1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_ingress.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
154 #include "net-sysfs.h"
156 #define MAX_GRO_SKBS 8
158 /* This should be increased if a protocol with a bigger head is added. */
159 #define GRO_MAX_HEAD (MAX_HEADER + 128)
161 static DEFINE_SPINLOCK(ptype_lock);
162 static DEFINE_SPINLOCK(offload_lock);
163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
164 struct list_head ptype_all __read_mostly; /* Taps */
165 static struct list_head offload_base __read_mostly;
167 static int netif_rx_internal(struct sk_buff *skb);
168 static int call_netdevice_notifiers_info(unsigned long val,
169 struct netdev_notifier_info *info);
170 static int call_netdevice_notifiers_extack(unsigned long val,
171 struct net_device *dev,
172 struct netlink_ext_ack *extack);
173 static struct napi_struct *napi_by_id(unsigned int napi_id);
176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base_head list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 static DEFINE_MUTEX(ifalias_mutex);
199 /* protects napi_hash addition/deletion and napi_gen_id */
200 static DEFINE_SPINLOCK(napi_hash_lock);
202 static unsigned int napi_gen_id = NR_CPUS;
203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
205 static DECLARE_RWSEM(devnet_rename_sem);
207 static inline void dev_base_seq_inc(struct net *net)
209 while (++net->dev_base_seq == 0)
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
225 static inline void rps_lock(struct softnet_data *sd)
228 spin_lock(&sd->input_pkt_queue.lock);
232 static inline void rps_unlock(struct softnet_data *sd)
235 spin_unlock(&sd->input_pkt_queue.lock);
239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
242 struct netdev_name_node *name_node;
244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
247 INIT_HLIST_NODE(&name_node->hlist);
248 name_node->dev = dev;
249 name_node->name = name;
253 static struct netdev_name_node *
254 netdev_name_node_head_alloc(struct net_device *dev)
256 struct netdev_name_node *name_node;
258 name_node = netdev_name_node_alloc(dev, dev->name);
261 INIT_LIST_HEAD(&name_node->list);
265 static void netdev_name_node_free(struct netdev_name_node *name_node)
270 static void netdev_name_node_add(struct net *net,
271 struct netdev_name_node *name_node)
273 hlist_add_head_rcu(&name_node->hlist,
274 dev_name_hash(net, name_node->name));
277 static void netdev_name_node_del(struct netdev_name_node *name_node)
279 hlist_del_rcu(&name_node->hlist);
282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
285 struct hlist_head *head = dev_name_hash(net, name);
286 struct netdev_name_node *name_node;
288 hlist_for_each_entry(name_node, head, hlist)
289 if (!strcmp(name_node->name, name))
294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
297 struct hlist_head *head = dev_name_hash(net, name);
298 struct netdev_name_node *name_node;
300 hlist_for_each_entry_rcu(name_node, head, hlist)
301 if (!strcmp(name_node->name, name))
306 bool netdev_name_in_use(struct net *net, const char *name)
308 return netdev_name_node_lookup(net, name);
310 EXPORT_SYMBOL(netdev_name_in_use);
312 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
314 struct netdev_name_node *name_node;
315 struct net *net = dev_net(dev);
317 name_node = netdev_name_node_lookup(net, name);
320 name_node = netdev_name_node_alloc(dev, name);
323 netdev_name_node_add(net, name_node);
324 /* The node that holds dev->name acts as a head of per-device list. */
325 list_add_tail(&name_node->list, &dev->name_node->list);
329 EXPORT_SYMBOL(netdev_name_node_alt_create);
331 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
333 list_del(&name_node->list);
334 netdev_name_node_del(name_node);
335 kfree(name_node->name);
336 netdev_name_node_free(name_node);
339 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
341 struct netdev_name_node *name_node;
342 struct net *net = dev_net(dev);
344 name_node = netdev_name_node_lookup(net, name);
347 /* lookup might have found our primary name or a name belonging
350 if (name_node == dev->name_node || name_node->dev != dev)
353 __netdev_name_node_alt_destroy(name_node);
357 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
359 static void netdev_name_node_alt_flush(struct net_device *dev)
361 struct netdev_name_node *name_node, *tmp;
363 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
364 __netdev_name_node_alt_destroy(name_node);
367 /* Device list insertion */
368 static void list_netdevice(struct net_device *dev)
370 struct net *net = dev_net(dev);
374 write_lock(&dev_base_lock);
375 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
376 netdev_name_node_add(net, dev->name_node);
377 hlist_add_head_rcu(&dev->index_hlist,
378 dev_index_hash(net, dev->ifindex));
379 write_unlock(&dev_base_lock);
381 dev_base_seq_inc(net);
384 /* Device list removal
385 * caller must respect a RCU grace period before freeing/reusing dev
387 static void unlist_netdevice(struct net_device *dev, bool lock)
391 /* Unlink dev from the device chain */
393 write_lock(&dev_base_lock);
394 list_del_rcu(&dev->dev_list);
395 netdev_name_node_del(dev->name_node);
396 hlist_del_rcu(&dev->index_hlist);
398 write_unlock(&dev_base_lock);
400 dev_base_seq_inc(dev_net(dev));
407 static RAW_NOTIFIER_HEAD(netdev_chain);
410 * Device drivers call our routines to queue packets here. We empty the
411 * queue in the local softnet handler.
414 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
415 EXPORT_PER_CPU_SYMBOL(softnet_data);
417 #ifdef CONFIG_LOCKDEP
419 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
420 * according to dev->type
422 static const unsigned short netdev_lock_type[] = {
423 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
424 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
425 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
426 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
427 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
428 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
429 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
430 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
431 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
432 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
433 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
434 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
435 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
436 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
437 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
439 static const char *const netdev_lock_name[] = {
440 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
441 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
442 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
443 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
444 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
445 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
446 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
447 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
448 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
449 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
450 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
451 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
452 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
453 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
454 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
456 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
457 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
459 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
463 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
464 if (netdev_lock_type[i] == dev_type)
466 /* the last key is used by default */
467 return ARRAY_SIZE(netdev_lock_type) - 1;
470 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
471 unsigned short dev_type)
475 i = netdev_lock_pos(dev_type);
476 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
477 netdev_lock_name[i]);
480 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
484 i = netdev_lock_pos(dev->type);
485 lockdep_set_class_and_name(&dev->addr_list_lock,
486 &netdev_addr_lock_key[i],
487 netdev_lock_name[i]);
490 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
491 unsigned short dev_type)
495 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
500 /*******************************************************************************
502 * Protocol management and registration routines
504 *******************************************************************************/
508 * Add a protocol ID to the list. Now that the input handler is
509 * smarter we can dispense with all the messy stuff that used to be
512 * BEWARE!!! Protocol handlers, mangling input packets,
513 * MUST BE last in hash buckets and checking protocol handlers
514 * MUST start from promiscuous ptype_all chain in net_bh.
515 * It is true now, do not change it.
516 * Explanation follows: if protocol handler, mangling packet, will
517 * be the first on list, it is not able to sense, that packet
518 * is cloned and should be copied-on-write, so that it will
519 * change it and subsequent readers will get broken packet.
523 static inline struct list_head *ptype_head(const struct packet_type *pt)
525 if (pt->type == htons(ETH_P_ALL))
526 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
528 return pt->dev ? &pt->dev->ptype_specific :
529 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
533 * dev_add_pack - add packet handler
534 * @pt: packet type declaration
536 * Add a protocol handler to the networking stack. The passed &packet_type
537 * is linked into kernel lists and may not be freed until it has been
538 * removed from the kernel lists.
540 * This call does not sleep therefore it can not
541 * guarantee all CPU's that are in middle of receiving packets
542 * will see the new packet type (until the next received packet).
545 void dev_add_pack(struct packet_type *pt)
547 struct list_head *head = ptype_head(pt);
549 spin_lock(&ptype_lock);
550 list_add_rcu(&pt->list, head);
551 spin_unlock(&ptype_lock);
553 EXPORT_SYMBOL(dev_add_pack);
556 * __dev_remove_pack - remove packet handler
557 * @pt: packet type declaration
559 * Remove a protocol handler that was previously added to the kernel
560 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
561 * from the kernel lists and can be freed or reused once this function
564 * The packet type might still be in use by receivers
565 * and must not be freed until after all the CPU's have gone
566 * through a quiescent state.
568 void __dev_remove_pack(struct packet_type *pt)
570 struct list_head *head = ptype_head(pt);
571 struct packet_type *pt1;
573 spin_lock(&ptype_lock);
575 list_for_each_entry(pt1, head, list) {
577 list_del_rcu(&pt->list);
582 pr_warn("dev_remove_pack: %p not found\n", pt);
584 spin_unlock(&ptype_lock);
586 EXPORT_SYMBOL(__dev_remove_pack);
589 * dev_remove_pack - remove packet handler
590 * @pt: packet type declaration
592 * Remove a protocol handler that was previously added to the kernel
593 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
594 * from the kernel lists and can be freed or reused once this function
597 * This call sleeps to guarantee that no CPU is looking at the packet
600 void dev_remove_pack(struct packet_type *pt)
602 __dev_remove_pack(pt);
606 EXPORT_SYMBOL(dev_remove_pack);
610 * dev_add_offload - register offload handlers
611 * @po: protocol offload declaration
613 * Add protocol offload handlers to the networking stack. The passed
614 * &proto_offload is linked into kernel lists and may not be freed until
615 * it has been removed from the kernel lists.
617 * This call does not sleep therefore it can not
618 * guarantee all CPU's that are in middle of receiving packets
619 * will see the new offload handlers (until the next received packet).
621 void dev_add_offload(struct packet_offload *po)
623 struct packet_offload *elem;
625 spin_lock(&offload_lock);
626 list_for_each_entry(elem, &offload_base, list) {
627 if (po->priority < elem->priority)
630 list_add_rcu(&po->list, elem->list.prev);
631 spin_unlock(&offload_lock);
633 EXPORT_SYMBOL(dev_add_offload);
636 * __dev_remove_offload - remove offload handler
637 * @po: packet offload declaration
639 * Remove a protocol offload handler that was previously added to the
640 * kernel offload handlers by dev_add_offload(). The passed &offload_type
641 * is removed from the kernel lists and can be freed or reused once this
644 * The packet type might still be in use by receivers
645 * and must not be freed until after all the CPU's have gone
646 * through a quiescent state.
648 static void __dev_remove_offload(struct packet_offload *po)
650 struct list_head *head = &offload_base;
651 struct packet_offload *po1;
653 spin_lock(&offload_lock);
655 list_for_each_entry(po1, head, list) {
657 list_del_rcu(&po->list);
662 pr_warn("dev_remove_offload: %p not found\n", po);
664 spin_unlock(&offload_lock);
668 * dev_remove_offload - remove packet offload handler
669 * @po: packet offload declaration
671 * Remove a packet offload handler that was previously added to the kernel
672 * offload handlers by dev_add_offload(). The passed &offload_type is
673 * removed from the kernel lists and can be freed or reused once this
676 * This call sleeps to guarantee that no CPU is looking at the packet
679 void dev_remove_offload(struct packet_offload *po)
681 __dev_remove_offload(po);
685 EXPORT_SYMBOL(dev_remove_offload);
687 /*******************************************************************************
689 * Device Interface Subroutines
691 *******************************************************************************/
694 * dev_get_iflink - get 'iflink' value of a interface
695 * @dev: targeted interface
697 * Indicates the ifindex the interface is linked to.
698 * Physical interfaces have the same 'ifindex' and 'iflink' values.
701 int dev_get_iflink(const struct net_device *dev)
703 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
704 return dev->netdev_ops->ndo_get_iflink(dev);
708 EXPORT_SYMBOL(dev_get_iflink);
711 * dev_fill_metadata_dst - Retrieve tunnel egress information.
712 * @dev: targeted interface
715 * For better visibility of tunnel traffic OVS needs to retrieve
716 * egress tunnel information for a packet. Following API allows
717 * user to get this info.
719 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
721 struct ip_tunnel_info *info;
723 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
726 info = skb_tunnel_info_unclone(skb);
729 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
732 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
734 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
736 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
738 int k = stack->num_paths++;
740 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
743 return &stack->path[k];
746 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
747 struct net_device_path_stack *stack)
749 const struct net_device *last_dev;
750 struct net_device_path_ctx ctx = {
753 struct net_device_path *path;
756 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
757 stack->num_paths = 0;
758 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
760 path = dev_fwd_path(stack);
764 memset(path, 0, sizeof(struct net_device_path));
765 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
769 if (WARN_ON_ONCE(last_dev == ctx.dev))
772 path = dev_fwd_path(stack);
775 path->type = DEV_PATH_ETHERNET;
780 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
783 * __dev_get_by_name - find a device by its name
784 * @net: the applicable net namespace
785 * @name: name to find
787 * Find an interface by name. Must be called under RTNL semaphore
788 * or @dev_base_lock. If the name is found a pointer to the device
789 * is returned. If the name is not found then %NULL is returned. The
790 * reference counters are not incremented so the caller must be
791 * careful with locks.
794 struct net_device *__dev_get_by_name(struct net *net, const char *name)
796 struct netdev_name_node *node_name;
798 node_name = netdev_name_node_lookup(net, name);
799 return node_name ? node_name->dev : NULL;
801 EXPORT_SYMBOL(__dev_get_by_name);
804 * dev_get_by_name_rcu - find a device by its name
805 * @net: the applicable net namespace
806 * @name: name to find
808 * Find an interface by name.
809 * If the name is found a pointer to the device is returned.
810 * If the name is not found then %NULL is returned.
811 * The reference counters are not incremented so the caller must be
812 * careful with locks. The caller must hold RCU lock.
815 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
817 struct netdev_name_node *node_name;
819 node_name = netdev_name_node_lookup_rcu(net, name);
820 return node_name ? node_name->dev : NULL;
822 EXPORT_SYMBOL(dev_get_by_name_rcu);
825 * dev_get_by_name - find a device by its name
826 * @net: the applicable net namespace
827 * @name: name to find
829 * Find an interface by name. This can be called from any
830 * context and does its own locking. The returned handle has
831 * the usage count incremented and the caller must use dev_put() to
832 * release it when it is no longer needed. %NULL is returned if no
833 * matching device is found.
836 struct net_device *dev_get_by_name(struct net *net, const char *name)
838 struct net_device *dev;
841 dev = dev_get_by_name_rcu(net, name);
846 EXPORT_SYMBOL(dev_get_by_name);
849 * __dev_get_by_index - find a device by its ifindex
850 * @net: the applicable net namespace
851 * @ifindex: index of device
853 * Search for an interface by index. Returns %NULL if the device
854 * is not found or a pointer to the device. The device has not
855 * had its reference counter increased so the caller must be careful
856 * about locking. The caller must hold either the RTNL semaphore
860 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
862 struct net_device *dev;
863 struct hlist_head *head = dev_index_hash(net, ifindex);
865 hlist_for_each_entry(dev, head, index_hlist)
866 if (dev->ifindex == ifindex)
871 EXPORT_SYMBOL(__dev_get_by_index);
874 * dev_get_by_index_rcu - find a device by its ifindex
875 * @net: the applicable net namespace
876 * @ifindex: index of device
878 * Search for an interface by index. Returns %NULL if the device
879 * is not found or a pointer to the device. The device has not
880 * had its reference counter increased so the caller must be careful
881 * about locking. The caller must hold RCU lock.
884 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
886 struct net_device *dev;
887 struct hlist_head *head = dev_index_hash(net, ifindex);
889 hlist_for_each_entry_rcu(dev, head, index_hlist)
890 if (dev->ifindex == ifindex)
895 EXPORT_SYMBOL(dev_get_by_index_rcu);
899 * dev_get_by_index - find a device by its ifindex
900 * @net: the applicable net namespace
901 * @ifindex: index of device
903 * Search for an interface by index. Returns NULL if the device
904 * is not found or a pointer to the device. The device returned has
905 * had a reference added and the pointer is safe until the user calls
906 * dev_put to indicate they have finished with it.
909 struct net_device *dev_get_by_index(struct net *net, int ifindex)
911 struct net_device *dev;
914 dev = dev_get_by_index_rcu(net, ifindex);
919 EXPORT_SYMBOL(dev_get_by_index);
922 * dev_get_by_napi_id - find a device by napi_id
923 * @napi_id: ID of the NAPI struct
925 * Search for an interface by NAPI ID. Returns %NULL if the device
926 * is not found or a pointer to the device. The device has not had
927 * its reference counter increased so the caller must be careful
928 * about locking. The caller must hold RCU lock.
931 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
933 struct napi_struct *napi;
935 WARN_ON_ONCE(!rcu_read_lock_held());
937 if (napi_id < MIN_NAPI_ID)
940 napi = napi_by_id(napi_id);
942 return napi ? napi->dev : NULL;
944 EXPORT_SYMBOL(dev_get_by_napi_id);
947 * netdev_get_name - get a netdevice name, knowing its ifindex.
948 * @net: network namespace
949 * @name: a pointer to the buffer where the name will be stored.
950 * @ifindex: the ifindex of the interface to get the name from.
952 int netdev_get_name(struct net *net, char *name, int ifindex)
954 struct net_device *dev;
957 down_read(&devnet_rename_sem);
960 dev = dev_get_by_index_rcu(net, ifindex);
966 strcpy(name, dev->name);
971 up_read(&devnet_rename_sem);
976 * dev_getbyhwaddr_rcu - find a device by its hardware address
977 * @net: the applicable net namespace
978 * @type: media type of device
979 * @ha: hardware address
981 * Search for an interface by MAC address. Returns NULL if the device
982 * is not found or a pointer to the device.
983 * The caller must hold RCU or RTNL.
984 * The returned device has not had its ref count increased
985 * and the caller must therefore be careful about locking
989 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
992 struct net_device *dev;
994 for_each_netdev_rcu(net, dev)
995 if (dev->type == type &&
996 !memcmp(dev->dev_addr, ha, dev->addr_len))
1001 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1003 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1005 struct net_device *dev, *ret = NULL;
1008 for_each_netdev_rcu(net, dev)
1009 if (dev->type == type) {
1017 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1020 * __dev_get_by_flags - find any device with given flags
1021 * @net: the applicable net namespace
1022 * @if_flags: IFF_* values
1023 * @mask: bitmask of bits in if_flags to check
1025 * Search for any interface with the given flags. Returns NULL if a device
1026 * is not found or a pointer to the device. Must be called inside
1027 * rtnl_lock(), and result refcount is unchanged.
1030 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1031 unsigned short mask)
1033 struct net_device *dev, *ret;
1038 for_each_netdev(net, dev) {
1039 if (((dev->flags ^ if_flags) & mask) == 0) {
1046 EXPORT_SYMBOL(__dev_get_by_flags);
1049 * dev_valid_name - check if name is okay for network device
1050 * @name: name string
1052 * Network device names need to be valid file names to
1053 * allow sysfs to work. We also disallow any kind of
1056 bool dev_valid_name(const char *name)
1060 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1062 if (!strcmp(name, ".") || !strcmp(name, ".."))
1066 if (*name == '/' || *name == ':' || isspace(*name))
1072 EXPORT_SYMBOL(dev_valid_name);
1075 * __dev_alloc_name - allocate a name for a device
1076 * @net: network namespace to allocate the device name in
1077 * @name: name format string
1078 * @buf: scratch buffer and result name string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1093 const int max_netdevices = 8*PAGE_SIZE;
1094 unsigned long *inuse;
1095 struct net_device *d;
1097 if (!dev_valid_name(name))
1100 p = strchr(name, '%');
1103 * Verify the string as this thing may have come from
1104 * the user. There must be either one "%d" and no other "%"
1107 if (p[1] != 'd' || strchr(p + 2, '%'))
1110 /* Use one page as a bit array of possible slots */
1111 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1115 for_each_netdev(net, d) {
1116 struct netdev_name_node *name_node;
1117 list_for_each_entry(name_node, &d->name_node->list, list) {
1118 if (!sscanf(name_node->name, name, &i))
1120 if (i < 0 || i >= max_netdevices)
1123 /* avoid cases where sscanf is not exact inverse of printf */
1124 snprintf(buf, IFNAMSIZ, name, i);
1125 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1128 if (!sscanf(d->name, name, &i))
1130 if (i < 0 || i >= max_netdevices)
1133 /* avoid cases where sscanf is not exact inverse of printf */
1134 snprintf(buf, IFNAMSIZ, name, i);
1135 if (!strncmp(buf, d->name, IFNAMSIZ))
1139 i = find_first_zero_bit(inuse, max_netdevices);
1140 free_page((unsigned long) inuse);
1143 snprintf(buf, IFNAMSIZ, name, i);
1144 if (!netdev_name_in_use(net, buf))
1147 /* It is possible to run out of possible slots
1148 * when the name is long and there isn't enough space left
1149 * for the digits, or if all bits are used.
1154 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1155 const char *want_name, char *out_name)
1159 if (!dev_valid_name(want_name))
1162 if (strchr(want_name, '%')) {
1163 ret = __dev_alloc_name(net, want_name, out_name);
1164 return ret < 0 ? ret : 0;
1165 } else if (netdev_name_in_use(net, want_name)) {
1167 } else if (out_name != want_name) {
1168 strscpy(out_name, want_name, IFNAMSIZ);
1174 static int dev_alloc_name_ns(struct net *net,
1175 struct net_device *dev,
1182 ret = __dev_alloc_name(net, name, buf);
1184 strscpy(dev->name, buf, IFNAMSIZ);
1189 * dev_alloc_name - allocate a name for a device
1191 * @name: name format string
1193 * Passed a format string - eg "lt%d" it will try and find a suitable
1194 * id. It scans list of devices to build up a free map, then chooses
1195 * the first empty slot. The caller must hold the dev_base or rtnl lock
1196 * while allocating the name and adding the device in order to avoid
1198 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1199 * Returns the number of the unit assigned or a negative errno code.
1202 int dev_alloc_name(struct net_device *dev, const char *name)
1204 return dev_alloc_name_ns(dev_net(dev), dev, name);
1206 EXPORT_SYMBOL(dev_alloc_name);
1208 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1214 ret = dev_prep_valid_name(net, dev, name, buf);
1216 strscpy(dev->name, buf, IFNAMSIZ);
1221 * dev_change_name - change name of a device
1223 * @newname: name (or format string) must be at least IFNAMSIZ
1225 * Change name of a device, can pass format strings "eth%d".
1228 int dev_change_name(struct net_device *dev, const char *newname)
1230 unsigned char old_assign_type;
1231 char oldname[IFNAMSIZ];
1237 BUG_ON(!dev_net(dev));
1241 /* Some auto-enslaved devices e.g. failover slaves are
1242 * special, as userspace might rename the device after
1243 * the interface had been brought up and running since
1244 * the point kernel initiated auto-enslavement. Allow
1245 * live name change even when these slave devices are
1248 * Typically, users of these auto-enslaving devices
1249 * don't actually care about slave name change, as
1250 * they are supposed to operate on master interface
1253 if (dev->flags & IFF_UP &&
1254 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1257 down_write(&devnet_rename_sem);
1259 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1260 up_write(&devnet_rename_sem);
1264 memcpy(oldname, dev->name, IFNAMSIZ);
1266 err = dev_get_valid_name(net, dev, newname);
1268 up_write(&devnet_rename_sem);
1272 if (oldname[0] && !strchr(oldname, '%'))
1273 netdev_info(dev, "renamed from %s\n", oldname);
1275 old_assign_type = dev->name_assign_type;
1276 dev->name_assign_type = NET_NAME_RENAMED;
1279 ret = device_rename(&dev->dev, dev->name);
1281 memcpy(dev->name, oldname, IFNAMSIZ);
1282 dev->name_assign_type = old_assign_type;
1283 up_write(&devnet_rename_sem);
1287 up_write(&devnet_rename_sem);
1289 netdev_adjacent_rename_links(dev, oldname);
1291 write_lock(&dev_base_lock);
1292 netdev_name_node_del(dev->name_node);
1293 write_unlock(&dev_base_lock);
1297 write_lock(&dev_base_lock);
1298 netdev_name_node_add(net, dev->name_node);
1299 write_unlock(&dev_base_lock);
1301 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1302 ret = notifier_to_errno(ret);
1305 /* err >= 0 after dev_alloc_name() or stores the first errno */
1308 down_write(&devnet_rename_sem);
1309 memcpy(dev->name, oldname, IFNAMSIZ);
1310 memcpy(oldname, newname, IFNAMSIZ);
1311 dev->name_assign_type = old_assign_type;
1312 old_assign_type = NET_NAME_RENAMED;
1315 pr_err("%s: name change rollback failed: %d\n",
1324 * dev_set_alias - change ifalias of a device
1326 * @alias: name up to IFALIASZ
1327 * @len: limit of bytes to copy from info
1329 * Set ifalias for a device,
1331 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1333 struct dev_ifalias *new_alias = NULL;
1335 if (len >= IFALIASZ)
1339 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1343 memcpy(new_alias->ifalias, alias, len);
1344 new_alias->ifalias[len] = 0;
1347 mutex_lock(&ifalias_mutex);
1348 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1349 mutex_is_locked(&ifalias_mutex));
1350 mutex_unlock(&ifalias_mutex);
1353 kfree_rcu(new_alias, rcuhead);
1357 EXPORT_SYMBOL(dev_set_alias);
1360 * dev_get_alias - get ifalias of a device
1362 * @name: buffer to store name of ifalias
1363 * @len: size of buffer
1365 * get ifalias for a device. Caller must make sure dev cannot go
1366 * away, e.g. rcu read lock or own a reference count to device.
1368 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1370 const struct dev_ifalias *alias;
1374 alias = rcu_dereference(dev->ifalias);
1376 ret = snprintf(name, len, "%s", alias->ifalias);
1383 * netdev_features_change - device changes features
1384 * @dev: device to cause notification
1386 * Called to indicate a device has changed features.
1388 void netdev_features_change(struct net_device *dev)
1390 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1392 EXPORT_SYMBOL(netdev_features_change);
1395 * netdev_state_change - device changes state
1396 * @dev: device to cause notification
1398 * Called to indicate a device has changed state. This function calls
1399 * the notifier chains for netdev_chain and sends a NEWLINK message
1400 * to the routing socket.
1402 void netdev_state_change(struct net_device *dev)
1404 if (dev->flags & IFF_UP) {
1405 struct netdev_notifier_change_info change_info = {
1409 call_netdevice_notifiers_info(NETDEV_CHANGE,
1411 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1414 EXPORT_SYMBOL(netdev_state_change);
1417 * __netdev_notify_peers - notify network peers about existence of @dev,
1418 * to be called when rtnl lock is already held.
1419 * @dev: network device
1421 * Generate traffic such that interested network peers are aware of
1422 * @dev, such as by generating a gratuitous ARP. This may be used when
1423 * a device wants to inform the rest of the network about some sort of
1424 * reconfiguration such as a failover event or virtual machine
1427 void __netdev_notify_peers(struct net_device *dev)
1430 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1431 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1433 EXPORT_SYMBOL(__netdev_notify_peers);
1436 * netdev_notify_peers - notify network peers about existence of @dev
1437 * @dev: network device
1439 * Generate traffic such that interested network peers are aware of
1440 * @dev, such as by generating a gratuitous ARP. This may be used when
1441 * a device wants to inform the rest of the network about some sort of
1442 * reconfiguration such as a failover event or virtual machine
1445 void netdev_notify_peers(struct net_device *dev)
1448 __netdev_notify_peers(dev);
1451 EXPORT_SYMBOL(netdev_notify_peers);
1453 static int napi_threaded_poll(void *data);
1455 static int napi_kthread_create(struct napi_struct *n)
1459 /* Create and wake up the kthread once to put it in
1460 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1461 * warning and work with loadavg.
1463 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1464 n->dev->name, n->napi_id);
1465 if (IS_ERR(n->thread)) {
1466 err = PTR_ERR(n->thread);
1467 pr_err("kthread_run failed with err %d\n", err);
1474 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1476 const struct net_device_ops *ops = dev->netdev_ops;
1481 if (!netif_device_present(dev)) {
1482 /* may be detached because parent is runtime-suspended */
1483 if (dev->dev.parent)
1484 pm_runtime_resume(dev->dev.parent);
1485 if (!netif_device_present(dev))
1489 /* Block netpoll from trying to do any rx path servicing.
1490 * If we don't do this there is a chance ndo_poll_controller
1491 * or ndo_poll may be running while we open the device
1493 netpoll_poll_disable(dev);
1495 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1496 ret = notifier_to_errno(ret);
1500 set_bit(__LINK_STATE_START, &dev->state);
1502 if (ops->ndo_validate_addr)
1503 ret = ops->ndo_validate_addr(dev);
1505 if (!ret && ops->ndo_open)
1506 ret = ops->ndo_open(dev);
1508 netpoll_poll_enable(dev);
1511 clear_bit(__LINK_STATE_START, &dev->state);
1513 dev->flags |= IFF_UP;
1514 dev_set_rx_mode(dev);
1516 add_device_randomness(dev->dev_addr, dev->addr_len);
1523 * dev_open - prepare an interface for use.
1524 * @dev: device to open
1525 * @extack: netlink extended ack
1527 * Takes a device from down to up state. The device's private open
1528 * function is invoked and then the multicast lists are loaded. Finally
1529 * the device is moved into the up state and a %NETDEV_UP message is
1530 * sent to the netdev notifier chain.
1532 * Calling this function on an active interface is a nop. On a failure
1533 * a negative errno code is returned.
1535 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1539 if (dev->flags & IFF_UP)
1542 ret = __dev_open(dev, extack);
1546 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1547 call_netdevice_notifiers(NETDEV_UP, dev);
1551 EXPORT_SYMBOL(dev_open);
1553 static void __dev_close_many(struct list_head *head)
1555 struct net_device *dev;
1560 list_for_each_entry(dev, head, close_list) {
1561 /* Temporarily disable netpoll until the interface is down */
1562 netpoll_poll_disable(dev);
1564 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1566 clear_bit(__LINK_STATE_START, &dev->state);
1568 /* Synchronize to scheduled poll. We cannot touch poll list, it
1569 * can be even on different cpu. So just clear netif_running().
