1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_ingress.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
154 #include "net-sysfs.h"
156 #define MAX_GRO_SKBS 8
158 /* This should be increased if a protocol with a bigger head is added. */
159 #define GRO_MAX_HEAD (MAX_HEADER + 128)
161 static DEFINE_SPINLOCK(ptype_lock);
162 static DEFINE_SPINLOCK(offload_lock);
163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
164 struct list_head ptype_all __read_mostly; /* Taps */
165 static struct list_head offload_base __read_mostly;
167 static int netif_rx_internal(struct sk_buff *skb);
168 static int call_netdevice_notifiers_info(unsigned long val,
169 struct netdev_notifier_info *info);
170 static int call_netdevice_notifiers_extack(unsigned long val,
171 struct net_device *dev,
172 struct netlink_ext_ack *extack);
173 static struct napi_struct *napi_by_id(unsigned int napi_id);
176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base_head list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 static DEFINE_MUTEX(ifalias_mutex);
199 /* protects napi_hash addition/deletion and napi_gen_id */
200 static DEFINE_SPINLOCK(napi_hash_lock);
202 static unsigned int napi_gen_id = NR_CPUS;
203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
205 static DECLARE_RWSEM(devnet_rename_sem);
207 static inline void dev_base_seq_inc(struct net *net)
209 while (++net->dev_base_seq == 0)
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
225 static inline void rps_lock(struct softnet_data *sd)
228 spin_lock(&sd->input_pkt_queue.lock);
232 static inline void rps_unlock(struct softnet_data *sd)
235 spin_unlock(&sd->input_pkt_queue.lock);
239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
242 struct netdev_name_node *name_node;
244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
247 INIT_HLIST_NODE(&name_node->hlist);
248 name_node->dev = dev;
249 name_node->name = name;
253 static struct netdev_name_node *
254 netdev_name_node_head_alloc(struct net_device *dev)
256 struct netdev_name_node *name_node;
258 name_node = netdev_name_node_alloc(dev, dev->name);
261 INIT_LIST_HEAD(&name_node->list);
265 static void netdev_name_node_free(struct netdev_name_node *name_node)
270 static void netdev_name_node_add(struct net *net,
271 struct netdev_name_node *name_node)
273 hlist_add_head_rcu(&name_node->hlist,
274 dev_name_hash(net, name_node->name));
277 static void netdev_name_node_del(struct netdev_name_node *name_node)
279 hlist_del_rcu(&name_node->hlist);
282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
285 struct hlist_head *head = dev_name_hash(net, name);
286 struct netdev_name_node *name_node;
288 hlist_for_each_entry(name_node, head, hlist)
289 if (!strcmp(name_node->name, name))
294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
297 struct hlist_head *head = dev_name_hash(net, name);
298 struct netdev_name_node *name_node;
300 hlist_for_each_entry_rcu(name_node, head, hlist)
301 if (!strcmp(name_node->name, name))
306 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
308 struct netdev_name_node *name_node;
309 struct net *net = dev_net(dev);
311 name_node = netdev_name_node_lookup(net, name);
314 name_node = netdev_name_node_alloc(dev, name);
317 netdev_name_node_add(net, name_node);
318 /* The node that holds dev->name acts as a head of per-device list. */
319 list_add_tail(&name_node->list, &dev->name_node->list);
323 EXPORT_SYMBOL(netdev_name_node_alt_create);
325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
327 list_del(&name_node->list);
328 netdev_name_node_del(name_node);
329 kfree(name_node->name);
330 netdev_name_node_free(name_node);
333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
335 struct netdev_name_node *name_node;
336 struct net *net = dev_net(dev);
338 name_node = netdev_name_node_lookup(net, name);
341 /* lookup might have found our primary name or a name belonging
344 if (name_node == dev->name_node || name_node->dev != dev)
347 __netdev_name_node_alt_destroy(name_node);
351 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
353 static void netdev_name_node_alt_flush(struct net_device *dev)
355 struct netdev_name_node *name_node, *tmp;
357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
358 __netdev_name_node_alt_destroy(name_node);
361 /* Device list insertion */
362 static void list_netdevice(struct net_device *dev)
364 struct net *net = dev_net(dev);
368 write_lock(&dev_base_lock);
369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
370 netdev_name_node_add(net, dev->name_node);
371 hlist_add_head_rcu(&dev->index_hlist,
372 dev_index_hash(net, dev->ifindex));
373 write_unlock(&dev_base_lock);
375 dev_base_seq_inc(net);
378 /* Device list removal
379 * caller must respect a RCU grace period before freeing/reusing dev
381 static void unlist_netdevice(struct net_device *dev, bool lock)
385 /* Unlink dev from the device chain */
387 write_lock(&dev_base_lock);
388 list_del_rcu(&dev->dev_list);
389 netdev_name_node_del(dev->name_node);
390 hlist_del_rcu(&dev->index_hlist);
392 write_unlock(&dev_base_lock);
394 dev_base_seq_inc(dev_net(dev));
401 static RAW_NOTIFIER_HEAD(netdev_chain);
404 * Device drivers call our routines to queue packets here. We empty the
405 * queue in the local softnet handler.
408 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
409 EXPORT_PER_CPU_SYMBOL(softnet_data);
411 #ifdef CONFIG_LOCKDEP
413 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
414 * according to dev->type
416 static const unsigned short netdev_lock_type[] = {
417 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
418 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
419 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
420 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
421 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
422 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
423 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
424 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
425 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
426 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
427 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
428 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
429 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
430 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
431 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
433 static const char *const netdev_lock_name[] = {
434 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
435 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
436 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
437 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
438 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
439 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
440 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
441 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
442 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
443 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
444 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
445 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
446 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
447 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
448 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
450 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
451 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
453 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
457 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
458 if (netdev_lock_type[i] == dev_type)
460 /* the last key is used by default */
461 return ARRAY_SIZE(netdev_lock_type) - 1;
464 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
465 unsigned short dev_type)
469 i = netdev_lock_pos(dev_type);
470 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
471 netdev_lock_name[i]);
474 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
478 i = netdev_lock_pos(dev->type);
479 lockdep_set_class_and_name(&dev->addr_list_lock,
480 &netdev_addr_lock_key[i],
481 netdev_lock_name[i]);
484 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
485 unsigned short dev_type)
489 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
494 /*******************************************************************************
496 * Protocol management and registration routines
498 *******************************************************************************/
502 * Add a protocol ID to the list. Now that the input handler is
503 * smarter we can dispense with all the messy stuff that used to be
506 * BEWARE!!! Protocol handlers, mangling input packets,
507 * MUST BE last in hash buckets and checking protocol handlers
508 * MUST start from promiscuous ptype_all chain in net_bh.
509 * It is true now, do not change it.
510 * Explanation follows: if protocol handler, mangling packet, will
511 * be the first on list, it is not able to sense, that packet
512 * is cloned and should be copied-on-write, so that it will
513 * change it and subsequent readers will get broken packet.
517 static inline struct list_head *ptype_head(const struct packet_type *pt)
519 if (pt->type == htons(ETH_P_ALL))
520 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
522 return pt->dev ? &pt->dev->ptype_specific :
523 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
527 * dev_add_pack - add packet handler
528 * @pt: packet type declaration
530 * Add a protocol handler to the networking stack. The passed &packet_type
531 * is linked into kernel lists and may not be freed until it has been
532 * removed from the kernel lists.
534 * This call does not sleep therefore it can not
535 * guarantee all CPU's that are in middle of receiving packets
536 * will see the new packet type (until the next received packet).
539 void dev_add_pack(struct packet_type *pt)
541 struct list_head *head = ptype_head(pt);
543 spin_lock(&ptype_lock);
544 list_add_rcu(&pt->list, head);
545 spin_unlock(&ptype_lock);
547 EXPORT_SYMBOL(dev_add_pack);
550 * __dev_remove_pack - remove packet handler
551 * @pt: packet type declaration
553 * Remove a protocol handler that was previously added to the kernel
554 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
555 * from the kernel lists and can be freed or reused once this function
558 * The packet type might still be in use by receivers
559 * and must not be freed until after all the CPU's have gone
560 * through a quiescent state.
562 void __dev_remove_pack(struct packet_type *pt)
564 struct list_head *head = ptype_head(pt);
565 struct packet_type *pt1;
567 spin_lock(&ptype_lock);
569 list_for_each_entry(pt1, head, list) {
571 list_del_rcu(&pt->list);
576 pr_warn("dev_remove_pack: %p not found\n", pt);
578 spin_unlock(&ptype_lock);
580 EXPORT_SYMBOL(__dev_remove_pack);
583 * dev_remove_pack - remove packet handler
584 * @pt: packet type declaration
586 * Remove a protocol handler that was previously added to the kernel
587 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
588 * from the kernel lists and can be freed or reused once this function
591 * This call sleeps to guarantee that no CPU is looking at the packet
594 void dev_remove_pack(struct packet_type *pt)
596 __dev_remove_pack(pt);
600 EXPORT_SYMBOL(dev_remove_pack);
604 * dev_add_offload - register offload handlers
605 * @po: protocol offload declaration
607 * Add protocol offload handlers to the networking stack. The passed
608 * &proto_offload is linked into kernel lists and may not be freed until
609 * it has been removed from the kernel lists.
611 * This call does not sleep therefore it can not
612 * guarantee all CPU's that are in middle of receiving packets
613 * will see the new offload handlers (until the next received packet).
615 void dev_add_offload(struct packet_offload *po)
617 struct packet_offload *elem;
619 spin_lock(&offload_lock);
620 list_for_each_entry(elem, &offload_base, list) {
621 if (po->priority < elem->priority)
624 list_add_rcu(&po->list, elem->list.prev);
625 spin_unlock(&offload_lock);
627 EXPORT_SYMBOL(dev_add_offload);
630 * __dev_remove_offload - remove offload handler
631 * @po: packet offload declaration
633 * Remove a protocol offload handler that was previously added to the
634 * kernel offload handlers by dev_add_offload(). The passed &offload_type
635 * is removed from the kernel lists and can be freed or reused once this
638 * The packet type might still be in use by receivers
639 * and must not be freed until after all the CPU's have gone
640 * through a quiescent state.
642 static void __dev_remove_offload(struct packet_offload *po)
644 struct list_head *head = &offload_base;
645 struct packet_offload *po1;
647 spin_lock(&offload_lock);
649 list_for_each_entry(po1, head, list) {
651 list_del_rcu(&po->list);
656 pr_warn("dev_remove_offload: %p not found\n", po);
658 spin_unlock(&offload_lock);
662 * dev_remove_offload - remove packet offload handler
663 * @po: packet offload declaration
665 * Remove a packet offload handler that was previously added to the kernel
666 * offload handlers by dev_add_offload(). The passed &offload_type is
667 * removed from the kernel lists and can be freed or reused once this
670 * This call sleeps to guarantee that no CPU is looking at the packet
673 void dev_remove_offload(struct packet_offload *po)
675 __dev_remove_offload(po);
679 EXPORT_SYMBOL(dev_remove_offload);
681 /*******************************************************************************
683 * Device Interface Subroutines
685 *******************************************************************************/
688 * dev_get_iflink - get 'iflink' value of a interface
689 * @dev: targeted interface
691 * Indicates the ifindex the interface is linked to.
692 * Physical interfaces have the same 'ifindex' and 'iflink' values.
695 int dev_get_iflink(const struct net_device *dev)
697 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
698 return dev->netdev_ops->ndo_get_iflink(dev);
702 EXPORT_SYMBOL(dev_get_iflink);
705 * dev_fill_metadata_dst - Retrieve tunnel egress information.
706 * @dev: targeted interface
709 * For better visibility of tunnel traffic OVS needs to retrieve
710 * egress tunnel information for a packet. Following API allows
711 * user to get this info.
713 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
715 struct ip_tunnel_info *info;
717 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
720 info = skb_tunnel_info_unclone(skb);
723 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
726 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
728 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
730 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
732 int k = stack->num_paths++;
734 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
737 return &stack->path[k];
740 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
741 struct net_device_path_stack *stack)
743 const struct net_device *last_dev;
744 struct net_device_path_ctx ctx = {
747 struct net_device_path *path;
750 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
751 stack->num_paths = 0;
752 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
754 path = dev_fwd_path(stack);
758 memset(path, 0, sizeof(struct net_device_path));
759 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
763 if (WARN_ON_ONCE(last_dev == ctx.dev))
766 path = dev_fwd_path(stack);
769 path->type = DEV_PATH_ETHERNET;
774 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
777 * __dev_get_by_name - find a device by its name
778 * @net: the applicable net namespace
779 * @name: name to find
781 * Find an interface by name. Must be called under RTNL semaphore
782 * or @dev_base_lock. If the name is found a pointer to the device
783 * is returned. If the name is not found then %NULL is returned. The
784 * reference counters are not incremented so the caller must be
785 * careful with locks.
788 struct net_device *__dev_get_by_name(struct net *net, const char *name)
790 struct netdev_name_node *node_name;
792 node_name = netdev_name_node_lookup(net, name);
793 return node_name ? node_name->dev : NULL;
795 EXPORT_SYMBOL(__dev_get_by_name);
798 * dev_get_by_name_rcu - find a device by its name
799 * @net: the applicable net namespace
800 * @name: name to find
802 * Find an interface by name.
803 * If the name is found a pointer to the device is returned.
804 * If the name is not found then %NULL is returned.
805 * The reference counters are not incremented so the caller must be
806 * careful with locks. The caller must hold RCU lock.
809 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
811 struct netdev_name_node *node_name;
813 node_name = netdev_name_node_lookup_rcu(net, name);
814 return node_name ? node_name->dev : NULL;
816 EXPORT_SYMBOL(dev_get_by_name_rcu);
819 * dev_get_by_name - find a device by its name
820 * @net: the applicable net namespace
821 * @name: name to find
823 * Find an interface by name. This can be called from any
824 * context and does its own locking. The returned handle has
825 * the usage count incremented and the caller must use dev_put() to
826 * release it when it is no longer needed. %NULL is returned if no
827 * matching device is found.
830 struct net_device *dev_get_by_name(struct net *net, const char *name)
832 struct net_device *dev;
835 dev = dev_get_by_name_rcu(net, name);
840 EXPORT_SYMBOL(dev_get_by_name);
843 * __dev_get_by_index - find a device by its ifindex
844 * @net: the applicable net namespace
845 * @ifindex: index of device
847 * Search for an interface by index. Returns %NULL if the device
848 * is not found or a pointer to the device. The device has not
849 * had its reference counter increased so the caller must be careful
850 * about locking. The caller must hold either the RTNL semaphore
854 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
856 struct net_device *dev;
857 struct hlist_head *head = dev_index_hash(net, ifindex);
859 hlist_for_each_entry(dev, head, index_hlist)
860 if (dev->ifindex == ifindex)
865 EXPORT_SYMBOL(__dev_get_by_index);
868 * dev_get_by_index_rcu - find a device by its ifindex
869 * @net: the applicable net namespace
870 * @ifindex: index of device
872 * Search for an interface by index. Returns %NULL if the device
873 * is not found or a pointer to the device. The device has not
874 * had its reference counter increased so the caller must be careful
875 * about locking. The caller must hold RCU lock.
878 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
880 struct net_device *dev;
881 struct hlist_head *head = dev_index_hash(net, ifindex);
883 hlist_for_each_entry_rcu(dev, head, index_hlist)
884 if (dev->ifindex == ifindex)
889 EXPORT_SYMBOL(dev_get_by_index_rcu);
893 * dev_get_by_index - find a device by its ifindex
894 * @net: the applicable net namespace
895 * @ifindex: index of device
897 * Search for an interface by index. Returns NULL if the device
898 * is not found or a pointer to the device. The device returned has
899 * had a reference added and the pointer is safe until the user calls
900 * dev_put to indicate they have finished with it.
903 struct net_device *dev_get_by_index(struct net *net, int ifindex)
905 struct net_device *dev;
908 dev = dev_get_by_index_rcu(net, ifindex);
913 EXPORT_SYMBOL(dev_get_by_index);
916 * dev_get_by_napi_id - find a device by napi_id
917 * @napi_id: ID of the NAPI struct
919 * Search for an interface by NAPI ID. Returns %NULL if the device
920 * is not found or a pointer to the device. The device has not had
921 * its reference counter increased so the caller must be careful
922 * about locking. The caller must hold RCU lock.
925 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
927 struct napi_struct *napi;
929 WARN_ON_ONCE(!rcu_read_lock_held());
931 if (napi_id < MIN_NAPI_ID)
934 napi = napi_by_id(napi_id);
936 return napi ? napi->dev : NULL;
938 EXPORT_SYMBOL(dev_get_by_napi_id);
941 * netdev_get_name - get a netdevice name, knowing its ifindex.
942 * @net: network namespace
943 * @name: a pointer to the buffer where the name will be stored.
944 * @ifindex: the ifindex of the interface to get the name from.
946 int netdev_get_name(struct net *net, char *name, int ifindex)
948 struct net_device *dev;
951 down_read(&devnet_rename_sem);
954 dev = dev_get_by_index_rcu(net, ifindex);
960 strcpy(name, dev->name);
965 up_read(&devnet_rename_sem);
970 * dev_getbyhwaddr_rcu - find a device by its hardware address
971 * @net: the applicable net namespace
972 * @type: media type of device
973 * @ha: hardware address
975 * Search for an interface by MAC address. Returns NULL if the device
976 * is not found or a pointer to the device.
977 * The caller must hold RCU or RTNL.
978 * The returned device has not had its ref count increased
979 * and the caller must therefore be careful about locking
983 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
986 struct net_device *dev;
988 for_each_netdev_rcu(net, dev)
989 if (dev->type == type &&
990 !memcmp(dev->dev_addr, ha, dev->addr_len))
995 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
997 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
999 struct net_device *dev, *ret = NULL;
1002 for_each_netdev_rcu(net, dev)
1003 if (dev->type == type) {
1011 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1014 * __dev_get_by_flags - find any device with given flags
1015 * @net: the applicable net namespace
1016 * @if_flags: IFF_* values
1017 * @mask: bitmask of bits in if_flags to check
1019 * Search for any interface with the given flags. Returns NULL if a device
1020 * is not found or a pointer to the device. Must be called inside
1021 * rtnl_lock(), and result refcount is unchanged.
1024 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1025 unsigned short mask)
1027 struct net_device *dev, *ret;
1032 for_each_netdev(net, dev) {
1033 if (((dev->flags ^ if_flags) & mask) == 0) {
1040 EXPORT_SYMBOL(__dev_get_by_flags);
1043 * dev_valid_name - check if name is okay for network device
1044 * @name: name string
1046 * Network device names need to be valid file names to
1047 * allow sysfs to work. We also disallow any kind of
1050 bool dev_valid_name(const char *name)
1054 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1056 if (!strcmp(name, ".") || !strcmp(name, ".."))
1060 if (*name == '/' || *name == ':' || isspace(*name))
1066 EXPORT_SYMBOL(dev_valid_name);
1069 * __dev_alloc_name - allocate a name for a device
1070 * @net: network namespace to allocate the device name in
1071 * @name: name format string
1072 * @buf: scratch buffer and result name string
1074 * Passed a format string - eg "lt%d" it will try and find a suitable
1075 * id. It scans list of devices to build up a free map, then chooses
1076 * the first empty slot. The caller must hold the dev_base or rtnl lock
1077 * while allocating the name and adding the device in order to avoid
1079 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1080 * Returns the number of the unit assigned or a negative errno code.
1083 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1087 const int max_netdevices = 8*PAGE_SIZE;
1088 unsigned long *inuse;
1089 struct net_device *d;
1091 if (!dev_valid_name(name))
1094 p = strchr(name, '%');
1097 * Verify the string as this thing may have come from
1098 * the user. There must be either one "%d" and no other "%"
1101 if (p[1] != 'd' || strchr(p + 2, '%'))
1104 /* Use one page as a bit array of possible slots */
1105 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1109 for_each_netdev(net, d) {
1110 struct netdev_name_node *name_node;
1111 list_for_each_entry(name_node, &d->name_node->list, list) {
1112 if (!sscanf(name_node->name, name, &i))
1114 if (i < 0 || i >= max_netdevices)
1117 /* avoid cases where sscanf is not exact inverse of printf */
1118 snprintf(buf, IFNAMSIZ, name, i);
1119 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1122 if (!sscanf(d->name, name, &i))
1124 if (i < 0 || i >= max_netdevices)
1127 /* avoid cases where sscanf is not exact inverse of printf */
1128 snprintf(buf, IFNAMSIZ, name, i);
1129 if (!strncmp(buf, d->name, IFNAMSIZ))
1133 i = find_first_zero_bit(inuse, max_netdevices);
1134 free_page((unsigned long) inuse);
1137 snprintf(buf, IFNAMSIZ, name, i);
1138 if (!__dev_get_by_name(net, buf))
1141 /* It is possible to run out of possible slots
1142 * when the name is long and there isn't enough space left
1143 * for the digits, or if all bits are used.
1148 static int dev_alloc_name_ns(struct net *net,
1149 struct net_device *dev,
1156 ret = __dev_alloc_name(net, name, buf);
1158 strlcpy(dev->name, buf, IFNAMSIZ);
1163 * dev_alloc_name - allocate a name for a device
1165 * @name: name format string
1167 * Passed a format string - eg "lt%d" it will try and find a suitable
1168 * id. It scans list of devices to build up a free map, then chooses
1169 * the first empty slot. The caller must hold the dev_base or rtnl lock
1170 * while allocating the name and adding the device in order to avoid
1172 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1173 * Returns the number of the unit assigned or a negative errno code.
1176 int dev_alloc_name(struct net_device *dev, const char *name)
1178 return dev_alloc_name_ns(dev_net(dev), dev, name);
1180 EXPORT_SYMBOL(dev_alloc_name);
1182 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1187 if (!dev_valid_name(name))
1190 if (strchr(name, '%'))
1191 return dev_alloc_name_ns(net, dev, name);
1192 else if (__dev_get_by_name(net, name))
1194 else if (dev->name != name)
1195 strlcpy(dev->name, name, IFNAMSIZ);
1201 * dev_change_name - change name of a device
1203 * @newname: name (or format string) must be at least IFNAMSIZ
1205 * Change name of a device, can pass format strings "eth%d".
1208 int dev_change_name(struct net_device *dev, const char *newname)
1210 unsigned char old_assign_type;
1211 char oldname[IFNAMSIZ];
1217 BUG_ON(!dev_net(dev));
1221 /* Some auto-enslaved devices e.g. failover slaves are
1222 * special, as userspace might rename the device after
1223 * the interface had been brought up and running since
1224 * the point kernel initiated auto-enslavement. Allow
1225 * live name change even when these slave devices are
1228 * Typically, users of these auto-enslaving devices
1229 * don't actually care about slave name change, as
1230 * they are supposed to operate on master interface
1233 if (dev->flags & IFF_UP &&
1234 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1237 down_write(&devnet_rename_sem);
1239 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1240 up_write(&devnet_rename_sem);
1244 memcpy(oldname, dev->name, IFNAMSIZ);
1246 err = dev_get_valid_name(net, dev, newname);
1248 up_write(&devnet_rename_sem);
1252 if (oldname[0] && !strchr(oldname, '%'))
1253 netdev_info(dev, "renamed from %s\n", oldname);
1255 old_assign_type = dev->name_assign_type;
1256 dev->name_assign_type = NET_NAME_RENAMED;
1259 ret = device_rename(&dev->dev, dev->name);
1261 memcpy(dev->name, oldname, IFNAMSIZ);
1262 dev->name_assign_type = old_assign_type;
1263 up_write(&devnet_rename_sem);
1267 up_write(&devnet_rename_sem);
1269 netdev_adjacent_rename_links(dev, oldname);
1271 write_lock(&dev_base_lock);
1272 netdev_name_node_del(dev->name_node);
1273 write_unlock(&dev_base_lock);
1277 write_lock(&dev_base_lock);
1278 netdev_name_node_add(net, dev->name_node);
1279 write_unlock(&dev_base_lock);
1281 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1282 ret = notifier_to_errno(ret);
1285 /* err >= 0 after dev_alloc_name() or stores the first errno */
1288 down_write(&devnet_rename_sem);
1289 memcpy(dev->name, oldname, IFNAMSIZ);
1290 memcpy(oldname, newname, IFNAMSIZ);
1291 dev->name_assign_type = old_assign_type;
1292 old_assign_type = NET_NAME_RENAMED;
1295 pr_err("%s: name change rollback failed: %d\n",
1304 * dev_set_alias - change ifalias of a device
1306 * @alias: name up to IFALIASZ
1307 * @len: limit of bytes to copy from info
1309 * Set ifalias for a device,
1311 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1313 struct dev_ifalias *new_alias = NULL;
1315 if (len >= IFALIASZ)
1319 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1323 memcpy(new_alias->ifalias, alias, len);
1324 new_alias->ifalias[len] = 0;
1327 mutex_lock(&ifalias_mutex);
1328 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1329 mutex_is_locked(&ifalias_mutex));
1330 mutex_unlock(&ifalias_mutex);
1333 kfree_rcu(new_alias, rcuhead);
1337 EXPORT_SYMBOL(dev_set_alias);
1340 * dev_get_alias - get ifalias of a device
1342 * @name: buffer to store name of ifalias
1343 * @len: size of buffer
1345 * get ifalias for a device. Caller must make sure dev cannot go
1346 * away, e.g. rcu read lock or own a reference count to device.
1348 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1350 const struct dev_ifalias *alias;
1354 alias = rcu_dereference(dev->ifalias);
1356 ret = snprintf(name, len, "%s", alias->ifalias);
1363 * netdev_features_change - device changes features
1364 * @dev: device to cause notification
1366 * Called to indicate a device has changed features.
1368 void netdev_features_change(struct net_device *dev)
1370 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1372 EXPORT_SYMBOL(netdev_features_change);
1375 * netdev_state_change - device changes state
1376 * @dev: device to cause notification
1378 * Called to indicate a device has changed state. This function calls
1379 * the notifier chains for netdev_chain and sends a NEWLINK message
1380 * to the routing socket.
1382 void netdev_state_change(struct net_device *dev)
1384 if (dev->flags & IFF_UP) {
1385 struct netdev_notifier_change_info change_info = {
1389 call_netdevice_notifiers_info(NETDEV_CHANGE,
1391 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1394 EXPORT_SYMBOL(netdev_state_change);
1397 * __netdev_notify_peers - notify network peers about existence of @dev,
1398 * to be called when rtnl lock is already held.
1399 * @dev: network device
1401 * Generate traffic such that interested network peers are aware of
1402 * @dev, such as by generating a gratuitous ARP. This may be used when
1403 * a device wants to inform the rest of the network about some sort of
1404 * reconfiguration such as a failover event or virtual machine
1407 void __netdev_notify_peers(struct net_device *dev)
1410 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1411 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1413 EXPORT_SYMBOL(__netdev_notify_peers);
1416 * netdev_notify_peers - notify network peers about existence of @dev
1417 * @dev: network device
1419 * Generate traffic such that interested network peers are aware of
1420 * @dev, such as by generating a gratuitous ARP. This may be used when
1421 * a device wants to inform the rest of the network about some sort of
1422 * reconfiguration such as a failover event or virtual machine
1425 void netdev_notify_peers(struct net_device *dev)
1428 __netdev_notify_peers(dev);
1431 EXPORT_SYMBOL(netdev_notify_peers);
1433 static int napi_threaded_poll(void *data);
1435 static int napi_kthread_create(struct napi_struct *n)
1439 /* Create and wake up the kthread once to put it in
1440 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1441 * warning and work with loadavg.
1443 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1444 n->dev->name, n->napi_id);
1445 if (IS_ERR(n->thread)) {
1446 err = PTR_ERR(n->thread);
1447 pr_err("kthread_run failed with err %d\n", err);
1454 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1456 const struct net_device_ops *ops = dev->netdev_ops;
1461 if (!netif_device_present(dev)) {
1462 /* may be detached because parent is runtime-suspended */
1463 if (dev->dev.parent)
1464 pm_runtime_resume(dev->dev.parent);
1465 if (!netif_device_present(dev))
1469 /* Block netpoll from trying to do any rx path servicing.
