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/string.h>
84 #include <linux/socket.h>
85 #include <linux/sockios.h>
86 #include <linux/errno.h>
87 #include <linux/interrupt.h>
88 #include <linux/if_ether.h>
89 #include <linux/netdevice.h>
90 #include <linux/etherdevice.h>
91 #include <linux/ethtool.h>
92 #include <linux/skbuff.h>
93 #include <linux/bpf.h>
94 #include <linux/bpf_trace.h>
95 #include <net/net_namespace.h>
97 #include <net/busy_poll.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/stat.h>
101 #include <net/dst_metadata.h>
102 #include <net/pkt_sched.h>
103 #include <net/pkt_cls.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
139 #include <linux/crash_dump.h>
140 #include <linux/sctp.h>
141 #include <net/udp_tunnel.h>
142 #include <linux/net_namespace.h>
143 #include <linux/indirect_call_wrapper.h>
144 #include <net/devlink.h>
146 #include "net-sysfs.h"
148 #define MAX_GRO_SKBS 8
150 /* This should be increased if a protocol with a bigger head is added. */
151 #define GRO_MAX_HEAD (MAX_HEADER + 128)
153 static DEFINE_SPINLOCK(ptype_lock);
154 static DEFINE_SPINLOCK(offload_lock);
155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
156 struct list_head ptype_all __read_mostly; /* Taps */
157 static struct list_head offload_base __read_mostly;
159 static int netif_rx_internal(struct sk_buff *skb);
160 static int call_netdevice_notifiers_info(unsigned long val,
161 struct netdev_notifier_info *info);
162 static int call_netdevice_notifiers_extack(unsigned long val,
163 struct net_device *dev,
164 struct netlink_ext_ack *extack);
165 static struct napi_struct *napi_by_id(unsigned int napi_id);
168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base_head list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 DEFINE_RWLOCK(dev_base_lock);
187 EXPORT_SYMBOL(dev_base_lock);
189 static DEFINE_MUTEX(ifalias_mutex);
191 /* protects napi_hash addition/deletion and napi_gen_id */
192 static DEFINE_SPINLOCK(napi_hash_lock);
194 static unsigned int napi_gen_id = NR_CPUS;
195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
197 static seqcount_t devnet_rename_seq;
199 static inline void dev_base_seq_inc(struct net *net)
201 while (++net->dev_base_seq == 0)
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217 static inline void rps_lock(struct softnet_data *sd)
220 spin_lock(&sd->input_pkt_queue.lock);
224 static inline void rps_unlock(struct softnet_data *sd)
227 spin_unlock(&sd->input_pkt_queue.lock);
231 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
234 struct netdev_name_node *name_node;
236 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
239 INIT_HLIST_NODE(&name_node->hlist);
240 name_node->dev = dev;
241 name_node->name = name;
245 static struct netdev_name_node *
246 netdev_name_node_head_alloc(struct net_device *dev)
248 struct netdev_name_node *name_node;
250 name_node = netdev_name_node_alloc(dev, dev->name);
253 INIT_LIST_HEAD(&name_node->list);
257 static void netdev_name_node_free(struct netdev_name_node *name_node)
262 static void netdev_name_node_add(struct net *net,
263 struct netdev_name_node *name_node)
265 hlist_add_head_rcu(&name_node->hlist,
266 dev_name_hash(net, name_node->name));
269 static void netdev_name_node_del(struct netdev_name_node *name_node)
271 hlist_del_rcu(&name_node->hlist);
274 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
277 struct hlist_head *head = dev_name_hash(net, name);
278 struct netdev_name_node *name_node;
280 hlist_for_each_entry(name_node, head, hlist)
281 if (!strcmp(name_node->name, name))
286 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
289 struct hlist_head *head = dev_name_hash(net, name);
290 struct netdev_name_node *name_node;
292 hlist_for_each_entry_rcu(name_node, head, hlist)
293 if (!strcmp(name_node->name, name))
298 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
300 struct netdev_name_node *name_node;
301 struct net *net = dev_net(dev);
303 name_node = netdev_name_node_lookup(net, name);
306 name_node = netdev_name_node_alloc(dev, name);
309 netdev_name_node_add(net, name_node);
310 /* The node that holds dev->name acts as a head of per-device list. */
311 list_add_tail(&name_node->list, &dev->name_node->list);
315 EXPORT_SYMBOL(netdev_name_node_alt_create);
317 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
319 list_del(&name_node->list);
320 netdev_name_node_del(name_node);
321 kfree(name_node->name);
322 netdev_name_node_free(name_node);
325 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
327 struct netdev_name_node *name_node;
328 struct net *net = dev_net(dev);
330 name_node = netdev_name_node_lookup(net, name);
333 /* lookup might have found our primary name or a name belonging
336 if (name_node == dev->name_node || name_node->dev != dev)
339 __netdev_name_node_alt_destroy(name_node);
343 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
345 static void netdev_name_node_alt_flush(struct net_device *dev)
347 struct netdev_name_node *name_node, *tmp;
349 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
350 __netdev_name_node_alt_destroy(name_node);
353 /* Device list insertion */
354 static void list_netdevice(struct net_device *dev)
356 struct net *net = dev_net(dev);
360 write_lock_bh(&dev_base_lock);
361 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
362 netdev_name_node_add(net, dev->name_node);
363 hlist_add_head_rcu(&dev->index_hlist,
364 dev_index_hash(net, dev->ifindex));
365 write_unlock_bh(&dev_base_lock);
367 dev_base_seq_inc(net);
370 /* Device list removal
371 * caller must respect a RCU grace period before freeing/reusing dev
373 static void unlist_netdevice(struct net_device *dev)
377 /* Unlink dev from the device chain */
378 write_lock_bh(&dev_base_lock);
379 list_del_rcu(&dev->dev_list);
380 netdev_name_node_del(dev->name_node);
381 hlist_del_rcu(&dev->index_hlist);
382 write_unlock_bh(&dev_base_lock);
384 dev_base_seq_inc(dev_net(dev));
391 static RAW_NOTIFIER_HEAD(netdev_chain);
394 * Device drivers call our routines to queue packets here. We empty the
395 * queue in the local softnet handler.
398 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
399 EXPORT_PER_CPU_SYMBOL(softnet_data);
401 #ifdef CONFIG_LOCKDEP
403 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
404 * according to dev->type
406 static const unsigned short netdev_lock_type[] = {
407 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
408 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
409 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
410 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
411 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
412 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
413 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
414 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
415 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
416 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
417 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
418 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
419 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
420 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
421 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
423 static const char *const netdev_lock_name[] = {
424 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
425 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
426 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
427 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
428 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
429 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
430 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
431 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
432 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
433 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
434 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
435 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
436 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
437 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
438 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
440 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
442 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
446 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
447 if (netdev_lock_type[i] == dev_type)
449 /* the last key is used by default */
450 return ARRAY_SIZE(netdev_lock_type) - 1;
453 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
454 unsigned short dev_type)
458 i = netdev_lock_pos(dev_type);
459 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
460 netdev_lock_name[i]);
463 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
464 unsigned short dev_type)
469 /*******************************************************************************
471 * Protocol management and registration routines
473 *******************************************************************************/
477 * Add a protocol ID to the list. Now that the input handler is
478 * smarter we can dispense with all the messy stuff that used to be
481 * BEWARE!!! Protocol handlers, mangling input packets,
482 * MUST BE last in hash buckets and checking protocol handlers
483 * MUST start from promiscuous ptype_all chain in net_bh.
484 * It is true now, do not change it.
485 * Explanation follows: if protocol handler, mangling packet, will
486 * be the first on list, it is not able to sense, that packet
487 * is cloned and should be copied-on-write, so that it will
488 * change it and subsequent readers will get broken packet.
492 static inline struct list_head *ptype_head(const struct packet_type *pt)
494 if (pt->type == htons(ETH_P_ALL))
495 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
497 return pt->dev ? &pt->dev->ptype_specific :
498 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
502 * dev_add_pack - add packet handler
503 * @pt: packet type declaration
505 * Add a protocol handler to the networking stack. The passed &packet_type
506 * is linked into kernel lists and may not be freed until it has been
507 * removed from the kernel lists.
509 * This call does not sleep therefore it can not
510 * guarantee all CPU's that are in middle of receiving packets
511 * will see the new packet type (until the next received packet).
514 void dev_add_pack(struct packet_type *pt)
516 struct list_head *head = ptype_head(pt);
518 spin_lock(&ptype_lock);
519 list_add_rcu(&pt->list, head);
520 spin_unlock(&ptype_lock);
522 EXPORT_SYMBOL(dev_add_pack);
525 * __dev_remove_pack - remove packet handler
526 * @pt: packet type declaration
528 * Remove a protocol handler that was previously added to the kernel
529 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
530 * from the kernel lists and can be freed or reused once this function
533 * The packet type might still be in use by receivers
534 * and must not be freed until after all the CPU's have gone
535 * through a quiescent state.
537 void __dev_remove_pack(struct packet_type *pt)
539 struct list_head *head = ptype_head(pt);
540 struct packet_type *pt1;
542 spin_lock(&ptype_lock);
544 list_for_each_entry(pt1, head, list) {
546 list_del_rcu(&pt->list);
551 pr_warn("dev_remove_pack: %p not found\n", pt);
553 spin_unlock(&ptype_lock);
555 EXPORT_SYMBOL(__dev_remove_pack);
558 * dev_remove_pack - remove packet handler
559 * @pt: packet type declaration
561 * Remove a protocol handler that was previously added to the kernel
562 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
563 * from the kernel lists and can be freed or reused once this function
566 * This call sleeps to guarantee that no CPU is looking at the packet
569 void dev_remove_pack(struct packet_type *pt)
571 __dev_remove_pack(pt);
575 EXPORT_SYMBOL(dev_remove_pack);
579 * dev_add_offload - register offload handlers
580 * @po: protocol offload declaration
582 * Add protocol offload handlers to the networking stack. The passed
583 * &proto_offload is linked into kernel lists and may not be freed until
584 * it has been removed from the kernel lists.
586 * This call does not sleep therefore it can not
587 * guarantee all CPU's that are in middle of receiving packets
588 * will see the new offload handlers (until the next received packet).
590 void dev_add_offload(struct packet_offload *po)
592 struct packet_offload *elem;
594 spin_lock(&offload_lock);
595 list_for_each_entry(elem, &offload_base, list) {
596 if (po->priority < elem->priority)
599 list_add_rcu(&po->list, elem->list.prev);
600 spin_unlock(&offload_lock);
602 EXPORT_SYMBOL(dev_add_offload);
605 * __dev_remove_offload - remove offload handler
606 * @po: packet offload declaration
608 * Remove a protocol offload handler that was previously added to the
609 * kernel offload handlers by dev_add_offload(). The passed &offload_type
610 * is removed from the kernel lists and can be freed or reused once this
613 * The packet type might still be in use by receivers
614 * and must not be freed until after all the CPU's have gone
615 * through a quiescent state.
617 static void __dev_remove_offload(struct packet_offload *po)
619 struct list_head *head = &offload_base;
620 struct packet_offload *po1;
622 spin_lock(&offload_lock);
624 list_for_each_entry(po1, head, list) {
626 list_del_rcu(&po->list);
631 pr_warn("dev_remove_offload: %p not found\n", po);
633 spin_unlock(&offload_lock);
637 * dev_remove_offload - remove packet offload handler
638 * @po: packet offload declaration
640 * Remove a packet offload handler that was previously added to the kernel
641 * offload handlers by dev_add_offload(). The passed &offload_type is
642 * removed from the kernel lists and can be freed or reused once this
645 * This call sleeps to guarantee that no CPU is looking at the packet
648 void dev_remove_offload(struct packet_offload *po)
650 __dev_remove_offload(po);
654 EXPORT_SYMBOL(dev_remove_offload);
656 /******************************************************************************
658 * Device Boot-time Settings Routines
660 ******************************************************************************/
662 /* Boot time configuration table */
663 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
666 * netdev_boot_setup_add - add new setup entry
667 * @name: name of the device
668 * @map: configured settings for the device
670 * Adds new setup entry to the dev_boot_setup list. The function
671 * returns 0 on error and 1 on success. This is a generic routine to
674 static int netdev_boot_setup_add(char *name, struct ifmap *map)
676 struct netdev_boot_setup *s;
680 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
681 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
682 memset(s[i].name, 0, sizeof(s[i].name));
683 strlcpy(s[i].name, name, IFNAMSIZ);
684 memcpy(&s[i].map, map, sizeof(s[i].map));
689 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
693 * netdev_boot_setup_check - check boot time settings
694 * @dev: the netdevice
696 * Check boot time settings for the device.
697 * The found settings are set for the device to be used
698 * later in the device probing.
699 * Returns 0 if no settings found, 1 if they are.
701 int netdev_boot_setup_check(struct net_device *dev)
703 struct netdev_boot_setup *s = dev_boot_setup;
706 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
707 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
708 !strcmp(dev->name, s[i].name)) {
709 dev->irq = s[i].map.irq;
710 dev->base_addr = s[i].map.base_addr;
711 dev->mem_start = s[i].map.mem_start;
712 dev->mem_end = s[i].map.mem_end;
718 EXPORT_SYMBOL(netdev_boot_setup_check);
722 * netdev_boot_base - get address from boot time settings
723 * @prefix: prefix for network device
724 * @unit: id for network device
726 * Check boot time settings for the base address of device.
727 * The found settings are set for the device to be used
728 * later in the device probing.
729 * Returns 0 if no settings found.
731 unsigned long netdev_boot_base(const char *prefix, int unit)
733 const struct netdev_boot_setup *s = dev_boot_setup;
737 sprintf(name, "%s%d", prefix, unit);
740 * If device already registered then return base of 1
741 * to indicate not to probe for this interface
743 if (__dev_get_by_name(&init_net, name))
746 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
747 if (!strcmp(name, s[i].name))
748 return s[i].map.base_addr;
753 * Saves at boot time configured settings for any netdevice.
755 int __init netdev_boot_setup(char *str)
760 str = get_options(str, ARRAY_SIZE(ints), ints);
765 memset(&map, 0, sizeof(map));
769 map.base_addr = ints[2];
771 map.mem_start = ints[3];
773 map.mem_end = ints[4];
775 /* Add new entry to the list */
776 return netdev_boot_setup_add(str, &map);
779 __setup("netdev=", netdev_boot_setup);
781 /*******************************************************************************
783 * Device Interface Subroutines
785 *******************************************************************************/
788 * dev_get_iflink - get 'iflink' value of a interface
789 * @dev: targeted interface
791 * Indicates the ifindex the interface is linked to.
792 * Physical interfaces have the same 'ifindex' and 'iflink' values.
795 int dev_get_iflink(const struct net_device *dev)
797 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
798 return dev->netdev_ops->ndo_get_iflink(dev);
802 EXPORT_SYMBOL(dev_get_iflink);
805 * dev_fill_metadata_dst - Retrieve tunnel egress information.
806 * @dev: targeted interface
809 * For better visibility of tunnel traffic OVS needs to retrieve
810 * egress tunnel information for a packet. Following API allows
811 * user to get this info.
813 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
815 struct ip_tunnel_info *info;
817 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
820 info = skb_tunnel_info_unclone(skb);
823 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
826 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
828 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
831 * __dev_get_by_name - find a device by its name
832 * @net: the applicable net namespace
833 * @name: name to find
835 * Find an interface by name. Must be called under RTNL semaphore
836 * or @dev_base_lock. If the name is found a pointer to the device
837 * is returned. If the name is not found then %NULL is returned. The
838 * reference counters are not incremented so the caller must be
839 * careful with locks.
842 struct net_device *__dev_get_by_name(struct net *net, const char *name)
844 struct netdev_name_node *node_name;
846 node_name = netdev_name_node_lookup(net, name);
847 return node_name ? node_name->dev : NULL;
849 EXPORT_SYMBOL(__dev_get_by_name);
852 * dev_get_by_name_rcu - find a device by its name
853 * @net: the applicable net namespace
854 * @name: name to find
856 * Find an interface by name.
857 * If the name is found a pointer to the device is returned.
858 * If the name is not found then %NULL is returned.
859 * The reference counters are not incremented so the caller must be
860 * careful with locks. The caller must hold RCU lock.
863 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
865 struct netdev_name_node *node_name;
867 node_name = netdev_name_node_lookup_rcu(net, name);
868 return node_name ? node_name->dev : NULL;
870 EXPORT_SYMBOL(dev_get_by_name_rcu);
873 * dev_get_by_name - find a device by its name
874 * @net: the applicable net namespace
875 * @name: name to find
877 * Find an interface by name. This can be called from any
878 * context and does its own locking. The returned handle has
879 * the usage count incremented and the caller must use dev_put() to
880 * release it when it is no longer needed. %NULL is returned if no
881 * matching device is found.
884 struct net_device *dev_get_by_name(struct net *net, const char *name)
886 struct net_device *dev;
889 dev = dev_get_by_name_rcu(net, name);
895 EXPORT_SYMBOL(dev_get_by_name);
898 * __dev_get_by_index - find a device by its ifindex
899 * @net: the applicable net namespace
900 * @ifindex: index of device
902 * Search for an interface by index. Returns %NULL if the device
903 * is not found or a pointer to the device. The device has not
904 * had its reference counter increased so the caller must be careful
905 * about locking. The caller must hold either the RTNL semaphore
909 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
911 struct net_device *dev;
912 struct hlist_head *head = dev_index_hash(net, ifindex);
914 hlist_for_each_entry(dev, head, index_hlist)
915 if (dev->ifindex == ifindex)
920 EXPORT_SYMBOL(__dev_get_by_index);
923 * dev_get_by_index_rcu - find a device by its ifindex
924 * @net: the applicable net namespace
925 * @ifindex: index of device
927 * Search for an interface by index. Returns %NULL if the device
928 * is not found or a pointer to the device. The device has not
929 * had its reference counter increased so the caller must be careful
930 * about locking. The caller must hold RCU lock.
933 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
935 struct net_device *dev;
936 struct hlist_head *head = dev_index_hash(net, ifindex);
938 hlist_for_each_entry_rcu(dev, head, index_hlist)
939 if (dev->ifindex == ifindex)
944 EXPORT_SYMBOL(dev_get_by_index_rcu);
948 * dev_get_by_index - find a device by its ifindex
949 * @net: the applicable net namespace
950 * @ifindex: index of device
952 * Search for an interface by index. Returns NULL if the device
953 * is not found or a pointer to the device. The device returned has
954 * had a reference added and the pointer is safe until the user calls
955 * dev_put to indicate they have finished with it.
958 struct net_device *dev_get_by_index(struct net *net, int ifindex)
960 struct net_device *dev;
963 dev = dev_get_by_index_rcu(net, ifindex);
969 EXPORT_SYMBOL(dev_get_by_index);
972 * dev_get_by_napi_id - find a device by napi_id
973 * @napi_id: ID of the NAPI struct
975 * Search for an interface by NAPI ID. Returns %NULL if the device
976 * is not found or a pointer to the device. The device has not had
977 * its reference counter increased so the caller must be careful
978 * about locking. The caller must hold RCU lock.
981 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
983 struct napi_struct *napi;
985 WARN_ON_ONCE(!rcu_read_lock_held());
987 if (napi_id < MIN_NAPI_ID)
990 napi = napi_by_id(napi_id);
992 return napi ? napi->dev : NULL;
994 EXPORT_SYMBOL(dev_get_by_napi_id);
997 * netdev_get_name - get a netdevice name, knowing its ifindex.
998 * @net: network namespace
999 * @name: a pointer to the buffer where the name will be stored.
1000 * @ifindex: the ifindex of the interface to get the name from.
1002 * The use of raw_seqcount_begin() and cond_resched() before
1003 * retrying is required as we want to give the writers a chance
1004 * to complete when CONFIG_PREEMPTION is not set.
1006 int netdev_get_name(struct net *net, char *name, int ifindex)
1008 struct net_device *dev;
1012 seq = raw_seqcount_begin(&devnet_rename_seq);
1014 dev = dev_get_by_index_rcu(net, ifindex);
1020 strcpy(name, dev->name);
1022 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
1031 * dev_getbyhwaddr_rcu - find a device by its hardware address
1032 * @net: the applicable net namespace
1033 * @type: media type of device
1034 * @ha: hardware address
1036 * Search for an interface by MAC address. Returns NULL if the device
1037 * is not found or a pointer to the device.
1038 * The caller must hold RCU or RTNL.
1039 * The returned device has not had its ref count increased
1040 * and the caller must therefore be careful about locking
1044 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1047 struct net_device *dev;
1049 for_each_netdev_rcu(net, dev)
1050 if (dev->type == type &&
1051 !memcmp(dev->dev_addr, ha, dev->addr_len))
1056 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1058 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1060 struct net_device *dev;
1063 for_each_netdev(net, dev)
1064 if (dev->type == type)
1069 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
1071 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1073 struct net_device *dev, *ret = NULL;
1076 for_each_netdev_rcu(net, dev)
1077 if (dev->type == type) {
1085 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1088 * __dev_get_by_flags - find any device with given flags
1089 * @net: the applicable net namespace
1090 * @if_flags: IFF_* values
1091 * @mask: bitmask of bits in if_flags to check
1093 * Search for any interface with the given flags. Returns NULL if a device
1094 * is not found or a pointer to the device. Must be called inside
1095 * rtnl_lock(), and result refcount is unchanged.
1098 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1099 unsigned short mask)
1101 struct net_device *dev, *ret;
1106 for_each_netdev(net, dev) {
1107 if (((dev->flags ^ if_flags) & mask) == 0) {
1114 EXPORT_SYMBOL(__dev_get_by_flags);
1117 * dev_valid_name - check if name is okay for network device
1118 * @name: name string
1120 * Network device names need to be valid file names to
1121 * to allow sysfs to work. We also disallow any kind of
1124 bool dev_valid_name(const char *name)
1128 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1130 if (!strcmp(name, ".") || !strcmp(name, ".."))
1134 if (*name == '/' || *name == ':' || isspace(*name))
1140 EXPORT_SYMBOL(dev_valid_name);
1143 * __dev_alloc_name - allocate a name for a device
1144 * @net: network namespace to allocate the device name in
1145 * @name: name format string
1146 * @buf: scratch buffer and result name string
1148 * Passed a format string - eg "lt%d" it will try and find a suitable
1149 * id. It scans list of devices to build up a free map, then chooses
1150 * the first empty slot. The caller must hold the dev_base or rtnl lock
1151 * while allocating the name and adding the device in order to avoid
1153 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1154 * Returns the number of the unit assigned or a negative errno code.
1157 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1161 const int max_netdevices = 8*PAGE_SIZE;
1162 unsigned long *inuse;
1163 struct net_device *d;
1165 if (!dev_valid_name(name))
1168 p = strchr(name, '%');
1171 * Verify the string as this thing may have come from
1172 * the user. There must be either one "%d" and no other "%"
1175 if (p[1] != 'd' || strchr(p + 2, '%'))
1178 /* Use one page as a bit array of possible slots */
1179 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1183 for_each_netdev(net, d) {
1184 if (!sscanf(d->name, name, &i))
1186 if (i < 0 || i >= max_netdevices)
1189 /* avoid cases where sscanf is not exact inverse of printf */
1190 snprintf(buf, IFNAMSIZ, name, i);
1191 if (!strncmp(buf, d->name, IFNAMSIZ))
1195 i = find_first_zero_bit(inuse, max_netdevices);
1196 free_page((unsigned long) inuse);
1199 snprintf(buf, IFNAMSIZ, name, i);
1200 if (!__dev_get_by_name(net, buf))
1203 /* It is possible to run out of possible slots
1204 * when the name is long and there isn't enough space left
1205 * for the digits, or if all bits are used.
1210 static int dev_alloc_name_ns(struct net *net,
1211 struct net_device *dev,
1218 ret = __dev_alloc_name(net, name, buf);
1220 strlcpy(dev->name, buf, IFNAMSIZ);
1225 * dev_alloc_name - allocate a name for a device
1227 * @name: name format string
1229 * Passed a format string - eg "lt%d" it will try and find a suitable
1230 * id. It scans list of devices to build up a free map, then chooses
1231 * the first empty slot. The caller must hold the dev_base or rtnl lock
1232 * while allocating the name and adding the device in order to avoid
1234 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1235 * Returns the number of the unit assigned or a negative errno code.
1238 int dev_alloc_name(struct net_device *dev, const char *name)
1240 return dev_alloc_name_ns(dev_net(dev), dev, name);
1242 EXPORT_SYMBOL(dev_alloc_name);
1244 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1249 if (!dev_valid_name(name))
1252 if (strchr(name, '%'))
1253 return dev_alloc_name_ns(net, dev, name);
1254 else if (__dev_get_by_name(net, name))
1256 else if (dev->name != name)
1257 strlcpy(dev->name, name, IFNAMSIZ);
1263 * dev_change_name - change name of a device
1265 * @newname: name (or format string) must be at least IFNAMSIZ
1267 * Change name of a device, can pass format strings "eth%d".