1571 * dev->stop() will invoke napi_disable() on all of it's
1572 * napi_struct instances on this device.
1574 smp_mb__after_atomic(); /* Commit netif_running(). */
1577 dev_deactivate_many(head);
1579 list_for_each_entry(dev, head, close_list) {
1580 const struct net_device_ops *ops = dev->netdev_ops;
1583 * Call the device specific close. This cannot fail.
1584 * Only if device is UP
1586 * We allow it to be called even after a DETACH hot-plug
1592 dev->flags &= ~IFF_UP;
1593 netpoll_poll_enable(dev);
1597 static void __dev_close(struct net_device *dev)
1601 list_add(&dev->close_list, &single);
1602 __dev_close_many(&single);
1606 void dev_close_many(struct list_head *head, bool unlink)
1608 struct net_device *dev, *tmp;
1610 /* Remove the devices that don't need to be closed */
1611 list_for_each_entry_safe(dev, tmp, head, close_list)
1612 if (!(dev->flags & IFF_UP))
1613 list_del_init(&dev->close_list);
1615 __dev_close_many(head);
1617 list_for_each_entry_safe(dev, tmp, head, close_list) {
1618 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1619 call_netdevice_notifiers(NETDEV_DOWN, dev);
1621 list_del_init(&dev->close_list);
1624 EXPORT_SYMBOL(dev_close_many);
1627 * dev_close - shutdown an interface.
1628 * @dev: device to shutdown
1630 * This function moves an active device into down state. A
1631 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1632 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1635 void dev_close(struct net_device *dev)
1637 if (dev->flags & IFF_UP) {
1640 list_add(&dev->close_list, &single);
1641 dev_close_many(&single, true);
1645 EXPORT_SYMBOL(dev_close);
1649 * dev_disable_lro - disable Large Receive Offload on a device
1652 * Disable Large Receive Offload (LRO) on a net device. Must be
1653 * called under RTNL. This is needed if received packets may be
1654 * forwarded to another interface.
1656 void dev_disable_lro(struct net_device *dev)
1658 struct net_device *lower_dev;
1659 struct list_head *iter;
1661 dev->wanted_features &= ~NETIF_F_LRO;
1662 netdev_update_features(dev);
1664 if (unlikely(dev->features & NETIF_F_LRO))
1665 netdev_WARN(dev, "failed to disable LRO!\n");
1667 netdev_for_each_lower_dev(dev, lower_dev, iter)
1668 dev_disable_lro(lower_dev);
1670 EXPORT_SYMBOL(dev_disable_lro);
1673 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1676 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1677 * called under RTNL. This is needed if Generic XDP is installed on
1680 static void dev_disable_gro_hw(struct net_device *dev)
1682 dev->wanted_features &= ~NETIF_F_GRO_HW;
1683 netdev_update_features(dev);
1685 if (unlikely(dev->features & NETIF_F_GRO_HW))
1686 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1689 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1692 case NETDEV_##val: \
1693 return "NETDEV_" __stringify(val);
1695 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1696 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1697 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1698 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1699 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1700 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1701 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1702 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1703 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1707 return "UNKNOWN_NETDEV_EVENT";
1709 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1711 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1712 struct net_device *dev)
1714 struct netdev_notifier_info info = {
1718 return nb->notifier_call(nb, val, &info);
1721 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1722 struct net_device *dev)
1726 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1727 err = notifier_to_errno(err);
1731 if (!(dev->flags & IFF_UP))
1734 call_netdevice_notifier(nb, NETDEV_UP, dev);
1738 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1739 struct net_device *dev)
1741 if (dev->flags & IFF_UP) {
1742 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1744 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1746 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1749 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1752 struct net_device *dev;
1755 for_each_netdev(net, dev) {
1756 err = call_netdevice_register_notifiers(nb, dev);
1763 for_each_netdev_continue_reverse(net, dev)
1764 call_netdevice_unregister_notifiers(nb, dev);
1768 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1771 struct net_device *dev;
1773 for_each_netdev(net, dev)
1774 call_netdevice_unregister_notifiers(nb, dev);
1777 static int dev_boot_phase = 1;
1780 * register_netdevice_notifier - register a network notifier block
1783 * Register a notifier to be called when network device events occur.
1784 * The notifier passed is linked into the kernel structures and must
1785 * not be reused until it has been unregistered. A negative errno code
1786 * is returned on a failure.
1788 * When registered all registration and up events are replayed
1789 * to the new notifier to allow device to have a race free
1790 * view of the network device list.
1793 int register_netdevice_notifier(struct notifier_block *nb)
1798 /* Close race with setup_net() and cleanup_net() */
1799 down_write(&pernet_ops_rwsem);
1801 err = raw_notifier_chain_register(&netdev_chain, nb);
1807 err = call_netdevice_register_net_notifiers(nb, net);
1814 up_write(&pernet_ops_rwsem);
1818 for_each_net_continue_reverse(net)
1819 call_netdevice_unregister_net_notifiers(nb, net);
1821 raw_notifier_chain_unregister(&netdev_chain, nb);
1824 EXPORT_SYMBOL(register_netdevice_notifier);
1827 * unregister_netdevice_notifier - unregister a network notifier block
1830 * Unregister a notifier previously registered by
1831 * register_netdevice_notifier(). The notifier is unlinked into the
1832 * kernel structures and may then be reused. A negative errno code
1833 * is returned on a failure.
1835 * After unregistering unregister and down device events are synthesized
1836 * for all devices on the device list to the removed notifier to remove
1837 * the need for special case cleanup code.
1840 int unregister_netdevice_notifier(struct notifier_block *nb)
1845 /* Close race with setup_net() and cleanup_net() */
1846 down_write(&pernet_ops_rwsem);
1848 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1853 call_netdevice_unregister_net_notifiers(nb, net);
1857 up_write(&pernet_ops_rwsem);
1860 EXPORT_SYMBOL(unregister_netdevice_notifier);
1862 static int __register_netdevice_notifier_net(struct net *net,
1863 struct notifier_block *nb,
1864 bool ignore_call_fail)
1868 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1874 err = call_netdevice_register_net_notifiers(nb, net);
1875 if (err && !ignore_call_fail)
1876 goto chain_unregister;
1881 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1885 static int __unregister_netdevice_notifier_net(struct net *net,
1886 struct notifier_block *nb)
1890 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1894 call_netdevice_unregister_net_notifiers(nb, net);
1899 * register_netdevice_notifier_net - register a per-netns network notifier block
1900 * @net: network namespace
1903 * Register a notifier to be called when network device events occur.
1904 * The notifier passed is linked into the kernel structures and must
1905 * not be reused until it has been unregistered. A negative errno code
1906 * is returned on a failure.
1908 * When registered all registration and up events are replayed
1909 * to the new notifier to allow device to have a race free
1910 * view of the network device list.
1913 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1918 err = __register_netdevice_notifier_net(net, nb, false);
1922 EXPORT_SYMBOL(register_netdevice_notifier_net);
1925 * unregister_netdevice_notifier_net - unregister a per-netns
1926 * network notifier block
1927 * @net: network namespace
1930 * Unregister a notifier previously registered by
1931 * register_netdevice_notifier(). The notifier is unlinked into the
1932 * kernel structures and may then be reused. A negative errno code
1933 * is returned on a failure.
1935 * After unregistering unregister and down device events are synthesized
1936 * for all devices on the device list to the removed notifier to remove
1937 * the need for special case cleanup code.
1940 int unregister_netdevice_notifier_net(struct net *net,
1941 struct notifier_block *nb)
1946 err = __unregister_netdevice_notifier_net(net, nb);
1950 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1952 int register_netdevice_notifier_dev_net(struct net_device *dev,
1953 struct notifier_block *nb,
1954 struct netdev_net_notifier *nn)
1959 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1962 list_add(&nn->list, &dev->net_notifier_list);
1967 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1969 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1970 struct notifier_block *nb,
1971 struct netdev_net_notifier *nn)
1976 list_del(&nn->list);
1977 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1981 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1983 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1986 struct netdev_net_notifier *nn;
1988 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1989 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1990 __register_netdevice_notifier_net(net, nn->nb, true);
1995 * call_netdevice_notifiers_info - call all network notifier blocks
1996 * @val: value passed unmodified to notifier function
1997 * @info: notifier information data
1999 * Call all network notifier blocks. Parameters and return value
2000 * are as for raw_notifier_call_chain().
2003 static int call_netdevice_notifiers_info(unsigned long val,
2004 struct netdev_notifier_info *info)
2006 struct net *net = dev_net(info->dev);
2011 /* Run per-netns notifier block chain first, then run the global one.
2012 * Hopefully, one day, the global one is going to be removed after
2013 * all notifier block registrators get converted to be per-netns.
2015 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2016 if (ret & NOTIFY_STOP_MASK)
2018 return raw_notifier_call_chain(&netdev_chain, val, info);
2021 static int call_netdevice_notifiers_extack(unsigned long val,
2022 struct net_device *dev,
2023 struct netlink_ext_ack *extack)
2025 struct netdev_notifier_info info = {
2030 return call_netdevice_notifiers_info(val, &info);
2034 * call_netdevice_notifiers - call all network notifier blocks
2035 * @val: value passed unmodified to notifier function
2036 * @dev: net_device pointer passed unmodified to notifier function
2038 * Call all network notifier blocks. Parameters and return value
2039 * are as for raw_notifier_call_chain().
2042 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2044 return call_netdevice_notifiers_extack(val, dev, NULL);
2046 EXPORT_SYMBOL(call_netdevice_notifiers);
2049 * call_netdevice_notifiers_mtu - call all network notifier blocks
2050 * @val: value passed unmodified to notifier function
2051 * @dev: net_device pointer passed unmodified to notifier function
2052 * @arg: additional u32 argument passed to the notifier function
2054 * Call all network notifier blocks. Parameters and return value
2055 * are as for raw_notifier_call_chain().
2057 static int call_netdevice_notifiers_mtu(unsigned long val,
2058 struct net_device *dev, u32 arg)
2060 struct netdev_notifier_info_ext info = {
2065 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2067 return call_netdevice_notifiers_info(val, &info.info);
2070 #ifdef CONFIG_NET_INGRESS
2071 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2073 void net_inc_ingress_queue(void)
2075 static_branch_inc(&ingress_needed_key);
2077 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2079 void net_dec_ingress_queue(void)
2081 static_branch_dec(&ingress_needed_key);
2083 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2086 #ifdef CONFIG_NET_EGRESS
2087 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2089 void net_inc_egress_queue(void)
2091 static_branch_inc(&egress_needed_key);
2093 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2095 void net_dec_egress_queue(void)
2097 static_branch_dec(&egress_needed_key);
2099 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2102 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2103 #ifdef CONFIG_JUMP_LABEL
2104 static atomic_t netstamp_needed_deferred;
2105 static atomic_t netstamp_wanted;
2106 static void netstamp_clear(struct work_struct *work)
2108 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2111 wanted = atomic_add_return(deferred, &netstamp_wanted);
2113 static_branch_enable(&netstamp_needed_key);
2115 static_branch_disable(&netstamp_needed_key);
2117 static DECLARE_WORK(netstamp_work, netstamp_clear);
2120 void net_enable_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_inc(&netstamp_needed_deferred);
2133 schedule_work(&netstamp_work);
2135 static_branch_inc(&netstamp_needed_key);
2138 EXPORT_SYMBOL(net_enable_timestamp);
2140 void net_disable_timestamp(void)
2142 #ifdef CONFIG_JUMP_LABEL
2146 wanted = atomic_read(&netstamp_wanted);
2149 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2152 atomic_dec(&netstamp_needed_deferred);
2153 schedule_work(&netstamp_work);
2155 static_branch_dec(&netstamp_needed_key);
2158 EXPORT_SYMBOL(net_disable_timestamp);
2160 static inline void net_timestamp_set(struct sk_buff *skb)
2163 if (static_branch_unlikely(&netstamp_needed_key))
2164 __net_timestamp(skb);
2167 #define net_timestamp_check(COND, SKB) \
2168 if (static_branch_unlikely(&netstamp_needed_key)) { \
2169 if ((COND) && !(SKB)->tstamp) \
2170 __net_timestamp(SKB); \
2173 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2175 return __is_skb_forwardable(dev, skb, true);
2177 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2179 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2182 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2185 skb->protocol = eth_type_trans(skb, dev);
2186 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2192 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2194 return __dev_forward_skb2(dev, skb, true);
2196 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2199 * dev_forward_skb - loopback an skb to another netif
2201 * @dev: destination network device
2202 * @skb: buffer to forward
2205 * NET_RX_SUCCESS (no congestion)
2206 * NET_RX_DROP (packet was dropped, but freed)
2208 * dev_forward_skb can be used for injecting an skb from the
2209 * start_xmit function of one device into the receive queue
2210 * of another device.
2212 * The receiving device may be in another namespace, so
2213 * we have to clear all information in the skb that could
2214 * impact namespace isolation.
2216 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2218 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2220 EXPORT_SYMBOL_GPL(dev_forward_skb);
2222 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2224 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2227 static inline int deliver_skb(struct sk_buff *skb,
2228 struct packet_type *pt_prev,
2229 struct net_device *orig_dev)
2231 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2233 refcount_inc(&skb->users);
2234 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2237 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2238 struct packet_type **pt,
2239 struct net_device *orig_dev,
2241 struct list_head *ptype_list)
2243 struct packet_type *ptype, *pt_prev = *pt;
2245 list_for_each_entry_rcu(ptype, ptype_list, list) {
2246 if (ptype->type != type)
2249 deliver_skb(skb, pt_prev, orig_dev);
2255 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2257 if (!ptype->af_packet_priv || !skb->sk)
2260 if (ptype->id_match)
2261 return ptype->id_match(ptype, skb->sk);
2262 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2269 * dev_nit_active - return true if any network interface taps are in use
2271 * @dev: network device to check for the presence of taps
2273 bool dev_nit_active(struct net_device *dev)
2275 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2277 EXPORT_SYMBOL_GPL(dev_nit_active);
2280 * Support routine. Sends outgoing frames to any network
2281 * taps currently in use.
2284 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2286 struct packet_type *ptype;
2287 struct sk_buff *skb2 = NULL;
2288 struct packet_type *pt_prev = NULL;
2289 struct list_head *ptype_list = &ptype_all;
2293 list_for_each_entry_rcu(ptype, ptype_list, list) {
2294 if (ptype->ignore_outgoing)
2297 /* Never send packets back to the socket
2298 * they originated from - MvS (miquels@drinkel.ow.org)
2300 if (skb_loop_sk(ptype, skb))
2304 deliver_skb(skb2, pt_prev, skb->dev);
2309 /* need to clone skb, done only once */
2310 skb2 = skb_clone(skb, GFP_ATOMIC);
2314 net_timestamp_set(skb2);
2316 /* skb->nh should be correctly
2317 * set by sender, so that the second statement is
2318 * just protection against buggy protocols.
2320 skb_reset_mac_header(skb2);
2322 if (skb_network_header(skb2) < skb2->data ||
2323 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2324 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2325 ntohs(skb2->protocol),
2327 skb_reset_network_header(skb2);
2330 skb2->transport_header = skb2->network_header;
2331 skb2->pkt_type = PACKET_OUTGOING;
2335 if (ptype_list == &ptype_all) {
2336 ptype_list = &dev->ptype_all;
2341 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2342 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2348 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2351 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2352 * @dev: Network device
2353 * @txq: number of queues available
2355 * If real_num_tx_queues is changed the tc mappings may no longer be
2356 * valid. To resolve this verify the tc mapping remains valid and if
2357 * not NULL the mapping. With no priorities mapping to this
2358 * offset/count pair it will no longer be used. In the worst case TC0
2359 * is invalid nothing can be done so disable priority mappings. If is
2360 * expected that drivers will fix this mapping if they can before
2361 * calling netif_set_real_num_tx_queues.
2363 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2366 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2368 /* If TC0 is invalidated disable TC mapping */
2369 if (tc->offset + tc->count > txq) {
2370 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2375 /* Invalidated prio to tc mappings set to TC0 */
2376 for (i = 1; i < TC_BITMASK + 1; i++) {
2377 int q = netdev_get_prio_tc_map(dev, i);
2379 tc = &dev->tc_to_txq[q];
2380 if (tc->offset + tc->count > txq) {
2381 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2383 netdev_set_prio_tc_map(dev, i, 0);
2388 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2391 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2394 /* walk through the TCs and see if it falls into any of them */
2395 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2396 if ((txq - tc->offset) < tc->count)
2400 /* didn't find it, just return -1 to indicate no match */
2406 EXPORT_SYMBOL(netdev_txq_to_tc);
2409 static struct static_key xps_needed __read_mostly;
2410 static struct static_key xps_rxqs_needed __read_mostly;
2411 static DEFINE_MUTEX(xps_map_mutex);
2412 #define xmap_dereference(P) \
2413 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2415 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2416 struct xps_dev_maps *old_maps, int tci, u16 index)
2418 struct xps_map *map = NULL;
2422 map = xmap_dereference(dev_maps->attr_map[tci]);
2426 for (pos = map->len; pos--;) {
2427 if (map->queues[pos] != index)
2431 map->queues[pos] = map->queues[--map->len];
2436 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2437 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2438 kfree_rcu(map, rcu);
2445 static bool remove_xps_queue_cpu(struct net_device *dev,
2446 struct xps_dev_maps *dev_maps,
2447 int cpu, u16 offset, u16 count)
2449 int num_tc = dev_maps->num_tc;
2450 bool active = false;
2453 for (tci = cpu * num_tc; num_tc--; tci++) {
2456 for (i = count, j = offset; i--; j++) {
2457 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2467 static void reset_xps_maps(struct net_device *dev,
2468 struct xps_dev_maps *dev_maps,
2469 enum xps_map_type type)
2471 static_key_slow_dec_cpuslocked(&xps_needed);
2472 if (type == XPS_RXQS)
2473 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2475 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2477 kfree_rcu(dev_maps, rcu);
2480 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2481 u16 offset, u16 count)
2483 struct xps_dev_maps *dev_maps;
2484 bool active = false;
2487 dev_maps = xmap_dereference(dev->xps_maps[type]);
2491 for (j = 0; j < dev_maps->nr_ids; j++)
2492 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2494 reset_xps_maps(dev, dev_maps, type);
2496 if (type == XPS_CPUS) {
2497 for (i = offset + (count - 1); count--; i--)
2498 netdev_queue_numa_node_write(
2499 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2503 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2506 if (!static_key_false(&xps_needed))
2510 mutex_lock(&xps_map_mutex);
2512 if (static_key_false(&xps_rxqs_needed))
2513 clean_xps_maps(dev, XPS_RXQS, offset, count);
2515 clean_xps_maps(dev, XPS_CPUS, offset, count);
2517 mutex_unlock(&xps_map_mutex);
2521 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2523 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2526 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2527 u16 index, bool is_rxqs_map)
2529 struct xps_map *new_map;
2530 int alloc_len = XPS_MIN_MAP_ALLOC;
2533 for (pos = 0; map && pos < map->len; pos++) {
2534 if (map->queues[pos] != index)
2539 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2541 if (pos < map->alloc_len)
2544 alloc_len = map->alloc_len * 2;
2547 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2551 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2553 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2554 cpu_to_node(attr_index));
2558 for (i = 0; i < pos; i++)
2559 new_map->queues[i] = map->queues[i];
2560 new_map->alloc_len = alloc_len;
2566 /* Copy xps maps at a given index */
2567 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2568 struct xps_dev_maps *new_dev_maps, int index,
2569 int tc, bool skip_tc)
2571 int i, tci = index * dev_maps->num_tc;
2572 struct xps_map *map;
2574 /* copy maps belonging to foreign traffic classes */
2575 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2576 if (i == tc && skip_tc)
2579 /* fill in the new device map from the old device map */
2580 map = xmap_dereference(dev_maps->attr_map[tci]);
2581 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2585 /* Must be called under cpus_read_lock */
2586 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2587 u16 index, enum xps_map_type type)
2589 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2590 const unsigned long *online_mask = NULL;
2591 bool active = false, copy = false;
2592 int i, j, tci, numa_node_id = -2;
2593 int maps_sz, num_tc = 1, tc = 0;
2594 struct xps_map *map, *new_map;
2595 unsigned int nr_ids;
2597 WARN_ON_ONCE(index >= dev->num_tx_queues);
2600 /* Do not allow XPS on subordinate device directly */
2601 num_tc = dev->num_tc;
2605 /* If queue belongs to subordinate dev use its map */
2606 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2608 tc = netdev_txq_to_tc(dev, index);
2613 mutex_lock(&xps_map_mutex);
2615 dev_maps = xmap_dereference(dev->xps_maps[type]);
2616 if (type == XPS_RXQS) {
2617 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2618 nr_ids = dev->num_rx_queues;
2620 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2621 if (num_possible_cpus() > 1)
2622 online_mask = cpumask_bits(cpu_online_mask);
2623 nr_ids = nr_cpu_ids;
2626 if (maps_sz < L1_CACHE_BYTES)
2627 maps_sz = L1_CACHE_BYTES;
2629 /* The old dev_maps could be larger or smaller than the one we're
2630 * setting up now, as dev->num_tc or nr_ids could have been updated in
2631 * between. We could try to be smart, but let's be safe instead and only
2632 * copy foreign traffic classes if the two map sizes match.
2635 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2638 /* allocate memory for queue storage */
2639 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2641 if (!new_dev_maps) {
2642 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2643 if (!new_dev_maps) {
2644 mutex_unlock(&xps_map_mutex);
2648 new_dev_maps->nr_ids = nr_ids;
2649 new_dev_maps->num_tc = num_tc;
2652 tci = j * num_tc + tc;
2653 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2655 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2659 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2663 goto out_no_new_maps;
2666 /* Increment static keys at most once per type */
2667 static_key_slow_inc_cpuslocked(&xps_needed);
2668 if (type == XPS_RXQS)
2669 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2672 for (j = 0; j < nr_ids; j++) {
2673 bool skip_tc = false;
2675 tci = j * num_tc + tc;
2676 if (netif_attr_test_mask(j, mask, nr_ids) &&
2677 netif_attr_test_online(j, online_mask, nr_ids)) {
2678 /* add tx-queue to CPU/rx-queue maps */
2683 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2684 while ((pos < map->len) && (map->queues[pos] != index))
2687 if (pos == map->len)
2688 map->queues[map->len++] = index;
2690 if (type == XPS_CPUS) {
2691 if (numa_node_id == -2)
2692 numa_node_id = cpu_to_node(j);
2693 else if (numa_node_id != cpu_to_node(j))
2700 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2704 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2706 /* Cleanup old maps */
2708 goto out_no_old_maps;
2710 for (j = 0; j < dev_maps->nr_ids; j++) {
2711 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2712 map = xmap_dereference(dev_maps->attr_map[tci]);
2717 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2722 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2723 kfree_rcu(map, rcu);
2727 old_dev_maps = dev_maps;
2730 dev_maps = new_dev_maps;
2734 if (type == XPS_CPUS)
2735 /* update Tx queue numa node */
2736 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2737 (numa_node_id >= 0) ?
2738 numa_node_id : NUMA_NO_NODE);
2743 /* removes tx-queue from unused CPUs/rx-queues */
2744 for (j = 0; j < dev_maps->nr_ids; j++) {
2745 tci = j * dev_maps->num_tc;
2747 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2749 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2750 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2753 active |= remove_xps_queue(dev_maps,
2754 copy ? old_dev_maps : NULL,
2760 kfree_rcu(old_dev_maps, rcu);
2762 /* free map if not active */
2764 reset_xps_maps(dev, dev_maps, type);
2767 mutex_unlock(&xps_map_mutex);
2771 /* remove any maps that we added */
2772 for (j = 0; j < nr_ids; j++) {
2773 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2774 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2776 xmap_dereference(dev_maps->attr_map[tci]) :
2778 if (new_map && new_map != map)
2783 mutex_unlock(&xps_map_mutex);
2785 kfree(new_dev_maps);
2788 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2790 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2796 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2801 EXPORT_SYMBOL(netif_set_xps_queue);
2804 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2806 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2808 /* Unbind any subordinate channels */
2809 while (txq-- != &dev->_tx[0]) {
2811 netdev_unbind_sb_channel(dev, txq->sb_dev);
2815 void netdev_reset_tc(struct net_device *dev)
2818 netif_reset_xps_queues_gt(dev, 0);
2820 netdev_unbind_all_sb_channels(dev);
2822 /* Reset TC configuration of device */
2824 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2825 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2827 EXPORT_SYMBOL(netdev_reset_tc);
2829 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2831 if (tc >= dev->num_tc)
2835 netif_reset_xps_queues(dev, offset, count);
2837 dev->tc_to_txq[tc].count = count;
2838 dev->tc_to_txq[tc].offset = offset;
2841 EXPORT_SYMBOL(netdev_set_tc_queue);
2843 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2845 if (num_tc > TC_MAX_QUEUE)
2849 netif_reset_xps_queues_gt(dev, 0);
2851 netdev_unbind_all_sb_channels(dev);
2853 dev->num_tc = num_tc;
2856 EXPORT_SYMBOL(netdev_set_num_tc);
2858 void netdev_unbind_sb_channel(struct net_device *dev,
2859 struct net_device *sb_dev)
2861 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2864 netif_reset_xps_queues_gt(sb_dev, 0);
2866 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2867 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2869 while (txq-- != &dev->_tx[0]) {
2870 if (txq->sb_dev == sb_dev)
2874 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2876 int netdev_bind_sb_channel_queue(struct net_device *dev,
2877 struct net_device *sb_dev,
2878 u8 tc, u16 count, u16 offset)
2880 /* Make certain the sb_dev and dev are already configured */
2881 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2884 /* We cannot hand out queues we don't have */
2885 if ((offset + count) > dev->real_num_tx_queues)
2888 /* Record the mapping */
2889 sb_dev->tc_to_txq[tc].count = count;
2890 sb_dev->tc_to_txq[tc].offset = offset;
2892 /* Provide a way for Tx queue to find the tc_to_txq map or
2893 * XPS map for itself.
2896 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2900 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2902 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2904 /* Do not use a multiqueue device to represent a subordinate channel */
2905 if (netif_is_multiqueue(dev))
2908 /* We allow channels 1 - 32767 to be used for subordinate channels.
2909 * Channel 0 is meant to be "native" mode and used only to represent
2910 * the main root device. We allow writing 0 to reset the device back
2911 * to normal mode after being used as a subordinate channel.
2913 if (channel > S16_MAX)
2916 dev->num_tc = -channel;
2920 EXPORT_SYMBOL(netdev_set_sb_channel);
2923 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2924 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2926 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2931 disabling = txq < dev->real_num_tx_queues;
2933 if (txq < 1 || txq > dev->num_tx_queues)
2936 if (dev->reg_state == NETREG_REGISTERED ||
2937 dev->reg_state == NETREG_UNREGISTERING) {
2940 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2946 netif_setup_tc(dev, txq);
2948 dev_qdisc_change_real_num_tx(dev, txq);
2950 dev->real_num_tx_queues = txq;
2954 qdisc_reset_all_tx_gt(dev, txq);
2956 netif_reset_xps_queues_gt(dev, txq);
2960 dev->real_num_tx_queues = txq;
2965 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2969 * netif_set_real_num_rx_queues - set actual number of RX queues used
2970 * @dev: Network device
2971 * @rxq: Actual number of RX queues
2973 * This must be called either with the rtnl_lock held or before
2974 * registration of the net device. Returns 0 on success, or a
2975 * negative error code. If called before registration, it always
2978 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2982 if (rxq < 1 || rxq > dev->num_rx_queues)
2985 if (dev->reg_state == NETREG_REGISTERED) {
2988 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2994 dev->real_num_rx_queues = rxq;
2997 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3001 * netif_set_real_num_queues - set actual number of RX and TX queues used
3002 * @dev: Network device
3003 * @txq: Actual number of TX queues
3004 * @rxq: Actual number of RX queues
3006 * Set the real number of both TX and RX queues.
3007 * Does nothing if the number of queues is already correct.
3009 int netif_set_real_num_queues(struct net_device *dev,
3010 unsigned int txq, unsigned int rxq)
3012 unsigned int old_rxq = dev->real_num_rx_queues;
3015 if (txq < 1 || txq > dev->num_tx_queues ||
3016 rxq < 1 || rxq > dev->num_rx_queues)
3019 /* Start from increases, so the error path only does decreases -
3020 * decreases can't fail.
3022 if (rxq > dev->real_num_rx_queues) {
3023 err = netif_set_real_num_rx_queues(dev, rxq);
3027 if (txq > dev->real_num_tx_queues) {
3028 err = netif_set_real_num_tx_queues(dev, txq);
3032 if (rxq < dev->real_num_rx_queues)
3033 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3034 if (txq < dev->real_num_tx_queues)
3035 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3039 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3042 EXPORT_SYMBOL(netif_set_real_num_queues);
3045 * netif_get_num_default_rss_queues - default number of RSS queues
3047 * This routine should set an upper limit on the number of RSS queues
3048 * used by default by multiqueue devices.
3050 int netif_get_num_default_rss_queues(void)
3052 return is_kdump_kernel() ?
