 upper_priv, upper_info, &priv, extack);
8217 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8219 static void __netdev_upper_dev_unlink(struct net_device *dev,
8220 struct net_device *upper_dev,
8221 struct netdev_nested_priv *priv)
8223 struct netdev_notifier_changeupper_info changeupper_info = {
8227 .upper_dev = upper_dev,
8233 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8235 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8236 &changeupper_info.info);
8238 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8240 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8241 &changeupper_info.info);
8243 __netdev_update_upper_level(dev, NULL);
8244 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8246 __netdev_update_lower_level(upper_dev, priv);
8247 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8252 * netdev_upper_dev_unlink - Removes a link to upper device
8254 * @upper_dev: new upper device
8256 * Removes a link to device which is upper to this one. The caller must hold
8259 void netdev_upper_dev_unlink(struct net_device *dev,
8260 struct net_device *upper_dev)
8262 struct netdev_nested_priv priv = {
8263 .flags = NESTED_SYNC_TODO,
8267 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8269 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8271 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8272 struct net_device *lower_dev,
8275 struct netdev_adjacent *adj;
8277 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8281 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8286 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8287 struct net_device *lower_dev)
8289 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8292 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8293 struct net_device *lower_dev)
8295 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8298 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8299 struct net_device *new_dev,
8300 struct net_device *dev,
8301 struct netlink_ext_ack *extack)
8303 struct netdev_nested_priv priv = {
8312 if (old_dev && new_dev != old_dev)
8313 netdev_adjacent_dev_disable(dev, old_dev);
8314 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8317 if (old_dev && new_dev != old_dev)
8318 netdev_adjacent_dev_enable(dev, old_dev);
8324 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8326 void netdev_adjacent_change_commit(struct net_device *old_dev,
8327 struct net_device *new_dev,
8328 struct net_device *dev)
8330 struct netdev_nested_priv priv = {
8331 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8335 if (!new_dev || !old_dev)
8338 if (new_dev == old_dev)
8341 netdev_adjacent_dev_enable(dev, old_dev);
8342 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8344 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8346 void netdev_adjacent_change_abort(struct net_device *old_dev,
8347 struct net_device *new_dev,
8348 struct net_device *dev)
8350 struct netdev_nested_priv priv = {
8358 if (old_dev && new_dev != old_dev)
8359 netdev_adjacent_dev_enable(dev, old_dev);
8361 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8363 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8366 * netdev_bonding_info_change - Dispatch event about slave change
8368 * @bonding_info: info to dispatch
8370 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8371 * The caller must hold the RTNL lock.
8373 void netdev_bonding_info_change(struct net_device *dev,
8374 struct netdev_bonding_info *bonding_info)
8376 struct netdev_notifier_bonding_info info = {
8380 memcpy(&info.bonding_info, bonding_info,
8381 sizeof(struct netdev_bonding_info));
8382 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8385 EXPORT_SYMBOL(netdev_bonding_info_change);
8388 * netdev_get_xmit_slave - Get the xmit slave of master device
8391 * @all_slaves: assume all the slaves are active
8393 * The reference counters are not incremented so the caller must be
8394 * careful with locks. The caller must hold RCU lock.
8395 * %NULL is returned if no slave is found.
8398 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8399 struct sk_buff *skb,
8402 const struct net_device_ops *ops = dev->netdev_ops;
8404 if (!ops->ndo_get_xmit_slave)
8406 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8408 EXPORT_SYMBOL(netdev_get_xmit_slave);
8410 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8413 const struct net_device_ops *ops = dev->netdev_ops;
8415 if (!ops->ndo_sk_get_lower_dev)
8417 return ops->ndo_sk_get_lower_dev(dev, sk);
8421 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8425 * %NULL is returned if no lower device is found.
8428 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8431 struct net_device *lower;
8433 lower = netdev_sk_get_lower_dev(dev, sk);
8436 lower = netdev_sk_get_lower_dev(dev, sk);
8441 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8443 static void netdev_adjacent_add_links(struct net_device *dev)
8445 struct netdev_adjacent *iter;
8447 struct net *net = dev_net(dev);
8449 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8450 if (!net_eq(net, dev_net(iter->dev)))
8452 netdev_adjacent_sysfs_add(iter->dev, dev,
8453 &iter->dev->adj_list.lower);
8454 netdev_adjacent_sysfs_add(dev, iter->dev,
8455 &dev->adj_list.upper);
8458 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8459 if (!net_eq(net, dev_net(iter->dev)))
8461 netdev_adjacent_sysfs_add(iter->dev, dev,
8462 &iter->dev->adj_list.upper);
8463 netdev_adjacent_sysfs_add(dev, iter->dev,
8464 &dev->adj_list.lower);
8468 static void netdev_adjacent_del_links(struct net_device *dev)
8470 struct netdev_adjacent *iter;
8472 struct net *net = dev_net(dev);
8474 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8475 if (!net_eq(net, dev_net(iter->dev)))
8477 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8478 &iter->dev->adj_list.lower);
8479 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8480 &dev->adj_list.upper);
8483 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8484 if (!net_eq(net, dev_net(iter->dev)))
8486 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8487 &iter->dev->adj_list.upper);
8488 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8489 &dev->adj_list.lower);
8493 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8495 struct netdev_adjacent *iter;
8497 struct net *net = dev_net(dev);
8499 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8500 if (!net_eq(net, dev_net(iter->dev)))
8502 netdev_adjacent_sysfs_del(iter->dev, oldname,
8503 &iter->dev->adj_list.lower);
8504 netdev_adjacent_sysfs_add(iter->dev, dev,
8505 &iter->dev->adj_list.lower);
8508 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8509 if (!net_eq(net, dev_net(iter->dev)))
8511 netdev_adjacent_sysfs_del(iter->dev, oldname,
8512 &iter->dev->adj_list.upper);
8513 netdev_adjacent_sysfs_add(iter->dev, dev,
8514 &iter->dev->adj_list.upper);
8518 void *netdev_lower_dev_get_private(struct net_device *dev,
8519 struct net_device *lower_dev)
8521 struct netdev_adjacent *lower;
8525 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8529 return lower->private;
8531 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8535 * netdev_lower_state_changed - Dispatch event about lower device state change
8536 * @lower_dev: device
8537 * @lower_state_info: state to dispatch
8539 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8540 * The caller must hold the RTNL lock.
8542 void netdev_lower_state_changed(struct net_device *lower_dev,
8543 void *lower_state_info)
8545 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8546 .info.dev = lower_dev,
8550 changelowerstate_info.lower_state_info = lower_state_info;
8551 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8552 &changelowerstate_info.info);
8554 EXPORT_SYMBOL(netdev_lower_state_changed);
8556 static void dev_change_rx_flags(struct net_device *dev, int flags)
8558 const struct net_device_ops *ops = dev->netdev_ops;
8560 if (ops->ndo_change_rx_flags)
8561 ops->ndo_change_rx_flags(dev, flags);
8564 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8566 unsigned int old_flags = dev->flags;
8572 dev->flags |= IFF_PROMISC;
8573 dev->promiscuity += inc;
8574 if (dev->promiscuity == 0) {
8577 * If inc causes overflow, untouch promisc and return error.
8580 dev->flags &= ~IFF_PROMISC;
8582 dev->promiscuity -= inc;
8583 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8587 if (dev->flags != old_flags) {
8588 pr_info("device %s %s promiscuous mode\n",
8590 dev->flags & IFF_PROMISC ? "entered" : "left");
8591 if (audit_enabled) {
8592 current_uid_gid(&uid, &gid);
8593 audit_log(audit_context(), GFP_ATOMIC,
8594 AUDIT_ANOM_PROMISCUOUS,
8595 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8596 dev->name, (dev->flags & IFF_PROMISC),
8597 (old_flags & IFF_PROMISC),
8598 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8599 from_kuid(&init_user_ns, uid),
8600 from_kgid(&init_user_ns, gid),
8601 audit_get_sessionid(current));
8604 dev_change_rx_flags(dev, IFF_PROMISC);
8607 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8612 * dev_set_promiscuity - update promiscuity count on a device
8616 * Add or remove promiscuity from a device. While the count in the device
8617 * remains above zero the interface remains promiscuous. Once it hits zero
8618 * the device reverts back to normal filtering operation. A negative inc
8619 * value is used to drop promiscuity on the device.
8620 * Return 0 if successful or a negative errno code on error.
8622 int dev_set_promiscuity(struct net_device *dev, int inc)
8624 unsigned int old_flags = dev->flags;
8627 err = __dev_set_promiscuity(dev, inc, true);
8630 if (dev->flags != old_flags)
8631 dev_set_rx_mode(dev);
8634 EXPORT_SYMBOL(dev_set_promiscuity);
8636 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8638 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8642 dev->flags |= IFF_ALLMULTI;
8643 dev->allmulti += inc;
8644 if (dev->allmulti == 0) {
8647 * If inc causes overflow, untouch allmulti and return error.
8650 dev->flags &= ~IFF_ALLMULTI;
8652 dev->allmulti -= inc;
8653 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8657 if (dev->flags ^ old_flags) {
8658 dev_change_rx_flags(dev, IFF_ALLMULTI);
8659 dev_set_rx_mode(dev);
8661 __dev_notify_flags(dev, old_flags,
8662 dev->gflags ^ old_gflags);
8668 * dev_set_allmulti - update allmulti count on a device
8672 * Add or remove reception of all multicast frames to a device. While the
8673 * count in the device remains above zero the interface remains listening
8674 * to all interfaces. Once it hits zero the device reverts back to normal
8675 * filtering operation. A negative @inc value is used to drop the counter
8676 * when releasing a resource needing all multicasts.