1270 int dev_change_name(struct net_device *dev, const char *newname)
1272 unsigned char old_assign_type;
1273 char oldname[IFNAMSIZ];
1279 BUG_ON(!dev_net(dev));
1283 /* Some auto-enslaved devices e.g. failover slaves are
1284 * special, as userspace might rename the device after
1285 * the interface had been brought up and running since
1286 * the point kernel initiated auto-enslavement. Allow
1287 * live name change even when these slave devices are
1290 * Typically, users of these auto-enslaving devices
1291 * don't actually care about slave name change, as
1292 * they are supposed to operate on master interface
1295 if (dev->flags & IFF_UP &&
1296 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1299 write_seqcount_begin(&devnet_rename_seq);
1301 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1302 write_seqcount_end(&devnet_rename_seq);
1306 memcpy(oldname, dev->name, IFNAMSIZ);
1308 err = dev_get_valid_name(net, dev, newname);
1310 write_seqcount_end(&devnet_rename_seq);
1314 if (oldname[0] && !strchr(oldname, '%'))
1315 netdev_info(dev, "renamed from %s\n", oldname);
1317 old_assign_type = dev->name_assign_type;
1318 dev->name_assign_type = NET_NAME_RENAMED;
1321 ret = device_rename(&dev->dev, dev->name);
1323 memcpy(dev->name, oldname, IFNAMSIZ);
1324 dev->name_assign_type = old_assign_type;
1325 write_seqcount_end(&devnet_rename_seq);
1329 write_seqcount_end(&devnet_rename_seq);
1331 netdev_adjacent_rename_links(dev, oldname);
1333 write_lock_bh(&dev_base_lock);
1334 netdev_name_node_del(dev->name_node);
1335 write_unlock_bh(&dev_base_lock);
1339 write_lock_bh(&dev_base_lock);
1340 netdev_name_node_add(net, dev->name_node);
1341 write_unlock_bh(&dev_base_lock);
1343 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1344 ret = notifier_to_errno(ret);
1347 /* err >= 0 after dev_alloc_name() or stores the first errno */
1350 write_seqcount_begin(&devnet_rename_seq);
1351 memcpy(dev->name, oldname, IFNAMSIZ);
1352 memcpy(oldname, newname, IFNAMSIZ);
1353 dev->name_assign_type = old_assign_type;
1354 old_assign_type = NET_NAME_RENAMED;
1357 pr_err("%s: name change rollback failed: %d\n",
1366 * dev_set_alias - change ifalias of a device
1368 * @alias: name up to IFALIASZ
1369 * @len: limit of bytes to copy from info
1371 * Set ifalias for a device,
1373 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1375 struct dev_ifalias *new_alias = NULL;
1377 if (len >= IFALIASZ)
1381 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1385 memcpy(new_alias->ifalias, alias, len);
1386 new_alias->ifalias[len] = 0;
1389 mutex_lock(&ifalias_mutex);
1390 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1391 mutex_is_locked(&ifalias_mutex));
1392 mutex_unlock(&ifalias_mutex);
1395 kfree_rcu(new_alias, rcuhead);
1399 EXPORT_SYMBOL(dev_set_alias);
1402 * dev_get_alias - get ifalias of a device
1404 * @name: buffer to store name of ifalias
1405 * @len: size of buffer
1407 * get ifalias for a device. Caller must make sure dev cannot go
1408 * away, e.g. rcu read lock or own a reference count to device.
1410 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1412 const struct dev_ifalias *alias;
1416 alias = rcu_dereference(dev->ifalias);
1418 ret = snprintf(name, len, "%s", alias->ifalias);
1425 * netdev_features_change - device changes features
1426 * @dev: device to cause notification
1428 * Called to indicate a device has changed features.
1430 void netdev_features_change(struct net_device *dev)
1432 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1434 EXPORT_SYMBOL(netdev_features_change);
1437 * netdev_state_change - device changes state
1438 * @dev: device to cause notification
1440 * Called to indicate a device has changed state. This function calls
1441 * the notifier chains for netdev_chain and sends a NEWLINK message
1442 * to the routing socket.
1444 void netdev_state_change(struct net_device *dev)
1446 if (dev->flags & IFF_UP) {
1447 struct netdev_notifier_change_info change_info = {
1451 call_netdevice_notifiers_info(NETDEV_CHANGE,
1453 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1456 EXPORT_SYMBOL(netdev_state_change);
1459 * netdev_notify_peers - notify network peers about existence of @dev
1460 * @dev: network device
1462 * Generate traffic such that interested network peers are aware of
1463 * @dev, such as by generating a gratuitous ARP. This may be used when
1464 * a device wants to inform the rest of the network about some sort of
1465 * reconfiguration such as a failover event or virtual machine
1468 void netdev_notify_peers(struct net_device *dev)
1471 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1472 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1475 EXPORT_SYMBOL(netdev_notify_peers);
1477 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1479 const struct net_device_ops *ops = dev->netdev_ops;
1484 if (!netif_device_present(dev))
1487 /* Block netpoll from trying to do any rx path servicing.
1488 * If we don't do this there is a chance ndo_poll_controller
1489 * or ndo_poll may be running while we open the device
1491 netpoll_poll_disable(dev);
1493 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1494 ret = notifier_to_errno(ret);
1498 set_bit(__LINK_STATE_START, &dev->state);
1500 if (ops->ndo_validate_addr)
1501 ret = ops->ndo_validate_addr(dev);
1503 if (!ret && ops->ndo_open)
1504 ret = ops->ndo_open(dev);
1506 netpoll_poll_enable(dev);
1509 clear_bit(__LINK_STATE_START, &dev->state);
1511 dev->flags |= IFF_UP;
1512 dev_set_rx_mode(dev);
1514 add_device_randomness(dev->dev_addr, dev->addr_len);
1521 * dev_open - prepare an interface for use.
1522 * @dev: device to open
1523 * @extack: netlink extended ack
1525 * Takes a device from down to up state. The device's private open
1526 * function is invoked and then the multicast lists are loaded. Finally
1527 * the device is moved into the up state and a %NETDEV_UP message is
1528 * sent to the netdev notifier chain.
1530 * Calling this function on an active interface is a nop. On a failure
1531 * a negative errno code is returned.
1533 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1537 if (dev->flags & IFF_UP)
1540 ret = __dev_open(dev, extack);
1544 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1545 call_netdevice_notifiers(NETDEV_UP, dev);
1549 EXPORT_SYMBOL(dev_open);
1551 static void __dev_close_many(struct list_head *head)
1553 struct net_device *dev;
1558 list_for_each_entry(dev, head, close_list) {
1559 /* Temporarily disable netpoll until the interface is down */
1560 netpoll_poll_disable(dev);
1562 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1564 clear_bit(__LINK_STATE_START, &dev->state);
1566 /* Synchronize to scheduled poll. We cannot touch poll list, it
1567 * can be even on different cpu. So just clear netif_running().
1569 * dev->stop() will invoke napi_disable() on all of it's
1570 * napi_struct instances on this device.
1572 smp_mb__after_atomic(); /* Commit netif_running(). */
1575 dev_deactivate_many(head);
1577 list_for_each_entry(dev, head, close_list) {
1578 const struct net_device_ops *ops = dev->netdev_ops;
1581 * Call the device specific close. This cannot fail.
1582 * Only if device is UP
1584 * We allow it to be called even after a DETACH hot-plug
1590 dev->flags &= ~IFF_UP;
1591 netpoll_poll_enable(dev);
1595 static void __dev_close(struct net_device *dev)
1599 list_add(&dev->close_list, &single);
1600 __dev_close_many(&single);
1604 void dev_close_many(struct list_head *head, bool unlink)
1606 struct net_device *dev, *tmp;
1608 /* Remove the devices that don't need to be closed */
1609 list_for_each_entry_safe(dev, tmp, head, close_list)
1610 if (!(dev->flags & IFF_UP))
1611 list_del_init(&dev->close_list);
1613 __dev_close_many(head);
1615 list_for_each_entry_safe(dev, tmp, head, close_list) {
1616 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1617 call_netdevice_notifiers(NETDEV_DOWN, dev);
1619 list_del_init(&dev->close_list);
1622 EXPORT_SYMBOL(dev_close_many);
1625 * dev_close - shutdown an interface.
1626 * @dev: device to shutdown
1628 * This function moves an active device into down state. A
1629 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1630 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1633 void dev_close(struct net_device *dev)
1635 if (dev->flags & IFF_UP) {
1638 list_add(&dev->close_list, &single);
1639 dev_close_many(&single, true);
1643 EXPORT_SYMBOL(dev_close);
1647 * dev_disable_lro - disable Large Receive Offload on a device
1650 * Disable Large Receive Offload (LRO) on a net device. Must be
1651 * called under RTNL. This is needed if received packets may be
1652 * forwarded to another interface.
1654 void dev_disable_lro(struct net_device *dev)
1656 struct net_device *lower_dev;
1657 struct list_head *iter;
1659 dev->wanted_features &= ~NETIF_F_LRO;
1660 netdev_update_features(dev);
1662 if (unlikely(dev->features & NETIF_F_LRO))
1663 netdev_WARN(dev, "failed to disable LRO!\n");
1665 netdev_for_each_lower_dev(dev, lower_dev, iter)
1666 dev_disable_lro(lower_dev);
1668 EXPORT_SYMBOL(dev_disable_lro);
1671 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1674 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1675 * called under RTNL. This is needed if Generic XDP is installed on
1678 static void dev_disable_gro_hw(struct net_device *dev)
1680 dev->wanted_features &= ~NETIF_F_GRO_HW;
1681 netdev_update_features(dev);
1683 if (unlikely(dev->features & NETIF_F_GRO_HW))
1684 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1687 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1690 case NETDEV_##val: \
1691 return "NETDEV_" __stringify(val);
1693 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1694 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1695 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1696 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1697 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1698 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1699 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1700 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1701 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1705 return "UNKNOWN_NETDEV_EVENT";
1707 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1709 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1710 struct net_device *dev)
1712 struct netdev_notifier_info info = {
1716 return nb->notifier_call(nb, val, &info);
1719 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1720 struct net_device *dev)
1724 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1725 err = notifier_to_errno(err);
1729 if (!(dev->flags & IFF_UP))
1732 call_netdevice_notifier(nb, NETDEV_UP, dev);
1736 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1737 struct net_device *dev)
1739 if (dev->flags & IFF_UP) {
1740 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1742 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1744 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1747 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1750 struct net_device *dev;
1753 for_each_netdev(net, dev) {
1754 err = call_netdevice_register_notifiers(nb, dev);
1761 for_each_netdev_continue_reverse(net, dev)
1762 call_netdevice_unregister_notifiers(nb, dev);
1766 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1769 struct net_device *dev;
1771 for_each_netdev(net, dev)
1772 call_netdevice_unregister_notifiers(nb, dev);
1775 static int dev_boot_phase = 1;
1778 * register_netdevice_notifier - register a network notifier block
1781 * Register a notifier to be called when network device events occur.
1782 * The notifier passed is linked into the kernel structures and must
1783 * not be reused until it has been unregistered. A negative errno code
1784 * is returned on a failure.
1786 * When registered all registration and up events are replayed
1787 * to the new notifier to allow device to have a race free
1788 * view of the network device list.
1791 int register_netdevice_notifier(struct notifier_block *nb)
1796 /* Close race with setup_net() and cleanup_net() */
1797 down_write(&pernet_ops_rwsem);
1799 err = raw_notifier_chain_register(&netdev_chain, nb);
1805 err = call_netdevice_register_net_notifiers(nb, net);
1812 up_write(&pernet_ops_rwsem);
1816 for_each_net_continue_reverse(net)
1817 call_netdevice_unregister_net_notifiers(nb, net);
1819 raw_notifier_chain_unregister(&netdev_chain, nb);
1822 EXPORT_SYMBOL(register_netdevice_notifier);
1825 * unregister_netdevice_notifier - unregister a network notifier block
1828 * Unregister a notifier previously registered by
1829 * register_netdevice_notifier(). The notifier is unlinked into the
1830 * kernel structures and may then be reused. A negative errno code
1831 * is returned on a failure.
1833 * After unregistering unregister and down device events are synthesized
1834 * for all devices on the device list to the removed notifier to remove
1835 * the need for special case cleanup code.
1838 int unregister_netdevice_notifier(struct notifier_block *nb)
1843 /* Close race with setup_net() and cleanup_net() */
1844 down_write(&pernet_ops_rwsem);
1846 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1851 call_netdevice_unregister_net_notifiers(nb, net);
1855 up_write(&pernet_ops_rwsem);
1858 EXPORT_SYMBOL(unregister_netdevice_notifier);
1860 static int __register_netdevice_notifier_net(struct net *net,
1861 struct notifier_block *nb,
1862 bool ignore_call_fail)
1866 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1872 err = call_netdevice_register_net_notifiers(nb, net);
1873 if (err && !ignore_call_fail)
1874 goto chain_unregister;
1879 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1883 static int __unregister_netdevice_notifier_net(struct net *net,
1884 struct notifier_block *nb)
1888 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1892 call_netdevice_unregister_net_notifiers(nb, net);
1897 * register_netdevice_notifier_net - register a per-netns network notifier block
1898 * @net: network namespace
1901 * Register a notifier to be called when network device events occur.
1902 * The notifier passed is linked into the kernel structures and must
1903 * not be reused until it has been unregistered. A negative errno code
1904 * is returned on a failure.
1906 * When registered all registration and up events are replayed
1907 * to the new notifier to allow device to have a race free
1908 * view of the network device list.
1911 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1916 err = __register_netdevice_notifier_net(net, nb, false);
1920 EXPORT_SYMBOL(register_netdevice_notifier_net);
1923 * unregister_netdevice_notifier_net - unregister a per-netns
1924 * network notifier block
1925 * @net: network namespace
1928 * Unregister a notifier previously registered by
1929 * register_netdevice_notifier(). The notifier is unlinked into the
1930 * kernel structures and may then be reused. A negative errno code
1931 * is returned on a failure.
1933 * After unregistering unregister and down device events are synthesized
1934 * for all devices on the device list to the removed notifier to remove
1935 * the need for special case cleanup code.
1938 int unregister_netdevice_notifier_net(struct net *net,
1939 struct notifier_block *nb)
1944 err = __unregister_netdevice_notifier_net(net, nb);
1948 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1950 int register_netdevice_notifier_dev_net(struct net_device *dev,
1951 struct notifier_block *nb,
1952 struct netdev_net_notifier *nn)
1957 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1960 list_add(&nn->list, &dev->net_notifier_list);
1965 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1967 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1968 struct notifier_block *nb,
1969 struct netdev_net_notifier *nn)
1974 list_del(&nn->list);
1975 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1979 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1981 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1984 struct netdev_net_notifier *nn;
1986 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1987 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1988 __register_netdevice_notifier_net(net, nn->nb, true);
1993 * call_netdevice_notifiers_info - call all network notifier blocks
1994 * @val: value passed unmodified to notifier function
1995 * @info: notifier information data
1997 * Call all network notifier blocks. Parameters and return value
1998 * are as for raw_notifier_call_chain().
2001 static int call_netdevice_notifiers_info(unsigned long val,
2002 struct netdev_notifier_info *info)
2004 struct net *net = dev_net(info->dev);
2009 /* Run per-netns notifier block chain first, then run the global one.
2010 * Hopefully, one day, the global one is going to be removed after
2011 * all notifier block registrators get converted to be per-netns.
2013 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2014 if (ret & NOTIFY_STOP_MASK)
2016 return raw_notifier_call_chain(&netdev_chain, val, info);
2019 static int call_netdevice_notifiers_extack(unsigned long val,
2020 struct net_device *dev,
2021 struct netlink_ext_ack *extack)
2023 struct netdev_notifier_info info = {
2028 return call_netdevice_notifiers_info(val, &info);
2032 * call_netdevice_notifiers - call all network notifier blocks
2033 * @val: value passed unmodified to notifier function
2034 * @dev: net_device pointer passed unmodified to notifier function
2036 * Call all network notifier blocks. Parameters and return value
2037 * are as for raw_notifier_call_chain().
2040 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2042 return call_netdevice_notifiers_extack(val, dev, NULL);
2044 EXPORT_SYMBOL(call_netdevice_notifiers);
2047 * call_netdevice_notifiers_mtu - call all network notifier blocks
2048 * @val: value passed unmodified to notifier function
2049 * @dev: net_device pointer passed unmodified to notifier function
2050 * @arg: additional u32 argument passed to the notifier function
2052 * Call all network notifier blocks. Parameters and return value
2053 * are as for raw_notifier_call_chain().
2055 static int call_netdevice_notifiers_mtu(unsigned long val,
2056 struct net_device *dev, u32 arg)
2058 struct netdev_notifier_info_ext info = {
2063 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2065 return call_netdevice_notifiers_info(val, &info.info);
2068 #ifdef CONFIG_NET_INGRESS
2069 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2071 void net_inc_ingress_queue(void)
2073 static_branch_inc(&ingress_needed_key);
2075 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2077 void net_dec_ingress_queue(void)
2079 static_branch_dec(&ingress_needed_key);
2081 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2084 #ifdef CONFIG_NET_EGRESS
2085 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2087 void net_inc_egress_queue(void)
2089 static_branch_inc(&egress_needed_key);
2091 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2093 void net_dec_egress_queue(void)
2095 static_branch_dec(&egress_needed_key);
2097 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2100 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2101 #ifdef CONFIG_JUMP_LABEL
2102 static atomic_t netstamp_needed_deferred;
2103 static atomic_t netstamp_wanted;
2104 static void netstamp_clear(struct work_struct *work)
2106 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2109 wanted = atomic_add_return(deferred, &netstamp_wanted);
2111 static_branch_enable(&netstamp_needed_key);
2113 static_branch_disable(&netstamp_needed_key);
2115 static DECLARE_WORK(netstamp_work, netstamp_clear);
2118 void net_enable_timestamp(void)
2120 #ifdef CONFIG_JUMP_LABEL
2124 wanted = atomic_read(&netstamp_wanted);
2127 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2130 atomic_inc(&netstamp_needed_deferred);
2131 schedule_work(&netstamp_work);
2133 static_branch_inc(&netstamp_needed_key);
2136 EXPORT_SYMBOL(net_enable_timestamp);
2138 void net_disable_timestamp(void)
2140 #ifdef CONFIG_JUMP_LABEL
2144 wanted = atomic_read(&netstamp_wanted);
2147 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2150 atomic_dec(&netstamp_needed_deferred);
2151 schedule_work(&netstamp_work);
2153 static_branch_dec(&netstamp_needed_key);
2156 EXPORT_SYMBOL(net_disable_timestamp);
2158 static inline void net_timestamp_set(struct sk_buff *skb)
2161 if (static_branch_unlikely(&netstamp_needed_key))
2162 __net_timestamp(skb);
2165 #define net_timestamp_check(COND, SKB) \
2166 if (static_branch_unlikely(&netstamp_needed_key)) { \
2167 if ((COND) && !(SKB)->tstamp) \
2168 __net_timestamp(SKB); \
2171 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2175 if (!(dev->flags & IFF_UP))
2178 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2179 if (skb->len <= len)
2182 /* if TSO is enabled, we don't care about the length as the packet
2183 * could be forwarded without being segmented before
2185 if (skb_is_gso(skb))
2190 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2192 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2194 int ret = ____dev_forward_skb(dev, skb);
2197 skb->protocol = eth_type_trans(skb, dev);
2198 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2203 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2206 * dev_forward_skb - loopback an skb to another netif
2208 * @dev: destination network device
2209 * @skb: buffer to forward
2212 * NET_RX_SUCCESS (no congestion)
2213 * NET_RX_DROP (packet was dropped, but freed)
2215 * dev_forward_skb can be used for injecting an skb from the
2216 * start_xmit function of one device into the receive queue
2217 * of another device.
2219 * The receiving device may be in another namespace, so
2220 * we have to clear all information in the skb that could
2221 * impact namespace isolation.
2223 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2225 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2227 EXPORT_SYMBOL_GPL(dev_forward_skb);
2229 static inline int deliver_skb(struct sk_buff *skb,
2230 struct packet_type *pt_prev,
2231 struct net_device *orig_dev)
2233 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2235 refcount_inc(&skb->users);
2236 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2239 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2240 struct packet_type **pt,
2241 struct net_device *orig_dev,
2243 struct list_head *ptype_list)
2245 struct packet_type *ptype, *pt_prev = *pt;
2247 list_for_each_entry_rcu(ptype, ptype_list, list) {
2248 if (ptype->type != type)
2251 deliver_skb(skb, pt_prev, orig_dev);
2257 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2259 if (!ptype->af_packet_priv || !skb->sk)
2262 if (ptype->id_match)
2263 return ptype->id_match(ptype, skb->sk);
2264 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2271 * dev_nit_active - return true if any network interface taps are in use
2273 * @dev: network device to check for the presence of taps
2275 bool dev_nit_active(struct net_device *dev)
2277 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2279 EXPORT_SYMBOL_GPL(dev_nit_active);
2282 * Support routine. Sends outgoing frames to any network
2283 * taps currently in use.
2286 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2288 struct packet_type *ptype;
2289 struct sk_buff *skb2 = NULL;
2290 struct packet_type *pt_prev = NULL;
2291 struct list_head *ptype_list = &ptype_all;
2295 list_for_each_entry_rcu(ptype, ptype_list, list) {
2296 if (ptype->ignore_outgoing)
2299 /* Never send packets back to the socket
2300 * they originated from - MvS (miquels@drinkel.ow.org)
2302 if (skb_loop_sk(ptype, skb))
2306 deliver_skb(skb2, pt_prev, skb->dev);
2311 /* need to clone skb, done only once */
2312 skb2 = skb_clone(skb, GFP_ATOMIC);
2316 net_timestamp_set(skb2);
2318 /* skb->nh should be correctly
2319 * set by sender, so that the second statement is
2320 * just protection against buggy protocols.
2322 skb_reset_mac_header(skb2);
2324 if (skb_network_header(skb2) < skb2->data ||
2325 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2326 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2327 ntohs(skb2->protocol),
2329 skb_reset_network_header(skb2);
2332 skb2->transport_header = skb2->network_header;
2333 skb2->pkt_type = PACKET_OUTGOING;
2337 if (ptype_list == &ptype_all) {
2338 ptype_list = &dev->ptype_all;
2343 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2344 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2350 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2353 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2354 * @dev: Network device
2355 * @txq: number of queues available
2357 * If real_num_tx_queues is changed the tc mappings may no longer be
2358 * valid. To resolve this verify the tc mapping remains valid and if
2359 * not NULL the mapping. With no priorities mapping to this
2360 * offset/count pair it will no longer be used. In the worst case TC0
2361 * is invalid nothing can be done so disable priority mappings. If is
2362 * expected that drivers will fix this mapping if they can before
2363 * calling netif_set_real_num_tx_queues.
2365 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2368 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2370 /* If TC0 is invalidated disable TC mapping */
2371 if (tc->offset + tc->count > txq) {
2372 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2377 /* Invalidated prio to tc mappings set to TC0 */
2378 for (i = 1; i < TC_BITMASK + 1; i++) {
2379 int q = netdev_get_prio_tc_map(dev, i);
2381 tc = &dev->tc_to_txq[q];
2382 if (tc->offset + tc->count > txq) {
2383 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2385 netdev_set_prio_tc_map(dev, i, 0);
2390 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2393 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2396 /* walk through the TCs and see if it falls into any of them */
2397 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2398 if ((txq - tc->offset) < tc->count)
2402 /* didn't find it, just return -1 to indicate no match */
2408 EXPORT_SYMBOL(netdev_txq_to_tc);
2411 struct static_key xps_needed __read_mostly;
2412 EXPORT_SYMBOL(xps_needed);
2413 struct static_key xps_rxqs_needed __read_mostly;
2414 EXPORT_SYMBOL(xps_rxqs_needed);
2415 static DEFINE_MUTEX(xps_map_mutex);
2416 #define xmap_dereference(P) \
2417 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2419 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2422 struct xps_map *map = NULL;
2426 map = xmap_dereference(dev_maps->attr_map[tci]);
2430 for (pos = map->len; pos--;) {
2431 if (map->queues[pos] != index)
2435 map->queues[pos] = map->queues[--map->len];
2439 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2440 kfree_rcu(map, rcu);
2447 static bool remove_xps_queue_cpu(struct net_device *dev,
2448 struct xps_dev_maps *dev_maps,
2449 int cpu, u16 offset, u16 count)
2451 int num_tc = dev->num_tc ? : 1;
2452 bool active = false;
2455 for (tci = cpu * num_tc; num_tc--; tci++) {
2458 for (i = count, j = offset; i--; j++) {
2459 if (!remove_xps_queue(dev_maps, tci, j))
2469 static void reset_xps_maps(struct net_device *dev,
2470 struct xps_dev_maps *dev_maps,
2474 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2475 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2477 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2479 static_key_slow_dec_cpuslocked(&xps_needed);
2480 kfree_rcu(dev_maps, rcu);
2483 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2484 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2485 u16 offset, u16 count, bool is_rxqs_map)
2487 bool active = false;
2490 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2492 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2495 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2498 for (i = offset + (count - 1); count--; i--) {
2499 netdev_queue_numa_node_write(
2500 netdev_get_tx_queue(dev, i),
2506 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2509 const unsigned long *possible_mask = NULL;
2510 struct xps_dev_maps *dev_maps;
2511 unsigned int nr_ids;
2513 if (!static_key_false(&xps_needed))
2517 mutex_lock(&xps_map_mutex);
2519 if (static_key_false(&xps_rxqs_needed)) {
2520 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2522 nr_ids = dev->num_rx_queues;
2523 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2524 offset, count, true);
2528 dev_maps = xmap_dereference(dev->xps_cpus_map);
2532 if (num_possible_cpus() > 1)
2533 possible_mask = cpumask_bits(cpu_possible_mask);
2534 nr_ids = nr_cpu_ids;
2535 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2539 mutex_unlock(&xps_map_mutex);
2543 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2545 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2548 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2549 u16 index, bool is_rxqs_map)
2551 struct xps_map *new_map;
2552 int alloc_len = XPS_MIN_MAP_ALLOC;
2555 for (pos = 0; map && pos < map->len; pos++) {
2556 if (map->queues[pos] != index)
2561 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2563 if (pos < map->alloc_len)
2566 alloc_len = map->alloc_len * 2;
2569 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2573 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2575 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2576 cpu_to_node(attr_index));
2580 for (i = 0; i < pos; i++)
2581 new_map->queues[i] = map->queues[i];
2582 new_map->alloc_len = alloc_len;
2588 /* Must be called under cpus_read_lock */
2589 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2590 u16 index, bool is_rxqs_map)
2592 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2593 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2594 int i, j, tci, numa_node_id = -2;
2595 int maps_sz, num_tc = 1, tc = 0;
2596 struct xps_map *map, *new_map;
2597 bool active = false;
2598 unsigned int nr_ids;
2601 /* Do not allow XPS on subordinate device directly */
2602 num_tc = dev->num_tc;
2606 /* If queue belongs to subordinate dev use its map */
2607 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2609 tc = netdev_txq_to_tc(dev, index);
2614 mutex_lock(&xps_map_mutex);
2616 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2617 dev_maps = xmap_dereference(dev->xps_rxqs_map);
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 possible_mask = cpumask_bits(cpu_possible_mask);
2625 dev_maps = xmap_dereference(dev->xps_cpus_map);
2626 nr_ids = nr_cpu_ids;
2629 if (maps_sz < L1_CACHE_BYTES)
2630 maps_sz = L1_CACHE_BYTES;
2632 /* allocate memory for queue storage */
2633 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2636 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2637 if (!new_dev_maps) {
2638 mutex_unlock(&xps_map_mutex);
2642 tci = j * num_tc + tc;
2643 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2646 map = expand_xps_map(map, j, index, is_rxqs_map);
2650 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2654 goto out_no_new_maps;
2657 /* Increment static keys at most once per type */
2658 static_key_slow_inc_cpuslocked(&xps_needed);
2660 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2663 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2665 /* copy maps belonging to foreign traffic classes */
2666 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2667 /* fill in the new device map from the old device map */
2668 map = xmap_dereference(dev_maps->attr_map[tci]);
2669 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2672 /* We need to explicitly update tci as prevous loop
2673 * could break out early if dev_maps is NULL.