1470 * If we don't do this there is a chance ndo_poll_controller
1471 * or ndo_poll may be running while we open the device
1473 netpoll_poll_disable(dev);
1475 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1476 ret = notifier_to_errno(ret);
1480 set_bit(__LINK_STATE_START, &dev->state);
1482 if (ops->ndo_validate_addr)
1483 ret = ops->ndo_validate_addr(dev);
1485 if (!ret && ops->ndo_open)
1486 ret = ops->ndo_open(dev);
1488 netpoll_poll_enable(dev);
1491 clear_bit(__LINK_STATE_START, &dev->state);
1493 dev->flags |= IFF_UP;
1494 dev_set_rx_mode(dev);
1496 add_device_randomness(dev->dev_addr, dev->addr_len);
1503 * dev_open - prepare an interface for use.
1504 * @dev: device to open
1505 * @extack: netlink extended ack
1507 * Takes a device from down to up state. The device's private open
1508 * function is invoked and then the multicast lists are loaded. Finally
1509 * the device is moved into the up state and a %NETDEV_UP message is
1510 * sent to the netdev notifier chain.
1512 * Calling this function on an active interface is a nop. On a failure
1513 * a negative errno code is returned.
1515 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1519 if (dev->flags & IFF_UP)
1522 ret = __dev_open(dev, extack);
1526 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1527 call_netdevice_notifiers(NETDEV_UP, dev);
1531 EXPORT_SYMBOL(dev_open);
1533 static void __dev_close_many(struct list_head *head)
1535 struct net_device *dev;
1540 list_for_each_entry(dev, head, close_list) {
1541 /* Temporarily disable netpoll until the interface is down */
1542 netpoll_poll_disable(dev);
1544 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1546 clear_bit(__LINK_STATE_START, &dev->state);
1548 /* Synchronize to scheduled poll. We cannot touch poll list, it
1549 * can be even on different cpu. So just clear netif_running().
1551 * dev->stop() will invoke napi_disable() on all of it's
1552 * napi_struct instances on this device.
1554 smp_mb__after_atomic(); /* Commit netif_running(). */
1557 dev_deactivate_many(head);
1559 list_for_each_entry(dev, head, close_list) {
1560 const struct net_device_ops *ops = dev->netdev_ops;
1563 * Call the device specific close. This cannot fail.
1564 * Only if device is UP
1566 * We allow it to be called even after a DETACH hot-plug
1572 dev->flags &= ~IFF_UP;
1573 netpoll_poll_enable(dev);
1577 static void __dev_close(struct net_device *dev)
1581 list_add(&dev->close_list, &single);
1582 __dev_close_many(&single);
1586 void dev_close_many(struct list_head *head, bool unlink)
1588 struct net_device *dev, *tmp;
1590 /* Remove the devices that don't need to be closed */
1591 list_for_each_entry_safe(dev, tmp, head, close_list)
1592 if (!(dev->flags & IFF_UP))
1593 list_del_init(&dev->close_list);
1595 __dev_close_many(head);
1597 list_for_each_entry_safe(dev, tmp, head, close_list) {
1598 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1599 call_netdevice_notifiers(NETDEV_DOWN, dev);
1601 list_del_init(&dev->close_list);
1604 EXPORT_SYMBOL(dev_close_many);
1607 * dev_close - shutdown an interface.
1608 * @dev: device to shutdown
1610 * This function moves an active device into down state. A
1611 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1612 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1615 void dev_close(struct net_device *dev)
1617 if (dev->flags & IFF_UP) {
1620 list_add(&dev->close_list, &single);
1621 dev_close_many(&single, true);
1625 EXPORT_SYMBOL(dev_close);
1629 * dev_disable_lro - disable Large Receive Offload on a device
1632 * Disable Large Receive Offload (LRO) on a net device. Must be
1633 * called under RTNL. This is needed if received packets may be
1634 * forwarded to another interface.
1636 void dev_disable_lro(struct net_device *dev)
1638 struct net_device *lower_dev;
1639 struct list_head *iter;
1641 dev->wanted_features &= ~NETIF_F_LRO;
1642 netdev_update_features(dev);
1644 if (unlikely(dev->features & NETIF_F_LRO))
1645 netdev_WARN(dev, "failed to disable LRO!\n");
1647 netdev_for_each_lower_dev(dev, lower_dev, iter)
1648 dev_disable_lro(lower_dev);
1650 EXPORT_SYMBOL(dev_disable_lro);
1653 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1656 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1657 * called under RTNL. This is needed if Generic XDP is installed on
1660 static void dev_disable_gro_hw(struct net_device *dev)
1662 dev->wanted_features &= ~NETIF_F_GRO_HW;
1663 netdev_update_features(dev);
1665 if (unlikely(dev->features & NETIF_F_GRO_HW))
1666 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1669 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1672 case NETDEV_##val: \
1673 return "NETDEV_" __stringify(val);
1675 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1676 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1677 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1678 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1679 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1680 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1681 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1682 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1683 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1687 return "UNKNOWN_NETDEV_EVENT";
1689 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1691 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1692 struct net_device *dev)
1694 struct netdev_notifier_info info = {
1698 return nb->notifier_call(nb, val, &info);
1701 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1702 struct net_device *dev)
1706 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1707 err = notifier_to_errno(err);
1711 if (!(dev->flags & IFF_UP))
1714 call_netdevice_notifier(nb, NETDEV_UP, dev);
1718 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1719 struct net_device *dev)
1721 if (dev->flags & IFF_UP) {
1722 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1724 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1726 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1729 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1732 struct net_device *dev;
1735 for_each_netdev(net, dev) {
1736 err = call_netdevice_register_notifiers(nb, dev);
1743 for_each_netdev_continue_reverse(net, dev)
1744 call_netdevice_unregister_notifiers(nb, dev);
1748 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1751 struct net_device *dev;
1753 for_each_netdev(net, dev)
1754 call_netdevice_unregister_notifiers(nb, dev);
1757 static int dev_boot_phase = 1;
1760 * register_netdevice_notifier - register a network notifier block
1763 * Register a notifier to be called when network device events occur.
1764 * The notifier passed is linked into the kernel structures and must
1765 * not be reused until it has been unregistered. A negative errno code
1766 * is returned on a failure.
1768 * When registered all registration and up events are replayed
1769 * to the new notifier to allow device to have a race free
1770 * view of the network device list.
1773 int register_netdevice_notifier(struct notifier_block *nb)
1778 /* Close race with setup_net() and cleanup_net() */
1779 down_write(&pernet_ops_rwsem);
1781 err = raw_notifier_chain_register(&netdev_chain, nb);
1787 err = call_netdevice_register_net_notifiers(nb, net);
1794 up_write(&pernet_ops_rwsem);
1798 for_each_net_continue_reverse(net)
1799 call_netdevice_unregister_net_notifiers(nb, net);
1801 raw_notifier_chain_unregister(&netdev_chain, nb);
1804 EXPORT_SYMBOL(register_netdevice_notifier);
1807 * unregister_netdevice_notifier - unregister a network notifier block
1810 * Unregister a notifier previously registered by
1811 * register_netdevice_notifier(). The notifier is unlinked into the
1812 * kernel structures and may then be reused. A negative errno code
1813 * is returned on a failure.
1815 * After unregistering unregister and down device events are synthesized
1816 * for all devices on the device list to the removed notifier to remove
1817 * the need for special case cleanup code.
1820 int unregister_netdevice_notifier(struct notifier_block *nb)
1825 /* Close race with setup_net() and cleanup_net() */
1826 down_write(&pernet_ops_rwsem);
1828 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1833 call_netdevice_unregister_net_notifiers(nb, net);
1837 up_write(&pernet_ops_rwsem);
1840 EXPORT_SYMBOL(unregister_netdevice_notifier);
1842 static int __register_netdevice_notifier_net(struct net *net,
1843 struct notifier_block *nb,
1844 bool ignore_call_fail)
1848 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1854 err = call_netdevice_register_net_notifiers(nb, net);
1855 if (err && !ignore_call_fail)
1856 goto chain_unregister;
1861 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1865 static int __unregister_netdevice_notifier_net(struct net *net,
1866 struct notifier_block *nb)
1870 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1874 call_netdevice_unregister_net_notifiers(nb, net);
1879 * register_netdevice_notifier_net - register a per-netns network notifier block
1880 * @net: network namespace
1883 * Register a notifier to be called when network device events occur.
1884 * The notifier passed is linked into the kernel structures and must
1885 * not be reused until it has been unregistered. A negative errno code
1886 * is returned on a failure.
1888 * When registered all registration and up events are replayed
1889 * to the new notifier to allow device to have a race free
1890 * view of the network device list.
1893 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1898 err = __register_netdevice_notifier_net(net, nb, false);
1902 EXPORT_SYMBOL(register_netdevice_notifier_net);
1905 * unregister_netdevice_notifier_net - unregister a per-netns
1906 * network notifier block
1907 * @net: network namespace
1910 * Unregister a notifier previously registered by
1911 * register_netdevice_notifier(). The notifier is unlinked into the
1912 * kernel structures and may then be reused. A negative errno code
1913 * is returned on a failure.
1915 * After unregistering unregister and down device events are synthesized
1916 * for all devices on the device list to the removed notifier to remove
1917 * the need for special case cleanup code.
1920 int unregister_netdevice_notifier_net(struct net *net,
1921 struct notifier_block *nb)
1926 err = __unregister_netdevice_notifier_net(net, nb);
1930 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1932 int register_netdevice_notifier_dev_net(struct net_device *dev,
1933 struct notifier_block *nb,
1934 struct netdev_net_notifier *nn)
1939 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1942 list_add(&nn->list, &dev->net_notifier_list);
1947 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1949 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1950 struct notifier_block *nb,
1951 struct netdev_net_notifier *nn)
1956 list_del(&nn->list);
1957 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1961 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1963 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1966 struct netdev_net_notifier *nn;
1968 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1969 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1970 __register_netdevice_notifier_net(net, nn->nb, true);
1975 * call_netdevice_notifiers_info - call all network notifier blocks
1976 * @val: value passed unmodified to notifier function
1977 * @info: notifier information data
1979 * Call all network notifier blocks. Parameters and return value
1980 * are as for raw_notifier_call_chain().
1983 static int call_netdevice_notifiers_info(unsigned long val,
1984 struct netdev_notifier_info *info)
1986 struct net *net = dev_net(info->dev);
1991 /* Run per-netns notifier block chain first, then run the global one.
1992 * Hopefully, one day, the global one is going to be removed after
1993 * all notifier block registrators get converted to be per-netns.
1995 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1996 if (ret & NOTIFY_STOP_MASK)
1998 return raw_notifier_call_chain(&netdev_chain, val, info);
2001 static int call_netdevice_notifiers_extack(unsigned long val,
2002 struct net_device *dev,
2003 struct netlink_ext_ack *extack)
2005 struct netdev_notifier_info info = {
2010 return call_netdevice_notifiers_info(val, &info);
2014 * call_netdevice_notifiers - call all network notifier blocks
2015 * @val: value passed unmodified to notifier function
2016 * @dev: net_device pointer passed unmodified to notifier function
2018 * Call all network notifier blocks. Parameters and return value
2019 * are as for raw_notifier_call_chain().
2022 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2024 return call_netdevice_notifiers_extack(val, dev, NULL);
2026 EXPORT_SYMBOL(call_netdevice_notifiers);
2029 * call_netdevice_notifiers_mtu - call all network notifier blocks
2030 * @val: value passed unmodified to notifier function
2031 * @dev: net_device pointer passed unmodified to notifier function
2032 * @arg: additional u32 argument passed to the notifier function
2034 * Call all network notifier blocks. Parameters and return value
2035 * are as for raw_notifier_call_chain().
2037 static int call_netdevice_notifiers_mtu(unsigned long val,
2038 struct net_device *dev, u32 arg)
2040 struct netdev_notifier_info_ext info = {
2045 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2047 return call_netdevice_notifiers_info(val, &info.info);
2050 #ifdef CONFIG_NET_INGRESS
2051 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2053 void net_inc_ingress_queue(void)
2055 static_branch_inc(&ingress_needed_key);
2057 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2059 void net_dec_ingress_queue(void)
2061 static_branch_dec(&ingress_needed_key);
2063 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2066 #ifdef CONFIG_NET_EGRESS
2067 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2069 void net_inc_egress_queue(void)
2071 static_branch_inc(&egress_needed_key);
2073 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2075 void net_dec_egress_queue(void)
2077 static_branch_dec(&egress_needed_key);
2079 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2082 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2083 #ifdef CONFIG_JUMP_LABEL
2084 static atomic_t netstamp_needed_deferred;
2085 static atomic_t netstamp_wanted;
2086 static void netstamp_clear(struct work_struct *work)
2088 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2091 wanted = atomic_add_return(deferred, &netstamp_wanted);
2093 static_branch_enable(&netstamp_needed_key);
2095 static_branch_disable(&netstamp_needed_key);
2097 static DECLARE_WORK(netstamp_work, netstamp_clear);
2100 void net_enable_timestamp(void)
2102 #ifdef CONFIG_JUMP_LABEL
2106 wanted = atomic_read(&netstamp_wanted);
2109 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2112 atomic_inc(&netstamp_needed_deferred);
2113 schedule_work(&netstamp_work);
2115 static_branch_inc(&netstamp_needed_key);
2118 EXPORT_SYMBOL(net_enable_timestamp);
2120 void net_disable_timestamp(void)
2122 #ifdef CONFIG_JUMP_LABEL
2126 wanted = atomic_read(&netstamp_wanted);
2129 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2132 atomic_dec(&netstamp_needed_deferred);
2133 schedule_work(&netstamp_work);
2135 static_branch_dec(&netstamp_needed_key);
2138 EXPORT_SYMBOL(net_disable_timestamp);
2140 static inline void net_timestamp_set(struct sk_buff *skb)
2143 if (static_branch_unlikely(&netstamp_needed_key))
2144 __net_timestamp(skb);
2147 #define net_timestamp_check(COND, SKB) \
2148 if (static_branch_unlikely(&netstamp_needed_key)) { \
2149 if ((COND) && !(SKB)->tstamp) \
2150 __net_timestamp(SKB); \
2153 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2155 return __is_skb_forwardable(dev, skb, true);
2157 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2159 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2162 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2165 skb->protocol = eth_type_trans(skb, dev);
2166 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2172 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2174 return __dev_forward_skb2(dev, skb, true);
2176 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2179 * dev_forward_skb - loopback an skb to another netif
2181 * @dev: destination network device
2182 * @skb: buffer to forward
2185 * NET_RX_SUCCESS (no congestion)
2186 * NET_RX_DROP (packet was dropped, but freed)
2188 * dev_forward_skb can be used for injecting an skb from the
2189 * start_xmit function of one device into the receive queue
2190 * of another device.
2192 * The receiving device may be in another namespace, so
2193 * we have to clear all information in the skb that could
2194 * impact namespace isolation.
2196 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2198 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2200 EXPORT_SYMBOL_GPL(dev_forward_skb);
2202 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2204 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2207 static inline int deliver_skb(struct sk_buff *skb,
2208 struct packet_type *pt_prev,
2209 struct net_device *orig_dev)
2211 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2213 refcount_inc(&skb->users);
2214 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2217 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2218 struct packet_type **pt,
2219 struct net_device *orig_dev,
2221 struct list_head *ptype_list)
2223 struct packet_type *ptype, *pt_prev = *pt;
2225 list_for_each_entry_rcu(ptype, ptype_list, list) {
2226 if (ptype->type != type)
2229 deliver_skb(skb, pt_prev, orig_dev);
2235 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2237 if (!ptype->af_packet_priv || !skb->sk)
2240 if (ptype->id_match)
2241 return ptype->id_match(ptype, skb->sk);
2242 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2249 * dev_nit_active - return true if any network interface taps are in use
2251 * @dev: network device to check for the presence of taps
2253 bool dev_nit_active(struct net_device *dev)
2255 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2257 EXPORT_SYMBOL_GPL(dev_nit_active);
2260 * Support routine. Sends outgoing frames to any network
2261 * taps currently in use.
2264 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2266 struct packet_type *ptype;
2267 struct sk_buff *skb2 = NULL;
2268 struct packet_type *pt_prev = NULL;
2269 struct list_head *ptype_list = &ptype_all;
2273 list_for_each_entry_rcu(ptype, ptype_list, list) {
2274 if (ptype->ignore_outgoing)
2277 /* Never send packets back to the socket
2278 * they originated from - MvS (miquels@drinkel.ow.org)
2280 if (skb_loop_sk(ptype, skb))
2284 deliver_skb(skb2, pt_prev, skb->dev);
2289 /* need to clone skb, done only once */
2290 skb2 = skb_clone(skb, GFP_ATOMIC);
2294 net_timestamp_set(skb2);
2296 /* skb->nh should be correctly
2297 * set by sender, so that the second statement is
2298 * just protection against buggy protocols.
2300 skb_reset_mac_header(skb2);
2302 if (skb_network_header(skb2) < skb2->data ||
2303 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2304 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2305 ntohs(skb2->protocol),
2307 skb_reset_network_header(skb2);
2310 skb2->transport_header = skb2->network_header;
2311 skb2->pkt_type = PACKET_OUTGOING;
2315 if (ptype_list == &ptype_all) {
2316 ptype_list = &dev->ptype_all;
2321 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2322 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2328 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2331 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2332 * @dev: Network device
2333 * @txq: number of queues available
2335 * If real_num_tx_queues is changed the tc mappings may no longer be
2336 * valid. To resolve this verify the tc mapping remains valid and if
2337 * not NULL the mapping. With no priorities mapping to this
2338 * offset/count pair it will no longer be used. In the worst case TC0
2339 * is invalid nothing can be done so disable priority mappings. If is
2340 * expected that drivers will fix this mapping if they can before
2341 * calling netif_set_real_num_tx_queues.
2343 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2346 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2348 /* If TC0 is invalidated disable TC mapping */
2349 if (tc->offset + tc->count > txq) {
2350 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2355 /* Invalidated prio to tc mappings set to TC0 */
2356 for (i = 1; i < TC_BITMASK + 1; i++) {
2357 int q = netdev_get_prio_tc_map(dev, i);
2359 tc = &dev->tc_to_txq[q];
2360 if (tc->offset + tc->count > txq) {
2361 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2363 netdev_set_prio_tc_map(dev, i, 0);
2368 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2371 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2374 /* walk through the TCs and see if it falls into any of them */
2375 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2376 if ((txq - tc->offset) < tc->count)
2380 /* didn't find it, just return -1 to indicate no match */
2386 EXPORT_SYMBOL(netdev_txq_to_tc);
2389 static struct static_key xps_needed __read_mostly;
2390 static struct static_key xps_rxqs_needed __read_mostly;
2391 static DEFINE_MUTEX(xps_map_mutex);
2392 #define xmap_dereference(P) \
2393 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2395 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2396 struct xps_dev_maps *old_maps, int tci, u16 index)
2398 struct xps_map *map = NULL;
2402 map = xmap_dereference(dev_maps->attr_map[tci]);
2406 for (pos = map->len; pos--;) {
2407 if (map->queues[pos] != index)
2411 map->queues[pos] = map->queues[--map->len];
2416 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2417 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2418 kfree_rcu(map, rcu);
2425 static bool remove_xps_queue_cpu(struct net_device *dev,
2426 struct xps_dev_maps *dev_maps,
2427 int cpu, u16 offset, u16 count)
2429 int num_tc = dev_maps->num_tc;
2430 bool active = false;
2433 for (tci = cpu * num_tc; num_tc--; tci++) {
2436 for (i = count, j = offset; i--; j++) {
2437 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2447 static void reset_xps_maps(struct net_device *dev,
2448 struct xps_dev_maps *dev_maps,
2449 enum xps_map_type type)
2451 static_key_slow_dec_cpuslocked(&xps_needed);
2452 if (type == XPS_RXQS)
2453 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2455 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2457 kfree_rcu(dev_maps, rcu);
2460 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2461 u16 offset, u16 count)
2463 struct xps_dev_maps *dev_maps;
2464 bool active = false;
2467 dev_maps = xmap_dereference(dev->xps_maps[type]);
2471 for (j = 0; j < dev_maps->nr_ids; j++)
2472 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2474 reset_xps_maps(dev, dev_maps, type);
2476 if (type == XPS_CPUS) {
2477 for (i = offset + (count - 1); count--; i--)
2478 netdev_queue_numa_node_write(
2479 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2483 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2486 if (!static_key_false(&xps_needed))
2490 mutex_lock(&xps_map_mutex);
2492 if (static_key_false(&xps_rxqs_needed))
2493 clean_xps_maps(dev, XPS_RXQS, offset, count);
2495 clean_xps_maps(dev, XPS_CPUS, offset, count);
2497 mutex_unlock(&xps_map_mutex);
2501 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2503 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2506 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2507 u16 index, bool is_rxqs_map)
2509 struct xps_map *new_map;
2510 int alloc_len = XPS_MIN_MAP_ALLOC;
2513 for (pos = 0; map && pos < map->len; pos++) {
2514 if (map->queues[pos] != index)
2519 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2521 if (pos < map->alloc_len)
2524 alloc_len = map->alloc_len * 2;
2527 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2531 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2533 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2534 cpu_to_node(attr_index));
2538 for (i = 0; i < pos; i++)
2539 new_map->queues[i] = map->queues[i];
2540 new_map->alloc_len = alloc_len;
2546 /* Copy xps maps at a given index */
2547 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2548 struct xps_dev_maps *new_dev_maps, int index,
2549 int tc, bool skip_tc)
2551 int i, tci = index * dev_maps->num_tc;
2552 struct xps_map *map;
2554 /* copy maps belonging to foreign traffic classes */
2555 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2556 if (i == tc && skip_tc)
2559 /* fill in the new device map from the old device map */
2560 map = xmap_dereference(dev_maps->attr_map[tci]);
2561 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2565 /* Must be called under cpus_read_lock */
2566 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2567 u16 index, enum xps_map_type type)
2569 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2570 const unsigned long *online_mask = NULL;
2571 bool active = false, copy = false;
2572 int i, j, tci, numa_node_id = -2;
2573 int maps_sz, num_tc = 1, tc = 0;
2574 struct xps_map *map, *new_map;
2575 unsigned int nr_ids;
2578 /* Do not allow XPS on subordinate device directly */
2579 num_tc = dev->num_tc;
2583 /* If queue belongs to subordinate dev use its map */
2584 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2586 tc = netdev_txq_to_tc(dev, index);
2591 mutex_lock(&xps_map_mutex);
2593 dev_maps = xmap_dereference(dev->xps_maps[type]);
2594 if (type == XPS_RXQS) {
2595 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2596 nr_ids = dev->num_rx_queues;
2598 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2599 if (num_possible_cpus() > 1)
2600 online_mask = cpumask_bits(cpu_online_mask);
2601 nr_ids = nr_cpu_ids;
2604 if (maps_sz < L1_CACHE_BYTES)
2605 maps_sz = L1_CACHE_BYTES;
2607 /* The old dev_maps could be larger or smaller than the one we're
2608 * setting up now, as dev->num_tc or nr_ids could have been updated in
2609 * between. We could try to be smart, but let's be safe instead and only
2610 * copy foreign traffic classes if the two map sizes match.
2613 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2616 /* allocate memory for queue storage */
2617 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2619 if (!new_dev_maps) {
2620 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2621 if (!new_dev_maps) {
2622 mutex_unlock(&xps_map_mutex);
2626 new_dev_maps->nr_ids = nr_ids;
2627 new_dev_maps->num_tc = num_tc;
2630 tci = j * num_tc + tc;
2631 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2633 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2637 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2641 goto out_no_new_maps;
2644 /* Increment static keys at most once per type */
2645 static_key_slow_inc_cpuslocked(&xps_needed);
2646 if (type == XPS_RXQS)
2647 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2650 for (j = 0; j < nr_ids; j++) {
2651 bool skip_tc = false;
2653 tci = j * num_tc + tc;
2654 if (netif_attr_test_mask(j, mask, nr_ids) &&
2655 netif_attr_test_online(j, online_mask, nr_ids)) {
2656 /* add tx-queue to CPU/rx-queue maps */
2661 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2662 while ((pos < map->len) && (map->queues[pos] != index))
2665 if (pos == map->len)
2666 map->queues[map->len++] = index;
2668 if (type == XPS_CPUS) {
2669 if (numa_node_id == -2)
2670 numa_node_id = cpu_to_node(j);
2671 else if (numa_node_id != cpu_to_node(j))
2678 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2682 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2684 /* Cleanup old maps */
2686 goto out_no_old_maps;
2688 for (j = 0; j < dev_maps->nr_ids; j++) {
2689 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2690 map = xmap_dereference(dev_maps->attr_map[tci]);
2695 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2700 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2701 kfree_rcu(map, rcu);
2705 old_dev_maps = dev_maps;
2708 dev_maps = new_dev_maps;
2712 if (type == XPS_CPUS)
2713 /* update Tx queue numa node */
2714 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2715 (numa_node_id >= 0) ?
2716 numa_node_id : NUMA_NO_NODE);
2721 /* removes tx-queue from unused CPUs/rx-queues */
2722 for (j = 0; j < dev_maps->nr_ids; j++) {
2723 tci = j * dev_maps->num_tc;
2725 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2727 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2728 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2731 active |= remove_xps_queue(dev_maps,
2732 copy ? old_dev_maps : NULL,
2738 kfree_rcu(old_dev_maps, rcu);
2740 /* free map if not active */
2742 reset_xps_maps(dev, dev_maps, type);
2745 mutex_unlock(&xps_map_mutex);
2749 /* remove any maps that we added */
2750 for (j = 0; j < nr_ids; j++) {
2751 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2752 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2754 xmap_dereference(dev_maps->attr_map[tci]) :
2756 if (new_map && new_map != map)
2761 mutex_unlock(&xps_map_mutex);
2763 kfree(new_dev_maps);
2766 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2768 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2774 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2779 EXPORT_SYMBOL(netif_set_xps_queue);
2782 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2784 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2786 /* Unbind any subordinate channels */
2787 while (txq-- != &dev->_tx[0]) {
2789 netdev_unbind_sb_channel(dev, txq->sb_dev);
2793 void netdev_reset_tc(struct net_device *dev)
2796 netif_reset_xps_queues_gt(dev, 0);
2798 netdev_unbind_all_sb_channels(dev);
2800 /* Reset TC configuration of device */
2802 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2803 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2805 EXPORT_SYMBOL(netdev_reset_tc);
2807 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2809 if (tc >= dev->num_tc)
2813 netif_reset_xps_queues(dev, offset, count);
2815 dev->tc_to_txq[tc].count = count;
2816 dev->tc_to_txq[tc].offset = offset;
2819 EXPORT_SYMBOL(netdev_set_tc_queue);
2821 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2823 if (num_tc > TC_MAX_QUEUE)
2827 netif_reset_xps_queues_gt(dev, 0);
2829 netdev_unbind_all_sb_channels(dev);
2831 dev->num_tc = num_tc;
2834 EXPORT_SYMBOL(netdev_set_num_tc);
2836 void netdev_unbind_sb_channel(struct net_device *dev,
2837 struct net_device *sb_dev)
2839 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2842 netif_reset_xps_queues_gt(sb_dev, 0);
2844 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2845 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2847 while (txq-- != &dev->_tx[0]) {
2848 if (txq->sb_dev == sb_dev)
2852 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2854 int netdev_bind_sb_channel_queue(struct net_device *dev,
2855 struct net_device *sb_dev,
2856 u8 tc, u16 count, u16 offset)
2858 /* Make certain the sb_dev and dev are already configured */
2859 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2862 /* We cannot hand out queues we don't have */
2863 if ((offset + count) > dev->real_num_tx_queues)
2866 /* Record the mapping */
2867 sb_dev->tc_to_txq[tc].count = count;
2868 sb_dev->tc_to_txq[tc].offset = offset;
2870 /* Provide a way for Tx queue to find the tc_to_txq map or
2871 * XPS map for itself.
2874 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2878 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2880 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2882 /* Do not use a multiqueue device to represent a subordinate channel */
2883 if (netif_is_multiqueue(dev))
2886 /* We allow channels 1 - 32767 to be used for subordinate channels.
2887 * Channel 0 is meant to be "native" mode and used only to represent
2888 * the main root device. We allow writing 0 to reset the device back
2889 * to normal mode after being used as a subordinate channel.