3053 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3055 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3057 static void __netif_reschedule(struct Qdisc *q)
3059 struct softnet_data *sd;
3060 unsigned long flags;
3062 local_irq_save(flags);
3063 sd = this_cpu_ptr(&softnet_data);
3064 q->next_sched = NULL;
3065 *sd->output_queue_tailp = q;
3066 sd->output_queue_tailp = &q->next_sched;
3067 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3068 local_irq_restore(flags);
3071 void __netif_schedule(struct Qdisc *q)
3073 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3074 __netif_reschedule(q);
3076 EXPORT_SYMBOL(__netif_schedule);
3078 struct dev_kfree_skb_cb {
3079 enum skb_free_reason reason;
3082 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3084 return (struct dev_kfree_skb_cb *)skb->cb;
3087 void netif_schedule_queue(struct netdev_queue *txq)
3090 if (!netif_xmit_stopped(txq)) {
3091 struct Qdisc *q = rcu_dereference(txq->qdisc);
3093 __netif_schedule(q);
3097 EXPORT_SYMBOL(netif_schedule_queue);
3099 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3101 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3105 q = rcu_dereference(dev_queue->qdisc);
3106 __netif_schedule(q);
3110 EXPORT_SYMBOL(netif_tx_wake_queue);
3112 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3114 unsigned long flags;
3119 if (likely(refcount_read(&skb->users) == 1)) {
3121 refcount_set(&skb->users, 0);
3122 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3125 get_kfree_skb_cb(skb)->reason = reason;
3126 local_irq_save(flags);
3127 skb->next = __this_cpu_read(softnet_data.completion_queue);
3128 __this_cpu_write(softnet_data.completion_queue, skb);
3129 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3130 local_irq_restore(flags);
3132 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3134 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3136 if (in_hardirq() || irqs_disabled())
3137 __dev_kfree_skb_irq(skb, reason);
3138 else if (unlikely(reason == SKB_REASON_DROPPED))
3143 EXPORT_SYMBOL(__dev_kfree_skb_any);
3147 * netif_device_detach - mark device as removed
3148 * @dev: network device
3150 * Mark device as removed from system and therefore no longer available.
3152 void netif_device_detach(struct net_device *dev)
3154 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3155 netif_running(dev)) {
3156 netif_tx_stop_all_queues(dev);
3159 EXPORT_SYMBOL(netif_device_detach);
3162 * netif_device_attach - mark device as attached
3163 * @dev: network device
3165 * Mark device as attached from system and restart if needed.
3167 void netif_device_attach(struct net_device *dev)
3169 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3170 netif_running(dev)) {
3171 netif_tx_wake_all_queues(dev);
3172 __netdev_watchdog_up(dev);
3175 EXPORT_SYMBOL(netif_device_attach);
3178 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3179 * to be used as a distribution range.
3181 static u16 skb_tx_hash(const struct net_device *dev,
3182 const struct net_device *sb_dev,
3183 struct sk_buff *skb)
3187 u16 qcount = dev->real_num_tx_queues;
3190 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3192 qoffset = sb_dev->tc_to_txq[tc].offset;
3193 qcount = sb_dev->tc_to_txq[tc].count;
3194 if (unlikely(!qcount)) {
3195 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3196 sb_dev->name, qoffset, tc);
3198 qcount = dev->real_num_tx_queues;
3202 if (skb_rx_queue_recorded(skb)) {
3203 hash = skb_get_rx_queue(skb);
3204 if (hash >= qoffset)
3206 while (unlikely(hash >= qcount))
3208 return hash + qoffset;
3211 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3214 static void skb_warn_bad_offload(const struct sk_buff *skb)
3216 static const netdev_features_t null_features;
3217 struct net_device *dev = skb->dev;
3218 const char *name = "";
3220 if (!net_ratelimit())
3224 if (dev->dev.parent)
3225 name = dev_driver_string(dev->dev.parent);
3227 name = netdev_name(dev);
3229 skb_dump(KERN_WARNING, skb, false);
3230 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3231 name, dev ? &dev->features : &null_features,
3232 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3236 * Invalidate hardware checksum when packet is to be mangled, and
3237 * complete checksum manually on outgoing path.
3239 int skb_checksum_help(struct sk_buff *skb)
3242 int ret = 0, offset;
3244 if (skb->ip_summed == CHECKSUM_COMPLETE)
3245 goto out_set_summed;
3247 if (unlikely(skb_is_gso(skb))) {
3248 skb_warn_bad_offload(skb);
3252 /* Before computing a checksum, we should make sure no frag could
3253 * be modified by an external entity : checksum could be wrong.
3255 if (skb_has_shared_frag(skb)) {
3256 ret = __skb_linearize(skb);
3261 offset = skb_checksum_start_offset(skb);
3263 if (WARN_ON_ONCE(offset >= skb_headlen(skb)))
3266 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3268 offset += skb->csum_offset;
3269 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb)))
3272 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3276 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3278 skb->ip_summed = CHECKSUM_NONE;
3282 EXPORT_SYMBOL(skb_checksum_help);
3284 int skb_crc32c_csum_help(struct sk_buff *skb)
3287 int ret = 0, offset, start;
3289 if (skb->ip_summed != CHECKSUM_PARTIAL)
3292 if (unlikely(skb_is_gso(skb)))
3295 /* Before computing a checksum, we should make sure no frag could
3296 * be modified by an external entity : checksum could be wrong.
3298 if (unlikely(skb_has_shared_frag(skb))) {
3299 ret = __skb_linearize(skb);
3303 start = skb_checksum_start_offset(skb);
3304 offset = start + offsetof(struct sctphdr, checksum);
3305 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3310 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3314 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3315 skb->len - start, ~(__u32)0,
3317 *(__le32 *)(skb->data + offset) = crc32c_csum;
3318 skb->ip_summed = CHECKSUM_NONE;
3319 skb->csum_not_inet = 0;
3324 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3326 __be16 type = skb->protocol;
3328 /* Tunnel gso handlers can set protocol to ethernet. */
3329 if (type == htons(ETH_P_TEB)) {
3332 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3335 eth = (struct ethhdr *)skb->data;
3336 type = eth->h_proto;
3339 return vlan_get_protocol_and_depth(skb, type, depth);
3343 * skb_mac_gso_segment - mac layer segmentation handler.
3344 * @skb: buffer to segment
3345 * @features: features for the output path (see dev->features)
3347 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3348 netdev_features_t features)
3350 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3351 struct packet_offload *ptype;
3352 int vlan_depth = skb->mac_len;
3353 __be16 type = skb_network_protocol(skb, &vlan_depth);
3355 if (unlikely(!type))
3356 return ERR_PTR(-EINVAL);
3358 __skb_pull(skb, vlan_depth);
3361 list_for_each_entry_rcu(ptype, &offload_base, list) {
3362 if (ptype->type == type && ptype->callbacks.gso_segment) {
3363 segs = ptype->callbacks.gso_segment(skb, features);
3369 __skb_push(skb, skb->data - skb_mac_header(skb));
3373 EXPORT_SYMBOL(skb_mac_gso_segment);
3376 /* openvswitch calls this on rx path, so we need a different check.
3378 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3381 return skb->ip_summed != CHECKSUM_PARTIAL &&
3382 skb->ip_summed != CHECKSUM_UNNECESSARY;
3384 return skb->ip_summed == CHECKSUM_NONE;
3388 * __skb_gso_segment - Perform segmentation on skb.
3389 * @skb: buffer to segment
3390 * @features: features for the output path (see dev->features)
3391 * @tx_path: whether it is called in TX path
3393 * This function segments the given skb and returns a list of segments.
3395 * It may return NULL if the skb requires no segmentation. This is
3396 * only possible when GSO is used for verifying header integrity.
3398 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3400 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3401 netdev_features_t features, bool tx_path)
3403 struct sk_buff *segs;
3405 if (unlikely(skb_needs_check(skb, tx_path))) {
3408 /* We're going to init ->check field in TCP or UDP header */
3409 err = skb_cow_head(skb, 0);
3411 return ERR_PTR(err);
3414 /* Only report GSO partial support if it will enable us to
3415 * support segmentation on this frame without needing additional
3418 if (features & NETIF_F_GSO_PARTIAL) {
3419 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3420 struct net_device *dev = skb->dev;
3422 partial_features |= dev->features & dev->gso_partial_features;
3423 if (!skb_gso_ok(skb, features | partial_features))
3424 features &= ~NETIF_F_GSO_PARTIAL;
3427 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3428 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3430 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3431 SKB_GSO_CB(skb)->encap_level = 0;
3433 skb_reset_mac_header(skb);
3434 skb_reset_mac_len(skb);
3436 segs = skb_mac_gso_segment(skb, features);
3438 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3439 skb_warn_bad_offload(skb);
3443 EXPORT_SYMBOL(__skb_gso_segment);
3445 /* Take action when hardware reception checksum errors are detected. */
3447 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3449 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3450 skb_dump(KERN_ERR, skb, true);
3454 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3456 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3458 EXPORT_SYMBOL(netdev_rx_csum_fault);
3461 /* XXX: check that highmem exists at all on the given machine. */
3462 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3464 #ifdef CONFIG_HIGHMEM
3467 if (!(dev->features & NETIF_F_HIGHDMA)) {
3468 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3469 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3471 if (PageHighMem(skb_frag_page(frag)))
3479 /* If MPLS offload request, verify we are testing hardware MPLS features
3480 * instead of standard features for the netdev.
3482 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3483 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3484 netdev_features_t features,
3487 if (eth_p_mpls(type))
3488 features &= skb->dev->mpls_features;
3493 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3494 netdev_features_t features,
3501 static netdev_features_t harmonize_features(struct sk_buff *skb,
3502 netdev_features_t features)
3506 type = skb_network_protocol(skb, NULL);
3507 features = net_mpls_features(skb, features, type);
3509 if (skb->ip_summed != CHECKSUM_NONE &&
3510 !can_checksum_protocol(features, type)) {
3511 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3513 if (illegal_highdma(skb->dev, skb))
3514 features &= ~NETIF_F_SG;
3519 netdev_features_t passthru_features_check(struct sk_buff *skb,
3520 struct net_device *dev,
3521 netdev_features_t features)
3525 EXPORT_SYMBOL(passthru_features_check);
3527 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3528 struct net_device *dev,
3529 netdev_features_t features)
3531 return vlan_features_check(skb, features);
3534 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3535 struct net_device *dev,
3536 netdev_features_t features)
3538 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3540 if (gso_segs > dev->gso_max_segs)
3541 return features & ~NETIF_F_GSO_MASK;
3543 if (!skb_shinfo(skb)->gso_type) {
3544 skb_warn_bad_offload(skb);
3545 return features & ~NETIF_F_GSO_MASK;
3548 /* Support for GSO partial features requires software
3549 * intervention before we can actually process the packets
3550 * so we need to strip support for any partial features now
3551 * and we can pull them back in after we have partially
3552 * segmented the frame.
3554 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3555 features &= ~dev->gso_partial_features;
3557 /* Make sure to clear the IPv4 ID mangling feature if the
3558 * IPv4 header has the potential to be fragmented.
3560 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3561 struct iphdr *iph = skb->encapsulation ?
3562 inner_ip_hdr(skb) : ip_hdr(skb);
3564 if (!(iph->frag_off & htons(IP_DF)))
3565 features &= ~NETIF_F_TSO_MANGLEID;
3571 netdev_features_t netif_skb_features(struct sk_buff *skb)
3573 struct net_device *dev = skb->dev;
3574 netdev_features_t features = dev->features;
3576 if (skb_is_gso(skb))
3577 features = gso_features_check(skb, dev, features);
3579 /* If encapsulation offload request, verify we are testing
3580 * hardware encapsulation features instead of standard
3581 * features for the netdev
3583 if (skb->encapsulation)
3584 features &= dev->hw_enc_features;
3586 if (skb_vlan_tagged(skb))
3587 features = netdev_intersect_features(features,
3588 dev->vlan_features |
3589 NETIF_F_HW_VLAN_CTAG_TX |
3590 NETIF_F_HW_VLAN_STAG_TX);
3592 if (dev->netdev_ops->ndo_features_check)
3593 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3596 features &= dflt_features_check(skb, dev, features);
3598 return harmonize_features(skb, features);
3600 EXPORT_SYMBOL(netif_skb_features);
3602 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3603 struct netdev_queue *txq, bool more)
3608 if (dev_nit_active(dev))
3609 dev_queue_xmit_nit(skb, dev);
3612 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3613 trace_net_dev_start_xmit(skb, dev);
3614 rc = netdev_start_xmit(skb, dev, txq, more);
3615 trace_net_dev_xmit(skb, rc, dev, len);
3620 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3621 struct netdev_queue *txq, int *ret)
3623 struct sk_buff *skb = first;
3624 int rc = NETDEV_TX_OK;
3627 struct sk_buff *next = skb->next;
3629 skb_mark_not_on_list(skb);
3630 rc = xmit_one(skb, dev, txq, next != NULL);
3631 if (unlikely(!dev_xmit_complete(rc))) {
3637 if (netif_tx_queue_stopped(txq) && skb) {
3638 rc = NETDEV_TX_BUSY;
3648 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3649 netdev_features_t features)
3651 if (skb_vlan_tag_present(skb) &&
3652 !vlan_hw_offload_capable(features, skb->vlan_proto))
3653 skb = __vlan_hwaccel_push_inside(skb);
3657 int skb_csum_hwoffload_help(struct sk_buff *skb,
3658 const netdev_features_t features)
3660 if (unlikely(skb_csum_is_sctp(skb)))
3661 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3662 skb_crc32c_csum_help(skb);
3664 if (features & NETIF_F_HW_CSUM)
3667 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3668 switch (skb->csum_offset) {
3669 case offsetof(struct tcphdr, check):
3670 case offsetof(struct udphdr, check):
3675 return skb_checksum_help(skb);
3677 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3679 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3681 netdev_features_t features;
3683 features = netif_skb_features(skb);
3684 skb = validate_xmit_vlan(skb, features);
3688 skb = sk_validate_xmit_skb(skb, dev);
3692 if (netif_needs_gso(skb, features)) {
3693 struct sk_buff *segs;
3695 segs = skb_gso_segment(skb, features);
3703 if (skb_needs_linearize(skb, features) &&
3704 __skb_linearize(skb))
3707 /* If packet is not checksummed and device does not
3708 * support checksumming for this protocol, complete
3709 * checksumming here.
3711 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3712 if (skb->encapsulation)
3713 skb_set_inner_transport_header(skb,
3714 skb_checksum_start_offset(skb));
3716 skb_set_transport_header(skb,
3717 skb_checksum_start_offset(skb));
3718 if (skb_csum_hwoffload_help(skb, features))
3723 skb = validate_xmit_xfrm(skb, features, again);
3730 atomic_long_inc(&dev->tx_dropped);
3734 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3736 struct sk_buff *next, *head = NULL, *tail;
3738 for (; skb != NULL; skb = next) {
3740 skb_mark_not_on_list(skb);
3742 /* in case skb wont be segmented, point to itself */
3745 skb = validate_xmit_skb(skb, dev, again);
3753 /* If skb was segmented, skb->prev points to
3754 * the last segment. If not, it still contains skb.
3760 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3762 static void qdisc_pkt_len_init(struct sk_buff *skb)
3764 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3766 qdisc_skb_cb(skb)->pkt_len = skb->len;
3768 /* To get more precise estimation of bytes sent on wire,
3769 * we add to pkt_len the headers size of all segments
3771 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3772 unsigned int hdr_len;
3773 u16 gso_segs = shinfo->gso_segs;
3775 /* mac layer + network layer */
3776 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3778 /* + transport layer */
3779 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3780 const struct tcphdr *th;
3781 struct tcphdr _tcphdr;
3783 th = skb_header_pointer(skb, skb_transport_offset(skb),
3784 sizeof(_tcphdr), &_tcphdr);
3786 hdr_len += __tcp_hdrlen(th);
3788 struct udphdr _udphdr;
3790 if (skb_header_pointer(skb, skb_transport_offset(skb),
3791 sizeof(_udphdr), &_udphdr))
3792 hdr_len += sizeof(struct udphdr);
3795 if (shinfo->gso_type & SKB_GSO_DODGY)
3796 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3799 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3803 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3804 struct sk_buff **to_free,
3805 struct netdev_queue *txq)
3809 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3810 if (rc == NET_XMIT_SUCCESS)
3811 trace_qdisc_enqueue(q, txq, skb);
3815 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3816 struct net_device *dev,
3817 struct netdev_queue *txq)
3819 spinlock_t *root_lock = qdisc_lock(q);
3820 struct sk_buff *to_free = NULL;
3824 qdisc_calculate_pkt_len(skb, q);
3826 if (q->flags & TCQ_F_NOLOCK) {
3827 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3828 qdisc_run_begin(q)) {
3829 /* Retest nolock_qdisc_is_empty() within the protection
3830 * of q->seqlock to protect from racing with requeuing.
3832 if (unlikely(!nolock_qdisc_is_empty(q))) {
3833 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3840 qdisc_bstats_cpu_update(q, skb);
3841 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3842 !nolock_qdisc_is_empty(q))
3846 return NET_XMIT_SUCCESS;
3849 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3853 if (unlikely(to_free))
3854 kfree_skb_list(to_free);
3859 * Heuristic to force contended enqueues to serialize on a
3860 * separate lock before trying to get qdisc main lock.
3861 * This permits qdisc->running owner to get the lock more
3862 * often and dequeue packets faster.
3864 contended = qdisc_is_running(q);
3865 if (unlikely(contended))
3866 spin_lock(&q->busylock);
3868 spin_lock(root_lock);
3869 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3870 __qdisc_drop(skb, &to_free);
3872 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3873 qdisc_run_begin(q)) {
3875 * This is a work-conserving queue; there are no old skbs
3876 * waiting to be sent out; and the qdisc is not running -
3877 * xmit the skb directly.
3880 qdisc_bstats_update(q, skb);
3882 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3883 if (unlikely(contended)) {
3884 spin_unlock(&q->busylock);
3891 rc = NET_XMIT_SUCCESS;
3893 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3894 if (qdisc_run_begin(q)) {
3895 if (unlikely(contended)) {
3896 spin_unlock(&q->busylock);
3903 spin_unlock(root_lock);
3904 if (unlikely(to_free))
3905 kfree_skb_list(to_free);
3906 if (unlikely(contended))
3907 spin_unlock(&q->busylock);
3911 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3912 static void skb_update_prio(struct sk_buff *skb)
3914 const struct netprio_map *map;
3915 const struct sock *sk;
3916 unsigned int prioidx;
3920 map = rcu_dereference_bh(skb->dev->priomap);
3923 sk = skb_to_full_sk(skb);
3927 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3929 if (prioidx < map->priomap_len)
3930 skb->priority = map->priomap[prioidx];
3933 #define skb_update_prio(skb)
3937 * dev_loopback_xmit - loop back @skb
3938 * @net: network namespace this loopback is happening in
3939 * @sk: sk needed to be a netfilter okfn
3940 * @skb: buffer to transmit
3942 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3944 skb_reset_mac_header(skb);
3945 __skb_pull(skb, skb_network_offset(skb));
3946 skb->pkt_type = PACKET_LOOPBACK;
3947 if (skb->ip_summed == CHECKSUM_NONE)
3948 skb->ip_summed = CHECKSUM_UNNECESSARY;
3949 WARN_ON(!skb_dst(skb));
3954 EXPORT_SYMBOL(dev_loopback_xmit);
3956 #ifdef CONFIG_NET_EGRESS
3957 static struct sk_buff *
3958 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3960 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3961 struct tcf_result cl_res;
3966 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3967 tc_skb_cb(skb)->mru = 0;
3968 tc_skb_cb(skb)->post_ct = false;
3969 mini_qdisc_bstats_cpu_update(miniq, skb);
3971 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3973 case TC_ACT_RECLASSIFY:
3974 skb->tc_index = TC_H_MIN(cl_res.classid);
3977 mini_qdisc_qstats_cpu_drop(miniq);
3978 *ret = NET_XMIT_DROP;
3984 *ret = NET_XMIT_SUCCESS;
3987 case TC_ACT_REDIRECT:
3988 /* No need to push/pop skb's mac_header here on egress! */
3989 skb_do_redirect(skb);
3990 *ret = NET_XMIT_SUCCESS;
3998 #endif /* CONFIG_NET_EGRESS */
4001 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4002 struct xps_dev_maps *dev_maps, unsigned int tci)
4004 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4005 struct xps_map *map;
4006 int queue_index = -1;
4008 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4011 tci *= dev_maps->num_tc;
4014 map = rcu_dereference(dev_maps->attr_map[tci]);
4017 queue_index = map->queues[0];
4019 queue_index = map->queues[reciprocal_scale(
4020 skb_get_hash(skb), map->len)];
4021 if (unlikely(queue_index >= dev->real_num_tx_queues))
4028 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4029 struct sk_buff *skb)
4032 struct xps_dev_maps *dev_maps;
4033 struct sock *sk = skb->sk;
4034 int queue_index = -1;
4036 if (!static_key_false(&xps_needed))
4040 if (!static_key_false(&xps_rxqs_needed))
4043 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4045 int tci = sk_rx_queue_get(sk);
4048 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4053 if (queue_index < 0) {
4054 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4056 unsigned int tci = skb->sender_cpu - 1;
4058 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4070 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4071 struct net_device *sb_dev)
4075 EXPORT_SYMBOL(dev_pick_tx_zero);
4077 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4078 struct net_device *sb_dev)
4080 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4082 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4084 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4085 struct net_device *sb_dev)
4087 struct sock *sk = skb->sk;
4088 int queue_index = sk_tx_queue_get(sk);
4090 sb_dev = sb_dev ? : dev;
4092 if (queue_index < 0 || skb->ooo_okay ||
4093 queue_index >= dev->real_num_tx_queues) {
4094 int new_index = get_xps_queue(dev, sb_dev, skb);
4097 new_index = skb_tx_hash(dev, sb_dev, skb);
4099 if (queue_index != new_index && sk &&
4101 rcu_access_pointer(sk->sk_dst_cache))
4102 sk_tx_queue_set(sk, new_index);
4104 queue_index = new_index;
4109 EXPORT_SYMBOL(netdev_pick_tx);
4111 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4112 struct sk_buff *skb,
4113 struct net_device *sb_dev)
4115 int queue_index = 0;
4118 u32 sender_cpu = skb->sender_cpu - 1;
4120 if (sender_cpu >= (u32)NR_CPUS)
4121 skb->sender_cpu = raw_smp_processor_id() + 1;
4124 if (dev->real_num_tx_queues != 1) {
4125 const struct net_device_ops *ops = dev->netdev_ops;
4127 if (ops->ndo_select_queue)
4128 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4130 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4132 queue_index = netdev_cap_txqueue(dev, queue_index);
4135 skb_set_queue_mapping(skb, queue_index);
4136 return netdev_get_tx_queue(dev, queue_index);
4140 * __dev_queue_xmit - transmit a buffer
4141 * @skb: buffer to transmit
4142 * @sb_dev: suboordinate device used for L2 forwarding offload
4144 * Queue a buffer for transmission to a network device. The caller must
4145 * have set the device and priority and built the buffer before calling
4146 * this function. The function can be called from an interrupt.
4148 * A negative errno code is returned on a failure. A success does not
4149 * guarantee the frame will be transmitted as it may be dropped due
4150 * to congestion or traffic shaping.
4152 * -----------------------------------------------------------------------------------
4153 * I notice this method can also return errors from the queue disciplines,
4154 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4157 * Regardless of the return value, the skb is consumed, so it is currently
4158 * difficult to retry a send to this method. (You can bump the ref count
4159 * before sending to hold a reference for retry if you are careful.)
4161 * When calling this method, interrupts MUST be enabled. This is because
4162 * the BH enable code must have IRQs enabled so that it will not deadlock.
4165 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4167 struct net_device *dev = skb->dev;
4168 struct netdev_queue *txq;
4173 skb_reset_mac_header(skb);
4174 skb_assert_len(skb);
4176 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4177 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4179 /* Disable soft irqs for various locks below. Also
4180 * stops preemption for RCU.
4184 skb_update_prio(skb);
4186 qdisc_pkt_len_init(skb);
4187 #ifdef CONFIG_NET_CLS_ACT
4188 skb->tc_at_ingress = 0;
4189 # ifdef CONFIG_NET_EGRESS
4190 if (static_branch_unlikely(&egress_needed_key)) {
4191 skb = sch_handle_egress(skb, &rc, dev);
4197 /* If device/qdisc don't need skb->dst, release it right now while
4198 * its hot in this cpu cache.
4200 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4205 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4206 q = rcu_dereference_bh(txq->qdisc);
4208 trace_net_dev_queue(skb);
4210 rc = __dev_xmit_skb(skb, q, dev, txq);
4214 /* The device has no queue. Common case for software devices:
4215 * loopback, all the sorts of tunnels...
4217 * Really, it is unlikely that netif_tx_lock protection is necessary
4218 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4220 * However, it is possible, that they rely on protection
4223 * Check this and shot the lock. It is not prone from deadlocks.
4224 *Either shot noqueue qdisc, it is even simpler 8)
4226 if (dev->flags & IFF_UP) {
4227 int cpu = smp_processor_id(); /* ok because BHs are off */
4229 /* Other cpus might concurrently change txq->xmit_lock_owner
4230 * to -1 or to their cpu id, but not to our id.
4232 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4233 if (dev_xmit_recursion())
4234 goto recursion_alert;
4236 skb = validate_xmit_skb(skb, dev, &again);
4240 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4241 HARD_TX_LOCK(dev, txq, cpu);
4243 if (!netif_xmit_stopped(txq)) {
4244 dev_xmit_recursion_inc();
4245 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4246 dev_xmit_recursion_dec();
4247 if (dev_xmit_complete(rc)) {
4248 HARD_TX_UNLOCK(dev, txq);
4252 HARD_TX_UNLOCK(dev, txq);
4253 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4256 /* Recursion is detected! It is possible,
4260 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4266 rcu_read_unlock_bh();
4268 atomic_long_inc(&dev->tx_dropped);
4269 kfree_skb_list(skb);
4272 rcu_read_unlock_bh();
4276 int dev_queue_xmit(struct sk_buff *skb)
4278 return __dev_queue_xmit(skb, NULL);
4280 EXPORT_SYMBOL(dev_queue_xmit);
4282 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4284 return __dev_queue_xmit(skb, sb_dev);
4286 EXPORT_SYMBOL(dev_queue_xmit_accel);
4288 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4290 struct net_device *dev = skb->dev;
4291 struct sk_buff *orig_skb = skb;
4292 struct netdev_queue *txq;
4293 int ret = NETDEV_TX_BUSY;
4296 if (unlikely(!netif_running(dev) ||
4297 !netif_carrier_ok(dev)))
4300 skb = validate_xmit_skb_list(skb, dev, &again);
4301 if (skb != orig_skb)
4304 skb_set_queue_mapping(skb, queue_id);
4305 txq = skb_get_tx_queue(dev, skb);
4306 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4310 dev_xmit_recursion_inc();
4311 HARD_TX_LOCK(dev, txq, smp_processor_id());
4312 if (!netif_xmit_frozen_or_drv_stopped(txq))
4313 ret = netdev_start_xmit(skb, dev, txq, false);
4314 HARD_TX_UNLOCK(dev, txq);
4315 dev_xmit_recursion_dec();
4320 atomic_long_inc(&dev->tx_dropped);
4321 kfree_skb_list(skb);
4322 return NET_XMIT_DROP;
4324 EXPORT_SYMBOL(__dev_direct_xmit);
4326 /*************************************************************************
4328 *************************************************************************/
4330 int netdev_max_backlog __read_mostly = 1000;
4331 EXPORT_SYMBOL(netdev_max_backlog);
4333 int netdev_tstamp_prequeue __read_mostly = 1;
4334 int netdev_budget __read_mostly = 300;
4335 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4336 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4337 int weight_p __read_mostly = 64; /* old backlog weight */
4338 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4339 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4340 int dev_rx_weight __read_mostly = 64;
4341 int dev_tx_weight __read_mostly = 64;
4342 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4343 int gro_normal_batch __read_mostly = 8;
4345 /* Called with irq disabled */
4346 static inline void ____napi_schedule(struct softnet_data *sd,
4347 struct napi_struct *napi)
4349 struct task_struct *thread;
4351 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4352 /* Paired with smp_mb__before_atomic() in
4353 * napi_enable()/dev_set_threaded().
4354 * Use READ_ONCE() to guarantee a complete
4355 * read on napi->thread. Only call
4356 * wake_up_process() when it's not NULL.
4358 thread = READ_ONCE(napi->thread);
4360 /* Avoid doing set_bit() if the thread is in
4361 * INTERRUPTIBLE state, cause napi_thread_wait()
4362 * makes sure to proceed with napi polling
4363 * if the thread is explicitly woken from here.
4365 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4366 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4367 wake_up_process(thread);
4372 list_add_tail(&napi->poll_list, &sd->poll_list);
4373 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4378 /* One global table that all flow-based protocols share. */
4379 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4380 EXPORT_SYMBOL(rps_sock_flow_table);
4381 u32 rps_cpu_mask __read_mostly;
4382 EXPORT_SYMBOL(rps_cpu_mask);
4384 struct static_key_false rps_needed __read_mostly;
4385 EXPORT_SYMBOL(rps_needed);
4386 struct static_key_false rfs_needed __read_mostly;
4387 EXPORT_SYMBOL(rfs_needed);
4389 static struct rps_dev_flow *
4390 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4391 struct rps_dev_flow *rflow, u16 next_cpu)
4393 if (next_cpu < nr_cpu_ids) {
4394 #ifdef CONFIG_RFS_ACCEL
4395 struct netdev_rx_queue *rxqueue;
4396 struct rps_dev_flow_table *flow_table;
4397 struct rps_dev_flow *old_rflow;
4402 /* Should we steer this flow to a different hardware queue? */
4403 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4404 !(dev->features & NETIF_F_NTUPLE))
4406 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4407 if (rxq_index == skb_get_rx_queue(skb))
4410 rxqueue = dev->_rx + rxq_index;
4411 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4414 flow_id = skb_get_hash(skb) & flow_table->mask;
4415 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4416 rxq_index, flow_id);
4420 rflow = &flow_table->flows[flow_id];
4422 if (old_rflow->filter == rflow->filter)
4423 old_rflow->filter = RPS_NO_FILTER;
4427 per_cpu(softnet_data, next_cpu).input_queue_head;
4430 rflow->cpu = next_cpu;
4435 * get_rps_cpu is called from netif_receive_skb and returns the target
4436 * CPU from the RPS map of the receiving queue for a given skb.
4437 * rcu_read_lock must be held on entry.