2675 tci = j * num_tc + tc;
2677 if (netif_attr_test_mask(j, mask, nr_ids) &&
2678 netif_attr_test_online(j, online_mask, nr_ids)) {
2679 /* add tx-queue to CPU/rx-queue maps */
2682 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2683 while ((pos < map->len) && (map->queues[pos] != index))
2686 if (pos == map->len)
2687 map->queues[map->len++] = index;
2690 if (numa_node_id == -2)
2691 numa_node_id = cpu_to_node(j);
2692 else if (numa_node_id != cpu_to_node(j))
2696 } else if (dev_maps) {
2697 /* fill in the new device map from the old device map */
2698 map = xmap_dereference(dev_maps->attr_map[tci]);
2699 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2702 /* copy maps belonging to foreign traffic classes */
2703 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2704 /* fill in the new device map from the old device map */
2705 map = xmap_dereference(dev_maps->attr_map[tci]);
2706 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2711 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2713 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2715 /* Cleanup old maps */
2717 goto out_no_old_maps;
2719 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2721 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2722 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2723 map = xmap_dereference(dev_maps->attr_map[tci]);
2724 if (map && map != new_map)
2725 kfree_rcu(map, rcu);
2729 kfree_rcu(dev_maps, rcu);
2732 dev_maps = new_dev_maps;
2737 /* update Tx queue numa node */
2738 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2739 (numa_node_id >= 0) ?
2740 numa_node_id : NUMA_NO_NODE);
2746 /* removes tx-queue from unused CPUs/rx-queues */
2747 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2749 for (i = tc, tci = j * num_tc; i--; tci++)
2750 active |= remove_xps_queue(dev_maps, tci, index);
2751 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2752 !netif_attr_test_online(j, online_mask, nr_ids))
2753 active |= remove_xps_queue(dev_maps, tci, index);
2754 for (i = num_tc - tc, tci++; --i; tci++)
2755 active |= remove_xps_queue(dev_maps, tci, index);
2758 /* free map if not active */
2760 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2763 mutex_unlock(&xps_map_mutex);
2767 /* remove any maps that we added */
2768 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2770 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2771 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2773 xmap_dereference(dev_maps->attr_map[tci]) :
2775 if (new_map && new_map != map)
2780 mutex_unlock(&xps_map_mutex);
2782 kfree(new_dev_maps);
2785 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2787 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2793 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2798 EXPORT_SYMBOL(netif_set_xps_queue);
2801 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2803 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2805 /* Unbind any subordinate channels */
2806 while (txq-- != &dev->_tx[0]) {
2808 netdev_unbind_sb_channel(dev, txq->sb_dev);
2812 void netdev_reset_tc(struct net_device *dev)
2815 netif_reset_xps_queues_gt(dev, 0);
2817 netdev_unbind_all_sb_channels(dev);
2819 /* Reset TC configuration of device */
2821 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2822 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2824 EXPORT_SYMBOL(netdev_reset_tc);
2826 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2828 if (tc >= dev->num_tc)
2832 netif_reset_xps_queues(dev, offset, count);
2834 dev->tc_to_txq[tc].count = count;
2835 dev->tc_to_txq[tc].offset = offset;
2838 EXPORT_SYMBOL(netdev_set_tc_queue);
2840 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2842 if (num_tc > TC_MAX_QUEUE)
2846 netif_reset_xps_queues_gt(dev, 0);
2848 netdev_unbind_all_sb_channels(dev);
2850 dev->num_tc = num_tc;
2853 EXPORT_SYMBOL(netdev_set_num_tc);
2855 void netdev_unbind_sb_channel(struct net_device *dev,
2856 struct net_device *sb_dev)
2858 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2861 netif_reset_xps_queues_gt(sb_dev, 0);
2863 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2864 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2866 while (txq-- != &dev->_tx[0]) {
2867 if (txq->sb_dev == sb_dev)
2871 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2873 int netdev_bind_sb_channel_queue(struct net_device *dev,
2874 struct net_device *sb_dev,
2875 u8 tc, u16 count, u16 offset)
2877 /* Make certain the sb_dev and dev are already configured */
2878 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2881 /* We cannot hand out queues we don't have */
2882 if ((offset + count) > dev->real_num_tx_queues)
2885 /* Record the mapping */
2886 sb_dev->tc_to_txq[tc].count = count;
2887 sb_dev->tc_to_txq[tc].offset = offset;
2889 /* Provide a way for Tx queue to find the tc_to_txq map or
2890 * XPS map for itself.
2893 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2897 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2899 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2901 /* Do not use a multiqueue device to represent a subordinate channel */
2902 if (netif_is_multiqueue(dev))
2905 /* We allow channels 1 - 32767 to be used for subordinate channels.
2906 * Channel 0 is meant to be "native" mode and used only to represent
2907 * the main root device. We allow writing 0 to reset the device back
2908 * to normal mode after being used as a subordinate channel.
2910 if (channel > S16_MAX)
2913 dev->num_tc = -channel;
2917 EXPORT_SYMBOL(netdev_set_sb_channel);
2920 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2921 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2923 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2928 disabling = txq < dev->real_num_tx_queues;
2930 if (txq < 1 || txq > dev->num_tx_queues)
2933 if (dev->reg_state == NETREG_REGISTERED ||
2934 dev->reg_state == NETREG_UNREGISTERING) {
2937 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2943 netif_setup_tc(dev, txq);
2945 dev->real_num_tx_queues = txq;
2949 qdisc_reset_all_tx_gt(dev, txq);
2951 netif_reset_xps_queues_gt(dev, txq);
2955 dev->real_num_tx_queues = txq;
2960 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2964 * netif_set_real_num_rx_queues - set actual number of RX queues used
2965 * @dev: Network device
2966 * @rxq: Actual number of RX queues
2968 * This must be called either with the rtnl_lock held or before
2969 * registration of the net device. Returns 0 on success, or a
2970 * negative error code. If called before registration, it always
2973 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2977 if (rxq < 1 || rxq > dev->num_rx_queues)
2980 if (dev->reg_state == NETREG_REGISTERED) {
2983 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2989 dev->real_num_rx_queues = rxq;
2992 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2996 * netif_get_num_default_rss_queues - default number of RSS queues
2998 * This routine should set an upper limit on the number of RSS queues
2999 * used by default by multiqueue devices.
3001 int netif_get_num_default_rss_queues(void)
3003 return is_kdump_kernel() ?
3004 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3006 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3008 static void __netif_reschedule(struct Qdisc *q)
3010 struct softnet_data *sd;
3011 unsigned long flags;
3013 local_irq_save(flags);
3014 sd = this_cpu_ptr(&softnet_data);
3015 q->next_sched = NULL;
3016 *sd->output_queue_tailp = q;
3017 sd->output_queue_tailp = &q->next_sched;
3018 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3019 local_irq_restore(flags);
3022 void __netif_schedule(struct Qdisc *q)
3024 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3025 __netif_reschedule(q);
3027 EXPORT_SYMBOL(__netif_schedule);
3029 struct dev_kfree_skb_cb {
3030 enum skb_free_reason reason;
3033 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3035 return (struct dev_kfree_skb_cb *)skb->cb;
3038 void netif_schedule_queue(struct netdev_queue *txq)
3041 if (!netif_xmit_stopped(txq)) {
3042 struct Qdisc *q = rcu_dereference(txq->qdisc);
3044 __netif_schedule(q);
3048 EXPORT_SYMBOL(netif_schedule_queue);
3050 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3052 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3056 q = rcu_dereference(dev_queue->qdisc);
3057 __netif_schedule(q);
3061 EXPORT_SYMBOL(netif_tx_wake_queue);
3063 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3065 unsigned long flags;
3070 if (likely(refcount_read(&skb->users) == 1)) {
3072 refcount_set(&skb->users, 0);
3073 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3076 get_kfree_skb_cb(skb)->reason = reason;
3077 local_irq_save(flags);
3078 skb->next = __this_cpu_read(softnet_data.completion_queue);
3079 __this_cpu_write(softnet_data.completion_queue, skb);
3080 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3081 local_irq_restore(flags);
3083 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3085 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3087 if (in_irq() || irqs_disabled())
3088 __dev_kfree_skb_irq(skb, reason);
3092 EXPORT_SYMBOL(__dev_kfree_skb_any);
3096 * netif_device_detach - mark device as removed
3097 * @dev: network device
3099 * Mark device as removed from system and therefore no longer available.
3101 void netif_device_detach(struct net_device *dev)
3103 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3104 netif_running(dev)) {
3105 netif_tx_stop_all_queues(dev);
3108 EXPORT_SYMBOL(netif_device_detach);
3111 * netif_device_attach - mark device as attached
3112 * @dev: network device
3114 * Mark device as attached from system and restart if needed.
3116 void netif_device_attach(struct net_device *dev)
3118 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3119 netif_running(dev)) {
3120 netif_tx_wake_all_queues(dev);
3121 __netdev_watchdog_up(dev);
3124 EXPORT_SYMBOL(netif_device_attach);
3127 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3128 * to be used as a distribution range.
3130 static u16 skb_tx_hash(const struct net_device *dev,
3131 const struct net_device *sb_dev,
3132 struct sk_buff *skb)
3136 u16 qcount = dev->real_num_tx_queues;
3139 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3141 qoffset = sb_dev->tc_to_txq[tc].offset;
3142 qcount = sb_dev->tc_to_txq[tc].count;
3145 if (skb_rx_queue_recorded(skb)) {
3146 hash = skb_get_rx_queue(skb);
3147 if (hash >= qoffset)
3149 while (unlikely(hash >= qcount))
3151 return hash + qoffset;
3154 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3157 static void skb_warn_bad_offload(const struct sk_buff *skb)
3159 static const netdev_features_t null_features;
3160 struct net_device *dev = skb->dev;
3161 const char *name = "";
3163 if (!net_ratelimit())
3167 if (dev->dev.parent)
3168 name = dev_driver_string(dev->dev.parent);
3170 name = netdev_name(dev);
3172 skb_dump(KERN_WARNING, skb, false);
3173 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3174 name, dev ? &dev->features : &null_features,
3175 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3179 * Invalidate hardware checksum when packet is to be mangled, and
3180 * complete checksum manually on outgoing path.
3182 int skb_checksum_help(struct sk_buff *skb)
3185 int ret = 0, offset;
3187 if (skb->ip_summed == CHECKSUM_COMPLETE)
3188 goto out_set_summed;
3190 if (unlikely(skb_shinfo(skb)->gso_size)) {
3191 skb_warn_bad_offload(skb);
3195 /* Before computing a checksum, we should make sure no frag could
3196 * be modified by an external entity : checksum could be wrong.
3198 if (skb_has_shared_frag(skb)) {
3199 ret = __skb_linearize(skb);
3204 offset = skb_checksum_start_offset(skb);
3205 BUG_ON(offset >= skb_headlen(skb));
3206 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3208 offset += skb->csum_offset;
3209 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3211 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3215 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3217 skb->ip_summed = CHECKSUM_NONE;
3221 EXPORT_SYMBOL(skb_checksum_help);
3223 int skb_crc32c_csum_help(struct sk_buff *skb)
3226 int ret = 0, offset, start;
3228 if (skb->ip_summed != CHECKSUM_PARTIAL)
3231 if (unlikely(skb_is_gso(skb)))
3234 /* Before computing a checksum, we should make sure no frag could
3235 * be modified by an external entity : checksum could be wrong.
3237 if (unlikely(skb_has_shared_frag(skb))) {
3238 ret = __skb_linearize(skb);
3242 start = skb_checksum_start_offset(skb);
3243 offset = start + offsetof(struct sctphdr, checksum);
3244 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3249 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3253 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3254 skb->len - start, ~(__u32)0,
3256 *(__le32 *)(skb->data + offset) = crc32c_csum;
3257 skb->ip_summed = CHECKSUM_NONE;
3258 skb->csum_not_inet = 0;
3263 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3265 __be16 type = skb->protocol;
3267 /* Tunnel gso handlers can set protocol to ethernet. */
3268 if (type == htons(ETH_P_TEB)) {
3271 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3274 eth = (struct ethhdr *)skb->data;
3275 type = eth->h_proto;
3278 return __vlan_get_protocol(skb, type, depth);
3282 * skb_mac_gso_segment - mac layer segmentation handler.
3283 * @skb: buffer to segment
3284 * @features: features for the output path (see dev->features)
3286 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3287 netdev_features_t features)
3289 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3290 struct packet_offload *ptype;
3291 int vlan_depth = skb->mac_len;
3292 __be16 type = skb_network_protocol(skb, &vlan_depth);
3294 if (unlikely(!type))
3295 return ERR_PTR(-EINVAL);
3297 __skb_pull(skb, vlan_depth);
3300 list_for_each_entry_rcu(ptype, &offload_base, list) {
3301 if (ptype->type == type && ptype->callbacks.gso_segment) {
3302 segs = ptype->callbacks.gso_segment(skb, features);
3308 __skb_push(skb, skb->data - skb_mac_header(skb));
3312 EXPORT_SYMBOL(skb_mac_gso_segment);
3315 /* openvswitch calls this on rx path, so we need a different check.
3317 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3320 return skb->ip_summed != CHECKSUM_PARTIAL &&
3321 skb->ip_summed != CHECKSUM_UNNECESSARY;
3323 return skb->ip_summed == CHECKSUM_NONE;
3327 * __skb_gso_segment - Perform segmentation on skb.
3328 * @skb: buffer to segment
3329 * @features: features for the output path (see dev->features)
3330 * @tx_path: whether it is called in TX path
3332 * This function segments the given skb and returns a list of segments.
3334 * It may return NULL if the skb requires no segmentation. This is
3335 * only possible when GSO is used for verifying header integrity.
3337 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3339 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3340 netdev_features_t features, bool tx_path)
3342 struct sk_buff *segs;
3344 if (unlikely(skb_needs_check(skb, tx_path))) {
3347 /* We're going to init ->check field in TCP or UDP header */
3348 err = skb_cow_head(skb, 0);
3350 return ERR_PTR(err);
3353 /* Only report GSO partial support if it will enable us to
3354 * support segmentation on this frame without needing additional
3357 if (features & NETIF_F_GSO_PARTIAL) {
3358 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3359 struct net_device *dev = skb->dev;
3361 partial_features |= dev->features & dev->gso_partial_features;
3362 if (!skb_gso_ok(skb, features | partial_features))
3363 features &= ~NETIF_F_GSO_PARTIAL;
3366 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3367 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3369 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3370 SKB_GSO_CB(skb)->encap_level = 0;
3372 skb_reset_mac_header(skb);
3373 skb_reset_mac_len(skb);
3375 segs = skb_mac_gso_segment(skb, features);
3377 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3378 skb_warn_bad_offload(skb);
3382 EXPORT_SYMBOL(__skb_gso_segment);
3384 /* Take action when hardware reception checksum errors are detected. */
3386 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3388 if (net_ratelimit()) {
3389 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3390 skb_dump(KERN_ERR, skb, true);
3394 EXPORT_SYMBOL(netdev_rx_csum_fault);
3397 /* XXX: check that highmem exists at all on the given machine. */
3398 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3400 #ifdef CONFIG_HIGHMEM
3403 if (!(dev->features & NETIF_F_HIGHDMA)) {
3404 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3405 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3407 if (PageHighMem(skb_frag_page(frag)))
3415 /* If MPLS offload request, verify we are testing hardware MPLS features
3416 * instead of standard features for the netdev.
3418 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3419 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3420 netdev_features_t features,
3423 if (eth_p_mpls(type))
3424 features &= skb->dev->mpls_features;
3429 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3430 netdev_features_t features,
3437 static netdev_features_t harmonize_features(struct sk_buff *skb,
3438 netdev_features_t features)
3443 type = skb_network_protocol(skb, &tmp);
3444 features = net_mpls_features(skb, features, type);
3446 if (skb->ip_summed != CHECKSUM_NONE &&
3447 !can_checksum_protocol(features, type)) {
3448 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3450 if (illegal_highdma(skb->dev, skb))
3451 features &= ~NETIF_F_SG;
3456 netdev_features_t passthru_features_check(struct sk_buff *skb,
3457 struct net_device *dev,
3458 netdev_features_t features)
3462 EXPORT_SYMBOL(passthru_features_check);
3464 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3465 struct net_device *dev,
3466 netdev_features_t features)
3468 return vlan_features_check(skb, features);
3471 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3472 struct net_device *dev,
3473 netdev_features_t features)
3475 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3477 if (gso_segs > dev->gso_max_segs)
3478 return features & ~NETIF_F_GSO_MASK;
3480 /* Support for GSO partial features requires software
3481 * intervention before we can actually process the packets
3482 * so we need to strip support for any partial features now
3483 * and we can pull them back in after we have partially
3484 * segmented the frame.
3486 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3487 features &= ~dev->gso_partial_features;
3489 /* Make sure to clear the IPv4 ID mangling feature if the
3490 * IPv4 header has the potential to be fragmented.
3492 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3493 struct iphdr *iph = skb->encapsulation ?
3494 inner_ip_hdr(skb) : ip_hdr(skb);
3496 if (!(iph->frag_off & htons(IP_DF)))
3497 features &= ~NETIF_F_TSO_MANGLEID;
3503 netdev_features_t netif_skb_features(struct sk_buff *skb)
3505 struct net_device *dev = skb->dev;
3506 netdev_features_t features = dev->features;
3508 if (skb_is_gso(skb))
3509 features = gso_features_check(skb, dev, features);
3511 /* If encapsulation offload request, verify we are testing
3512 * hardware encapsulation features instead of standard
3513 * features for the netdev
3515 if (skb->encapsulation)
3516 features &= dev->hw_enc_features;
3518 if (skb_vlan_tagged(skb))
3519 features = netdev_intersect_features(features,
3520 dev->vlan_features |
3521 NETIF_F_HW_VLAN_CTAG_TX |
3522 NETIF_F_HW_VLAN_STAG_TX);
3524 if (dev->netdev_ops->ndo_features_check)
3525 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3528 features &= dflt_features_check(skb, dev, features);
3530 return harmonize_features(skb, features);
3532 EXPORT_SYMBOL(netif_skb_features);
3534 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3535 struct netdev_queue *txq, bool more)
3540 if (dev_nit_active(dev))
3541 dev_queue_xmit_nit(skb, dev);
3544 trace_net_dev_start_xmit(skb, dev);
3545 rc = netdev_start_xmit(skb, dev, txq, more);
3546 trace_net_dev_xmit(skb, rc, dev, len);
3551 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3552 struct netdev_queue *txq, int *ret)
3554 struct sk_buff *skb = first;
3555 int rc = NETDEV_TX_OK;
3558 struct sk_buff *next = skb->next;
3560 skb_mark_not_on_list(skb);
3561 rc = xmit_one(skb, dev, txq, next != NULL);
3562 if (unlikely(!dev_xmit_complete(rc))) {
3568 if (netif_tx_queue_stopped(txq) && skb) {
3569 rc = NETDEV_TX_BUSY;
3579 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3580 netdev_features_t features)
3582 if (skb_vlan_tag_present(skb) &&
3583 !vlan_hw_offload_capable(features, skb->vlan_proto))
3584 skb = __vlan_hwaccel_push_inside(skb);
3588 int skb_csum_hwoffload_help(struct sk_buff *skb,
3589 const netdev_features_t features)
3591 if (unlikely(skb->csum_not_inet))
3592 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3593 skb_crc32c_csum_help(skb);
3595 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3597 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3599 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3601 netdev_features_t features;
3603 features = netif_skb_features(skb);
3604 skb = validate_xmit_vlan(skb, features);
3608 skb = sk_validate_xmit_skb(skb, dev);
3612 if (netif_needs_gso(skb, features)) {
3613 struct sk_buff *segs;
3615 segs = skb_gso_segment(skb, features);
3623 if (skb_needs_linearize(skb, features) &&
3624 __skb_linearize(skb))
3627 /* If packet is not checksummed and device does not
3628 * support checksumming for this protocol, complete
3629 * checksumming here.
3631 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3632 if (skb->encapsulation)
3633 skb_set_inner_transport_header(skb,
3634 skb_checksum_start_offset(skb));
3636 skb_set_transport_header(skb,
3637 skb_checksum_start_offset(skb));
3638 if (skb_csum_hwoffload_help(skb, features))
3643 skb = validate_xmit_xfrm(skb, features, again);
3650 atomic_long_inc(&dev->tx_dropped);
3654 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3656 struct sk_buff *next, *head = NULL, *tail;
3658 for (; skb != NULL; skb = next) {
3660 skb_mark_not_on_list(skb);
3662 /* in case skb wont be segmented, point to itself */
3665 skb = validate_xmit_skb(skb, dev, again);
3673 /* If skb was segmented, skb->prev points to
3674 * the last segment. If not, it still contains skb.
3680 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3682 static void qdisc_pkt_len_init(struct sk_buff *skb)
3684 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3686 qdisc_skb_cb(skb)->pkt_len = skb->len;
3688 /* To get more precise estimation of bytes sent on wire,
3689 * we add to pkt_len the headers size of all segments
3691 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3692 unsigned int hdr_len;
3693 u16 gso_segs = shinfo->gso_segs;
3695 /* mac layer + network layer */
3696 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3698 /* + transport layer */
3699 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3700 const struct tcphdr *th;
3701 struct tcphdr _tcphdr;
3703 th = skb_header_pointer(skb, skb_transport_offset(skb),
3704 sizeof(_tcphdr), &_tcphdr);
3706 hdr_len += __tcp_hdrlen(th);
3708 struct udphdr _udphdr;
3710 if (skb_header_pointer(skb, skb_transport_offset(skb),
3711 sizeof(_udphdr), &_udphdr))
3712 hdr_len += sizeof(struct udphdr);
3715 if (shinfo->gso_type & SKB_GSO_DODGY)
3716 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3719 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3723 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3724 struct net_device *dev,
3725 struct netdev_queue *txq)
3727 spinlock_t *root_lock = qdisc_lock(q);
3728 struct sk_buff *to_free = NULL;
3732 qdisc_calculate_pkt_len(skb, q);
3734 if (q->flags & TCQ_F_NOLOCK) {
3735 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3738 if (unlikely(to_free))
3739 kfree_skb_list(to_free);
3744 * Heuristic to force contended enqueues to serialize on a
3745 * separate lock before trying to get qdisc main lock.
3746 * This permits qdisc->running owner to get the lock more
3747 * often and dequeue packets faster.
3749 contended = qdisc_is_running(q);
3750 if (unlikely(contended))
3751 spin_lock(&q->busylock);
3753 spin_lock(root_lock);
3754 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3755 __qdisc_drop(skb, &to_free);
3757 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3758 qdisc_run_begin(q)) {
3760 * This is a work-conserving queue; there are no old skbs
3761 * waiting to be sent out; and the qdisc is not running -
3762 * xmit the skb directly.