8677 * Return 0 if successful or a negative errno code on error.
8680 int dev_set_allmulti(struct net_device *dev, int inc)
8682 return __dev_set_allmulti(dev, inc, true);
8684 EXPORT_SYMBOL(dev_set_allmulti);
8687 * Upload unicast and multicast address lists to device and
8688 * configure RX filtering. When the device doesn't support unicast
8689 * filtering it is put in promiscuous mode while unicast addresses
8692 void __dev_set_rx_mode(struct net_device *dev)
8694 const struct net_device_ops *ops = dev->netdev_ops;
8696 /* dev_open will call this function so the list will stay sane. */
8697 if (!(dev->flags&IFF_UP))
8700 if (!netif_device_present(dev))
8703 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8704 /* Unicast addresses changes may only happen under the rtnl,
8705 * therefore calling __dev_set_promiscuity here is safe.
8707 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8708 __dev_set_promiscuity(dev, 1, false);
8709 dev->uc_promisc = true;
8710 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8711 __dev_set_promiscuity(dev, -1, false);
8712 dev->uc_promisc = false;
8716 if (ops->ndo_set_rx_mode)
8717 ops->ndo_set_rx_mode(dev);
8720 void dev_set_rx_mode(struct net_device *dev)
8722 netif_addr_lock_bh(dev);
8723 __dev_set_rx_mode(dev);
8724 netif_addr_unlock_bh(dev);
8728 * dev_get_flags - get flags reported to userspace
8731 * Get the combination of flag bits exported through APIs to userspace.
8733 unsigned int dev_get_flags(const struct net_device *dev)
8737 flags = (dev->flags & ~(IFF_PROMISC |
8742 (dev->gflags & (IFF_PROMISC |
8745 if (netif_running(dev)) {
8746 if (netif_oper_up(dev))
8747 flags |= IFF_RUNNING;
8748 if (netif_carrier_ok(dev))
8749 flags |= IFF_LOWER_UP;
8750 if (netif_dormant(dev))
8751 flags |= IFF_DORMANT;
8756 EXPORT_SYMBOL(dev_get_flags);
8758 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8759 struct netlink_ext_ack *extack)
8761 unsigned int old_flags = dev->flags;
8767 * Set the flags on our device.
8770 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8771 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8773 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8777 * Load in the correct multicast list now the flags have changed.
8780 if ((old_flags ^ flags) & IFF_MULTICAST)
8781 dev_change_rx_flags(dev, IFF_MULTICAST);
8783 dev_set_rx_mode(dev);
8786 * Have we downed the interface. We handle IFF_UP ourselves
8787 * according to user attempts to set it, rather than blindly
8792 if ((old_flags ^ flags) & IFF_UP) {
8793 if (old_flags & IFF_UP)
8796 ret = __dev_open(dev, extack);
8799 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8800 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8801 unsigned int old_flags = dev->flags;
8803 dev->gflags ^= IFF_PROMISC;
8805 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8806 if (dev->flags != old_flags)
8807 dev_set_rx_mode(dev);
8810 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8811 * is important. Some (broken) drivers set IFF_PROMISC, when
8812 * IFF_ALLMULTI is requested not asking us and not reporting.
8814 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8815 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8817 dev->gflags ^= IFF_ALLMULTI;
8818 __dev_set_allmulti(dev, inc, false);
8824 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8825 unsigned int gchanges)
8827 unsigned int changes = dev->flags ^ old_flags;
8830 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8832 if (changes & IFF_UP) {
8833 if (dev->flags & IFF_UP)
8834 call_netdevice_notifiers(NETDEV_UP, dev);
8836 call_netdevice_notifiers(NETDEV_DOWN, dev);
8839 if (dev->flags & IFF_UP &&
8840 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8841 struct netdev_notifier_change_info change_info = {
8845 .flags_changed = changes,
8848 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8853 * dev_change_flags - change device settings
8855 * @flags: device state flags
8856 * @extack: netlink extended ack
8858 * Change settings on device based state flags. The flags are
8859 * in the userspace exported format.
8861 int dev_change_flags(struct net_device *dev, unsigned int flags,
8862 struct netlink_ext_ack *extack)
8865 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8867 ret = __dev_change_flags(dev, flags, extack);
8871 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8872 __dev_notify_flags(dev, old_flags, changes);
8875 EXPORT_SYMBOL(dev_change_flags);
8877 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8879 const struct net_device_ops *ops = dev->netdev_ops;
8881 if (ops->ndo_change_mtu)
8882 return ops->ndo_change_mtu(dev, new_mtu);
8884 /* Pairs with all the lockless reads of dev->mtu in the stack */
8885 WRITE_ONCE(dev->mtu, new_mtu);
8888 EXPORT_SYMBOL(__dev_set_mtu);
8890 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8891 struct netlink_ext_ack *extack)
8893 /* MTU must be positive, and in range */
8894 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8895 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8899 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8900 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8907 * dev_set_mtu_ext - Change maximum transfer unit
8909 * @new_mtu: new transfer unit
8910 * @extack: netlink extended ack
8912 * Change the maximum transfer size of the network device.
8914 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8915 struct netlink_ext_ack *extack)
8919 if (new_mtu == dev->mtu)
8922 err = dev_validate_mtu(dev, new_mtu, extack);
8926 if (!netif_device_present(dev))
8929 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8930 err = notifier_to_errno(err);
8934 orig_mtu = dev->mtu;
8935 err = __dev_set_mtu(dev, new_mtu);
8938 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8940 err = notifier_to_errno(err);
8942 /* setting mtu back and notifying everyone again,
8943 * so that they have a chance to revert changes.
8945 __dev_set_mtu(dev, orig_mtu);
8946 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8953 int dev_set_mtu(struct net_device *dev, int new_mtu)
8955 struct netlink_ext_ack extack;
8958 memset(&extack, 0, sizeof(extack));
8959 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8960 if (err && extack._msg)
8961 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8964 EXPORT_SYMBOL(dev_set_mtu);
8967 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8969 * @new_len: new tx queue length
8971 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8973 unsigned int orig_len = dev->tx_queue_len;
8976 if (new_len != (unsigned int)new_len)
8979 if (new_len != orig_len) {
8980 dev->tx_queue_len = new_len;
8981 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8982 res = notifier_to_errno(res);
8985 res = dev_qdisc_change_tx_queue_len(dev);
8993 netdev_err(dev, "refused to change device tx_queue_len\n");
8994 dev->tx_queue_len = orig_len;
8999 * dev_set_group - Change group this device belongs to
9001 * @new_group: group this device should belong to
9003 void dev_set_group(struct net_device *dev, int new_group)
9005 dev->group = new_group;
9007 EXPORT_SYMBOL(dev_set_group);
9010 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9012 * @addr: new address
9013 * @extack: netlink extended ack
9015 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9016 struct netlink_ext_ack *extack)
9018 struct netdev_notifier_pre_changeaddr_info info = {
9020 .info.extack = extack,
9025 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9026 return notifier_to_errno(rc);
9028 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9031 * dev_set_mac_address - Change Media Access Control Address
9034 * @extack: netlink extended ack
9036 * Change the hardware (MAC) address of the device
9038 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9039 struct netlink_ext_ack *extack)
9041 const struct net_device_ops *ops = dev->netdev_ops;
9044 if (!ops->ndo_set_mac_address)
9046 if (sa->sa_family != dev->type)
9048 if (!netif_device_present(dev))
9050 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9053 err = ops->ndo_set_mac_address(dev, sa);
9056 dev->addr_assign_type = NET_ADDR_SET;
9057 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9058 add_device_randomness(dev->dev_addr, dev->addr_len);
9061 EXPORT_SYMBOL(dev_set_mac_address);
9063 static DECLARE_RWSEM(dev_addr_sem);
9065 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9066 struct netlink_ext_ack *extack)
9070 down_write(&dev_addr_sem);
9071 ret = dev_set_mac_address(dev, sa, extack);
9072 up_write(&dev_addr_sem);
9075 EXPORT_SYMBOL(dev_set_mac_address_user);
9077 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9079 size_t size = sizeof(sa->sa_data);
9080 struct net_device *dev;
9083 down_read(&dev_addr_sem);
9086 dev = dev_get_by_name_rcu(net, dev_name);
9092 memset(sa->sa_data, 0, size);
9094 memcpy(sa->sa_data, dev->dev_addr,
9095 min_t(size_t, size, dev->addr_len));
9096 sa->sa_family = dev->type;
9100 up_read(&dev_addr_sem);
9103 EXPORT_SYMBOL(dev_get_mac_address);
9106 * dev_change_carrier - Change device carrier
9108 * @new_carrier: new value
9110 * Change device carrier
9112 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9114 const struct net_device_ops *ops = dev->netdev_ops;
9116 if (!ops->ndo_change_carrier)
9118 if (!netif_device_present(dev))
9120 return ops->ndo_change_carrier(dev, new_carrier);
9122 EXPORT_SYMBOL(dev_change_carrier);
9125 * dev_get_phys_port_id - Get device physical port ID
9129 * Get device physical port ID
9131 int dev_get_phys_port_id(struct net_device *dev,
9132 struct netdev_phys_item_id *ppid)
9134 const struct net_device_ops *ops = dev->netdev_ops;
9136 if (!ops->ndo_get_phys_port_id)
9138 return ops->ndo_get_phys_port_id(dev, ppid);
9140 EXPORT_SYMBOL(dev_get_phys_port_id);
9143 * dev_get_phys_port_name - Get device physical port name
9146 * @len: limit of bytes to copy to name
9148 * Get device physical port name
9150 int dev_get_phys_port_name(struct net_device *dev,
9151 char *name, size_t len)
9153 const struct net_device_ops *ops = dev->netdev_ops;
9156 if (ops->ndo_get_phys_port_name) {
9157 err = ops->ndo_get_phys_port_name(dev, name, len);
9158 if (err != -EOPNOTSUPP)
9161 return devlink_compat_phys_port_name_get(dev, name, len);
9163 EXPORT_SYMBOL(dev_get_phys_port_name);
9166 * dev_get_port_parent_id - Get the device's port parent identifier
9167 * @dev: network device
9168 * @ppid: pointer to a storage for the port's parent identifier
9169 * @recurse: allow/disallow recursion to lower devices
9171 * Get the devices's port parent identifier
9173 int dev_get_port_parent_id(struct net_device *dev,
9174 struct netdev_phys_item_id *ppid,
9177 const struct net_device_ops *ops = dev->netdev_ops;
9178 struct netdev_phys_item_id first = { };
9179 struct net_device *lower_dev;
9180 struct list_head *iter;
9183 if (ops->ndo_get_port_parent_id) {
9184 err = ops->ndo_get_port_parent_id(dev, ppid);
9185 if (err != -EOPNOTSUPP)
9189 err = devlink_compat_switch_id_get(dev, ppid);
9190 if (!recurse || err != -EOPNOTSUPP)
9193 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9194 err = dev_get_port_parent_id(lower_dev, ppid, true);
9199 else if (memcmp(&first, ppid, sizeof(*ppid)))
9205 EXPORT_SYMBOL(dev_get_port_parent_id);
9208 * netdev_port_same_parent_id - Indicate if two network devices have
9209 * the same port parent identifier
9210 * @a: first network device
9211 * @b: second network device
9213 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9215 struct netdev_phys_item_id a_id = { };
9216 struct netdev_phys_item_id b_id = { };
9218 if (dev_get_port_parent_id(a, &a_id, true) ||