2891 if (channel > S16_MAX)
2894 dev->num_tc = -channel;
2898 EXPORT_SYMBOL(netdev_set_sb_channel);
2901 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2902 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2904 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2909 disabling = txq < dev->real_num_tx_queues;
2911 if (txq < 1 || txq > dev->num_tx_queues)
2914 if (dev->reg_state == NETREG_REGISTERED ||
2915 dev->reg_state == NETREG_UNREGISTERING) {
2918 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2924 netif_setup_tc(dev, txq);
2926 dev_qdisc_change_real_num_tx(dev, txq);
2928 dev->real_num_tx_queues = txq;
2932 qdisc_reset_all_tx_gt(dev, txq);
2934 netif_reset_xps_queues_gt(dev, txq);
2938 dev->real_num_tx_queues = txq;
2943 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2947 * netif_set_real_num_rx_queues - set actual number of RX queues used
2948 * @dev: Network device
2949 * @rxq: Actual number of RX queues
2951 * This must be called either with the rtnl_lock held or before
2952 * registration of the net device. Returns 0 on success, or a
2953 * negative error code. If called before registration, it always
2956 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2960 if (rxq < 1 || rxq > dev->num_rx_queues)
2963 if (dev->reg_state == NETREG_REGISTERED) {
2966 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2972 dev->real_num_rx_queues = rxq;
2975 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2979 * netif_set_real_num_queues - set actual number of RX and TX queues used
2980 * @dev: Network device
2981 * @txq: Actual number of TX queues
2982 * @rxq: Actual number of RX queues
2984 * Set the real number of both TX and RX queues.
2985 * Does nothing if the number of queues is already correct.
2987 int netif_set_real_num_queues(struct net_device *dev,
2988 unsigned int txq, unsigned int rxq)
2990 unsigned int old_rxq = dev->real_num_rx_queues;
2993 if (txq < 1 || txq > dev->num_tx_queues ||
2994 rxq < 1 || rxq > dev->num_rx_queues)
2997 /* Start from increases, so the error path only does decreases -
2998 * decreases can't fail.
3000 if (rxq > dev->real_num_rx_queues) {
3001 err = netif_set_real_num_rx_queues(dev, rxq);
3005 if (txq > dev->real_num_tx_queues) {
3006 err = netif_set_real_num_tx_queues(dev, txq);
3010 if (rxq < dev->real_num_rx_queues)
3011 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3012 if (txq < dev->real_num_tx_queues)
3013 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3017 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3020 EXPORT_SYMBOL(netif_set_real_num_queues);
3023 * netif_get_num_default_rss_queues - default number of RSS queues
3025 * This routine should set an upper limit on the number of RSS queues
3026 * used by default by multiqueue devices.
3028 int netif_get_num_default_rss_queues(void)
3030 return is_kdump_kernel() ?
3031 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3033 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3035 static void __netif_reschedule(struct Qdisc *q)
3037 struct softnet_data *sd;
3038 unsigned long flags;
3040 local_irq_save(flags);
3041 sd = this_cpu_ptr(&softnet_data);
3042 q->next_sched = NULL;
3043 *sd->output_queue_tailp = q;
3044 sd->output_queue_tailp = &q->next_sched;
3045 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3046 local_irq_restore(flags);
3049 void __netif_schedule(struct Qdisc *q)
3051 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3052 __netif_reschedule(q);
3054 EXPORT_SYMBOL(__netif_schedule);
3056 struct dev_kfree_skb_cb {
3057 enum skb_free_reason reason;
3060 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3062 return (struct dev_kfree_skb_cb *)skb->cb;
3065 void netif_schedule_queue(struct netdev_queue *txq)
3068 if (!netif_xmit_stopped(txq)) {
3069 struct Qdisc *q = rcu_dereference(txq->qdisc);
3071 __netif_schedule(q);
3075 EXPORT_SYMBOL(netif_schedule_queue);
3077 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3079 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3083 q = rcu_dereference(dev_queue->qdisc);
3084 __netif_schedule(q);
3088 EXPORT_SYMBOL(netif_tx_wake_queue);
3090 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3092 unsigned long flags;
3097 if (likely(refcount_read(&skb->users) == 1)) {
3099 refcount_set(&skb->users, 0);
3100 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3103 get_kfree_skb_cb(skb)->reason = reason;
3104 local_irq_save(flags);
3105 skb->next = __this_cpu_read(softnet_data.completion_queue);
3106 __this_cpu_write(softnet_data.completion_queue, skb);
3107 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3108 local_irq_restore(flags);
3110 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3112 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3114 if (in_hardirq() || irqs_disabled())
3115 __dev_kfree_skb_irq(skb, reason);
3119 EXPORT_SYMBOL(__dev_kfree_skb_any);
3123 * netif_device_detach - mark device as removed
3124 * @dev: network device
3126 * Mark device as removed from system and therefore no longer available.
3128 void netif_device_detach(struct net_device *dev)
3130 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3131 netif_running(dev)) {
3132 netif_tx_stop_all_queues(dev);
3135 EXPORT_SYMBOL(netif_device_detach);
3138 * netif_device_attach - mark device as attached
3139 * @dev: network device
3141 * Mark device as attached from system and restart if needed.
3143 void netif_device_attach(struct net_device *dev)
3145 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3146 netif_running(dev)) {
3147 netif_tx_wake_all_queues(dev);
3148 __netdev_watchdog_up(dev);
3151 EXPORT_SYMBOL(netif_device_attach);
3154 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3155 * to be used as a distribution range.
3157 static u16 skb_tx_hash(const struct net_device *dev,
3158 const struct net_device *sb_dev,
3159 struct sk_buff *skb)
3163 u16 qcount = dev->real_num_tx_queues;
3166 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3168 qoffset = sb_dev->tc_to_txq[tc].offset;
3169 qcount = sb_dev->tc_to_txq[tc].count;
3170 if (unlikely(!qcount)) {
3171 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3172 sb_dev->name, qoffset, tc);
3174 qcount = dev->real_num_tx_queues;
3178 if (skb_rx_queue_recorded(skb)) {
3179 hash = skb_get_rx_queue(skb);
3180 if (hash >= qoffset)
3182 while (unlikely(hash >= qcount))
3184 return hash + qoffset;
3187 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3190 static void skb_warn_bad_offload(const struct sk_buff *skb)
3192 static const netdev_features_t null_features;
3193 struct net_device *dev = skb->dev;
3194 const char *name = "";
3196 if (!net_ratelimit())
3200 if (dev->dev.parent)
3201 name = dev_driver_string(dev->dev.parent);
3203 name = netdev_name(dev);
3205 skb_dump(KERN_WARNING, skb, false);
3206 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3207 name, dev ? &dev->features : &null_features,
3208 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3212 * Invalidate hardware checksum when packet is to be mangled, and
3213 * complete checksum manually on outgoing path.
3215 int skb_checksum_help(struct sk_buff *skb)
3218 int ret = 0, offset;
3220 if (skb->ip_summed == CHECKSUM_COMPLETE)
3221 goto out_set_summed;
3223 if (unlikely(skb_is_gso(skb))) {
3224 skb_warn_bad_offload(skb);
3228 /* Before computing a checksum, we should make sure no frag could
3229 * be modified by an external entity : checksum could be wrong.
3231 if (skb_has_shared_frag(skb)) {
3232 ret = __skb_linearize(skb);
3237 offset = skb_checksum_start_offset(skb);
3239 if (WARN_ON_ONCE(offset >= skb_headlen(skb)))
3242 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3244 offset += skb->csum_offset;
3245 if (WARN_ON_ONCE(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 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
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 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3937 struct tcf_result cl_res;
3942 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3943 tc_skb_cb(skb)->mru = 0;
3944 tc_skb_cb(skb)->post_ct = false;
3945 mini_qdisc_bstats_cpu_update(miniq, skb);
3947 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3949 case TC_ACT_RECLASSIFY:
3950 skb->tc_index = TC_H_MIN(cl_res.classid);
3953 mini_qdisc_qstats_cpu_drop(miniq);
3954 *ret = NET_XMIT_DROP;
3960 *ret = NET_XMIT_SUCCESS;
3963 case TC_ACT_REDIRECT:
3964 /* No need to push/pop skb's mac_header here on egress! */
3965 skb_do_redirect(skb);
3966 *ret = NET_XMIT_SUCCESS;
3974 #endif /* CONFIG_NET_EGRESS */
3977 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3978 struct xps_dev_maps *dev_maps, unsigned int tci)
3980 int tc = netdev_get_prio_tc_map(dev, skb->priority);
3981 struct xps_map *map;
3982 int queue_index = -1;
3984 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3987 tci *= dev_maps->num_tc;
3990 map = rcu_dereference(dev_maps->attr_map[tci]);
3993 queue_index = map->queues[0];
3995 queue_index = map->queues[reciprocal_scale(
3996 skb_get_hash(skb), map->len)];
3997 if (unlikely(queue_index >= dev->real_num_tx_queues))
4004 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4005 struct sk_buff *skb)
4008 struct xps_dev_maps *dev_maps;
4009 struct sock *sk = skb->sk;
4010 int queue_index = -1;
4012 if (!static_key_false(&xps_needed))
4016 if (!static_key_false(&xps_rxqs_needed))
4019 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4021 int tci = sk_rx_queue_get(sk);
4024 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4029 if (queue_index < 0) {
4030 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4032 unsigned int tci = skb->sender_cpu - 1;
4034 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4046 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4047 struct net_device *sb_dev)
4051 EXPORT_SYMBOL(dev_pick_tx_zero);
4053 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4054 struct net_device *sb_dev)
4056 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4058 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4060 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4061 struct net_device *sb_dev)
4063 struct sock *sk = skb->sk;
4064 int queue_index = sk_tx_queue_get(sk);
4066 sb_dev = sb_dev ? : dev;
4068 if (queue_index < 0 || skb->ooo_okay ||
4069 queue_index >= dev->real_num_tx_queues) {
4070 int new_index = get_xps_queue(dev, sb_dev, skb);
4073 new_index = skb_tx_hash(dev, sb_dev, skb);
4075 if (queue_index != new_index && sk &&
4077 rcu_access_pointer(sk->sk_dst_cache))
4078 sk_tx_queue_set(sk, new_index);
4080 queue_index = new_index;
4085 EXPORT_SYMBOL(netdev_pick_tx);
4087 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4088 struct sk_buff *skb,
4089 struct net_device *sb_dev)
4091 int queue_index = 0;
4094 u32 sender_cpu = skb->sender_cpu - 1;
4096 if (sender_cpu >= (u32)NR_CPUS)
4097 skb->sender_cpu = raw_smp_processor_id() + 1;
4100 if (dev->real_num_tx_queues != 1) {
4101 const struct net_device_ops *ops = dev->netdev_ops;
4103 if (ops->ndo_select_queue)
4104 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4106 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4108 queue_index = netdev_cap_txqueue(dev, queue_index);
4111 skb_set_queue_mapping(skb, queue_index);
4112 return netdev_get_tx_queue(dev, queue_index);
4116 * __dev_queue_xmit - transmit a buffer
4117 * @skb: buffer to transmit
4118 * @sb_dev: suboordinate device used for L2 forwarding offload
4120 * Queue a buffer for transmission to a network device. The caller must
4121 * have set the device and priority and built the buffer before calling
4122 * this function. The function can be called from an interrupt.
4124 * A negative errno code is returned on a failure. A success does not
4125 * guarantee the frame will be transmitted as it may be dropped due
4126 * to congestion or traffic shaping.
4128 * -----------------------------------------------------------------------------------
4129 * I notice this method can also return errors from the queue disciplines,
4130 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4133 * Regardless of the return value, the skb is consumed, so it is currently
4134 * difficult to retry a send to this method. (You can bump the ref count
4135 * before sending to hold a reference for retry if you are careful.)
4137 * When calling this method, interrupts MUST be enabled. This is because
4138 * the BH enable code must have IRQs enabled so that it will not deadlock.
4141 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4143 struct net_device *dev = skb->dev;
4144 struct netdev_queue *txq;
4149 skb_reset_mac_header(skb);
4151 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4152 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4154 /* Disable soft irqs for various locks below. Also
4155 * stops preemption for RCU.
4159 skb_update_prio(skb);
4161 qdisc_pkt_len_init(skb);
4162 #ifdef CONFIG_NET_CLS_ACT
4163 skb->tc_at_ingress = 0;
4164 # ifdef CONFIG_NET_EGRESS
4165 if (static_branch_unlikely(&egress_needed_key)) {
4166 skb = sch_handle_egress(skb, &rc, dev);
4172 /* If device/qdisc don't need skb->dst, release it right now while
4173 * its hot in this cpu cache.
4175 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4180 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4181 q = rcu_dereference_bh(txq->qdisc);
4183 trace_net_dev_queue(skb);
4185 rc = __dev_xmit_skb(skb, q, dev, txq);
4189 /* The device has no queue. Common case for software devices:
4190 * loopback, all the sorts of tunnels...
4192 * Really, it is unlikely that netif_tx_lock protection is necessary
4193 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4195 * However, it is possible, that they rely on protection
4198 * Check this and shot the lock. It is not prone from deadlocks.
4199 *Either shot noqueue qdisc, it is even simpler 8)
4201 if (dev->flags & IFF_UP) {
4202 int cpu = smp_processor_id(); /* ok because BHs are off */
4204 /* Other cpus might concurrently change txq->xmit_lock_owner
4205 * to -1 or to their cpu id, but not to our id.
4207 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4208 if (dev_xmit_recursion())
4209 goto recursion_alert;
4211 skb = validate_xmit_skb(skb, dev, &again);
4215 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4216 HARD_TX_LOCK(dev, txq, cpu);
4218 if (!netif_xmit_stopped(txq)) {
4219 dev_xmit_recursion_inc();
4220 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4221 dev_xmit_recursion_dec();
4222 if (dev_xmit_complete(rc)) {
4223 HARD_TX_UNLOCK(dev, txq);
4227 HARD_TX_UNLOCK(dev, txq);
4228 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4231 /* Recursion is detected! It is possible,
4235 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4241 rcu_read_unlock_bh();
4243 atomic_long_inc(&dev->tx_dropped);
4244 kfree_skb_list(skb);
4247 rcu_read_unlock_bh();
4251 int dev_queue_xmit(struct sk_buff *skb)
4253 return __dev_queue_xmit(skb, NULL);
4255 EXPORT_SYMBOL(dev_queue_xmit);
4257 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4259 return __dev_queue_xmit(skb, sb_dev);
4261 EXPORT_SYMBOL(dev_queue_xmit_accel);
4263 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4265 struct net_device *dev = skb->dev;
4266 struct sk_buff *orig_skb = skb;
4267 struct netdev_queue *txq;
4268 int ret = NETDEV_TX_BUSY;
4271 if (unlikely(!netif_running(dev) ||
4272 !netif_carrier_ok(dev)))
4275 skb = validate_xmit_skb_list(skb, dev, &again);
4276 if (skb != orig_skb)
4279 skb_set_queue_mapping(skb, queue_id);
4280 txq = skb_get_tx_queue(dev, skb);
4281 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4285 dev_xmit_recursion_inc();
4286 HARD_TX_LOCK(dev, txq, smp_processor_id());
4287 if (!netif_xmit_frozen_or_drv_stopped(txq))
4288 ret = netdev_start_xmit(skb, dev, txq, false);
4289 HARD_TX_UNLOCK(dev, txq);
4290 dev_xmit_recursion_dec();
4295 atomic_long_inc(&dev->tx_dropped);
4296 kfree_skb_list(skb);
4297 return NET_XMIT_DROP;
4299 EXPORT_SYMBOL(__dev_direct_xmit);
4301 /*************************************************************************
4303 *************************************************************************/
4305 int netdev_max_backlog __read_mostly = 1000;
4306 EXPORT_SYMBOL(netdev_max_backlog);
4308 int netdev_tstamp_prequeue __read_mostly = 1;
4309 int netdev_budget __read_mostly = 300;
4310 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4311 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4312 int weight_p __read_mostly = 64; /* old backlog weight */
4313 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4314 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4315 int dev_rx_weight __read_mostly = 64;
4316 int dev_tx_weight __read_mostly = 64;
4317 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4318 int gro_normal_batch __read_mostly = 8;
4320 /* Called with irq disabled */
4321 static inline void ____napi_schedule(struct softnet_data *sd,
4322 struct napi_struct *napi)
4324 struct task_struct *thread;
4326 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4327 /* Paired with smp_mb__before_atomic() in
4328 * napi_enable()/dev_set_threaded().
4329 * Use READ_ONCE() to guarantee a complete
4330 * read on napi->thread. Only call
4331 * wake_up_process() when it's not NULL.
4333 thread = READ_ONCE(napi->thread);
4335 /* Avoid doing set_bit() if the thread is in
4336 * INTERRUPTIBLE state, cause napi_thread_wait()
4337 * makes sure to proceed with napi polling
4338 * if the thread is explicitly woken from here.
4340 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4341 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4342 wake_up_process(thread);
4347 list_add_tail(&napi->poll_list, &sd->poll_list);
4348 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4353 /* One global table that all flow-based protocols share. */
4354 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4355 EXPORT_SYMBOL(rps_sock_flow_table);
4356 u32 rps_cpu_mask __read_mostly;
4357 EXPORT_SYMBOL(rps_cpu_mask);
4359 struct static_key_false rps_needed __read_mostly;
4360 EXPORT_SYMBOL(rps_needed);
4361 struct static_key_false rfs_needed __read_mostly;
4362 EXPORT_SYMBOL(rfs_needed);
4364 static struct rps_dev_flow *
4365 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4366 struct rps_dev_flow *rflow, u16 next_cpu)
4368 if (next_cpu < nr_cpu_ids) {
4369 #ifdef CONFIG_RFS_ACCEL
4370 struct netdev_rx_queue *rxqueue;
4371 struct rps_dev_flow_table *flow_table;
4372 struct rps_dev_flow *old_rflow;
4377 /* Should we steer this flow to a different hardware queue? */
4378 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4379 !(dev->features & NETIF_F_NTUPLE))
4381 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4382 if (rxq_index == skb_get_rx_queue(skb))
4385 rxqueue = dev->_rx + rxq_index;
4386 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4389 flow_id = skb_get_hash(skb) & flow_table->mask;
4390 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4391 rxq_index, flow_id);
4395 rflow = &flow_table->flows[flow_id];
4397 if (old_rflow->filter == rflow->filter)
4398 old_rflow->filter = RPS_NO_FILTER;
4402 per_cpu(softnet_data, next_cpu).input_queue_head;
4405 rflow->cpu = next_cpu;
4410 * get_rps_cpu is called from netif_receive_skb and returns the target
4411 * CPU from the RPS map of the receiving queue for a given skb.
4412 * rcu_read_lock must be held on entry.
4414 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4415 struct rps_dev_flow **rflowp)
4417 const struct rps_sock_flow_table *sock_flow_table;
4418 struct netdev_rx_queue *rxqueue = dev->_rx;
4419 struct rps_dev_flow_table *flow_table;
4420 struct rps_map *map;
4425 if (skb_rx_queue_recorded(skb)) {
4426 u16 index = skb_get_rx_queue(skb);
4428 if (unlikely(index >= dev->real_num_rx_queues)) {
4429 WARN_ONCE(dev->real_num_rx_queues > 1,
4430 "%s received packet on queue %u, but number "
4431 "of RX queues is %u\n",
4432 dev->name, index, dev->real_num_rx_queues);
4438 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4440 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4441 map = rcu_dereference(rxqueue->rps_map);
4442 if (!flow_table && !map)
4445 skb_reset_network_header(skb);
4446 hash = skb_get_hash(skb);
4450 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4451 if (flow_table && sock_flow_table) {
4452 struct rps_dev_flow *rflow;
4456 /* First check into global flow table if there is a match */
4457 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4458 if ((ident ^ hash) & ~rps_cpu_mask)
4461 next_cpu = ident & rps_cpu_mask;
4463 /* OK, now we know there is a match,
4464 * we can look at the local (per receive queue) flow table
4466 rflow = &flow_table->flows[hash & flow_table->mask];
4470 * If the desired CPU (where last recvmsg was done) is
4471 * different from current CPU (one in the rx-queue flow
4472 * table entry), switch if one of the following holds:
4473 * - Current CPU is unset (>= nr_cpu_ids).
4474 * - Current CPU is offline.
4475 * - The current CPU's queue tail has advanced beyond the
4476 * last packet that was enqueued using this table entry.
4477 * This guarantees that all previous packets for the flow
4478 * have been dequeued, thus preserving in order delivery.
4480 if (unlikely(tcpu != next_cpu) &&
4481 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4482 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4483 rflow->last_qtail)) >= 0)) {
4485 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4488 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4498 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4499 if (cpu_online(tcpu)) {
4509 #ifdef CONFIG_RFS_ACCEL
4512 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4513 * @dev: Device on which the filter was set
4514 * @rxq_index: RX queue index
4515 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4516 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4518 * Drivers that implement ndo_rx_flow_steer() should periodically call
4519 * this function for each installed filter and remove the filters for
4520 * which it returns %true.
4522 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4523 u32 flow_id, u16 filter_id)
4525 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4526 struct rps_dev_flow_table *flow_table;
4527 struct rps_dev_flow *rflow;
4532 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4533 if (flow_table && flow_id <= flow_table->mask) {
4534 rflow = &flow_table->flows[flow_id];
4535 cpu = READ_ONCE(rflow->cpu);
4536 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4537 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4538 rflow->last_qtail) <
4539 (int)(10 * flow_table->mask)))
4545 EXPORT_SYMBOL(rps_may_expire_flow);
4547 #endif /* CONFIG_RFS_ACCEL */
4549 /* Called from hardirq (IPI) context */
4550 static void rps_trigger_softirq(void *data)
4552 struct softnet_data *sd = data;
4554 ____napi_schedule(sd, &sd->backlog);
4558 #endif /* CONFIG_RPS */
4561 * Check if this softnet_data structure is another cpu one
4562 * If yes, queue it to our IPI list and return 1
4565 static int rps_ipi_queued(struct softnet_data *sd)
4568 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4571 sd->rps_ipi_next = mysd->rps_ipi_list;
4572 mysd->rps_ipi_list = sd;
4574 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4577 #endif /* CONFIG_RPS */
4581 #ifdef CONFIG_NET_FLOW_LIMIT
4582 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4585 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4587 #ifdef CONFIG_NET_FLOW_LIMIT
4588 struct sd_flow_limit *fl;
4589 struct softnet_data *sd;
4590 unsigned int old_flow, new_flow;
4592 if (qlen < (netdev_max_backlog >> 1))
4595 sd = this_cpu_ptr(&softnet_data);
4598 fl = rcu_dereference(sd->flow_limit);
4600 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4601 old_flow = fl->history[fl->history_head];
4602 fl->history[fl->history_head] = new_flow;
4605 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4607 if (likely(fl->buckets[old_flow]))
4608 fl->buckets[old_flow]--;
4610 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4622 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4623 * queue (may be a remote CPU queue).
4625 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4626 unsigned int *qtail)
4628 struct softnet_data *sd;
4629 unsigned long flags;
4632 sd = &per_cpu(softnet_data, cpu);
4634 local_irq_save(flags);
4637 if (!netif_running(skb->dev))
4639 qlen = skb_queue_len(&sd->input_pkt_queue);
4640 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4643 __skb_queue_tail(&sd->input_pkt_queue, skb);
4644 input_queue_tail_incr_save(sd, qtail);
4646 local_irq_restore(flags);
4647 return NET_RX_SUCCESS;
4650 /* Schedule NAPI for backlog device
4651 * We can use non atomic operation since we own the queue lock
4653 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4654 if (!rps_ipi_queued(sd))
4655 ____napi_schedule(sd, &sd->backlog);
4664 local_irq_restore(flags);
4666 atomic_long_inc(&skb->dev->rx_dropped);
4671 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4673 struct net_device *dev = skb->dev;
4674 struct netdev_rx_queue *rxqueue;
4678 if (skb_rx_queue_recorded(skb)) {
4679 u16 index = skb_get_rx_queue(skb);
4681 if (unlikely(index >= dev->real_num_rx_queues)) {
4682 WARN_ONCE(dev->real_num_rx_queues > 1,
4683 "%s received packet on queue %u, but number "
4684 "of RX queues is %u\n",
4685 dev->name, index, dev->real_num_rx_queues);
4687 return rxqueue; /* Return first rxqueue */
4694 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4695 struct bpf_prog *xdp_prog)
4697 void *orig_data, *orig_data_end, *hard_start;
4698 struct netdev_rx_queue *rxqueue;
4699 bool orig_bcast, orig_host;
4700 u32 mac_len, frame_sz;
4701 __be16 orig_eth_type;
4706 /* The XDP program wants to see the packet starting at the MAC
4709 mac_len = skb->data - skb_mac_header(skb);
4710 hard_start = skb->data - skb_headroom(skb);
4712 /* SKB "head" area always have tailroom for skb_shared_info */
4713 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4714 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4716 rxqueue = netif_get_rxqueue(skb);
4717 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4718 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4719 skb_headlen(skb) + mac_len, true);
4721 orig_data_end = xdp->data_end;
4722 orig_data = xdp->data;
4723 eth = (struct ethhdr *)xdp->data;
4724 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4725 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4726 orig_eth_type = eth->h_proto;
4728 act = bpf_prog_run_xdp(xdp_prog, xdp);
4730 /* check if bpf_xdp_adjust_head was used */
4731 off = xdp->data - orig_data;
4734 __skb_pull(skb, off);
4736 __skb_push(skb, -off);
4738 skb->mac_header += off;
4739 skb_reset_network_header(skb);
4742 /* check if bpf_xdp_adjust_tail was used */
4743 off = xdp->data_end - orig_data_end;
4745 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4746 skb->len += off; /* positive on grow, negative on shrink */
4749 /* check if XDP changed eth hdr such SKB needs update */
4750 eth = (struct ethhdr *)xdp->data;
4751 if ((orig_eth_type != eth->h_proto) ||
4752 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4753 skb->dev->dev_addr)) ||
4754 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4755 __skb_push(skb, ETH_HLEN);
4756 skb->pkt_type = PACKET_HOST;
4757 skb->protocol = eth_type_trans(skb, skb->dev);
4760 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4761 * before calling us again on redirect path. We do not call do_redirect
4762 * as we leave that up to the caller.
4764 * Caller is responsible for managing lifetime of skb (i.e. calling
4765 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4770 __skb_push(skb, mac_len);
4773 metalen = xdp->data - xdp->data_meta;
4775 skb_metadata_set(skb, metalen);
4782 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4783 struct xdp_buff *xdp,
4784 struct bpf_prog *xdp_prog)
4788 /* Reinjected packets coming from act_mirred or similar should
4789 * not get XDP generic processing.
4791 if (skb_is_redirected(skb))
4794 /* XDP packets must be linear and must have sufficient headroom
4795 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4796 * native XDP provides, thus we need to do it here as well.
4798 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4799 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4800 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4801 int troom = skb->tail + skb->data_len - skb->end;
4803 /* In case we have to go down the path and also linearize,
4804 * then lets do the pskb_expand_head() work just once here.
4806 if (pskb_expand_head(skb,
4807 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4808 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4810 if (skb_linearize(skb))
4814 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4821 bpf_warn_invalid_xdp_action(act);
4824 trace_xdp_exception(skb->dev, xdp_prog, act);
4835 /* When doing generic XDP we have to bypass the qdisc layer and the
4836 * network taps in order to match in-driver-XDP behavior.