3765 qdisc_bstats_update(q, skb);
3767 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3768 if (unlikely(contended)) {
3769 spin_unlock(&q->busylock);
3776 rc = NET_XMIT_SUCCESS;
3778 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3779 if (qdisc_run_begin(q)) {
3780 if (unlikely(contended)) {
3781 spin_unlock(&q->busylock);
3788 spin_unlock(root_lock);
3789 if (unlikely(to_free))
3790 kfree_skb_list(to_free);
3791 if (unlikely(contended))
3792 spin_unlock(&q->busylock);
3796 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3797 static void skb_update_prio(struct sk_buff *skb)
3799 const struct netprio_map *map;
3800 const struct sock *sk;
3801 unsigned int prioidx;
3805 map = rcu_dereference_bh(skb->dev->priomap);
3808 sk = skb_to_full_sk(skb);
3812 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3814 if (prioidx < map->priomap_len)
3815 skb->priority = map->priomap[prioidx];
3818 #define skb_update_prio(skb)
3822 * dev_loopback_xmit - loop back @skb
3823 * @net: network namespace this loopback is happening in
3824 * @sk: sk needed to be a netfilter okfn
3825 * @skb: buffer to transmit
3827 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3829 skb_reset_mac_header(skb);
3830 __skb_pull(skb, skb_network_offset(skb));
3831 skb->pkt_type = PACKET_LOOPBACK;
3832 skb->ip_summed = CHECKSUM_UNNECESSARY;
3833 WARN_ON(!skb_dst(skb));
3838 EXPORT_SYMBOL(dev_loopback_xmit);
3840 #ifdef CONFIG_NET_EGRESS
3841 static struct sk_buff *
3842 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3844 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3845 struct tcf_result cl_res;
3850 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3851 mini_qdisc_bstats_cpu_update(miniq, skb);
3853 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3855 case TC_ACT_RECLASSIFY:
3856 skb->tc_index = TC_H_MIN(cl_res.classid);
3859 mini_qdisc_qstats_cpu_drop(miniq);
3860 *ret = NET_XMIT_DROP;
3866 *ret = NET_XMIT_SUCCESS;
3869 case TC_ACT_REDIRECT:
3870 /* No need to push/pop skb's mac_header here on egress! */
3871 skb_do_redirect(skb);
3872 *ret = NET_XMIT_SUCCESS;
3880 #endif /* CONFIG_NET_EGRESS */
3883 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3884 struct xps_dev_maps *dev_maps, unsigned int tci)
3886 struct xps_map *map;
3887 int queue_index = -1;
3891 tci += netdev_get_prio_tc_map(dev, skb->priority);
3894 map = rcu_dereference(dev_maps->attr_map[tci]);
3897 queue_index = map->queues[0];
3899 queue_index = map->queues[reciprocal_scale(
3900 skb_get_hash(skb), map->len)];
3901 if (unlikely(queue_index >= dev->real_num_tx_queues))
3908 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3909 struct sk_buff *skb)
3912 struct xps_dev_maps *dev_maps;
3913 struct sock *sk = skb->sk;
3914 int queue_index = -1;
3916 if (!static_key_false(&xps_needed))
3920 if (!static_key_false(&xps_rxqs_needed))
3923 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3925 int tci = sk_rx_queue_get(sk);
3927 if (tci >= 0 && tci < dev->num_rx_queues)
3928 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3933 if (queue_index < 0) {
3934 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3936 unsigned int tci = skb->sender_cpu - 1;
3938 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3950 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3951 struct net_device *sb_dev)
3955 EXPORT_SYMBOL(dev_pick_tx_zero);
3957 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3958 struct net_device *sb_dev)
3960 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3962 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3964 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3965 struct net_device *sb_dev)
3967 struct sock *sk = skb->sk;
3968 int queue_index = sk_tx_queue_get(sk);
3970 sb_dev = sb_dev ? : dev;
3972 if (queue_index < 0 || skb->ooo_okay ||
3973 queue_index >= dev->real_num_tx_queues) {
3974 int new_index = get_xps_queue(dev, sb_dev, skb);
3977 new_index = skb_tx_hash(dev, sb_dev, skb);
3979 if (queue_index != new_index && sk &&
3981 rcu_access_pointer(sk->sk_dst_cache))
3982 sk_tx_queue_set(sk, new_index);
3984 queue_index = new_index;
3989 EXPORT_SYMBOL(netdev_pick_tx);
3991 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3992 struct sk_buff *skb,
3993 struct net_device *sb_dev)
3995 int queue_index = 0;
3998 u32 sender_cpu = skb->sender_cpu - 1;
4000 if (sender_cpu >= (u32)NR_CPUS)
4001 skb->sender_cpu = raw_smp_processor_id() + 1;
4004 if (dev->real_num_tx_queues != 1) {
4005 const struct net_device_ops *ops = dev->netdev_ops;
4007 if (ops->ndo_select_queue)
4008 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4010 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4012 queue_index = netdev_cap_txqueue(dev, queue_index);
4015 skb_set_queue_mapping(skb, queue_index);
4016 return netdev_get_tx_queue(dev, queue_index);
4020 * __dev_queue_xmit - transmit a buffer
4021 * @skb: buffer to transmit
4022 * @sb_dev: suboordinate device used for L2 forwarding offload
4024 * Queue a buffer for transmission to a network device. The caller must
4025 * have set the device and priority and built the buffer before calling
4026 * this function. The function can be called from an interrupt.
4028 * A negative errno code is returned on a failure. A success does not
4029 * guarantee the frame will be transmitted as it may be dropped due
4030 * to congestion or traffic shaping.
4032 * -----------------------------------------------------------------------------------
4033 * I notice this method can also return errors from the queue disciplines,
4034 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4037 * Regardless of the return value, the skb is consumed, so it is currently
4038 * difficult to retry a send to this method. (You can bump the ref count
4039 * before sending to hold a reference for retry if you are careful.)
4041 * When calling this method, interrupts MUST be enabled. This is because
4042 * the BH enable code must have IRQs enabled so that it will not deadlock.
4045 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4047 struct net_device *dev = skb->dev;
4048 struct netdev_queue *txq;
4053 skb_reset_mac_header(skb);
4055 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4056 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
4058 /* Disable soft irqs for various locks below. Also
4059 * stops preemption for RCU.
4063 skb_update_prio(skb);
4065 qdisc_pkt_len_init(skb);
4066 #ifdef CONFIG_NET_CLS_ACT
4067 skb->tc_at_ingress = 0;
4068 # ifdef CONFIG_NET_EGRESS
4069 if (static_branch_unlikely(&egress_needed_key)) {
4070 skb = sch_handle_egress(skb, &rc, dev);
4076 /* If device/qdisc don't need skb->dst, release it right now while
4077 * its hot in this cpu cache.
4079 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4084 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4085 q = rcu_dereference_bh(txq->qdisc);
4087 trace_net_dev_queue(skb);
4089 rc = __dev_xmit_skb(skb, q, dev, txq);
4093 /* The device has no queue. Common case for software devices:
4094 * loopback, all the sorts of tunnels...
4096 * Really, it is unlikely that netif_tx_lock protection is necessary
4097 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4099 * However, it is possible, that they rely on protection
4102 * Check this and shot the lock. It is not prone from deadlocks.
4103 *Either shot noqueue qdisc, it is even simpler 8)
4105 if (dev->flags & IFF_UP) {
4106 int cpu = smp_processor_id(); /* ok because BHs are off */
4108 if (txq->xmit_lock_owner != cpu) {
4109 if (dev_xmit_recursion())
4110 goto recursion_alert;
4112 skb = validate_xmit_skb(skb, dev, &again);
4116 HARD_TX_LOCK(dev, txq, cpu);
4118 if (!netif_xmit_stopped(txq)) {
4119 dev_xmit_recursion_inc();
4120 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4121 dev_xmit_recursion_dec();
4122 if (dev_xmit_complete(rc)) {
4123 HARD_TX_UNLOCK(dev, txq);
4127 HARD_TX_UNLOCK(dev, txq);
4128 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4131 /* Recursion is detected! It is possible,
4135 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4141 rcu_read_unlock_bh();
4143 atomic_long_inc(&dev->tx_dropped);
4144 kfree_skb_list(skb);
4147 rcu_read_unlock_bh();
4151 int dev_queue_xmit(struct sk_buff *skb)
4153 return __dev_queue_xmit(skb, NULL);
4155 EXPORT_SYMBOL(dev_queue_xmit);
4157 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4159 return __dev_queue_xmit(skb, sb_dev);
4161 EXPORT_SYMBOL(dev_queue_xmit_accel);
4163 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4165 struct net_device *dev = skb->dev;
4166 struct sk_buff *orig_skb = skb;
4167 struct netdev_queue *txq;
4168 int ret = NETDEV_TX_BUSY;
4171 if (unlikely(!netif_running(dev) ||
4172 !netif_carrier_ok(dev)))
4175 skb = validate_xmit_skb_list(skb, dev, &again);
4176 if (skb != orig_skb)
4179 skb_set_queue_mapping(skb, queue_id);
4180 txq = skb_get_tx_queue(dev, skb);
4184 HARD_TX_LOCK(dev, txq, smp_processor_id());
4185 if (!netif_xmit_frozen_or_drv_stopped(txq))
4186 ret = netdev_start_xmit(skb, dev, txq, false);
4187 HARD_TX_UNLOCK(dev, txq);
4191 if (!dev_xmit_complete(ret))
4196 atomic_long_inc(&dev->tx_dropped);
4197 kfree_skb_list(skb);
4198 return NET_XMIT_DROP;
4200 EXPORT_SYMBOL(dev_direct_xmit);
4202 /*************************************************************************
4204 *************************************************************************/
4206 int netdev_max_backlog __read_mostly = 1000;
4207 EXPORT_SYMBOL(netdev_max_backlog);
4209 int netdev_tstamp_prequeue __read_mostly = 1;
4210 int netdev_budget __read_mostly = 300;
4211 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4212 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4213 int weight_p __read_mostly = 64; /* old backlog weight */
4214 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4215 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4216 int dev_rx_weight __read_mostly = 64;
4217 int dev_tx_weight __read_mostly = 64;
4218 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4219 int gro_normal_batch __read_mostly = 8;
4221 /* Called with irq disabled */
4222 static inline void ____napi_schedule(struct softnet_data *sd,
4223 struct napi_struct *napi)
4225 list_add_tail(&napi->poll_list, &sd->poll_list);
4226 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4231 /* One global table that all flow-based protocols share. */
4232 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4233 EXPORT_SYMBOL(rps_sock_flow_table);
4234 u32 rps_cpu_mask __read_mostly;
4235 EXPORT_SYMBOL(rps_cpu_mask);
4237 struct static_key_false rps_needed __read_mostly;
4238 EXPORT_SYMBOL(rps_needed);
4239 struct static_key_false rfs_needed __read_mostly;
4240 EXPORT_SYMBOL(rfs_needed);
4242 static struct rps_dev_flow *
4243 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4244 struct rps_dev_flow *rflow, u16 next_cpu)
4246 if (next_cpu < nr_cpu_ids) {
4247 #ifdef CONFIG_RFS_ACCEL
4248 struct netdev_rx_queue *rxqueue;
4249 struct rps_dev_flow_table *flow_table;
4250 struct rps_dev_flow *old_rflow;
4255 /* Should we steer this flow to a different hardware queue? */
4256 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4257 !(dev->features & NETIF_F_NTUPLE))
4259 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4260 if (rxq_index == skb_get_rx_queue(skb))
4263 rxqueue = dev->_rx + rxq_index;
4264 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4267 flow_id = skb_get_hash(skb) & flow_table->mask;
4268 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4269 rxq_index, flow_id);
4273 rflow = &flow_table->flows[flow_id];
4275 if (old_rflow->filter == rflow->filter)
4276 old_rflow->filter = RPS_NO_FILTER;
4280 per_cpu(softnet_data, next_cpu).input_queue_head;
4283 rflow->cpu = next_cpu;
4288 * get_rps_cpu is called from netif_receive_skb and returns the target
4289 * CPU from the RPS map of the receiving queue for a given skb.
4290 * rcu_read_lock must be held on entry.
4292 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4293 struct rps_dev_flow **rflowp)
4295 const struct rps_sock_flow_table *sock_flow_table;
4296 struct netdev_rx_queue *rxqueue = dev->_rx;
4297 struct rps_dev_flow_table *flow_table;
4298 struct rps_map *map;
4303 if (skb_rx_queue_recorded(skb)) {
4304 u16 index = skb_get_rx_queue(skb);
4306 if (unlikely(index >= dev->real_num_rx_queues)) {
4307 WARN_ONCE(dev->real_num_rx_queues > 1,
4308 "%s received packet on queue %u, but number "
4309 "of RX queues is %u\n",
4310 dev->name, index, dev->real_num_rx_queues);
4316 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4318 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4319 map = rcu_dereference(rxqueue->rps_map);
4320 if (!flow_table && !map)
4323 skb_reset_network_header(skb);
4324 hash = skb_get_hash(skb);
4328 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4329 if (flow_table && sock_flow_table) {
4330 struct rps_dev_flow *rflow;
4334 /* First check into global flow table if there is a match */
4335 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4336 if ((ident ^ hash) & ~rps_cpu_mask)
4339 next_cpu = ident & rps_cpu_mask;
4341 /* OK, now we know there is a match,
4342 * we can look at the local (per receive queue) flow table
4344 rflow = &flow_table->flows[hash & flow_table->mask];
4348 * If the desired CPU (where last recvmsg was done) is
4349 * different from current CPU (one in the rx-queue flow
4350 * table entry), switch if one of the following holds:
4351 * - Current CPU is unset (>= nr_cpu_ids).
4352 * - Current CPU is offline.
4353 * - The current CPU's queue tail has advanced beyond the
4354 * last packet that was enqueued using this table entry.
4355 * This guarantees that all previous packets for the flow
4356 * have been dequeued, thus preserving in order delivery.
4358 if (unlikely(tcpu != next_cpu) &&
4359 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4360 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4361 rflow->last_qtail)) >= 0)) {
4363 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4366 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4376 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4377 if (cpu_online(tcpu)) {
4387 #ifdef CONFIG_RFS_ACCEL
4390 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4391 * @dev: Device on which the filter was set
4392 * @rxq_index: RX queue index
4393 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4394 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4396 * Drivers that implement ndo_rx_flow_steer() should periodically call
4397 * this function for each installed filter and remove the filters for
4398 * which it returns %true.
4400 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4401 u32 flow_id, u16 filter_id)
4403 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4404 struct rps_dev_flow_table *flow_table;
4405 struct rps_dev_flow *rflow;
4410 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4411 if (flow_table && flow_id <= flow_table->mask) {
4412 rflow = &flow_table->flows[flow_id];
4413 cpu = READ_ONCE(rflow->cpu);
4414 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4415 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4416 rflow->last_qtail) <
4417 (int)(10 * flow_table->mask)))
4423 EXPORT_SYMBOL(rps_may_expire_flow);
4425 #endif /* CONFIG_RFS_ACCEL */
4427 /* Called from hardirq (IPI) context */
4428 static void rps_trigger_softirq(void *data)
4430 struct softnet_data *sd = data;
4432 ____napi_schedule(sd, &sd->backlog);
4436 #endif /* CONFIG_RPS */
4439 * Check if this softnet_data structure is another cpu one
4440 * If yes, queue it to our IPI list and return 1
4443 static int rps_ipi_queued(struct softnet_data *sd)
4446 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4449 sd->rps_ipi_next = mysd->rps_ipi_list;
4450 mysd->rps_ipi_list = sd;
4452 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4455 #endif /* CONFIG_RPS */
4459 #ifdef CONFIG_NET_FLOW_LIMIT
4460 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4463 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4465 #ifdef CONFIG_NET_FLOW_LIMIT
4466 struct sd_flow_limit *fl;
4467 struct softnet_data *sd;
4468 unsigned int old_flow, new_flow;
4470 if (qlen < (netdev_max_backlog >> 1))
4473 sd = this_cpu_ptr(&softnet_data);
4476 fl = rcu_dereference(sd->flow_limit);
4478 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4479 old_flow = fl->history[fl->history_head];
4480 fl->history[fl->history_head] = new_flow;
4483 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4485 if (likely(fl->buckets[old_flow]))
4486 fl->buckets[old_flow]--;
4488 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4500 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4501 * queue (may be a remote CPU queue).
4503 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4504 unsigned int *qtail)
4506 struct softnet_data *sd;
4507 unsigned long flags;
4510 sd = &per_cpu(softnet_data, cpu);
4512 local_irq_save(flags);
4515 if (!netif_running(skb->dev))
4517 qlen = skb_queue_len(&sd->input_pkt_queue);
4518 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4521 __skb_queue_tail(&sd->input_pkt_queue, skb);
4522 input_queue_tail_incr_save(sd, qtail);
4524 local_irq_restore(flags);
4525 return NET_RX_SUCCESS;
4528 /* Schedule NAPI for backlog device
4529 * We can use non atomic operation since we own the queue lock
4531 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4532 if (!rps_ipi_queued(sd))
4533 ____napi_schedule(sd, &sd->backlog);
4542 local_irq_restore(flags);
4544 atomic_long_inc(&skb->dev->rx_dropped);
4549 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4551 struct net_device *dev = skb->dev;
4552 struct netdev_rx_queue *rxqueue;
4556 if (skb_rx_queue_recorded(skb)) {
4557 u16 index = skb_get_rx_queue(skb);
4559 if (unlikely(index >= dev->real_num_rx_queues)) {
4560 WARN_ONCE(dev->real_num_rx_queues > 1,
4561 "%s received packet on queue %u, but number "
4562 "of RX queues is %u\n",
4563 dev->name, index, dev->real_num_rx_queues);
4565 return rxqueue; /* Return first rxqueue */
4572 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4573 struct xdp_buff *xdp,
4574 struct bpf_prog *xdp_prog)
4576 struct netdev_rx_queue *rxqueue;
4577 void *orig_data, *orig_data_end;
4578 u32 metalen, act = XDP_DROP;
4579 __be16 orig_eth_type;
4585 /* Reinjected packets coming from act_mirred or similar should
4586 * not get XDP generic processing.
4588 if (skb_is_redirected(skb))
4591 /* XDP packets must be linear and must have sufficient headroom
4592 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4593 * native XDP provides, thus we need to do it here as well.
4595 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4596 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4597 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4598 int troom = skb->tail + skb->data_len - skb->end;
4600 /* In case we have to go down the path and also linearize,
4601 * then lets do the pskb_expand_head() work just once here.
4603 if (pskb_expand_head(skb,
4604 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4605 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4607 if (skb_linearize(skb))
4611 /* The XDP program wants to see the packet starting at the MAC
4614 mac_len = skb->data - skb_mac_header(skb);
4615 hlen = skb_headlen(skb) + mac_len;
4616 xdp->data = skb->data - mac_len;
4617 xdp->data_meta = xdp->data;
4618 xdp->data_end = xdp->data + hlen;
4619 xdp->data_hard_start = skb->data - skb_headroom(skb);
4621 /* SKB "head" area always have tailroom for skb_shared_info */
4622 xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start;
4623 xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4625 orig_data_end = xdp->data_end;
4626 orig_data = xdp->data;
4627 eth = (struct ethhdr *)xdp->data;
4628 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4629 orig_eth_type = eth->h_proto;
4631 rxqueue = netif_get_rxqueue(skb);
4632 xdp->rxq = &rxqueue->xdp_rxq;
4634 act = bpf_prog_run_xdp(xdp_prog, xdp);
4636 /* check if bpf_xdp_adjust_head was used */
4637 off = xdp->data - orig_data;
4640 __skb_pull(skb, off);
4642 __skb_push(skb, -off);
4644 skb->mac_header += off;
4645 skb_reset_network_header(skb);
4648 /* check if bpf_xdp_adjust_tail was used */
4649 off = xdp->data_end - orig_data_end;
4651 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4652 skb->len += off; /* positive on grow, negative on shrink */
4655 /* check if XDP changed eth hdr such SKB needs update */
4656 eth = (struct ethhdr *)xdp->data;
4657 if ((orig_eth_type != eth->h_proto) ||
4658 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4659 __skb_push(skb, ETH_HLEN);
4660 skb->protocol = eth_type_trans(skb, skb->dev);
4666 __skb_push(skb, mac_len);
4669 metalen = xdp->data - xdp->data_meta;
4671 skb_metadata_set(skb, metalen);
4674 bpf_warn_invalid_xdp_action(act);
4677 trace_xdp_exception(skb->dev, xdp_prog, act);
4688 /* When doing generic XDP we have to bypass the qdisc layer and the
4689 * network taps in order to match in-driver-XDP behavior.
4691 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4693 struct net_device *dev = skb->dev;
4694 struct netdev_queue *txq;
4695 bool free_skb = true;
4698 txq = netdev_core_pick_tx(dev, skb, NULL);
4699 cpu = smp_processor_id();
4700 HARD_TX_LOCK(dev, txq, cpu);
4701 if (!netif_xmit_stopped(txq)) {
4702 rc = netdev_start_xmit(skb, dev, txq, 0);
4703 if (dev_xmit_complete(rc))
4706 HARD_TX_UNLOCK(dev, txq);
4708 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4713 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4715 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4718 struct xdp_buff xdp;
4722 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4723 if (act != XDP_PASS) {
4726 err = xdp_do_generic_redirect(skb->dev, skb,
4732 generic_xdp_tx(skb, xdp_prog);
4743 EXPORT_SYMBOL_GPL(do_xdp_generic);
4745 static int netif_rx_internal(struct sk_buff *skb)
4749 net_timestamp_check(netdev_tstamp_prequeue, skb);
4751 trace_netif_rx(skb);
4754 if (static_branch_unlikely(&rps_needed)) {
4755 struct rps_dev_flow voidflow, *rflow = &voidflow;
4761 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4763 cpu = smp_processor_id();
4765 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4774 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4781 * netif_rx - post buffer to the network code
4782 * @skb: buffer to post
4784 * This function receives a packet from a device driver and queues it for
4785 * the upper (protocol) levels to process. It always succeeds. The buffer
4786 * may be dropped during processing for congestion control or by the
4790 * NET_RX_SUCCESS (no congestion)
4791 * NET_RX_DROP (packet was dropped)
4795 int netif_rx(struct sk_buff *skb)
4799 trace_netif_rx_entry(skb);
4801 ret = netif_rx_internal(skb);
4802 trace_netif_rx_exit(ret);
4806 EXPORT_SYMBOL(netif_rx);
4808 int netif_rx_ni(struct sk_buff *skb)
4812 trace_netif_rx_ni_entry(skb);
4815 err = netif_rx_internal(skb);
4816 if (local_softirq_pending())
4819 trace_netif_rx_ni_exit(err);
4823 EXPORT_SYMBOL(netif_rx_ni);
4825 static __latent_entropy void net_tx_action(struct softirq_action *h)
4827 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4829 if (sd->completion_queue) {
4830 struct sk_buff *clist;
4832 local_irq_disable();
4833 clist = sd->completion_queue;
4834 sd->completion_queue = NULL;
4838 struct sk_buff *skb = clist;
4840 clist = clist->next;
4842 WARN_ON(refcount_read(&skb->users));
4843 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4844 trace_consume_skb(skb);
4846 trace_kfree_skb(skb, net_tx_action);
4848 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4851 __kfree_skb_defer(skb);
4854 __kfree_skb_flush();
4857 if (sd->output_queue) {
4860 local_irq_disable();
4861 head = sd->output_queue;
4862 sd->output_queue = NULL;
4863 sd->output_queue_tailp = &sd->output_queue;
4867 struct Qdisc *q = head;
4868 spinlock_t *root_lock = NULL;
4870 head = head->next_sched;
4872 if (!(q->flags & TCQ_F_NOLOCK)) {
4873 root_lock = qdisc_lock(q);
4874 spin_lock(root_lock);
4876 /* We need to make sure head->next_sched is read
4877 * before clearing __QDISC_STATE_SCHED
4879 smp_mb__before_atomic();
4880 clear_bit(__QDISC_STATE_SCHED, &q->state);
4883 spin_unlock(root_lock);
4887 xfrm_dev_backlog(sd);
4890 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4891 /* This hook is defined here for ATM LANE */
4892 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4893 unsigned char *addr) __read_mostly;
4894 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4897 static inline struct sk_buff *
4898 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4899 struct net_device *orig_dev)
4901 #ifdef CONFIG_NET_CLS_ACT
4902 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4903 struct tcf_result cl_res;
4905 /* If there's at least one ingress present somewhere (so
4906 * we get here via enabled static key), remaining devices
4907 * that are not configured with an ingress qdisc will bail
4914 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4918 qdisc_skb_cb(skb)->pkt_len = skb->len;
4919 skb->tc_at_ingress = 1;
4920 mini_qdisc_bstats_cpu_update(miniq, skb);
4922 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
4925 case TC_ACT_RECLASSIFY:
4926 skb->tc_index = TC_H_MIN(cl_res.classid);
4929 mini_qdisc_qstats_cpu_drop(miniq);
4937 case TC_ACT_REDIRECT:
4938 /* skb_mac_header check was done by cls/act_bpf, so
4939 * we can safely push the L2 header back before
4940 * redirecting to another netdev
4942 __skb_push(skb, skb->mac_len);
4943 skb_do_redirect(skb);
4945 case TC_ACT_CONSUMED:
4950 #endif /* CONFIG_NET_CLS_ACT */
4955 * netdev_is_rx_handler_busy - check if receive handler is registered
4956 * @dev: device to check
4958 * Check if a receive handler is already registered for a given device.
4959 * Return true if there one.
4961 * The caller must hold the rtnl_mutex.
4963 bool netdev_is_rx_handler_busy(struct net_device *dev)
4966 return dev && rtnl_dereference(dev->rx_handler);
4968 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4971 * netdev_rx_handler_register - register receive handler
4972 * @dev: device to register a handler for
4973 * @rx_handler: receive handler to register
4974 * @rx_handler_data: data pointer that is used by rx handler
4976 * Register a receive handler for a device. This handler will then be
4977 * called from __netif_receive_skb. A negative errno code is returned
4980 * The caller must hold the rtnl_mutex.
4982 * For a general description of rx_handler, see enum rx_handler_result.
4984 int netdev_rx_handler_register(struct net_device *dev,
4985 rx_handler_func_t *rx_handler,
4986 void *rx_handler_data)
4988 if (netdev_is_rx_handler_busy(dev))
4991 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4994 /* Note: rx_handler_data must be set before rx_handler */
4995 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4996 rcu_assign_pointer(dev->rx_handler, rx_handler);
5000 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5003 * netdev_rx_handler_unregister - unregister receive handler
5004 * @dev: device to unregister a handler from
5006 * Unregister a receive handler from a device.
5008 * The caller must hold the rtnl_mutex.
5010 void netdev_rx_handler_unregister(struct net_device *dev)
5014 RCU_INIT_POINTER(dev->rx_handler, NULL);
5015 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5016 * section has a guarantee to see a non NULL rx_handler_data
5020 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5022 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5025 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5026 * the special handling of PFMEMALLOC skbs.