9219 dev_get_port_parent_id(b, &b_id, true))
9222 return netdev_phys_item_id_same(&a_id, &b_id);
9224 EXPORT_SYMBOL(netdev_port_same_parent_id);
9227 * dev_change_proto_down - update protocol port state information
9229 * @proto_down: new value
9231 * This info can be used by switch drivers to set the phys state of the
9234 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9236 const struct net_device_ops *ops = dev->netdev_ops;
9238 if (!ops->ndo_change_proto_down)
9240 if (!netif_device_present(dev))
9242 return ops->ndo_change_proto_down(dev, proto_down);
9244 EXPORT_SYMBOL(dev_change_proto_down);
9247 * dev_change_proto_down_generic - generic implementation for
9248 * ndo_change_proto_down that sets carrier according to
9252 * @proto_down: new value
9254 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9257 netif_carrier_off(dev);
9259 netif_carrier_on(dev);
9260 dev->proto_down = proto_down;
9263 EXPORT_SYMBOL(dev_change_proto_down_generic);
9266 * dev_change_proto_down_reason - proto down reason
9269 * @mask: proto down mask
9270 * @value: proto down value
9272 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9278 dev->proto_down_reason = value;
9280 for_each_set_bit(b, &mask, 32) {
9281 if (value & (1 << b))
9282 dev->proto_down_reason |= BIT(b);
9284 dev->proto_down_reason &= ~BIT(b);
9288 EXPORT_SYMBOL(dev_change_proto_down_reason);
9290 struct bpf_xdp_link {
9291 struct bpf_link link;
9292 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9296 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9298 if (flags & XDP_FLAGS_HW_MODE)
9300 if (flags & XDP_FLAGS_DRV_MODE)
9301 return XDP_MODE_DRV;
9302 if (flags & XDP_FLAGS_SKB_MODE)
9303 return XDP_MODE_SKB;
9304 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9307 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9311 return generic_xdp_install;
9314 return dev->netdev_ops->ndo_bpf;
9320 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9321 enum bpf_xdp_mode mode)
9323 return dev->xdp_state[mode].link;
9326 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9327 enum bpf_xdp_mode mode)
9329 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9332 return link->link.prog;
9333 return dev->xdp_state[mode].prog;
9336 u8 dev_xdp_prog_count(struct net_device *dev)
9341 for (i = 0; i < __MAX_XDP_MODE; i++)
9342 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9346 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9348 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9350 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9352 return prog ? prog->aux->id : 0;
9355 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9356 struct bpf_xdp_link *link)
9358 dev->xdp_state[mode].link = link;
9359 dev->xdp_state[mode].prog = NULL;
9362 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9363 struct bpf_prog *prog)
9365 dev->xdp_state[mode].link = NULL;
9366 dev->xdp_state[mode].prog = prog;
9369 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9370 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9371 u32 flags, struct bpf_prog *prog)
9373 struct netdev_bpf xdp;
9376 memset(&xdp, 0, sizeof(xdp));
9377 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9378 xdp.extack = extack;
9382 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9383 * "moved" into driver), so they don't increment it on their own, but
9384 * they do decrement refcnt when program is detached or replaced.
9385 * Given net_device also owns link/prog, we need to bump refcnt here
9386 * to prevent drivers from underflowing it.
9390 err = bpf_op(dev, &xdp);
9397 if (mode != XDP_MODE_HW)
9398 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9403 static void dev_xdp_uninstall(struct net_device *dev)
9405 struct bpf_xdp_link *link;
9406 struct bpf_prog *prog;
9407 enum bpf_xdp_mode mode;
9412 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9413 prog = dev_xdp_prog(dev, mode);
9417 bpf_op = dev_xdp_bpf_op(dev, mode);
9421 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9423 /* auto-detach link from net device */
9424 link = dev_xdp_link(dev, mode);
9430 dev_xdp_set_link(dev, mode, NULL);
9434 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9435 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9436 struct bpf_prog *old_prog, u32 flags)
9438 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9439 struct bpf_prog *cur_prog;
9440 struct net_device *upper;
9441 struct list_head *iter;
9442 enum bpf_xdp_mode mode;
9448 /* either link or prog attachment, never both */
9449 if (link && (new_prog || old_prog))
9451 /* link supports only XDP mode flags */
9452 if (link && (flags & ~XDP_FLAGS_MODES)) {
9453 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9456 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9457 if (num_modes > 1) {
9458 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9461 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9462 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9463 NL_SET_ERR_MSG(extack,
9464 "More than one program loaded, unset mode is ambiguous");
9467 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9468 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9469 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9473 mode = dev_xdp_mode(dev, flags);
9474 /* can't replace attached link */
9475 if (dev_xdp_link(dev, mode)) {
9476 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9480 /* don't allow if an upper device already has a program */
9481 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9482 if (dev_xdp_prog_count(upper) > 0) {
9483 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9488 cur_prog = dev_xdp_prog(dev, mode);
9489 /* can't replace attached prog with link */
9490 if (link && cur_prog) {
9491 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9494 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9495 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9499 /* put effective new program into new_prog */
9501 new_prog = link->link.prog;
9504 bool offload = mode == XDP_MODE_HW;
9505 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9506 ? XDP_MODE_DRV : XDP_MODE_SKB;
9508 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9509 NL_SET_ERR_MSG(extack, "XDP program already attached");
9512 if (!offload && dev_xdp_prog(dev, other_mode)) {
9513 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9516 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9517 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9520 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9521 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9524 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9525 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9530 /* don't call drivers if the effective program didn't change */
9531 if (new_prog != cur_prog) {
9532 bpf_op = dev_xdp_bpf_op(dev, mode);
9534 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9538 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9544 dev_xdp_set_link(dev, mode, link);
9546 dev_xdp_set_prog(dev, mode, new_prog);
9548 bpf_prog_put(cur_prog);
9553 static int dev_xdp_attach_link(struct net_device *dev,
9554 struct netlink_ext_ack *extack,
9555 struct bpf_xdp_link *link)
9557 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9560 static int dev_xdp_detach_link(struct net_device *dev,
9561 struct netlink_ext_ack *extack,
9562 struct bpf_xdp_link *link)
9564 enum bpf_xdp_mode mode;
9569 mode = dev_xdp_mode(dev, link->flags);
9570 if (dev_xdp_link(dev, mode) != link)
9573 bpf_op = dev_xdp_bpf_op(dev, mode);
9574 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9575 dev_xdp_set_link(dev, mode, NULL);
9579 static void bpf_xdp_link_release(struct bpf_link *link)
9581 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9585 /* if racing with net_device's tear down, xdp_link->dev might be
9586 * already NULL, in which case link was already auto-detached
9588 if (xdp_link->dev) {
9589 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9590 xdp_link->dev = NULL;
9596 static int bpf_xdp_link_detach(struct bpf_link *link)
9598 bpf_xdp_link_release(link);
9602 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9604 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9609 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9610 struct seq_file *seq)
9612 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9617 ifindex = xdp_link->dev->ifindex;
9620 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9623 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9624 struct bpf_link_info *info)
9626 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9631 ifindex = xdp_link->dev->ifindex;
9634 info->xdp.ifindex = ifindex;
9638 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9639 struct bpf_prog *old_prog)
9641 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9642 enum bpf_xdp_mode mode;
9648 /* link might have been auto-released already, so fail */
9649 if (!xdp_link->dev) {
9654 if (old_prog && link->prog != old_prog) {
9658 old_prog = link->prog;
9659 if (old_prog == new_prog) {
9660 /* no-op, don't disturb drivers */
9661 bpf_prog_put(new_prog);
9665 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9666 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9667 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9668 xdp_link->flags, new_prog);
9672 old_prog = xchg(&link->prog, new_prog);
9673 bpf_prog_put(old_prog);
9680 static const struct bpf_link_ops bpf_xdp_link_lops = {
9681 .release = bpf_xdp_link_release,
9682 .dealloc = bpf_xdp_link_dealloc,
9683 .detach = bpf_xdp_link_detach,
9684 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9685 .fill_link_info = bpf_xdp_link_fill_link_info,
9686 .update_prog = bpf_xdp_link_update,
9689 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9691 struct net *net = current->nsproxy->net_ns;
9692 struct bpf_link_primer link_primer;
9693 struct bpf_xdp_link *link;
9694 struct net_device *dev;
9698 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9704 link = kzalloc(sizeof(*link), GFP_USER);
9710 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9712 link->flags = attr->link_create.flags;
9714 err = bpf_link_prime(&link->link, &link_primer);
9720 err = dev_xdp_attach_link(dev, NULL, link);
9725 bpf_link_cleanup(&link_primer);
9729 fd = bpf_link_settle(&link_primer);
9730 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9743 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9745 * @extack: netlink extended ack
9746 * @fd: new program fd or negative value to clear
9747 * @expected_fd: old program fd that userspace expects to replace or clear
9748 * @flags: xdp-related flags
9750 * Set or clear a bpf program for a device
9752 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9753 int fd, int expected_fd, u32 flags)
9755 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9756 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9762 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9763 mode != XDP_MODE_SKB);
9764 if (IS_ERR(new_prog))
9765 return PTR_ERR(new_prog);
9768 if (expected_fd >= 0) {
9769 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9770 mode != XDP_MODE_SKB);
9771 if (IS_ERR(old_prog)) {
9772 err = PTR_ERR(old_prog);
9778 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9781 if (err && new_prog)
9782 bpf_prog_put(new_prog);
9784 bpf_prog_put(old_prog);
9789 * dev_new_index - allocate an ifindex
9790 * @net: the applicable net namespace
9792 * Returns a suitable unique value for a new device interface
9793 * number. The caller must hold the rtnl semaphore or the
9794 * dev_base_lock to be sure it remains unique.