4838 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4840 struct net_device *dev = skb->dev;
4841 struct netdev_queue *txq;
4842 bool free_skb = true;
4845 txq = netdev_core_pick_tx(dev, skb, NULL);
4846 cpu = smp_processor_id();
4847 HARD_TX_LOCK(dev, txq, cpu);
4848 if (!netif_xmit_stopped(txq)) {
4849 rc = netdev_start_xmit(skb, dev, txq, 0);
4850 if (dev_xmit_complete(rc))
4853 HARD_TX_UNLOCK(dev, txq);
4855 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4860 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4862 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4865 struct xdp_buff xdp;
4869 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4870 if (act != XDP_PASS) {
4873 err = xdp_do_generic_redirect(skb->dev, skb,
4879 generic_xdp_tx(skb, xdp_prog);
4890 EXPORT_SYMBOL_GPL(do_xdp_generic);
4892 static int netif_rx_internal(struct sk_buff *skb)
4896 net_timestamp_check(netdev_tstamp_prequeue, skb);
4898 trace_netif_rx(skb);
4901 if (static_branch_unlikely(&rps_needed)) {
4902 struct rps_dev_flow voidflow, *rflow = &voidflow;
4908 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4910 cpu = smp_processor_id();
4912 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4921 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4928 * netif_rx - post buffer to the network code
4929 * @skb: buffer to post
4931 * This function receives a packet from a device driver and queues it for
4932 * the upper (protocol) levels to process. It always succeeds. The buffer
4933 * may be dropped during processing for congestion control or by the
4937 * NET_RX_SUCCESS (no congestion)
4938 * NET_RX_DROP (packet was dropped)
4942 int netif_rx(struct sk_buff *skb)
4946 trace_netif_rx_entry(skb);
4948 ret = netif_rx_internal(skb);
4949 trace_netif_rx_exit(ret);
4953 EXPORT_SYMBOL(netif_rx);
4955 int netif_rx_ni(struct sk_buff *skb)
4959 trace_netif_rx_ni_entry(skb);
4962 err = netif_rx_internal(skb);
4963 if (local_softirq_pending())
4966 trace_netif_rx_ni_exit(err);
4970 EXPORT_SYMBOL(netif_rx_ni);
4972 int netif_rx_any_context(struct sk_buff *skb)
4975 * If invoked from contexts which do not invoke bottom half
4976 * processing either at return from interrupt or when softrqs are
4977 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4981 return netif_rx(skb);
4983 return netif_rx_ni(skb);
4985 EXPORT_SYMBOL(netif_rx_any_context);
4987 static __latent_entropy void net_tx_action(struct softirq_action *h)
4989 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4991 if (sd->completion_queue) {
4992 struct sk_buff *clist;
4994 local_irq_disable();
4995 clist = sd->completion_queue;
4996 sd->completion_queue = NULL;
5000 struct sk_buff *skb = clist;
5002 clist = clist->next;
5004 WARN_ON(refcount_read(&skb->users));
5005 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5006 trace_consume_skb(skb);
5008 trace_kfree_skb(skb, net_tx_action);
5010 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5013 __kfree_skb_defer(skb);
5017 if (sd->output_queue) {
5020 local_irq_disable();
5021 head = sd->output_queue;
5022 sd->output_queue = NULL;
5023 sd->output_queue_tailp = &sd->output_queue;
5029 struct Qdisc *q = head;
5030 spinlock_t *root_lock = NULL;
5032 head = head->next_sched;
5034 /* We need to make sure head->next_sched is read
5035 * before clearing __QDISC_STATE_SCHED
5037 smp_mb__before_atomic();
5039 if (!(q->flags & TCQ_F_NOLOCK)) {
5040 root_lock = qdisc_lock(q);
5041 spin_lock(root_lock);
5042 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5044 /* There is a synchronize_net() between
5045 * STATE_DEACTIVATED flag being set and
5046 * qdisc_reset()/some_qdisc_is_busy() in
5047 * dev_deactivate(), so we can safely bail out
5048 * early here to avoid data race between
5049 * qdisc_deactivate() and some_qdisc_is_busy()
5050 * for lockless qdisc.
5052 clear_bit(__QDISC_STATE_SCHED, &q->state);
5056 clear_bit(__QDISC_STATE_SCHED, &q->state);
5059 spin_unlock(root_lock);
5065 xfrm_dev_backlog(sd);
5068 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5069 /* This hook is defined here for ATM LANE */
5070 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5071 unsigned char *addr) __read_mostly;
5072 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5075 static inline struct sk_buff *
5076 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5077 struct net_device *orig_dev, bool *another)
5079 #ifdef CONFIG_NET_CLS_ACT
5080 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5081 struct tcf_result cl_res;
5083 /* If there's at least one ingress present somewhere (so
5084 * we get here via enabled static key), remaining devices
5085 * that are not configured with an ingress qdisc will bail
5092 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5096 qdisc_skb_cb(skb)->pkt_len = skb->len;
5097 tc_skb_cb(skb)->mru = 0;
5098 tc_skb_cb(skb)->post_ct = false;
5099 skb->tc_at_ingress = 1;
5100 mini_qdisc_bstats_cpu_update(miniq, skb);
5102 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5104 case TC_ACT_RECLASSIFY:
5105 skb->tc_index = TC_H_MIN(cl_res.classid);
5108 mini_qdisc_qstats_cpu_drop(miniq);
5116 case TC_ACT_REDIRECT:
5117 /* skb_mac_header check was done by cls/act_bpf, so
5118 * we can safely push the L2 header back before
5119 * redirecting to another netdev
5121 __skb_push(skb, skb->mac_len);
5122 if (skb_do_redirect(skb) == -EAGAIN) {
5123 __skb_pull(skb, skb->mac_len);
5128 case TC_ACT_CONSUMED:
5133 #endif /* CONFIG_NET_CLS_ACT */
5138 * netdev_is_rx_handler_busy - check if receive handler is registered
5139 * @dev: device to check
5141 * Check if a receive handler is already registered for a given device.
5142 * Return true if there one.
5144 * The caller must hold the rtnl_mutex.
5146 bool netdev_is_rx_handler_busy(struct net_device *dev)
5149 return dev && rtnl_dereference(dev->rx_handler);
5151 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5154 * netdev_rx_handler_register - register receive handler
5155 * @dev: device to register a handler for
5156 * @rx_handler: receive handler to register
5157 * @rx_handler_data: data pointer that is used by rx handler
5159 * Register a receive handler for a device. This handler will then be
5160 * called from __netif_receive_skb. A negative errno code is returned
5163 * The caller must hold the rtnl_mutex.
5165 * For a general description of rx_handler, see enum rx_handler_result.
5167 int netdev_rx_handler_register(struct net_device *dev,
5168 rx_handler_func_t *rx_handler,
5169 void *rx_handler_data)
5171 if (netdev_is_rx_handler_busy(dev))
5174 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5177 /* Note: rx_handler_data must be set before rx_handler */
5178 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5179 rcu_assign_pointer(dev->rx_handler, rx_handler);
5183 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5186 * netdev_rx_handler_unregister - unregister receive handler
5187 * @dev: device to unregister a handler from
5189 * Unregister a receive handler from a device.
5191 * The caller must hold the rtnl_mutex.
5193 void netdev_rx_handler_unregister(struct net_device *dev)
5197 RCU_INIT_POINTER(dev->rx_handler, NULL);
5198 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5199 * section has a guarantee to see a non NULL rx_handler_data
5203 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5205 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5208 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5209 * the special handling of PFMEMALLOC skbs.
5211 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5213 switch (skb->protocol) {
5214 case htons(ETH_P_ARP):
5215 case htons(ETH_P_IP):
5216 case htons(ETH_P_IPV6):
5217 case htons(ETH_P_8021Q):
5218 case htons(ETH_P_8021AD):
5225 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5226 int *ret, struct net_device *orig_dev)
5228 if (nf_hook_ingress_active(skb)) {
5232 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5237 ingress_retval = nf_hook_ingress(skb);
5239 return ingress_retval;
5244 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5245 struct packet_type **ppt_prev)
5247 struct packet_type *ptype, *pt_prev;
5248 rx_handler_func_t *rx_handler;
5249 struct sk_buff *skb = *pskb;
5250 struct net_device *orig_dev;
5251 bool deliver_exact = false;
5252 int ret = NET_RX_DROP;
5255 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5257 trace_netif_receive_skb(skb);
5259 orig_dev = skb->dev;
5261 skb_reset_network_header(skb);
5262 if (!skb_transport_header_was_set(skb))
5263 skb_reset_transport_header(skb);
5264 skb_reset_mac_len(skb);
5269 skb->skb_iif = skb->dev->ifindex;
5271 __this_cpu_inc(softnet_data.processed);
5273 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5277 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5280 if (ret2 != XDP_PASS) {
5286 if (eth_type_vlan(skb->protocol)) {
5287 skb = skb_vlan_untag(skb);
5292 if (skb_skip_tc_classify(skb))
5298 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5300 ret = deliver_skb(skb, pt_prev, orig_dev);
5304 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5306 ret = deliver_skb(skb, pt_prev, orig_dev);
5311 #ifdef CONFIG_NET_INGRESS
5312 if (static_branch_unlikely(&ingress_needed_key)) {
5313 bool another = false;
5315 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5322 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5326 skb_reset_redirect(skb);
5328 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5331 if (skb_vlan_tag_present(skb)) {
5333 ret = deliver_skb(skb, pt_prev, orig_dev);
5336 if (vlan_do_receive(&skb))
5338 else if (unlikely(!skb))
5342 rx_handler = rcu_dereference(skb->dev->rx_handler);
5345 ret = deliver_skb(skb, pt_prev, orig_dev);
5348 switch (rx_handler(&skb)) {
5349 case RX_HANDLER_CONSUMED:
5350 ret = NET_RX_SUCCESS;
5352 case RX_HANDLER_ANOTHER:
5354 case RX_HANDLER_EXACT:
5355 deliver_exact = true;
5357 case RX_HANDLER_PASS:
5364 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5366 if (skb_vlan_tag_get_id(skb)) {
5367 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5370 skb->pkt_type = PACKET_OTHERHOST;
5371 } else if (eth_type_vlan(skb->protocol)) {
5372 /* Outer header is 802.1P with vlan 0, inner header is
5373 * 802.1Q or 802.1AD and vlan_do_receive() above could
5374 * not find vlan dev for vlan id 0.
5376 __vlan_hwaccel_clear_tag(skb);
5377 skb = skb_vlan_untag(skb);
5380 if (vlan_do_receive(&skb))
5381 /* After stripping off 802.1P header with vlan 0
5382 * vlan dev is found for inner header.
5385 else if (unlikely(!skb))
5388 /* We have stripped outer 802.1P vlan 0 header.
5389 * But could not find vlan dev.
5390 * check again for vlan id to set OTHERHOST.
5394 /* Note: we might in the future use prio bits
5395 * and set skb->priority like in vlan_do_receive()
5396 * For the time being, just ignore Priority Code Point
5398 __vlan_hwaccel_clear_tag(skb);
5401 type = skb->protocol;
5403 /* deliver only exact match when indicated */
5404 if (likely(!deliver_exact)) {
5405 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5406 &ptype_base[ntohs(type) &
5410 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5411 &orig_dev->ptype_specific);
5413 if (unlikely(skb->dev != orig_dev)) {
5414 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5415 &skb->dev->ptype_specific);
5419 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5421 *ppt_prev = pt_prev;
5425 atomic_long_inc(&skb->dev->rx_dropped);
5427 atomic_long_inc(&skb->dev->rx_nohandler);
5429 /* Jamal, now you will not able to escape explaining
5430 * me how you were going to use this. :-)
5436 /* The invariant here is that if *ppt_prev is not NULL
5437 * then skb should also be non-NULL.
5439 * Apparently *ppt_prev assignment above holds this invariant due to
5440 * skb dereferencing near it.
5446 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5448 struct net_device *orig_dev = skb->dev;
5449 struct packet_type *pt_prev = NULL;
5452 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5454 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5455 skb->dev, pt_prev, orig_dev);
5460 * netif_receive_skb_core - special purpose version of netif_receive_skb
5461 * @skb: buffer to process
5463 * More direct receive version of netif_receive_skb(). It should
5464 * only be used by callers that have a need to skip RPS and Generic XDP.
5465 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5467 * This function may only be called from softirq context and interrupts
5468 * should be enabled.
5470 * Return values (usually ignored):
5471 * NET_RX_SUCCESS: no congestion
5472 * NET_RX_DROP: packet was dropped
5474 int netif_receive_skb_core(struct sk_buff *skb)
5479 ret = __netif_receive_skb_one_core(skb, false);
5484 EXPORT_SYMBOL(netif_receive_skb_core);
5486 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5487 struct packet_type *pt_prev,
5488 struct net_device *orig_dev)
5490 struct sk_buff *skb, *next;
5494 if (list_empty(head))
5496 if (pt_prev->list_func != NULL)
5497 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5498 ip_list_rcv, head, pt_prev, orig_dev);
5500 list_for_each_entry_safe(skb, next, head, list) {
5501 skb_list_del_init(skb);
5502 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5506 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5508 /* Fast-path assumptions:
5509 * - There is no RX handler.
5510 * - Only one packet_type matches.
5511 * If either of these fails, we will end up doing some per-packet
5512 * processing in-line, then handling the 'last ptype' for the whole
5513 * sublist. This can't cause out-of-order delivery to any single ptype,
5514 * because the 'last ptype' must be constant across the sublist, and all
5515 * other ptypes are handled per-packet.
5517 /* Current (common) ptype of sublist */
5518 struct packet_type *pt_curr = NULL;
5519 /* Current (common) orig_dev of sublist */
5520 struct net_device *od_curr = NULL;
5521 struct list_head sublist;
5522 struct sk_buff *skb, *next;
5524 INIT_LIST_HEAD(&sublist);
5525 list_for_each_entry_safe(skb, next, head, list) {
5526 struct net_device *orig_dev = skb->dev;
5527 struct packet_type *pt_prev = NULL;
5529 skb_list_del_init(skb);
5530 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5533 if (pt_curr != pt_prev || od_curr != orig_dev) {
5534 /* dispatch old sublist */
5535 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5536 /* start new sublist */
5537 INIT_LIST_HEAD(&sublist);
5541 list_add_tail(&skb->list, &sublist);
5544 /* dispatch final sublist */
5545 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5548 static int __netif_receive_skb(struct sk_buff *skb)
5552 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5553 unsigned int noreclaim_flag;
5556 * PFMEMALLOC skbs are special, they should
5557 * - be delivered to SOCK_MEMALLOC sockets only
5558 * - stay away from userspace
5559 * - have bounded memory usage
5561 * Use PF_MEMALLOC as this saves us from propagating the allocation
5562 * context down to all allocation sites.
5564 noreclaim_flag = memalloc_noreclaim_save();
5565 ret = __netif_receive_skb_one_core(skb, true);
5566 memalloc_noreclaim_restore(noreclaim_flag);
5568 ret = __netif_receive_skb_one_core(skb, false);
5573 static void __netif_receive_skb_list(struct list_head *head)
5575 unsigned long noreclaim_flag = 0;
5576 struct sk_buff *skb, *next;
5577 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5579 list_for_each_entry_safe(skb, next, head, list) {
5580 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5581 struct list_head sublist;
5583 /* Handle the previous sublist */
5584 list_cut_before(&sublist, head, &skb->list);
5585 if (!list_empty(&sublist))
5586 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5587 pfmemalloc = !pfmemalloc;
5588 /* See comments in __netif_receive_skb */
5590 noreclaim_flag = memalloc_noreclaim_save();
5592 memalloc_noreclaim_restore(noreclaim_flag);
5595 /* Handle the remaining sublist */
5596 if (!list_empty(head))
5597 __netif_receive_skb_list_core(head, pfmemalloc);
5598 /* Restore pflags */
5600 memalloc_noreclaim_restore(noreclaim_flag);
5603 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5605 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5606 struct bpf_prog *new = xdp->prog;
5609 switch (xdp->command) {
5610 case XDP_SETUP_PROG:
5611 rcu_assign_pointer(dev->xdp_prog, new);
5616 static_branch_dec(&generic_xdp_needed_key);
5617 } else if (new && !old) {
5618 static_branch_inc(&generic_xdp_needed_key);
5619 dev_disable_lro(dev);
5620 dev_disable_gro_hw(dev);
5632 static int netif_receive_skb_internal(struct sk_buff *skb)
5636 net_timestamp_check(netdev_tstamp_prequeue, skb);
5638 if (skb_defer_rx_timestamp(skb))
5639 return NET_RX_SUCCESS;
5643 if (static_branch_unlikely(&rps_needed)) {
5644 struct rps_dev_flow voidflow, *rflow = &voidflow;
5645 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5648 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5654 ret = __netif_receive_skb(skb);
5659 static void netif_receive_skb_list_internal(struct list_head *head)
5661 struct sk_buff *skb, *next;
5662 struct list_head sublist;
5664 INIT_LIST_HEAD(&sublist);
5665 list_for_each_entry_safe(skb, next, head, list) {
5666 net_timestamp_check(netdev_tstamp_prequeue, skb);
5667 skb_list_del_init(skb);
5668 if (!skb_defer_rx_timestamp(skb))
5669 list_add_tail(&skb->list, &sublist);
5671 list_splice_init(&sublist, head);
5675 if (static_branch_unlikely(&rps_needed)) {
5676 list_for_each_entry_safe(skb, next, head, list) {
5677 struct rps_dev_flow voidflow, *rflow = &voidflow;
5678 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5681 /* Will be handled, remove from list */
5682 skb_list_del_init(skb);
5683 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5688 __netif_receive_skb_list(head);
5693 * netif_receive_skb - process receive buffer from network
5694 * @skb: buffer to process
5696 * netif_receive_skb() is the main receive data processing function.
5697 * It always succeeds. The buffer may be dropped during processing
5698 * for congestion control or by the protocol layers.
5700 * This function may only be called from softirq context and interrupts
5701 * should be enabled.
5703 * Return values (usually ignored):
5704 * NET_RX_SUCCESS: no congestion
5705 * NET_RX_DROP: packet was dropped
5707 int netif_receive_skb(struct sk_buff *skb)
5711 trace_netif_receive_skb_entry(skb);
5713 ret = netif_receive_skb_internal(skb);
5714 trace_netif_receive_skb_exit(ret);
5718 EXPORT_SYMBOL(netif_receive_skb);
5721 * netif_receive_skb_list - process many receive buffers from network
5722 * @head: list of skbs to process.
5724 * Since return value of netif_receive_skb() is normally ignored, and
5725 * wouldn't be meaningful for a list, this function returns void.
5727 * This function may only be called from softirq context and interrupts
5728 * should be enabled.
5730 void netif_receive_skb_list(struct list_head *head)
5732 struct sk_buff *skb;
5734 if (list_empty(head))
5736 if (trace_netif_receive_skb_list_entry_enabled()) {
5737 list_for_each_entry(skb, head, list)
5738 trace_netif_receive_skb_list_entry(skb);
5740 netif_receive_skb_list_internal(head);
5741 trace_netif_receive_skb_list_exit(0);
5743 EXPORT_SYMBOL(netif_receive_skb_list);
5745 static DEFINE_PER_CPU(struct work_struct, flush_works);
5747 /* Network device is going away, flush any packets still pending */
5748 static void flush_backlog(struct work_struct *work)
5750 struct sk_buff *skb, *tmp;
5751 struct softnet_data *sd;
5754 sd = this_cpu_ptr(&softnet_data);
5756 local_irq_disable();
5758 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5759 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5760 __skb_unlink(skb, &sd->input_pkt_queue);
5761 dev_kfree_skb_irq(skb);
5762 input_queue_head_incr(sd);
5768 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5769 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5770 __skb_unlink(skb, &sd->process_queue);
5772 input_queue_head_incr(sd);
5778 static bool flush_required(int cpu)
5780 #if IS_ENABLED(CONFIG_RPS)
5781 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5784 local_irq_disable();
5787 /* as insertion into process_queue happens with the rps lock held,
5788 * process_queue access may race only with dequeue
5790 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5791 !skb_queue_empty_lockless(&sd->process_queue);
5797 /* without RPS we can't safely check input_pkt_queue: during a
5798 * concurrent remote skb_queue_splice() we can detect as empty both
5799 * input_pkt_queue and process_queue even if the latter could end-up
5800 * containing a lot of packets.
5805 static void flush_all_backlogs(void)
5807 static cpumask_t flush_cpus;
5810 /* since we are under rtnl lock protection we can use static data
5811 * for the cpumask and avoid allocating on stack the possibly
5818 cpumask_clear(&flush_cpus);
5819 for_each_online_cpu(cpu) {
5820 if (flush_required(cpu)) {
5821 queue_work_on(cpu, system_highpri_wq,
5822 per_cpu_ptr(&flush_works, cpu));
5823 cpumask_set_cpu(cpu, &flush_cpus);
5827 /* we can have in flight packet[s] on the cpus we are not flushing,
5828 * synchronize_net() in unregister_netdevice_many() will take care of
5831 for_each_cpu(cpu, &flush_cpus)
5832 flush_work(per_cpu_ptr(&flush_works, cpu));
5837 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5838 static void gro_normal_list(struct napi_struct *napi)
5840 if (!napi->rx_count)
5842 netif_receive_skb_list_internal(&napi->rx_list);
5843 INIT_LIST_HEAD(&napi->rx_list);
5847 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5848 * pass the whole batch up to the stack.
5850 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5852 list_add_tail(&skb->list, &napi->rx_list);
5853 napi->rx_count += segs;
5854 if (napi->rx_count >= gro_normal_batch)
5855 gro_normal_list(napi);
5858 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5860 struct packet_offload *ptype;
5861 __be16 type = skb->protocol;
5862 struct list_head *head = &offload_base;
5865 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5867 if (NAPI_GRO_CB(skb)->count == 1) {
5868 skb_shinfo(skb)->gso_size = 0;
5873 list_for_each_entry_rcu(ptype, head, list) {
5874 if (ptype->type != type || !ptype->callbacks.gro_complete)
5877 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5878 ipv6_gro_complete, inet_gro_complete,
5885 WARN_ON(&ptype->list == head);
5887 return NET_RX_SUCCESS;
5891 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5892 return NET_RX_SUCCESS;
5895 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5898 struct list_head *head = &napi->gro_hash[index].list;
5899 struct sk_buff *skb, *p;
5901 list_for_each_entry_safe_reverse(skb, p, head, list) {
5902 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5904 skb_list_del_init(skb);
5905 napi_gro_complete(napi, skb);
5906 napi->gro_hash[index].count--;
5909 if (!napi->gro_hash[index].count)
5910 __clear_bit(index, &napi->gro_bitmask);
5913 /* napi->gro_hash[].list contains packets ordered by age.
5914 * youngest packets at the head of it.
5915 * Complete skbs in reverse order to reduce latencies.
5917 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5919 unsigned long bitmask = napi->gro_bitmask;
5920 unsigned int i, base = ~0U;
5922 while ((i = ffs(bitmask)) != 0) {
5925 __napi_gro_flush_chain(napi, base, flush_old);
5928 EXPORT_SYMBOL(napi_gro_flush);
5930 static void gro_list_prepare(const struct list_head *head,
5931 const struct sk_buff *skb)
5933 unsigned int maclen = skb->dev->hard_header_len;
5934 u32 hash = skb_get_hash_raw(skb);
5937 list_for_each_entry(p, head, list) {
5938 unsigned long diffs;
5940 NAPI_GRO_CB(p)->flush = 0;
5942 if (hash != skb_get_hash_raw(p)) {
5943 NAPI_GRO_CB(p)->same_flow = 0;
5947 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5948 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5949 if (skb_vlan_tag_present(p))
5950 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5951 diffs |= skb_metadata_differs(p, skb);
5952 if (maclen == ETH_HLEN)
5953 diffs |= compare_ether_header(skb_mac_header(p),
5954 skb_mac_header(skb));
5956 diffs = memcmp(skb_mac_header(p),
5957 skb_mac_header(skb),
5960 /* in most common scenarions 'slow_gro' is 0
5961 * otherwise we are already on some slower paths
5962 * either skip all the infrequent tests altogether or
5963 * avoid trying too hard to skip each of them individually
5965 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
5966 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5967 struct tc_skb_ext *skb_ext;
5968 struct tc_skb_ext *p_ext;
5971 diffs |= p->sk != skb->sk;
5972 diffs |= skb_metadata_dst_cmp(p, skb);
5973 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
5975 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5976 skb_ext = skb_ext_find(skb, TC_SKB_EXT);
5977 p_ext = skb_ext_find(p, TC_SKB_EXT);
5979 diffs |= (!!p_ext) ^ (!!skb_ext);
5980 if (!diffs && unlikely(skb_ext))
5981 diffs |= p_ext->chain ^ skb_ext->chain;
5985 NAPI_GRO_CB(p)->same_flow = !diffs;
5989 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
5991 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5992 const skb_frag_t *frag0 = &pinfo->frags[0];
5994 NAPI_GRO_CB(skb)->data_offset = 0;
5995 NAPI_GRO_CB(skb)->frag0 = NULL;
5996 NAPI_GRO_CB(skb)->frag0_len = 0;
5998 if (!skb_headlen(skb) && pinfo->nr_frags &&
5999 !PageHighMem(skb_frag_page(frag0)) &&
6000 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
6001 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
6002 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
6003 skb_frag_size(frag0),
6004 skb->end - skb->tail);
6008 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
6010 struct skb_shared_info *pinfo = skb_shinfo(skb);
6012 BUG_ON(skb->end - skb->tail < grow);
6014 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6016 skb->data_len -= grow;
6019 skb_frag_off_add(&pinfo->frags[0], grow);
6020 skb_frag_size_sub(&pinfo->frags[0], grow);
6022 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6023 skb_frag_unref(skb, 0);
6024 memmove(pinfo->frags, pinfo->frags + 1,
6025 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
6029 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6031 struct sk_buff *oldest;
6033 oldest = list_last_entry(head, struct sk_buff, list);
6035 /* We are called with head length >= MAX_GRO_SKBS, so this is
6038 if (WARN_ON_ONCE(!oldest))
6041 /* Do not adjust napi->gro_hash[].count, caller is adding a new
6044 skb_list_del_init(oldest);
6045 napi_gro_complete(napi, oldest);
6048 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6050 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6051 struct gro_list *gro_list = &napi->gro_hash[bucket];
6052 struct list_head *head = &offload_base;
6053 struct packet_offload *ptype;
6054 __be16 type = skb->protocol;
6055 struct sk_buff *pp = NULL;
6056 enum gro_result ret;
6060 if (netif_elide_gro(skb->dev))
6063 gro_list_prepare(&gro_list->list, skb);
6066 list_for_each_entry_rcu(ptype, head, list) {
6067 if (ptype->type != type || !ptype->callbacks.gro_receive)
6070 skb_set_network_header(skb, skb_gro_offset(skb));
6071 skb_reset_mac_len(skb);
6072 NAPI_GRO_CB(skb)->same_flow = 0;
6073 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6074 NAPI_GRO_CB(skb)->free = 0;
6075 NAPI_GRO_CB(skb)->encap_mark = 0;
6076 NAPI_GRO_CB(skb)->recursion_counter = 0;
6077 NAPI_GRO_CB(skb)->is_fou = 0;
6078 NAPI_GRO_CB(skb)->is_atomic = 1;
6079 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6081 /* Setup for GRO checksum validation */
6082 switch (skb->ip_summed) {
6083 case CHECKSUM_COMPLETE:
6084 NAPI_GRO_CB(skb)->csum = skb->csum;
6085 NAPI_GRO_CB(skb)->csum_valid = 1;
6086 NAPI_GRO_CB(skb)->csum_cnt = 0;
6088 case CHECKSUM_UNNECESSARY:
6089 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6090 NAPI_GRO_CB(skb)->csum_valid = 0;
6093 NAPI_GRO_CB(skb)->csum_cnt = 0;
6094 NAPI_GRO_CB(skb)->csum_valid = 0;
6097 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6098 ipv6_gro_receive, inet_gro_receive,
6099 &gro_list->list, skb);
6104 if (&ptype->list == head)
6107 if (PTR_ERR(pp) == -EINPROGRESS) {
6112 same_flow = NAPI_GRO_CB(skb)->same_flow;
6113 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6116 skb_list_del_init(pp);
6117 napi_gro_complete(napi, pp);
6124 if (NAPI_GRO_CB(skb)->flush)
6127 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6128 gro_flush_oldest(napi, &gro_list->list);
6132 NAPI_GRO_CB(skb)->count = 1;
6133 NAPI_GRO_CB(skb)->age = jiffies;
6134 NAPI_GRO_CB(skb)->last = skb;
6135 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6136 list_add(&skb->list, &gro_list->list);
6140 grow = skb_gro_offset(skb) - skb_headlen(skb);
6142 gro_pull_from_frag0(skb, grow);
6144 if (gro_list->count) {
6145 if (!test_bit(bucket, &napi->gro_bitmask))
6146 __set_bit(bucket, &napi->gro_bitmask);
6147 } else if (test_bit(bucket, &napi->gro_bitmask)) {
6148 __clear_bit(bucket, &napi->gro_bitmask);
6158 struct packet_offload *gro_find_receive_by_type(__be16 type)
6160 struct list_head *offload_head = &offload_base;
6161 struct packet_offload *ptype;
6163 list_for_each_entry_rcu(ptype, offload_head, list) {
6164 if (ptype->type != type || !ptype->callbacks.gro_receive)
6170 EXPORT_SYMBOL(gro_find_receive_by_type);
6172 struct packet_offload *gro_find_complete_by_type(__be16 type)
6174 struct list_head *offload_head = &offload_base;
6175 struct packet_offload *ptype;
6177 list_for_each_entry_rcu(ptype, offload_head, list) {
6178 if (ptype->type != type || !ptype->callbacks.gro_complete)
6184 EXPORT_SYMBOL(gro_find_complete_by_type);
6186 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6187 struct sk_buff *skb,
6192 gro_normal_one(napi, skb, 1);
6195 case GRO_MERGED_FREE:
6196 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6197 napi_skb_free_stolen_head(skb);
6198 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6201 __kfree_skb_defer(skb);
6213 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6217 skb_mark_napi_id(skb, napi);
6218 trace_napi_gro_receive_entry(skb);
6220 skb_gro_reset_offset(skb, 0);
6222 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6223 trace_napi_gro_receive_exit(ret);
6227 EXPORT_SYMBOL(napi_gro_receive);
6229 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6231 if (unlikely(skb->pfmemalloc)) {
6235 __skb_pull(skb, skb_headlen(skb));
6236 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6237 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6238 __vlan_hwaccel_clear_tag(skb);
6239 skb->dev = napi->dev;
6242 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6243 skb->pkt_type = PACKET_HOST;
6245 skb->encapsulation = 0;
6246 skb_shinfo(skb)->gso_type = 0;
6247 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6248 if (unlikely(skb->slow_gro)) {
6258 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6260 struct sk_buff *skb = napi->skb;
6263 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6266 skb_mark_napi_id(skb, napi);
6271 EXPORT_SYMBOL(napi_get_frags);
6273 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6274 struct sk_buff *skb,
6280 __skb_push(skb, ETH_HLEN);
6281 skb->protocol = eth_type_trans(skb, skb->dev);
6282 if (ret == GRO_NORMAL)
6283 gro_normal_one(napi, skb, 1);
6286 case GRO_MERGED_FREE:
6287 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6288 napi_skb_free_stolen_head(skb);
6290 napi_reuse_skb(napi, skb);
6301 /* Upper GRO stack assumes network header starts at gro_offset=0
6302 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6303 * We copy ethernet header into skb->data to have a common layout.