5028 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5030 switch (skb->protocol) {
5031 case htons(ETH_P_ARP):
5032 case htons(ETH_P_IP):
5033 case htons(ETH_P_IPV6):
5034 case htons(ETH_P_8021Q):
5035 case htons(ETH_P_8021AD):
5042 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5043 int *ret, struct net_device *orig_dev)
5045 if (nf_hook_ingress_active(skb)) {
5049 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5054 ingress_retval = nf_hook_ingress(skb);
5056 return ingress_retval;
5061 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5062 struct packet_type **ppt_prev)
5064 struct packet_type *ptype, *pt_prev;
5065 rx_handler_func_t *rx_handler;
5066 struct sk_buff *skb = *pskb;
5067 struct net_device *orig_dev;
5068 bool deliver_exact = false;
5069 int ret = NET_RX_DROP;
5072 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5074 trace_netif_receive_skb(skb);
5076 orig_dev = skb->dev;
5078 skb_reset_network_header(skb);
5079 if (!skb_transport_header_was_set(skb))
5080 skb_reset_transport_header(skb);
5081 skb_reset_mac_len(skb);
5086 skb->skb_iif = skb->dev->ifindex;
5088 __this_cpu_inc(softnet_data.processed);
5090 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5094 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5097 if (ret2 != XDP_PASS) {
5101 skb_reset_mac_len(skb);
5104 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5105 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5106 skb = skb_vlan_untag(skb);
5111 if (skb_skip_tc_classify(skb))
5117 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5119 ret = deliver_skb(skb, pt_prev, orig_dev);
5123 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5125 ret = deliver_skb(skb, pt_prev, orig_dev);
5130 #ifdef CONFIG_NET_INGRESS
5131 if (static_branch_unlikely(&ingress_needed_key)) {
5132 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5136 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5140 skb_reset_redirect(skb);
5142 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5145 if (skb_vlan_tag_present(skb)) {
5147 ret = deliver_skb(skb, pt_prev, orig_dev);
5150 if (vlan_do_receive(&skb))
5152 else if (unlikely(!skb))
5156 rx_handler = rcu_dereference(skb->dev->rx_handler);
5159 ret = deliver_skb(skb, pt_prev, orig_dev);
5162 switch (rx_handler(&skb)) {
5163 case RX_HANDLER_CONSUMED:
5164 ret = NET_RX_SUCCESS;
5166 case RX_HANDLER_ANOTHER:
5168 case RX_HANDLER_EXACT:
5169 deliver_exact = true;
5170 case RX_HANDLER_PASS:
5177 if (unlikely(skb_vlan_tag_present(skb))) {
5179 if (skb_vlan_tag_get_id(skb)) {
5180 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5183 skb->pkt_type = PACKET_OTHERHOST;
5184 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5185 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5186 /* Outer header is 802.1P with vlan 0, inner header is
5187 * 802.1Q or 802.1AD and vlan_do_receive() above could
5188 * not find vlan dev for vlan id 0.
5190 __vlan_hwaccel_clear_tag(skb);
5191 skb = skb_vlan_untag(skb);
5194 if (vlan_do_receive(&skb))
5195 /* After stripping off 802.1P header with vlan 0
5196 * vlan dev is found for inner header.
5199 else if (unlikely(!skb))
5202 /* We have stripped outer 802.1P vlan 0 header.
5203 * But could not find vlan dev.
5204 * check again for vlan id to set OTHERHOST.
5208 /* Note: we might in the future use prio bits
5209 * and set skb->priority like in vlan_do_receive()
5210 * For the time being, just ignore Priority Code Point
5212 __vlan_hwaccel_clear_tag(skb);
5215 type = skb->protocol;
5217 /* deliver only exact match when indicated */
5218 if (likely(!deliver_exact)) {
5219 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5220 &ptype_base[ntohs(type) &
5224 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5225 &orig_dev->ptype_specific);
5227 if (unlikely(skb->dev != orig_dev)) {
5228 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5229 &skb->dev->ptype_specific);
5233 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5235 *ppt_prev = pt_prev;
5239 atomic_long_inc(&skb->dev->rx_dropped);
5241 atomic_long_inc(&skb->dev->rx_nohandler);
5243 /* Jamal, now you will not able to escape explaining
5244 * me how you were going to use this. :-)
5250 /* The invariant here is that if *ppt_prev is not NULL
5251 * then skb should also be non-NULL.
5253 * Apparently *ppt_prev assignment above holds this invariant due to
5254 * skb dereferencing near it.
5260 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5262 struct net_device *orig_dev = skb->dev;
5263 struct packet_type *pt_prev = NULL;
5266 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5268 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5269 skb->dev, pt_prev, orig_dev);
5274 * netif_receive_skb_core - special purpose version of netif_receive_skb
5275 * @skb: buffer to process
5277 * More direct receive version of netif_receive_skb(). It should
5278 * only be used by callers that have a need to skip RPS and Generic XDP.
5279 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5281 * This function may only be called from softirq context and interrupts
5282 * should be enabled.
5284 * Return values (usually ignored):
5285 * NET_RX_SUCCESS: no congestion
5286 * NET_RX_DROP: packet was dropped
5288 int netif_receive_skb_core(struct sk_buff *skb)
5293 ret = __netif_receive_skb_one_core(skb, false);
5298 EXPORT_SYMBOL(netif_receive_skb_core);
5300 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5301 struct packet_type *pt_prev,
5302 struct net_device *orig_dev)
5304 struct sk_buff *skb, *next;
5308 if (list_empty(head))
5310 if (pt_prev->list_func != NULL)
5311 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5312 ip_list_rcv, head, pt_prev, orig_dev);
5314 list_for_each_entry_safe(skb, next, head, list) {
5315 skb_list_del_init(skb);
5316 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5320 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5322 /* Fast-path assumptions:
5323 * - There is no RX handler.
5324 * - Only one packet_type matches.
5325 * If either of these fails, we will end up doing some per-packet
5326 * processing in-line, then handling the 'last ptype' for the whole
5327 * sublist. This can't cause out-of-order delivery to any single ptype,
5328 * because the 'last ptype' must be constant across the sublist, and all
5329 * other ptypes are handled per-packet.
5331 /* Current (common) ptype of sublist */
5332 struct packet_type *pt_curr = NULL;
5333 /* Current (common) orig_dev of sublist */
5334 struct net_device *od_curr = NULL;
5335 struct list_head sublist;
5336 struct sk_buff *skb, *next;
5338 INIT_LIST_HEAD(&sublist);
5339 list_for_each_entry_safe(skb, next, head, list) {
5340 struct net_device *orig_dev = skb->dev;
5341 struct packet_type *pt_prev = NULL;
5343 skb_list_del_init(skb);
5344 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5347 if (pt_curr != pt_prev || od_curr != orig_dev) {
5348 /* dispatch old sublist */
5349 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5350 /* start new sublist */
5351 INIT_LIST_HEAD(&sublist);
5355 list_add_tail(&skb->list, &sublist);
5358 /* dispatch final sublist */
5359 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5362 static int __netif_receive_skb(struct sk_buff *skb)
5366 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5367 unsigned int noreclaim_flag;
5370 * PFMEMALLOC skbs are special, they should
5371 * - be delivered to SOCK_MEMALLOC sockets only
5372 * - stay away from userspace
5373 * - have bounded memory usage
5375 * Use PF_MEMALLOC as this saves us from propagating the allocation
5376 * context down to all allocation sites.
5378 noreclaim_flag = memalloc_noreclaim_save();
5379 ret = __netif_receive_skb_one_core(skb, true);
5380 memalloc_noreclaim_restore(noreclaim_flag);
5382 ret = __netif_receive_skb_one_core(skb, false);
5387 static void __netif_receive_skb_list(struct list_head *head)
5389 unsigned long noreclaim_flag = 0;
5390 struct sk_buff *skb, *next;
5391 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5393 list_for_each_entry_safe(skb, next, head, list) {
5394 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5395 struct list_head sublist;
5397 /* Handle the previous sublist */
5398 list_cut_before(&sublist, head, &skb->list);
5399 if (!list_empty(&sublist))
5400 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5401 pfmemalloc = !pfmemalloc;
5402 /* See comments in __netif_receive_skb */
5404 noreclaim_flag = memalloc_noreclaim_save();
5406 memalloc_noreclaim_restore(noreclaim_flag);
5409 /* Handle the remaining sublist */
5410 if (!list_empty(head))
5411 __netif_receive_skb_list_core(head, pfmemalloc);
5412 /* Restore pflags */
5414 memalloc_noreclaim_restore(noreclaim_flag);
5417 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5419 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5420 struct bpf_prog *new = xdp->prog;
5426 /* generic XDP does not work with DEVMAPs that can
5427 * have a bpf_prog installed on an entry
5429 for (i = 0; i < new->aux->used_map_cnt; i++) {
5430 if (dev_map_can_have_prog(new->aux->used_maps[i]))
5435 switch (xdp->command) {
5436 case XDP_SETUP_PROG:
5437 rcu_assign_pointer(dev->xdp_prog, new);
5442 static_branch_dec(&generic_xdp_needed_key);
5443 } else if (new && !old) {
5444 static_branch_inc(&generic_xdp_needed_key);
5445 dev_disable_lro(dev);
5446 dev_disable_gro_hw(dev);
5450 case XDP_QUERY_PROG:
5451 xdp->prog_id = old ? old->aux->id : 0;
5462 static int netif_receive_skb_internal(struct sk_buff *skb)
5466 net_timestamp_check(netdev_tstamp_prequeue, skb);
5468 if (skb_defer_rx_timestamp(skb))
5469 return NET_RX_SUCCESS;
5473 if (static_branch_unlikely(&rps_needed)) {
5474 struct rps_dev_flow voidflow, *rflow = &voidflow;
5475 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5478 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5484 ret = __netif_receive_skb(skb);
5489 static void netif_receive_skb_list_internal(struct list_head *head)
5491 struct sk_buff *skb, *next;
5492 struct list_head sublist;
5494 INIT_LIST_HEAD(&sublist);
5495 list_for_each_entry_safe(skb, next, head, list) {
5496 net_timestamp_check(netdev_tstamp_prequeue, skb);
5497 skb_list_del_init(skb);
5498 if (!skb_defer_rx_timestamp(skb))
5499 list_add_tail(&skb->list, &sublist);
5501 list_splice_init(&sublist, head);
5505 if (static_branch_unlikely(&rps_needed)) {
5506 list_for_each_entry_safe(skb, next, head, list) {
5507 struct rps_dev_flow voidflow, *rflow = &voidflow;
5508 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5511 /* Will be handled, remove from list */
5512 skb_list_del_init(skb);
5513 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5518 __netif_receive_skb_list(head);
5523 * netif_receive_skb - process receive buffer from network
5524 * @skb: buffer to process
5526 * netif_receive_skb() is the main receive data processing function.
5527 * It always succeeds. The buffer may be dropped during processing
5528 * for congestion control or by the protocol layers.
5530 * This function may only be called from softirq context and interrupts
5531 * should be enabled.
5533 * Return values (usually ignored):
5534 * NET_RX_SUCCESS: no congestion
5535 * NET_RX_DROP: packet was dropped
5537 int netif_receive_skb(struct sk_buff *skb)
5541 trace_netif_receive_skb_entry(skb);
5543 ret = netif_receive_skb_internal(skb);
5544 trace_netif_receive_skb_exit(ret);
5548 EXPORT_SYMBOL(netif_receive_skb);
5551 * netif_receive_skb_list - process many receive buffers from network
5552 * @head: list of skbs to process.
5554 * Since return value of netif_receive_skb() is normally ignored, and
5555 * wouldn't be meaningful for a list, this function returns void.
5557 * This function may only be called from softirq context and interrupts
5558 * should be enabled.
5560 void netif_receive_skb_list(struct list_head *head)
5562 struct sk_buff *skb;
5564 if (list_empty(head))
5566 if (trace_netif_receive_skb_list_entry_enabled()) {
5567 list_for_each_entry(skb, head, list)
5568 trace_netif_receive_skb_list_entry(skb);
5570 netif_receive_skb_list_internal(head);
5571 trace_netif_receive_skb_list_exit(0);
5573 EXPORT_SYMBOL(netif_receive_skb_list);
5575 DEFINE_PER_CPU(struct work_struct, flush_works);
5577 /* Network device is going away, flush any packets still pending */
5578 static void flush_backlog(struct work_struct *work)
5580 struct sk_buff *skb, *tmp;
5581 struct softnet_data *sd;
5584 sd = this_cpu_ptr(&softnet_data);
5586 local_irq_disable();
5588 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5589 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5590 __skb_unlink(skb, &sd->input_pkt_queue);
5592 input_queue_head_incr(sd);
5598 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5599 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5600 __skb_unlink(skb, &sd->process_queue);
5602 input_queue_head_incr(sd);
5608 static void flush_all_backlogs(void)
5614 for_each_online_cpu(cpu)
5615 queue_work_on(cpu, system_highpri_wq,
5616 per_cpu_ptr(&flush_works, cpu));
5618 for_each_online_cpu(cpu)
5619 flush_work(per_cpu_ptr(&flush_works, cpu));
5624 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5625 static void gro_normal_list(struct napi_struct *napi)
5627 if (!napi->rx_count)
5629 netif_receive_skb_list_internal(&napi->rx_list);
5630 INIT_LIST_HEAD(&napi->rx_list);
5634 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5635 * pass the whole batch up to the stack.
5637 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5639 list_add_tail(&skb->list, &napi->rx_list);
5640 if (++napi->rx_count >= gro_normal_batch)
5641 gro_normal_list(napi);
5644 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5645 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5646 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5648 struct packet_offload *ptype;
5649 __be16 type = skb->protocol;
5650 struct list_head *head = &offload_base;
5653 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5655 if (NAPI_GRO_CB(skb)->count == 1) {
5656 skb_shinfo(skb)->gso_size = 0;
5661 list_for_each_entry_rcu(ptype, head, list) {
5662 if (ptype->type != type || !ptype->callbacks.gro_complete)
5665 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5666 ipv6_gro_complete, inet_gro_complete,
5673 WARN_ON(&ptype->list == head);
5675 return NET_RX_SUCCESS;
5679 gro_normal_one(napi, skb);
5680 return NET_RX_SUCCESS;
5683 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5686 struct list_head *head = &napi->gro_hash[index].list;
5687 struct sk_buff *skb, *p;
5689 list_for_each_entry_safe_reverse(skb, p, head, list) {
5690 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5692 skb_list_del_init(skb);
5693 napi_gro_complete(napi, skb);
5694 napi->gro_hash[index].count--;
5697 if (!napi->gro_hash[index].count)
5698 __clear_bit(index, &napi->gro_bitmask);
5701 /* napi->gro_hash[].list contains packets ordered by age.
5702 * youngest packets at the head of it.
5703 * Complete skbs in reverse order to reduce latencies.
5705 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5707 unsigned long bitmask = napi->gro_bitmask;
5708 unsigned int i, base = ~0U;
5710 while ((i = ffs(bitmask)) != 0) {
5713 __napi_gro_flush_chain(napi, base, flush_old);
5716 EXPORT_SYMBOL(napi_gro_flush);
5718 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5719 struct sk_buff *skb)
5721 unsigned int maclen = skb->dev->hard_header_len;
5722 u32 hash = skb_get_hash_raw(skb);
5723 struct list_head *head;
5726 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5727 list_for_each_entry(p, head, list) {
5728 unsigned long diffs;
5730 NAPI_GRO_CB(p)->flush = 0;
5732 if (hash != skb_get_hash_raw(p)) {
5733 NAPI_GRO_CB(p)->same_flow = 0;
5737 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5738 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5739 if (skb_vlan_tag_present(p))
5740 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5741 diffs |= skb_metadata_dst_cmp(p, skb);
5742 diffs |= skb_metadata_differs(p, skb);
5743 if (maclen == ETH_HLEN)
5744 diffs |= compare_ether_header(skb_mac_header(p),
5745 skb_mac_header(skb));
5747 diffs = memcmp(skb_mac_header(p),
5748 skb_mac_header(skb),
5750 NAPI_GRO_CB(p)->same_flow = !diffs;
5756 static void skb_gro_reset_offset(struct sk_buff *skb)
5758 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5759 const skb_frag_t *frag0 = &pinfo->frags[0];
5761 NAPI_GRO_CB(skb)->data_offset = 0;
5762 NAPI_GRO_CB(skb)->frag0 = NULL;
5763 NAPI_GRO_CB(skb)->frag0_len = 0;
5765 if (!skb_headlen(skb) && pinfo->nr_frags &&
5766 !PageHighMem(skb_frag_page(frag0))) {
5767 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5768 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5769 skb_frag_size(frag0),
5770 skb->end - skb->tail);
5774 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5776 struct skb_shared_info *pinfo = skb_shinfo(skb);
5778 BUG_ON(skb->end - skb->tail < grow);
5780 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5782 skb->data_len -= grow;
5785 skb_frag_off_add(&pinfo->frags[0], grow);
5786 skb_frag_size_sub(&pinfo->frags[0], grow);
5788 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5789 skb_frag_unref(skb, 0);
5790 memmove(pinfo->frags, pinfo->frags + 1,
5791 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5795 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5797 struct sk_buff *oldest;
5799 oldest = list_last_entry(head, struct sk_buff, list);
5801 /* We are called with head length >= MAX_GRO_SKBS, so this is
5804 if (WARN_ON_ONCE(!oldest))
5807 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5810 skb_list_del_init(oldest);
5811 napi_gro_complete(napi, oldest);
5814 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5816 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5818 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5820 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5821 struct list_head *head = &offload_base;
5822 struct packet_offload *ptype;
5823 __be16 type = skb->protocol;
5824 struct list_head *gro_head;
5825 struct sk_buff *pp = NULL;
5826 enum gro_result ret;
5830 if (netif_elide_gro(skb->dev))
5833 gro_head = gro_list_prepare(napi, skb);
5836 list_for_each_entry_rcu(ptype, head, list) {
5837 if (ptype->type != type || !ptype->callbacks.gro_receive)
5840 skb_set_network_header(skb, skb_gro_offset(skb));
5841 skb_reset_mac_len(skb);
5842 NAPI_GRO_CB(skb)->same_flow = 0;
5843 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5844 NAPI_GRO_CB(skb)->free = 0;
5845 NAPI_GRO_CB(skb)->encap_mark = 0;
5846 NAPI_GRO_CB(skb)->recursion_counter = 0;
5847 NAPI_GRO_CB(skb)->is_fou = 0;
5848 NAPI_GRO_CB(skb)->is_atomic = 1;
5849 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5851 /* Setup for GRO checksum validation */
5852 switch (skb->ip_summed) {
5853 case CHECKSUM_COMPLETE:
5854 NAPI_GRO_CB(skb)->csum = skb->csum;
5855 NAPI_GRO_CB(skb)->csum_valid = 1;
5856 NAPI_GRO_CB(skb)->csum_cnt = 0;
5858 case CHECKSUM_UNNECESSARY:
5859 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5860 NAPI_GRO_CB(skb)->csum_valid = 0;
5863 NAPI_GRO_CB(skb)->csum_cnt = 0;
5864 NAPI_GRO_CB(skb)->csum_valid = 0;
5867 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5868 ipv6_gro_receive, inet_gro_receive,
5874 if (&ptype->list == head)
5877 if (PTR_ERR(pp) == -EINPROGRESS) {
5882 same_flow = NAPI_GRO_CB(skb)->same_flow;
5883 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5886 skb_list_del_init(pp);
5887 napi_gro_complete(napi, pp);
5888 napi->gro_hash[hash].count--;
5894 if (NAPI_GRO_CB(skb)->flush)
5897 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5898 gro_flush_oldest(napi, gro_head);
5900 napi->gro_hash[hash].count++;
5902 NAPI_GRO_CB(skb)->count = 1;
5903 NAPI_GRO_CB(skb)->age = jiffies;
5904 NAPI_GRO_CB(skb)->last = skb;
5905 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5906 list_add(&skb->list, gro_head);
5910 grow = skb_gro_offset(skb) - skb_headlen(skb);
5912 gro_pull_from_frag0(skb, grow);
5914 if (napi->gro_hash[hash].count) {
5915 if (!test_bit(hash, &napi->gro_bitmask))
5916 __set_bit(hash, &napi->gro_bitmask);
5917 } else if (test_bit(hash, &napi->gro_bitmask)) {
5918 __clear_bit(hash, &napi->gro_bitmask);
5928 struct packet_offload *gro_find_receive_by_type(__be16 type)
5930 struct list_head *offload_head = &offload_base;
5931 struct packet_offload *ptype;
5933 list_for_each_entry_rcu(ptype, offload_head, list) {
5934 if (ptype->type != type || !ptype->callbacks.gro_receive)
5940 EXPORT_SYMBOL(gro_find_receive_by_type);
5942 struct packet_offload *gro_find_complete_by_type(__be16 type)
5944 struct list_head *offload_head = &offload_base;
5945 struct packet_offload *ptype;
5947 list_for_each_entry_rcu(ptype, offload_head, list) {
5948 if (ptype->type != type || !ptype->callbacks.gro_complete)
5954 EXPORT_SYMBOL(gro_find_complete_by_type);
5956 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5960 kmem_cache_free(skbuff_head_cache, skb);
5963 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5964 struct sk_buff *skb,
5969 gro_normal_one(napi, skb);
5976 case GRO_MERGED_FREE:
5977 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5978 napi_skb_free_stolen_head(skb);
5992 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5996 skb_mark_napi_id(skb, napi);
5997 trace_napi_gro_receive_entry(skb);
5999 skb_gro_reset_offset(skb);
6001 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6002 trace_napi_gro_receive_exit(ret);
6006 EXPORT_SYMBOL(napi_gro_receive);
6008 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6010 if (unlikely(skb->pfmemalloc)) {
6014 __skb_pull(skb, skb_headlen(skb));
6015 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6016 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6017 __vlan_hwaccel_clear_tag(skb);
6018 skb->dev = napi->dev;
6021 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6022 skb->pkt_type = PACKET_HOST;
6024 skb->encapsulation = 0;
6025 skb_shinfo(skb)->gso_type = 0;
6026 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6032 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6034 struct sk_buff *skb = napi->skb;
6037 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6040 skb_mark_napi_id(skb, napi);
6045 EXPORT_SYMBOL(napi_get_frags);
6047 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6048 struct sk_buff *skb,
6054 __skb_push(skb, ETH_HLEN);
6055 skb->protocol = eth_type_trans(skb, skb->dev);
6056 if (ret == GRO_NORMAL)
6057 gro_normal_one(napi, skb);
6061 napi_reuse_skb(napi, skb);
6064 case GRO_MERGED_FREE:
6065 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6066 napi_skb_free_stolen_head(skb);
6068 napi_reuse_skb(napi, skb);
6079 /* Upper GRO stack assumes network header starts at gro_offset=0
6080 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6081 * We copy ethernet header into skb->data to have a common layout.
6083 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6085 struct sk_buff *skb = napi->skb;
6086 const struct ethhdr *eth;
6087 unsigned int hlen = sizeof(*eth);
6091 skb_reset_mac_header(skb);
6092 skb_gro_reset_offset(skb);
6094 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6095 eth = skb_gro_header_slow(skb, hlen, 0);
6096 if (unlikely(!eth)) {
6097 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6098 __func__, napi->dev->name);
6099 napi_reuse_skb(napi, skb);
6103 eth = (const struct ethhdr *)skb->data;
6104 gro_pull_from_frag0(skb, hlen);
6105 NAPI_GRO_CB(skb)->frag0 += hlen;
6106 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6108 __skb_pull(skb, hlen);
6111 * This works because the only protocols we care about don't require
6113 * We'll fix it up properly in napi_frags_finish()
6115 skb->protocol = eth->h_proto;
6120 gro_result_t napi_gro_frags(struct napi_struct *napi)
6123 struct sk_buff *skb = napi_frags_skb(napi);
6128 trace_napi_gro_frags_entry(skb);
6130 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6131 trace_napi_gro_frags_exit(ret);
6135 EXPORT_SYMBOL(napi_gro_frags);
6137 /* Compute the checksum from gro_offset and return the folded value
6138 * after adding in any pseudo checksum.
6140 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6145 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6147 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6148 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6149 /* See comments in __skb_checksum_complete(). */
6151 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6152 !skb->csum_complete_sw)
6153 netdev_rx_csum_fault(skb->dev, skb);
6156 NAPI_GRO_CB(skb)->csum = wsum;
6157 NAPI_GRO_CB(skb)->csum_valid = 1;
6161 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6163 static void net_rps_send_ipi(struct softnet_data *remsd)
6167 struct softnet_data *next = remsd->rps_ipi_next;
6169 if (cpu_online(remsd->cpu))
6170 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6177 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6178 * Note: called with local irq disabled, but exits with local irq enabled.
6180 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6183 struct softnet_data *remsd = sd->rps_ipi_list;
6186 sd->rps_ipi_list = NULL;
6190 /* Send pending IPI's to kick RPS processing on remote cpus. */
6191 net_rps_send_ipi(remsd);
6197 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6200 return sd->rps_ipi_list != NULL;
6206 static int process_backlog(struct napi_struct *napi, int quota)
6208 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6212 /* Check if we have pending ipi, its better to send them now,
6213 * not waiting net_rx_action() end.
6215 if (sd_has_rps_ipi_waiting(sd)) {
6216 local_irq_disable();
6217 net_rps_action_and_irq_enable(sd);
6220 napi->weight = dev_rx_weight;
6222 struct sk_buff *skb;
6224 while ((skb = __skb_dequeue(&sd->process_queue))) {
6226 __netif_receive_skb(skb);
6228 input_queue_head_incr(sd);
6229 if (++work >= quota)
6234 local_irq_disable();
6236 if (skb_queue_empty(&sd->input_pkt_queue)) {
6238 * Inline a custom version of __napi_complete().
6239 * only current cpu owns and manipulates this napi,
6240 * and NAPI_STATE_SCHED is the only possible flag set
6242 * We can use a plain write instead of clear_bit(),
6243 * and we dont need an smp_mb() memory barrier.
6248 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6249 &sd->process_queue);
6259 * __napi_schedule - schedule for receive
6260 * @n: entry to schedule
6262 * The entry's receive function will be scheduled to run.
6263 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6265 void __napi_schedule(struct napi_struct *n)
6267 unsigned long flags;
6269 local_irq_save(flags);
6270 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6271 local_irq_restore(flags);
6273 EXPORT_SYMBOL(__napi_schedule);
6276 * napi_schedule_prep - check if napi can be scheduled
6279 * Test if NAPI routine is already running, and if not mark
6280 * it as running. This is used as a condition variable
6281 * insure only one NAPI poll instance runs. We also make
6282 * sure there is no pending NAPI disable.
6284 bool napi_schedule_prep(struct napi_struct *n)
6286 unsigned long val, new;
6289 val = READ_ONCE(n->state);
6290 if (unlikely(val & NAPIF_STATE_DISABLE))
6292 new = val | NAPIF_STATE_SCHED;
6294 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6295 * This was suggested by Alexander Duyck, as compiler
6296 * emits better code than :
6297 * if (val & NAPIF_STATE_SCHED)
6298 * new |= NAPIF_STATE_MISSED;
6300 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6302 } while (cmpxchg(&n->state, val, new) != val);
6304 return !(val & NAPIF_STATE_SCHED);
6306 EXPORT_SYMBOL(napi_schedule_prep);
6309 * __napi_schedule_irqoff - schedule for receive
6310 * @n: entry to schedule
6312 * Variant of __napi_schedule() assuming hard irqs are masked
6314 void __napi_schedule_irqoff(struct napi_struct *n)
6316 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6318 EXPORT_SYMBOL(__napi_schedule_irqoff);
6320 bool napi_complete_done(struct napi_struct *n, int work_done)
6322 unsigned long flags, val, new, timeout = 0;
6326 * 1) Don't let napi dequeue from the cpu poll list
6327 * just in case its running on a different cpu.