4439 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4440 struct rps_dev_flow **rflowp)
4442 const struct rps_sock_flow_table *sock_flow_table;
4443 struct netdev_rx_queue *rxqueue = dev->_rx;
4444 struct rps_dev_flow_table *flow_table;
4445 struct rps_map *map;
4450 if (skb_rx_queue_recorded(skb)) {
4451 u16 index = skb_get_rx_queue(skb);
4453 if (unlikely(index >= dev->real_num_rx_queues)) {
4454 WARN_ONCE(dev->real_num_rx_queues > 1,
4455 "%s received packet on queue %u, but number "
4456 "of RX queues is %u\n",
4457 dev->name, index, dev->real_num_rx_queues);
4463 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4465 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4466 map = rcu_dereference(rxqueue->rps_map);
4467 if (!flow_table && !map)
4470 skb_reset_network_header(skb);
4471 hash = skb_get_hash(skb);
4475 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4476 if (flow_table && sock_flow_table) {
4477 struct rps_dev_flow *rflow;
4481 /* First check into global flow table if there is a match.
4482 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4484 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4485 if ((ident ^ hash) & ~rps_cpu_mask)
4488 next_cpu = ident & rps_cpu_mask;
4490 /* OK, now we know there is a match,
4491 * we can look at the local (per receive queue) flow table
4493 rflow = &flow_table->flows[hash & flow_table->mask];
4497 * If the desired CPU (where last recvmsg was done) is
4498 * different from current CPU (one in the rx-queue flow
4499 * table entry), switch if one of the following holds:
4500 * - Current CPU is unset (>= nr_cpu_ids).
4501 * - Current CPU is offline.
4502 * - The current CPU's queue tail has advanced beyond the
4503 * last packet that was enqueued using this table entry.
4504 * This guarantees that all previous packets for the flow
4505 * have been dequeued, thus preserving in order delivery.
4507 if (unlikely(tcpu != next_cpu) &&
4508 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4509 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4510 rflow->last_qtail)) >= 0)) {
4512 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4515 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4525 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4526 if (cpu_online(tcpu)) {
4536 #ifdef CONFIG_RFS_ACCEL
4539 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4540 * @dev: Device on which the filter was set
4541 * @rxq_index: RX queue index
4542 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4543 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4545 * Drivers that implement ndo_rx_flow_steer() should periodically call
4546 * this function for each installed filter and remove the filters for
4547 * which it returns %true.
4549 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4550 u32 flow_id, u16 filter_id)
4552 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4553 struct rps_dev_flow_table *flow_table;
4554 struct rps_dev_flow *rflow;
4559 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4560 if (flow_table && flow_id <= flow_table->mask) {
4561 rflow = &flow_table->flows[flow_id];
4562 cpu = READ_ONCE(rflow->cpu);
4563 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4564 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4565 rflow->last_qtail) <
4566 (int)(10 * flow_table->mask)))
4572 EXPORT_SYMBOL(rps_may_expire_flow);
4574 #endif /* CONFIG_RFS_ACCEL */
4576 /* Called from hardirq (IPI) context */
4577 static void rps_trigger_softirq(void *data)
4579 struct softnet_data *sd = data;
4581 ____napi_schedule(sd, &sd->backlog);
4585 #endif /* CONFIG_RPS */
4588 * Check if this softnet_data structure is another cpu one
4589 * If yes, queue it to our IPI list and return 1
4592 static int rps_ipi_queued(struct softnet_data *sd)
4595 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4598 sd->rps_ipi_next = mysd->rps_ipi_list;
4599 mysd->rps_ipi_list = sd;
4601 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4604 #endif /* CONFIG_RPS */
4608 #ifdef CONFIG_NET_FLOW_LIMIT
4609 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4612 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4614 #ifdef CONFIG_NET_FLOW_LIMIT
4615 struct sd_flow_limit *fl;
4616 struct softnet_data *sd;
4617 unsigned int old_flow, new_flow;
4619 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4622 sd = this_cpu_ptr(&softnet_data);
4625 fl = rcu_dereference(sd->flow_limit);
4627 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4628 old_flow = fl->history[fl->history_head];
4629 fl->history[fl->history_head] = new_flow;
4632 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4634 if (likely(fl->buckets[old_flow]))
4635 fl->buckets[old_flow]--;
4637 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4649 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4650 * queue (may be a remote CPU queue).
4652 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4653 unsigned int *qtail)
4655 struct softnet_data *sd;
4656 unsigned long flags;
4659 sd = &per_cpu(softnet_data, cpu);
4661 local_irq_save(flags);
4664 if (!netif_running(skb->dev))
4666 qlen = skb_queue_len(&sd->input_pkt_queue);
4667 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4670 __skb_queue_tail(&sd->input_pkt_queue, skb);
4671 input_queue_tail_incr_save(sd, qtail);
4673 local_irq_restore(flags);
4674 return NET_RX_SUCCESS;
4677 /* Schedule NAPI for backlog device
4678 * We can use non atomic operation since we own the queue lock
4680 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4681 if (!rps_ipi_queued(sd))
4682 ____napi_schedule(sd, &sd->backlog);
4691 local_irq_restore(flags);
4693 atomic_long_inc(&skb->dev->rx_dropped);
4698 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4700 struct net_device *dev = skb->dev;
4701 struct netdev_rx_queue *rxqueue;
4705 if (skb_rx_queue_recorded(skb)) {
4706 u16 index = skb_get_rx_queue(skb);
4708 if (unlikely(index >= dev->real_num_rx_queues)) {
4709 WARN_ONCE(dev->real_num_rx_queues > 1,
4710 "%s received packet on queue %u, but number "
4711 "of RX queues is %u\n",
4712 dev->name, index, dev->real_num_rx_queues);
4714 return rxqueue; /* Return first rxqueue */
4721 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4722 struct bpf_prog *xdp_prog)
4724 void *orig_data, *orig_data_end, *hard_start;
4725 struct netdev_rx_queue *rxqueue;
4726 bool orig_bcast, orig_host;
4727 u32 mac_len, frame_sz;
4728 __be16 orig_eth_type;
4733 /* The XDP program wants to see the packet starting at the MAC
4736 mac_len = skb->data - skb_mac_header(skb);
4737 hard_start = skb->data - skb_headroom(skb);
4739 /* SKB "head" area always have tailroom for skb_shared_info */
4740 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4741 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4743 rxqueue = netif_get_rxqueue(skb);
4744 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4745 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4746 skb_headlen(skb) + mac_len, true);
4748 orig_data_end = xdp->data_end;
4749 orig_data = xdp->data;
4750 eth = (struct ethhdr *)xdp->data;
4751 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4752 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4753 orig_eth_type = eth->h_proto;
4755 act = bpf_prog_run_xdp(xdp_prog, xdp);
4757 /* check if bpf_xdp_adjust_head was used */
4758 off = xdp->data - orig_data;
4761 __skb_pull(skb, off);
4763 __skb_push(skb, -off);
4765 skb->mac_header += off;
4766 skb_reset_network_header(skb);
4769 /* check if bpf_xdp_adjust_tail was used */
4770 off = xdp->data_end - orig_data_end;
4772 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4773 skb->len += off; /* positive on grow, negative on shrink */
4776 /* check if XDP changed eth hdr such SKB needs update */
4777 eth = (struct ethhdr *)xdp->data;
4778 if ((orig_eth_type != eth->h_proto) ||
4779 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4780 skb->dev->dev_addr)) ||
4781 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4782 __skb_push(skb, ETH_HLEN);
4783 skb->pkt_type = PACKET_HOST;
4784 skb->protocol = eth_type_trans(skb, skb->dev);
4787 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4788 * before calling us again on redirect path. We do not call do_redirect
4789 * as we leave that up to the caller.
4791 * Caller is responsible for managing lifetime of skb (i.e. calling
4792 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4797 __skb_push(skb, mac_len);
4800 metalen = xdp->data - xdp->data_meta;
4802 skb_metadata_set(skb, metalen);
4809 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4810 struct xdp_buff *xdp,
4811 struct bpf_prog *xdp_prog)
4815 /* Reinjected packets coming from act_mirred or similar should
4816 * not get XDP generic processing.
4818 if (skb_is_redirected(skb))
4821 /* XDP packets must be linear and must have sufficient headroom
4822 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4823 * native XDP provides, thus we need to do it here as well.
4825 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4826 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4827 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4828 int troom = skb->tail + skb->data_len - skb->end;
4830 /* In case we have to go down the path and also linearize,
4831 * then lets do the pskb_expand_head() work just once here.
4833 if (pskb_expand_head(skb,
4834 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4835 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4837 if (skb_linearize(skb))
4841 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4848 bpf_warn_invalid_xdp_action(act);
4851 trace_xdp_exception(skb->dev, xdp_prog, act);
4862 /* When doing generic XDP we have to bypass the qdisc layer and the
4863 * network taps in order to match in-driver-XDP behavior.
4865 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4867 struct net_device *dev = skb->dev;
4868 struct netdev_queue *txq;
4869 bool free_skb = true;
4872 txq = netdev_core_pick_tx(dev, skb, NULL);
4873 cpu = smp_processor_id();
4874 HARD_TX_LOCK(dev, txq, cpu);
4875 if (!netif_xmit_stopped(txq)) {
4876 rc = netdev_start_xmit(skb, dev, txq, 0);
4877 if (dev_xmit_complete(rc))
4880 HARD_TX_UNLOCK(dev, txq);
4882 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4887 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4889 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4892 struct xdp_buff xdp;
4896 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4897 if (act != XDP_PASS) {
4900 err = xdp_do_generic_redirect(skb->dev, skb,
4906 generic_xdp_tx(skb, xdp_prog);
4917 EXPORT_SYMBOL_GPL(do_xdp_generic);
4919 static int netif_rx_internal(struct sk_buff *skb)
4923 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4925 trace_netif_rx(skb);
4928 if (static_branch_unlikely(&rps_needed)) {
4929 struct rps_dev_flow voidflow, *rflow = &voidflow;
4935 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4937 cpu = smp_processor_id();
4939 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4948 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4955 * netif_rx - post buffer to the network code
4956 * @skb: buffer to post
4958 * This function receives a packet from a device driver and queues it for
4959 * the upper (protocol) levels to process. It always succeeds. The buffer
4960 * may be dropped during processing for congestion control or by the
4964 * NET_RX_SUCCESS (no congestion)
4965 * NET_RX_DROP (packet was dropped)
4969 int netif_rx(struct sk_buff *skb)
4973 trace_netif_rx_entry(skb);
4975 ret = netif_rx_internal(skb);
4976 trace_netif_rx_exit(ret);
4980 EXPORT_SYMBOL(netif_rx);
4982 int netif_rx_ni(struct sk_buff *skb)
4986 trace_netif_rx_ni_entry(skb);
4989 err = netif_rx_internal(skb);
4990 if (local_softirq_pending())
4993 trace_netif_rx_ni_exit(err);
4997 EXPORT_SYMBOL(netif_rx_ni);
4999 int netif_rx_any_context(struct sk_buff *skb)
5002 * If invoked from contexts which do not invoke bottom half
5003 * processing either at return from interrupt or when softrqs are
5004 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
5008 return netif_rx(skb);
5010 return netif_rx_ni(skb);
5012 EXPORT_SYMBOL(netif_rx_any_context);
5014 static __latent_entropy void net_tx_action(struct softirq_action *h)
5016 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5018 if (sd->completion_queue) {
5019 struct sk_buff *clist;
5021 local_irq_disable();
5022 clist = sd->completion_queue;
5023 sd->completion_queue = NULL;
5027 struct sk_buff *skb = clist;
5029 clist = clist->next;
5031 WARN_ON(refcount_read(&skb->users));
5032 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5033 trace_consume_skb(skb);
5035 trace_kfree_skb(skb, net_tx_action,
5036 SKB_DROP_REASON_NOT_SPECIFIED);
5038 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5041 __kfree_skb_defer(skb);
5045 if (sd->output_queue) {
5048 local_irq_disable();
5049 head = sd->output_queue;
5050 sd->output_queue = NULL;
5051 sd->output_queue_tailp = &sd->output_queue;
5057 struct Qdisc *q = head;
5058 spinlock_t *root_lock = NULL;
5060 head = head->next_sched;
5062 /* We need to make sure head->next_sched is read
5063 * before clearing __QDISC_STATE_SCHED
5065 smp_mb__before_atomic();
5067 if (!(q->flags & TCQ_F_NOLOCK)) {
5068 root_lock = qdisc_lock(q);
5069 spin_lock(root_lock);
5070 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5072 /* There is a synchronize_net() between
5073 * STATE_DEACTIVATED flag being set and
5074 * qdisc_reset()/some_qdisc_is_busy() in
5075 * dev_deactivate(), so we can safely bail out
5076 * early here to avoid data race between
5077 * qdisc_deactivate() and some_qdisc_is_busy()
5078 * for lockless qdisc.
5080 clear_bit(__QDISC_STATE_SCHED, &q->state);
5084 clear_bit(__QDISC_STATE_SCHED, &q->state);
5087 spin_unlock(root_lock);
5093 xfrm_dev_backlog(sd);
5096 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5097 /* This hook is defined here for ATM LANE */
5098 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5099 unsigned char *addr) __read_mostly;
5100 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5103 static inline struct sk_buff *
5104 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5105 struct net_device *orig_dev, bool *another)
5107 #ifdef CONFIG_NET_CLS_ACT
5108 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5109 struct tcf_result cl_res;
5111 /* If there's at least one ingress present somewhere (so
5112 * we get here via enabled static key), remaining devices
5113 * that are not configured with an ingress qdisc will bail
5120 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5124 qdisc_skb_cb(skb)->pkt_len = skb->len;
5125 tc_skb_cb(skb)->mru = 0;
5126 tc_skb_cb(skb)->post_ct = false;
5127 skb->tc_at_ingress = 1;
5128 mini_qdisc_bstats_cpu_update(miniq, skb);
5130 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5132 case TC_ACT_RECLASSIFY:
5133 skb->tc_index = TC_H_MIN(cl_res.classid);
5136 mini_qdisc_qstats_cpu_drop(miniq);
5144 case TC_ACT_REDIRECT:
5145 /* skb_mac_header check was done by cls/act_bpf, so
5146 * we can safely push the L2 header back before
5147 * redirecting to another netdev
5149 __skb_push(skb, skb->mac_len);
5150 if (skb_do_redirect(skb) == -EAGAIN) {
5151 __skb_pull(skb, skb->mac_len);
5156 case TC_ACT_CONSUMED:
5161 #endif /* CONFIG_NET_CLS_ACT */
5166 * netdev_is_rx_handler_busy - check if receive handler is registered
5167 * @dev: device to check
5169 * Check if a receive handler is already registered for a given device.
5170 * Return true if there one.
5172 * The caller must hold the rtnl_mutex.
5174 bool netdev_is_rx_handler_busy(struct net_device *dev)
5177 return dev && rtnl_dereference(dev->rx_handler);
5179 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5182 * netdev_rx_handler_register - register receive handler
5183 * @dev: device to register a handler for
5184 * @rx_handler: receive handler to register
5185 * @rx_handler_data: data pointer that is used by rx handler
5187 * Register a receive handler for a device. This handler will then be
5188 * called from __netif_receive_skb. A negative errno code is returned
5191 * The caller must hold the rtnl_mutex.
5193 * For a general description of rx_handler, see enum rx_handler_result.
5195 int netdev_rx_handler_register(struct net_device *dev,
5196 rx_handler_func_t *rx_handler,
5197 void *rx_handler_data)
5199 if (netdev_is_rx_handler_busy(dev))
5202 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5205 /* Note: rx_handler_data must be set before rx_handler */
5206 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5207 rcu_assign_pointer(dev->rx_handler, rx_handler);
5211 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5214 * netdev_rx_handler_unregister - unregister receive handler
5215 * @dev: device to unregister a handler from
5217 * Unregister a receive handler from a device.
5219 * The caller must hold the rtnl_mutex.
5221 void netdev_rx_handler_unregister(struct net_device *dev)
5225 RCU_INIT_POINTER(dev->rx_handler, NULL);
5226 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5227 * section has a guarantee to see a non NULL rx_handler_data
5231 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5233 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5236 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5237 * the special handling of PFMEMALLOC skbs.
5239 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5241 switch (skb->protocol) {
5242 case htons(ETH_P_ARP):
5243 case htons(ETH_P_IP):
5244 case htons(ETH_P_IPV6):
5245 case htons(ETH_P_8021Q):
5246 case htons(ETH_P_8021AD):
5253 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5254 int *ret, struct net_device *orig_dev)
5256 if (nf_hook_ingress_active(skb)) {
5260 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5265 ingress_retval = nf_hook_ingress(skb);
5267 return ingress_retval;
5272 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5273 struct packet_type **ppt_prev)
5275 struct packet_type *ptype, *pt_prev;
5276 rx_handler_func_t *rx_handler;
5277 struct sk_buff *skb = *pskb;
5278 struct net_device *orig_dev;
5279 bool deliver_exact = false;
5280 int ret = NET_RX_DROP;
5283 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5285 trace_netif_receive_skb(skb);
5287 orig_dev = skb->dev;
5289 skb_reset_network_header(skb);
5290 if (!skb_transport_header_was_set(skb))
5291 skb_reset_transport_header(skb);
5292 skb_reset_mac_len(skb);
5297 skb->skb_iif = skb->dev->ifindex;
5299 __this_cpu_inc(softnet_data.processed);
5301 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5305 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5308 if (ret2 != XDP_PASS) {
5314 if (eth_type_vlan(skb->protocol)) {
5315 skb = skb_vlan_untag(skb);
5320 if (skb_skip_tc_classify(skb))
5326 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5328 ret = deliver_skb(skb, pt_prev, orig_dev);
5332 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5334 ret = deliver_skb(skb, pt_prev, orig_dev);
5339 #ifdef CONFIG_NET_INGRESS
5340 if (static_branch_unlikely(&ingress_needed_key)) {
5341 bool another = false;
5343 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5350 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5354 skb_reset_redirect(skb);
5356 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5359 if (skb_vlan_tag_present(skb)) {
5361 ret = deliver_skb(skb, pt_prev, orig_dev);
5364 if (vlan_do_receive(&skb))
5366 else if (unlikely(!skb))
5370 rx_handler = rcu_dereference(skb->dev->rx_handler);
5373 ret = deliver_skb(skb, pt_prev, orig_dev);
5376 switch (rx_handler(&skb)) {
5377 case RX_HANDLER_CONSUMED:
5378 ret = NET_RX_SUCCESS;
5380 case RX_HANDLER_ANOTHER:
5382 case RX_HANDLER_EXACT:
5383 deliver_exact = true;
5385 case RX_HANDLER_PASS:
5392 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5394 if (skb_vlan_tag_get_id(skb)) {
5395 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5398 skb->pkt_type = PACKET_OTHERHOST;
5399 } else if (eth_type_vlan(skb->protocol)) {
5400 /* Outer header is 802.1P with vlan 0, inner header is
5401 * 802.1Q or 802.1AD and vlan_do_receive() above could
5402 * not find vlan dev for vlan id 0.
5404 __vlan_hwaccel_clear_tag(skb);
5405 skb = skb_vlan_untag(skb);
5408 if (vlan_do_receive(&skb))
5409 /* After stripping off 802.1P header with vlan 0
5410 * vlan dev is found for inner header.
5413 else if (unlikely(!skb))
5416 /* We have stripped outer 802.1P vlan 0 header.
5417 * But could not find vlan dev.
5418 * check again for vlan id to set OTHERHOST.
5422 /* Note: we might in the future use prio bits
5423 * and set skb->priority like in vlan_do_receive()
5424 * For the time being, just ignore Priority Code Point
5426 __vlan_hwaccel_clear_tag(skb);
5429 type = skb->protocol;
5431 /* deliver only exact match when indicated */
5432 if (likely(!deliver_exact)) {
5433 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5434 &ptype_base[ntohs(type) &
5438 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5439 &orig_dev->ptype_specific);
5441 if (unlikely(skb->dev != orig_dev)) {
5442 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5443 &skb->dev->ptype_specific);
5447 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5449 *ppt_prev = pt_prev;
5453 atomic_long_inc(&skb->dev->rx_dropped);
5455 atomic_long_inc(&skb->dev->rx_nohandler);
5457 /* Jamal, now you will not able to escape explaining
5458 * me how you were going to use this. :-)
5464 /* The invariant here is that if *ppt_prev is not NULL
5465 * then skb should also be non-NULL.
5467 * Apparently *ppt_prev assignment above holds this invariant due to
5468 * skb dereferencing near it.
5474 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5476 struct net_device *orig_dev = skb->dev;
5477 struct packet_type *pt_prev = NULL;
5480 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5482 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5483 skb->dev, pt_prev, orig_dev);
5488 * netif_receive_skb_core - special purpose version of netif_receive_skb
5489 * @skb: buffer to process
5491 * More direct receive version of netif_receive_skb(). It should
5492 * only be used by callers that have a need to skip RPS and Generic XDP.
5493 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5495 * This function may only be called from softirq context and interrupts
5496 * should be enabled.
5498 * Return values (usually ignored):
5499 * NET_RX_SUCCESS: no congestion
5500 * NET_RX_DROP: packet was dropped
5502 int netif_receive_skb_core(struct sk_buff *skb)
5507 ret = __netif_receive_skb_one_core(skb, false);
5512 EXPORT_SYMBOL(netif_receive_skb_core);
5514 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5515 struct packet_type *pt_prev,
5516 struct net_device *orig_dev)
5518 struct sk_buff *skb, *next;
5522 if (list_empty(head))
5524 if (pt_prev->list_func != NULL)
5525 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5526 ip_list_rcv, head, pt_prev, orig_dev);
5528 list_for_each_entry_safe(skb, next, head, list) {
5529 skb_list_del_init(skb);
5530 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5534 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5536 /* Fast-path assumptions:
5537 * - There is no RX handler.
5538 * - Only one packet_type matches.
5539 * If either of these fails, we will end up doing some per-packet
5540 * processing in-line, then handling the 'last ptype' for the whole
5541 * sublist. This can't cause out-of-order delivery to any single ptype,
5542 * because the 'last ptype' must be constant across the sublist, and all
5543 * other ptypes are handled per-packet.
5545 /* Current (common) ptype of sublist */
5546 struct packet_type *pt_curr = NULL;
5547 /* Current (common) orig_dev of sublist */
5548 struct net_device *od_curr = NULL;
5549 struct list_head sublist;
5550 struct sk_buff *skb, *next;
5552 INIT_LIST_HEAD(&sublist);
5553 list_for_each_entry_safe(skb, next, head, list) {
5554 struct net_device *orig_dev = skb->dev;
5555 struct packet_type *pt_prev = NULL;
5557 skb_list_del_init(skb);
5558 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5561 if (pt_curr != pt_prev || od_curr != orig_dev) {
5562 /* dispatch old sublist */
5563 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5564 /* start new sublist */
5565 INIT_LIST_HEAD(&sublist);
5569 list_add_tail(&skb->list, &sublist);
5572 /* dispatch final sublist */
5573 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5576 static int __netif_receive_skb(struct sk_buff *skb)
5580 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5581 unsigned int noreclaim_flag;
5584 * PFMEMALLOC skbs are special, they should
5585 * - be delivered to SOCK_MEMALLOC sockets only
5586 * - stay away from userspace
5587 * - have bounded memory usage
5589 * Use PF_MEMALLOC as this saves us from propagating the allocation
5590 * context down to all allocation sites.
5592 noreclaim_flag = memalloc_noreclaim_save();
5593 ret = __netif_receive_skb_one_core(skb, true);
5594 memalloc_noreclaim_restore(noreclaim_flag);
5596 ret = __netif_receive_skb_one_core(skb, false);
5601 static void __netif_receive_skb_list(struct list_head *head)
5603 unsigned long noreclaim_flag = 0;
5604 struct sk_buff *skb, *next;
5605 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5607 list_for_each_entry_safe(skb, next, head, list) {
5608 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5609 struct list_head sublist;
5611 /* Handle the previous sublist */
5612 list_cut_before(&sublist, head, &skb->list);
5613 if (!list_empty(&sublist))
5614 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5615 pfmemalloc = !pfmemalloc;
5616 /* See comments in __netif_receive_skb */
5618 noreclaim_flag = memalloc_noreclaim_save();
5620 memalloc_noreclaim_restore(noreclaim_flag);
5623 /* Handle the remaining sublist */
5624 if (!list_empty(head))
5625 __netif_receive_skb_list_core(head, pfmemalloc);
5626 /* Restore pflags */
5628 memalloc_noreclaim_restore(noreclaim_flag);
5631 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5633 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5634 struct bpf_prog *new = xdp->prog;
5637 switch (xdp->command) {
5638 case XDP_SETUP_PROG:
5639 rcu_assign_pointer(dev->xdp_prog, new);
5644 static_branch_dec(&generic_xdp_needed_key);
5645 } else if (new && !old) {
5646 static_branch_inc(&generic_xdp_needed_key);
5647 dev_disable_lro(dev);
5648 dev_disable_gro_hw(dev);
5660 static int netif_receive_skb_internal(struct sk_buff *skb)
5664 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5666 if (skb_defer_rx_timestamp(skb))
5667 return NET_RX_SUCCESS;
5671 if (static_branch_unlikely(&rps_needed)) {
5672 struct rps_dev_flow voidflow, *rflow = &voidflow;
5673 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5676 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5682 ret = __netif_receive_skb(skb);
5687 static void netif_receive_skb_list_internal(struct list_head *head)
5689 struct sk_buff *skb, *next;
5690 struct list_head sublist;
5692 INIT_LIST_HEAD(&sublist);
5693 list_for_each_entry_safe(skb, next, head, list) {
5694 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5695 skb_list_del_init(skb);
5696 if (!skb_defer_rx_timestamp(skb))
5697 list_add_tail(&skb->list, &sublist);
5699 list_splice_init(&sublist, head);
5703 if (static_branch_unlikely(&rps_needed)) {
5704 list_for_each_entry_safe(skb, next, head, list) {
5705 struct rps_dev_flow voidflow, *rflow = &voidflow;
5706 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5709 /* Will be handled, remove from list */
5710 skb_list_del_init(skb);
5711 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5716 __netif_receive_skb_list(head);
5721 * netif_receive_skb - process receive buffer from network
5722 * @skb: buffer to process
5724 * netif_receive_skb() is the main receive data processing function.
5725 * It always succeeds. The buffer may be dropped during processing
5726 * for congestion control or by the protocol layers.
5728 * This function may only be called from softirq context and interrupts
5729 * should be enabled.
5731 * Return values (usually ignored):
5732 * NET_RX_SUCCESS: no congestion
5733 * NET_RX_DROP: packet was dropped
5735 int netif_receive_skb(struct sk_buff *skb)
5739 trace_netif_receive_skb_entry(skb);
5741 ret = netif_receive_skb_internal(skb);
5742 trace_netif_receive_skb_exit(ret);
5746 EXPORT_SYMBOL(netif_receive_skb);
5749 * netif_receive_skb_list - process many receive buffers from network
5750 * @head: list of skbs to process.
5752 * Since return value of netif_receive_skb() is normally ignored, and
5753 * wouldn't be meaningful for a list, this function returns void.
5755 * This function may only be called from softirq context and interrupts
5756 * should be enabled.
5758 void netif_receive_skb_list(struct list_head *head)
5760 struct sk_buff *skb;
5762 if (list_empty(head))
5764 if (trace_netif_receive_skb_list_entry_enabled()) {
5765 list_for_each_entry(skb, head, list)
5766 trace_netif_receive_skb_list_entry(skb);
5768 netif_receive_skb_list_internal(head);
5769 trace_netif_receive_skb_list_exit(0);
5771 EXPORT_SYMBOL(netif_receive_skb_list);
5773 static DEFINE_PER_CPU(struct work_struct, flush_works);
5775 /* Network device is going away, flush any packets still pending */
5776 static void flush_backlog(struct work_struct *work)
5778 struct sk_buff *skb, *tmp;
5779 struct softnet_data *sd;
5782 sd = this_cpu_ptr(&softnet_data);
5784 local_irq_disable();
5786 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5787 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5788 __skb_unlink(skb, &sd->input_pkt_queue);
5789 dev_kfree_skb_irq(skb);
5790 input_queue_head_incr(sd);
5796 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5797 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5798 __skb_unlink(skb, &sd->process_queue);
5800 input_queue_head_incr(sd);
5806 static bool flush_required(int cpu)
5808 #if IS_ENABLED(CONFIG_RPS)
5809 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5812 local_irq_disable();
5815 /* as insertion into process_queue happens with the rps lock held,
5816 * process_queue access may race only with dequeue
5818 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5819 !skb_queue_empty_lockless(&sd->process_queue);
5825 /* without RPS we can't safely check input_pkt_queue: during a
5826 * concurrent remote skb_queue_splice() we can detect as empty both
5827 * input_pkt_queue and process_queue even if the latter could end-up
5828 * containing a lot of packets.
5833 static void flush_all_backlogs(void)
5835 static cpumask_t flush_cpus;
5838 /* since we are under rtnl lock protection we can use static data
5839 * for the cpumask and avoid allocating on stack the possibly
5846 cpumask_clear(&flush_cpus);
5847 for_each_online_cpu(cpu) {
5848 if (flush_required(cpu)) {
5849 queue_work_on(cpu, system_highpri_wq,
5850 per_cpu_ptr(&flush_works, cpu));
5851 cpumask_set_cpu(cpu, &flush_cpus);
5855 /* we can have in flight packet[s] on the cpus we are not flushing,
5856 * synchronize_net() in unregister_netdevice_many() will take care of
5859 for_each_cpu(cpu, &flush_cpus)
5860 flush_work(per_cpu_ptr(&flush_works, cpu));
5865 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5866 static void gro_normal_list(struct napi_struct *napi)
5868 if (!napi->rx_count)
5870 netif_receive_skb_list_internal(&napi->rx_list);
5871 INIT_LIST_HEAD(&napi->rx_list);
5875 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5876 * pass the whole batch up to the stack.