6305 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6307 struct sk_buff *skb = napi->skb;
6308 const struct ethhdr *eth;
6309 unsigned int hlen = sizeof(*eth);
6313 skb_reset_mac_header(skb);
6314 skb_gro_reset_offset(skb, hlen);
6316 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6317 eth = skb_gro_header_slow(skb, hlen, 0);
6318 if (unlikely(!eth)) {
6319 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6320 __func__, napi->dev->name);
6321 napi_reuse_skb(napi, skb);
6325 eth = (const struct ethhdr *)skb->data;
6326 gro_pull_from_frag0(skb, hlen);
6327 NAPI_GRO_CB(skb)->frag0 += hlen;
6328 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6330 __skb_pull(skb, hlen);
6333 * This works because the only protocols we care about don't require
6335 * We'll fix it up properly in napi_frags_finish()
6337 skb->protocol = eth->h_proto;
6342 gro_result_t napi_gro_frags(struct napi_struct *napi)
6345 struct sk_buff *skb = napi_frags_skb(napi);
6347 trace_napi_gro_frags_entry(skb);
6349 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6350 trace_napi_gro_frags_exit(ret);
6354 EXPORT_SYMBOL(napi_gro_frags);
6356 /* Compute the checksum from gro_offset and return the folded value
6357 * after adding in any pseudo checksum.
6359 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6364 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6366 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6367 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6368 /* See comments in __skb_checksum_complete(). */
6370 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6371 !skb->csum_complete_sw)
6372 netdev_rx_csum_fault(skb->dev, skb);
6375 NAPI_GRO_CB(skb)->csum = wsum;
6376 NAPI_GRO_CB(skb)->csum_valid = 1;
6380 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6382 static void net_rps_send_ipi(struct softnet_data *remsd)
6386 struct softnet_data *next = remsd->rps_ipi_next;
6388 if (cpu_online(remsd->cpu))
6389 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6396 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6397 * Note: called with local irq disabled, but exits with local irq enabled.
6399 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6402 struct softnet_data *remsd = sd->rps_ipi_list;
6405 sd->rps_ipi_list = NULL;
6409 /* Send pending IPI's to kick RPS processing on remote cpus. */
6410 net_rps_send_ipi(remsd);
6416 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6419 return sd->rps_ipi_list != NULL;
6425 static int process_backlog(struct napi_struct *napi, int quota)
6427 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6431 /* Check if we have pending ipi, its better to send them now,
6432 * not waiting net_rx_action() end.
6434 if (sd_has_rps_ipi_waiting(sd)) {
6435 local_irq_disable();
6436 net_rps_action_and_irq_enable(sd);
6439 napi->weight = dev_rx_weight;
6441 struct sk_buff *skb;
6443 while ((skb = __skb_dequeue(&sd->process_queue))) {
6445 __netif_receive_skb(skb);
6447 input_queue_head_incr(sd);
6448 if (++work >= quota)
6453 local_irq_disable();
6455 if (skb_queue_empty(&sd->input_pkt_queue)) {
6457 * Inline a custom version of __napi_complete().
6458 * only current cpu owns and manipulates this napi,
6459 * and NAPI_STATE_SCHED is the only possible flag set
6461 * We can use a plain write instead of clear_bit(),
6462 * and we dont need an smp_mb() memory barrier.
6467 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6468 &sd->process_queue);
6478 * __napi_schedule - schedule for receive
6479 * @n: entry to schedule
6481 * The entry's receive function will be scheduled to run.
6482 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6484 void __napi_schedule(struct napi_struct *n)
6486 unsigned long flags;
6488 local_irq_save(flags);
6489 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6490 local_irq_restore(flags);
6492 EXPORT_SYMBOL(__napi_schedule);
6495 * napi_schedule_prep - check if napi can be scheduled
6498 * Test if NAPI routine is already running, and if not mark
6499 * it as running. This is used as a condition variable to
6500 * insure only one NAPI poll instance runs. We also make
6501 * sure there is no pending NAPI disable.
6503 bool napi_schedule_prep(struct napi_struct *n)
6505 unsigned long val, new;
6508 val = READ_ONCE(n->state);
6509 if (unlikely(val & NAPIF_STATE_DISABLE))
6511 new = val | NAPIF_STATE_SCHED;
6513 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6514 * This was suggested by Alexander Duyck, as compiler
6515 * emits better code than :
6516 * if (val & NAPIF_STATE_SCHED)
6517 * new |= NAPIF_STATE_MISSED;
6519 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6521 } while (cmpxchg(&n->state, val, new) != val);
6523 return !(val & NAPIF_STATE_SCHED);
6525 EXPORT_SYMBOL(napi_schedule_prep);
6528 * __napi_schedule_irqoff - schedule for receive
6529 * @n: entry to schedule
6531 * Variant of __napi_schedule() assuming hard irqs are masked.
6533 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6534 * because the interrupt disabled assumption might not be true
6535 * due to force-threaded interrupts and spinlock substitution.
6537 void __napi_schedule_irqoff(struct napi_struct *n)
6539 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6540 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6544 EXPORT_SYMBOL(__napi_schedule_irqoff);
6546 bool napi_complete_done(struct napi_struct *n, int work_done)
6548 unsigned long flags, val, new, timeout = 0;
6552 * 1) Don't let napi dequeue from the cpu poll list
6553 * just in case its running on a different cpu.
6554 * 2) If we are busy polling, do nothing here, we have
6555 * the guarantee we will be called later.
6557 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6558 NAPIF_STATE_IN_BUSY_POLL)))
6563 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6564 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6566 if (n->defer_hard_irqs_count > 0) {
6567 n->defer_hard_irqs_count--;
6568 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6572 if (n->gro_bitmask) {
6573 /* When the NAPI instance uses a timeout and keeps postponing
6574 * it, we need to bound somehow the time packets are kept in
6577 napi_gro_flush(n, !!timeout);
6582 if (unlikely(!list_empty(&n->poll_list))) {
6583 /* If n->poll_list is not empty, we need to mask irqs */
6584 local_irq_save(flags);
6585 list_del_init(&n->poll_list);
6586 local_irq_restore(flags);
6590 val = READ_ONCE(n->state);
6592 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6594 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6595 NAPIF_STATE_SCHED_THREADED |
6596 NAPIF_STATE_PREFER_BUSY_POLL);
6598 /* If STATE_MISSED was set, leave STATE_SCHED set,
6599 * because we will call napi->poll() one more time.
6600 * This C code was suggested by Alexander Duyck to help gcc.
6602 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6604 } while (cmpxchg(&n->state, val, new) != val);
6606 if (unlikely(val & NAPIF_STATE_MISSED)) {
6612 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6613 HRTIMER_MODE_REL_PINNED);
6616 EXPORT_SYMBOL(napi_complete_done);
6618 /* must be called under rcu_read_lock(), as we dont take a reference */
6619 static struct napi_struct *napi_by_id(unsigned int napi_id)
6621 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6622 struct napi_struct *napi;
6624 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6625 if (napi->napi_id == napi_id)
6631 #if defined(CONFIG_NET_RX_BUSY_POLL)
6633 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6635 if (!skip_schedule) {
6636 gro_normal_list(napi);
6637 __napi_schedule(napi);
6641 if (napi->gro_bitmask) {
6642 /* flush too old packets
6643 * If HZ < 1000, flush all packets.
6645 napi_gro_flush(napi, HZ >= 1000);
6648 gro_normal_list(napi);
6649 clear_bit(NAPI_STATE_SCHED, &napi->state);
6652 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6655 bool skip_schedule = false;
6656 unsigned long timeout;
6659 /* Busy polling means there is a high chance device driver hard irq
6660 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6661 * set in napi_schedule_prep().
6662 * Since we are about to call napi->poll() once more, we can safely
6663 * clear NAPI_STATE_MISSED.
6665 * Note: x86 could use a single "lock and ..." instruction
6666 * to perform these two clear_bit()
6668 clear_bit(NAPI_STATE_MISSED, &napi->state);
6669 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6673 if (prefer_busy_poll) {
6674 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6675 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6676 if (napi->defer_hard_irqs_count && timeout) {
6677 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6678 skip_schedule = true;
6682 /* All we really want here is to re-enable device interrupts.
6683 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6685 rc = napi->poll(napi, budget);
6686 /* We can't gro_normal_list() here, because napi->poll() might have
6687 * rearmed the napi (napi_complete_done()) in which case it could
6688 * already be running on another CPU.
6690 trace_napi_poll(napi, rc, budget);
6691 netpoll_poll_unlock(have_poll_lock);
6693 __busy_poll_stop(napi, skip_schedule);
6697 void napi_busy_loop(unsigned int napi_id,
6698 bool (*loop_end)(void *, unsigned long),
6699 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6701 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6702 int (*napi_poll)(struct napi_struct *napi, int budget);
6703 void *have_poll_lock = NULL;
6704 struct napi_struct *napi;
6711 napi = napi_by_id(napi_id);
6721 unsigned long val = READ_ONCE(napi->state);
6723 /* If multiple threads are competing for this napi,
6724 * we avoid dirtying napi->state as much as we can.
6726 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6727 NAPIF_STATE_IN_BUSY_POLL)) {
6728 if (prefer_busy_poll)
6729 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6732 if (cmpxchg(&napi->state, val,
6733 val | NAPIF_STATE_IN_BUSY_POLL |
6734 NAPIF_STATE_SCHED) != val) {
6735 if (prefer_busy_poll)
6736 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6739 have_poll_lock = netpoll_poll_lock(napi);
6740 napi_poll = napi->poll;
6742 work = napi_poll(napi, budget);
6743 trace_napi_poll(napi, work, budget);
6744 gro_normal_list(napi);
6747 __NET_ADD_STATS(dev_net(napi->dev),
6748 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6751 if (!loop_end || loop_end(loop_end_arg, start_time))
6754 if (unlikely(need_resched())) {
6756 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6760 if (loop_end(loop_end_arg, start_time))
6767 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6772 EXPORT_SYMBOL(napi_busy_loop);
6774 #endif /* CONFIG_NET_RX_BUSY_POLL */
6776 static void napi_hash_add(struct napi_struct *napi)
6778 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6781 spin_lock(&napi_hash_lock);
6783 /* 0..NR_CPUS range is reserved for sender_cpu use */
6785 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6786 napi_gen_id = MIN_NAPI_ID;
6787 } while (napi_by_id(napi_gen_id));
6788 napi->napi_id = napi_gen_id;
6790 hlist_add_head_rcu(&napi->napi_hash_node,
6791 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6793 spin_unlock(&napi_hash_lock);
6796 /* Warning : caller is responsible to make sure rcu grace period
6797 * is respected before freeing memory containing @napi
6799 static void napi_hash_del(struct napi_struct *napi)
6801 spin_lock(&napi_hash_lock);
6803 hlist_del_init_rcu(&napi->napi_hash_node);
6805 spin_unlock(&napi_hash_lock);
6808 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6810 struct napi_struct *napi;
6812 napi = container_of(timer, struct napi_struct, timer);
6814 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6815 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6817 if (!napi_disable_pending(napi) &&
6818 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6819 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6820 __napi_schedule_irqoff(napi);
6823 return HRTIMER_NORESTART;
6826 static void init_gro_hash(struct napi_struct *napi)
6830 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6831 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6832 napi->gro_hash[i].count = 0;
6834 napi->gro_bitmask = 0;
6837 int dev_set_threaded(struct net_device *dev, bool threaded)
6839 struct napi_struct *napi;
6842 if (dev->threaded == threaded)
6846 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6847 if (!napi->thread) {
6848 err = napi_kthread_create(napi);
6857 dev->threaded = threaded;
6859 /* Make sure kthread is created before THREADED bit
6862 smp_mb__before_atomic();
6864 /* Setting/unsetting threaded mode on a napi might not immediately
6865 * take effect, if the current napi instance is actively being
6866 * polled. In this case, the switch between threaded mode and
6867 * softirq mode will happen in the next round of napi_schedule().
6868 * This should not cause hiccups/stalls to the live traffic.
6870 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6872 set_bit(NAPI_STATE_THREADED, &napi->state);
6874 clear_bit(NAPI_STATE_THREADED, &napi->state);
6879 EXPORT_SYMBOL(dev_set_threaded);
6881 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6882 int (*poll)(struct napi_struct *, int), int weight)
6884 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6887 INIT_LIST_HEAD(&napi->poll_list);
6888 INIT_HLIST_NODE(&napi->napi_hash_node);
6889 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6890 napi->timer.function = napi_watchdog;
6891 init_gro_hash(napi);
6893 INIT_LIST_HEAD(&napi->rx_list);
6896 if (weight > NAPI_POLL_WEIGHT)
6897 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6899 napi->weight = weight;
6901 #ifdef CONFIG_NETPOLL
6902 napi->poll_owner = -1;
6904 set_bit(NAPI_STATE_SCHED, &napi->state);
6905 set_bit(NAPI_STATE_NPSVC, &napi->state);
6906 list_add_rcu(&napi->dev_list, &dev->napi_list);
6907 napi_hash_add(napi);
6908 /* Create kthread for this napi if dev->threaded is set.
6909 * Clear dev->threaded if kthread creation failed so that
6910 * threaded mode will not be enabled in napi_enable().
6912 if (dev->threaded && napi_kthread_create(napi))
6915 EXPORT_SYMBOL(netif_napi_add);
6917 void napi_disable(struct napi_struct *n)
6920 set_bit(NAPI_STATE_DISABLE, &n->state);
6922 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6924 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6927 hrtimer_cancel(&n->timer);
6929 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6930 clear_bit(NAPI_STATE_DISABLE, &n->state);
6931 clear_bit(NAPI_STATE_THREADED, &n->state);
6933 EXPORT_SYMBOL(napi_disable);
6936 * napi_enable - enable NAPI scheduling
6939 * Resume NAPI from being scheduled on this context.
6940 * Must be paired with napi_disable.
6942 void napi_enable(struct napi_struct *n)
6944 unsigned long val, new;
6947 val = READ_ONCE(n->state);
6948 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6950 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6951 if (n->dev->threaded && n->thread)
6952 new |= NAPIF_STATE_THREADED;
6953 } while (cmpxchg(&n->state, val, new) != val);
6955 EXPORT_SYMBOL(napi_enable);
6957 static void flush_gro_hash(struct napi_struct *napi)
6961 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6962 struct sk_buff *skb, *n;
6964 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6966 napi->gro_hash[i].count = 0;
6970 /* Must be called in process context */
6971 void __netif_napi_del(struct napi_struct *napi)
6973 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6976 napi_hash_del(napi);
6977 list_del_rcu(&napi->dev_list);
6978 napi_free_frags(napi);
6980 flush_gro_hash(napi);
6981 napi->gro_bitmask = 0;
6984 kthread_stop(napi->thread);
6985 napi->thread = NULL;
6988 EXPORT_SYMBOL(__netif_napi_del);
6990 static int __napi_poll(struct napi_struct *n, bool *repoll)
6996 /* This NAPI_STATE_SCHED test is for avoiding a race
6997 * with netpoll's poll_napi(). Only the entity which
6998 * obtains the lock and sees NAPI_STATE_SCHED set will
6999 * actually make the ->poll() call. Therefore we avoid
7000 * accidentally calling ->poll() when NAPI is not scheduled.
7003 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
7004 work = n->poll(n, weight);
7005 trace_napi_poll(n, work, weight);
7008 if (unlikely(work > weight))
7009 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7010 n->poll, work, weight);
7012 if (likely(work < weight))
7015 /* Drivers must not modify the NAPI state if they
7016 * consume the entire weight. In such cases this code
7017 * still "owns" the NAPI instance and therefore can
7018 * move the instance around on the list at-will.
7020 if (unlikely(napi_disable_pending(n))) {
7025 /* The NAPI context has more processing work, but busy-polling
7026 * is preferred. Exit early.
7028 if (napi_prefer_busy_poll(n)) {
7029 if (napi_complete_done(n, work)) {
7030 /* If timeout is not set, we need to make sure
7031 * that the NAPI is re-scheduled.
7038 if (n->gro_bitmask) {
7039 /* flush too old packets
7040 * If HZ < 1000, flush all packets.
7042 napi_gro_flush(n, HZ >= 1000);
7047 /* Some drivers may have called napi_schedule
7048 * prior to exhausting their budget.
7050 if (unlikely(!list_empty(&n->poll_list))) {
7051 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7052 n->dev ? n->dev->name : "backlog");
7061 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7063 bool do_repoll = false;
7067 list_del_init(&n->poll_list);
7069 have = netpoll_poll_lock(n);
7071 work = __napi_poll(n, &do_repoll);
7074 list_add_tail(&n->poll_list, repoll);
7076 netpoll_poll_unlock(have);
7081 static int napi_thread_wait(struct napi_struct *napi)
7085 set_current_state(TASK_INTERRUPTIBLE);
7087 while (!kthread_should_stop()) {
7088 /* Testing SCHED_THREADED bit here to make sure the current
7089 * kthread owns this napi and could poll on this napi.
7090 * Testing SCHED bit is not enough because SCHED bit might be
7091 * set by some other busy poll thread or by napi_disable().
7093 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7094 WARN_ON(!list_empty(&napi->poll_list));
7095 __set_current_state(TASK_RUNNING);
7100 /* woken being true indicates this thread owns this napi. */
7102 set_current_state(TASK_INTERRUPTIBLE);
7104 __set_current_state(TASK_RUNNING);
7109 static int napi_threaded_poll(void *data)
7111 struct napi_struct *napi = data;
7114 while (!napi_thread_wait(napi)) {
7116 bool repoll = false;
7120 have = netpoll_poll_lock(napi);
7121 __napi_poll(napi, &repoll);
7122 netpoll_poll_unlock(have);
7135 static __latent_entropy void net_rx_action(struct softirq_action *h)
7137 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7138 unsigned long time_limit = jiffies +
7139 usecs_to_jiffies(netdev_budget_usecs);
7140 int budget = netdev_budget;
7144 local_irq_disable();
7145 list_splice_init(&sd->poll_list, &list);
7149 struct napi_struct *n;
7151 if (list_empty(&list)) {
7152 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7157 n = list_first_entry(&list, struct napi_struct, poll_list);
7158 budget -= napi_poll(n, &repoll);
7160 /* If softirq window is exhausted then punt.
7161 * Allow this to run for 2 jiffies since which will allow
7162 * an average latency of 1.5/HZ.
7164 if (unlikely(budget <= 0 ||
7165 time_after_eq(jiffies, time_limit))) {
7171 local_irq_disable();
7173 list_splice_tail_init(&sd->poll_list, &list);
7174 list_splice_tail(&repoll, &list);
7175 list_splice(&list, &sd->poll_list);
7176 if (!list_empty(&sd->poll_list))
7177 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7179 net_rps_action_and_irq_enable(sd);
7182 struct netdev_adjacent {
7183 struct net_device *dev;
7185 /* upper master flag, there can only be one master device per list */
7188 /* lookup ignore flag */
7191 /* counter for the number of times this device was added to us */
7194 /* private field for the users */
7197 struct list_head list;
7198 struct rcu_head rcu;
7201 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7202 struct list_head *adj_list)
7204 struct netdev_adjacent *adj;
7206 list_for_each_entry(adj, adj_list, list) {
7207 if (adj->dev == adj_dev)
7213 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7214 struct netdev_nested_priv *priv)
7216 struct net_device *dev = (struct net_device *)priv->data;
7218 return upper_dev == dev;
7222 * netdev_has_upper_dev - Check if device is linked to an upper device
7224 * @upper_dev: upper device to check
7226 * Find out if a device is linked to specified upper device and return true
7227 * in case it is. Note that this checks only immediate upper device,
7228 * not through a complete stack of devices. The caller must hold the RTNL lock.
7230 bool netdev_has_upper_dev(struct net_device *dev,
7231 struct net_device *upper_dev)
7233 struct netdev_nested_priv priv = {
7234 .data = (void *)upper_dev,
7239 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7242 EXPORT_SYMBOL(netdev_has_upper_dev);
7245 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7247 * @upper_dev: upper device to check
7249 * Find out if a device is linked to specified upper device and return true
7250 * in case it is. Note that this checks the entire upper device chain.
7251 * The caller must hold rcu lock.
7254 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7255 struct net_device *upper_dev)
7257 struct netdev_nested_priv priv = {
7258 .data = (void *)upper_dev,
7261 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7264 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7267 * netdev_has_any_upper_dev - Check if device is linked to some device
7270 * Find out if a device is linked to an upper device and return true in case
7271 * it is. The caller must hold the RTNL lock.
7273 bool netdev_has_any_upper_dev(struct net_device *dev)
7277 return !list_empty(&dev->adj_list.upper);
7279 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7282 * netdev_master_upper_dev_get - Get master upper device
7285 * Find a master upper device and return pointer to it or NULL in case
7286 * it's not there. The caller must hold the RTNL lock.
7288 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7290 struct netdev_adjacent *upper;
7294 if (list_empty(&dev->adj_list.upper))
7297 upper = list_first_entry(&dev->adj_list.upper,
7298 struct netdev_adjacent, list);
7299 if (likely(upper->master))
7303 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7305 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7307 struct netdev_adjacent *upper;
7311 if (list_empty(&dev->adj_list.upper))
7314 upper = list_first_entry(&dev->adj_list.upper,
7315 struct netdev_adjacent, list);
7316 if (likely(upper->master) && !upper->ignore)
7322 * netdev_has_any_lower_dev - Check if device is linked to some device
7325 * Find out if a device is linked to a lower device and return true in case
7326 * it is. The caller must hold the RTNL lock.
7328 static bool netdev_has_any_lower_dev(struct net_device *dev)
7332 return !list_empty(&dev->adj_list.lower);
7335 void *netdev_adjacent_get_private(struct list_head *adj_list)
7337 struct netdev_adjacent *adj;
7339 adj = list_entry(adj_list, struct netdev_adjacent, list);
7341 return adj->private;
7343 EXPORT_SYMBOL(netdev_adjacent_get_private);
7346 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7348 * @iter: list_head ** of the current position
7350 * Gets the next device from the dev's upper list, starting from iter
7351 * position. The caller must hold RCU read lock.
7353 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7354 struct list_head **iter)
7356 struct netdev_adjacent *upper;
7358 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7360 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7362 if (&upper->list == &dev->adj_list.upper)
7365 *iter = &upper->list;
7369 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7371 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7372 struct list_head **iter,
7375 struct netdev_adjacent *upper;
7377 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7379 if (&upper->list == &dev->adj_list.upper)
7382 *iter = &upper->list;
7383 *ignore = upper->ignore;
7388 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7389 struct list_head **iter)
7391 struct netdev_adjacent *upper;
7393 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7395 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7397 if (&upper->list == &dev->adj_list.upper)
7400 *iter = &upper->list;
7405 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7406 int (*fn)(struct net_device *dev,
7407 struct netdev_nested_priv *priv),
7408 struct netdev_nested_priv *priv)
7410 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7411 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7416 iter = &dev->adj_list.upper;
7420 ret = fn(now, priv);
7427 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7434 niter = &udev->adj_list.upper;
7435 dev_stack[cur] = now;
7436 iter_stack[cur++] = iter;
7443 next = dev_stack[--cur];
7444 niter = iter_stack[cur];
7454 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7455 int (*fn)(struct net_device *dev,
7456 struct netdev_nested_priv *priv),
7457 struct netdev_nested_priv *priv)
7459 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7460 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7464 iter = &dev->adj_list.upper;
7468 ret = fn(now, priv);
7475 udev = netdev_next_upper_dev_rcu(now, &iter);
7480 niter = &udev->adj_list.upper;
7481 dev_stack[cur] = now;
7482 iter_stack[cur++] = iter;
7489 next = dev_stack[--cur];
7490 niter = iter_stack[cur];
7499 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7501 static bool __netdev_has_upper_dev(struct net_device *dev,
7502 struct net_device *upper_dev)
7504 struct netdev_nested_priv priv = {
7506 .data = (void *)upper_dev,
7511 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7516 * netdev_lower_get_next_private - Get the next ->private from the
7517 * lower neighbour list
7519 * @iter: list_head ** of the current position
7521 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7522 * list, starting from iter position. The caller must hold either hold the
7523 * RTNL lock or its own locking that guarantees that the neighbour lower
7524 * list will remain unchanged.
7526 void *netdev_lower_get_next_private(struct net_device *dev,
7527 struct list_head **iter)
7529 struct netdev_adjacent *lower;
7531 lower = list_entry(*iter, struct netdev_adjacent, list);
7533 if (&lower->list == &dev->adj_list.lower)
7536 *iter = lower->list.next;
7538 return lower->private;
7540 EXPORT_SYMBOL(netdev_lower_get_next_private);
7543 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7544 * lower neighbour list, RCU
7547 * @iter: list_head ** of the current position
7549 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7550 * list, starting from iter position. The caller must hold RCU read lock.
7552 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7553 struct list_head **iter)
7555 struct netdev_adjacent *lower;
7557 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7559 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7561 if (&lower->list == &dev->adj_list.lower)
7564 *iter = &lower->list;
7566 return lower->private;
7568 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7571 * netdev_lower_get_next - Get the next device from the lower neighbour
7574 * @iter: list_head ** of the current position
7576 * Gets the next netdev_adjacent from the dev's lower neighbour
7577 * list, starting from iter position. The caller must hold RTNL lock or
7578 * its own locking that guarantees that the neighbour lower
7579 * list will remain unchanged.