6328 * 2) If we are busy polling, do nothing here, we have
6329 * the guarantee we will be called later.
6331 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6332 NAPIF_STATE_IN_BUSY_POLL)))
6337 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6338 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6340 if (n->defer_hard_irqs_count > 0) {
6341 n->defer_hard_irqs_count--;
6342 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6346 if (n->gro_bitmask) {
6347 /* When the NAPI instance uses a timeout and keeps postponing
6348 * it, we need to bound somehow the time packets are kept in
6351 napi_gro_flush(n, !!timeout);
6356 if (unlikely(!list_empty(&n->poll_list))) {
6357 /* If n->poll_list is not empty, we need to mask irqs */
6358 local_irq_save(flags);
6359 list_del_init(&n->poll_list);
6360 local_irq_restore(flags);
6364 val = READ_ONCE(n->state);
6366 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6368 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6370 /* If STATE_MISSED was set, leave STATE_SCHED set,
6371 * because we will call napi->poll() one more time.
6372 * This C code was suggested by Alexander Duyck to help gcc.
6374 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6376 } while (cmpxchg(&n->state, val, new) != val);
6378 if (unlikely(val & NAPIF_STATE_MISSED)) {
6384 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6385 HRTIMER_MODE_REL_PINNED);
6388 EXPORT_SYMBOL(napi_complete_done);
6390 /* must be called under rcu_read_lock(), as we dont take a reference */
6391 static struct napi_struct *napi_by_id(unsigned int napi_id)
6393 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6394 struct napi_struct *napi;
6396 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6397 if (napi->napi_id == napi_id)
6403 #if defined(CONFIG_NET_RX_BUSY_POLL)
6405 #define BUSY_POLL_BUDGET 8
6407 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6411 /* Busy polling means there is a high chance device driver hard irq
6412 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6413 * set in napi_schedule_prep().
6414 * Since we are about to call napi->poll() once more, we can safely
6415 * clear NAPI_STATE_MISSED.
6417 * Note: x86 could use a single "lock and ..." instruction
6418 * to perform these two clear_bit()
6420 clear_bit(NAPI_STATE_MISSED, &napi->state);
6421 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6425 /* All we really want here is to re-enable device interrupts.
6426 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6428 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6429 /* We can't gro_normal_list() here, because napi->poll() might have
6430 * rearmed the napi (napi_complete_done()) in which case it could
6431 * already be running on another CPU.
6433 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6434 netpoll_poll_unlock(have_poll_lock);
6435 if (rc == BUSY_POLL_BUDGET) {
6436 /* As the whole budget was spent, we still own the napi so can
6437 * safely handle the rx_list.
6439 gro_normal_list(napi);
6440 __napi_schedule(napi);
6445 void napi_busy_loop(unsigned int napi_id,
6446 bool (*loop_end)(void *, unsigned long),
6449 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6450 int (*napi_poll)(struct napi_struct *napi, int budget);
6451 void *have_poll_lock = NULL;
6452 struct napi_struct *napi;
6459 napi = napi_by_id(napi_id);
6469 unsigned long val = READ_ONCE(napi->state);
6471 /* If multiple threads are competing for this napi,
6472 * we avoid dirtying napi->state as much as we can.
6474 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6475 NAPIF_STATE_IN_BUSY_POLL))
6477 if (cmpxchg(&napi->state, val,
6478 val | NAPIF_STATE_IN_BUSY_POLL |
6479 NAPIF_STATE_SCHED) != val)
6481 have_poll_lock = netpoll_poll_lock(napi);
6482 napi_poll = napi->poll;
6484 work = napi_poll(napi, BUSY_POLL_BUDGET);
6485 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6486 gro_normal_list(napi);
6489 __NET_ADD_STATS(dev_net(napi->dev),
6490 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6493 if (!loop_end || loop_end(loop_end_arg, start_time))
6496 if (unlikely(need_resched())) {
6498 busy_poll_stop(napi, have_poll_lock);
6502 if (loop_end(loop_end_arg, start_time))
6509 busy_poll_stop(napi, have_poll_lock);
6514 EXPORT_SYMBOL(napi_busy_loop);
6516 #endif /* CONFIG_NET_RX_BUSY_POLL */
6518 static void napi_hash_add(struct napi_struct *napi)
6520 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6521 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6524 spin_lock(&napi_hash_lock);
6526 /* 0..NR_CPUS range is reserved for sender_cpu use */
6528 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6529 napi_gen_id = MIN_NAPI_ID;
6530 } while (napi_by_id(napi_gen_id));
6531 napi->napi_id = napi_gen_id;
6533 hlist_add_head_rcu(&napi->napi_hash_node,
6534 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6536 spin_unlock(&napi_hash_lock);
6539 /* Warning : caller is responsible to make sure rcu grace period
6540 * is respected before freeing memory containing @napi
6542 bool napi_hash_del(struct napi_struct *napi)
6544 bool rcu_sync_needed = false;
6546 spin_lock(&napi_hash_lock);
6548 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6549 rcu_sync_needed = true;
6550 hlist_del_rcu(&napi->napi_hash_node);
6552 spin_unlock(&napi_hash_lock);
6553 return rcu_sync_needed;
6555 EXPORT_SYMBOL_GPL(napi_hash_del);
6557 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6559 struct napi_struct *napi;
6561 napi = container_of(timer, struct napi_struct, timer);
6563 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6564 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6566 if (!napi_disable_pending(napi) &&
6567 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6568 __napi_schedule_irqoff(napi);
6570 return HRTIMER_NORESTART;
6573 static void init_gro_hash(struct napi_struct *napi)
6577 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6578 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6579 napi->gro_hash[i].count = 0;
6581 napi->gro_bitmask = 0;
6584 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6585 int (*poll)(struct napi_struct *, int), int weight)
6587 INIT_LIST_HEAD(&napi->poll_list);
6588 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6589 napi->timer.function = napi_watchdog;
6590 init_gro_hash(napi);
6592 INIT_LIST_HEAD(&napi->rx_list);
6595 if (weight > NAPI_POLL_WEIGHT)
6596 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6598 napi->weight = weight;
6599 list_add(&napi->dev_list, &dev->napi_list);
6601 #ifdef CONFIG_NETPOLL
6602 napi->poll_owner = -1;
6604 set_bit(NAPI_STATE_SCHED, &napi->state);
6605 napi_hash_add(napi);
6607 EXPORT_SYMBOL(netif_napi_add);
6609 void napi_disable(struct napi_struct *n)
6612 set_bit(NAPI_STATE_DISABLE, &n->state);
6614 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6616 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6619 hrtimer_cancel(&n->timer);
6621 clear_bit(NAPI_STATE_DISABLE, &n->state);
6623 EXPORT_SYMBOL(napi_disable);
6625 static void flush_gro_hash(struct napi_struct *napi)
6629 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6630 struct sk_buff *skb, *n;
6632 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6634 napi->gro_hash[i].count = 0;
6638 /* Must be called in process context */
6639 void netif_napi_del(struct napi_struct *napi)
6642 if (napi_hash_del(napi))
6644 list_del_init(&napi->dev_list);
6645 napi_free_frags(napi);
6647 flush_gro_hash(napi);
6648 napi->gro_bitmask = 0;
6650 EXPORT_SYMBOL(netif_napi_del);
6652 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6657 list_del_init(&n->poll_list);
6659 have = netpoll_poll_lock(n);
6663 /* This NAPI_STATE_SCHED test is for avoiding a race
6664 * with netpoll's poll_napi(). Only the entity which
6665 * obtains the lock and sees NAPI_STATE_SCHED set will
6666 * actually make the ->poll() call. Therefore we avoid
6667 * accidentally calling ->poll() when NAPI is not scheduled.
6670 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6671 work = n->poll(n, weight);
6672 trace_napi_poll(n, work, weight);
6675 WARN_ON_ONCE(work > weight);
6677 if (likely(work < weight))
6680 /* Drivers must not modify the NAPI state if they
6681 * consume the entire weight. In such cases this code
6682 * still "owns" the NAPI instance and therefore can
6683 * move the instance around on the list at-will.
6685 if (unlikely(napi_disable_pending(n))) {
6690 if (n->gro_bitmask) {
6691 /* flush too old packets
6692 * If HZ < 1000, flush all packets.
6694 napi_gro_flush(n, HZ >= 1000);
6699 /* Some drivers may have called napi_schedule
6700 * prior to exhausting their budget.
6702 if (unlikely(!list_empty(&n->poll_list))) {
6703 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6704 n->dev ? n->dev->name : "backlog");
6708 list_add_tail(&n->poll_list, repoll);
6711 netpoll_poll_unlock(have);
6716 static __latent_entropy void net_rx_action(struct softirq_action *h)
6718 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6719 unsigned long time_limit = jiffies +
6720 usecs_to_jiffies(netdev_budget_usecs);
6721 int budget = netdev_budget;
6725 local_irq_disable();
6726 list_splice_init(&sd->poll_list, &list);
6730 struct napi_struct *n;
6732 if (list_empty(&list)) {
6733 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6738 n = list_first_entry(&list, struct napi_struct, poll_list);
6739 budget -= napi_poll(n, &repoll);
6741 /* If softirq window is exhausted then punt.
6742 * Allow this to run for 2 jiffies since which will allow
6743 * an average latency of 1.5/HZ.
6745 if (unlikely(budget <= 0 ||
6746 time_after_eq(jiffies, time_limit))) {
6752 local_irq_disable();
6754 list_splice_tail_init(&sd->poll_list, &list);
6755 list_splice_tail(&repoll, &list);
6756 list_splice(&list, &sd->poll_list);
6757 if (!list_empty(&sd->poll_list))
6758 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6760 net_rps_action_and_irq_enable(sd);
6762 __kfree_skb_flush();
6765 struct netdev_adjacent {
6766 struct net_device *dev;
6768 /* upper master flag, there can only be one master device per list */
6771 /* lookup ignore flag */
6774 /* counter for the number of times this device was added to us */
6777 /* private field for the users */
6780 struct list_head list;
6781 struct rcu_head rcu;
6784 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6785 struct list_head *adj_list)
6787 struct netdev_adjacent *adj;
6789 list_for_each_entry(adj, adj_list, list) {
6790 if (adj->dev == adj_dev)
6796 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6798 struct net_device *dev = data;
6800 return upper_dev == dev;
6804 * netdev_has_upper_dev - Check if device is linked to an upper device
6806 * @upper_dev: upper device to check
6808 * Find out if a device is linked to specified upper device and return true
6809 * in case it is. Note that this checks only immediate upper device,
6810 * not through a complete stack of devices. The caller must hold the RTNL lock.
6812 bool netdev_has_upper_dev(struct net_device *dev,
6813 struct net_device *upper_dev)
6817 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6820 EXPORT_SYMBOL(netdev_has_upper_dev);
6823 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6825 * @upper_dev: upper device to check
6827 * Find out if a device is linked to specified upper device and return true
6828 * in case it is. Note that this checks the entire upper device chain.
6829 * The caller must hold rcu lock.
6832 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6833 struct net_device *upper_dev)
6835 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6838 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6841 * netdev_has_any_upper_dev - Check if device is linked to some device
6844 * Find out if a device is linked to an upper device and return true in case
6845 * it is. The caller must hold the RTNL lock.
6847 bool netdev_has_any_upper_dev(struct net_device *dev)
6851 return !list_empty(&dev->adj_list.upper);
6853 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6856 * netdev_master_upper_dev_get - Get master upper device
6859 * Find a master upper device and return pointer to it or NULL in case
6860 * it's not there. The caller must hold the RTNL lock.
6862 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6864 struct netdev_adjacent *upper;
6868 if (list_empty(&dev->adj_list.upper))
6871 upper = list_first_entry(&dev->adj_list.upper,
6872 struct netdev_adjacent, list);
6873 if (likely(upper->master))
6877 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6879 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6881 struct netdev_adjacent *upper;
6885 if (list_empty(&dev->adj_list.upper))
6888 upper = list_first_entry(&dev->adj_list.upper,
6889 struct netdev_adjacent, list);
6890 if (likely(upper->master) && !upper->ignore)
6896 * netdev_has_any_lower_dev - Check if device is linked to some device
6899 * Find out if a device is linked to a lower device and return true in case
6900 * it is. The caller must hold the RTNL lock.
6902 static bool netdev_has_any_lower_dev(struct net_device *dev)
6906 return !list_empty(&dev->adj_list.lower);
6909 void *netdev_adjacent_get_private(struct list_head *adj_list)
6911 struct netdev_adjacent *adj;
6913 adj = list_entry(adj_list, struct netdev_adjacent, list);
6915 return adj->private;
6917 EXPORT_SYMBOL(netdev_adjacent_get_private);
6920 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6922 * @iter: list_head ** of the current position
6924 * Gets the next device from the dev's upper list, starting from iter
6925 * position. The caller must hold RCU read lock.
6927 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6928 struct list_head **iter)
6930 struct netdev_adjacent *upper;
6932 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6934 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6936 if (&upper->list == &dev->adj_list.upper)
6939 *iter = &upper->list;
6943 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6945 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6946 struct list_head **iter,
6949 struct netdev_adjacent *upper;
6951 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6953 if (&upper->list == &dev->adj_list.upper)
6956 *iter = &upper->list;
6957 *ignore = upper->ignore;
6962 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6963 struct list_head **iter)
6965 struct netdev_adjacent *upper;
6967 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6969 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6971 if (&upper->list == &dev->adj_list.upper)
6974 *iter = &upper->list;
6979 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6980 int (*fn)(struct net_device *dev,
6984 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6985 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6990 iter = &dev->adj_list.upper;
6994 ret = fn(now, data);
7001 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7008 niter = &udev->adj_list.upper;
7009 dev_stack[cur] = now;
7010 iter_stack[cur++] = iter;
7017 next = dev_stack[--cur];
7018 niter = iter_stack[cur];
7028 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7029 int (*fn)(struct net_device *dev,
7033 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7034 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7038 iter = &dev->adj_list.upper;
7042 ret = fn(now, data);
7049 udev = netdev_next_upper_dev_rcu(now, &iter);
7054 niter = &udev->adj_list.upper;
7055 dev_stack[cur] = now;
7056 iter_stack[cur++] = iter;
7063 next = dev_stack[--cur];
7064 niter = iter_stack[cur];
7073 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7075 static bool __netdev_has_upper_dev(struct net_device *dev,
7076 struct net_device *upper_dev)
7080 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7085 * netdev_lower_get_next_private - Get the next ->private from the
7086 * lower neighbour list
7088 * @iter: list_head ** of the current position
7090 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7091 * list, starting from iter position. The caller must hold either hold the
7092 * RTNL lock or its own locking that guarantees that the neighbour lower
7093 * list will remain unchanged.
7095 void *netdev_lower_get_next_private(struct net_device *dev,
7096 struct list_head **iter)
7098 struct netdev_adjacent *lower;
7100 lower = list_entry(*iter, struct netdev_adjacent, list);
7102 if (&lower->list == &dev->adj_list.lower)
7105 *iter = lower->list.next;
7107 return lower->private;
7109 EXPORT_SYMBOL(netdev_lower_get_next_private);
7112 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7113 * lower neighbour list, RCU
7116 * @iter: list_head ** of the current position
7118 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7119 * list, starting from iter position. The caller must hold RCU read lock.
7121 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7122 struct list_head **iter)
7124 struct netdev_adjacent *lower;
7126 WARN_ON_ONCE(!rcu_read_lock_held());
7128 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7130 if (&lower->list == &dev->adj_list.lower)
7133 *iter = &lower->list;
7135 return lower->private;
7137 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7140 * netdev_lower_get_next - Get the next device from the lower neighbour
7143 * @iter: list_head ** of the current position
7145 * Gets the next netdev_adjacent from the dev's lower neighbour
7146 * list, starting from iter position. The caller must hold RTNL lock or
7147 * its own locking that guarantees that the neighbour lower
7148 * list will remain unchanged.
7150 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7152 struct netdev_adjacent *lower;
7154 lower = list_entry(*iter, struct netdev_adjacent, list);
7156 if (&lower->list == &dev->adj_list.lower)
7159 *iter = lower->list.next;
7163 EXPORT_SYMBOL(netdev_lower_get_next);
7165 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7166 struct list_head **iter)
7168 struct netdev_adjacent *lower;
7170 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7172 if (&lower->list == &dev->adj_list.lower)
7175 *iter = &lower->list;
7180 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7181 struct list_head **iter,
7184 struct netdev_adjacent *lower;
7186 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7188 if (&lower->list == &dev->adj_list.lower)
7191 *iter = &lower->list;
7192 *ignore = lower->ignore;
7197 int netdev_walk_all_lower_dev(struct net_device *dev,
7198 int (*fn)(struct net_device *dev,
7202 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7203 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7207 iter = &dev->adj_list.lower;
7211 ret = fn(now, data);
7218 ldev = netdev_next_lower_dev(now, &iter);
7223 niter = &ldev->adj_list.lower;
7224 dev_stack[cur] = now;
7225 iter_stack[cur++] = iter;
7232 next = dev_stack[--cur];
7233 niter = iter_stack[cur];
7242 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7244 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7245 int (*fn)(struct net_device *dev,
7249 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7250 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7255 iter = &dev->adj_list.lower;
7259 ret = fn(now, data);
7266 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7273 niter = &ldev->adj_list.lower;
7274 dev_stack[cur] = now;
7275 iter_stack[cur++] = iter;
7282 next = dev_stack[--cur];
7283 niter = iter_stack[cur];
7293 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7294 struct list_head **iter)
7296 struct netdev_adjacent *lower;
7298 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7299 if (&lower->list == &dev->adj_list.lower)
7302 *iter = &lower->list;
7306 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7308 static u8 __netdev_upper_depth(struct net_device *dev)
7310 struct net_device *udev;
7311 struct list_head *iter;
7315 for (iter = &dev->adj_list.upper,
7316 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7318 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7321 if (max_depth < udev->upper_level)
7322 max_depth = udev->upper_level;
7328 static u8 __netdev_lower_depth(struct net_device *dev)
7330 struct net_device *ldev;
7331 struct list_head *iter;
7335 for (iter = &dev->adj_list.lower,
7336 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7338 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7341 if (max_depth < ldev->lower_level)
7342 max_depth = ldev->lower_level;
7348 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7350 dev->upper_level = __netdev_upper_depth(dev) + 1;
7354 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7356 dev->lower_level = __netdev_lower_depth(dev) + 1;
7360 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7361 int (*fn)(struct net_device *dev,
7365 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7366 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7370 iter = &dev->adj_list.lower;
7374 ret = fn(now, data);
7381 ldev = netdev_next_lower_dev_rcu(now, &iter);
7386 niter = &ldev->adj_list.lower;
7387 dev_stack[cur] = now;
7388 iter_stack[cur++] = iter;
7395 next = dev_stack[--cur];
7396 niter = iter_stack[cur];
7405 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7408 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7409 * lower neighbour list, RCU
7413 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7414 * list. The caller must hold RCU read lock.
7416 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7418 struct netdev_adjacent *lower;
7420 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7421 struct netdev_adjacent, list);
7423 return lower->private;
7426 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7429 * netdev_master_upper_dev_get_rcu - Get master upper device
7432 * Find a master upper device and return pointer to it or NULL in case
7433 * it's not there. The caller must hold the RCU read lock.
7435 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7437 struct netdev_adjacent *upper;
7439 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7440 struct netdev_adjacent, list);
7441 if (upper && likely(upper->master))
7445 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7447 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7448 struct net_device *adj_dev,
7449 struct list_head *dev_list)
7451 char linkname[IFNAMSIZ+7];
7453 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7454 "upper_%s" : "lower_%s", adj_dev->name);
7455 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7458 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7460 struct list_head *dev_list)
7462 char linkname[IFNAMSIZ+7];
7464 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7465 "upper_%s" : "lower_%s", name);
7466 sysfs_remove_link(&(dev->dev.kobj), linkname);
7469 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7470 struct net_device *adj_dev,
7471 struct list_head *dev_list)
7473 return (dev_list == &dev->adj_list.upper ||
7474 dev_list == &dev->adj_list.lower) &&
7475 net_eq(dev_net(dev), dev_net(adj_dev));
7478 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7479 struct net_device *adj_dev,
7480 struct list_head *dev_list,
7481 void *private, bool master)
7483 struct netdev_adjacent *adj;
7486 adj = __netdev_find_adj(adj_dev, dev_list);
7490 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7491 dev->name, adj_dev->name, adj->ref_nr);
7496 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7501 adj->master = master;
7503 adj->private = private;
7504 adj->ignore = false;
7507 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7508 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7510 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7511 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7516 /* Ensure that master link is always the first item in list. */
7518 ret = sysfs_create_link(&(dev->dev.kobj),
7519 &(adj_dev->dev.kobj), "master");
7521 goto remove_symlinks;
7523 list_add_rcu(&adj->list, dev_list);
7525 list_add_tail_rcu(&adj->list, dev_list);
7531 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7532 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7540 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7541 struct net_device *adj_dev,
7543 struct list_head *dev_list)
7545 struct netdev_adjacent *adj;
7547 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7548 dev->name, adj_dev->name, ref_nr);
7550 adj = __netdev_find_adj(adj_dev, dev_list);
7553 pr_err("Adjacency does not exist for device %s from %s\n",
7554 dev->name, adj_dev->name);
7559 if (adj->ref_nr > ref_nr) {
7560 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7561 dev->name, adj_dev->name, ref_nr,
7562 adj->ref_nr - ref_nr);
7563 adj->ref_nr -= ref_nr;
7568 sysfs_remove_link(&(dev->dev.kobj), "master");
7570 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7571 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7573 list_del_rcu(&adj->list);
7574 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7575 adj_dev->name, dev->name, adj_dev->name);
7577 kfree_rcu(adj, rcu);
7580 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7581 struct net_device *upper_dev,
7582 struct list_head *up_list,
7583 struct list_head *down_list,
7584 void *private, bool master)
7588 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7593 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7596 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7603 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7604 struct net_device *upper_dev,
7606 struct list_head *up_list,
7607 struct list_head *down_list)
7609 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7610 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7613 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7614 struct net_device *upper_dev,
7615 void *private, bool master)
7617 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7618 &dev->adj_list.upper,
7619 &upper_dev->adj_list.lower,
7623 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7624 struct net_device *upper_dev)
7626 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7627 &dev->adj_list.upper,
7628 &upper_dev->adj_list.lower);
7631 static int __netdev_upper_dev_link(struct net_device *dev,
7632 struct net_device *upper_dev, bool master,
7633 void *upper_priv, void *upper_info,
7634 struct netlink_ext_ack *extack)
7636 struct netdev_notifier_changeupper_info changeupper_info = {
7641 .upper_dev = upper_dev,
7644 .upper_info = upper_info,
7646 struct net_device *master_dev;
7651 if (dev == upper_dev)
7654 /* To prevent loops, check if dev is not upper device to upper_dev. */
7655 if (__netdev_has_upper_dev(upper_dev, dev))
7658 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7662 if (__netdev_has_upper_dev(dev, upper_dev))
7665 master_dev = __netdev_master_upper_dev_get(dev);
7667 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7670 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7671 &changeupper_info.info);
7672 ret = notifier_to_errno(ret);
7676 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7681 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7682 &changeupper_info.info);
7683 ret = notifier_to_errno(ret);
7687 __netdev_update_upper_level(dev, NULL);
7688 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7690 __netdev_update_lower_level(upper_dev, NULL);
7691 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7697 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7703 * netdev_upper_dev_link - Add a link to the upper device
7705 * @upper_dev: new upper device
7706 * @extack: netlink extended ack
7708 * Adds a link to device which is upper to this one. The caller must hold
7709 * the RTNL lock. On a failure a negative errno code is returned.
7710 * On success the reference counts are adjusted and the function
7713 int netdev_upper_dev_link(struct net_device *dev,
7714 struct net_device *upper_dev,
7715 struct netlink_ext_ack *extack)
7717 return __netdev_upper_dev_link(dev, upper_dev, false,
7718 NULL, NULL, extack);
7720 EXPORT_SYMBOL(netdev_upper_dev_link);
7723 * netdev_master_upper_dev_link - Add a master link to the upper device
7725 * @upper_dev: new upper device
7726 * @upper_priv: upper device private
7727 * @upper_info: upper info to be passed down via notifier
7728 * @extack: netlink extended ack
7730 * Adds a link to device which is upper to this one. In this case, only
7731 * one master upper device can be linked, although other non-master devices
7732 * might be linked as well. The caller must hold the RTNL lock.
7733 * On a failure a negative errno code is returned. On success the reference
7734 * counts are adjusted and the function returns zero.