9796 static int dev_new_index(struct net *net)
9798 int ifindex = net->ifindex;
9803 if (!__dev_get_by_index(net, ifindex))
9804 return net->ifindex = ifindex;
9808 /* Delayed registration/unregisteration */
9809 static LIST_HEAD(net_todo_list);
9810 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9812 static void net_set_todo(struct net_device *dev)
9814 list_add_tail(&dev->todo_list, &net_todo_list);
9815 dev_net(dev)->dev_unreg_count++;
9818 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9819 struct net_device *upper, netdev_features_t features)
9821 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9822 netdev_features_t feature;
9825 for_each_netdev_feature(upper_disables, feature_bit) {
9826 feature = __NETIF_F_BIT(feature_bit);
9827 if (!(upper->wanted_features & feature)
9828 && (features & feature)) {
9829 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9830 &feature, upper->name);
9831 features &= ~feature;
9838 static void netdev_sync_lower_features(struct net_device *upper,
9839 struct net_device *lower, netdev_features_t features)
9841 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9842 netdev_features_t feature;
9845 for_each_netdev_feature(upper_disables, feature_bit) {
9846 feature = __NETIF_F_BIT(feature_bit);
9847 if (!(features & feature) && (lower->features & feature)) {
9848 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9849 &feature, lower->name);
9850 lower->wanted_features &= ~feature;
9851 __netdev_update_features(lower);
9853 if (unlikely(lower->features & feature))
9854 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9855 &feature, lower->name);
9857 netdev_features_change(lower);
9862 static netdev_features_t netdev_fix_features(struct net_device *dev,
9863 netdev_features_t features)
9865 /* Fix illegal checksum combinations */
9866 if ((features & NETIF_F_HW_CSUM) &&
9867 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9868 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9869 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9872 /* TSO requires that SG is present as well. */
9873 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9874 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9875 features &= ~NETIF_F_ALL_TSO;
9878 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9879 !(features & NETIF_F_IP_CSUM)) {
9880 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9881 features &= ~NETIF_F_TSO;
9882 features &= ~NETIF_F_TSO_ECN;
9885 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9886 !(features & NETIF_F_IPV6_CSUM)) {
9887 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9888 features &= ~NETIF_F_TSO6;
9891 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9892 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9893 features &= ~NETIF_F_TSO_MANGLEID;
9895 /* TSO ECN requires that TSO is present as well. */
9896 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9897 features &= ~NETIF_F_TSO_ECN;
9899 /* Software GSO depends on SG. */
9900 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9901 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9902 features &= ~NETIF_F_GSO;
9905 /* GSO partial features require GSO partial be set */
9906 if ((features & dev->gso_partial_features) &&
9907 !(features & NETIF_F_GSO_PARTIAL)) {
9909 "Dropping partially supported GSO features since no GSO partial.\n");
9910 features &= ~dev->gso_partial_features;
9913 if (!(features & NETIF_F_RXCSUM)) {
9914 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9915 * successfully merged by hardware must also have the
9916 * checksum verified by hardware. If the user does not
9917 * want to enable RXCSUM, logically, we should disable GRO_HW.
9919 if (features & NETIF_F_GRO_HW) {
9920 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9921 features &= ~NETIF_F_GRO_HW;
9925 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9926 if (features & NETIF_F_RXFCS) {
9927 if (features & NETIF_F_LRO) {
9928 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9929 features &= ~NETIF_F_LRO;
9932 if (features & NETIF_F_GRO_HW) {
9933 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9934 features &= ~NETIF_F_GRO_HW;
9938 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9939 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9940 features &= ~NETIF_F_LRO;
9943 if (features & NETIF_F_HW_TLS_TX) {
9944 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9945 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9946 bool hw_csum = features & NETIF_F_HW_CSUM;
9948 if (!ip_csum && !hw_csum) {
9949 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9950 features &= ~NETIF_F_HW_TLS_TX;
9954 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9955 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9956 features &= ~NETIF_F_HW_TLS_RX;
9962 int __netdev_update_features(struct net_device *dev)
9964 struct net_device *upper, *lower;
9965 netdev_features_t features;
9966 struct list_head *iter;
9971 features = netdev_get_wanted_features(dev);
9973 if (dev->netdev_ops->ndo_fix_features)
9974 features = dev->netdev_ops->ndo_fix_features(dev, features);
9976 /* driver might be less strict about feature dependencies */
9977 features = netdev_fix_features(dev, features);
9979 /* some features can't be enabled if they're off on an upper device */
9980 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9981 features = netdev_sync_upper_features(dev, upper, features);
9983 if (dev->features == features)
9986 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9987 &dev->features, &features);
9989 if (dev->netdev_ops->ndo_set_features)
9990 err = dev->netdev_ops->ndo_set_features(dev, features);
9994 if (unlikely(err < 0)) {
9996 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9997 err, &features, &dev->features);
9998 /* return non-0 since some features might have changed and
9999 * it's better to fire a spurious notification than miss it
10005 /* some features must be disabled on lower devices when disabled
10006 * on an upper device (think: bonding master or bridge)
10008 netdev_for_each_lower_dev(dev, lower, iter)
10009 netdev_sync_lower_features(dev, lower, features);
10012 netdev_features_t diff = features ^ dev->features;
10014 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10015 /* udp_tunnel_{get,drop}_rx_info both need
10016 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10017 * device, or they won't do anything.
10018 * Thus we need to update dev->features
10019 * *before* calling udp_tunnel_get_rx_info,
10020 * but *after* calling udp_tunnel_drop_rx_info.
10022 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10023 dev->features = features;
10024 udp_tunnel_get_rx_info(dev);
10026 udp_tunnel_drop_rx_info(dev);
10030 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10031 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10032 dev->features = features;
10033 err |= vlan_get_rx_ctag_filter_info(dev);
10035 vlan_drop_rx_ctag_filter_info(dev);
10039 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10040 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10041 dev->features = features;
10042 err |= vlan_get_rx_stag_filter_info(dev);
10044 vlan_drop_rx_stag_filter_info(dev);
10048 dev->features = features;
10051 return err < 0 ? 0 : 1;
10055 * netdev_update_features - recalculate device features
10056 * @dev: the device to check
10058 * Recalculate dev->features set and send notifications if it
10059 * has changed. Should be called after driver or hardware dependent
10060 * conditions might have changed that influence the features.
10062 void netdev_update_features(struct net_device *dev)
10064 if (__netdev_update_features(dev))
10065 netdev_features_change(dev);
10067 EXPORT_SYMBOL(netdev_update_features);
10070 * netdev_change_features - recalculate device features
10071 * @dev: the device to check
10073 * Recalculate dev->features set and send notifications even
10074 * if they have not changed. Should be called instead of
10075 * netdev_update_features() if also dev->vlan_features might
10076 * have changed to allow the changes to be propagated to stacked
10079 void netdev_change_features(struct net_device *dev)
10081 __netdev_update_features(dev);
10082 netdev_features_change(dev);
10084 EXPORT_SYMBOL(netdev_change_features);
10087 * netif_stacked_transfer_operstate - transfer operstate
10088 * @rootdev: the root or lower level device to transfer state from
10089 * @dev: the device to transfer operstate to
10091 * Transfer operational state from root to device. This is normally
10092 * called when a stacking relationship exists between the root
10093 * device and the device(a leaf device).