7581 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7583 struct netdev_adjacent *lower;
7585 lower = list_entry(*iter, struct netdev_adjacent, list);
7587 if (&lower->list == &dev->adj_list.lower)
7590 *iter = lower->list.next;
7594 EXPORT_SYMBOL(netdev_lower_get_next);
7596 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7597 struct list_head **iter)
7599 struct netdev_adjacent *lower;
7601 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7603 if (&lower->list == &dev->adj_list.lower)
7606 *iter = &lower->list;
7611 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7612 struct list_head **iter,
7615 struct netdev_adjacent *lower;
7617 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7619 if (&lower->list == &dev->adj_list.lower)
7622 *iter = &lower->list;
7623 *ignore = lower->ignore;
7628 int netdev_walk_all_lower_dev(struct net_device *dev,
7629 int (*fn)(struct net_device *dev,
7630 struct netdev_nested_priv *priv),
7631 struct netdev_nested_priv *priv)
7633 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7634 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7638 iter = &dev->adj_list.lower;
7642 ret = fn(now, priv);
7649 ldev = netdev_next_lower_dev(now, &iter);
7654 niter = &ldev->adj_list.lower;
7655 dev_stack[cur] = now;
7656 iter_stack[cur++] = iter;
7663 next = dev_stack[--cur];
7664 niter = iter_stack[cur];
7673 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7675 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7676 int (*fn)(struct net_device *dev,
7677 struct netdev_nested_priv *priv),
7678 struct netdev_nested_priv *priv)
7680 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7681 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7686 iter = &dev->adj_list.lower;
7690 ret = fn(now, priv);
7697 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7704 niter = &ldev->adj_list.lower;
7705 dev_stack[cur] = now;
7706 iter_stack[cur++] = iter;
7713 next = dev_stack[--cur];
7714 niter = iter_stack[cur];
7724 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7725 struct list_head **iter)
7727 struct netdev_adjacent *lower;
7729 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7730 if (&lower->list == &dev->adj_list.lower)
7733 *iter = &lower->list;
7737 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7739 static u8 __netdev_upper_depth(struct net_device *dev)
7741 struct net_device *udev;
7742 struct list_head *iter;
7746 for (iter = &dev->adj_list.upper,
7747 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7749 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7752 if (max_depth < udev->upper_level)
7753 max_depth = udev->upper_level;
7759 static u8 __netdev_lower_depth(struct net_device *dev)
7761 struct net_device *ldev;
7762 struct list_head *iter;
7766 for (iter = &dev->adj_list.lower,
7767 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7769 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7772 if (max_depth < ldev->lower_level)
7773 max_depth = ldev->lower_level;
7779 static int __netdev_update_upper_level(struct net_device *dev,
7780 struct netdev_nested_priv *__unused)
7782 dev->upper_level = __netdev_upper_depth(dev) + 1;
7786 static int __netdev_update_lower_level(struct net_device *dev,
7787 struct netdev_nested_priv *priv)
7789 dev->lower_level = __netdev_lower_depth(dev) + 1;
7791 #ifdef CONFIG_LOCKDEP
7795 if (priv->flags & NESTED_SYNC_IMM)
7796 dev->nested_level = dev->lower_level - 1;
7797 if (priv->flags & NESTED_SYNC_TODO)
7798 net_unlink_todo(dev);
7803 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7804 int (*fn)(struct net_device *dev,
7805 struct netdev_nested_priv *priv),
7806 struct netdev_nested_priv *priv)
7808 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7809 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7813 iter = &dev->adj_list.lower;
7817 ret = fn(now, priv);
7824 ldev = netdev_next_lower_dev_rcu(now, &iter);
7829 niter = &ldev->adj_list.lower;
7830 dev_stack[cur] = now;
7831 iter_stack[cur++] = iter;
7838 next = dev_stack[--cur];
7839 niter = iter_stack[cur];
7848 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7851 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7852 * lower neighbour list, RCU
7856 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7857 * list. The caller must hold RCU read lock.
7859 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7861 struct netdev_adjacent *lower;
7863 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7864 struct netdev_adjacent, list);
7866 return lower->private;
7869 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7872 * netdev_master_upper_dev_get_rcu - Get master upper device
7875 * Find a master upper device and return pointer to it or NULL in case
7876 * it's not there. The caller must hold the RCU read lock.
7878 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7880 struct netdev_adjacent *upper;
7882 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7883 struct netdev_adjacent, list);
7884 if (upper && likely(upper->master))
7888 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7890 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7891 struct net_device *adj_dev,
7892 struct list_head *dev_list)
7894 char linkname[IFNAMSIZ+7];
7896 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7897 "upper_%s" : "lower_%s", adj_dev->name);
7898 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7901 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7903 struct list_head *dev_list)
7905 char linkname[IFNAMSIZ+7];
7907 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7908 "upper_%s" : "lower_%s", name);
7909 sysfs_remove_link(&(dev->dev.kobj), linkname);
7912 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7913 struct net_device *adj_dev,
7914 struct list_head *dev_list)
7916 return (dev_list == &dev->adj_list.upper ||
7917 dev_list == &dev->adj_list.lower) &&
7918 net_eq(dev_net(dev), dev_net(adj_dev));
7921 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7922 struct net_device *adj_dev,
7923 struct list_head *dev_list,
7924 void *private, bool master)
7926 struct netdev_adjacent *adj;
7929 adj = __netdev_find_adj(adj_dev, dev_list);
7933 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7934 dev->name, adj_dev->name, adj->ref_nr);
7939 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7944 adj->master = master;
7946 adj->private = private;
7947 adj->ignore = false;
7950 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7951 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7953 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7954 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7959 /* Ensure that master link is always the first item in list. */
7961 ret = sysfs_create_link(&(dev->dev.kobj),
7962 &(adj_dev->dev.kobj), "master");
7964 goto remove_symlinks;
7966 list_add_rcu(&adj->list, dev_list);
7968 list_add_tail_rcu(&adj->list, dev_list);
7974 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7975 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7983 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7984 struct net_device *adj_dev,
7986 struct list_head *dev_list)
7988 struct netdev_adjacent *adj;
7990 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7991 dev->name, adj_dev->name, ref_nr);
7993 adj = __netdev_find_adj(adj_dev, dev_list);
7996 pr_err("Adjacency does not exist for device %s from %s\n",
7997 dev->name, adj_dev->name);
8002 if (adj->ref_nr > ref_nr) {
8003 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8004 dev->name, adj_dev->name, ref_nr,
8005 adj->ref_nr - ref_nr);
8006 adj->ref_nr -= ref_nr;
8011 sysfs_remove_link(&(dev->dev.kobj), "master");
8013 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8014 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8016 list_del_rcu(&adj->list);
8017 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8018 adj_dev->name, dev->name, adj_dev->name);
8020 kfree_rcu(adj, rcu);
8023 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8024 struct net_device *upper_dev,
8025 struct list_head *up_list,
8026 struct list_head *down_list,
8027 void *private, bool master)
8031 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8036 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8039 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8046 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8047 struct net_device *upper_dev,
8049 struct list_head *up_list,
8050 struct list_head *down_list)
8052 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8053 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8056 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8057 struct net_device *upper_dev,
8058 void *private, bool master)
8060 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8061 &dev->adj_list.upper,
8062 &upper_dev->adj_list.lower,
8066 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8067 struct net_device *upper_dev)
8069 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8070 &dev->adj_list.upper,
8071 &upper_dev->adj_list.lower);
8074 static int __netdev_upper_dev_link(struct net_device *dev,
8075 struct net_device *upper_dev, bool master,
8076 void *upper_priv, void *upper_info,
8077 struct netdev_nested_priv *priv,
8078 struct netlink_ext_ack *extack)
8080 struct netdev_notifier_changeupper_info changeupper_info = {
8085 .upper_dev = upper_dev,
8088 .upper_info = upper_info,
8090 struct net_device *master_dev;
8095 if (dev == upper_dev)
8098 /* To prevent loops, check if dev is not upper device to upper_dev. */
8099 if (__netdev_has_upper_dev(upper_dev, dev))
8102 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8106 if (__netdev_has_upper_dev(dev, upper_dev))
8109 master_dev = __netdev_master_upper_dev_get(dev);
8111 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8114 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8115 &changeupper_info.info);
8116 ret = notifier_to_errno(ret);
8120 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8125 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8126 &changeupper_info.info);
8127 ret = notifier_to_errno(ret);
8131 __netdev_update_upper_level(dev, NULL);
8132 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8134 __netdev_update_lower_level(upper_dev, priv);
8135 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8141 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8147 * netdev_upper_dev_link - Add a link to the upper device
8149 * @upper_dev: new upper device
8150 * @extack: netlink extended ack
8152 * Adds a link to device which is upper to this one. The caller must hold
8153 * the RTNL lock. On a failure a negative errno code is returned.
8154 * On success the reference counts are adjusted and the function
8157 int netdev_upper_dev_link(struct net_device *dev,
8158 struct net_device *upper_dev,
8159 struct netlink_ext_ack *extack)
8161 struct netdev_nested_priv priv = {
8162 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8166 return __netdev_upper_dev_link(dev, upper_dev, false,
8167 NULL, NULL, &priv, extack);
8169 EXPORT_SYMBOL(netdev_upper_dev_link);
8172 * netdev_master_upper_dev_link - Add a master link to the upper device
8174 * @upper_dev: new upper device
8175 * @upper_priv: upper device private
8176 * @upper_info: upper info to be passed down via notifier
8177 * @extack: netlink extended ack
8179 * Adds a link to device which is upper to this one. In this case, only
8180 * one master upper device can be linked, although other non-master devices
8181 * might be linked as well. The caller must hold the RTNL lock.
8182 * On a failure a negative errno code is returned. On success the reference
8183 * counts are adjusted and the function returns zero.
8185 int netdev_master_upper_dev_link(struct net_device *dev,
8186 struct net_device *upper_dev,
8187 void *upper_priv, void *upper_info,
8188 struct netlink_ext_ack *extack)
8190 struct netdev_nested_priv priv = {
8191 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8195 return __netdev_upper_dev_link(dev, upper_dev, true,
8196 upper_priv, upper_info, &priv, extack);
8198 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8200 static void __netdev_upper_dev_unlink(struct net_device *dev,
8201 struct net_device *upper_dev,
8202 struct netdev_nested_priv *priv)
8204 struct netdev_notifier_changeupper_info changeupper_info = {
8208 .upper_dev = upper_dev,
8214 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8216 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8217 &changeupper_info.info);
8219 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8221 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8222 &changeupper_info.info);
8224 __netdev_update_upper_level(dev, NULL);
8225 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8227 __netdev_update_lower_level(upper_dev, priv);
8228 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8233 * netdev_upper_dev_unlink - Removes a link to upper device
8235 * @upper_dev: new upper device
8237 * Removes a link to device which is upper to this one. The caller must hold
8240 void netdev_upper_dev_unlink(struct net_device *dev,
8241 struct net_device *upper_dev)
8243 struct netdev_nested_priv priv = {
8244 .flags = NESTED_SYNC_TODO,
8248 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8250 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8252 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8253 struct net_device *lower_dev,
8256 struct netdev_adjacent *adj;
8258 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8262 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8267 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8268 struct net_device *lower_dev)
8270 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8273 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8274 struct net_device *lower_dev)
8276 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8279 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8280 struct net_device *new_dev,
8281 struct net_device *dev,
8282 struct netlink_ext_ack *extack)
8284 struct netdev_nested_priv priv = {
8293 if (old_dev && new_dev != old_dev)
8294 netdev_adjacent_dev_disable(dev, old_dev);
8295 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8298 if (old_dev && new_dev != old_dev)
8299 netdev_adjacent_dev_enable(dev, old_dev);
8305 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8307 void netdev_adjacent_change_commit(struct net_device *old_dev,
8308 struct net_device *new_dev,
8309 struct net_device *dev)
8311 struct netdev_nested_priv priv = {
8312 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8316 if (!new_dev || !old_dev)
8319 if (new_dev == old_dev)
8322 netdev_adjacent_dev_enable(dev, old_dev);
8323 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8325 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8327 void netdev_adjacent_change_abort(struct net_device *old_dev,
8328 struct net_device *new_dev,
8329 struct net_device *dev)
8331 struct netdev_nested_priv priv = {
8339 if (old_dev && new_dev != old_dev)
8340 netdev_adjacent_dev_enable(dev, old_dev);
8342 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8344 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8347 * netdev_bonding_info_change - Dispatch event about slave change
8349 * @bonding_info: info to dispatch
8351 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8352 * The caller must hold the RTNL lock.
8354 void netdev_bonding_info_change(struct net_device *dev,
8355 struct netdev_bonding_info *bonding_info)
8357 struct netdev_notifier_bonding_info info = {
8361 memcpy(&info.bonding_info, bonding_info,
8362 sizeof(struct netdev_bonding_info));
8363 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8366 EXPORT_SYMBOL(netdev_bonding_info_change);
8369 * netdev_get_xmit_slave - Get the xmit slave of master device
8372 * @all_slaves: assume all the slaves are active
8374 * The reference counters are not incremented so the caller must be
8375 * careful with locks. The caller must hold RCU lock.
8376 * %NULL is returned if no slave is found.
8379 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8380 struct sk_buff *skb,
8383 const struct net_device_ops *ops = dev->netdev_ops;
8385 if (!ops->ndo_get_xmit_slave)
8387 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8389 EXPORT_SYMBOL(netdev_get_xmit_slave);
8391 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8394 const struct net_device_ops *ops = dev->netdev_ops;
8396 if (!ops->ndo_sk_get_lower_dev)
8398 return ops->ndo_sk_get_lower_dev(dev, sk);
8402 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8406 * %NULL is returned if no lower device is found.
8409 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8412 struct net_device *lower;
8414 lower = netdev_sk_get_lower_dev(dev, sk);
8417 lower = netdev_sk_get_lower_dev(dev, sk);
8422 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8424 static void netdev_adjacent_add_links(struct net_device *dev)
8426 struct netdev_adjacent *iter;
8428 struct net *net = dev_net(dev);
8430 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8431 if (!net_eq(net, dev_net(iter->dev)))
8433 netdev_adjacent_sysfs_add(iter->dev, dev,
8434 &iter->dev->adj_list.lower);
8435 netdev_adjacent_sysfs_add(dev, iter->dev,
8436 &dev->adj_list.upper);
8439 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8440 if (!net_eq(net, dev_net(iter->dev)))
8442 netdev_adjacent_sysfs_add(iter->dev, dev,
8443 &iter->dev->adj_list.upper);
8444 netdev_adjacent_sysfs_add(dev, iter->dev,
8445 &dev->adj_list.lower);
8449 static void netdev_adjacent_del_links(struct net_device *dev)
8451 struct netdev_adjacent *iter;
8453 struct net *net = dev_net(dev);
8455 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8456 if (!net_eq(net, dev_net(iter->dev)))
8458 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8459 &iter->dev->adj_list.lower);
8460 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8461 &dev->adj_list.upper);
8464 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8465 if (!net_eq(net, dev_net(iter->dev)))
8467 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8468 &iter->dev->adj_list.upper);
8469 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8470 &dev->adj_list.lower);
8474 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8476 struct netdev_adjacent *iter;
8478 struct net *net = dev_net(dev);
8480 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8481 if (!net_eq(net, dev_net(iter->dev)))
8483 netdev_adjacent_sysfs_del(iter->dev, oldname,
8484 &iter->dev->adj_list.lower);
8485 netdev_adjacent_sysfs_add(iter->dev, dev,
8486 &iter->dev->adj_list.lower);
8489 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8490 if (!net_eq(net, dev_net(iter->dev)))
8492 netdev_adjacent_sysfs_del(iter->dev, oldname,
8493 &iter->dev->adj_list.upper);
8494 netdev_adjacent_sysfs_add(iter->dev, dev,
8495 &iter->dev->adj_list.upper);
8499 void *netdev_lower_dev_get_private(struct net_device *dev,
8500 struct net_device *lower_dev)
8502 struct netdev_adjacent *lower;
8506 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8510 return lower->private;
8512 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8516 * netdev_lower_state_changed - Dispatch event about lower device state change
8517 * @lower_dev: device
8518 * @lower_state_info: state to dispatch
8520 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8521 * The caller must hold the RTNL lock.
8523 void netdev_lower_state_changed(struct net_device *lower_dev,
8524 void *lower_state_info)
8526 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8527 .info.dev = lower_dev,
8531 changelowerstate_info.lower_state_info = lower_state_info;
8532 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8533 &changelowerstate_info.info);
8535 EXPORT_SYMBOL(netdev_lower_state_changed);
8537 static void dev_change_rx_flags(struct net_device *dev, int flags)
8539 const struct net_device_ops *ops = dev->netdev_ops;
8541 if (ops->ndo_change_rx_flags)
8542 ops->ndo_change_rx_flags(dev, flags);
8545 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8547 unsigned int old_flags = dev->flags;
8553 dev->flags |= IFF_PROMISC;
8554 dev->promiscuity += inc;
8555 if (dev->promiscuity == 0) {
8558 * If inc causes overflow, untouch promisc and return error.
8561 dev->flags &= ~IFF_PROMISC;
8563 dev->promiscuity -= inc;
8564 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8569 if (dev->flags != old_flags) {
8570 pr_info("device %s %s promiscuous mode\n",
8572 dev->flags & IFF_PROMISC ? "entered" : "left");
8573 if (audit_enabled) {
8574 current_uid_gid(&uid, &gid);
8575 audit_log(audit_context(), GFP_ATOMIC,
8576 AUDIT_ANOM_PROMISCUOUS,
8577 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8578 dev->name, (dev->flags & IFF_PROMISC),
8579 (old_flags & IFF_PROMISC),
8580 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8581 from_kuid(&init_user_ns, uid),
8582 from_kgid(&init_user_ns, gid),
8583 audit_get_sessionid(current));
8586 dev_change_rx_flags(dev, IFF_PROMISC);
8589 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8594 * dev_set_promiscuity - update promiscuity count on a device
8598 * Add or remove promiscuity from a device. While the count in the device
8599 * remains above zero the interface remains promiscuous. Once it hits zero
8600 * the device reverts back to normal filtering operation. A negative inc
8601 * value is used to drop promiscuity on the device.
8602 * Return 0 if successful or a negative errno code on error.
8604 int dev_set_promiscuity(struct net_device *dev, int inc)
8606 unsigned int old_flags = dev->flags;
8609 err = __dev_set_promiscuity(dev, inc, true);
8612 if (dev->flags != old_flags)
8613 dev_set_rx_mode(dev);
8616 EXPORT_SYMBOL(dev_set_promiscuity);
8618 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8620 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8624 dev->flags |= IFF_ALLMULTI;
8625 dev->allmulti += inc;
8626 if (dev->allmulti == 0) {
8629 * If inc causes overflow, untouch allmulti and return error.
8632 dev->flags &= ~IFF_ALLMULTI;
8634 dev->allmulti -= inc;
8635 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8640 if (dev->flags ^ old_flags) {
8641 dev_change_rx_flags(dev, IFF_ALLMULTI);
8642 dev_set_rx_mode(dev);
8644 __dev_notify_flags(dev, old_flags,
8645 dev->gflags ^ old_gflags);
8651 * dev_set_allmulti - update allmulti count on a device
8655 * Add or remove reception of all multicast frames to a device. While the
8656 * count in the device remains above zero the interface remains listening
8657 * to all interfaces. Once it hits zero the device reverts back to normal
8658 * filtering operation. A negative @inc value is used to drop the counter
8659 * when releasing a resource needing all multicasts.
8660 * Return 0 if successful or a negative errno code on error.
8663 int dev_set_allmulti(struct net_device *dev, int inc)
8665 return __dev_set_allmulti(dev, inc, true);
8667 EXPORT_SYMBOL(dev_set_allmulti);
8670 * Upload unicast and multicast address lists to device and
8671 * configure RX filtering. When the device doesn't support unicast
8672 * filtering it is put in promiscuous mode while unicast addresses
8675 void __dev_set_rx_mode(struct net_device *dev)
8677 const struct net_device_ops *ops = dev->netdev_ops;
8679 /* dev_open will call this function so the list will stay sane. */
8680 if (!(dev->flags&IFF_UP))
8683 if (!netif_device_present(dev))
8686 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8687 /* Unicast addresses changes may only happen under the rtnl,
8688 * therefore calling __dev_set_promiscuity here is safe.
8690 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8691 __dev_set_promiscuity(dev, 1, false);
8692 dev->uc_promisc = true;
8693 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8694 __dev_set_promiscuity(dev, -1, false);
8695 dev->uc_promisc = false;
8699 if (ops->ndo_set_rx_mode)
8700 ops->ndo_set_rx_mode(dev);
8703 void dev_set_rx_mode(struct net_device *dev)
8705 netif_addr_lock_bh(dev);
8706 __dev_set_rx_mode(dev);
8707 netif_addr_unlock_bh(dev);
8711 * dev_get_flags - get flags reported to userspace
8714 * Get the combination of flag bits exported through APIs to userspace.
8716 unsigned int dev_get_flags(const struct net_device *dev)
8720 flags = (dev->flags & ~(IFF_PROMISC |
8725 (dev->gflags & (IFF_PROMISC |
8728 if (netif_running(dev)) {
8729 if (netif_oper_up(dev))
8730 flags |= IFF_RUNNING;
8731 if (netif_carrier_ok(dev))
8732 flags |= IFF_LOWER_UP;
8733 if (netif_dormant(dev))
8734 flags |= IFF_DORMANT;
8739 EXPORT_SYMBOL(dev_get_flags);
8741 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8742 struct netlink_ext_ack *extack)
8744 unsigned int old_flags = dev->flags;
8750 * Set the flags on our device.
8753 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8754 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8756 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8760 * Load in the correct multicast list now the flags have changed.
8763 if ((old_flags ^ flags) & IFF_MULTICAST)
8764 dev_change_rx_flags(dev, IFF_MULTICAST);
8766 dev_set_rx_mode(dev);
8769 * Have we downed the interface. We handle IFF_UP ourselves
8770 * according to user attempts to set it, rather than blindly
8775 if ((old_flags ^ flags) & IFF_UP) {
8776 if (old_flags & IFF_UP)
8779 ret = __dev_open(dev, extack);
8782 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8783 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8784 unsigned int old_flags = dev->flags;
8786 dev->gflags ^= IFF_PROMISC;
8788 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8789 if (dev->flags != old_flags)
8790 dev_set_rx_mode(dev);
8793 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8794 * is important. Some (broken) drivers set IFF_PROMISC, when
8795 * IFF_ALLMULTI is requested not asking us and not reporting.
8797 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8798 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8800 dev->gflags ^= IFF_ALLMULTI;
8801 __dev_set_allmulti(dev, inc, false);
8807 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8808 unsigned int gchanges)
8810 unsigned int changes = dev->flags ^ old_flags;
8813 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8815 if (changes & IFF_UP) {
8816 if (dev->flags & IFF_UP)
8817 call_netdevice_notifiers(NETDEV_UP, dev);
8819 call_netdevice_notifiers(NETDEV_DOWN, dev);
8822 if (dev->flags & IFF_UP &&
8823 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8824 struct netdev_notifier_change_info change_info = {
8828 .flags_changed = changes,
8831 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8836 * dev_change_flags - change device settings
8838 * @flags: device state flags
8839 * @extack: netlink extended ack
8841 * Change settings on device based state flags. The flags are
8842 * in the userspace exported format.
8844 int dev_change_flags(struct net_device *dev, unsigned int flags,
8845 struct netlink_ext_ack *extack)
8848 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8850 ret = __dev_change_flags(dev, flags, extack);
8854 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8855 __dev_notify_flags(dev, old_flags, changes);
8858 EXPORT_SYMBOL(dev_change_flags);
8860 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8862 const struct net_device_ops *ops = dev->netdev_ops;
8864 if (ops->ndo_change_mtu)
8865 return ops->ndo_change_mtu(dev, new_mtu);
8867 /* Pairs with all the lockless reads of dev->mtu in the stack */
8868 WRITE_ONCE(dev->mtu, new_mtu);
8871 EXPORT_SYMBOL(__dev_set_mtu);
8873 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8874 struct netlink_ext_ack *extack)
8876 /* MTU must be positive, and in range */
8877 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8878 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8882 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8883 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8890 * dev_set_mtu_ext - Change maximum transfer unit
8892 * @new_mtu: new transfer unit
8893 * @extack: netlink extended ack
8895 * Change the maximum transfer size of the network device.
8897 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8898 struct netlink_ext_ack *extack)
8902 if (new_mtu == dev->mtu)
8905 err = dev_validate_mtu(dev, new_mtu, extack);
8909 if (!netif_device_present(dev))
8912 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8913 err = notifier_to_errno(err);
8917 orig_mtu = dev->mtu;
8918 err = __dev_set_mtu(dev, new_mtu);
8921 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8923 err = notifier_to_errno(err);
8925 /* setting mtu back and notifying everyone again,
8926 * so that they have a chance to revert changes.