7736 int netdev_master_upper_dev_link(struct net_device *dev,
7737 struct net_device *upper_dev,
7738 void *upper_priv, void *upper_info,
7739 struct netlink_ext_ack *extack)
7741 return __netdev_upper_dev_link(dev, upper_dev, true,
7742 upper_priv, upper_info, extack);
7744 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7747 * netdev_upper_dev_unlink - Removes a link to upper device
7749 * @upper_dev: new upper device
7751 * Removes a link to device which is upper to this one. The caller must hold
7754 void netdev_upper_dev_unlink(struct net_device *dev,
7755 struct net_device *upper_dev)
7757 struct netdev_notifier_changeupper_info changeupper_info = {
7761 .upper_dev = upper_dev,
7767 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7769 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7770 &changeupper_info.info);
7772 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7774 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7775 &changeupper_info.info);
7777 __netdev_update_upper_level(dev, NULL);
7778 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7780 __netdev_update_lower_level(upper_dev, NULL);
7781 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7784 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7786 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7787 struct net_device *lower_dev,
7790 struct netdev_adjacent *adj;
7792 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7796 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7801 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7802 struct net_device *lower_dev)
7804 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7807 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7808 struct net_device *lower_dev)
7810 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7813 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7814 struct net_device *new_dev,
7815 struct net_device *dev,
7816 struct netlink_ext_ack *extack)
7823 if (old_dev && new_dev != old_dev)
7824 netdev_adjacent_dev_disable(dev, old_dev);
7826 err = netdev_upper_dev_link(new_dev, dev, extack);
7828 if (old_dev && new_dev != old_dev)
7829 netdev_adjacent_dev_enable(dev, old_dev);
7835 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7837 void netdev_adjacent_change_commit(struct net_device *old_dev,
7838 struct net_device *new_dev,
7839 struct net_device *dev)
7841 if (!new_dev || !old_dev)
7844 if (new_dev == old_dev)
7847 netdev_adjacent_dev_enable(dev, old_dev);
7848 netdev_upper_dev_unlink(old_dev, dev);
7850 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7852 void netdev_adjacent_change_abort(struct net_device *old_dev,
7853 struct net_device *new_dev,
7854 struct net_device *dev)
7859 if (old_dev && new_dev != old_dev)
7860 netdev_adjacent_dev_enable(dev, old_dev);
7862 netdev_upper_dev_unlink(new_dev, dev);
7864 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7867 * netdev_bonding_info_change - Dispatch event about slave change
7869 * @bonding_info: info to dispatch
7871 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7872 * The caller must hold the RTNL lock.
7874 void netdev_bonding_info_change(struct net_device *dev,
7875 struct netdev_bonding_info *bonding_info)
7877 struct netdev_notifier_bonding_info info = {
7881 memcpy(&info.bonding_info, bonding_info,
7882 sizeof(struct netdev_bonding_info));
7883 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7886 EXPORT_SYMBOL(netdev_bonding_info_change);
7889 * netdev_get_xmit_slave - Get the xmit slave of master device
7891 * @all_slaves: assume all the slaves are active
7893 * The reference counters are not incremented so the caller must be
7894 * careful with locks. The caller must hold RCU lock.
7895 * %NULL is returned if no slave is found.
7898 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
7899 struct sk_buff *skb,
7902 const struct net_device_ops *ops = dev->netdev_ops;
7904 if (!ops->ndo_get_xmit_slave)
7906 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
7908 EXPORT_SYMBOL(netdev_get_xmit_slave);
7910 static void netdev_adjacent_add_links(struct net_device *dev)
7912 struct netdev_adjacent *iter;
7914 struct net *net = dev_net(dev);
7916 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7917 if (!net_eq(net, dev_net(iter->dev)))
7919 netdev_adjacent_sysfs_add(iter->dev, dev,
7920 &iter->dev->adj_list.lower);
7921 netdev_adjacent_sysfs_add(dev, iter->dev,
7922 &dev->adj_list.upper);
7925 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7926 if (!net_eq(net, dev_net(iter->dev)))
7928 netdev_adjacent_sysfs_add(iter->dev, dev,
7929 &iter->dev->adj_list.upper);
7930 netdev_adjacent_sysfs_add(dev, iter->dev,
7931 &dev->adj_list.lower);
7935 static void netdev_adjacent_del_links(struct net_device *dev)
7937 struct netdev_adjacent *iter;
7939 struct net *net = dev_net(dev);
7941 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7942 if (!net_eq(net, dev_net(iter->dev)))
7944 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7945 &iter->dev->adj_list.lower);
7946 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7947 &dev->adj_list.upper);
7950 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7951 if (!net_eq(net, dev_net(iter->dev)))
7953 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7954 &iter->dev->adj_list.upper);
7955 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7956 &dev->adj_list.lower);
7960 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7962 struct netdev_adjacent *iter;
7964 struct net *net = dev_net(dev);
7966 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7967 if (!net_eq(net, dev_net(iter->dev)))
7969 netdev_adjacent_sysfs_del(iter->dev, oldname,
7970 &iter->dev->adj_list.lower);
7971 netdev_adjacent_sysfs_add(iter->dev, dev,
7972 &iter->dev->adj_list.lower);
7975 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7976 if (!net_eq(net, dev_net(iter->dev)))
7978 netdev_adjacent_sysfs_del(iter->dev, oldname,
7979 &iter->dev->adj_list.upper);
7980 netdev_adjacent_sysfs_add(iter->dev, dev,
7981 &iter->dev->adj_list.upper);
7985 void *netdev_lower_dev_get_private(struct net_device *dev,
7986 struct net_device *lower_dev)
7988 struct netdev_adjacent *lower;
7992 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7996 return lower->private;
7998 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8002 * netdev_lower_change - Dispatch event about lower device state change
8003 * @lower_dev: device
8004 * @lower_state_info: state to dispatch
8006 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8007 * The caller must hold the RTNL lock.
8009 void netdev_lower_state_changed(struct net_device *lower_dev,
8010 void *lower_state_info)
8012 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8013 .info.dev = lower_dev,
8017 changelowerstate_info.lower_state_info = lower_state_info;
8018 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8019 &changelowerstate_info.info);
8021 EXPORT_SYMBOL(netdev_lower_state_changed);
8023 static void dev_change_rx_flags(struct net_device *dev, int flags)
8025 const struct net_device_ops *ops = dev->netdev_ops;
8027 if (ops->ndo_change_rx_flags)
8028 ops->ndo_change_rx_flags(dev, flags);
8031 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8033 unsigned int old_flags = dev->flags;
8039 dev->flags |= IFF_PROMISC;
8040 dev->promiscuity += inc;
8041 if (dev->promiscuity == 0) {
8044 * If inc causes overflow, untouch promisc and return error.
8047 dev->flags &= ~IFF_PROMISC;
8049 dev->promiscuity -= inc;
8050 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8055 if (dev->flags != old_flags) {
8056 pr_info("device %s %s promiscuous mode\n",
8058 dev->flags & IFF_PROMISC ? "entered" : "left");
8059 if (audit_enabled) {
8060 current_uid_gid(&uid, &gid);
8061 audit_log(audit_context(), GFP_ATOMIC,
8062 AUDIT_ANOM_PROMISCUOUS,
8063 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8064 dev->name, (dev->flags & IFF_PROMISC),
8065 (old_flags & IFF_PROMISC),
8066 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8067 from_kuid(&init_user_ns, uid),
8068 from_kgid(&init_user_ns, gid),
8069 audit_get_sessionid(current));
8072 dev_change_rx_flags(dev, IFF_PROMISC);
8075 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8080 * dev_set_promiscuity - update promiscuity count on a device
8084 * Add or remove promiscuity from a device. While the count in the device
8085 * remains above zero the interface remains promiscuous. Once it hits zero
8086 * the device reverts back to normal filtering operation. A negative inc
8087 * value is used to drop promiscuity on the device.
8088 * Return 0 if successful or a negative errno code on error.
8090 int dev_set_promiscuity(struct net_device *dev, int inc)
8092 unsigned int old_flags = dev->flags;
8095 err = __dev_set_promiscuity(dev, inc, true);
8098 if (dev->flags != old_flags)
8099 dev_set_rx_mode(dev);
8102 EXPORT_SYMBOL(dev_set_promiscuity);
8104 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8106 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8110 dev->flags |= IFF_ALLMULTI;
8111 dev->allmulti += inc;
8112 if (dev->allmulti == 0) {
8115 * If inc causes overflow, untouch allmulti and return error.
8118 dev->flags &= ~IFF_ALLMULTI;
8120 dev->allmulti -= inc;
8121 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8126 if (dev->flags ^ old_flags) {
8127 dev_change_rx_flags(dev, IFF_ALLMULTI);
8128 dev_set_rx_mode(dev);
8130 __dev_notify_flags(dev, old_flags,
8131 dev->gflags ^ old_gflags);
8137 * dev_set_allmulti - update allmulti count on a device
8141 * Add or remove reception of all multicast frames to a device. While the
8142 * count in the device remains above zero the interface remains listening
8143 * to all interfaces. Once it hits zero the device reverts back to normal
8144 * filtering operation. A negative @inc value is used to drop the counter
8145 * when releasing a resource needing all multicasts.
8146 * Return 0 if successful or a negative errno code on error.
8149 int dev_set_allmulti(struct net_device *dev, int inc)
8151 return __dev_set_allmulti(dev, inc, true);
8153 EXPORT_SYMBOL(dev_set_allmulti);
8156 * Upload unicast and multicast address lists to device and
8157 * configure RX filtering. When the device doesn't support unicast
8158 * filtering it is put in promiscuous mode while unicast addresses
8161 void __dev_set_rx_mode(struct net_device *dev)
8163 const struct net_device_ops *ops = dev->netdev_ops;
8165 /* dev_open will call this function so the list will stay sane. */
8166 if (!(dev->flags&IFF_UP))
8169 if (!netif_device_present(dev))
8172 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8173 /* Unicast addresses changes may only happen under the rtnl,
8174 * therefore calling __dev_set_promiscuity here is safe.
8176 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8177 __dev_set_promiscuity(dev, 1, false);
8178 dev->uc_promisc = true;
8179 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8180 __dev_set_promiscuity(dev, -1, false);
8181 dev->uc_promisc = false;
8185 if (ops->ndo_set_rx_mode)
8186 ops->ndo_set_rx_mode(dev);
8189 void dev_set_rx_mode(struct net_device *dev)
8191 netif_addr_lock_bh(dev);
8192 __dev_set_rx_mode(dev);
8193 netif_addr_unlock_bh(dev);
8197 * dev_get_flags - get flags reported to userspace
8200 * Get the combination of flag bits exported through APIs to userspace.
8202 unsigned int dev_get_flags(const struct net_device *dev)
8206 flags = (dev->flags & ~(IFF_PROMISC |
8211 (dev->gflags & (IFF_PROMISC |
8214 if (netif_running(dev)) {
8215 if (netif_oper_up(dev))
8216 flags |= IFF_RUNNING;
8217 if (netif_carrier_ok(dev))
8218 flags |= IFF_LOWER_UP;
8219 if (netif_dormant(dev))
8220 flags |= IFF_DORMANT;
8225 EXPORT_SYMBOL(dev_get_flags);
8227 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8228 struct netlink_ext_ack *extack)
8230 unsigned int old_flags = dev->flags;
8236 * Set the flags on our device.
8239 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8240 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8242 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8246 * Load in the correct multicast list now the flags have changed.
8249 if ((old_flags ^ flags) & IFF_MULTICAST)
8250 dev_change_rx_flags(dev, IFF_MULTICAST);
8252 dev_set_rx_mode(dev);
8255 * Have we downed the interface. We handle IFF_UP ourselves
8256 * according to user attempts to set it, rather than blindly
8261 if ((old_flags ^ flags) & IFF_UP) {
8262 if (old_flags & IFF_UP)
8265 ret = __dev_open(dev, extack);
8268 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8269 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8270 unsigned int old_flags = dev->flags;
8272 dev->gflags ^= IFF_PROMISC;
8274 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8275 if (dev->flags != old_flags)
8276 dev_set_rx_mode(dev);
8279 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8280 * is important. Some (broken) drivers set IFF_PROMISC, when
8281 * IFF_ALLMULTI is requested not asking us and not reporting.
8283 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8284 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8286 dev->gflags ^= IFF_ALLMULTI;
8287 __dev_set_allmulti(dev, inc, false);
8293 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8294 unsigned int gchanges)
8296 unsigned int changes = dev->flags ^ old_flags;
8299 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8301 if (changes & IFF_UP) {
8302 if (dev->flags & IFF_UP)
8303 call_netdevice_notifiers(NETDEV_UP, dev);
8305 call_netdevice_notifiers(NETDEV_DOWN, dev);
8308 if (dev->flags & IFF_UP &&
8309 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8310 struct netdev_notifier_change_info change_info = {
8314 .flags_changed = changes,
8317 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8322 * dev_change_flags - change device settings
8324 * @flags: device state flags
8325 * @extack: netlink extended ack
8327 * Change settings on device based state flags. The flags are
8328 * in the userspace exported format.
8330 int dev_change_flags(struct net_device *dev, unsigned int flags,
8331 struct netlink_ext_ack *extack)
8334 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8336 ret = __dev_change_flags(dev, flags, extack);
8340 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8341 __dev_notify_flags(dev, old_flags, changes);
8344 EXPORT_SYMBOL(dev_change_flags);
8346 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8348 const struct net_device_ops *ops = dev->netdev_ops;
8350 if (ops->ndo_change_mtu)
8351 return ops->ndo_change_mtu(dev, new_mtu);
8353 /* Pairs with all the lockless reads of dev->mtu in the stack */
8354 WRITE_ONCE(dev->mtu, new_mtu);
8357 EXPORT_SYMBOL(__dev_set_mtu);
8359 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8360 struct netlink_ext_ack *extack)
8362 /* MTU must be positive, and in range */
8363 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8364 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8368 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8369 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8376 * dev_set_mtu_ext - Change maximum transfer unit
8378 * @new_mtu: new transfer unit
8379 * @extack: netlink extended ack
8381 * Change the maximum transfer size of the network device.
8383 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8384 struct netlink_ext_ack *extack)
8388 if (new_mtu == dev->mtu)
8391 err = dev_validate_mtu(dev, new_mtu, extack);
8395 if (!netif_device_present(dev))
8398 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8399 err = notifier_to_errno(err);
8403 orig_mtu = dev->mtu;
8404 err = __dev_set_mtu(dev, new_mtu);
8407 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8409 err = notifier_to_errno(err);
8411 /* setting mtu back and notifying everyone again,
8412 * so that they have a chance to revert changes.
8414 __dev_set_mtu(dev, orig_mtu);
8415 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8422 int dev_set_mtu(struct net_device *dev, int new_mtu)
8424 struct netlink_ext_ack extack;
8427 memset(&extack, 0, sizeof(extack));
8428 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8429 if (err && extack._msg)
8430 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8433 EXPORT_SYMBOL(dev_set_mtu);
8436 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8438 * @new_len: new tx queue length
8440 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8442 unsigned int orig_len = dev->tx_queue_len;
8445 if (new_len != (unsigned int)new_len)
8448 if (new_len != orig_len) {
8449 dev->tx_queue_len = new_len;
8450 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8451 res = notifier_to_errno(res);
8454 res = dev_qdisc_change_tx_queue_len(dev);
8462 netdev_err(dev, "refused to change device tx_queue_len\n");
8463 dev->tx_queue_len = orig_len;
8468 * dev_set_group - Change group this device belongs to
8470 * @new_group: group this device should belong to
8472 void dev_set_group(struct net_device *dev, int new_group)
8474 dev->group = new_group;
8476 EXPORT_SYMBOL(dev_set_group);
8479 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8481 * @addr: new address
8482 * @extack: netlink extended ack
8484 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8485 struct netlink_ext_ack *extack)
8487 struct netdev_notifier_pre_changeaddr_info info = {
8489 .info.extack = extack,
8494 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8495 return notifier_to_errno(rc);
8497 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8500 * dev_set_mac_address - Change Media Access Control Address
8503 * @extack: netlink extended ack
8505 * Change the hardware (MAC) address of the device
8507 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8508 struct netlink_ext_ack *extack)
8510 const struct net_device_ops *ops = dev->netdev_ops;
8513 if (!ops->ndo_set_mac_address)
8515 if (sa->sa_family != dev->type)
8517 if (!netif_device_present(dev))
8519 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8522 err = ops->ndo_set_mac_address(dev, sa);
8525 dev->addr_assign_type = NET_ADDR_SET;
8526 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8527 add_device_randomness(dev->dev_addr, dev->addr_len);
8530 EXPORT_SYMBOL(dev_set_mac_address);
8533 * dev_change_carrier - Change device carrier
8535 * @new_carrier: new value
8537 * Change device carrier
8539 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8541 const struct net_device_ops *ops = dev->netdev_ops;
8543 if (!ops->ndo_change_carrier)
8545 if (!netif_device_present(dev))
8547 return ops->ndo_change_carrier(dev, new_carrier);
8549 EXPORT_SYMBOL(dev_change_carrier);
8552 * dev_get_phys_port_id - Get device physical port ID
8556 * Get device physical port ID
8558 int dev_get_phys_port_id(struct net_device *dev,
8559 struct netdev_phys_item_id *ppid)
8561 const struct net_device_ops *ops = dev->netdev_ops;
8563 if (!ops->ndo_get_phys_port_id)
8565 return ops->ndo_get_phys_port_id(dev, ppid);
8567 EXPORT_SYMBOL(dev_get_phys_port_id);
8570 * dev_get_phys_port_name - Get device physical port name
8573 * @len: limit of bytes to copy to name
8575 * Get device physical port name
8577 int dev_get_phys_port_name(struct net_device *dev,
8578 char *name, size_t len)
8580 const struct net_device_ops *ops = dev->netdev_ops;
8583 if (ops->ndo_get_phys_port_name) {
8584 err = ops->ndo_get_phys_port_name(dev, name, len);
8585 if (err != -EOPNOTSUPP)
8588 return devlink_compat_phys_port_name_get(dev, name, len);
8590 EXPORT_SYMBOL(dev_get_phys_port_name);
8593 * dev_get_port_parent_id - Get the device's port parent identifier
8594 * @dev: network device
8595 * @ppid: pointer to a storage for the port's parent identifier
8596 * @recurse: allow/disallow recursion to lower devices
8598 * Get the devices's port parent identifier
8600 int dev_get_port_parent_id(struct net_device *dev,
8601 struct netdev_phys_item_id *ppid,
8604 const struct net_device_ops *ops = dev->netdev_ops;
8605 struct netdev_phys_item_id first = { };
8606 struct net_device *lower_dev;
8607 struct list_head *iter;
8610 if (ops->ndo_get_port_parent_id) {
8611 err = ops->ndo_get_port_parent_id(dev, ppid);
8612 if (err != -EOPNOTSUPP)
8616 err = devlink_compat_switch_id_get(dev, ppid);
8617 if (!err || err != -EOPNOTSUPP)
8623 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8624 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8629 else if (memcmp(&first, ppid, sizeof(*ppid)))
8635 EXPORT_SYMBOL(dev_get_port_parent_id);
8638 * netdev_port_same_parent_id - Indicate if two network devices have
8639 * the same port parent identifier
8640 * @a: first network device
8641 * @b: second network device
8643 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8645 struct netdev_phys_item_id a_id = { };
8646 struct netdev_phys_item_id b_id = { };
8648 if (dev_get_port_parent_id(a, &a_id, true) ||
8649 dev_get_port_parent_id(b, &b_id, true))
8652 return netdev_phys_item_id_same(&a_id, &b_id);
8654 EXPORT_SYMBOL(netdev_port_same_parent_id);
8657 * dev_change_proto_down - update protocol port state information
8659 * @proto_down: new value
8661 * This info can be used by switch drivers to set the phys state of the
8664 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8666 const struct net_device_ops *ops = dev->netdev_ops;
8668 if (!ops->ndo_change_proto_down)
8670 if (!netif_device_present(dev))
8672 return ops->ndo_change_proto_down(dev, proto_down);
8674 EXPORT_SYMBOL(dev_change_proto_down);
8677 * dev_change_proto_down_generic - generic implementation for
8678 * ndo_change_proto_down that sets carrier according to
8682 * @proto_down: new value
8684 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8687 netif_carrier_off(dev);
8689 netif_carrier_on(dev);
8690 dev->proto_down = proto_down;
8693 EXPORT_SYMBOL(dev_change_proto_down_generic);
8695 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8696 enum bpf_netdev_command cmd)
8698 struct netdev_bpf xdp;
8703 memset(&xdp, 0, sizeof(xdp));
8706 /* Query must always succeed. */
8707 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8712 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8713 struct netlink_ext_ack *extack, u32 flags,
8714 struct bpf_prog *prog)
8716 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8717 struct bpf_prog *prev_prog = NULL;
8718 struct netdev_bpf xdp;
8722 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8724 if (IS_ERR(prev_prog))
8728 memset(&xdp, 0, sizeof(xdp));
8729 if (flags & XDP_FLAGS_HW_MODE)
8730 xdp.command = XDP_SETUP_PROG_HW;
8732 xdp.command = XDP_SETUP_PROG;
8733 xdp.extack = extack;
8737 err = bpf_op(dev, &xdp);
8739 bpf_prog_change_xdp(prev_prog, prog);
8742 bpf_prog_put(prev_prog);
8747 static void dev_xdp_uninstall(struct net_device *dev)
8749 struct netdev_bpf xdp;
8752 /* Remove generic XDP */
8753 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8755 /* Remove from the driver */
8756 ndo_bpf = dev->netdev_ops->ndo_bpf;
8760 memset(&xdp, 0, sizeof(xdp));
8761 xdp.command = XDP_QUERY_PROG;
8762 WARN_ON(ndo_bpf(dev, &xdp));
8764 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8767 /* Remove HW offload */
8768 memset(&xdp, 0, sizeof(xdp));
8769 xdp.command = XDP_QUERY_PROG_HW;
8770 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8771 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8776 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8778 * @extack: netlink extended ack
8779 * @fd: new program fd or negative value to clear
8780 * @expected_fd: old program fd that userspace expects to replace or clear
8781 * @flags: xdp-related flags
8783 * Set or clear a bpf program for a device
8785 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8786 int fd, int expected_fd, u32 flags)
8788 const struct net_device_ops *ops = dev->netdev_ops;
8789 enum bpf_netdev_command query;
8790 u32 prog_id, expected_id = 0;
8791 bpf_op_t bpf_op, bpf_chk;
8792 struct bpf_prog *prog;
8798 offload = flags & XDP_FLAGS_HW_MODE;
8799 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8801 bpf_op = bpf_chk = ops->ndo_bpf;
8802 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8803 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8806 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8807 bpf_op = generic_xdp_install;
8808 if (bpf_op == bpf_chk)
8809 bpf_chk = generic_xdp_install;
8811 prog_id = __dev_xdp_query(dev, bpf_op, query);
8812 if (flags & XDP_FLAGS_REPLACE) {
8813 if (expected_fd >= 0) {
8814 prog = bpf_prog_get_type_dev(expected_fd,
8816 bpf_op == ops->ndo_bpf);
8818 return PTR_ERR(prog);
8819 expected_id = prog->aux->id;
8823 if (prog_id != expected_id) {
8824 NL_SET_ERR_MSG(extack, "Active program does not match expected");
8829 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8830 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8834 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8835 NL_SET_ERR_MSG(extack, "XDP program already attached");
8839 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8840 bpf_op == ops->ndo_bpf);
8842 return PTR_ERR(prog);
8844 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8845 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8850 if (prog->expected_attach_type == BPF_XDP_DEVMAP) {
8851 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
8856 /* prog->aux->id may be 0 for orphaned device-bound progs */
8857 if (prog->aux->id && prog->aux->id == prog_id) {
8867 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8868 if (err < 0 && prog)
8875 * dev_new_index - allocate an ifindex
8876 * @net: the applicable net namespace
8878 * Returns a suitable unique value for a new device interface
8879 * number. The caller must hold the rtnl semaphore or the
8880 * dev_base_lock to be sure it remains unique.
8882 static int dev_new_index(struct net *net)
8884 int ifindex = net->ifindex;
8889 if (!__dev_get_by_index(net, ifindex))
8890 return net->ifindex = ifindex;
8894 /* Delayed registration/unregisteration */
8895 static LIST_HEAD(net_todo_list);
8896 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8898 static void net_set_todo(struct net_device *dev)
8900 list_add_tail(&dev->todo_list, &net_todo_list);
8901 dev_net(dev)->dev_unreg_count++;
8904 static void rollback_registered_many(struct list_head *head)
8906 struct net_device *dev, *tmp;
8907 LIST_HEAD(close_head);
8909 BUG_ON(dev_boot_phase);
8912 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8913 /* Some devices call without registering
8914 * for initialization unwind. Remove those
8915 * devices and proceed with the remaining.
8917 if (dev->reg_state == NETREG_UNINITIALIZED) {
8918 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8922 list_del(&dev->unreg_list);
8925 dev->dismantle = true;
8926 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8929 /* If device is running, close it first. */
8930 list_for_each_entry(dev, head, unreg_list)
8931 list_add_tail(&dev->close_list, &close_head);
8932 dev_close_many(&close_head, true);
8934 list_for_each_entry(dev, head, unreg_list) {
8935 /* And unlink it from device chain. */
8936 unlist_netdevice(dev);
8938 dev->reg_state = NETREG_UNREGISTERING;
8940 flush_all_backlogs();
8944 list_for_each_entry(dev, head, unreg_list) {
8945 struct sk_buff *skb = NULL;
8947 /* Shutdown queueing discipline. */
8950 dev_xdp_uninstall(dev);
8952 /* Notify protocols, that we are about to destroy
8953 * this device. They should clean all the things.