10095 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10096 struct net_device *dev)
10098 if (rootdev->operstate == IF_OPER_DORMANT)
10099 netif_dormant_on(dev);
10101 netif_dormant_off(dev);
10103 if (rootdev->operstate == IF_OPER_TESTING)
10104 netif_testing_on(dev);
10106 netif_testing_off(dev);
10108 if (netif_carrier_ok(rootdev))
10109 netif_carrier_on(dev);
10111 netif_carrier_off(dev);
10113 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10115 static int netif_alloc_rx_queues(struct net_device *dev)
10117 unsigned int i, count = dev->num_rx_queues;
10118 struct netdev_rx_queue *rx;
10119 size_t sz = count * sizeof(*rx);
10124 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10130 for (i = 0; i < count; i++) {
10133 /* XDP RX-queue setup */
10134 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10141 /* Rollback successful reg's and free other resources */
10143 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10149 static void netif_free_rx_queues(struct net_device *dev)
10151 unsigned int i, count = dev->num_rx_queues;
10153 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10157 for (i = 0; i < count; i++)
10158 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10163 static void netdev_init_one_queue(struct net_device *dev,
10164 struct netdev_queue *queue, void *_unused)
10166 /* Initialize queue lock */
10167 spin_lock_init(&queue->_xmit_lock);
10168 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10169 queue->xmit_lock_owner = -1;
10170 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10173 dql_init(&queue->dql, HZ);
10177 static void netif_free_tx_queues(struct net_device *dev)
10182 static int netif_alloc_netdev_queues(struct net_device *dev)
10184 unsigned int count = dev->num_tx_queues;
10185 struct netdev_queue *tx;
10186 size_t sz = count * sizeof(*tx);
10188 if (count < 1 || count > 0xffff)
10191 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10197 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10198 spin_lock_init(&dev->tx_global_lock);
10203 void netif_tx_stop_all_queues(struct net_device *dev)
10207 for (i = 0; i < dev->num_tx_queues; i++) {
10208 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10210 netif_tx_stop_queue(txq);
10213 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10216 * register_netdevice - register a network device
10217 * @dev: device to register
10219 * Take a completed network device structure and add it to the kernel
10220 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10221 * chain. 0 is returned on success. A negative errno code is returned
10222 * on a failure to set up the device, or if the name is a duplicate.
10224 * Callers must hold the rtnl semaphore. You may want
10225 * register_netdev() instead of this.
10228 * The locking appears insufficient to guarantee two parallel registers
10229 * will not get the same name.
10232 int register_netdevice(struct net_device *dev)
10235 struct net *net = dev_net(dev);
10237 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10238 NETDEV_FEATURE_COUNT);
10239 BUG_ON(dev_boot_phase);
10244 /* When net_device's are persistent, this will be fatal. */
10245 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10248 ret = ethtool_check_ops(dev->ethtool_ops);
10252 spin_lock_init(&dev->addr_list_lock);
10253 netdev_set_addr_lockdep_class(dev);
10255 ret = dev_get_valid_name(net, dev, dev->name);
10260 dev->name_node = netdev_name_node_head_alloc(dev);
10261 if (!dev->name_node)
10264 /* Init, if this function is available */
10265 if (dev->netdev_ops->ndo_init) {
10266 ret = dev->netdev_ops->ndo_init(dev);
10270 goto err_free_name;
10274 if (((dev->hw_features | dev->features) &
10275 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10276 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10277 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10278 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10285 dev->ifindex = dev_new_index(net);
10286 else if (__dev_get_by_index(net, dev->ifindex))
10289 /* Transfer changeable features to wanted_features and enable
10290 * software offloads (GSO and GRO).
10292 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10293 dev->features |= NETIF_F_SOFT_FEATURES;
10295 if (dev->udp_tunnel_nic_info) {
10296 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10297 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10300 dev->wanted_features = dev->features & dev->hw_features;
10302 if (!(dev->flags & IFF_LOOPBACK))
10303 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10305 /* If IPv4 TCP segmentation offload is supported we should also
10306 * allow the device to enable segmenting the frame with the option
10307 * of ignoring a static IP ID value. This doesn't enable the
10308 * feature itself but allows the user to enable it later.
10310 if (dev->hw_features & NETIF_F_TSO)
10311 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10312 if (dev->vlan_features & NETIF_F_TSO)
10313 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10314 if (dev->mpls_features & NETIF_F_TSO)
10315 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10316 if (dev->hw_enc_features & NETIF_F_TSO)
10317 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10319 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10321 dev->vlan_features |= NETIF_F_HIGHDMA;
10323 /* Make NETIF_F_SG inheritable to tunnel devices.
10325 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10327 /* Make NETIF_F_SG inheritable to MPLS.
10329 dev->mpls_features |= NETIF_F_SG;
10331 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10332 ret = notifier_to_errno(ret);
10336 ret = netdev_register_kobject(dev);
10338 dev->reg_state = NETREG_UNREGISTERED;
10341 dev->reg_state = NETREG_REGISTERED;
10343 __netdev_update_features(dev);
10346 * Default initial state at registry is that the
10347 * device is present.
10350 set_bit(__LINK_STATE_PRESENT, &dev->state);
10352 linkwatch_init_dev(dev);
10354 dev_init_scheduler(dev);
10356 list_netdevice(dev);
10357 add_device_randomness(dev->dev_addr, dev->addr_len);
10359 /* If the device has permanent device address, driver should
10360 * set dev_addr and also addr_assign_type should be set to
10361 * NET_ADDR_PERM (default value).
10363 if (dev->addr_assign_type == NET_ADDR_PERM)
10364 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10366 /* Notify protocols, that a new device appeared. */
10367 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10368 ret = notifier_to_errno(ret);
10370 /* Expect explicit free_netdev() on failure */
10371 dev->needs_free_netdev = false;
10372 unregister_netdevice_queue(dev, NULL);
10376 * Prevent userspace races by waiting until the network
10377 * device is fully setup before sending notifications.
10379 if (!dev->rtnl_link_ops ||
10380 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10381 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10387 if (dev->netdev_ops->ndo_uninit)
10388 dev->netdev_ops->ndo_uninit(dev);
10389 if (dev->priv_destructor)
10390 dev->priv_destructor(dev);
10392 netdev_name_node_free(dev->name_node);
10395 EXPORT_SYMBOL(register_netdevice);
10398 * init_dummy_netdev - init a dummy network device for NAPI
10399 * @dev: device to init
10401 * This takes a network device structure and initialize the minimum
10402 * amount of fields so it can be used to schedule NAPI polls without
10403 * registering a full blown interface. This is to be used by drivers
10404 * that need to tie several hardware interfaces to a single NAPI
10405 * poll scheduler due to HW limitations.
10407 int init_dummy_netdev(struct net_device *dev)
10409 /* Clear everything. Note we don't initialize spinlocks
10410 * are they aren't supposed to be taken by any of the
10411 * NAPI code and this dummy netdev is supposed to be
10412 * only ever used for NAPI polls
10414 memset(dev, 0, sizeof(struct net_device));
10416 /* make sure we BUG if trying to hit standard
10417 * register/unregister code path
10419 dev->reg_state = NETREG_DUMMY;
10421 /* NAPI wants this */
10422 INIT_LIST_HEAD(&dev->napi_list);
10424 /* a dummy interface is started by default */
10425 set_bit(__LINK_STATE_PRESENT, &dev->state);
10426 set_bit(__LINK_STATE_START, &dev->state);
10428 /* napi_busy_loop stats accounting wants this */
10429 dev_net_set(dev, &init_net);
10431 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10432 * because users of this 'device' dont need to change
10438 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10442 * register_netdev - register a network device
10443 * @dev: device to register
10445 * Take a completed network device structure and add it to the kernel
10446 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10447 * chain. 0 is returned on success. A negative errno code is returned
10448 * on a failure to set up the device, or if the name is a duplicate.
10450 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10451 * and expands the device name if you passed a format string to
10454 int register_netdev(struct net_device *dev)
10458 if (rtnl_lock_killable())
10460 err = register_netdevice(dev);
10464 EXPORT_SYMBOL(register_netdev);
10466 int netdev_refcnt_read(const struct net_device *dev)
10468 #ifdef CONFIG_PCPU_DEV_REFCNT
10471 for_each_possible_cpu(i)
10472 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10475 return refcount_read(&dev->dev_refcnt);
10478 EXPORT_SYMBOL(netdev_refcnt_read);
10480 int netdev_unregister_timeout_secs __read_mostly = 10;
10482 #define WAIT_REFS_MIN_MSECS 1
10483 #define WAIT_REFS_MAX_MSECS 250
10485 * netdev_wait_allrefs - wait until all references are gone.
10486 * @dev: target net_device
10488 * This is called when unregistering network devices.
10490 * Any protocol or device that holds a reference should register
10491 * for netdevice notification, and cleanup and put back the
10492 * reference if they receive an UNREGISTER event.
10493 * We can get stuck here if buggy protocols don't correctly
10496 static void netdev_wait_allrefs(struct net_device *dev)
10498 unsigned long rebroadcast_time, warning_time;
10499 int wait = 0, refcnt;
10501 linkwatch_forget_dev(dev);
10503 rebroadcast_time = warning_time = jiffies;
10504 refcnt = netdev_refcnt_read(dev);
10506 while (refcnt != 1) {
10507 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10510 /* Rebroadcast unregister notification */
10511 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10517 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10519 /* We must not have linkwatch events
10520 * pending on unregister. If this
10521 * happens, we simply run the queue
10522 * unscheduled, resulting in a noop
10525 linkwatch_run_queue();
10530 rebroadcast_time = jiffies;
10535 wait = WAIT_REFS_MIN_MSECS;
10538 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10541 refcnt = netdev_refcnt_read(dev);
10544 time_after(jiffies, warning_time +
10545 netdev_unregister_timeout_secs * HZ)) {
10546 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10547 dev->name, refcnt);
10548 warning_time = jiffies;
10553 /* The sequence is:
10557 * register_netdevice(x1);
10558 * register_netdevice(x2);
10560 * unregister_netdevice(y1);
10561 * unregister_netdevice(y2);
10567 * We are invoked by rtnl_unlock().