5878 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5880 list_add_tail(&skb->list, &napi->rx_list);
5881 napi->rx_count += segs;
5882 if (napi->rx_count >= gro_normal_batch)
5883 gro_normal_list(napi);
5886 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5888 struct packet_offload *ptype;
5889 __be16 type = skb->protocol;
5890 struct list_head *head = &offload_base;
5893 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5895 if (NAPI_GRO_CB(skb)->count == 1) {
5896 skb_shinfo(skb)->gso_size = 0;
5901 list_for_each_entry_rcu(ptype, head, list) {
5902 if (ptype->type != type || !ptype->callbacks.gro_complete)
5905 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5906 ipv6_gro_complete, inet_gro_complete,
5913 WARN_ON(&ptype->list == head);
5915 return NET_RX_SUCCESS;
5919 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5920 return NET_RX_SUCCESS;
5923 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5926 struct list_head *head = &napi->gro_hash[index].list;
5927 struct sk_buff *skb, *p;
5929 list_for_each_entry_safe_reverse(skb, p, head, list) {
5930 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5932 skb_list_del_init(skb);
5933 napi_gro_complete(napi, skb);
5934 napi->gro_hash[index].count--;
5937 if (!napi->gro_hash[index].count)
5938 __clear_bit(index, &napi->gro_bitmask);
5941 /* napi->gro_hash[].list contains packets ordered by age.
5942 * youngest packets at the head of it.
5943 * Complete skbs in reverse order to reduce latencies.
5945 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5947 unsigned long bitmask = napi->gro_bitmask;
5948 unsigned int i, base = ~0U;
5950 while ((i = ffs(bitmask)) != 0) {
5953 __napi_gro_flush_chain(napi, base, flush_old);
5956 EXPORT_SYMBOL(napi_gro_flush);
5958 static void gro_list_prepare(const struct list_head *head,
5959 const struct sk_buff *skb)
5961 unsigned int maclen = skb->dev->hard_header_len;
5962 u32 hash = skb_get_hash_raw(skb);
5965 list_for_each_entry(p, head, list) {
5966 unsigned long diffs;
5968 NAPI_GRO_CB(p)->flush = 0;
5970 if (hash != skb_get_hash_raw(p)) {
5971 NAPI_GRO_CB(p)->same_flow = 0;
5975 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5976 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5977 if (skb_vlan_tag_present(p))
5978 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5979 diffs |= skb_metadata_differs(p, skb);
5980 if (maclen == ETH_HLEN)
5981 diffs |= compare_ether_header(skb_mac_header(p),
5982 skb_mac_header(skb));
5984 diffs = memcmp(skb_mac_header(p),
5985 skb_mac_header(skb),
5988 /* in most common scenarions 'slow_gro' is 0
5989 * otherwise we are already on some slower paths
5990 * either skip all the infrequent tests altogether or
5991 * avoid trying too hard to skip each of them individually
5993 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
5994 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5995 struct tc_skb_ext *skb_ext;
5996 struct tc_skb_ext *p_ext;
5999 diffs |= p->sk != skb->sk;
6000 diffs |= skb_metadata_dst_cmp(p, skb);
6001 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
6003 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
6004 skb_ext = skb_ext_find(skb, TC_SKB_EXT);
6005 p_ext = skb_ext_find(p, TC_SKB_EXT);
6007 diffs |= (!!p_ext) ^ (!!skb_ext);
6008 if (!diffs && unlikely(skb_ext))
6009 diffs |= p_ext->chain ^ skb_ext->chain;
6013 NAPI_GRO_CB(p)->same_flow = !diffs;
6017 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
6019 const struct skb_shared_info *pinfo = skb_shinfo(skb);
6020 const skb_frag_t *frag0 = &pinfo->frags[0];
6022 NAPI_GRO_CB(skb)->data_offset = 0;
6023 NAPI_GRO_CB(skb)->frag0 = NULL;
6024 NAPI_GRO_CB(skb)->frag0_len = 0;
6026 if (!skb_headlen(skb) && pinfo->nr_frags &&
6027 !PageHighMem(skb_frag_page(frag0)) &&
6028 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
6029 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
6030 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
6031 skb_frag_size(frag0),
6032 skb->end - skb->tail);
6036 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
6038 struct skb_shared_info *pinfo = skb_shinfo(skb);
6040 BUG_ON(skb->end - skb->tail < grow);
6042 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
6044 skb->data_len -= grow;
6047 skb_frag_off_add(&pinfo->frags[0], grow);
6048 skb_frag_size_sub(&pinfo->frags[0], grow);
6050 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
6051 skb_frag_unref(skb, 0);
6052 memmove(pinfo->frags, pinfo->frags + 1,
6053 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
6057 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6059 struct sk_buff *oldest;
6061 oldest = list_last_entry(head, struct sk_buff, list);
6063 /* We are called with head length >= MAX_GRO_SKBS, so this is
6066 if (WARN_ON_ONCE(!oldest))
6069 /* Do not adjust napi->gro_hash[].count, caller is adding a new
6072 skb_list_del_init(oldest);
6073 napi_gro_complete(napi, oldest);
6076 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6078 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6079 struct gro_list *gro_list = &napi->gro_hash[bucket];
6080 struct list_head *head = &offload_base;
6081 struct packet_offload *ptype;
6082 __be16 type = skb->protocol;
6083 struct sk_buff *pp = NULL;
6084 enum gro_result ret;
6088 if (netif_elide_gro(skb->dev))
6091 gro_list_prepare(&gro_list->list, skb);
6094 list_for_each_entry_rcu(ptype, head, list) {
6095 if (ptype->type != type || !ptype->callbacks.gro_receive)
6098 skb_set_network_header(skb, skb_gro_offset(skb));
6099 skb_reset_mac_len(skb);
6100 NAPI_GRO_CB(skb)->same_flow = 0;
6101 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6102 NAPI_GRO_CB(skb)->free = 0;
6103 NAPI_GRO_CB(skb)->encap_mark = 0;
6104 NAPI_GRO_CB(skb)->recursion_counter = 0;
6105 NAPI_GRO_CB(skb)->is_fou = 0;
6106 NAPI_GRO_CB(skb)->is_atomic = 1;
6107 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6109 /* Setup for GRO checksum validation */
6110 switch (skb->ip_summed) {
6111 case CHECKSUM_COMPLETE:
6112 NAPI_GRO_CB(skb)->csum = skb->csum;
6113 NAPI_GRO_CB(skb)->csum_valid = 1;
6114 NAPI_GRO_CB(skb)->csum_cnt = 0;
6116 case CHECKSUM_UNNECESSARY:
6117 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6118 NAPI_GRO_CB(skb)->csum_valid = 0;
6121 NAPI_GRO_CB(skb)->csum_cnt = 0;
6122 NAPI_GRO_CB(skb)->csum_valid = 0;
6125 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6126 ipv6_gro_receive, inet_gro_receive,
6127 &gro_list->list, skb);
6132 if (&ptype->list == head)
6135 if (PTR_ERR(pp) == -EINPROGRESS) {
6140 same_flow = NAPI_GRO_CB(skb)->same_flow;
6141 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6144 skb_list_del_init(pp);
6145 napi_gro_complete(napi, pp);
6152 if (NAPI_GRO_CB(skb)->flush)
6155 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6156 gro_flush_oldest(napi, &gro_list->list);
6160 NAPI_GRO_CB(skb)->count = 1;
6161 NAPI_GRO_CB(skb)->age = jiffies;
6162 NAPI_GRO_CB(skb)->last = skb;
6163 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6164 list_add(&skb->list, &gro_list->list);
6168 grow = skb_gro_offset(skb) - skb_headlen(skb);
6170 gro_pull_from_frag0(skb, grow);
6172 if (gro_list->count) {
6173 if (!test_bit(bucket, &napi->gro_bitmask))
6174 __set_bit(bucket, &napi->gro_bitmask);
6175 } else if (test_bit(bucket, &napi->gro_bitmask)) {
6176 __clear_bit(bucket, &napi->gro_bitmask);
6186 struct packet_offload *gro_find_receive_by_type(__be16 type)
6188 struct list_head *offload_head = &offload_base;
6189 struct packet_offload *ptype;
6191 list_for_each_entry_rcu(ptype, offload_head, list) {
6192 if (ptype->type != type || !ptype->callbacks.gro_receive)
6198 EXPORT_SYMBOL(gro_find_receive_by_type);
6200 struct packet_offload *gro_find_complete_by_type(__be16 type)
6202 struct list_head *offload_head = &offload_base;
6203 struct packet_offload *ptype;
6205 list_for_each_entry_rcu(ptype, offload_head, list) {
6206 if (ptype->type != type || !ptype->callbacks.gro_complete)
6212 EXPORT_SYMBOL(gro_find_complete_by_type);
6214 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6215 struct sk_buff *skb,
6220 gro_normal_one(napi, skb, 1);
6223 case GRO_MERGED_FREE:
6224 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6225 napi_skb_free_stolen_head(skb);
6226 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
6229 __kfree_skb_defer(skb);
6241 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6245 skb_mark_napi_id(skb, napi);
6246 trace_napi_gro_receive_entry(skb);
6248 skb_gro_reset_offset(skb, 0);
6250 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6251 trace_napi_gro_receive_exit(ret);
6255 EXPORT_SYMBOL(napi_gro_receive);
6257 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6259 if (unlikely(skb->pfmemalloc)) {
6263 __skb_pull(skb, skb_headlen(skb));
6264 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6265 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6266 __vlan_hwaccel_clear_tag(skb);
6267 skb->dev = napi->dev;
6270 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6271 skb->pkt_type = PACKET_HOST;
6273 skb->encapsulation = 0;
6274 skb_shinfo(skb)->gso_type = 0;
6275 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6276 if (unlikely(skb->slow_gro)) {
6286 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6288 struct sk_buff *skb = napi->skb;
6291 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6294 skb_mark_napi_id(skb, napi);
6299 EXPORT_SYMBOL(napi_get_frags);
6301 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6302 struct sk_buff *skb,
6308 __skb_push(skb, ETH_HLEN);
6309 skb->protocol = eth_type_trans(skb, skb->dev);
6310 if (ret == GRO_NORMAL)
6311 gro_normal_one(napi, skb, 1);
6314 case GRO_MERGED_FREE:
6315 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6316 napi_skb_free_stolen_head(skb);
6318 napi_reuse_skb(napi, skb);
6329 /* Upper GRO stack assumes network header starts at gro_offset=0
6330 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6331 * We copy ethernet header into skb->data to have a common layout.
6333 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6335 struct sk_buff *skb = napi->skb;
6336 const struct ethhdr *eth;
6337 unsigned int hlen = sizeof(*eth);
6341 skb_reset_mac_header(skb);
6342 skb_gro_reset_offset(skb, hlen);
6344 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6345 eth = skb_gro_header_slow(skb, hlen, 0);
6346 if (unlikely(!eth)) {
6347 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6348 __func__, napi->dev->name);
6349 napi_reuse_skb(napi, skb);
6353 eth = (const struct ethhdr *)skb->data;
6354 gro_pull_from_frag0(skb, hlen);
6355 NAPI_GRO_CB(skb)->frag0 += hlen;
6356 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6358 __skb_pull(skb, hlen);
6361 * This works because the only protocols we care about don't require
6363 * We'll fix it up properly in napi_frags_finish()
6365 skb->protocol = eth->h_proto;
6370 gro_result_t napi_gro_frags(struct napi_struct *napi)
6373 struct sk_buff *skb = napi_frags_skb(napi);
6375 trace_napi_gro_frags_entry(skb);
6377 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6378 trace_napi_gro_frags_exit(ret);
6382 EXPORT_SYMBOL(napi_gro_frags);
6384 /* Compute the checksum from gro_offset and return the folded value
6385 * after adding in any pseudo checksum.
6387 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6392 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6394 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6395 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6396 /* See comments in __skb_checksum_complete(). */
6398 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6399 !skb->csum_complete_sw)
6400 netdev_rx_csum_fault(skb->dev, skb);
6403 NAPI_GRO_CB(skb)->csum = wsum;
6404 NAPI_GRO_CB(skb)->csum_valid = 1;
6408 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6410 static void net_rps_send_ipi(struct softnet_data *remsd)
6414 struct softnet_data *next = remsd->rps_ipi_next;
6416 if (cpu_online(remsd->cpu))
6417 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6424 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6425 * Note: called with local irq disabled, but exits with local irq enabled.
6427 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6430 struct softnet_data *remsd = sd->rps_ipi_list;
6433 sd->rps_ipi_list = NULL;
6437 /* Send pending IPI's to kick RPS processing on remote cpus. */
6438 net_rps_send_ipi(remsd);
6444 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6447 return sd->rps_ipi_list != NULL;
6453 static int process_backlog(struct napi_struct *napi, int quota)
6455 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6459 /* Check if we have pending ipi, its better to send them now,
6460 * not waiting net_rx_action() end.
6462 if (sd_has_rps_ipi_waiting(sd)) {
6463 local_irq_disable();
6464 net_rps_action_and_irq_enable(sd);
6467 napi->weight = READ_ONCE(dev_rx_weight);
6469 struct sk_buff *skb;
6471 while ((skb = __skb_dequeue(&sd->process_queue))) {
6473 __netif_receive_skb(skb);
6475 input_queue_head_incr(sd);
6476 if (++work >= quota)
6481 local_irq_disable();
6483 if (skb_queue_empty(&sd->input_pkt_queue)) {
6485 * Inline a custom version of __napi_complete().
6486 * only current cpu owns and manipulates this napi,
6487 * and NAPI_STATE_SCHED is the only possible flag set
6489 * We can use a plain write instead of clear_bit(),
6490 * and we dont need an smp_mb() memory barrier.
6495 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6496 &sd->process_queue);
6506 * __napi_schedule - schedule for receive
6507 * @n: entry to schedule
6509 * The entry's receive function will be scheduled to run.
6510 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6512 void __napi_schedule(struct napi_struct *n)
6514 unsigned long flags;
6516 local_irq_save(flags);
6517 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6518 local_irq_restore(flags);
6520 EXPORT_SYMBOL(__napi_schedule);
6523 * napi_schedule_prep - check if napi can be scheduled
6526 * Test if NAPI routine is already running, and if not mark
6527 * it as running. This is used as a condition variable to
6528 * insure only one NAPI poll instance runs. We also make
6529 * sure there is no pending NAPI disable.
6531 bool napi_schedule_prep(struct napi_struct *n)
6533 unsigned long val, new;
6536 val = READ_ONCE(n->state);
6537 if (unlikely(val & NAPIF_STATE_DISABLE))
6539 new = val | NAPIF_STATE_SCHED;
6541 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6542 * This was suggested by Alexander Duyck, as compiler
6543 * emits better code than :
6544 * if (val & NAPIF_STATE_SCHED)
6545 * new |= NAPIF_STATE_MISSED;
6547 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6549 } while (cmpxchg(&n->state, val, new) != val);
6551 return !(val & NAPIF_STATE_SCHED);
6553 EXPORT_SYMBOL(napi_schedule_prep);
6556 * __napi_schedule_irqoff - schedule for receive
6557 * @n: entry to schedule
6559 * Variant of __napi_schedule() assuming hard irqs are masked.
6561 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6562 * because the interrupt disabled assumption might not be true
6563 * due to force-threaded interrupts and spinlock substitution.
6565 void __napi_schedule_irqoff(struct napi_struct *n)
6567 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6568 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6572 EXPORT_SYMBOL(__napi_schedule_irqoff);
6574 bool napi_complete_done(struct napi_struct *n, int work_done)
6576 unsigned long flags, val, new, timeout = 0;
6580 * 1) Don't let napi dequeue from the cpu poll list
6581 * just in case its running on a different cpu.
6582 * 2) If we are busy polling, do nothing here, we have
6583 * the guarantee we will be called later.
6585 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6586 NAPIF_STATE_IN_BUSY_POLL)))
6591 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6592 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6594 if (n->defer_hard_irqs_count > 0) {
6595 n->defer_hard_irqs_count--;
6596 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6600 if (n->gro_bitmask) {
6601 /* When the NAPI instance uses a timeout and keeps postponing
6602 * it, we need to bound somehow the time packets are kept in
6605 napi_gro_flush(n, !!timeout);
6610 if (unlikely(!list_empty(&n->poll_list))) {
6611 /* If n->poll_list is not empty, we need to mask irqs */
6612 local_irq_save(flags);
6613 list_del_init(&n->poll_list);
6614 local_irq_restore(flags);
6618 val = READ_ONCE(n->state);
6620 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6622 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6623 NAPIF_STATE_SCHED_THREADED |
6624 NAPIF_STATE_PREFER_BUSY_POLL);
6626 /* If STATE_MISSED was set, leave STATE_SCHED set,
6627 * because we will call napi->poll() one more time.
6628 * This C code was suggested by Alexander Duyck to help gcc.
6630 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6632 } while (cmpxchg(&n->state, val, new) != val);
6634 if (unlikely(val & NAPIF_STATE_MISSED)) {
6640 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6641 HRTIMER_MODE_REL_PINNED);
6644 EXPORT_SYMBOL(napi_complete_done);
6646 /* must be called under rcu_read_lock(), as we dont take a reference */
6647 static struct napi_struct *napi_by_id(unsigned int napi_id)
6649 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6650 struct napi_struct *napi;
6652 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6653 if (napi->napi_id == napi_id)
6659 #if defined(CONFIG_NET_RX_BUSY_POLL)
6661 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6663 if (!skip_schedule) {
6664 gro_normal_list(napi);
6665 __napi_schedule(napi);
6669 if (napi->gro_bitmask) {
6670 /* flush too old packets
6671 * If HZ < 1000, flush all packets.
6673 napi_gro_flush(napi, HZ >= 1000);
6676 gro_normal_list(napi);
6677 clear_bit(NAPI_STATE_SCHED, &napi->state);
6680 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6683 bool skip_schedule = false;
6684 unsigned long timeout;
6687 /* Busy polling means there is a high chance device driver hard irq
6688 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6689 * set in napi_schedule_prep().
6690 * Since we are about to call napi->poll() once more, we can safely
6691 * clear NAPI_STATE_MISSED.
6693 * Note: x86 could use a single "lock and ..." instruction
6694 * to perform these two clear_bit()
6696 clear_bit(NAPI_STATE_MISSED, &napi->state);
6697 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6701 if (prefer_busy_poll) {
6702 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6703 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6704 if (napi->defer_hard_irqs_count && timeout) {
6705 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6706 skip_schedule = true;
6710 /* All we really want here is to re-enable device interrupts.
6711 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6713 rc = napi->poll(napi, budget);
6714 /* We can't gro_normal_list() here, because napi->poll() might have
6715 * rearmed the napi (napi_complete_done()) in which case it could
6716 * already be running on another CPU.
6718 trace_napi_poll(napi, rc, budget);
6719 netpoll_poll_unlock(have_poll_lock);
6721 __busy_poll_stop(napi, skip_schedule);
6725 void napi_busy_loop(unsigned int napi_id,
6726 bool (*loop_end)(void *, unsigned long),
6727 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6729 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6730 int (*napi_poll)(struct napi_struct *napi, int budget);
6731 void *have_poll_lock = NULL;
6732 struct napi_struct *napi;
6739 napi = napi_by_id(napi_id);
6749 unsigned long val = READ_ONCE(napi->state);
6751 /* If multiple threads are competing for this napi,
6752 * we avoid dirtying napi->state as much as we can.
6754 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6755 NAPIF_STATE_IN_BUSY_POLL)) {
6756 if (prefer_busy_poll)
6757 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6760 if (cmpxchg(&napi->state, val,
6761 val | NAPIF_STATE_IN_BUSY_POLL |
6762 NAPIF_STATE_SCHED) != val) {
6763 if (prefer_busy_poll)
6764 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6767 have_poll_lock = netpoll_poll_lock(napi);
6768 napi_poll = napi->poll;
6770 work = napi_poll(napi, budget);
6771 trace_napi_poll(napi, work, budget);
6772 gro_normal_list(napi);
6775 __NET_ADD_STATS(dev_net(napi->dev),
6776 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6779 if (!loop_end || loop_end(loop_end_arg, start_time))
6782 if (unlikely(need_resched())) {
6784 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6788 if (loop_end(loop_end_arg, start_time))
6795 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6800 EXPORT_SYMBOL(napi_busy_loop);
6802 #endif /* CONFIG_NET_RX_BUSY_POLL */
6804 static void napi_hash_add(struct napi_struct *napi)
6806 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6809 spin_lock(&napi_hash_lock);
6811 /* 0..NR_CPUS range is reserved for sender_cpu use */
6813 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6814 napi_gen_id = MIN_NAPI_ID;
6815 } while (napi_by_id(napi_gen_id));
6816 napi->napi_id = napi_gen_id;
6818 hlist_add_head_rcu(&napi->napi_hash_node,
6819 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6821 spin_unlock(&napi_hash_lock);
6824 /* Warning : caller is responsible to make sure rcu grace period
6825 * is respected before freeing memory containing @napi
6827 static void napi_hash_del(struct napi_struct *napi)
6829 spin_lock(&napi_hash_lock);
6831 hlist_del_init_rcu(&napi->napi_hash_node);
6833 spin_unlock(&napi_hash_lock);
6836 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6838 struct napi_struct *napi;
6840 napi = container_of(timer, struct napi_struct, timer);
6842 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6843 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6845 if (!napi_disable_pending(napi) &&
6846 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6847 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6848 __napi_schedule_irqoff(napi);
6851 return HRTIMER_NORESTART;
6854 static void init_gro_hash(struct napi_struct *napi)
6858 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6859 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6860 napi->gro_hash[i].count = 0;
6862 napi->gro_bitmask = 0;
6865 int dev_set_threaded(struct net_device *dev, bool threaded)
6867 struct napi_struct *napi;
6870 if (dev->threaded == threaded)
6874 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6875 if (!napi->thread) {
6876 err = napi_kthread_create(napi);
6885 dev->threaded = threaded;
6887 /* Make sure kthread is created before THREADED bit
6890 smp_mb__before_atomic();
6892 /* Setting/unsetting threaded mode on a napi might not immediately
6893 * take effect, if the current napi instance is actively being
6894 * polled. In this case, the switch between threaded mode and
6895 * softirq mode will happen in the next round of napi_schedule().
6896 * This should not cause hiccups/stalls to the live traffic.
6898 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6900 set_bit(NAPI_STATE_THREADED, &napi->state);
6902 clear_bit(NAPI_STATE_THREADED, &napi->state);
6907 EXPORT_SYMBOL(dev_set_threaded);
6909 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6910 int (*poll)(struct napi_struct *, int), int weight)
6912 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6915 INIT_LIST_HEAD(&napi->poll_list);
6916 INIT_HLIST_NODE(&napi->napi_hash_node);
6917 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6918 napi->timer.function = napi_watchdog;
6919 init_gro_hash(napi);
6921 INIT_LIST_HEAD(&napi->rx_list);
6924 if (weight > NAPI_POLL_WEIGHT)
6925 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6927 napi->weight = weight;
6929 #ifdef CONFIG_NETPOLL
6930 napi->poll_owner = -1;
6932 set_bit(NAPI_STATE_SCHED, &napi->state);
6933 set_bit(NAPI_STATE_NPSVC, &napi->state);
6934 list_add_rcu(&napi->dev_list, &dev->napi_list);
6935 napi_hash_add(napi);
6936 /* Create kthread for this napi if dev->threaded is set.
6937 * Clear dev->threaded if kthread creation failed so that
6938 * threaded mode will not be enabled in napi_enable().
6940 if (dev->threaded && napi_kthread_create(napi))
6943 EXPORT_SYMBOL(netif_napi_add);
6945 void napi_disable(struct napi_struct *n)
6948 set_bit(NAPI_STATE_DISABLE, &n->state);
6950 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6952 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6955 hrtimer_cancel(&n->timer);
6957 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6958 clear_bit(NAPI_STATE_DISABLE, &n->state);
6959 clear_bit(NAPI_STATE_THREADED, &n->state);
6961 EXPORT_SYMBOL(napi_disable);
6964 * napi_enable - enable NAPI scheduling
6967 * Resume NAPI from being scheduled on this context.
6968 * Must be paired with napi_disable.
6970 void napi_enable(struct napi_struct *n)
6972 unsigned long val, new;
6975 val = READ_ONCE(n->state);
6976 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6978 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6979 if (n->dev->threaded && n->thread)
6980 new |= NAPIF_STATE_THREADED;
6981 } while (cmpxchg(&n->state, val, new) != val);
6983 EXPORT_SYMBOL(napi_enable);
6985 static void flush_gro_hash(struct napi_struct *napi)
6989 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6990 struct sk_buff *skb, *n;
6992 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6994 napi->gro_hash[i].count = 0;
6998 /* Must be called in process context */
6999 void __netif_napi_del(struct napi_struct *napi)
7001 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
7004 napi_hash_del(napi);
7005 list_del_rcu(&napi->dev_list);
7006 napi_free_frags(napi);
7008 flush_gro_hash(napi);
7009 napi->gro_bitmask = 0;
7012 kthread_stop(napi->thread);
7013 napi->thread = NULL;
7016 EXPORT_SYMBOL(__netif_napi_del);
7018 static int __napi_poll(struct napi_struct *n, bool *repoll)
7024 /* This NAPI_STATE_SCHED test is for avoiding a race
7025 * with netpoll's poll_napi(). Only the entity which
7026 * obtains the lock and sees NAPI_STATE_SCHED set will
7027 * actually make the ->poll() call. Therefore we avoid
7028 * accidentally calling ->poll() when NAPI is not scheduled.
7031 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
7032 work = n->poll(n, weight);
7033 trace_napi_poll(n, work, weight);
7036 if (unlikely(work > weight))
7037 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7038 n->poll, work, weight);
7040 if (likely(work < weight))
7043 /* Drivers must not modify the NAPI state if they
7044 * consume the entire weight. In such cases this code
7045 * still "owns" the NAPI instance and therefore can
7046 * move the instance around on the list at-will.
7048 if (unlikely(napi_disable_pending(n))) {
7053 /* The NAPI context has more processing work, but busy-polling
7054 * is preferred. Exit early.
7056 if (napi_prefer_busy_poll(n)) {
7057 if (napi_complete_done(n, work)) {
7058 /* If timeout is not set, we need to make sure
7059 * that the NAPI is re-scheduled.
7066 if (n->gro_bitmask) {
7067 /* flush too old packets
7068 * If HZ < 1000, flush all packets.
7070 napi_gro_flush(n, HZ >= 1000);
7075 /* Some drivers may have called napi_schedule
7076 * prior to exhausting their budget.
7078 if (unlikely(!list_empty(&n->poll_list))) {
7079 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7080 n->dev ? n->dev->name : "backlog");
7089 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7091 bool do_repoll = false;
7095 list_del_init(&n->poll_list);
7097 have = netpoll_poll_lock(n);
7099 work = __napi_poll(n, &do_repoll);
7102 list_add_tail(&n->poll_list, repoll);
7104 netpoll_poll_unlock(have);
7109 static int napi_thread_wait(struct napi_struct *napi)
7113 set_current_state(TASK_INTERRUPTIBLE);
7115 while (!kthread_should_stop()) {
7116 /* Testing SCHED_THREADED bit here to make sure the current
7117 * kthread owns this napi and could poll on this napi.
7118 * Testing SCHED bit is not enough because SCHED bit might be
7119 * set by some other busy poll thread or by napi_disable().
7121 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7122 WARN_ON(!list_empty(&napi->poll_list));
7123 __set_current_state(TASK_RUNNING);
7128 /* woken being true indicates this thread owns this napi. */
7130 set_current_state(TASK_INTERRUPTIBLE);
7132 __set_current_state(TASK_RUNNING);
7137 static int napi_threaded_poll(void *data)
7139 struct napi_struct *napi = data;
7142 while (!napi_thread_wait(napi)) {
7144 bool repoll = false;
7148 have = netpoll_poll_lock(napi);
7149 __napi_poll(napi, &repoll);
7150 netpoll_poll_unlock(have);
7163 static __latent_entropy void net_rx_action(struct softirq_action *h)
7165 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7166 unsigned long time_limit = jiffies +
7167 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
7168 int budget = READ_ONCE(netdev_budget);
7172 local_irq_disable();
7173 list_splice_init(&sd->poll_list, &list);
7177 struct napi_struct *n;
7179 if (list_empty(&list)) {
7180 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7185 n = list_first_entry(&list, struct napi_struct, poll_list);
7186 budget -= napi_poll(n, &repoll);
7188 /* If softirq window is exhausted then punt.
7189 * Allow this to run for 2 jiffies since which will allow
7190 * an average latency of 1.5/HZ.
7192 if (unlikely(budget <= 0 ||
7193 time_after_eq(jiffies, time_limit))) {
7199 local_irq_disable();
7201 list_splice_tail_init(&sd->poll_list, &list);
7202 list_splice_tail(&repoll, &list);
7203 list_splice(&list, &sd->poll_list);
7204 if (!list_empty(&sd->poll_list))
7205 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7207 net_rps_action_and_irq_enable(sd);
7210 struct netdev_adjacent {
7211 struct net_device *dev;
7213 /* upper master flag, there can only be one master device per list */
7216 /* lookup ignore flag */
7219 /* counter for the number of times this device was added to us */
7222 /* private field for the users */
7225 struct list_head list;
7226 struct rcu_head rcu;
7229 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7230 struct list_head *adj_list)
7232 struct netdev_adjacent *adj;
7234 list_for_each_entry(adj, adj_list, list) {
7235 if (adj->dev == adj_dev)
7241 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7242 struct netdev_nested_priv *priv)
7244 struct net_device *dev = (struct net_device *)priv->data;
7246 return upper_dev == dev;
7250 * netdev_has_upper_dev - Check if device is linked to an upper device
7252 * @upper_dev: upper device to check
7254 * Find out if a device is linked to specified upper device and return true
7255 * in case it is. Note that this checks only immediate upper device,
7256 * not through a complete stack of devices. The caller must hold the RTNL lock.
7258 bool netdev_has_upper_dev(struct net_device *dev,
7259 struct net_device *upper_dev)
7261 struct netdev_nested_priv priv = {
7262 .data = (void *)upper_dev,
7267 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7270 EXPORT_SYMBOL(netdev_has_upper_dev);
7273 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7275 * @upper_dev: upper device to check
7277 * Find out if a device is linked to specified upper device and return true
7278 * in case it is. Note that this checks the entire upper device chain.