8928 __dev_set_mtu(dev, orig_mtu);
8929 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8936 int dev_set_mtu(struct net_device *dev, int new_mtu)
8938 struct netlink_ext_ack extack;
8941 memset(&extack, 0, sizeof(extack));
8942 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8943 if (err && extack._msg)
8944 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8947 EXPORT_SYMBOL(dev_set_mtu);
8950 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8952 * @new_len: new tx queue length
8954 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8956 unsigned int orig_len = dev->tx_queue_len;
8959 if (new_len != (unsigned int)new_len)
8962 if (new_len != orig_len) {
8963 dev->tx_queue_len = new_len;
8964 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8965 res = notifier_to_errno(res);
8968 res = dev_qdisc_change_tx_queue_len(dev);
8976 netdev_err(dev, "refused to change device tx_queue_len\n");
8977 dev->tx_queue_len = orig_len;
8982 * dev_set_group - Change group this device belongs to
8984 * @new_group: group this device should belong to
8986 void dev_set_group(struct net_device *dev, int new_group)
8988 dev->group = new_group;
8990 EXPORT_SYMBOL(dev_set_group);
8993 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8995 * @addr: new address
8996 * @extack: netlink extended ack
8998 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8999 struct netlink_ext_ack *extack)
9001 struct netdev_notifier_pre_changeaddr_info info = {
9003 .info.extack = extack,
9008 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9009 return notifier_to_errno(rc);
9011 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9014 * dev_set_mac_address - Change Media Access Control Address
9017 * @extack: netlink extended ack
9019 * Change the hardware (MAC) address of the device
9021 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9022 struct netlink_ext_ack *extack)
9024 const struct net_device_ops *ops = dev->netdev_ops;
9027 if (!ops->ndo_set_mac_address)
9029 if (sa->sa_family != dev->type)
9031 if (!netif_device_present(dev))
9033 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9036 err = ops->ndo_set_mac_address(dev, sa);
9039 dev->addr_assign_type = NET_ADDR_SET;
9040 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9041 add_device_randomness(dev->dev_addr, dev->addr_len);
9044 EXPORT_SYMBOL(dev_set_mac_address);
9046 static DECLARE_RWSEM(dev_addr_sem);
9048 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9049 struct netlink_ext_ack *extack)
9053 down_write(&dev_addr_sem);
9054 ret = dev_set_mac_address(dev, sa, extack);
9055 up_write(&dev_addr_sem);
9058 EXPORT_SYMBOL(dev_set_mac_address_user);
9060 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9062 size_t size = sizeof(sa->sa_data);
9063 struct net_device *dev;
9066 down_read(&dev_addr_sem);
9069 dev = dev_get_by_name_rcu(net, dev_name);
9075 memset(sa->sa_data, 0, size);
9077 memcpy(sa->sa_data, dev->dev_addr,
9078 min_t(size_t, size, dev->addr_len));
9079 sa->sa_family = dev->type;
9083 up_read(&dev_addr_sem);
9086 EXPORT_SYMBOL(dev_get_mac_address);
9089 * dev_change_carrier - Change device carrier
9091 * @new_carrier: new value
9093 * Change device carrier
9095 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9097 const struct net_device_ops *ops = dev->netdev_ops;
9099 if (!ops->ndo_change_carrier)
9101 if (!netif_device_present(dev))
9103 return ops->ndo_change_carrier(dev, new_carrier);
9105 EXPORT_SYMBOL(dev_change_carrier);
9108 * dev_get_phys_port_id - Get device physical port ID
9112 * Get device physical port ID
9114 int dev_get_phys_port_id(struct net_device *dev,
9115 struct netdev_phys_item_id *ppid)
9117 const struct net_device_ops *ops = dev->netdev_ops;
9119 if (!ops->ndo_get_phys_port_id)
9121 return ops->ndo_get_phys_port_id(dev, ppid);
9123 EXPORT_SYMBOL(dev_get_phys_port_id);
9126 * dev_get_phys_port_name - Get device physical port name
9129 * @len: limit of bytes to copy to name
9131 * Get device physical port name
9133 int dev_get_phys_port_name(struct net_device *dev,
9134 char *name, size_t len)
9136 const struct net_device_ops *ops = dev->netdev_ops;
9139 if (ops->ndo_get_phys_port_name) {
9140 err = ops->ndo_get_phys_port_name(dev, name, len);
9141 if (err != -EOPNOTSUPP)
9144 return devlink_compat_phys_port_name_get(dev, name, len);
9146 EXPORT_SYMBOL(dev_get_phys_port_name);
9149 * dev_get_port_parent_id - Get the device's port parent identifier
9150 * @dev: network device
9151 * @ppid: pointer to a storage for the port's parent identifier
9152 * @recurse: allow/disallow recursion to lower devices
9154 * Get the devices's port parent identifier
9156 int dev_get_port_parent_id(struct net_device *dev,
9157 struct netdev_phys_item_id *ppid,
9160 const struct net_device_ops *ops = dev->netdev_ops;
9161 struct netdev_phys_item_id first = { };
9162 struct net_device *lower_dev;
9163 struct list_head *iter;
9166 if (ops->ndo_get_port_parent_id) {
9167 err = ops->ndo_get_port_parent_id(dev, ppid);
9168 if (err != -EOPNOTSUPP)
9172 err = devlink_compat_switch_id_get(dev, ppid);
9173 if (!err || err != -EOPNOTSUPP)
9179 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9180 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9185 else if (memcmp(&first, ppid, sizeof(*ppid)))
9191 EXPORT_SYMBOL(dev_get_port_parent_id);
9194 * netdev_port_same_parent_id - Indicate if two network devices have
9195 * the same port parent identifier
9196 * @a: first network device
9197 * @b: second network device
9199 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9201 struct netdev_phys_item_id a_id = { };
9202 struct netdev_phys_item_id b_id = { };
9204 if (dev_get_port_parent_id(a, &a_id, true) ||
9205 dev_get_port_parent_id(b, &b_id, true))
9208 return netdev_phys_item_id_same(&a_id, &b_id);
9210 EXPORT_SYMBOL(netdev_port_same_parent_id);
9213 * dev_change_proto_down - update protocol port state information
9215 * @proto_down: new value
9217 * This info can be used by switch drivers to set the phys state of the
9220 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9222 const struct net_device_ops *ops = dev->netdev_ops;
9224 if (!ops->ndo_change_proto_down)
9226 if (!netif_device_present(dev))
9228 return ops->ndo_change_proto_down(dev, proto_down);
9230 EXPORT_SYMBOL(dev_change_proto_down);
9233 * dev_change_proto_down_generic - generic implementation for
9234 * ndo_change_proto_down that sets carrier according to
9238 * @proto_down: new value
9240 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9243 netif_carrier_off(dev);
9245 netif_carrier_on(dev);
9246 dev->proto_down = proto_down;
9249 EXPORT_SYMBOL(dev_change_proto_down_generic);
9252 * dev_change_proto_down_reason - proto down reason
9255 * @mask: proto down mask
9256 * @value: proto down value
9258 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9264 dev->proto_down_reason = value;
9266 for_each_set_bit(b, &mask, 32) {
9267 if (value & (1 << b))
9268 dev->proto_down_reason |= BIT(b);
9270 dev->proto_down_reason &= ~BIT(b);
9274 EXPORT_SYMBOL(dev_change_proto_down_reason);
9276 struct bpf_xdp_link {
9277 struct bpf_link link;
9278 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9282 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9284 if (flags & XDP_FLAGS_HW_MODE)
9286 if (flags & XDP_FLAGS_DRV_MODE)
9287 return XDP_MODE_DRV;
9288 if (flags & XDP_FLAGS_SKB_MODE)
9289 return XDP_MODE_SKB;
9290 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9293 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9297 return generic_xdp_install;
9300 return dev->netdev_ops->ndo_bpf;
9306 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9307 enum bpf_xdp_mode mode)
9309 return dev->xdp_state[mode].link;
9312 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9313 enum bpf_xdp_mode mode)
9315 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9318 return link->link.prog;
9319 return dev->xdp_state[mode].prog;
9322 u8 dev_xdp_prog_count(struct net_device *dev)
9327 for (i = 0; i < __MAX_XDP_MODE; i++)
9328 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9332 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9334 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9336 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9338 return prog ? prog->aux->id : 0;
9341 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9342 struct bpf_xdp_link *link)
9344 dev->xdp_state[mode].link = link;
9345 dev->xdp_state[mode].prog = NULL;
9348 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9349 struct bpf_prog *prog)
9351 dev->xdp_state[mode].link = NULL;
9352 dev->xdp_state[mode].prog = prog;
9355 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9356 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9357 u32 flags, struct bpf_prog *prog)
9359 struct netdev_bpf xdp;
9362 memset(&xdp, 0, sizeof(xdp));
9363 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9364 xdp.extack = extack;
9368 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9369 * "moved" into driver), so they don't increment it on their own, but
9370 * they do decrement refcnt when program is detached or replaced.
9371 * Given net_device also owns link/prog, we need to bump refcnt here
9372 * to prevent drivers from underflowing it.
9376 err = bpf_op(dev, &xdp);
9383 if (mode != XDP_MODE_HW)
9384 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9389 static void dev_xdp_uninstall(struct net_device *dev)
9391 struct bpf_xdp_link *link;
9392 struct bpf_prog *prog;
9393 enum bpf_xdp_mode mode;
9398 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9399 prog = dev_xdp_prog(dev, mode);
9403 bpf_op = dev_xdp_bpf_op(dev, mode);
9407 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9409 /* auto-detach link from net device */
9410 link = dev_xdp_link(dev, mode);
9416 dev_xdp_set_link(dev, mode, NULL);
9420 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9421 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9422 struct bpf_prog *old_prog, u32 flags)
9424 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9425 struct bpf_prog *cur_prog;
9426 struct net_device *upper;
9427 struct list_head *iter;
9428 enum bpf_xdp_mode mode;
9434 /* either link or prog attachment, never both */
9435 if (link && (new_prog || old_prog))
9437 /* link supports only XDP mode flags */
9438 if (link && (flags & ~XDP_FLAGS_MODES)) {
9439 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9442 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9443 if (num_modes > 1) {
9444 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9447 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9448 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9449 NL_SET_ERR_MSG(extack,
9450 "More than one program loaded, unset mode is ambiguous");
9453 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9454 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9455 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9459 mode = dev_xdp_mode(dev, flags);
9460 /* can't replace attached link */
9461 if (dev_xdp_link(dev, mode)) {
9462 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9466 /* don't allow if an upper device already has a program */
9467 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9468 if (dev_xdp_prog_count(upper) > 0) {
9469 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9474 cur_prog = dev_xdp_prog(dev, mode);
9475 /* can't replace attached prog with link */
9476 if (link && cur_prog) {
9477 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9480 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9481 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9485 /* put effective new program into new_prog */
9487 new_prog = link->link.prog;
9490 bool offload = mode == XDP_MODE_HW;
9491 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9492 ? XDP_MODE_DRV : XDP_MODE_SKB;
9494 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9495 NL_SET_ERR_MSG(extack, "XDP program already attached");
9498 if (!offload && dev_xdp_prog(dev, other_mode)) {
9499 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9502 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9503 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9506 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9507 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9510 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9511 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9516 /* don't call drivers if the effective program didn't change */
9517 if (new_prog != cur_prog) {
9518 bpf_op = dev_xdp_bpf_op(dev, mode);
9520 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9524 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9530 dev_xdp_set_link(dev, mode, link);
9532 dev_xdp_set_prog(dev, mode, new_prog);
9534 bpf_prog_put(cur_prog);
9539 static int dev_xdp_attach_link(struct net_device *dev,
9540 struct netlink_ext_ack *extack,
9541 struct bpf_xdp_link *link)
9543 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9546 static int dev_xdp_detach_link(struct net_device *dev,
9547 struct netlink_ext_ack *extack,
9548 struct bpf_xdp_link *link)
9550 enum bpf_xdp_mode mode;
9555 mode = dev_xdp_mode(dev, link->flags);
9556 if (dev_xdp_link(dev, mode) != link)
9559 bpf_op = dev_xdp_bpf_op(dev, mode);
9560 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9561 dev_xdp_set_link(dev, mode, NULL);
9565 static void bpf_xdp_link_release(struct bpf_link *link)
9567 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9571 /* if racing with net_device's tear down, xdp_link->dev might be
9572 * already NULL, in which case link was already auto-detached
9574 if (xdp_link->dev) {
9575 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9576 xdp_link->dev = NULL;
9582 static int bpf_xdp_link_detach(struct bpf_link *link)
9584 bpf_xdp_link_release(link);
9588 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9590 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9595 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9596 struct seq_file *seq)
9598 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9603 ifindex = xdp_link->dev->ifindex;
9606 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9609 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9610 struct bpf_link_info *info)
9612 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9617 ifindex = xdp_link->dev->ifindex;
9620 info->xdp.ifindex = ifindex;
9624 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9625 struct bpf_prog *old_prog)
9627 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9628 enum bpf_xdp_mode mode;
9634 /* link might have been auto-released already, so fail */
9635 if (!xdp_link->dev) {
9640 if (old_prog && link->prog != old_prog) {
9644 old_prog = link->prog;
9645 if (old_prog->type != new_prog->type ||
9646 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9651 if (old_prog == new_prog) {
9652 /* no-op, don't disturb drivers */
9653 bpf_prog_put(new_prog);
9657 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9658 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9659 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9660 xdp_link->flags, new_prog);
9664 old_prog = xchg(&link->prog, new_prog);
9665 bpf_prog_put(old_prog);
9672 static const struct bpf_link_ops bpf_xdp_link_lops = {
9673 .release = bpf_xdp_link_release,
9674 .dealloc = bpf_xdp_link_dealloc,
9675 .detach = bpf_xdp_link_detach,
9676 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9677 .fill_link_info = bpf_xdp_link_fill_link_info,
9678 .update_prog = bpf_xdp_link_update,
9681 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9683 struct net *net = current->nsproxy->net_ns;
9684 struct bpf_link_primer link_primer;
9685 struct bpf_xdp_link *link;
9686 struct net_device *dev;
9690 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9696 link = kzalloc(sizeof(*link), GFP_USER);
9702 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9704 link->flags = attr->link_create.flags;
9706 err = bpf_link_prime(&link->link, &link_primer);
9712 err = dev_xdp_attach_link(dev, NULL, link);
9717 bpf_link_cleanup(&link_primer);
9721 fd = bpf_link_settle(&link_primer);
9722 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9735 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9737 * @extack: netlink extended ack
9738 * @fd: new program fd or negative value to clear
9739 * @expected_fd: old program fd that userspace expects to replace or clear
9740 * @flags: xdp-related flags
9742 * Set or clear a bpf program for a device
9744 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9745 int fd, int expected_fd, u32 flags)
9747 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9748 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9754 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9755 mode != XDP_MODE_SKB);
9756 if (IS_ERR(new_prog))
9757 return PTR_ERR(new_prog);
9760 if (expected_fd >= 0) {
9761 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9762 mode != XDP_MODE_SKB);
9763 if (IS_ERR(old_prog)) {
9764 err = PTR_ERR(old_prog);
9770 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9773 if (err && new_prog)
9774 bpf_prog_put(new_prog);
9776 bpf_prog_put(old_prog);
9781 * dev_new_index - allocate an ifindex
9782 * @net: the applicable net namespace
9784 * Returns a suitable unique value for a new device interface
9785 * number. The caller must hold the rtnl semaphore or the
9786 * dev_base_lock to be sure it remains unique.
9788 static int dev_new_index(struct net *net)
9790 int ifindex = net->ifindex;
9795 if (!__dev_get_by_index(net, ifindex))
9796 return net->ifindex = ifindex;
9800 /* Delayed registration/unregisteration */
9801 static LIST_HEAD(net_todo_list);
9802 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9804 static void net_set_todo(struct net_device *dev)
9806 list_add_tail(&dev->todo_list, &net_todo_list);
9807 dev_net(dev)->dev_unreg_count++;
9810 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9811 struct net_device *upper, netdev_features_t features)
9813 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9814 netdev_features_t feature;
9817 for_each_netdev_feature(upper_disables, feature_bit) {
9818 feature = __NETIF_F_BIT(feature_bit);
9819 if (!(upper->wanted_features & feature)
9820 && (features & feature)) {
9821 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9822 &feature, upper->name);
9823 features &= ~feature;
9830 static void netdev_sync_lower_features(struct net_device *upper,
9831 struct net_device *lower, netdev_features_t features)
9833 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9834 netdev_features_t feature;
9837 for_each_netdev_feature(upper_disables, feature_bit) {
9838 feature = __NETIF_F_BIT(feature_bit);
9839 if (!(features & feature) && (lower->features & feature)) {
9840 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9841 &feature, lower->name);
9842 lower->wanted_features &= ~feature;
9843 __netdev_update_features(lower);
9845 if (unlikely(lower->features & feature))
9846 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9847 &feature, lower->name);
9849 netdev_features_change(lower);
9854 static netdev_features_t netdev_fix_features(struct net_device *dev,
9855 netdev_features_t features)
9857 /* Fix illegal checksum combinations */
9858 if ((features & NETIF_F_HW_CSUM) &&
9859 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9860 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9861 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9864 /* TSO requires that SG is present as well. */
9865 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9866 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9867 features &= ~NETIF_F_ALL_TSO;
9870 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9871 !(features & NETIF_F_IP_CSUM)) {
9872 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9873 features &= ~NETIF_F_TSO;
9874 features &= ~NETIF_F_TSO_ECN;
9877 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9878 !(features & NETIF_F_IPV6_CSUM)) {
9879 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9880 features &= ~NETIF_F_TSO6;
9883 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9884 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9885 features &= ~NETIF_F_TSO_MANGLEID;
9887 /* TSO ECN requires that TSO is present as well. */
9888 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9889 features &= ~NETIF_F_TSO_ECN;
9891 /* Software GSO depends on SG. */
9892 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9893 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9894 features &= ~NETIF_F_GSO;
9897 /* GSO partial features require GSO partial be set */
9898 if ((features & dev->gso_partial_features) &&
9899 !(features & NETIF_F_GSO_PARTIAL)) {
9901 "Dropping partially supported GSO features since no GSO partial.\n");
9902 features &= ~dev->gso_partial_features;
9905 if (!(features & NETIF_F_RXCSUM)) {
9906 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9907 * successfully merged by hardware must also have the
9908 * checksum verified by hardware. If the user does not
9909 * want to enable RXCSUM, logically, we should disable GRO_HW.
9911 if (features & NETIF_F_GRO_HW) {
9912 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9913 features &= ~NETIF_F_GRO_HW;
9917 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9918 if (features & NETIF_F_RXFCS) {
9919 if (features & NETIF_F_LRO) {
9920 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9921 features &= ~NETIF_F_LRO;
9924 if (features & NETIF_F_GRO_HW) {
9925 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9926 features &= ~NETIF_F_GRO_HW;
9930 if (features & NETIF_F_HW_TLS_TX) {
9931 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9932 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9933 bool hw_csum = features & NETIF_F_HW_CSUM;
9935 if (!ip_csum && !hw_csum) {
9936 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9937 features &= ~NETIF_F_HW_TLS_TX;
9941 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9942 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9943 features &= ~NETIF_F_HW_TLS_RX;
9949 int __netdev_update_features(struct net_device *dev)
9951 struct net_device *upper, *lower;
9952 netdev_features_t features;
9953 struct list_head *iter;
9958 features = netdev_get_wanted_features(dev);
9960 if (dev->netdev_ops->ndo_fix_features)
9961 features = dev->netdev_ops->ndo_fix_features(dev, features);
9963 /* driver might be less strict about feature dependencies */
9964 features = netdev_fix_features(dev, features);
9966 /* some features can't be enabled if they're off on an upper device */
9967 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9968 features = netdev_sync_upper_features(dev, upper, features);
9970 if (dev->features == features)
9973 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9974 &dev->features, &features);
9976 if (dev->netdev_ops->ndo_set_features)
9977 err = dev->netdev_ops->ndo_set_features(dev, features);
9981 if (unlikely(err < 0)) {
9983 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9984 err, &features, &dev->features);
9985 /* return non-0 since some features might have changed and
9986 * it's better to fire a spurious notification than miss it
9992 /* some features must be disabled on lower devices when disabled
9993 * on an upper device (think: bonding master or bridge)
9995 netdev_for_each_lower_dev(dev, lower, iter)
9996 netdev_sync_lower_features(dev, lower, features);
9999 netdev_features_t diff = features ^ dev->features;
10001 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10002 /* udp_tunnel_{get,drop}_rx_info both need
10003 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10004 * device, or they won't do anything.
10005 * Thus we need to update dev->features
10006 * *before* calling udp_tunnel_get_rx_info,
10007 * but *after* calling udp_tunnel_drop_rx_info.
10009 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10010 dev->features = features;
10011 udp_tunnel_get_rx_info(dev);
10013 udp_tunnel_drop_rx_info(dev);
10017 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10018 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10019 dev->features = features;
10020 err |= vlan_get_rx_ctag_filter_info(dev);
10022 vlan_drop_rx_ctag_filter_info(dev);
10026 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10027 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10028 dev->features = features;
10029 err |= vlan_get_rx_stag_filter_info(dev);
10031 vlan_drop_rx_stag_filter_info(dev);
10035 dev->features = features;
10038 return err < 0 ? 0 : 1;
10042 * netdev_update_features - recalculate device features
10043 * @dev: the device to check
10045 * Recalculate dev->features set and send notifications if it
10046 * has changed. Should be called after driver or hardware dependent
10047 * conditions might have changed that influence the features.
10049 void netdev_update_features(struct net_device *dev)
10051 if (__netdev_update_features(dev))
10052 netdev_features_change(dev);
10054 EXPORT_SYMBOL(netdev_update_features);
10057 * netdev_change_features - recalculate device features
10058 * @dev: the device to check
10060 * Recalculate dev->features set and send notifications even
10061 * if they have not changed. Should be called instead of
10062 * netdev_update_features() if also dev->vlan_features might
10063 * have changed to allow the changes to be propagated to stacked
10066 void netdev_change_features(struct net_device *dev)
10068 __netdev_update_features(dev);
10069 netdev_features_change(dev);
10071 EXPORT_SYMBOL(netdev_change_features);
10074 * netif_stacked_transfer_operstate - transfer operstate
10075 * @rootdev: the root or lower level device to transfer state from
10076 * @dev: the device to transfer operstate to
10078 * Transfer operational state from root to device. This is normally
10079 * called when a stacking relationship exists between the root
10080 * device and the device(a leaf device).
10082 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10083 struct net_device *dev)
10085 if (rootdev->operstate == IF_OPER_DORMANT)
10086 netif_dormant_on(dev);
10088 netif_dormant_off(dev);
10090 if (rootdev->operstate == IF_OPER_TESTING)
10091 netif_testing_on(dev);
10093 netif_testing_off(dev);
10095 if (netif_carrier_ok(rootdev))
10096 netif_carrier_on(dev);
10098 netif_carrier_off(dev);
10100 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10102 static int netif_alloc_rx_queues(struct net_device *dev)
10104 unsigned int i, count = dev->num_rx_queues;
10105 struct netdev_rx_queue *rx;
10106 size_t sz = count * sizeof(*rx);
10111 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10117 for (i = 0; i < count; i++) {
10120 /* XDP RX-queue setup */
10121 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10128 /* Rollback successful reg's and free other resources */
10130 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10136 static void netif_free_rx_queues(struct net_device *dev)
10138 unsigned int i, count = dev->num_rx_queues;
10140 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10144 for (i = 0; i < count; i++)
10145 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10150 static void netdev_init_one_queue(struct net_device *dev,
10151 struct netdev_queue *queue, void *_unused)
10153 /* Initialize queue lock */
10154 spin_lock_init(&queue->_xmit_lock);
10155 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10156 queue->xmit_lock_owner = -1;
10157 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10160 dql_init(&queue->dql, HZ);
10164 static void netif_free_tx_queues(struct net_device *dev)
10169 static int netif_alloc_netdev_queues(struct net_device *dev)
10171 unsigned int count = dev->num_tx_queues;
10172 struct netdev_queue *tx;
10173 size_t sz = count * sizeof(*tx);
10175 if (count < 1 || count > 0xffff)
10178 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10184 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10185 spin_lock_init(&dev->tx_global_lock);
10190 void netif_tx_stop_all_queues(struct net_device *dev)
10194 for (i = 0; i < dev->num_tx_queues; i++) {
10195 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10197 netif_tx_stop_queue(txq);
10200 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10203 * register_netdevice - register a network device
10204 * @dev: device to register
10206 * Take a completed network device structure and add it to the kernel
10207 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10208 * chain. 0 is returned on success. A negative errno code is returned
10209 * on a failure to set up the device, or if the name is a duplicate.
10211 * Callers must hold the rtnl semaphore. You may want
10212 * register_netdev() instead of this.
10215 * The locking appears insufficient to guarantee two parallel registers
10216 * will not get the same name.
10219 int register_netdevice(struct net_device *dev)
10222 struct net *net = dev_net(dev);
10224 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10225 NETDEV_FEATURE_COUNT);
10226 BUG_ON(dev_boot_phase);
10231 /* When net_device's are persistent, this will be fatal. */
10232 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10235 ret = ethtool_check_ops(dev->ethtool_ops);
10239 spin_lock_init(&dev->addr_list_lock);
10240 netdev_set_addr_lockdep_class(dev);
10242 ret = dev_get_valid_name(net, dev, dev->name);
10247 dev->name_node = netdev_name_node_head_alloc(dev);
10248 if (!dev->name_node)
10251 /* Init, if this function is available */
10252 if (dev->netdev_ops->ndo_init) {
10253 ret = dev->netdev_ops->ndo_init(dev);
10257 goto err_free_name;
10261 if (((dev->hw_features | dev->features) &
10262 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10263 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10264 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10265 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10272 dev->ifindex = dev_new_index(net);
10273 else if (__dev_get_by_index(net, dev->ifindex))
10276 /* Transfer changeable features to wanted_features and enable
10277 * software offloads (GSO and GRO).
10279 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10280 dev->features |= NETIF_F_SOFT_FEATURES;
10282 if (dev->udp_tunnel_nic_info) {
10283 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10284 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10287 dev->wanted_features = dev->features & dev->hw_features;
10289 if (!(dev->flags & IFF_LOOPBACK))
10290 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10292 /* If IPv4 TCP segmentation offload is supported we should also
10293 * allow the device to enable segmenting the frame with the option
10294 * of ignoring a static IP ID value. This doesn't enable the
10295 * feature itself but allows the user to enable it later.
10297 if (dev->hw_features & NETIF_F_TSO)
10298 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10299 if (dev->vlan_features & NETIF_F_TSO)
10300 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10301 if (dev->mpls_features & NETIF_F_TSO)
10302 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10303 if (dev->hw_enc_features & NETIF_F_TSO)
10304 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10306 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10308 dev->vlan_features |= NETIF_F_HIGHDMA;
10310 /* Make NETIF_F_SG inheritable to tunnel devices.
10312 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10314 /* Make NETIF_F_SG inheritable to MPLS.
10316 dev->mpls_features |= NETIF_F_SG;
10318 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10319 ret = notifier_to_errno(ret);
10323 ret = netdev_register_kobject(dev);
10324 write_lock(&dev_base_lock);
10325 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10326 write_unlock(&dev_base_lock);
10330 __netdev_update_features(dev);
10333 * Default initial state at registry is that the
10334 * device is present.
10337 set_bit(__LINK_STATE_PRESENT, &dev->state);
10339 linkwatch_init_dev(dev);
10341 dev_init_scheduler(dev);
10343 list_netdevice(dev);
10344 add_device_randomness(dev->dev_addr, dev->addr_len);
10346 /* If the device has permanent device address, driver should
10347 * set dev_addr and also addr_assign_type should be set to
10348 * NET_ADDR_PERM (default value).
10350 if (dev->addr_assign_type == NET_ADDR_PERM)
10351 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10353 /* Notify protocols, that a new device appeared. */
10354 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10355 ret = notifier_to_errno(ret);
10357 /* Expect explicit free_netdev() on failure */
10358 dev->needs_free_netdev = false;
10359 unregister_netdevice_queue(dev, NULL);
10363 * Prevent userspace races by waiting until the network
10364 * device is fully setup before sending notifications.
10366 if (!dev->rtnl_link_ops ||
10367 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10368 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10374 if (dev->netdev_ops->ndo_uninit)
10375 dev->netdev_ops->ndo_uninit(dev);
10376 if (dev->priv_destructor)
10377 dev->priv_destructor(dev);
10379 netdev_name_node_free(dev->name_node);
10382 EXPORT_SYMBOL(register_netdevice);
10385 * init_dummy_netdev - init a dummy network device for NAPI
10386 * @dev: device to init
10388 * This takes a network device structure and initialize the minimum
10389 * amount of fields so it can be used to schedule NAPI polls without
10390 * registering a full blown interface. This is to be used by drivers
10391 * that need to tie several hardware interfaces to a single NAPI
10392 * poll scheduler due to HW limitations.
10394 int init_dummy_netdev(struct net_device *dev)
10396 /* Clear everything. Note we don't initialize spinlocks
10397 * are they aren't supposed to be taken by any of the
10398 * NAPI code and this dummy netdev is supposed to be
10399 * only ever used for NAPI polls
10401 memset(dev, 0, sizeof(struct net_device));
10403 /* make sure we BUG if trying to hit standard
10404 * register/unregister code path
10406 dev->reg_state = NETREG_DUMMY;
10408 /* NAPI wants this */
10409 INIT_LIST_HEAD(&dev->napi_list);
10411 /* a dummy interface is started by default */
10412 set_bit(__LINK_STATE_PRESENT, &dev->state);
10413 set_bit(__LINK_STATE_START, &dev->state);
10415 /* napi_busy_loop stats accounting wants this */
10416 dev_net_set(dev, &init_net);
10418 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10419 * because users of this 'device' dont need to change
10425 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10429 * register_netdev - register a network device
10430 * @dev: device to register
10432 * Take a completed network device structure and add it to the kernel
10433 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10434 * chain. 0 is returned on success. A negative errno code is returned
10435 * on a failure to set up the device, or if the name is a duplicate.
10437 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10438 * and expands the device name if you passed a format string to
10441 int register_netdev(struct net_device *dev)
10445 if (rtnl_lock_killable())
10447 err = register_netdevice(dev);
10451 EXPORT_SYMBOL(register_netdev);
10453 int netdev_refcnt_read(const struct net_device *dev)
10455 #ifdef CONFIG_PCPU_DEV_REFCNT
10458 for_each_possible_cpu(i)
10459 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10462 return refcount_read(&dev->dev_refcnt);
10465 EXPORT_SYMBOL(netdev_refcnt_read);
10467 int netdev_unregister_timeout_secs __read_mostly = 10;
10469 #define WAIT_REFS_MIN_MSECS 1
10470 #define WAIT_REFS_MAX_MSECS 250
10472 * netdev_wait_allrefs - wait until all references are gone.
10473 * @dev: target net_device
10475 * This is called when unregistering network devices.
10477 * Any protocol or device that holds a reference should register
10478 * for netdevice notification, and cleanup and put back the
10479 * reference if they receive an UNREGISTER event.
10480 * We can get stuck here if buggy protocols don't correctly
10483 static void netdev_wait_allrefs(struct net_device *dev)
10485 unsigned long rebroadcast_time, warning_time;
10486 int wait = 0, refcnt;
10488 rebroadcast_time = warning_time = jiffies;
10489 refcnt = netdev_refcnt_read(dev);
10491 while (refcnt != 1) {
10492 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10495 /* Rebroadcast unregister notification */
10496 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10502 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10504 /* We must not have linkwatch events
10505 * pending on unregister. If this
10506 * happens, we simply run the queue
10507 * unscheduled, resulting in a noop
10510 linkwatch_run_queue();
10515 rebroadcast_time = jiffies;
10520 wait = WAIT_REFS_MIN_MSECS;
10523 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10526 refcnt = netdev_refcnt_read(dev);
10529 time_after(jiffies, warning_time +
10530 netdev_unregister_timeout_secs * HZ)) {
10531 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10532 dev->name, refcnt);
10533 warning_time = jiffies;
10538 /* The sequence is:
10542 * register_netdevice(x1);
10543 * register_netdevice(x2);
10545 * unregister_netdevice(y1);
10546 * unregister_netdevice(y2);
10552 * We are invoked by rtnl_unlock().