10568 * This allows us to deal with problems:
10569 * 1) We can delete sysfs objects which invoke hotplug
10570 * without deadlocking with linkwatch via keventd.
10571 * 2) Since we run with the RTNL semaphore not held, we can sleep
10572 * safely in order to wait for the netdev refcnt to drop to zero.
10574 * We must not return until all unregister events added during
10575 * the interval the lock was held have been completed.
10577 void netdev_run_todo(void)
10579 struct list_head list;
10580 #ifdef CONFIG_LOCKDEP
10581 struct list_head unlink_list;
10583 list_replace_init(&net_unlink_list, &unlink_list);
10585 while (!list_empty(&unlink_list)) {
10586 struct net_device *dev = list_first_entry(&unlink_list,
10589 list_del_init(&dev->unlink_list);
10590 dev->nested_level = dev->lower_level - 1;
10594 /* Snapshot list, allow later requests */
10595 list_replace_init(&net_todo_list, &list);
10600 /* Wait for rcu callbacks to finish before next phase */
10601 if (!list_empty(&list))
10604 while (!list_empty(&list)) {
10605 struct net_device *dev
10606 = list_first_entry(&list, struct net_device, todo_list);
10607 list_del(&dev->todo_list);
10609 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10610 pr_err("network todo '%s' but state %d\n",
10611 dev->name, dev->reg_state);
10616 dev->reg_state = NETREG_UNREGISTERED;
10618 netdev_wait_allrefs(dev);
10621 BUG_ON(netdev_refcnt_read(dev) != 1);
10622 BUG_ON(!list_empty(&dev->ptype_all));
10623 BUG_ON(!list_empty(&dev->ptype_specific));
10624 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10625 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10626 #if IS_ENABLED(CONFIG_DECNET)
10627 WARN_ON(dev->dn_ptr);
10629 if (dev->priv_destructor)
10630 dev->priv_destructor(dev);
10631 if (dev->needs_free_netdev)
10634 /* Report a network device has been unregistered */
10636 dev_net(dev)->dev_unreg_count--;
10638 wake_up(&netdev_unregistering_wq);
10640 /* Free network device */
10641 kobject_put(&dev->dev.kobj);
10645 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10646 * all the same fields in the same order as net_device_stats, with only
10647 * the type differing, but rtnl_link_stats64 may have additional fields
10648 * at the end for newer counters.
10650 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10651 const struct net_device_stats *netdev_stats)
10653 #if BITS_PER_LONG == 64
10654 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10655 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10656 /* zero out counters that only exist in rtnl_link_stats64 */
10657 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10658 sizeof(*stats64) - sizeof(*netdev_stats));
10660 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10661 const unsigned long *src = (const unsigned long *)netdev_stats;
10662 u64 *dst = (u64 *)stats64;
10664 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10665 for (i = 0; i < n; i++)
10667 /* zero out counters that only exist in rtnl_link_stats64 */
10668 memset((char *)stats64 + n * sizeof(u64), 0,
10669 sizeof(*stats64) - n * sizeof(u64));
10672 EXPORT_SYMBOL(netdev_stats_to_stats64);
10675 * dev_get_stats - get network device statistics
10676 * @dev: device to get statistics from
10677 * @storage: place to store stats
10679 * Get network statistics from device. Return @storage.
10680 * The device driver may provide its own method by setting
10681 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10682 * otherwise the internal statistics structure is used.
10684 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10685 struct rtnl_link_stats64 *storage)
10687 const struct net_device_ops *ops = dev->netdev_ops;
10689 if (ops->ndo_get_stats64) {
10690 memset(storage, 0, sizeof(*storage));
10691 ops->ndo_get_stats64(dev, storage);
10692 } else if (ops->ndo_get_stats) {
10693 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10695 netdev_stats_to_stats64(storage, &dev->stats);
10697 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10698 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10699 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10702 EXPORT_SYMBOL(dev_get_stats);
10705 * dev_fetch_sw_netstats - get per-cpu network device statistics
10706 * @s: place to store stats
10707 * @netstats: per-cpu network stats to read from
10709 * Read per-cpu network statistics and populate the related fields in @s.
10711 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10712 const struct pcpu_sw_netstats __percpu *netstats)
10716 for_each_possible_cpu(cpu) {
10717 const struct pcpu_sw_netstats *stats;
10718 struct pcpu_sw_netstats tmp;
10719 unsigned int start;
10721 stats = per_cpu_ptr(netstats, cpu);
10723 start = u64_stats_fetch_begin_irq(&stats->syncp);
10724 tmp.rx_packets = stats->rx_packets;
10725 tmp.rx_bytes = stats->rx_bytes;
10726 tmp.tx_packets = stats->tx_packets;
10727 tmp.tx_bytes = stats->tx_bytes;
10728 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10730 s->rx_packets += tmp.rx_packets;
10731 s->rx_bytes += tmp.rx_bytes;
10732 s->tx_packets += tmp.tx_packets;
10733 s->tx_bytes += tmp.tx_bytes;
10736 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10739 * dev_get_tstats64 - ndo_get_stats64 implementation
10740 * @dev: device to get statistics from
10741 * @s: place to store stats
10743 * Populate @s from dev->stats and dev->tstats. Can be used as
10744 * ndo_get_stats64() callback.
10746 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10748 netdev_stats_to_stats64(s, &dev->stats);
10749 dev_fetch_sw_netstats(s, dev->tstats);
10751 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10753 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10755 struct netdev_queue *queue = dev_ingress_queue(dev);
10757 #ifdef CONFIG_NET_CLS_ACT
10760 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10763 netdev_init_one_queue(dev, queue, NULL);
10764 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10765 queue->qdisc_sleeping = &noop_qdisc;
10766 rcu_assign_pointer(dev->ingress_queue, queue);
10771 static const struct ethtool_ops default_ethtool_ops;
10773 void netdev_set_default_ethtool_ops(struct net_device *dev,
10774 const struct ethtool_ops *ops)
10776 if (dev->ethtool_ops == &default_ethtool_ops)
10777 dev->ethtool_ops = ops;
10779 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10781 void netdev_freemem(struct net_device *dev)
10783 char *addr = (char *)dev - dev->padded;
10789 * alloc_netdev_mqs - allocate network device
10790 * @sizeof_priv: size of private data to allocate space for
10791 * @name: device name format string
10792 * @name_assign_type: origin of device name
10793 * @setup: callback to initialize device
10794 * @txqs: the number of TX subqueues to allocate
10795 * @rxqs: the number of RX subqueues to allocate
10797 * Allocates a struct net_device with private data area for driver use
10798 * and performs basic initialization. Also allocates subqueue structs
10799 * for each queue on the device.
10801 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10802 unsigned char name_assign_type,
10803 void (*setup)(struct net_device *),
10804 unsigned int txqs, unsigned int rxqs)
10806 struct net_device *dev;
10807 unsigned int alloc_size;
10808 struct net_device *p;
10810 BUG_ON(strlen(name) >= sizeof(dev->name));
10813 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10818 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10822 alloc_size = sizeof(struct net_device);
10824 /* ensure 32-byte alignment of private area */
10825 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10826 alloc_size += sizeof_priv;
10828 /* ensure 32-byte alignment of whole construct */
10829 alloc_size += NETDEV_ALIGN - 1;
10831 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10835 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10836 dev->padded = (char *)dev - (char *)p;
10838 #ifdef CONFIG_PCPU_DEV_REFCNT
10839 dev->pcpu_refcnt = alloc_percpu(int);
10840 if (!dev->pcpu_refcnt)
10844 refcount_set(&dev->dev_refcnt, 1);
10847 if (dev_addr_init(dev))
10853 dev_net_set(dev, &init_net);
10855 dev->gso_max_size = GSO_MAX_SIZE;
10856 dev->gso_max_segs = GSO_MAX_SEGS;
10857 dev->upper_level = 1;
10858 dev->lower_level = 1;
10859 #ifdef CONFIG_LOCKDEP
10860 dev->nested_level = 0;
10861 INIT_LIST_HEAD(&dev->unlink_list);
10864 INIT_LIST_HEAD(&dev->napi_list);
10865 INIT_LIST_HEAD(&dev->unreg_list);
10866 INIT_LIST_HEAD(&dev->close_list);
10867 INIT_LIST_HEAD(&dev->link_watch_list);
10868 INIT_LIST_HEAD(&dev->adj_list.upper);
10869 INIT_LIST_HEAD(&dev->adj_list.lower);
10870 INIT_LIST_HEAD(&dev->ptype_all);
10871 INIT_LIST_HEAD(&dev->ptype_specific);
10872 INIT_LIST_HEAD(&dev->net_notifier_list);
10873 #ifdef CONFIG_NET_SCHED
10874 hash_init(dev->qdisc_hash);
10876 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10879 if (!dev->tx_queue_len) {
10880 dev->priv_flags |= IFF_NO_QUEUE;
10881 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10884 dev->num_tx_queues = txqs;
10885 dev->real_num_tx_queues = txqs;
10886 if (netif_alloc_netdev_queues(dev))
10889 dev->num_rx_queues = rxqs;
10890 dev->real_num_rx_queues = rxqs;
10891 if (netif_alloc_rx_queues(dev))
10894 strcpy(dev->name, name);
10895 dev->name_assign_type = name_assign_type;
10896 dev->group = INIT_NETDEV_GROUP;
10897 if (!dev->ethtool_ops)
10898 dev->ethtool_ops = &default_ethtool_ops;
10900 nf_hook_netdev_init(dev);
10909 #ifdef CONFIG_PCPU_DEV_REFCNT
10910 free_percpu(dev->pcpu_refcnt);
10913 netdev_freemem(dev);
10916 EXPORT_SYMBOL(alloc_netdev_mqs);
10919 * free_netdev - free network device
10922 * This function does the last stage of destroying an allocated device
10923 * interface. The reference to the device object is released. If this
10924 * is the last reference then it will be freed.Must be called in process
10927 void free_netdev(struct net_device *dev)
10929 struct napi_struct *p, *n;
10933 /* When called immediately after register_netdevice() failed the unwind
10934 * handling may still be dismantling the device. Handle that case by
10935 * deferring the free.