7279 * The caller must hold rcu lock.
7282 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7283 struct net_device *upper_dev)
7285 struct netdev_nested_priv priv = {
7286 .data = (void *)upper_dev,
7289 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7292 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7295 * netdev_has_any_upper_dev - Check if device is linked to some device
7298 * Find out if a device is linked to an upper device and return true in case
7299 * it is. The caller must hold the RTNL lock.
7301 bool netdev_has_any_upper_dev(struct net_device *dev)
7305 return !list_empty(&dev->adj_list.upper);
7307 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7310 * netdev_master_upper_dev_get - Get master upper device
7313 * Find a master upper device and return pointer to it or NULL in case
7314 * it's not there. The caller must hold the RTNL lock.
7316 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7318 struct netdev_adjacent *upper;
7322 if (list_empty(&dev->adj_list.upper))
7325 upper = list_first_entry(&dev->adj_list.upper,
7326 struct netdev_adjacent, list);
7327 if (likely(upper->master))
7331 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7333 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7335 struct netdev_adjacent *upper;
7339 if (list_empty(&dev->adj_list.upper))
7342 upper = list_first_entry(&dev->adj_list.upper,
7343 struct netdev_adjacent, list);
7344 if (likely(upper->master) && !upper->ignore)
7350 * netdev_has_any_lower_dev - Check if device is linked to some device
7353 * Find out if a device is linked to a lower device and return true in case
7354 * it is. The caller must hold the RTNL lock.
7356 static bool netdev_has_any_lower_dev(struct net_device *dev)
7360 return !list_empty(&dev->adj_list.lower);
7363 void *netdev_adjacent_get_private(struct list_head *adj_list)
7365 struct netdev_adjacent *adj;
7367 adj = list_entry(adj_list, struct netdev_adjacent, list);
7369 return adj->private;
7371 EXPORT_SYMBOL(netdev_adjacent_get_private);
7374 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7376 * @iter: list_head ** of the current position
7378 * Gets the next device from the dev's upper list, starting from iter
7379 * position. The caller must hold RCU read lock.
7381 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7382 struct list_head **iter)
7384 struct netdev_adjacent *upper;
7386 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7388 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7390 if (&upper->list == &dev->adj_list.upper)
7393 *iter = &upper->list;
7397 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7399 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7400 struct list_head **iter,
7403 struct netdev_adjacent *upper;
7405 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7407 if (&upper->list == &dev->adj_list.upper)
7410 *iter = &upper->list;
7411 *ignore = upper->ignore;
7416 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7417 struct list_head **iter)
7419 struct netdev_adjacent *upper;
7421 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7423 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7425 if (&upper->list == &dev->adj_list.upper)
7428 *iter = &upper->list;
7433 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7434 int (*fn)(struct net_device *dev,
7435 struct netdev_nested_priv *priv),
7436 struct netdev_nested_priv *priv)
7438 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7439 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7444 iter = &dev->adj_list.upper;
7448 ret = fn(now, priv);
7455 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7462 niter = &udev->adj_list.upper;
7463 dev_stack[cur] = now;
7464 iter_stack[cur++] = iter;
7471 next = dev_stack[--cur];
7472 niter = iter_stack[cur];
7482 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7483 int (*fn)(struct net_device *dev,
7484 struct netdev_nested_priv *priv),
7485 struct netdev_nested_priv *priv)
7487 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7488 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7492 iter = &dev->adj_list.upper;
7496 ret = fn(now, priv);
7503 udev = netdev_next_upper_dev_rcu(now, &iter);
7508 niter = &udev->adj_list.upper;
7509 dev_stack[cur] = now;
7510 iter_stack[cur++] = iter;
7517 next = dev_stack[--cur];
7518 niter = iter_stack[cur];
7527 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7529 static bool __netdev_has_upper_dev(struct net_device *dev,
7530 struct net_device *upper_dev)
7532 struct netdev_nested_priv priv = {
7534 .data = (void *)upper_dev,
7539 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7544 * netdev_lower_get_next_private - Get the next ->private from the
7545 * lower neighbour list
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 either hold the
7551 * RTNL lock or its own locking that guarantees that the neighbour lower
7552 * list will remain unchanged.
7554 void *netdev_lower_get_next_private(struct net_device *dev,
7555 struct list_head **iter)
7557 struct netdev_adjacent *lower;
7559 lower = list_entry(*iter, struct netdev_adjacent, list);
7561 if (&lower->list == &dev->adj_list.lower)
7564 *iter = lower->list.next;
7566 return lower->private;
7568 EXPORT_SYMBOL(netdev_lower_get_next_private);
7571 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7572 * lower neighbour list, RCU
7575 * @iter: list_head ** of the current position
7577 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7578 * list, starting from iter position. The caller must hold RCU read lock.
7580 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7581 struct list_head **iter)
7583 struct netdev_adjacent *lower;
7585 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7587 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7589 if (&lower->list == &dev->adj_list.lower)
7592 *iter = &lower->list;
7594 return lower->private;
7596 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7599 * netdev_lower_get_next - Get the next device from the lower neighbour
7602 * @iter: list_head ** of the current position
7604 * Gets the next netdev_adjacent from the dev's lower neighbour
7605 * list, starting from iter position. The caller must hold RTNL lock or
7606 * its own locking that guarantees that the neighbour lower
7607 * list will remain unchanged.
7609 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7611 struct netdev_adjacent *lower;
7613 lower = list_entry(*iter, struct netdev_adjacent, list);
7615 if (&lower->list == &dev->adj_list.lower)
7618 *iter = lower->list.next;
7622 EXPORT_SYMBOL(netdev_lower_get_next);
7624 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7625 struct list_head **iter)
7627 struct netdev_adjacent *lower;
7629 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7631 if (&lower->list == &dev->adj_list.lower)
7634 *iter = &lower->list;
7639 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7640 struct list_head **iter,
7643 struct netdev_adjacent *lower;
7645 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7647 if (&lower->list == &dev->adj_list.lower)
7650 *iter = &lower->list;
7651 *ignore = lower->ignore;
7656 int netdev_walk_all_lower_dev(struct net_device *dev,
7657 int (*fn)(struct net_device *dev,
7658 struct netdev_nested_priv *priv),
7659 struct netdev_nested_priv *priv)
7661 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7662 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7666 iter = &dev->adj_list.lower;
7670 ret = fn(now, priv);
7677 ldev = netdev_next_lower_dev(now, &iter);
7682 niter = &ldev->adj_list.lower;
7683 dev_stack[cur] = now;
7684 iter_stack[cur++] = iter;
7691 next = dev_stack[--cur];
7692 niter = iter_stack[cur];
7701 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7703 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7704 int (*fn)(struct net_device *dev,
7705 struct netdev_nested_priv *priv),
7706 struct netdev_nested_priv *priv)
7708 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7709 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7714 iter = &dev->adj_list.lower;
7718 ret = fn(now, priv);
7725 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7732 niter = &ldev->adj_list.lower;
7733 dev_stack[cur] = now;
7734 iter_stack[cur++] = iter;
7741 next = dev_stack[--cur];
7742 niter = iter_stack[cur];
7752 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7753 struct list_head **iter)
7755 struct netdev_adjacent *lower;
7757 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7758 if (&lower->list == &dev->adj_list.lower)
7761 *iter = &lower->list;
7765 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7767 static u8 __netdev_upper_depth(struct net_device *dev)
7769 struct net_device *udev;
7770 struct list_head *iter;
7774 for (iter = &dev->adj_list.upper,
7775 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7777 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7780 if (max_depth < udev->upper_level)
7781 max_depth = udev->upper_level;
7787 static u8 __netdev_lower_depth(struct net_device *dev)
7789 struct net_device *ldev;
7790 struct list_head *iter;
7794 for (iter = &dev->adj_list.lower,
7795 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7797 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7800 if (max_depth < ldev->lower_level)
7801 max_depth = ldev->lower_level;
7807 static int __netdev_update_upper_level(struct net_device *dev,
7808 struct netdev_nested_priv *__unused)
7810 dev->upper_level = __netdev_upper_depth(dev) + 1;
7814 static int __netdev_update_lower_level(struct net_device *dev,
7815 struct netdev_nested_priv *priv)
7817 dev->lower_level = __netdev_lower_depth(dev) + 1;
7819 #ifdef CONFIG_LOCKDEP
7823 if (priv->flags & NESTED_SYNC_IMM)
7824 dev->nested_level = dev->lower_level - 1;
7825 if (priv->flags & NESTED_SYNC_TODO)
7826 net_unlink_todo(dev);
7831 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7832 int (*fn)(struct net_device *dev,
7833 struct netdev_nested_priv *priv),
7834 struct netdev_nested_priv *priv)
7836 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7837 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7841 iter = &dev->adj_list.lower;
7845 ret = fn(now, priv);
7852 ldev = netdev_next_lower_dev_rcu(now, &iter);
7857 niter = &ldev->adj_list.lower;
7858 dev_stack[cur] = now;
7859 iter_stack[cur++] = iter;
7866 next = dev_stack[--cur];
7867 niter = iter_stack[cur];
7876 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7879 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7880 * lower neighbour list, RCU
7884 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7885 * list. The caller must hold RCU read lock.
7887 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7889 struct netdev_adjacent *lower;
7891 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7892 struct netdev_adjacent, list);
7894 return lower->private;
7897 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7900 * netdev_master_upper_dev_get_rcu - Get master upper device
7903 * Find a master upper device and return pointer to it or NULL in case
7904 * it's not there. The caller must hold the RCU read lock.
7906 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7908 struct netdev_adjacent *upper;
7910 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7911 struct netdev_adjacent, list);
7912 if (upper && likely(upper->master))
7916 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7918 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7919 struct net_device *adj_dev,
7920 struct list_head *dev_list)
7922 char linkname[IFNAMSIZ+7];
7924 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7925 "upper_%s" : "lower_%s", adj_dev->name);
7926 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7929 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7931 struct list_head *dev_list)
7933 char linkname[IFNAMSIZ+7];
7935 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7936 "upper_%s" : "lower_%s", name);
7937 sysfs_remove_link(&(dev->dev.kobj), linkname);
7940 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7941 struct net_device *adj_dev,
7942 struct list_head *dev_list)
7944 return (dev_list == &dev->adj_list.upper ||
7945 dev_list == &dev->adj_list.lower) &&
7946 net_eq(dev_net(dev), dev_net(adj_dev));
7949 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7950 struct net_device *adj_dev,
7951 struct list_head *dev_list,
7952 void *private, bool master)
7954 struct netdev_adjacent *adj;
7957 adj = __netdev_find_adj(adj_dev, dev_list);
7961 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7962 dev->name, adj_dev->name, adj->ref_nr);
7967 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7972 adj->master = master;
7974 adj->private = private;
7975 adj->ignore = false;
7978 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7979 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7981 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7982 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7987 /* Ensure that master link is always the first item in list. */
7989 ret = sysfs_create_link(&(dev->dev.kobj),
7990 &(adj_dev->dev.kobj), "master");
7992 goto remove_symlinks;
7994 list_add_rcu(&adj->list, dev_list);
7996 list_add_tail_rcu(&adj->list, dev_list);
8002 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8003 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8011 static void __netdev_adjacent_dev_remove(struct net_device *dev,
8012 struct net_device *adj_dev,
8014 struct list_head *dev_list)
8016 struct netdev_adjacent *adj;
8018 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
8019 dev->name, adj_dev->name, ref_nr);
8021 adj = __netdev_find_adj(adj_dev, dev_list);
8024 pr_err("Adjacency does not exist for device %s from %s\n",
8025 dev->name, adj_dev->name);
8030 if (adj->ref_nr > ref_nr) {
8031 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8032 dev->name, adj_dev->name, ref_nr,
8033 adj->ref_nr - ref_nr);
8034 adj->ref_nr -= ref_nr;
8039 sysfs_remove_link(&(dev->dev.kobj), "master");
8041 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8042 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8044 list_del_rcu(&adj->list);
8045 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8046 adj_dev->name, dev->name, adj_dev->name);
8048 kfree_rcu(adj, rcu);
8051 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8052 struct net_device *upper_dev,
8053 struct list_head *up_list,
8054 struct list_head *down_list,
8055 void *private, bool master)
8059 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8064 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8067 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8074 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8075 struct net_device *upper_dev,
8077 struct list_head *up_list,
8078 struct list_head *down_list)
8080 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8081 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8084 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8085 struct net_device *upper_dev,
8086 void *private, bool master)
8088 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8089 &dev->adj_list.upper,
8090 &upper_dev->adj_list.lower,
8094 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8095 struct net_device *upper_dev)
8097 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8098 &dev->adj_list.upper,
8099 &upper_dev->adj_list.lower);
8102 static int __netdev_upper_dev_link(struct net_device *dev,
8103 struct net_device *upper_dev, bool master,
8104 void *upper_priv, void *upper_info,
8105 struct netdev_nested_priv *priv,
8106 struct netlink_ext_ack *extack)
8108 struct netdev_notifier_changeupper_info changeupper_info = {
8113 .upper_dev = upper_dev,
8116 .upper_info = upper_info,
8118 struct net_device *master_dev;
8123 if (dev == upper_dev)
8126 /* To prevent loops, check if dev is not upper device to upper_dev. */
8127 if (__netdev_has_upper_dev(upper_dev, dev))
8130 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8134 if (__netdev_has_upper_dev(dev, upper_dev))
8137 master_dev = __netdev_master_upper_dev_get(dev);
8139 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8142 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8143 &changeupper_info.info);
8144 ret = notifier_to_errno(ret);
8148 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8153 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8154 &changeupper_info.info);
8155 ret = notifier_to_errno(ret);
8159 __netdev_update_upper_level(dev, NULL);
8160 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8162 __netdev_update_lower_level(upper_dev, priv);
8163 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8169 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8175 * netdev_upper_dev_link - Add a link to the upper device
8177 * @upper_dev: new upper device
8178 * @extack: netlink extended ack
8180 * Adds a link to device which is upper to this one. The caller must hold
8181 * the RTNL lock. On a failure a negative errno code is returned.
8182 * On success the reference counts are adjusted and the function
8185 int netdev_upper_dev_link(struct net_device *dev,
8186 struct net_device *upper_dev,
8187 struct netlink_ext_ack *extack)
8189 struct netdev_nested_priv priv = {
8190 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8194 return __netdev_upper_dev_link(dev, upper_dev, false,
8195 NULL, NULL, &priv, extack);
8197 EXPORT_SYMBOL(netdev_upper_dev_link);
8200 * netdev_master_upper_dev_link - Add a master link to the upper device
8202 * @upper_dev: new upper device
8203 * @upper_priv: upper device private
8204 * @upper_info: upper info to be passed down via notifier
8205 * @extack: netlink extended ack
8207 * Adds a link to device which is upper to this one. In this case, only
8208 * one master upper device can be linked, although other non-master devices
8209 * might be linked as well. The caller must hold the RTNL lock.
8210 * On a failure a negative errno code is returned. On success the reference
8211 * counts are adjusted and the function returns zero.
8213 int netdev_master_upper_dev_link(struct net_device *dev,
8214 struct net_device *upper_dev,
8215 void *upper_priv, void *upper_info,
8216 struct netlink_ext_ack *extack)
8218 struct netdev_nested_priv priv = {
8219 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8223 return __netdev_upper_dev_link(dev, upper_dev, true,
8224 upper_priv, upper_info, &priv, extack);
8226 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8228 static void __netdev_upper_dev_unlink(struct net_device *dev,
8229 struct net_device *upper_dev,
8230 struct netdev_nested_priv *priv)
8232 struct netdev_notifier_changeupper_info changeupper_info = {
8236 .upper_dev = upper_dev,
8242 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8244 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8245 &changeupper_info.info);
8247 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8249 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8250 &changeupper_info.info);
8252 __netdev_update_upper_level(dev, NULL);
8253 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8255 __netdev_update_lower_level(upper_dev, priv);
8256 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8261 * netdev_upper_dev_unlink - Removes a link to upper device
8263 * @upper_dev: new upper device
8265 * Removes a link to device which is upper to this one. The caller must hold
8268 void netdev_upper_dev_unlink(struct net_device *dev,
8269 struct net_device *upper_dev)
8271 struct netdev_nested_priv priv = {
8272 .flags = NESTED_SYNC_TODO,
8276 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8278 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8280 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8281 struct net_device *lower_dev,
8284 struct netdev_adjacent *adj;
8286 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8290 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8295 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8296 struct net_device *lower_dev)
8298 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8301 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8302 struct net_device *lower_dev)
8304 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8307 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8308 struct net_device *new_dev,
8309 struct net_device *dev,
8310 struct netlink_ext_ack *extack)
8312 struct netdev_nested_priv priv = {
8321 if (old_dev && new_dev != old_dev)
8322 netdev_adjacent_dev_disable(dev, old_dev);
8323 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8326 if (old_dev && new_dev != old_dev)
8327 netdev_adjacent_dev_enable(dev, old_dev);
8333 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8335 void netdev_adjacent_change_commit(struct net_device *old_dev,
8336 struct net_device *new_dev,
8337 struct net_device *dev)
8339 struct netdev_nested_priv priv = {
8340 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8344 if (!new_dev || !old_dev)
8347 if (new_dev == old_dev)
8350 netdev_adjacent_dev_enable(dev, old_dev);
8351 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8353 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8355 void netdev_adjacent_change_abort(struct net_device *old_dev,
8356 struct net_device *new_dev,
8357 struct net_device *dev)
8359 struct netdev_nested_priv priv = {
8367 if (old_dev && new_dev != old_dev)
8368 netdev_adjacent_dev_enable(dev, old_dev);
8370 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8372 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8375 * netdev_bonding_info_change - Dispatch event about slave change
8377 * @bonding_info: info to dispatch
8379 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8380 * The caller must hold the RTNL lock.
8382 void netdev_bonding_info_change(struct net_device *dev,
8383 struct netdev_bonding_info *bonding_info)
8385 struct netdev_notifier_bonding_info info = {
8389 memcpy(&info.bonding_info, bonding_info,
8390 sizeof(struct netdev_bonding_info));
8391 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8394 EXPORT_SYMBOL(netdev_bonding_info_change);
8397 * netdev_get_xmit_slave - Get the xmit slave of master device
8400 * @all_slaves: assume all the slaves are active
8402 * The reference counters are not incremented so the caller must be
8403 * careful with locks. The caller must hold RCU lock.
8404 * %NULL is returned if no slave is found.
8407 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8408 struct sk_buff *skb,
8411 const struct net_device_ops *ops = dev->netdev_ops;
8413 if (!ops->ndo_get_xmit_slave)
8415 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8417 EXPORT_SYMBOL(netdev_get_xmit_slave);
8419 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8422 const struct net_device_ops *ops = dev->netdev_ops;
8424 if (!ops->ndo_sk_get_lower_dev)
8426 return ops->ndo_sk_get_lower_dev(dev, sk);
8430 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8434 * %NULL is returned if no lower device is found.
8437 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8440 struct net_device *lower;
8442 lower = netdev_sk_get_lower_dev(dev, sk);
8445 lower = netdev_sk_get_lower_dev(dev, sk);
8450 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8452 static void netdev_adjacent_add_links(struct net_device *dev)
8454 struct netdev_adjacent *iter;
8456 struct net *net = dev_net(dev);
8458 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8459 if (!net_eq(net, dev_net(iter->dev)))
8461 netdev_adjacent_sysfs_add(iter->dev, dev,
8462 &iter->dev->adj_list.lower);
8463 netdev_adjacent_sysfs_add(dev, iter->dev,
8464 &dev->adj_list.upper);
8467 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8468 if (!net_eq(net, dev_net(iter->dev)))
8470 netdev_adjacent_sysfs_add(iter->dev, dev,
8471 &iter->dev->adj_list.upper);
8472 netdev_adjacent_sysfs_add(dev, iter->dev,
8473 &dev->adj_list.lower);
8477 static void netdev_adjacent_del_links(struct net_device *dev)
8479 struct netdev_adjacent *iter;
8481 struct net *net = dev_net(dev);
8483 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8484 if (!net_eq(net, dev_net(iter->dev)))
8486 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8487 &iter->dev->adj_list.lower);
8488 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8489 &dev->adj_list.upper);
8492 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8493 if (!net_eq(net, dev_net(iter->dev)))
8495 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8496 &iter->dev->adj_list.upper);
8497 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8498 &dev->adj_list.lower);
8502 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8504 struct netdev_adjacent *iter;
8506 struct net *net = dev_net(dev);
8508 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8509 if (!net_eq(net, dev_net(iter->dev)))
8511 netdev_adjacent_sysfs_del(iter->dev, oldname,
8512 &iter->dev->adj_list.lower);
8513 netdev_adjacent_sysfs_add(iter->dev, dev,
8514 &iter->dev->adj_list.lower);
8517 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8518 if (!net_eq(net, dev_net(iter->dev)))
8520 netdev_adjacent_sysfs_del(iter->dev, oldname,
8521 &iter->dev->adj_list.upper);
8522 netdev_adjacent_sysfs_add(iter->dev, dev,
8523 &iter->dev->adj_list.upper);
8527 void *netdev_lower_dev_get_private(struct net_device *dev,
8528 struct net_device *lower_dev)
8530 struct netdev_adjacent *lower;
8534 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8538 return lower->private;
8540 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8544 * netdev_lower_state_changed - Dispatch event about lower device state change
8545 * @lower_dev: device
8546 * @lower_state_info: state to dispatch
8548 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8549 * The caller must hold the RTNL lock.
8551 void netdev_lower_state_changed(struct net_device *lower_dev,
8552 void *lower_state_info)
8554 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8555 .info.dev = lower_dev,
8559 changelowerstate_info.lower_state_info = lower_state_info;
8560 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8561 &changelowerstate_info.info);
8563 EXPORT_SYMBOL(netdev_lower_state_changed);
8565 static void dev_change_rx_flags(struct net_device *dev, int flags)
8567 const struct net_device_ops *ops = dev->netdev_ops;
8569 if (ops->ndo_change_rx_flags)
8570 ops->ndo_change_rx_flags(dev, flags);
8573 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8575 unsigned int old_flags = dev->flags;
8581 dev->flags |= IFF_PROMISC;
8582 dev->promiscuity += inc;
8583 if (dev->promiscuity == 0) {
8586 * If inc causes overflow, untouch promisc and return error.
8589 dev->flags &= ~IFF_PROMISC;
8591 dev->promiscuity -= inc;
8592 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8597 if (dev->flags != old_flags) {
8598 pr_info("device %s %s promiscuous mode\n",
8600 dev->flags & IFF_PROMISC ? "entered" : "left");
8601 if (audit_enabled) {
8602 current_uid_gid(&uid, &gid);
8603 audit_log(audit_context(), GFP_ATOMIC,
8604 AUDIT_ANOM_PROMISCUOUS,
8605 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8606 dev->name, (dev->flags & IFF_PROMISC),
8607 (old_flags & IFF_PROMISC),
8608 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8609 from_kuid(&init_user_ns, uid),
8610 from_kgid(&init_user_ns, gid),
8611 audit_get_sessionid(current));
8614 dev_change_rx_flags(dev, IFF_PROMISC);
8617 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8622 * dev_set_promiscuity - update promiscuity count on a device
8626 * Add or remove promiscuity from a device. While the count in the device
8627 * remains above zero the interface remains promiscuous. Once it hits zero
8628 * the device reverts back to normal filtering operation. A negative inc
8629 * value is used to drop promiscuity on the device.
8630 * Return 0 if successful or a negative errno code on error.
8632 int dev_set_promiscuity(struct net_device *dev, int inc)
8634 unsigned int old_flags = dev->flags;
8637 err = __dev_set_promiscuity(dev, inc, true);
8640 if (dev->flags != old_flags)
8641 dev_set_rx_mode(dev);
8644 EXPORT_SYMBOL(dev_set_promiscuity);
8646 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8648 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8652 dev->flags |= IFF_ALLMULTI;
8653 dev->allmulti += inc;
8654 if (dev->allmulti == 0) {
8657 * If inc causes overflow, untouch allmulti and return error.
8660 dev->flags &= ~IFF_ALLMULTI;
8662 dev->allmulti -= inc;
8663 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8668 if (dev->flags ^ old_flags) {
8669 dev_change_rx_flags(dev, IFF_ALLMULTI);
8670 dev_set_rx_mode(dev);
8672 __dev_notify_flags(dev, old_flags,
8673 dev->gflags ^ old_gflags);
8679 * dev_set_allmulti - update allmulti count on a device
8683 * Add or remove reception of all multicast frames to a device. While the
8684 * count in the device remains above zero the interface remains listening
8685 * to all interfaces. Once it hits zero the device reverts back to normal
8686 * filtering operation. A negative @inc value is used to drop the counter
8687 * when releasing a resource needing all multicasts.
8688 * Return 0 if successful or a negative errno code on error.
8691 int dev_set_allmulti(struct net_device *dev, int inc)
8693 return __dev_set_allmulti(dev, inc, true);
8695 EXPORT_SYMBOL(dev_set_allmulti);
8698 * Upload unicast and multicast address lists to device and
8699 * configure RX filtering. When the device doesn't support unicast
8700 * filtering it is put in promiscuous mode while unicast addresses
8703 void __dev_set_rx_mode(struct net_device *dev)
8705 const struct net_device_ops *ops = dev->netdev_ops;
8707 /* dev_open will call this function so the list will stay sane. */
8708 if (!(dev->flags&IFF_UP))
8711 if (!netif_device_present(dev))
8714 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8715 /* Unicast addresses changes may only happen under the rtnl,
8716 * therefore calling __dev_set_promiscuity here is safe.
8718 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8719 __dev_set_promiscuity(dev, 1, false);
8720 dev->uc_promisc = true;
8721 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8722 __dev_set_promiscuity(dev, -1, false);
8723 dev->uc_promisc = false;
8727 if (ops->ndo_set_rx_mode)
8728 ops->ndo_set_rx_mode(dev);
8731 void dev_set_rx_mode(struct net_device *dev)
8733 netif_addr_lock_bh(dev);
8734 __dev_set_rx_mode(dev);
8735 netif_addr_unlock_bh(dev);
8739 * dev_get_flags - get flags reported to userspace
8742 * Get the combination of flag bits exported through APIs to userspace.
8744 unsigned int dev_get_flags(const struct net_device *dev)
8748 flags = (dev->flags & ~(IFF_PROMISC |
8753 (dev->gflags & (IFF_PROMISC |
8756 if (netif_running(dev)) {
8757 if (netif_oper_up(dev))
8758 flags |= IFF_RUNNING;
8759 if (netif_carrier_ok(dev))
8760 flags |= IFF_LOWER_UP;
8761 if (netif_dormant(dev))
8762 flags |= IFF_DORMANT;
8767 EXPORT_SYMBOL(dev_get_flags);
8769 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8770 struct netlink_ext_ack *extack)
8772 unsigned int old_flags = dev->flags;
8778 * Set the flags on our device.
8781 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8782 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8784 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8788 * Load in the correct multicast list now the flags have changed.
8791 if ((old_flags ^ flags) & IFF_MULTICAST)
8792 dev_change_rx_flags(dev, IFF_MULTICAST);
8794 dev_set_rx_mode(dev);
8797 * Have we downed the interface. We handle IFF_UP ourselves
8798 * according to user attempts to set it, rather than blindly
8803 if ((old_flags ^ flags) & IFF_UP) {
8804 if (old_flags & IFF_UP)
8807 ret = __dev_open(dev, extack);
8810 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8811 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8812 unsigned int old_flags = dev->flags;
8814 dev->gflags ^= IFF_PROMISC;
8816 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8817 if (dev->flags != old_flags)
8818 dev_set_rx_mode(dev);
8821 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8822 * is important. Some (broken) drivers set IFF_PROMISC, when
8823 * IFF_ALLMULTI is requested not asking us and not reporting.
8825 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8826 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8828 dev->gflags ^= IFF_ALLMULTI;
8829 __dev_set_allmulti(dev, inc, false);
8835 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8836 unsigned int gchanges)
8838 unsigned int changes = dev->flags ^ old_flags;
8841 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8843 if (changes & IFF_UP) {
8844 if (dev->flags & IFF_UP)
8845 call_netdevice_notifiers(NETDEV_UP, dev);
8847 call_netdevice_notifiers(NETDEV_DOWN, dev);
8850 if (dev->flags & IFF_UP &&
8851 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8852 struct netdev_notifier_change_info change_info = {
8856 .flags_changed = changes,
8859 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8864 * dev_change_flags - change device settings
8866 * @flags: device state flags
8867 * @extack: netlink extended ack
8869 * Change settings on device based state flags. The flags are
8870 * in the userspace exported format.
8872 int dev_change_flags(struct net_device *dev, unsigned int flags,
8873 struct netlink_ext_ack *extack)
8876 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8878 ret = __dev_change_flags(dev, flags, extack);
8882 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8883 __dev_notify_flags(dev, old_flags, changes);
8886 EXPORT_SYMBOL(dev_change_flags);
8888 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8890 const struct net_device_ops *ops = dev->netdev_ops;
8892 if (ops->ndo_change_mtu)
8893 return ops->ndo_change_mtu(dev, new_mtu);
8895 /* Pairs with all the lockless reads of dev->mtu in the stack */
8896 WRITE_ONCE(dev->mtu, new_mtu);
8899 EXPORT_SYMBOL(__dev_set_mtu);
8901 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8902 struct netlink_ext_ack *extack)
8904 /* MTU must be positive, and in range */
8905 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8906 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8910 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8911 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8918 * dev_set_mtu_ext - Change maximum transfer unit
8920 * @new_mtu: new transfer unit
8921 * @extack: netlink extended ack
8923 * Change the maximum transfer size of the network device.
8925 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8926 struct netlink_ext_ack *extack)
8930 if (new_mtu == dev->mtu)
8933 err = dev_validate_mtu(dev, new_mtu, extack);
8937 if (!netif_device_present(dev))
8940 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8941 err = notifier_to_errno(err);
8945 orig_mtu = dev->mtu;
8946 err = __dev_set_mtu(dev, new_mtu);
8949 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8951 err = notifier_to_errno(err);
8953 /* setting mtu back and notifying everyone again,
8954 * so that they have a chance to revert changes.