10553 * This allows us to deal with problems:
10554 * 1) We can delete sysfs objects which invoke hotplug
10555 * without deadlocking with linkwatch via keventd.
10556 * 2) Since we run with the RTNL semaphore not held, we can sleep
10557 * safely in order to wait for the netdev refcnt to drop to zero.
10559 * We must not return until all unregister events added during
10560 * the interval the lock was held have been completed.
10562 void netdev_run_todo(void)
10564 struct list_head list;
10565 #ifdef CONFIG_LOCKDEP
10566 struct list_head unlink_list;
10568 list_replace_init(&net_unlink_list, &unlink_list);
10570 while (!list_empty(&unlink_list)) {
10571 struct net_device *dev = list_first_entry(&unlink_list,
10574 list_del_init(&dev->unlink_list);
10575 dev->nested_level = dev->lower_level - 1;
10579 /* Snapshot list, allow later requests */
10580 list_replace_init(&net_todo_list, &list);
10585 /* Wait for rcu callbacks to finish before next phase */
10586 if (!list_empty(&list))
10589 while (!list_empty(&list)) {
10590 struct net_device *dev
10591 = list_first_entry(&list, struct net_device, todo_list);
10592 list_del(&dev->todo_list);
10594 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10595 pr_err("network todo '%s' but state %d\n",
10596 dev->name, dev->reg_state);
10601 write_lock(&dev_base_lock);
10602 dev->reg_state = NETREG_UNREGISTERED;
10603 write_unlock(&dev_base_lock);
10604 linkwatch_forget_dev(dev);
10606 netdev_wait_allrefs(dev);
10609 BUG_ON(netdev_refcnt_read(dev) != 1);
10610 BUG_ON(!list_empty(&dev->ptype_all));
10611 BUG_ON(!list_empty(&dev->ptype_specific));
10612 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10613 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10614 #if IS_ENABLED(CONFIG_DECNET)
10615 WARN_ON(dev->dn_ptr);
10617 if (dev->priv_destructor)
10618 dev->priv_destructor(dev);
10619 if (dev->needs_free_netdev)
10622 /* Report a network device has been unregistered */
10624 dev_net(dev)->dev_unreg_count--;
10626 wake_up(&netdev_unregistering_wq);
10628 /* Free network device */
10629 kobject_put(&dev->dev.kobj);
10633 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10634 * all the same fields in the same order as net_device_stats, with only
10635 * the type differing, but rtnl_link_stats64 may have additional fields
10636 * at the end for newer counters.
10638 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10639 const struct net_device_stats *netdev_stats)
10641 #if BITS_PER_LONG == 64
10642 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10643 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10644 /* zero out counters that only exist in rtnl_link_stats64 */
10645 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10646 sizeof(*stats64) - sizeof(*netdev_stats));
10648 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10649 const unsigned long *src = (const unsigned long *)netdev_stats;
10650 u64 *dst = (u64 *)stats64;
10652 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10653 for (i = 0; i < n; i++)
10655 /* zero out counters that only exist in rtnl_link_stats64 */
10656 memset((char *)stats64 + n * sizeof(u64), 0,
10657 sizeof(*stats64) - n * sizeof(u64));
10660 EXPORT_SYMBOL(netdev_stats_to_stats64);
10663 * dev_get_stats - get network device statistics
10664 * @dev: device to get statistics from
10665 * @storage: place to store stats
10667 * Get network statistics from device. Return @storage.
10668 * The device driver may provide its own method by setting
10669 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10670 * otherwise the internal statistics structure is used.
10672 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10673 struct rtnl_link_stats64 *storage)
10675 const struct net_device_ops *ops = dev->netdev_ops;
10677 if (ops->ndo_get_stats64) {
10678 memset(storage, 0, sizeof(*storage));
10679 ops->ndo_get_stats64(dev, storage);
10680 } else if (ops->ndo_get_stats) {
10681 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10683 netdev_stats_to_stats64(storage, &dev->stats);
10685 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10686 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10687 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10690 EXPORT_SYMBOL(dev_get_stats);
10693 * dev_fetch_sw_netstats - get per-cpu network device statistics
10694 * @s: place to store stats
10695 * @netstats: per-cpu network stats to read from
10697 * Read per-cpu network statistics and populate the related fields in @s.
10699 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10700 const struct pcpu_sw_netstats __percpu *netstats)
10704 for_each_possible_cpu(cpu) {
10705 const struct pcpu_sw_netstats *stats;
10706 struct pcpu_sw_netstats tmp;
10707 unsigned int start;
10709 stats = per_cpu_ptr(netstats, cpu);
10711 start = u64_stats_fetch_begin_irq(&stats->syncp);
10712 tmp.rx_packets = stats->rx_packets;
10713 tmp.rx_bytes = stats->rx_bytes;
10714 tmp.tx_packets = stats->tx_packets;
10715 tmp.tx_bytes = stats->tx_bytes;
10716 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10718 s->rx_packets += tmp.rx_packets;
10719 s->rx_bytes += tmp.rx_bytes;
10720 s->tx_packets += tmp.tx_packets;
10721 s->tx_bytes += tmp.tx_bytes;
10724 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10727 * dev_get_tstats64 - ndo_get_stats64 implementation
10728 * @dev: device to get statistics from
10729 * @s: place to store stats
10731 * Populate @s from dev->stats and dev->tstats. Can be used as
10732 * ndo_get_stats64() callback.
10734 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10736 netdev_stats_to_stats64(s, &dev->stats);
10737 dev_fetch_sw_netstats(s, dev->tstats);
10739 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10741 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10743 struct netdev_queue *queue = dev_ingress_queue(dev);
10745 #ifdef CONFIG_NET_CLS_ACT
10748 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10751 netdev_init_one_queue(dev, queue, NULL);
10752 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10753 queue->qdisc_sleeping = &noop_qdisc;
10754 rcu_assign_pointer(dev->ingress_queue, queue);
10759 static const struct ethtool_ops default_ethtool_ops;
10761 void netdev_set_default_ethtool_ops(struct net_device *dev,
10762 const struct ethtool_ops *ops)
10764 if (dev->ethtool_ops == &default_ethtool_ops)
10765 dev->ethtool_ops = ops;
10767 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10769 void netdev_freemem(struct net_device *dev)
10771 char *addr = (char *)dev - dev->padded;
10777 * alloc_netdev_mqs - allocate network device
10778 * @sizeof_priv: size of private data to allocate space for
10779 * @name: device name format string
10780 * @name_assign_type: origin of device name
10781 * @setup: callback to initialize device
10782 * @txqs: the number of TX subqueues to allocate
10783 * @rxqs: the number of RX subqueues to allocate
10785 * Allocates a struct net_device with private data area for driver use
10786 * and performs basic initialization. Also allocates subqueue structs
10787 * for each queue on the device.
10789 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10790 unsigned char name_assign_type,
10791 void (*setup)(struct net_device *),
10792 unsigned int txqs, unsigned int rxqs)
10794 struct net_device *dev;
10795 unsigned int alloc_size;
10796 struct net_device *p;
10798 BUG_ON(strlen(name) >= sizeof(dev->name));
10801 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10806 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10810 alloc_size = sizeof(struct net_device);
10812 /* ensure 32-byte alignment of private area */
10813 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10814 alloc_size += sizeof_priv;
10816 /* ensure 32-byte alignment of whole construct */
10817 alloc_size += NETDEV_ALIGN - 1;
10819 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10823 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10824 dev->padded = (char *)dev - (char *)p;
10826 #ifdef CONFIG_PCPU_DEV_REFCNT
10827 dev->pcpu_refcnt = alloc_percpu(int);
10828 if (!dev->pcpu_refcnt)
10832 refcount_set(&dev->dev_refcnt, 1);
10835 if (dev_addr_init(dev))
10841 dev_net_set(dev, &init_net);
10843 dev->gso_max_size = GSO_MAX_SIZE;
10844 dev->gso_max_segs = GSO_MAX_SEGS;
10845 dev->upper_level = 1;
10846 dev->lower_level = 1;
10847 #ifdef CONFIG_LOCKDEP
10848 dev->nested_level = 0;
10849 INIT_LIST_HEAD(&dev->unlink_list);
10852 INIT_LIST_HEAD(&dev->napi_list);
10853 INIT_LIST_HEAD(&dev->unreg_list);
10854 INIT_LIST_HEAD(&dev->close_list);
10855 INIT_LIST_HEAD(&dev->link_watch_list);
10856 INIT_LIST_HEAD(&dev->adj_list.upper);
10857 INIT_LIST_HEAD(&dev->adj_list.lower);
10858 INIT_LIST_HEAD(&dev->ptype_all);
10859 INIT_LIST_HEAD(&dev->ptype_specific);
10860 INIT_LIST_HEAD(&dev->net_notifier_list);
10861 #ifdef CONFIG_NET_SCHED
10862 hash_init(dev->qdisc_hash);
10864 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10867 if (!dev->tx_queue_len) {
10868 dev->priv_flags |= IFF_NO_QUEUE;
10869 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10872 dev->num_tx_queues = txqs;
10873 dev->real_num_tx_queues = txqs;
10874 if (netif_alloc_netdev_queues(dev))
10877 dev->num_rx_queues = rxqs;
10878 dev->real_num_rx_queues = rxqs;
10879 if (netif_alloc_rx_queues(dev))
10882 strcpy(dev->name, name);
10883 dev->name_assign_type = name_assign_type;
10884 dev->group = INIT_NETDEV_GROUP;
10885 if (!dev->ethtool_ops)
10886 dev->ethtool_ops = &default_ethtool_ops;
10888 nf_hook_ingress_init(dev);
10897 #ifdef CONFIG_PCPU_DEV_REFCNT
10898 free_percpu(dev->pcpu_refcnt);
10901 netdev_freemem(dev);
10904 EXPORT_SYMBOL(alloc_netdev_mqs);
10907 * free_netdev - free network device
10910 * This function does the last stage of destroying an allocated device
10911 * interface. The reference to the device object is released. If this
10912 * is the last reference then it will be freed.Must be called in process
10915 void free_netdev(struct net_device *dev)
10917 struct napi_struct *p, *n;
10921 /* When called immediately after register_netdevice() failed the unwind
10922 * handling may still be dismantling the device. Handle that case by
10923 * deferring the free.
10925 if (dev->reg_state == NETREG_UNREGISTERING) {
10927 dev->needs_free_netdev = true;
10931 netif_free_tx_queues(dev);
10932 netif_free_rx_queues(dev);
10934 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10936 /* Flush device addresses */
10937 dev_addr_flush(dev);
10939 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10942 #ifdef CONFIG_PCPU_DEV_REFCNT
10943 free_percpu(dev->pcpu_refcnt);
10944 dev->pcpu_refcnt = NULL;
10946 free_percpu(dev->xdp_bulkq);
10947 dev->xdp_bulkq = NULL;
10949 /* Compatibility with error handling in drivers */
10950 if (dev->reg_state == NETREG_UNINITIALIZED) {
10951 netdev_freemem(dev);
10955 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10956 dev->reg_state = NETREG_RELEASED;
10958 /* will free via device release */
10959 put_device(&dev->dev);
10961 EXPORT_SYMBOL(free_netdev);
10964 * synchronize_net - Synchronize with packet receive processing
10966 * Wait for packets currently being received to be done.
10967 * Does not block later packets from starting.
10969 void synchronize_net(void)
10972 if (rtnl_is_locked())
10973 synchronize_rcu_expedited();
10977 EXPORT_SYMBOL(synchronize_net);
10980 * unregister_netdevice_queue - remove device from the kernel
10984 * This function shuts down a device interface and removes it
10985 * from the kernel tables.
10986 * If head not NULL, device is queued to be unregistered later.
10988 * Callers must hold the rtnl semaphore. You may want
10989 * unregister_netdev() instead of this.
10992 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10997 list_move_tail(&dev->unreg_list, head);
11001 list_add(&dev->unreg_list, &single);
11002 unregister_netdevice_many(&single);
11005 EXPORT_SYMBOL(unregister_netdevice_queue);
11008 * unregister_netdevice_many - unregister many devices
11009 * @head: list of devices
11011 * Note: As most callers use a stack allocated list_head,
11012 * we force a list_del() to make sure stack wont be corrupted later.
11014 void unregister_netdevice_many(struct list_head *head)
11016 struct net_device *dev, *tmp;
11017 LIST_HEAD(close_head);
11019 BUG_ON(dev_boot_phase);
11022 if (list_empty(head))
11025 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11026 /* Some devices call without registering
11027 * for initialization unwind. Remove those
11028 * devices and proceed with the remaining.
11030 if (dev->reg_state == NETREG_UNINITIALIZED) {
11031 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11035 list_del(&dev->unreg_list);
11038 dev->dismantle = true;
11039 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11042 /* If device is running, close it first. */
11043 list_for_each_entry(dev, head, unreg_list)
11044 list_add_tail(&dev->close_list, &close_head);
11045 dev_close_many(&close_head, true);
11047 list_for_each_entry(dev, head, unreg_list) {
11048 /* And unlink it from device chain. */
11049 write_lock(&dev_base_lock);
11050 unlist_netdevice(dev, false);
11051 dev->reg_state = NETREG_UNREGISTERING;
11052 write_unlock(&dev_base_lock);
11054 flush_all_backlogs();
11058 list_for_each_entry(dev, head, unreg_list) {
11059 struct sk_buff *skb = NULL;
11061 /* Shutdown queueing discipline. */
11064 dev_xdp_uninstall(dev);
11066 /* Notify protocols, that we are about to destroy
11067 * this device. They should clean all the things.
11069 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11071 if (!dev->rtnl_link_ops ||
11072 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11073 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11074 GFP_KERNEL, NULL, 0);
11077 * Flush the unicast and multicast chains
11082 netdev_name_node_alt_flush(dev);
11083 netdev_name_node_free(dev->name_node);
11085 if (dev->netdev_ops->ndo_uninit)
11086 dev->netdev_ops->ndo_uninit(dev);
11089 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11091 /* Notifier chain MUST detach us all upper devices. */
11092 WARN_ON(netdev_has_any_upper_dev(dev));
11093 WARN_ON(netdev_has_any_lower_dev(dev));
11095 /* Remove entries from kobject tree */
11096 netdev_unregister_kobject(dev);
11098 /* Remove XPS queueing entries */
11099 netif_reset_xps_queues_gt(dev, 0);
11105 list_for_each_entry(dev, head, unreg_list) {
11112 EXPORT_SYMBOL(unregister_netdevice_many);
11115 * unregister_netdev - remove device from the kernel
11118 * This function shuts down a device interface and removes it
11119 * from the kernel tables.
11121 * This is just a wrapper for unregister_netdevice that takes
11122 * the rtnl semaphore. In general you want to use this and not
11123 * unregister_netdevice.
11125 void unregister_netdev(struct net_device *dev)
11128 unregister_netdevice(dev);
11131 EXPORT_SYMBOL(unregister_netdev);
11134 * __dev_change_net_namespace - move device to different nethost namespace
11136 * @net: network namespace
11137 * @pat: If not NULL name pattern to try if the current device name
11138 * is already taken in the destination network namespace.
11139 * @new_ifindex: If not zero, specifies device index in the target
11142 * This function shuts down a device interface and moves it
11143 * to a new network namespace. On success 0 is returned, on
11144 * a failure a netagive errno code is returned.
11146 * Callers must hold the rtnl semaphore.
11149 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11150 const char *pat, int new_ifindex)
11152 struct net *net_old = dev_net(dev);
11157 /* Don't allow namespace local devices to be moved. */
11159 if (dev->features & NETIF_F_NETNS_LOCAL)
11162 /* Ensure the device has been registrered */
11163 if (dev->reg_state != NETREG_REGISTERED)
11166 /* Get out if there is nothing todo */
11168 if (net_eq(net_old, net))
11171 /* Pick the destination device name, and ensure
11172 * we can use it in the destination network namespace.
11175 if (__dev_get_by_name(net, dev->name)) {
11176 /* We get here if we can't use the current device name */
11179 err = dev_get_valid_name(net, dev, pat);
11184 /* Check that new_ifindex isn't used yet. */
11186 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11190 * And now a mini version of register_netdevice unregister_netdevice.
11193 /* If device is running close it first. */
11196 /* And unlink it from device chain */
11197 unlist_netdevice(dev, true);
11201 /* Shutdown queueing discipline. */
11204 /* Notify protocols, that we are about to destroy
11205 * this device. They should clean all the things.
11207 * Note that dev->reg_state stays at NETREG_REGISTERED.
11208 * This is wanted because this way 8021q and macvlan know
11209 * the device is just moving and can keep their slaves up.
11211 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11214 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11215 /* If there is an ifindex conflict assign a new one */
11216 if (!new_ifindex) {
11217 if (__dev_get_by_index(net, dev->ifindex))
11218 new_ifindex = dev_new_index(net);
11220 new_ifindex = dev->ifindex;
11223 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11227 * Flush the unicast and multicast chains
11232 /* Send a netdev-removed uevent to the old namespace */
11233 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11234 netdev_adjacent_del_links(dev);
11236 /* Move per-net netdevice notifiers that are following the netdevice */
11237 move_netdevice_notifiers_dev_net(dev, net);
11239 /* Actually switch the network namespace */
11240 dev_net_set(dev, net);
11241 dev->ifindex = new_ifindex;
11243 /* Send a netdev-add uevent to the new namespace */
11244 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11245 netdev_adjacent_add_links(dev);
11247 /* Fixup kobjects */
11248 err = device_rename(&dev->dev, dev->name);
11251 /* Adapt owner in case owning user namespace of target network
11252 * namespace is different from the original one.
11254 err = netdev_change_owner(dev, net_old, net);
11257 /* Add the device back in the hashes */
11258 list_netdevice(dev);
11260 /* Notify protocols, that a new device appeared. */
11261 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11264 * Prevent userspace races by waiting until the network
11265 * device is fully setup before sending notifications.
11267 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11274 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11276 static int dev_cpu_dead(unsigned int oldcpu)
11278 struct sk_buff **list_skb;
11279 struct sk_buff *skb;
11281 struct softnet_data *sd, *oldsd, *remsd = NULL;
11283 local_irq_disable();
11284 cpu = smp_processor_id();
11285 sd = &per_cpu(softnet_data, cpu);
11286 oldsd = &per_cpu(softnet_data, oldcpu);
11288 /* Find end of our completion_queue. */
11289 list_skb = &sd->completion_queue;
11291 list_skb = &(*list_skb)->next;
11292 /* Append completion queue from offline CPU. */
11293 *list_skb = oldsd->completion_queue;
11294 oldsd->completion_queue = NULL;
11296 /* Append output queue from offline CPU. */
11297 if (oldsd->output_queue) {
11298 *sd->output_queue_tailp = oldsd->output_queue;
11299 sd->output_queue_tailp = oldsd->output_queue_tailp;
11300 oldsd->output_queue = NULL;
11301 oldsd->output_queue_tailp = &oldsd->output_queue;
11303 /* Append NAPI poll list from offline CPU, with one exception :
11304 * process_backlog() must be called by cpu owning percpu backlog.
11305 * We properly handle process_queue & input_pkt_queue later.
11307 while (!list_empty(&oldsd->poll_list)) {
11308 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11309 struct napi_struct,
11312 list_del_init(&napi->poll_list);
11313 if (napi->poll == process_backlog)
11316 ____napi_schedule(sd, napi);
11319 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11320 local_irq_enable();
11323 remsd = oldsd->rps_ipi_list;
11324 oldsd->rps_ipi_list = NULL;
11326 /* send out pending IPI's on offline CPU */
11327 net_rps_send_ipi(remsd);
11329 /* Process offline CPU's input_pkt_queue */
11330 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11332 input_queue_head_incr(oldsd);
11334 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11336 input_queue_head_incr(oldsd);
11343 * netdev_increment_features - increment feature set by one
11344 * @all: current feature set
11345 * @one: new feature set
11346 * @mask: mask feature set
11348 * Computes a new feature set after adding a device with feature set
11349 * @one to the master device with current feature set @all. Will not
11350 * enable anything that is off in @mask. Returns the new feature set.
11352 netdev_features_t netdev_increment_features(netdev_features_t all,
11353 netdev_features_t one, netdev_features_t mask)
11355 if (mask & NETIF_F_HW_CSUM)
11356 mask |= NETIF_F_CSUM_MASK;
11357 mask |= NETIF_F_VLAN_CHALLENGED;
11359 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11360 all &= one | ~NETIF_F_ALL_FOR_ALL;
11362 /* If one device supports hw checksumming, set for all. */
11363 if (all & NETIF_F_HW_CSUM)
11364 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11368 EXPORT_SYMBOL(netdev_increment_features);
11370 static struct hlist_head * __net_init netdev_create_hash(void)
11373 struct hlist_head *hash;
11375 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11377 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11378 INIT_HLIST_HEAD(&hash[i]);
11383 /* Initialize per network namespace state */
11384 static int __net_init netdev_init(struct net *net)
11386 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11387 8 * sizeof_field(struct napi_struct, gro_bitmask));
11389 INIT_LIST_HEAD(&net->dev_base_head);
11391 net->dev_name_head = netdev_create_hash();
11392 if (net->dev_name_head == NULL)
11395 net->dev_index_head = netdev_create_hash();
11396 if (net->dev_index_head == NULL)
11399 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11404 kfree(net->dev_name_head);
11410 * netdev_drivername - network driver for the device
11411 * @dev: network device
11413 * Determine network driver for device.
11415 const char *netdev_drivername(const struct net_device *dev)
11417 const struct device_driver *driver;
11418 const struct device *parent;
11419 const char *empty = "";
11421 parent = dev->dev.parent;
11425 driver = parent->driver;
11426 if (driver && driver->name)
11427 return driver->name;
11431 static void __netdev_printk(const char *level, const struct net_device *dev,
11432 struct va_format *vaf)
11434 if (dev && dev->dev.parent) {
11435 dev_printk_emit(level[1] - '0',
11438 dev_driver_string(dev->dev.parent),
11439 dev_name(dev->dev.parent),
11440 netdev_name(dev), netdev_reg_state(dev),
11443 printk("%s%s%s: %pV",
11444 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11446 printk("%s(NULL net_device): %pV", level, vaf);
11450 void netdev_printk(const char *level, const struct net_device *dev,
11451 const char *format, ...)
11453 struct va_format vaf;
11456 va_start(args, format);
11461 __netdev_printk(level, dev, &vaf);
11465 EXPORT_SYMBOL(netdev_printk);
11467 #define define_netdev_printk_level(func, level) \
11468 void func(const struct net_device *dev, const char *fmt, ...) \
11470 struct va_format vaf; \
11473 va_start(args, fmt); \
11478 __netdev_printk(level, dev, &vaf); \
11482 EXPORT_SYMBOL(func);
11484 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11485 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11486 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11487 define_netdev_printk_level(netdev_err, KERN_ERR);
11488 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11489 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11490 define_netdev_printk_level(netdev_info, KERN_INFO);
11492 static void __net_exit netdev_exit(struct net *net)
11494 kfree(net->dev_name_head);
11495 kfree(net->dev_index_head);
11496 if (net != &init_net)
11497 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11500 static struct pernet_operations __net_initdata netdev_net_ops = {
11501 .init = netdev_init,
11502 .exit = netdev_exit,
11505 static void __net_exit default_device_exit(struct net *net)
11507 struct net_device *dev, *aux;
11509 * Push all migratable network devices back to the
11510 * initial network namespace
11513 for_each_netdev_safe(net, dev, aux) {
11515 char fb_name[IFNAMSIZ];
11517 /* Ignore unmoveable devices (i.e. loopback) */
11518 if (dev->features & NETIF_F_NETNS_LOCAL)
11521 /* Leave virtual devices for the generic cleanup */
11522 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11525 /* Push remaining network devices to init_net */
11526 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11527 if (__dev_get_by_name(&init_net, fb_name))
11528 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11529 err = dev_change_net_namespace(dev, &init_net, fb_name);
11531 pr_emerg("%s: failed to move %s to init_net: %d\n",
11532 __func__, dev->name, err);
11539 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11541 /* Return with the rtnl_lock held when there are no network
11542 * devices unregistering in any network namespace in net_list.
11545 bool unregistering;
11546 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11548 add_wait_queue(&netdev_unregistering_wq, &wait);
11550 unregistering = false;
11552 list_for_each_entry(net, net_list, exit_list) {
11553 if (net->dev_unreg_count > 0) {
11554 unregistering = true;
11558 if (!unregistering)
11562 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11564 remove_wait_queue(&netdev_unregistering_wq, &wait);
11567 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11569 /* At exit all network devices most be removed from a network
11570 * namespace. Do this in the reverse order of registration.
11571 * Do this across as many network namespaces as possible to
11572 * improve batching efficiency.
11574 struct net_device *dev;
11576 LIST_HEAD(dev_kill_list);
11578 /* To prevent network device cleanup code from dereferencing
11579 * loopback devices or network devices that have been freed
11580 * wait here for all pending unregistrations to complete,
11581 * before unregistring the loopback device and allowing the
11582 * network namespace be freed.
11584 * The netdev todo list containing all network devices
11585 * unregistrations that happen in default_device_exit_batch
11586 * will run in the rtnl_unlock() at the end of
11587 * default_device_exit_batch.
11589 rtnl_lock_unregistering(net_list);
11590 list_for_each_entry(net, net_list, exit_list) {
11591 for_each_netdev_reverse(net, dev) {
11592 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11593 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11595 unregister_netdevice_queue(dev, &dev_kill_list);
11598 unregister_netdevice_many(&dev_kill_list);
11602 static struct pernet_operations __net_initdata default_device_ops = {
11603 .exit = default_device_exit,
11604 .exit_batch = default_device_exit_batch,
11608 * Initialize the DEV module. At boot time this walks the device list and
11609 * unhooks any devices that fail to initialise (normally hardware not
11610 * present) and leaves us with a valid list of present and active devices.
11615 * This is called single threaded during boot, so no need
11616 * to take the rtnl semaphore.
11618 static int __init net_dev_init(void)
11620 int i, rc = -ENOMEM;
11622 BUG_ON(!dev_boot_phase);
11624 if (dev_proc_init())
11627 if (netdev_kobject_init())
11630 INIT_LIST_HEAD(&ptype_all);
11631 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11632 INIT_LIST_HEAD(&ptype_base[i]);
11634 INIT_LIST_HEAD(&offload_base);
11636 if (register_pernet_subsys(&netdev_net_ops))
11640 * Initialise the packet receive queues.
11643 for_each_possible_cpu(i) {
11644 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11645 struct softnet_data *sd = &per_cpu(softnet_data, i);
11647 INIT_WORK(flush, flush_backlog);
11649 skb_queue_head_init(&sd->input_pkt_queue);
11650 skb_queue_head_init(&sd->process_queue);
11651 #ifdef CONFIG_XFRM_OFFLOAD
11652 skb_queue_head_init(&sd->xfrm_backlog);
11654 INIT_LIST_HEAD(&sd->poll_list);
11655 sd->output_queue_tailp = &sd->output_queue;
11657 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11661 init_gro_hash(&sd->backlog);
11662 sd->backlog.poll = process_backlog;
11663 sd->backlog.weight = weight_p;
11666 dev_boot_phase = 0;
11668 /* The loopback device is special if any other network devices
11669 * is present in a network namespace the loopback device must
11670 * be present. Since we now dynamically allocate and free the
11671 * loopback device ensure this invariant is maintained by
11672 * keeping the loopback device as the first device on the
11673 * list of network devices. Ensuring the loopback devices
11674 * is the first device that appears and the last network device
11677 if (register_pernet_device(&loopback_net_ops))
11680 if (register_pernet_device(&default_device_ops))
11683 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11684 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11686 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11687 NULL, dev_cpu_dead);
11694 subsys_initcall(net_dev_init);