10937 if (dev->reg_state == NETREG_UNREGISTERING) {
10939 dev->needs_free_netdev = true;
10943 netif_free_tx_queues(dev);
10944 netif_free_rx_queues(dev);
10946 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10948 /* Flush device addresses */
10949 dev_addr_flush(dev);
10951 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10954 #ifdef CONFIG_PCPU_DEV_REFCNT
10955 free_percpu(dev->pcpu_refcnt);
10956 dev->pcpu_refcnt = NULL;
10958 free_percpu(dev->xdp_bulkq);
10959 dev->xdp_bulkq = NULL;
10961 /* Compatibility with error handling in drivers */
10962 if (dev->reg_state == NETREG_UNINITIALIZED) {
10963 netdev_freemem(dev);
10967 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10968 dev->reg_state = NETREG_RELEASED;
10970 /* will free via device release */
10971 put_device(&dev->dev);
10973 EXPORT_SYMBOL(free_netdev);
10976 * synchronize_net - Synchronize with packet receive processing
10978 * Wait for packets currently being received to be done.
10979 * Does not block later packets from starting.
10981 void synchronize_net(void)
10984 if (rtnl_is_locked())
10985 synchronize_rcu_expedited();
10989 EXPORT_SYMBOL(synchronize_net);
10992 * unregister_netdevice_queue - remove device from the kernel
10996 * This function shuts down a device interface and removes it
10997 * from the kernel tables.
10998 * If head not NULL, device is queued to be unregistered later.
11000 * Callers must hold the rtnl semaphore. You may want
11001 * unregister_netdev() instead of this.
11004 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11009 list_move_tail(&dev->unreg_list, head);
11013 list_add(&dev->unreg_list, &single);
11014 unregister_netdevice_many(&single);
11017 EXPORT_SYMBOL(unregister_netdevice_queue);
11020 * unregister_netdevice_many - unregister many devices
11021 * @head: list of devices
11023 * Note: As most callers use a stack allocated list_head,
11024 * we force a list_del() to make sure stack wont be corrupted later.
11026 void unregister_netdevice_many(struct list_head *head)
11028 struct net_device *dev, *tmp;
11029 LIST_HEAD(close_head);
11031 BUG_ON(dev_boot_phase);
11034 if (list_empty(head))
11037 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11038 /* Some devices call without registering
11039 * for initialization unwind. Remove those
11040 * devices and proceed with the remaining.
11042 if (dev->reg_state == NETREG_UNINITIALIZED) {
11043 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11047 list_del(&dev->unreg_list);
11050 dev->dismantle = true;
11051 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11054 /* If device is running, close it first. */
11055 list_for_each_entry(dev, head, unreg_list)
11056 list_add_tail(&dev->close_list, &close_head);
11057 dev_close_many(&close_head, true);
11059 list_for_each_entry(dev, head, unreg_list) {
11060 /* And unlink it from device chain. */
11061 unlist_netdevice(dev);
11063 dev->reg_state = NETREG_UNREGISTERING;
11065 flush_all_backlogs();
11069 list_for_each_entry(dev, head, unreg_list) {
11070 struct sk_buff *skb = NULL;
11072 /* Shutdown queueing discipline. */
11075 dev_xdp_uninstall(dev);
11077 /* Notify protocols, that we are about to destroy
11078 * this device. They should clean all the things.
11080 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11082 if (!dev->rtnl_link_ops ||
11083 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11084 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11085 GFP_KERNEL, NULL, 0);
11088 * Flush the unicast and multicast chains
11093 netdev_name_node_alt_flush(dev);
11094 netdev_name_node_free(dev->name_node);
11096 if (dev->netdev_ops->ndo_uninit)
11097 dev->netdev_ops->ndo_uninit(dev);
11100 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11102 /* Notifier chain MUST detach us all upper devices. */
11103 WARN_ON(netdev_has_any_upper_dev(dev));
11104 WARN_ON(netdev_has_any_lower_dev(dev));
11106 /* Remove entries from kobject tree */
11107 netdev_unregister_kobject(dev);
11109 /* Remove XPS queueing entries */
11110 netif_reset_xps_queues_gt(dev, 0);
11116 list_for_each_entry(dev, head, unreg_list) {
11123 EXPORT_SYMBOL(unregister_netdevice_many);
11126 * unregister_netdev - remove device from the kernel
11129 * This function shuts down a device interface and removes it
11130 * from the kernel tables.
11132 * This is just a wrapper for unregister_netdevice that takes
11133 * the rtnl semaphore. In general you want to use this and not
11134 * unregister_netdevice.
11136 void unregister_netdev(struct net_device *dev)
11139 unregister_netdevice(dev);
11142 EXPORT_SYMBOL(unregister_netdev);
11145 * __dev_change_net_namespace - move device to different nethost namespace
11147 * @net: network namespace
11148 * @pat: If not NULL name pattern to try if the current device name
11149 * is already taken in the destination network namespace.
11150 * @new_ifindex: If not zero, specifies device index in the target
11153 * This function shuts down a device interface and moves it
11154 * to a new network namespace. On success 0 is returned, on
11155 * a failure a netagive errno code is returned.
11157 * Callers must hold the rtnl semaphore.
11160 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11161 const char *pat, int new_ifindex)
11163 struct net *net_old = dev_net(dev);
11168 /* Don't allow namespace local devices to be moved. */
11170 if (dev->features & NETIF_F_NETNS_LOCAL)
11173 /* Ensure the device has been registrered */
11174 if (dev->reg_state != NETREG_REGISTERED)
11177 /* Get out if there is nothing todo */
11179 if (net_eq(net_old, net))
11182 /* Pick the destination device name, and ensure
11183 * we can use it in the destination network namespace.
11186 if (netdev_name_in_use(net, dev->name)) {
11187 /* We get here if we can't use the current device name */
11190 err = dev_get_valid_name(net, dev, pat);
11195 /* Check that new_ifindex isn't used yet. */
11197 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11201 * And now a mini version of register_netdevice unregister_netdevice.
11204 /* If device is running close it first. */
11207 /* And unlink it from device chain */
11208 unlist_netdevice(dev);
11212 /* Shutdown queueing discipline. */
11215 /* Notify protocols, that we are about to destroy
11216 * this device. They should clean all the things.
11218 * Note that dev->reg_state stays at NETREG_REGISTERED.
11219 * This is wanted because this way 8021q and macvlan know
11220 * the device is just moving and can keep their slaves up.
11222 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11225 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11226 /* If there is an ifindex conflict assign a new one */
11227 if (!new_ifindex) {
11228 if (__dev_get_by_index(net, dev->ifindex))
11229 new_ifindex = dev_new_index(net);
11231 new_ifindex = dev->ifindex;
11234 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11238 * Flush the unicast and multicast chains
11243 /* Send a netdev-removed uevent to the old namespace */
11244 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11245 netdev_adjacent_del_links(dev);
11247 /* Move per-net netdevice notifiers that are following the netdevice */
11248 move_netdevice_notifiers_dev_net(dev, net);
11250 /* Actually switch the network namespace */
11251 dev_net_set(dev, net);
11252 dev->ifindex = new_ifindex;
11254 /* Send a netdev-add uevent to the new namespace */
11255 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11256 netdev_adjacent_add_links(dev);
11258 /* Fixup kobjects */
11259 err = device_rename(&dev->dev, dev->name);
11262 /* Adapt owner in case owning user namespace of target network
11263 * namespace is different from the original one.
11265 err = netdev_change_owner(dev, net_old, net);
11268 /* Add the device back in the hashes */
11269 list_netdevice(dev);
11271 /* Notify protocols, that a new device appeared. */
11272 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11275 * Prevent userspace races by waiting until the network
11276 * device is fully setup before sending notifications.
11278 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11285 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11287 static int dev_cpu_dead(unsigned int oldcpu)
11289 struct sk_buff **list_skb;
11290 struct sk_buff *skb;
11292 struct softnet_data *sd, *oldsd, *remsd = NULL;
11294 local_irq_disable();
11295 cpu = smp_processor_id();
11296 sd = &per_cpu(softnet_data, cpu);
11297 oldsd = &per_cpu(softnet_data, oldcpu);
11299 /* Find end of our completion_queue. */
11300 list_skb = &sd->completion_queue;
11302 list_skb = &(*list_skb)->next;
11303 /* Append completion queue from offline CPU. */
11304 *list_skb = oldsd->completion_queue;
11305 oldsd->completion_queue = NULL;
11307 /* Append output queue from offline CPU. */
11308 if (oldsd->output_queue) {
11309 *sd->output_queue_tailp = oldsd->output_queue;
11310 sd->output_queue_tailp = oldsd->output_queue_tailp;
11311 oldsd->output_queue = NULL;
11312 oldsd->output_queue_tailp = &oldsd->output_queue;
11314 /* Append NAPI poll list from offline CPU, with one exception :
11315 * process_backlog() must be called by cpu owning percpu backlog.