8956 __dev_set_mtu(dev, orig_mtu);
8957 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8964 int dev_set_mtu(struct net_device *dev, int new_mtu)
8966 struct netlink_ext_ack extack;
8969 memset(&extack, 0, sizeof(extack));
8970 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8971 if (err && extack._msg)
8972 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8975 EXPORT_SYMBOL(dev_set_mtu);
8978 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8980 * @new_len: new tx queue length
8982 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8984 unsigned int orig_len = dev->tx_queue_len;
8987 if (new_len != (unsigned int)new_len)
8990 if (new_len != orig_len) {
8991 dev->tx_queue_len = new_len;
8992 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8993 res = notifier_to_errno(res);
8996 res = dev_qdisc_change_tx_queue_len(dev);
9004 netdev_err(dev, "refused to change device tx_queue_len\n");
9005 dev->tx_queue_len = orig_len;
9010 * dev_set_group - Change group this device belongs to
9012 * @new_group: group this device should belong to
9014 void dev_set_group(struct net_device *dev, int new_group)
9016 dev->group = new_group;
9018 EXPORT_SYMBOL(dev_set_group);
9021 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9023 * @addr: new address
9024 * @extack: netlink extended ack
9026 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9027 struct netlink_ext_ack *extack)
9029 struct netdev_notifier_pre_changeaddr_info info = {
9031 .info.extack = extack,
9036 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9037 return notifier_to_errno(rc);
9039 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9042 * dev_set_mac_address - Change Media Access Control Address
9045 * @extack: netlink extended ack
9047 * Change the hardware (MAC) address of the device
9049 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9050 struct netlink_ext_ack *extack)
9052 const struct net_device_ops *ops = dev->netdev_ops;
9055 if (!ops->ndo_set_mac_address)
9057 if (sa->sa_family != dev->type)
9059 if (!netif_device_present(dev))
9061 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9064 err = ops->ndo_set_mac_address(dev, sa);
9067 dev->addr_assign_type = NET_ADDR_SET;
9068 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9069 add_device_randomness(dev->dev_addr, dev->addr_len);
9072 EXPORT_SYMBOL(dev_set_mac_address);
9074 static DECLARE_RWSEM(dev_addr_sem);
9076 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9077 struct netlink_ext_ack *extack)
9081 down_write(&dev_addr_sem);
9082 ret = dev_set_mac_address(dev, sa, extack);
9083 up_write(&dev_addr_sem);
9086 EXPORT_SYMBOL(dev_set_mac_address_user);
9088 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9090 size_t size = sizeof(sa->sa_data);
9091 struct net_device *dev;
9094 down_read(&dev_addr_sem);
9097 dev = dev_get_by_name_rcu(net, dev_name);
9103 memset(sa->sa_data, 0, size);
9105 memcpy(sa->sa_data, dev->dev_addr,
9106 min_t(size_t, size, dev->addr_len));
9107 sa->sa_family = dev->type;
9111 up_read(&dev_addr_sem);
9114 EXPORT_SYMBOL(dev_get_mac_address);
9117 * dev_change_carrier - Change device carrier
9119 * @new_carrier: new value
9121 * Change device carrier
9123 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9125 const struct net_device_ops *ops = dev->netdev_ops;
9127 if (!ops->ndo_change_carrier)
9129 if (!netif_device_present(dev))
9131 return ops->ndo_change_carrier(dev, new_carrier);
9133 EXPORT_SYMBOL(dev_change_carrier);
9136 * dev_get_phys_port_id - Get device physical port ID
9140 * Get device physical port ID
9142 int dev_get_phys_port_id(struct net_device *dev,
9143 struct netdev_phys_item_id *ppid)
9145 const struct net_device_ops *ops = dev->netdev_ops;
9147 if (!ops->ndo_get_phys_port_id)
9149 return ops->ndo_get_phys_port_id(dev, ppid);
9151 EXPORT_SYMBOL(dev_get_phys_port_id);
9154 * dev_get_phys_port_name - Get device physical port name
9157 * @len: limit of bytes to copy to name
9159 * Get device physical port name
9161 int dev_get_phys_port_name(struct net_device *dev,
9162 char *name, size_t len)
9164 const struct net_device_ops *ops = dev->netdev_ops;
9167 if (ops->ndo_get_phys_port_name) {
9168 err = ops->ndo_get_phys_port_name(dev, name, len);
9169 if (err != -EOPNOTSUPP)
9172 return devlink_compat_phys_port_name_get(dev, name, len);
9174 EXPORT_SYMBOL(dev_get_phys_port_name);
9177 * dev_get_port_parent_id - Get the device's port parent identifier
9178 * @dev: network device
9179 * @ppid: pointer to a storage for the port's parent identifier
9180 * @recurse: allow/disallow recursion to lower devices
9182 * Get the devices's port parent identifier
9184 int dev_get_port_parent_id(struct net_device *dev,
9185 struct netdev_phys_item_id *ppid,
9188 const struct net_device_ops *ops = dev->netdev_ops;
9189 struct netdev_phys_item_id first = { };
9190 struct net_device *lower_dev;
9191 struct list_head *iter;
9194 if (ops->ndo_get_port_parent_id) {
9195 err = ops->ndo_get_port_parent_id(dev, ppid);
9196 if (err != -EOPNOTSUPP)
9200 err = devlink_compat_switch_id_get(dev, ppid);
9201 if (!err || err != -EOPNOTSUPP)
9207 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9208 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9213 else if (memcmp(&first, ppid, sizeof(*ppid)))
9219 EXPORT_SYMBOL(dev_get_port_parent_id);
9222 * netdev_port_same_parent_id - Indicate if two network devices have
9223 * the same port parent identifier
9224 * @a: first network device
9225 * @b: second network device
9227 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9229 struct netdev_phys_item_id a_id = { };
9230 struct netdev_phys_item_id b_id = { };
9232 if (dev_get_port_parent_id(a, &a_id, true) ||
9233 dev_get_port_parent_id(b, &b_id, true))
9236 return netdev_phys_item_id_same(&a_id, &b_id);
9238 EXPORT_SYMBOL(netdev_port_same_parent_id);
9241 * dev_change_proto_down - update protocol port state information
9243 * @proto_down: new value
9245 * This info can be used by switch drivers to set the phys state of the
9248 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9250 const struct net_device_ops *ops = dev->netdev_ops;
9252 if (!ops->ndo_change_proto_down)
9254 if (!netif_device_present(dev))
9256 return ops->ndo_change_proto_down(dev, proto_down);
9258 EXPORT_SYMBOL(dev_change_proto_down);
9261 * dev_change_proto_down_generic - generic implementation for
9262 * ndo_change_proto_down that sets carrier according to
9266 * @proto_down: new value
9268 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9271 netif_carrier_off(dev);
9273 netif_carrier_on(dev);
9274 dev->proto_down = proto_down;
9277 EXPORT_SYMBOL(dev_change_proto_down_generic);
9280 * dev_change_proto_down_reason - proto down reason
9283 * @mask: proto down mask
9284 * @value: proto down value
9286 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9292 dev->proto_down_reason = value;
9294 for_each_set_bit(b, &mask, 32) {
9295 if (value & (1 << b))
9296 dev->proto_down_reason |= BIT(b);
9298 dev->proto_down_reason &= ~BIT(b);
9302 EXPORT_SYMBOL(dev_change_proto_down_reason);
9304 struct bpf_xdp_link {
9305 struct bpf_link link;
9306 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9310 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9312 if (flags & XDP_FLAGS_HW_MODE)
9314 if (flags & XDP_FLAGS_DRV_MODE)
9315 return XDP_MODE_DRV;
9316 if (flags & XDP_FLAGS_SKB_MODE)
9317 return XDP_MODE_SKB;
9318 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9321 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9325 return generic_xdp_install;
9328 return dev->netdev_ops->ndo_bpf;
9334 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9335 enum bpf_xdp_mode mode)
9337 return dev->xdp_state[mode].link;
9340 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9341 enum bpf_xdp_mode mode)
9343 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9346 return link->link.prog;
9347 return dev->xdp_state[mode].prog;
9350 u8 dev_xdp_prog_count(struct net_device *dev)
9355 for (i = 0; i < __MAX_XDP_MODE; i++)
9356 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9360 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9362 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9364 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9366 return prog ? prog->aux->id : 0;
9369 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9370 struct bpf_xdp_link *link)
9372 dev->xdp_state[mode].link = link;
9373 dev->xdp_state[mode].prog = NULL;
9376 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9377 struct bpf_prog *prog)
9379 dev->xdp_state[mode].link = NULL;
9380 dev->xdp_state[mode].prog = prog;
9383 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9384 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9385 u32 flags, struct bpf_prog *prog)
9387 struct netdev_bpf xdp;
9390 memset(&xdp, 0, sizeof(xdp));
9391 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9392 xdp.extack = extack;
9396 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9397 * "moved" into driver), so they don't increment it on their own, but
9398 * they do decrement refcnt when program is detached or replaced.
9399 * Given net_device also owns link/prog, we need to bump refcnt here
9400 * to prevent drivers from underflowing it.
9404 err = bpf_op(dev, &xdp);
9411 if (mode != XDP_MODE_HW)
9412 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9417 static void dev_xdp_uninstall(struct net_device *dev)
9419 struct bpf_xdp_link *link;
9420 struct bpf_prog *prog;
9421 enum bpf_xdp_mode mode;
9426 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9427 prog = dev_xdp_prog(dev, mode);
9431 bpf_op = dev_xdp_bpf_op(dev, mode);
9435 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9437 /* auto-detach link from net device */
9438 link = dev_xdp_link(dev, mode);
9444 dev_xdp_set_link(dev, mode, NULL);
9448 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9449 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9450 struct bpf_prog *old_prog, u32 flags)
9452 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9453 struct bpf_prog *cur_prog;
9454 struct net_device *upper;
9455 struct list_head *iter;
9456 enum bpf_xdp_mode mode;
9462 /* either link or prog attachment, never both */
9463 if (link && (new_prog || old_prog))
9465 /* link supports only XDP mode flags */
9466 if (link && (flags & ~XDP_FLAGS_MODES)) {
9467 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9470 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9471 if (num_modes > 1) {
9472 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9475 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9476 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9477 NL_SET_ERR_MSG(extack,
9478 "More than one program loaded, unset mode is ambiguous");
9481 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9482 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9483 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9487 mode = dev_xdp_mode(dev, flags);
9488 /* can't replace attached link */
9489 if (dev_xdp_link(dev, mode)) {
9490 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9494 /* don't allow if an upper device already has a program */
9495 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9496 if (dev_xdp_prog_count(upper) > 0) {
9497 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9502 cur_prog = dev_xdp_prog(dev, mode);
9503 /* can't replace attached prog with link */
9504 if (link && cur_prog) {
9505 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9508 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9509 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9513 /* put effective new program into new_prog */
9515 new_prog = link->link.prog;
9518 bool offload = mode == XDP_MODE_HW;
9519 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9520 ? XDP_MODE_DRV : XDP_MODE_SKB;
9522 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9523 NL_SET_ERR_MSG(extack, "XDP program already attached");
9526 if (!offload && dev_xdp_prog(dev, other_mode)) {
9527 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9530 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9531 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9534 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9535 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9538 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9539 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9544 /* don't call drivers if the effective program didn't change */
9545 if (new_prog != cur_prog) {
9546 bpf_op = dev_xdp_bpf_op(dev, mode);
9548 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9552 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9558 dev_xdp_set_link(dev, mode, link);
9560 dev_xdp_set_prog(dev, mode, new_prog);
9562 bpf_prog_put(cur_prog);
9567 static int dev_xdp_attach_link(struct net_device *dev,
9568 struct netlink_ext_ack *extack,
9569 struct bpf_xdp_link *link)
9571 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9574 static int dev_xdp_detach_link(struct net_device *dev,
9575 struct netlink_ext_ack *extack,
9576 struct bpf_xdp_link *link)
9578 enum bpf_xdp_mode mode;
9583 mode = dev_xdp_mode(dev, link->flags);
9584 if (dev_xdp_link(dev, mode) != link)
9587 bpf_op = dev_xdp_bpf_op(dev, mode);
9588 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9589 dev_xdp_set_link(dev, mode, NULL);
9593 static void bpf_xdp_link_release(struct bpf_link *link)
9595 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9599 /* if racing with net_device's tear down, xdp_link->dev might be
9600 * already NULL, in which case link was already auto-detached
9602 if (xdp_link->dev) {
9603 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9604 xdp_link->dev = NULL;
9610 static int bpf_xdp_link_detach(struct bpf_link *link)
9612 bpf_xdp_link_release(link);
9616 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9618 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9623 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9624 struct seq_file *seq)
9626 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9631 ifindex = xdp_link->dev->ifindex;
9634 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9637 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9638 struct bpf_link_info *info)
9640 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9645 ifindex = xdp_link->dev->ifindex;
9648 info->xdp.ifindex = ifindex;
9652 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9653 struct bpf_prog *old_prog)
9655 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9656 enum bpf_xdp_mode mode;
9662 /* link might have been auto-released already, so fail */
9663 if (!xdp_link->dev) {
9668 if (old_prog && link->prog != old_prog) {
9672 old_prog = link->prog;
9673 if (old_prog->type != new_prog->type ||
9674 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9679 if (old_prog == new_prog) {
9680 /* no-op, don't disturb drivers */
9681 bpf_prog_put(new_prog);
9685 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9686 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9687 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9688 xdp_link->flags, new_prog);
9692 old_prog = xchg(&link->prog, new_prog);
9693 bpf_prog_put(old_prog);
9700 static const struct bpf_link_ops bpf_xdp_link_lops = {
9701 .release = bpf_xdp_link_release,
9702 .dealloc = bpf_xdp_link_dealloc,
9703 .detach = bpf_xdp_link_detach,
9704 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9705 .fill_link_info = bpf_xdp_link_fill_link_info,
9706 .update_prog = bpf_xdp_link_update,
9709 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9711 struct net *net = current->nsproxy->net_ns;
9712 struct bpf_link_primer link_primer;
9713 struct bpf_xdp_link *link;
9714 struct net_device *dev;
9718 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9724 link = kzalloc(sizeof(*link), GFP_USER);
9730 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9732 link->flags = attr->link_create.flags;
9734 err = bpf_link_prime(&link->link, &link_primer);
9740 err = dev_xdp_attach_link(dev, NULL, link);
9745 bpf_link_cleanup(&link_primer);
9749 fd = bpf_link_settle(&link_primer);
9750 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9763 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9765 * @extack: netlink extended ack
9766 * @fd: new program fd or negative value to clear
9767 * @expected_fd: old program fd that userspace expects to replace or clear
9768 * @flags: xdp-related flags
9770 * Set or clear a bpf program for a device
9772 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9773 int fd, int expected_fd, u32 flags)
9775 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9776 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9782 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9783 mode != XDP_MODE_SKB);
9784 if (IS_ERR(new_prog))
9785 return PTR_ERR(new_prog);
9788 if (expected_fd >= 0) {
9789 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9790 mode != XDP_MODE_SKB);
9791 if (IS_ERR(old_prog)) {
9792 err = PTR_ERR(old_prog);
9798 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9801 if (err && new_prog)
9802 bpf_prog_put(new_prog);
9804 bpf_prog_put(old_prog);
9809 * dev_new_index - allocate an ifindex
9810 * @net: the applicable net namespace
9812 * Returns a suitable unique value for a new device interface
9813 * number. The caller must hold the rtnl semaphore or the
9814 * dev_base_lock to be sure it remains unique.
9816 static int dev_new_index(struct net *net)
9818 int ifindex = net->ifindex;
9823 if (!__dev_get_by_index(net, ifindex))
9824 return net->ifindex = ifindex;
9828 /* Delayed registration/unregisteration */
9829 static LIST_HEAD(net_todo_list);
9830 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9832 static void net_set_todo(struct net_device *dev)
9834 list_add_tail(&dev->todo_list, &net_todo_list);
9835 dev_net(dev)->dev_unreg_count++;
9838 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9839 struct net_device *upper, netdev_features_t features)
9841 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9842 netdev_features_t feature;
9845 for_each_netdev_feature(upper_disables, feature_bit) {
9846 feature = __NETIF_F_BIT(feature_bit);
9847 if (!(upper->wanted_features & feature)
9848 && (features & feature)) {
9849 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9850 &feature, upper->name);
9851 features &= ~feature;
9858 static void netdev_sync_lower_features(struct net_device *upper,
9859 struct net_device *lower, netdev_features_t features)
9861 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9862 netdev_features_t feature;
9865 for_each_netdev_feature(upper_disables, feature_bit) {
9866 feature = __NETIF_F_BIT(feature_bit);
9867 if (!(features & feature) && (lower->features & feature)) {
9868 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9869 &feature, lower->name);
9870 lower->wanted_features &= ~feature;
9871 __netdev_update_features(lower);
9873 if (unlikely(lower->features & feature))
9874 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9875 &feature, lower->name);
9877 netdev_features_change(lower);
9882 static netdev_features_t netdev_fix_features(struct net_device *dev,
9883 netdev_features_t features)
9885 /* Fix illegal checksum combinations */
9886 if ((features & NETIF_F_HW_CSUM) &&
9887 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9888 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9889 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9892 /* TSO requires that SG is present as well. */
9893 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9894 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9895 features &= ~NETIF_F_ALL_TSO;
9898 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9899 !(features & NETIF_F_IP_CSUM)) {
9900 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9901 features &= ~NETIF_F_TSO;
9902 features &= ~NETIF_F_TSO_ECN;
9905 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9906 !(features & NETIF_F_IPV6_CSUM)) {
9907 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9908 features &= ~NETIF_F_TSO6;
9911 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9912 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9913 features &= ~NETIF_F_TSO_MANGLEID;
9915 /* TSO ECN requires that TSO is present as well. */
9916 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9917 features &= ~NETIF_F_TSO_ECN;
9919 /* Software GSO depends on SG. */
9920 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9921 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9922 features &= ~NETIF_F_GSO;
9925 /* GSO partial features require GSO partial be set */
9926 if ((features & dev->gso_partial_features) &&
9927 !(features & NETIF_F_GSO_PARTIAL)) {
9929 "Dropping partially supported GSO features since no GSO partial.\n");
9930 features &= ~dev->gso_partial_features;
9933 if (!(features & NETIF_F_RXCSUM)) {
9934 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9935 * successfully merged by hardware must also have the
9936 * checksum verified by hardware. If the user does not
9937 * want to enable RXCSUM, logically, we should disable GRO_HW.
9939 if (features & NETIF_F_GRO_HW) {
9940 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9941 features &= ~NETIF_F_GRO_HW;
9945 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9946 if (features & NETIF_F_RXFCS) {
9947 if (features & NETIF_F_LRO) {
9948 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9949 features &= ~NETIF_F_LRO;
9952 if (features & NETIF_F_GRO_HW) {
9953 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9954 features &= ~NETIF_F_GRO_HW;
9958 if (features & NETIF_F_HW_TLS_TX) {
9959 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9960 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9961 bool hw_csum = features & NETIF_F_HW_CSUM;
9963 if (!ip_csum && !hw_csum) {
9964 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9965 features &= ~NETIF_F_HW_TLS_TX;
9969 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9970 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9971 features &= ~NETIF_F_HW_TLS_RX;
9977 int __netdev_update_features(struct net_device *dev)
9979 struct net_device *upper, *lower;
9980 netdev_features_t features;
9981 struct list_head *iter;
9986 features = netdev_get_wanted_features(dev);
9988 if (dev->netdev_ops->ndo_fix_features)
9989 features = dev->netdev_ops->ndo_fix_features(dev, features);
9991 /* driver might be less strict about feature dependencies */
9992 features = netdev_fix_features(dev, features);
9994 /* some features can't be enabled if they're off on an upper device */
9995 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9996 features = netdev_sync_upper_features(dev, upper, features);
9998 if (dev->features == features)
10001 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10002 &dev->features, &features);
10004 if (dev->netdev_ops->ndo_set_features)
10005 err = dev->netdev_ops->ndo_set_features(dev, features);
10009 if (unlikely(err < 0)) {
10011 "set_features() failed (%d); wanted %pNF, left %pNF\n",
10012 err, &features, &dev->features);
10013 /* return non-0 since some features might have changed and
10014 * it's better to fire a spurious notification than miss it
10020 /* some features must be disabled on lower devices when disabled
10021 * on an upper device (think: bonding master or bridge)
10023 netdev_for_each_lower_dev(dev, lower, iter)
10024 netdev_sync_lower_features(dev, lower, features);
10027 netdev_features_t diff = features ^ dev->features;
10029 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10030 /* udp_tunnel_{get,drop}_rx_info both need
10031 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10032 * device, or they won't do anything.
10033 * Thus we need to update dev->features
10034 * *before* calling udp_tunnel_get_rx_info,
10035 * but *after* calling udp_tunnel_drop_rx_info.
10037 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10038 dev->features = features;
10039 udp_tunnel_get_rx_info(dev);
10041 udp_tunnel_drop_rx_info(dev);
10045 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10046 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10047 dev->features = features;
10048 err |= vlan_get_rx_ctag_filter_info(dev);
10050 vlan_drop_rx_ctag_filter_info(dev);
10054 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10055 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10056 dev->features = features;
10057 err |= vlan_get_rx_stag_filter_info(dev);
10059 vlan_drop_rx_stag_filter_info(dev);
10063 dev->features = features;
10066 return err < 0 ? 0 : 1;
10070 * netdev_update_features - recalculate device features
10071 * @dev: the device to check
10073 * Recalculate dev->features set and send notifications if it
10074 * has changed. Should be called after driver or hardware dependent
10075 * conditions might have changed that influence the features.
10077 void netdev_update_features(struct net_device *dev)
10079 if (__netdev_update_features(dev))
10080 netdev_features_change(dev);
10082 EXPORT_SYMBOL(netdev_update_features);
10085 * netdev_change_features - recalculate device features
10086 * @dev: the device to check
10088 * Recalculate dev->features set and send notifications even
10089 * if they have not changed. Should be called instead of
10090 * netdev_update_features() if also dev->vlan_features might
10091 * have changed to allow the changes to be propagated to stacked
10094 void netdev_change_features(struct net_device *dev)
10096 __netdev_update_features(dev);
10097 netdev_features_change(dev);
10099 EXPORT_SYMBOL(netdev_change_features);
10102 * netif_stacked_transfer_operstate - transfer operstate
10103 * @rootdev: the root or lower level device to transfer state from
10104 * @dev: the device to transfer operstate to
10106 * Transfer operational state from root to device. This is normally
10107 * called when a stacking relationship exists between the root
10108 * device and the device(a leaf device).
10110 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10111 struct net_device *dev)
10113 if (rootdev->operstate == IF_OPER_DORMANT)
10114 netif_dormant_on(dev);
10116 netif_dormant_off(dev);
10118 if (rootdev->operstate == IF_OPER_TESTING)
10119 netif_testing_on(dev);
10121 netif_testing_off(dev);
10123 if (netif_carrier_ok(rootdev))
10124 netif_carrier_on(dev);
10126 netif_carrier_off(dev);
10128 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10130 static int netif_alloc_rx_queues(struct net_device *dev)
10132 unsigned int i, count = dev->num_rx_queues;
10133 struct netdev_rx_queue *rx;
10134 size_t sz = count * sizeof(*rx);
10139 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10145 for (i = 0; i < count; i++) {
10148 /* XDP RX-queue setup */
10149 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10156 /* Rollback successful reg's and free other resources */
10158 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10164 static void netif_free_rx_queues(struct net_device *dev)
10166 unsigned int i, count = dev->num_rx_queues;
10168 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10172 for (i = 0; i < count; i++)
10173 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10178 static void netdev_init_one_queue(struct net_device *dev,
10179 struct netdev_queue *queue, void *_unused)
10181 /* Initialize queue lock */
10182 spin_lock_init(&queue->_xmit_lock);
10183 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10184 queue->xmit_lock_owner = -1;
10185 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10188 dql_init(&queue->dql, HZ);
10192 static void netif_free_tx_queues(struct net_device *dev)
10197 static int netif_alloc_netdev_queues(struct net_device *dev)
10199 unsigned int count = dev->num_tx_queues;
10200 struct netdev_queue *tx;
10201 size_t sz = count * sizeof(*tx);
10203 if (count < 1 || count > 0xffff)
10206 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10212 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10213 spin_lock_init(&dev->tx_global_lock);
10218 void netif_tx_stop_all_queues(struct net_device *dev)
10222 for (i = 0; i < dev->num_tx_queues; i++) {
10223 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10225 netif_tx_stop_queue(txq);
10228 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10231 * register_netdevice - register a network device
10232 * @dev: device to register
10234 * Take a completed network device structure and add it to the kernel
10235 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10236 * chain. 0 is returned on success. A negative errno code is returned
10237 * on a failure to set up the device, or if the name is a duplicate.
10239 * Callers must hold the rtnl semaphore. You may want
10240 * register_netdev() instead of this.
10243 * The locking appears insufficient to guarantee two parallel registers
10244 * will not get the same name.
10247 int register_netdevice(struct net_device *dev)
10250 struct net *net = dev_net(dev);
10252 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10253 NETDEV_FEATURE_COUNT);
10254 BUG_ON(dev_boot_phase);
10259 /* When net_device's are persistent, this will be fatal. */
10260 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10263 ret = ethtool_check_ops(dev->ethtool_ops);
10267 spin_lock_init(&dev->addr_list_lock);
10268 netdev_set_addr_lockdep_class(dev);
10270 ret = dev_get_valid_name(net, dev, dev->name);
10275 dev->name_node = netdev_name_node_head_alloc(dev);
10276 if (!dev->name_node)
10279 /* Init, if this function is available */
10280 if (dev->netdev_ops->ndo_init) {
10281 ret = dev->netdev_ops->ndo_init(dev);
10285 goto err_free_name;
10289 if (((dev->hw_features | dev->features) &
10290 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10291 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10292 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10293 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10300 dev->ifindex = dev_new_index(net);
10301 else if (__dev_get_by_index(net, dev->ifindex))
10304 /* Transfer changeable features to wanted_features and enable
10305 * software offloads (GSO and GRO).
10307 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10308 dev->features |= NETIF_F_SOFT_FEATURES;
10310 if (dev->udp_tunnel_nic_info) {
10311 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10312 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10315 dev->wanted_features = dev->features & dev->hw_features;
10317 if (!(dev->flags & IFF_LOOPBACK))
10318 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10320 /* If IPv4 TCP segmentation offload is supported we should also
10321 * allow the device to enable segmenting the frame with the option
10322 * of ignoring a static IP ID value. This doesn't enable the
10323 * feature itself but allows the user to enable it later.
10325 if (dev->hw_features & NETIF_F_TSO)
10326 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10327 if (dev->vlan_features & NETIF_F_TSO)
10328 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10329 if (dev->mpls_features & NETIF_F_TSO)
10330 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10331 if (dev->hw_enc_features & NETIF_F_TSO)
10332 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10334 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10336 dev->vlan_features |= NETIF_F_HIGHDMA;
10338 /* Make NETIF_F_SG inheritable to tunnel devices.
10340 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10342 /* Make NETIF_F_SG inheritable to MPLS.
10344 dev->mpls_features |= NETIF_F_SG;
10346 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10347 ret = notifier_to_errno(ret);
10351 ret = netdev_register_kobject(dev);
10352 write_lock(&dev_base_lock);
10353 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10354 write_unlock(&dev_base_lock);
10358 __netdev_update_features(dev);
10361 * Default initial state at registry is that the
10362 * device is present.
10365 set_bit(__LINK_STATE_PRESENT, &dev->state);
10367 linkwatch_init_dev(dev);
10369 dev_init_scheduler(dev);
10371 list_netdevice(dev);
10372 add_device_randomness(dev->dev_addr, dev->addr_len);
10374 /* If the device has permanent device address, driver should
10375 * set dev_addr and also addr_assign_type should be set to
10376 * NET_ADDR_PERM (default value).
10378 if (dev->addr_assign_type == NET_ADDR_PERM)
10379 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10381 /* Notify protocols, that a new device appeared. */
10382 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10383 ret = notifier_to_errno(ret);
10385 /* Expect explicit free_netdev() on failure */
10386 dev->needs_free_netdev = false;
10387 unregister_netdevice_queue(dev, NULL);
10391 * Prevent userspace races by waiting until the network
10392 * device is fully setup before sending notifications.
10394 if (!dev->rtnl_link_ops ||
10395 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10396 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10402 if (dev->netdev_ops->ndo_uninit)
10403 dev->netdev_ops->ndo_uninit(dev);
10404 if (dev->priv_destructor)
10405 dev->priv_destructor(dev);
10407 netdev_name_node_free(dev->name_node);
10410 EXPORT_SYMBOL(register_netdevice);
10413 * init_dummy_netdev - init a dummy network device for NAPI
10414 * @dev: device to init
10416 * This takes a network device structure and initialize the minimum
10417 * amount of fields so it can be used to schedule NAPI polls without
10418 * registering a full blown interface. This is to be used by drivers
10419 * that need to tie several hardware interfaces to a single NAPI
10420 * poll scheduler due to HW limitations.
10422 int init_dummy_netdev(struct net_device *dev)
10424 /* Clear everything. Note we don't initialize spinlocks
10425 * are they aren't supposed to be taken by any of the
10426 * NAPI code and this dummy netdev is supposed to be
10427 * only ever used for NAPI polls
10429 memset(dev, 0, sizeof(struct net_device));
10431 /* make sure we BUG if trying to hit standard
10432 * register/unregister code path
10434 dev->reg_state = NETREG_DUMMY;
10436 /* NAPI wants this */
10437 INIT_LIST_HEAD(&dev->napi_list);
10439 /* a dummy interface is started by default */
10440 set_bit(__LINK_STATE_PRESENT, &dev->state);
10441 set_bit(__LINK_STATE_START, &dev->state);
10443 /* napi_busy_loop stats accounting wants this */
10444 dev_net_set(dev, &init_net);
10446 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10447 * because users of this 'device' dont need to change
10453 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10457 * register_netdev - register a network device
10458 * @dev: device to register
10460 * Take a completed network device structure and add it to the kernel
10461 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10462 * chain. 0 is returned on success. A negative errno code is returned
10463 * on a failure to set up the device, or if the name is a duplicate.