11316 * We properly handle process_queue & input_pkt_queue later.
11318 while (!list_empty(&oldsd->poll_list)) {
11319 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11320 struct napi_struct,
11323 list_del_init(&napi->poll_list);
11324 if (napi->poll == process_backlog)
11327 ____napi_schedule(sd, napi);
11330 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11331 local_irq_enable();
11334 remsd = oldsd->rps_ipi_list;
11335 oldsd->rps_ipi_list = NULL;
11337 /* send out pending IPI's on offline CPU */
11338 net_rps_send_ipi(remsd);
11340 /* Process offline CPU's input_pkt_queue */
11341 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11343 input_queue_head_incr(oldsd);
11345 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11347 input_queue_head_incr(oldsd);
11354 * netdev_increment_features - increment feature set by one
11355 * @all: current feature set
11356 * @one: new feature set
11357 * @mask: mask feature set
11359 * Computes a new feature set after adding a device with feature set
11360 * @one to the master device with current feature set @all. Will not
11361 * enable anything that is off in @mask. Returns the new feature set.
11363 netdev_features_t netdev_increment_features(netdev_features_t all,
11364 netdev_features_t one, netdev_features_t mask)
11366 if (mask & NETIF_F_HW_CSUM)
11367 mask |= NETIF_F_CSUM_MASK;
11368 mask |= NETIF_F_VLAN_CHALLENGED;
11370 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11371 all &= one | ~NETIF_F_ALL_FOR_ALL;
11373 /* If one device supports hw checksumming, set for all. */
11374 if (all & NETIF_F_HW_CSUM)
11375 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11379 EXPORT_SYMBOL(netdev_increment_features);
11381 static struct hlist_head * __net_init netdev_create_hash(void)
11384 struct hlist_head *hash;
11386 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11388 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11389 INIT_HLIST_HEAD(&hash[i]);
11394 /* Initialize per network namespace state */
11395 static int __net_init netdev_init(struct net *net)
11397 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11398 8 * sizeof_field(struct napi_struct, gro_bitmask));
11400 if (net != &init_net)
11401 INIT_LIST_HEAD(&net->dev_base_head);
11403 net->dev_name_head = netdev_create_hash();
11404 if (net->dev_name_head == NULL)
11407 net->dev_index_head = netdev_create_hash();
11408 if (net->dev_index_head == NULL)
11411 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11416 kfree(net->dev_name_head);
11422 * netdev_drivername - network driver for the device
11423 * @dev: network device
11425 * Determine network driver for device.
11427 const char *netdev_drivername(const struct net_device *dev)
11429 const struct device_driver *driver;
11430 const struct device *parent;
11431 const char *empty = "";
11433 parent = dev->dev.parent;
11437 driver = parent->driver;
11438 if (driver && driver->name)
11439 return driver->name;
11443 static void __netdev_printk(const char *level, const struct net_device *dev,
11444 struct va_format *vaf)
11446 if (dev && dev->dev.parent) {
11447 dev_printk_emit(level[1] - '0',
11450 dev_driver_string(dev->dev.parent),
11451 dev_name(dev->dev.parent),
11452 netdev_name(dev), netdev_reg_state(dev),
11455 printk("%s%s%s: %pV",
11456 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11458 printk("%s(NULL net_device): %pV", level, vaf);
11462 void netdev_printk(const char *level, const struct net_device *dev,
11463 const char *format, ...)
11465 struct va_format vaf;
11468 va_start(args, format);
11473 __netdev_printk(level, dev, &vaf);
11477 EXPORT_SYMBOL(netdev_printk);
11479 #define define_netdev_printk_level(func, level) \
11480 void func(const struct net_device *dev, const char *fmt, ...) \
11482 struct va_format vaf; \
11485 va_start(args, fmt); \
11490 __netdev_printk(level, dev, &vaf); \
11494 EXPORT_SYMBOL(func);
11496 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11497 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11498 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11499 define_netdev_printk_level(netdev_err, KERN_ERR);
11500 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11501 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11502 define_netdev_printk_level(netdev_info, KERN_INFO);
11504 static void __net_exit netdev_exit(struct net *net)
11506 kfree(net->dev_name_head);
11507 kfree(net->dev_index_head);
11508 if (net != &init_net)
11509 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11512 static struct pernet_operations __net_initdata netdev_net_ops = {
11513 .init = netdev_init,
11514 .exit = netdev_exit,
11517 static void __net_exit default_device_exit(struct net *net)
11519 struct net_device *dev, *aux;
11521 * Push all migratable network devices back to the
11522 * initial network namespace
11525 for_each_netdev_safe(net, dev, aux) {
11527 char fb_name[IFNAMSIZ];
11529 /* Ignore unmoveable devices (i.e. loopback) */
11530 if (dev->features & NETIF_F_NETNS_LOCAL)
11533 /* Leave virtual devices for the generic cleanup */
11534 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11537 /* Push remaining network devices to init_net */
11538 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11539 if (netdev_name_in_use(&init_net, fb_name))
11540 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11541 err = dev_change_net_namespace(dev, &init_net, fb_name);
11543 pr_emerg("%s: failed to move %s to init_net: %d\n",
11544 __func__, dev->name, err);
11551 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11553 /* Return with the rtnl_lock held when there are no network
11554 * devices unregistering in any network namespace in net_list.
11557 bool unregistering;
11558 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11560 add_wait_queue(&netdev_unregistering_wq, &wait);
11562 unregistering = false;
11564 list_for_each_entry(net, net_list, exit_list) {
11565 if (net->dev_unreg_count > 0) {
11566 unregistering = true;
11570 if (!unregistering)
11574 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11576 remove_wait_queue(&netdev_unregistering_wq, &wait);
11579 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11581 /* At exit all network devices most be removed from a network
11582 * namespace. Do this in the reverse order of registration.
11583 * Do this across as many network namespaces as possible to
11584 * improve batching efficiency.
11586 struct net_device *dev;
11588 LIST_HEAD(dev_kill_list);
11590 /* To prevent network device cleanup code from dereferencing
11591 * loopback devices or network devices that have been freed
11592 * wait here for all pending unregistrations to complete,
11593 * before unregistring the loopback device and allowing the
11594 * network namespace be freed.
11596 * The netdev todo list containing all network devices
11597 * unregistrations that happen in default_device_exit_batch
11598 * will run in the rtnl_unlock() at the end of
11599 * default_device_exit_batch.
11601 rtnl_lock_unregistering(net_list);
11602 list_for_each_entry(net, net_list, exit_list) {
11603 for_each_netdev_reverse(net, dev) {
11604 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11605 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11607 unregister_netdevice_queue(dev, &dev_kill_list);
11610 unregister_netdevice_many(&dev_kill_list);
11614 static struct pernet_operations __net_initdata default_device_ops = {
11615 .exit = default_device_exit,
11616 .exit_batch = default_device_exit_batch,
11620 * Initialize the DEV module. At boot time this walks the device list and
11621 * unhooks any devices that fail to initialise (normally hardware not
11622 * present) and leaves us with a valid list of present and active devices.
11627 * This is called single threaded during boot, so no need
11628 * to take the rtnl semaphore.
11630 static int __init net_dev_init(void)
11632 int i, rc = -ENOMEM;
11634 BUG_ON(!dev_boot_phase);
11636 if (dev_proc_init())
11639 if (netdev_kobject_init())
11642 INIT_LIST_HEAD(&ptype_all);
11643 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11644 INIT_LIST_HEAD(&ptype_base[i]);
11646 INIT_LIST_HEAD(&offload_base);
11648 if (register_pernet_subsys(&netdev_net_ops))
11652 * Initialise the packet receive queues.
11655 for_each_possible_cpu(i) {
11656 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11657 struct softnet_data *sd = &per_cpu(softnet_data, i);
11659 INIT_WORK(flush, flush_backlog);
11661 skb_queue_head_init(&sd->input_pkt_queue);
11662 skb_queue_head_init(&sd->process_queue);
11663 #ifdef CONFIG_XFRM_OFFLOAD
11664 skb_queue_head_init(&sd->xfrm_backlog);
11666 INIT_LIST_HEAD(&sd->poll_list);
11667 sd->output_queue_tailp = &sd->output_queue;
11669 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11673 init_gro_hash(&sd->backlog);
11674 sd->backlog.poll = process_backlog;
11675 sd->backlog.weight = weight_p;
11678 dev_boot_phase = 0;
11680 /* The loopback device is special if any other network devices
11681 * is present in a network namespace the loopback device must
11682 * be present. Since we now dynamically allocate and free the
11683 * loopback device ensure this invariant is maintained by
11684 * keeping the loopback device as the first device on the
11685 * list of network devices. Ensuring the loopback devices
11686 * is the first device that appears and the last network device
11689 if (register_pernet_device(&loopback_net_ops))
11692 if (register_pernet_device(&default_device_ops))
11695 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11696 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11698 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11699 NULL, dev_cpu_dead);
11706 subsys_initcall(net_dev_init);