10465 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10466 * and expands the device name if you passed a format string to
10469 int register_netdev(struct net_device *dev)
10473 if (rtnl_lock_killable())
10475 err = register_netdevice(dev);
10479 EXPORT_SYMBOL(register_netdev);
10481 int netdev_refcnt_read(const struct net_device *dev)
10483 #ifdef CONFIG_PCPU_DEV_REFCNT
10486 for_each_possible_cpu(i)
10487 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10490 return refcount_read(&dev->dev_refcnt);
10493 EXPORT_SYMBOL(netdev_refcnt_read);
10495 int netdev_unregister_timeout_secs __read_mostly = 10;
10497 #define WAIT_REFS_MIN_MSECS 1
10498 #define WAIT_REFS_MAX_MSECS 250
10500 * netdev_wait_allrefs - wait until all references are gone.
10501 * @dev: target net_device
10503 * This is called when unregistering network devices.
10505 * Any protocol or device that holds a reference should register
10506 * for netdevice notification, and cleanup and put back the
10507 * reference if they receive an UNREGISTER event.
10508 * We can get stuck here if buggy protocols don't correctly
10511 static void netdev_wait_allrefs(struct net_device *dev)
10513 unsigned long rebroadcast_time, warning_time;
10514 int wait = 0, refcnt;
10516 rebroadcast_time = warning_time = jiffies;
10517 refcnt = netdev_refcnt_read(dev);
10519 while (refcnt != 1) {
10520 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10523 /* Rebroadcast unregister notification */
10524 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10530 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10532 /* We must not have linkwatch events
10533 * pending on unregister. If this
10534 * happens, we simply run the queue
10535 * unscheduled, resulting in a noop
10538 linkwatch_run_queue();
10543 rebroadcast_time = jiffies;
10548 wait = WAIT_REFS_MIN_MSECS;
10551 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10554 refcnt = netdev_refcnt_read(dev);
10557 time_after(jiffies, warning_time +
10558 netdev_unregister_timeout_secs * HZ)) {
10559 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10560 dev->name, refcnt);
10561 warning_time = jiffies;
10566 /* The sequence is:
10570 * register_netdevice(x1);
10571 * register_netdevice(x2);
10573 * unregister_netdevice(y1);
10574 * unregister_netdevice(y2);
10580 * We are invoked by rtnl_unlock().
10581 * This allows us to deal with problems:
10582 * 1) We can delete sysfs objects which invoke hotplug
10583 * without deadlocking with linkwatch via keventd.
10584 * 2) Since we run with the RTNL semaphore not held, we can sleep
10585 * safely in order to wait for the netdev refcnt to drop to zero.
10587 * We must not return until all unregister events added during
10588 * the interval the lock was held have been completed.
10590 void netdev_run_todo(void)
10592 struct list_head list;
10593 #ifdef CONFIG_LOCKDEP
10594 struct list_head unlink_list;
10596 list_replace_init(&net_unlink_list, &unlink_list);
10598 while (!list_empty(&unlink_list)) {
10599 struct net_device *dev = list_first_entry(&unlink_list,
10602 list_del_init(&dev->unlink_list);
10603 dev->nested_level = dev->lower_level - 1;
10607 /* Snapshot list, allow later requests */
10608 list_replace_init(&net_todo_list, &list);
10613 /* Wait for rcu callbacks to finish before next phase */
10614 if (!list_empty(&list))
10617 while (!list_empty(&list)) {
10618 struct net_device *dev
10619 = list_first_entry(&list, struct net_device, todo_list);
10620 list_del(&dev->todo_list);
10622 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10623 pr_err("network todo '%s' but state %d\n",
10624 dev->name, dev->reg_state);
10629 write_lock(&dev_base_lock);
10630 dev->reg_state = NETREG_UNREGISTERED;
10631 write_unlock(&dev_base_lock);
10632 linkwatch_forget_dev(dev);
10634 netdev_wait_allrefs(dev);
10637 BUG_ON(netdev_refcnt_read(dev) != 1);
10638 BUG_ON(!list_empty(&dev->ptype_all));
10639 BUG_ON(!list_empty(&dev->ptype_specific));
10640 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10641 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10643 if (dev->priv_destructor)
10644 dev->priv_destructor(dev);
10645 if (dev->needs_free_netdev)
10648 /* Report a network device has been unregistered */
10650 dev_net(dev)->dev_unreg_count--;
10652 wake_up(&netdev_unregistering_wq);
10654 /* Free network device */
10655 kobject_put(&dev->dev.kobj);
10659 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10660 * all the same fields in the same order as net_device_stats, with only
10661 * the type differing, but rtnl_link_stats64 may have additional fields
10662 * at the end for newer counters.
10664 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10665 const struct net_device_stats *netdev_stats)
10667 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10668 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10669 u64 *dst = (u64 *)stats64;
10671 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10672 for (i = 0; i < n; i++)
10673 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10674 /* zero out counters that only exist in rtnl_link_stats64 */
10675 memset((char *)stats64 + n * sizeof(u64), 0,
10676 sizeof(*stats64) - n * sizeof(u64));
10678 EXPORT_SYMBOL(netdev_stats_to_stats64);
10681 * dev_get_stats - get network device statistics
10682 * @dev: device to get statistics from
10683 * @storage: place to store stats
10685 * Get network statistics from device. Return @storage.
10686 * The device driver may provide its own method by setting
10687 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10688 * otherwise the internal statistics structure is used.
10690 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10691 struct rtnl_link_stats64 *storage)
10693 const struct net_device_ops *ops = dev->netdev_ops;
10695 if (ops->ndo_get_stats64) {
10696 memset(storage, 0, sizeof(*storage));
10697 ops->ndo_get_stats64(dev, storage);
10698 } else if (ops->ndo_get_stats) {
10699 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10701 netdev_stats_to_stats64(storage, &dev->stats);
10703 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10704 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10705 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10708 EXPORT_SYMBOL(dev_get_stats);
10711 * dev_fetch_sw_netstats - get per-cpu network device statistics
10712 * @s: place to store stats
10713 * @netstats: per-cpu network stats to read from
10715 * Read per-cpu network statistics and populate the related fields in @s.
10717 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10718 const struct pcpu_sw_netstats __percpu *netstats)
10722 for_each_possible_cpu(cpu) {
10723 const struct pcpu_sw_netstats *stats;
10724 struct pcpu_sw_netstats tmp;
10725 unsigned int start;
10727 stats = per_cpu_ptr(netstats, cpu);
10729 start = u64_stats_fetch_begin_irq(&stats->syncp);
10730 tmp.rx_packets = stats->rx_packets;
10731 tmp.rx_bytes = stats->rx_bytes;
10732 tmp.tx_packets = stats->tx_packets;
10733 tmp.tx_bytes = stats->tx_bytes;
10734 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10736 s->rx_packets += tmp.rx_packets;
10737 s->rx_bytes += tmp.rx_bytes;
10738 s->tx_packets += tmp.tx_packets;
10739 s->tx_bytes += tmp.tx_bytes;
10742 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10745 * dev_get_tstats64 - ndo_get_stats64 implementation
10746 * @dev: device to get statistics from
10747 * @s: place to store stats
10749 * Populate @s from dev->stats and dev->tstats. Can be used as
10750 * ndo_get_stats64() callback.
10752 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10754 netdev_stats_to_stats64(s, &dev->stats);
10755 dev_fetch_sw_netstats(s, dev->tstats);
10757 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10759 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10761 struct netdev_queue *queue = dev_ingress_queue(dev);
10763 #ifdef CONFIG_NET_CLS_ACT
10766 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10769 netdev_init_one_queue(dev, queue, NULL);
10770 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10771 queue->qdisc_sleeping = &noop_qdisc;
10772 rcu_assign_pointer(dev->ingress_queue, queue);
10777 static const struct ethtool_ops default_ethtool_ops;
10779 void netdev_set_default_ethtool_ops(struct net_device *dev,
10780 const struct ethtool_ops *ops)
10782 if (dev->ethtool_ops == &default_ethtool_ops)
10783 dev->ethtool_ops = ops;
10785 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10787 void netdev_freemem(struct net_device *dev)
10789 char *addr = (char *)dev - dev->padded;
10795 * alloc_netdev_mqs - allocate network device
10796 * @sizeof_priv: size of private data to allocate space for
10797 * @name: device name format string
10798 * @name_assign_type: origin of device name
10799 * @setup: callback to initialize device
10800 * @txqs: the number of TX subqueues to allocate
10801 * @rxqs: the number of RX subqueues to allocate
10803 * Allocates a struct net_device with private data area for driver use
10804 * and performs basic initialization. Also allocates subqueue structs
10805 * for each queue on the device.
10807 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10808 unsigned char name_assign_type,
10809 void (*setup)(struct net_device *),
10810 unsigned int txqs, unsigned int rxqs)
10812 struct net_device *dev;
10813 unsigned int alloc_size;
10814 struct net_device *p;
10816 BUG_ON(strlen(name) >= sizeof(dev->name));
10819 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10824 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10828 alloc_size = sizeof(struct net_device);
10830 /* ensure 32-byte alignment of private area */
10831 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10832 alloc_size += sizeof_priv;
10834 /* ensure 32-byte alignment of whole construct */
10835 alloc_size += NETDEV_ALIGN - 1;
10837 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10841 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10842 dev->padded = (char *)dev - (char *)p;
10844 #ifdef CONFIG_PCPU_DEV_REFCNT
10845 dev->pcpu_refcnt = alloc_percpu(int);
10846 if (!dev->pcpu_refcnt)
10850 refcount_set(&dev->dev_refcnt, 1);
10853 if (dev_addr_init(dev))
10859 dev_net_set(dev, &init_net);
10861 dev->gso_max_size = GSO_MAX_SIZE;
10862 dev->gso_max_segs = GSO_MAX_SEGS;
10863 dev->upper_level = 1;
10864 dev->lower_level = 1;
10865 #ifdef CONFIG_LOCKDEP
10866 dev->nested_level = 0;
10867 INIT_LIST_HEAD(&dev->unlink_list);
10870 INIT_LIST_HEAD(&dev->napi_list);
10871 INIT_LIST_HEAD(&dev->unreg_list);
10872 INIT_LIST_HEAD(&dev->close_list);
10873 INIT_LIST_HEAD(&dev->link_watch_list);
10874 INIT_LIST_HEAD(&dev->adj_list.upper);
10875 INIT_LIST_HEAD(&dev->adj_list.lower);
10876 INIT_LIST_HEAD(&dev->ptype_all);
10877 INIT_LIST_HEAD(&dev->ptype_specific);
10878 INIT_LIST_HEAD(&dev->net_notifier_list);
10879 #ifdef CONFIG_NET_SCHED
10880 hash_init(dev->qdisc_hash);
10882 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10885 if (!dev->tx_queue_len) {
10886 dev->priv_flags |= IFF_NO_QUEUE;
10887 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10890 dev->num_tx_queues = txqs;
10891 dev->real_num_tx_queues = txqs;
10892 if (netif_alloc_netdev_queues(dev))
10895 dev->num_rx_queues = rxqs;
10896 dev->real_num_rx_queues = rxqs;
10897 if (netif_alloc_rx_queues(dev))
10900 strcpy(dev->name, name);
10901 dev->name_assign_type = name_assign_type;
10902 dev->group = INIT_NETDEV_GROUP;
10903 if (!dev->ethtool_ops)
10904 dev->ethtool_ops = &default_ethtool_ops;
10906 nf_hook_ingress_init(dev);
10915 #ifdef CONFIG_PCPU_DEV_REFCNT
10916 free_percpu(dev->pcpu_refcnt);
10919 netdev_freemem(dev);
10922 EXPORT_SYMBOL(alloc_netdev_mqs);
10925 * free_netdev - free network device
10928 * This function does the last stage of destroying an allocated device
10929 * interface. The reference to the device object is released. If this
10930 * is the last reference then it will be freed.Must be called in process
10933 void free_netdev(struct net_device *dev)
10935 struct napi_struct *p, *n;
10939 /* When called immediately after register_netdevice() failed the unwind
10940 * handling may still be dismantling the device. Handle that case by
10941 * deferring the free.
10943 if (dev->reg_state == NETREG_UNREGISTERING) {
10945 dev->needs_free_netdev = true;
10949 netif_free_tx_queues(dev);
10950 netif_free_rx_queues(dev);
10952 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10954 /* Flush device addresses */
10955 dev_addr_flush(dev);
10957 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10960 #ifdef CONFIG_PCPU_DEV_REFCNT
10961 free_percpu(dev->pcpu_refcnt);
10962 dev->pcpu_refcnt = NULL;
10964 free_percpu(dev->xdp_bulkq);
10965 dev->xdp_bulkq = NULL;
10967 /* Compatibility with error handling in drivers */
10968 if (dev->reg_state == NETREG_UNINITIALIZED) {
10969 netdev_freemem(dev);
10973 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10974 dev->reg_state = NETREG_RELEASED;
10976 /* will free via device release */
10977 put_device(&dev->dev);
10979 EXPORT_SYMBOL(free_netdev);
10982 * synchronize_net - Synchronize with packet receive processing
10984 * Wait for packets currently being received to be done.
10985 * Does not block later packets from starting.
10987 void synchronize_net(void)
10990 if (rtnl_is_locked())
10991 synchronize_rcu_expedited();
10995 EXPORT_SYMBOL(synchronize_net);
10998 * unregister_netdevice_queue - remove device from the kernel
11002 * This function shuts down a device interface and removes it
11003 * from the kernel tables.
11004 * If head not NULL, device is queued to be unregistered later.
11006 * Callers must hold the rtnl semaphore. You may want
11007 * unregister_netdev() instead of this.
11010 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11015 list_move_tail(&dev->unreg_list, head);
11019 list_add(&dev->unreg_list, &single);
11020 unregister_netdevice_many(&single);
11023 EXPORT_SYMBOL(unregister_netdevice_queue);
11026 * unregister_netdevice_many - unregister many devices
11027 * @head: list of devices
11029 * Note: As most callers use a stack allocated list_head,
11030 * we force a list_del() to make sure stack wont be corrupted later.
11032 void unregister_netdevice_many(struct list_head *head)
11034 struct net_device *dev, *tmp;
11035 LIST_HEAD(close_head);
11037 BUG_ON(dev_boot_phase);
11040 if (list_empty(head))
11043 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11044 /* Some devices call without registering
11045 * for initialization unwind. Remove those
11046 * devices and proceed with the remaining.
11048 if (dev->reg_state == NETREG_UNINITIALIZED) {
11049 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11053 list_del(&dev->unreg_list);
11056 dev->dismantle = true;
11057 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11060 /* If device is running, close it first. */
11061 list_for_each_entry(dev, head, unreg_list)
11062 list_add_tail(&dev->close_list, &close_head);
11063 dev_close_many(&close_head, true);
11065 list_for_each_entry(dev, head, unreg_list) {
11066 /* And unlink it from device chain. */
11067 write_lock(&dev_base_lock);
11068 unlist_netdevice(dev, false);
11069 dev->reg_state = NETREG_UNREGISTERING;
11070 write_unlock(&dev_base_lock);
11072 flush_all_backlogs();
11076 list_for_each_entry(dev, head, unreg_list) {
11077 struct sk_buff *skb = NULL;
11079 /* Shutdown queueing discipline. */
11082 dev_xdp_uninstall(dev);
11084 /* Notify protocols, that we are about to destroy
11085 * this device. They should clean all the things.
11087 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11089 if (!dev->rtnl_link_ops ||
11090 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11091 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11092 GFP_KERNEL, NULL, 0);
11095 * Flush the unicast and multicast chains
11100 netdev_name_node_alt_flush(dev);
11101 netdev_name_node_free(dev->name_node);
11103 if (dev->netdev_ops->ndo_uninit)
11104 dev->netdev_ops->ndo_uninit(dev);
11107 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11109 /* Notifier chain MUST detach us all upper devices. */
11110 WARN_ON(netdev_has_any_upper_dev(dev));
11111 WARN_ON(netdev_has_any_lower_dev(dev));
11113 /* Remove entries from kobject tree */
11114 netdev_unregister_kobject(dev);
11116 /* Remove XPS queueing entries */
11117 netif_reset_xps_queues_gt(dev, 0);
11123 list_for_each_entry(dev, head, unreg_list) {
11130 EXPORT_SYMBOL(unregister_netdevice_many);
11133 * unregister_netdev - remove device from the kernel
11136 * This function shuts down a device interface and removes it
11137 * from the kernel tables.
11139 * This is just a wrapper for unregister_netdevice that takes
11140 * the rtnl semaphore. In general you want to use this and not
11141 * unregister_netdevice.
11143 void unregister_netdev(struct net_device *dev)
11146 unregister_netdevice(dev);
11149 EXPORT_SYMBOL(unregister_netdev);
11152 * __dev_change_net_namespace - move device to different nethost namespace
11154 * @net: network namespace
11155 * @pat: If not NULL name pattern to try if the current device name
11156 * is already taken in the destination network namespace.
11157 * @new_ifindex: If not zero, specifies device index in the target
11160 * This function shuts down a device interface and moves it
11161 * to a new network namespace. On success 0 is returned, on
11162 * a failure a netagive errno code is returned.
11164 * Callers must hold the rtnl semaphore.
11167 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11168 const char *pat, int new_ifindex)
11170 struct net *net_old = dev_net(dev);
11171 char new_name[IFNAMSIZ] = {};
11176 /* Don't allow namespace local devices to be moved. */
11178 if (dev->features & NETIF_F_NETNS_LOCAL)
11181 /* Ensure the device has been registrered */
11182 if (dev->reg_state != NETREG_REGISTERED)
11185 /* Get out if there is nothing todo */
11187 if (net_eq(net_old, net))
11190 /* Pick the destination device name, and ensure
11191 * we can use it in the destination network namespace.
11194 if (netdev_name_in_use(net, dev->name)) {
11195 /* We get here if we can't use the current device name */
11198 err = dev_prep_valid_name(net, dev, pat, new_name);
11203 /* Check that new_ifindex isn't used yet. */
11205 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11209 * And now a mini version of register_netdevice unregister_netdevice.
11212 /* If device is running close it first. */
11215 /* And unlink it from device chain */
11216 unlist_netdevice(dev, true);
11220 /* Shutdown queueing discipline. */
11223 /* Notify protocols, that we are about to destroy
11224 * this device. They should clean all the things.
11226 * Note that dev->reg_state stays at NETREG_REGISTERED.
11227 * This is wanted because this way 8021q and macvlan know
11228 * the device is just moving and can keep their slaves up.
11230 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11233 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11234 /* If there is an ifindex conflict assign a new one */
11235 if (!new_ifindex) {
11236 if (__dev_get_by_index(net, dev->ifindex))
11237 new_ifindex = dev_new_index(net);
11239 new_ifindex = dev->ifindex;
11242 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11246 * Flush the unicast and multicast chains
11251 /* Send a netdev-removed uevent to the old namespace */
11252 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11253 netdev_adjacent_del_links(dev);
11255 /* Move per-net netdevice notifiers that are following the netdevice */
11256 move_netdevice_notifiers_dev_net(dev, net);
11258 /* Actually switch the network namespace */
11259 dev_net_set(dev, net);
11260 dev->ifindex = new_ifindex;
11262 /* Send a netdev-add uevent to the new namespace */
11263 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11264 netdev_adjacent_add_links(dev);
11266 if (new_name[0]) /* Rename the netdev to prepared name */
11267 strscpy(dev->name, new_name, IFNAMSIZ);
11269 /* Fixup kobjects */
11270 err = device_rename(&dev->dev, dev->name);
11273 /* Adapt owner in case owning user namespace of target network
11274 * namespace is different from the original one.
11276 err = netdev_change_owner(dev, net_old, net);
11279 /* Add the device back in the hashes */
11280 list_netdevice(dev);
11282 /* Notify protocols, that a new device appeared. */
11283 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11286 * Prevent userspace races by waiting until the network
11287 * device is fully setup before sending notifications.
11289 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11296 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11298 static int dev_cpu_dead(unsigned int oldcpu)
11300 struct sk_buff **list_skb;
11301 struct sk_buff *skb;
11303 struct softnet_data *sd, *oldsd, *remsd = NULL;
11305 local_irq_disable();
11306 cpu = smp_processor_id();
11307 sd = &per_cpu(softnet_data, cpu);
11308 oldsd = &per_cpu(softnet_data, oldcpu);
11310 /* Find end of our completion_queue. */
11311 list_skb = &sd->completion_queue;
11313 list_skb = &(*list_skb)->next;
11314 /* Append completion queue from offline CPU. */
11315 *list_skb = oldsd->completion_queue;
11316 oldsd->completion_queue = NULL;
11318 /* Append output queue from offline CPU. */
11319 if (oldsd->output_queue) {
11320 *sd->output_queue_tailp = oldsd->output_queue;
11321 sd->output_queue_tailp = oldsd->output_queue_tailp;
11322 oldsd->output_queue = NULL;
11323 oldsd->output_queue_tailp = &oldsd->output_queue;
11325 /* Append NAPI poll list from offline CPU, with one exception :
11326 * process_backlog() must be called by cpu owning percpu backlog.
11327 * We properly handle process_queue & input_pkt_queue later.
11329 while (!list_empty(&oldsd->poll_list)) {
11330 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11331 struct napi_struct,
11334 list_del_init(&napi->poll_list);
11335 if (napi->poll == process_backlog)
11338 ____napi_schedule(sd, napi);
11341 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11342 local_irq_enable();
11345 remsd = oldsd->rps_ipi_list;
11346 oldsd->rps_ipi_list = NULL;
11348 /* send out pending IPI's on offline CPU */
11349 net_rps_send_ipi(remsd);
11351 /* Process offline CPU's input_pkt_queue */
11352 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11354 input_queue_head_incr(oldsd);
11356 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11358 input_queue_head_incr(oldsd);
11365 * netdev_increment_features - increment feature set by one
11366 * @all: current feature set
11367 * @one: new feature set
11368 * @mask: mask feature set
11370 * Computes a new feature set after adding a device with feature set
11371 * @one to the master device with current feature set @all. Will not
11372 * enable anything that is off in @mask. Returns the new feature set.
11374 netdev_features_t netdev_increment_features(netdev_features_t all,
11375 netdev_features_t one, netdev_features_t mask)
11377 if (mask & NETIF_F_HW_CSUM)
11378 mask |= NETIF_F_CSUM_MASK;
11379 mask |= NETIF_F_VLAN_CHALLENGED;
11381 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11382 all &= one | ~NETIF_F_ALL_FOR_ALL;
11384 /* If one device supports hw checksumming, set for all. */
11385 if (all & NETIF_F_HW_CSUM)
11386 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11390 EXPORT_SYMBOL(netdev_increment_features);
11392 static struct hlist_head * __net_init netdev_create_hash(void)
11395 struct hlist_head *hash;
11397 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11399 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11400 INIT_HLIST_HEAD(&hash[i]);
11405 /* Initialize per network namespace state */
11406 static int __net_init netdev_init(struct net *net)
11408 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11409 8 * sizeof_field(struct napi_struct, gro_bitmask));
11411 INIT_LIST_HEAD(&net->dev_base_head);
11413 net->dev_name_head = netdev_create_hash();
11414 if (net->dev_name_head == NULL)
11417 net->dev_index_head = netdev_create_hash();
11418 if (net->dev_index_head == NULL)
11421 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11426 kfree(net->dev_name_head);
11432 * netdev_drivername - network driver for the device
11433 * @dev: network device
11435 * Determine network driver for device.
11437 const char *netdev_drivername(const struct net_device *dev)
11439 const struct device_driver *driver;
11440 const struct device *parent;
11441 const char *empty = "";
11443 parent = dev->dev.parent;
11447 driver = parent->driver;
11448 if (driver && driver->name)
11449 return driver->name;
11453 static void __netdev_printk(const char *level, const struct net_device *dev,
11454 struct va_format *vaf)
11456 if (dev && dev->dev.parent) {
11457 dev_printk_emit(level[1] - '0',
11460 dev_driver_string(dev->dev.parent),
11461 dev_name(dev->dev.parent),
11462 netdev_name(dev), netdev_reg_state(dev),
11465 printk("%s%s%s: %pV",
11466 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11468 printk("%s(NULL net_device): %pV", level, vaf);
11472 void netdev_printk(const char *level, const struct net_device *dev,
11473 const char *format, ...)
11475 struct va_format vaf;
11478 va_start(args, format);
11483 __netdev_printk(level, dev, &vaf);
11487 EXPORT_SYMBOL(netdev_printk);
11489 #define define_netdev_printk_level(func, level) \
11490 void func(const struct net_device *dev, const char *fmt, ...) \
11492 struct va_format vaf; \
11495 va_start(args, fmt); \
11500 __netdev_printk(level, dev, &vaf); \
11504 EXPORT_SYMBOL(func);
11506 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11507 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11508 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11509 define_netdev_printk_level(netdev_err, KERN_ERR);
11510 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11511 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11512 define_netdev_printk_level(netdev_info, KERN_INFO);
11514 static void __net_exit netdev_exit(struct net *net)
11516 kfree(net->dev_name_head);
11517 kfree(net->dev_index_head);
11518 if (net != &init_net)
11519 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11522 static struct pernet_operations __net_initdata netdev_net_ops = {
11523 .init = netdev_init,
11524 .exit = netdev_exit,
11527 static void __net_exit default_device_exit(struct net *net)
11529 struct net_device *dev, *aux;
11531 * Push all migratable network devices back to the
11532 * initial network namespace
11535 for_each_netdev_safe(net, dev, aux) {
11537 char fb_name[IFNAMSIZ];
11539 /* Ignore unmoveable devices (i.e. loopback) */
11540 if (dev->features & NETIF_F_NETNS_LOCAL)
11543 /* Leave virtual devices for the generic cleanup */
11544 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11547 /* Push remaining network devices to init_net */
11548 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11549 if (netdev_name_in_use(&init_net, fb_name))
11550 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11551 err = dev_change_net_namespace(dev, &init_net, fb_name);
11553 pr_emerg("%s: failed to move %s to init_net: %d\n",
11554 __func__, dev->name, err);
11561 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11563 /* Return with the rtnl_lock held when there are no network
11564 * devices unregistering in any network namespace in net_list.
11567 bool unregistering;
11568 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11570 add_wait_queue(&netdev_unregistering_wq, &wait);
11572 unregistering = false;
11574 list_for_each_entry(net, net_list, exit_list) {
11575 if (net->dev_unreg_count > 0) {
11576 unregistering = true;
11580 if (!unregistering)
11584 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11586 remove_wait_queue(&netdev_unregistering_wq, &wait);
11589 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11591 /* At exit all network devices most be removed from a network
11592 * namespace. Do this in the reverse order of registration.
11593 * Do this across as many network namespaces as possible to
11594 * improve batching efficiency.
11596 struct net_device *dev;
11598 LIST_HEAD(dev_kill_list);
11600 /* To prevent network device cleanup code from dereferencing
11601 * loopback devices or network devices that have been freed
11602 * wait here for all pending unregistrations to complete,
11603 * before unregistring the loopback device and allowing the
11604 * network namespace be freed.
11606 * The netdev todo list containing all network devices
11607 * unregistrations that happen in default_device_exit_batch
11608 * will run in the rtnl_unlock() at the end of
11609 * default_device_exit_batch.
11611 rtnl_lock_unregistering(net_list);
11612 list_for_each_entry(net, net_list, exit_list) {
11613 for_each_netdev_reverse(net, dev) {
11614 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11615 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11617 unregister_netdevice_queue(dev, &dev_kill_list);
11620 unregister_netdevice_many(&dev_kill_list);
11624 static struct pernet_operations __net_initdata default_device_ops = {
11625 .exit = default_device_exit,
11626 .exit_batch = default_device_exit_batch,
11630 * Initialize the DEV module. At boot time this walks the device list and
11631 * unhooks any devices that fail to initialise (normally hardware not
11632 * present) and leaves us with a valid list of present and active devices.
11637 * This is called single threaded during boot, so no need
11638 * to take the rtnl semaphore.
11640 static int __init net_dev_init(void)
11642 int i, rc = -ENOMEM;
11644 BUG_ON(!dev_boot_phase);
11646 if (dev_proc_init())
11649 if (netdev_kobject_init())
11652 INIT_LIST_HEAD(&ptype_all);
11653 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11654 INIT_LIST_HEAD(&ptype_base[i]);
11656 INIT_LIST_HEAD(&offload_base);
11658 if (register_pernet_subsys(&netdev_net_ops))
11662 * Initialise the packet receive queues.
11665 for_each_possible_cpu(i) {
11666 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11667 struct softnet_data *sd = &per_cpu(softnet_data, i);
11669 INIT_WORK(flush, flush_backlog);
11671 skb_queue_head_init(&sd->input_pkt_queue);
11672 skb_queue_head_init(&sd->process_queue);
11673 #ifdef CONFIG_XFRM_OFFLOAD
11674 skb_queue_head_init(&sd->xfrm_backlog);
11676 INIT_LIST_HEAD(&sd->poll_list);
11677 sd->output_queue_tailp = &sd->output_queue;
11679 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11683 init_gro_hash(&sd->backlog);
11684 sd->backlog.poll = process_backlog;
11685 sd->backlog.weight = weight_p;
11688 dev_boot_phase = 0;
11690 /* The loopback device is special if any other network devices
11691 * is present in a network namespace the loopback device must
11692 * be present. Since we now dynamically allocate and free the
11693 * loopback device ensure this invariant is maintained by
11694 * keeping the loopback device as the first device on the
11695 * list of network devices. Ensuring the loopback devices
11696 * is the first device that appears and the last network device
11699 if (register_pernet_device(&loopback_net_ops))
11702 if (register_pernet_device(&default_device_ops))
11705 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11706 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11708 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11709 NULL, dev_cpu_dead);
11716 subsys_initcall(net_dev_init);