8955 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8957 if (!dev->rtnl_link_ops ||
8958 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8959 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8960 GFP_KERNEL, NULL, 0);
8963 * Flush the unicast and multicast chains
8968 netdev_name_node_alt_flush(dev);
8969 netdev_name_node_free(dev->name_node);
8971 if (dev->netdev_ops->ndo_uninit)
8972 dev->netdev_ops->ndo_uninit(dev);
8975 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8977 /* Notifier chain MUST detach us all upper devices. */
8978 WARN_ON(netdev_has_any_upper_dev(dev));
8979 WARN_ON(netdev_has_any_lower_dev(dev));
8981 /* Remove entries from kobject tree */
8982 netdev_unregister_kobject(dev);
8984 /* Remove XPS queueing entries */
8985 netif_reset_xps_queues_gt(dev, 0);
8991 list_for_each_entry(dev, head, unreg_list)
8995 static void rollback_registered(struct net_device *dev)
8999 list_add(&dev->unreg_list, &single);
9000 rollback_registered_many(&single);
9004 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9005 struct net_device *upper, netdev_features_t features)
9007 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9008 netdev_features_t feature;
9011 for_each_netdev_feature(upper_disables, feature_bit) {
9012 feature = __NETIF_F_BIT(feature_bit);
9013 if (!(upper->wanted_features & feature)
9014 && (features & feature)) {
9015 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9016 &feature, upper->name);
9017 features &= ~feature;
9024 static void netdev_sync_lower_features(struct net_device *upper,
9025 struct net_device *lower, netdev_features_t features)
9027 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9028 netdev_features_t feature;
9031 for_each_netdev_feature(upper_disables, feature_bit) {
9032 feature = __NETIF_F_BIT(feature_bit);
9033 if (!(features & feature) && (lower->features & feature)) {
9034 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9035 &feature, lower->name);
9036 lower->wanted_features &= ~feature;
9037 __netdev_update_features(lower);
9039 if (unlikely(lower->features & feature))
9040 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9041 &feature, lower->name);
9043 netdev_features_change(lower);
9048 static netdev_features_t netdev_fix_features(struct net_device *dev,
9049 netdev_features_t features)
9051 /* Fix illegal checksum combinations */
9052 if ((features & NETIF_F_HW_CSUM) &&
9053 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9054 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9055 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9058 /* TSO requires that SG is present as well. */
9059 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9060 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9061 features &= ~NETIF_F_ALL_TSO;
9064 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9065 !(features & NETIF_F_IP_CSUM)) {
9066 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9067 features &= ~NETIF_F_TSO;
9068 features &= ~NETIF_F_TSO_ECN;
9071 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9072 !(features & NETIF_F_IPV6_CSUM)) {
9073 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9074 features &= ~NETIF_F_TSO6;
9077 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9078 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9079 features &= ~NETIF_F_TSO_MANGLEID;
9081 /* TSO ECN requires that TSO is present as well. */
9082 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9083 features &= ~NETIF_F_TSO_ECN;
9085 /* Software GSO depends on SG. */
9086 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9087 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9088 features &= ~NETIF_F_GSO;
9091 /* GSO partial features require GSO partial be set */
9092 if ((features & dev->gso_partial_features) &&
9093 !(features & NETIF_F_GSO_PARTIAL)) {
9095 "Dropping partially supported GSO features since no GSO partial.\n");
9096 features &= ~dev->gso_partial_features;
9099 if (!(features & NETIF_F_RXCSUM)) {
9100 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9101 * successfully merged by hardware must also have the
9102 * checksum verified by hardware. If the user does not
9103 * want to enable RXCSUM, logically, we should disable GRO_HW.
9105 if (features & NETIF_F_GRO_HW) {
9106 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9107 features &= ~NETIF_F_GRO_HW;
9111 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9112 if (features & NETIF_F_RXFCS) {
9113 if (features & NETIF_F_LRO) {
9114 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9115 features &= ~NETIF_F_LRO;
9118 if (features & NETIF_F_GRO_HW) {
9119 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9120 features &= ~NETIF_F_GRO_HW;
9127 int __netdev_update_features(struct net_device *dev)
9129 struct net_device *upper, *lower;
9130 netdev_features_t features;
9131 struct list_head *iter;
9136 features = netdev_get_wanted_features(dev);
9138 if (dev->netdev_ops->ndo_fix_features)
9139 features = dev->netdev_ops->ndo_fix_features(dev, features);
9141 /* driver might be less strict about feature dependencies */
9142 features = netdev_fix_features(dev, features);
9144 /* some features can't be enabled if they're off an an upper device */
9145 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9146 features = netdev_sync_upper_features(dev, upper, features);
9148 if (dev->features == features)
9151 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9152 &dev->features, &features);
9154 if (dev->netdev_ops->ndo_set_features)
9155 err = dev->netdev_ops->ndo_set_features(dev, features);
9159 if (unlikely(err < 0)) {
9161 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9162 err, &features, &dev->features);
9163 /* return non-0 since some features might have changed and
9164 * it's better to fire a spurious notification than miss it
9170 /* some features must be disabled on lower devices when disabled
9171 * on an upper device (think: bonding master or bridge)
9173 netdev_for_each_lower_dev(dev, lower, iter)
9174 netdev_sync_lower_features(dev, lower, features);
9177 netdev_features_t diff = features ^ dev->features;
9179 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9180 /* udp_tunnel_{get,drop}_rx_info both need
9181 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9182 * device, or they won't do anything.
9183 * Thus we need to update dev->features
9184 * *before* calling udp_tunnel_get_rx_info,
9185 * but *after* calling udp_tunnel_drop_rx_info.
9187 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9188 dev->features = features;
9189 udp_tunnel_get_rx_info(dev);
9191 udp_tunnel_drop_rx_info(dev);
9195 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9196 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9197 dev->features = features;
9198 err |= vlan_get_rx_ctag_filter_info(dev);
9200 vlan_drop_rx_ctag_filter_info(dev);
9204 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9205 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9206 dev->features = features;
9207 err |= vlan_get_rx_stag_filter_info(dev);
9209 vlan_drop_rx_stag_filter_info(dev);
9213 dev->features = features;
9216 return err < 0 ? 0 : 1;
9220 * netdev_update_features - recalculate device features
9221 * @dev: the device to check
9223 * Recalculate dev->features set and send notifications if it
9224 * has changed. Should be called after driver or hardware dependent
9225 * conditions might have changed that influence the features.
9227 void netdev_update_features(struct net_device *dev)
9229 if (__netdev_update_features(dev))
9230 netdev_features_change(dev);
9232 EXPORT_SYMBOL(netdev_update_features);
9235 * netdev_change_features - recalculate device features
9236 * @dev: the device to check
9238 * Recalculate dev->features set and send notifications even
9239 * if they have not changed. Should be called instead of
9240 * netdev_update_features() if also dev->vlan_features might
9241 * have changed to allow the changes to be propagated to stacked
9244 void netdev_change_features(struct net_device *dev)
9246 __netdev_update_features(dev);
9247 netdev_features_change(dev);
9249 EXPORT_SYMBOL(netdev_change_features);
9252 * netif_stacked_transfer_operstate - transfer operstate
9253 * @rootdev: the root or lower level device to transfer state from
9254 * @dev: the device to transfer operstate to
9256 * Transfer operational state from root to device. This is normally
9257 * called when a stacking relationship exists between the root
9258 * device and the device(a leaf device).
9260 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9261 struct net_device *dev)
9263 if (rootdev->operstate == IF_OPER_DORMANT)
9264 netif_dormant_on(dev);
9266 netif_dormant_off(dev);
9268 if (rootdev->operstate == IF_OPER_TESTING)
9269 netif_testing_on(dev);
9271 netif_testing_off(dev);
9273 if (netif_carrier_ok(rootdev))
9274 netif_carrier_on(dev);
9276 netif_carrier_off(dev);
9278 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9280 static int netif_alloc_rx_queues(struct net_device *dev)
9282 unsigned int i, count = dev->num_rx_queues;
9283 struct netdev_rx_queue *rx;
9284 size_t sz = count * sizeof(*rx);
9289 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9295 for (i = 0; i < count; i++) {
9298 /* XDP RX-queue setup */
9299 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9306 /* Rollback successful reg's and free other resources */
9308 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9314 static void netif_free_rx_queues(struct net_device *dev)
9316 unsigned int i, count = dev->num_rx_queues;
9318 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9322 for (i = 0; i < count; i++)
9323 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9328 static void netdev_init_one_queue(struct net_device *dev,
9329 struct netdev_queue *queue, void *_unused)
9331 /* Initialize queue lock */
9332 spin_lock_init(&queue->_xmit_lock);
9333 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9334 queue->xmit_lock_owner = -1;
9335 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9338 dql_init(&queue->dql, HZ);
9342 static void netif_free_tx_queues(struct net_device *dev)
9347 static int netif_alloc_netdev_queues(struct net_device *dev)
9349 unsigned int count = dev->num_tx_queues;
9350 struct netdev_queue *tx;
9351 size_t sz = count * sizeof(*tx);
9353 if (count < 1 || count > 0xffff)
9356 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9362 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9363 spin_lock_init(&dev->tx_global_lock);
9368 void netif_tx_stop_all_queues(struct net_device *dev)
9372 for (i = 0; i < dev->num_tx_queues; i++) {
9373 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9375 netif_tx_stop_queue(txq);
9378 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9380 void netdev_update_lockdep_key(struct net_device *dev)
9382 lockdep_unregister_key(&dev->addr_list_lock_key);
9383 lockdep_register_key(&dev->addr_list_lock_key);
9385 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9387 EXPORT_SYMBOL(netdev_update_lockdep_key);
9390 * register_netdevice - register a network device
9391 * @dev: device to register
9393 * Take a completed network device structure and add it to the kernel
9394 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9395 * chain. 0 is returned on success. A negative errno code is returned
9396 * on a failure to set up the device, or if the name is a duplicate.
9398 * Callers must hold the rtnl semaphore. You may want
9399 * register_netdev() instead of this.
9402 * The locking appears insufficient to guarantee two parallel registers
9403 * will not get the same name.
9406 int register_netdevice(struct net_device *dev)
9409 struct net *net = dev_net(dev);
9411 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9412 NETDEV_FEATURE_COUNT);
9413 BUG_ON(dev_boot_phase);
9418 /* When net_device's are persistent, this will be fatal. */
9419 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9422 ret = ethtool_check_ops(dev->ethtool_ops);
9426 spin_lock_init(&dev->addr_list_lock);
9427 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9429 ret = dev_get_valid_name(net, dev, dev->name);
9434 dev->name_node = netdev_name_node_head_alloc(dev);
9435 if (!dev->name_node)
9438 /* Init, if this function is available */
9439 if (dev->netdev_ops->ndo_init) {
9440 ret = dev->netdev_ops->ndo_init(dev);
9448 if (((dev->hw_features | dev->features) &
9449 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9450 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9451 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9452 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9459 dev->ifindex = dev_new_index(net);
9460 else if (__dev_get_by_index(net, dev->ifindex))
9463 /* Transfer changeable features to wanted_features and enable
9464 * software offloads (GSO and GRO).
9466 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9467 dev->features |= NETIF_F_SOFT_FEATURES;
9469 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9470 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9471 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9474 dev->wanted_features = dev->features & dev->hw_features;
9476 if (!(dev->flags & IFF_LOOPBACK))
9477 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9479 /* If IPv4 TCP segmentation offload is supported we should also
9480 * allow the device to enable segmenting the frame with the option
9481 * of ignoring a static IP ID value. This doesn't enable the
9482 * feature itself but allows the user to enable it later.
9484 if (dev->hw_features & NETIF_F_TSO)
9485 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9486 if (dev->vlan_features & NETIF_F_TSO)
9487 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9488 if (dev->mpls_features & NETIF_F_TSO)
9489 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9490 if (dev->hw_enc_features & NETIF_F_TSO)
9491 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9493 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9495 dev->vlan_features |= NETIF_F_HIGHDMA;
9497 /* Make NETIF_F_SG inheritable to tunnel devices.
9499 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9501 /* Make NETIF_F_SG inheritable to MPLS.
9503 dev->mpls_features |= NETIF_F_SG;
9505 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9506 ret = notifier_to_errno(ret);
9510 ret = netdev_register_kobject(dev);
9512 dev->reg_state = NETREG_UNREGISTERED;
9515 dev->reg_state = NETREG_REGISTERED;
9517 __netdev_update_features(dev);
9520 * Default initial state at registry is that the
9521 * device is present.
9524 set_bit(__LINK_STATE_PRESENT, &dev->state);
9526 linkwatch_init_dev(dev);
9528 dev_init_scheduler(dev);
9530 list_netdevice(dev);
9531 add_device_randomness(dev->dev_addr, dev->addr_len);
9533 /* If the device has permanent device address, driver should
9534 * set dev_addr and also addr_assign_type should be set to
9535 * NET_ADDR_PERM (default value).
9537 if (dev->addr_assign_type == NET_ADDR_PERM)
9538 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9540 /* Notify protocols, that a new device appeared. */
9541 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9542 ret = notifier_to_errno(ret);
9544 rollback_registered(dev);
9547 dev->reg_state = NETREG_UNREGISTERED;
9550 * Prevent userspace races by waiting until the network
9551 * device is fully setup before sending notifications.
9553 if (!dev->rtnl_link_ops ||
9554 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9555 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9561 if (dev->netdev_ops->ndo_uninit)
9562 dev->netdev_ops->ndo_uninit(dev);
9563 if (dev->priv_destructor)
9564 dev->priv_destructor(dev);
9566 netdev_name_node_free(dev->name_node);
9569 EXPORT_SYMBOL(register_netdevice);
9572 * init_dummy_netdev - init a dummy network device for NAPI
9573 * @dev: device to init
9575 * This takes a network device structure and initialize the minimum
9576 * amount of fields so it can be used to schedule NAPI polls without
9577 * registering a full blown interface. This is to be used by drivers
9578 * that need to tie several hardware interfaces to a single NAPI
9579 * poll scheduler due to HW limitations.
9581 int init_dummy_netdev(struct net_device *dev)
9583 /* Clear everything. Note we don't initialize spinlocks
9584 * are they aren't supposed to be taken by any of the
9585 * NAPI code and this dummy netdev is supposed to be
9586 * only ever used for NAPI polls
9588 memset(dev, 0, sizeof(struct net_device));
9590 /* make sure we BUG if trying to hit standard
9591 * register/unregister code path
9593 dev->reg_state = NETREG_DUMMY;
9595 /* NAPI wants this */
9596 INIT_LIST_HEAD(&dev->napi_list);
9598 /* a dummy interface is started by default */
9599 set_bit(__LINK_STATE_PRESENT, &dev->state);
9600 set_bit(__LINK_STATE_START, &dev->state);
9602 /* napi_busy_loop stats accounting wants this */
9603 dev_net_set(dev, &init_net);
9605 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9606 * because users of this 'device' dont need to change
9612 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9616 * register_netdev - register a network device
9617 * @dev: device to register
9619 * Take a completed network device structure and add it to the kernel
9620 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9621 * chain. 0 is returned on success. A negative errno code is returned
9622 * on a failure to set up the device, or if the name is a duplicate.
9624 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9625 * and expands the device name if you passed a format string to
9628 int register_netdev(struct net_device *dev)
9632 if (rtnl_lock_killable())
9634 err = register_netdevice(dev);
9638 EXPORT_SYMBOL(register_netdev);
9640 int netdev_refcnt_read(const struct net_device *dev)
9644 for_each_possible_cpu(i)
9645 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9648 EXPORT_SYMBOL(netdev_refcnt_read);
9651 * netdev_wait_allrefs - wait until all references are gone.
9652 * @dev: target net_device
9654 * This is called when unregistering network devices.
9656 * Any protocol or device that holds a reference should register
9657 * for netdevice notification, and cleanup and put back the
9658 * reference if they receive an UNREGISTER event.
9659 * We can get stuck here if buggy protocols don't correctly
9662 static void netdev_wait_allrefs(struct net_device *dev)
9664 unsigned long rebroadcast_time, warning_time;
9667 linkwatch_forget_dev(dev);
9669 rebroadcast_time = warning_time = jiffies;
9670 refcnt = netdev_refcnt_read(dev);
9672 while (refcnt != 0) {
9673 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9676 /* Rebroadcast unregister notification */
9677 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9683 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9685 /* We must not have linkwatch events
9686 * pending on unregister. If this
9687 * happens, we simply run the queue
9688 * unscheduled, resulting in a noop
9691 linkwatch_run_queue();
9696 rebroadcast_time = jiffies;
9701 refcnt = netdev_refcnt_read(dev);
9703 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9704 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9706 warning_time = jiffies;
9715 * register_netdevice(x1);
9716 * register_netdevice(x2);
9718 * unregister_netdevice(y1);
9719 * unregister_netdevice(y2);
9725 * We are invoked by rtnl_unlock().
9726 * This allows us to deal with problems:
9727 * 1) We can delete sysfs objects which invoke hotplug
9728 * without deadlocking with linkwatch via keventd.
9729 * 2) Since we run with the RTNL semaphore not held, we can sleep
9730 * safely in order to wait for the netdev refcnt to drop to zero.
9732 * We must not return until all unregister events added during
9733 * the interval the lock was held have been completed.
9735 void netdev_run_todo(void)
9737 struct list_head list;
9739 /* Snapshot list, allow later requests */
9740 list_replace_init(&net_todo_list, &list);
9745 /* Wait for rcu callbacks to finish before next phase */
9746 if (!list_empty(&list))
9749 while (!list_empty(&list)) {
9750 struct net_device *dev
9751 = list_first_entry(&list, struct net_device, todo_list);
9752 list_del(&dev->todo_list);
9754 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9755 pr_err("network todo '%s' but state %d\n",
9756 dev->name, dev->reg_state);
9761 dev->reg_state = NETREG_UNREGISTERED;
9763 netdev_wait_allrefs(dev);
9766 BUG_ON(netdev_refcnt_read(dev));
9767 BUG_ON(!list_empty(&dev->ptype_all));
9768 BUG_ON(!list_empty(&dev->ptype_specific));
9769 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9770 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9771 #if IS_ENABLED(CONFIG_DECNET)
9772 WARN_ON(dev->dn_ptr);
9774 if (dev->priv_destructor)
9775 dev->priv_destructor(dev);
9776 if (dev->needs_free_netdev)
9779 /* Report a network device has been unregistered */
9781 dev_net(dev)->dev_unreg_count--;
9783 wake_up(&netdev_unregistering_wq);
9785 /* Free network device */
9786 kobject_put(&dev->dev.kobj);
9790 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9791 * all the same fields in the same order as net_device_stats, with only
9792 * the type differing, but rtnl_link_stats64 may have additional fields
9793 * at the end for newer counters.
9795 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9796 const struct net_device_stats *netdev_stats)
9798 #if BITS_PER_LONG == 64
9799 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9800 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9801 /* zero out counters that only exist in rtnl_link_stats64 */
9802 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9803 sizeof(*stats64) - sizeof(*netdev_stats));
9805 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9806 const unsigned long *src = (const unsigned long *)netdev_stats;
9807 u64 *dst = (u64 *)stats64;
9809 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9810 for (i = 0; i < n; i++)
9812 /* zero out counters that only exist in rtnl_link_stats64 */
9813 memset((char *)stats64 + n * sizeof(u64), 0,
9814 sizeof(*stats64) - n * sizeof(u64));
9817 EXPORT_SYMBOL(netdev_stats_to_stats64);
9820 * dev_get_stats - get network device statistics
9821 * @dev: device to get statistics from
9822 * @storage: place to store stats
9824 * Get network statistics from device. Return @storage.
9825 * The device driver may provide its own method by setting
9826 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9827 * otherwise the internal statistics structure is used.
9829 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9830 struct rtnl_link_stats64 *storage)
9832 const struct net_device_ops *ops = dev->netdev_ops;
9834 if (ops->ndo_get_stats64) {
9835 memset(storage, 0, sizeof(*storage));
9836 ops->ndo_get_stats64(dev, storage);
9837 } else if (ops->ndo_get_stats) {
9838 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9840 netdev_stats_to_stats64(storage, &dev->stats);
9842 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9843 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9844 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9847 EXPORT_SYMBOL(dev_get_stats);
9849 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9851 struct netdev_queue *queue = dev_ingress_queue(dev);
9853 #ifdef CONFIG_NET_CLS_ACT
9856 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9859 netdev_init_one_queue(dev, queue, NULL);
9860 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9861 queue->qdisc_sleeping = &noop_qdisc;
9862 rcu_assign_pointer(dev->ingress_queue, queue);
9867 static const struct ethtool_ops default_ethtool_ops;
9869 void netdev_set_default_ethtool_ops(struct net_device *dev,
9870 const struct ethtool_ops *ops)
9872 if (dev->ethtool_ops == &default_ethtool_ops)
9873 dev->ethtool_ops = ops;
9875 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9877 void netdev_freemem(struct net_device *dev)
9879 char *addr = (char *)dev - dev->padded;
9885 * alloc_netdev_mqs - allocate network device
9886 * @sizeof_priv: size of private data to allocate space for
9887 * @name: device name format string
9888 * @name_assign_type: origin of device name
9889 * @setup: callback to initialize device
9890 * @txqs: the number of TX subqueues to allocate
9891 * @rxqs: the number of RX subqueues to allocate
9893 * Allocates a struct net_device with private data area for driver use
9894 * and performs basic initialization. Also allocates subqueue structs
9895 * for each queue on the device.
9897 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9898 unsigned char name_assign_type,
9899 void (*setup)(struct net_device *),
9900 unsigned int txqs, unsigned int rxqs)
9902 struct net_device *dev;
9903 unsigned int alloc_size;
9904 struct net_device *p;
9906 BUG_ON(strlen(name) >= sizeof(dev->name));
9909 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9914 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9918 alloc_size = sizeof(struct net_device);
9920 /* ensure 32-byte alignment of private area */
9921 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9922 alloc_size += sizeof_priv;
9924 /* ensure 32-byte alignment of whole construct */
9925 alloc_size += NETDEV_ALIGN - 1;
9927 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9931 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9932 dev->padded = (char *)dev - (char *)p;
9934 dev->pcpu_refcnt = alloc_percpu(int);
9935 if (!dev->pcpu_refcnt)
9938 if (dev_addr_init(dev))
9944 dev_net_set(dev, &init_net);
9946 lockdep_register_key(&dev->addr_list_lock_key);
9948 dev->gso_max_size = GSO_MAX_SIZE;
9949 dev->gso_max_segs = GSO_MAX_SEGS;
9950 dev->upper_level = 1;
9951 dev->lower_level = 1;
9953 INIT_LIST_HEAD(&dev->napi_list);
9954 INIT_LIST_HEAD(&dev->unreg_list);
9955 INIT_LIST_HEAD(&dev->close_list);
9956 INIT_LIST_HEAD(&dev->link_watch_list);
9957 INIT_LIST_HEAD(&dev->adj_list.upper);
9958 INIT_LIST_HEAD(&dev->adj_list.lower);
9959 INIT_LIST_HEAD(&dev->ptype_all);
9960 INIT_LIST_HEAD(&dev->ptype_specific);
9961 INIT_LIST_HEAD(&dev->net_notifier_list);
9962 #ifdef CONFIG_NET_SCHED
9963 hash_init(dev->qdisc_hash);
9965 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9968 if (!dev->tx_queue_len) {
9969 dev->priv_flags |= IFF_NO_QUEUE;
9970 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9973 dev->num_tx_queues = txqs;
9974 dev->real_num_tx_queues = txqs;
9975 if (netif_alloc_netdev_queues(dev))
9978 dev->num_rx_queues = rxqs;
9979 dev->real_num_rx_queues = rxqs;
9980 if (netif_alloc_rx_queues(dev))
9983 strcpy(dev->name, name);
9984 dev->name_assign_type = name_assign_type;
9985 dev->group = INIT_NETDEV_GROUP;
9986 if (!dev->ethtool_ops)
9987 dev->ethtool_ops = &default_ethtool_ops;
9989 nf_hook_ingress_init(dev);
9998 free_percpu(dev->pcpu_refcnt);
10000 netdev_freemem(dev);
10003 EXPORT_SYMBOL(alloc_netdev_mqs);
10006 * free_netdev - free network device
10009 * This function does the last stage of destroying an allocated device
10010 * interface. The reference to the device object is released. If this
10011 * is the last reference then it will be freed.Must be called in process
10014 void free_netdev(struct net_device *dev)
10016 struct napi_struct *p, *n;
10019 netif_free_tx_queues(dev);
10020 netif_free_rx_queues(dev);
10022 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10024 /* Flush device addresses */
10025 dev_addr_flush(dev);
10027 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10030 free_percpu(dev->pcpu_refcnt);
10031 dev->pcpu_refcnt = NULL;
10032 free_percpu(dev->xdp_bulkq);
10033 dev->xdp_bulkq = NULL;
10035 lockdep_unregister_key(&dev->addr_list_lock_key);
10037 /* Compatibility with error handling in drivers */
10038 if (dev->reg_state == NETREG_UNINITIALIZED) {
10039 netdev_freemem(dev);
10043 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10044 dev->reg_state = NETREG_RELEASED;
10046 /* will free via device release */
10047 put_device(&dev->dev);
10049 EXPORT_SYMBOL(free_netdev);
10052 * synchronize_net - Synchronize with packet receive processing
10054 * Wait for packets currently being received to be done.
10055 * Does not block later packets from starting.
10057 void synchronize_net(void)
10060 if (rtnl_is_locked())
10061 synchronize_rcu_expedited();
10065 EXPORT_SYMBOL(synchronize_net);
10068 * unregister_netdevice_queue - remove device from the kernel
10072 * This function shuts down a device interface and removes it
10073 * from the kernel tables.
10074 * If head not NULL, device is queued to be unregistered later.
10076 * Callers must hold the rtnl semaphore. You may want
10077 * unregister_netdev() instead of this.
10080 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10085 list_move_tail(&dev->unreg_list, head);
10087 rollback_registered(dev);
10088 /* Finish processing unregister after unlock */
10092 EXPORT_SYMBOL(unregister_netdevice_queue);
10095 * unregister_netdevice_many - unregister many devices
10096 * @head: list of devices
10098 * Note: As most callers use a stack allocated list_head,
10099 * we force a list_del() to make sure stack wont be corrupted later.
10101 void unregister_netdevice_many(struct list_head *head)
10103 struct net_device *dev;
10105 if (!list_empty(head)) {
10106 rollback_registered_many(head);
10107 list_for_each_entry(dev, head, unreg_list)
10112 EXPORT_SYMBOL(unregister_netdevice_many);
10115 * unregister_netdev - remove device from the kernel
10118 * This function shuts down a device interface and removes it
10119 * from the kernel tables.
10121 * This is just a wrapper for unregister_netdevice that takes
10122 * the rtnl semaphore. In general you want to use this and not
10123 * unregister_netdevice.
10125 void unregister_netdev(struct net_device *dev)
10128 unregister_netdevice(dev);
10131 EXPORT_SYMBOL(unregister_netdev);
10134 * dev_change_net_namespace - move device to different nethost namespace
10136 * @net: network namespace
10137 * @pat: If not NULL name pattern to try if the current device name
10138 * is already taken in the destination network namespace.
10140 * This function shuts down a device interface and moves it
10141 * to a new network namespace. On success 0 is returned, on
10142 * a failure a netagive errno code is returned.
10144 * Callers must hold the rtnl semaphore.
10147 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
10149 struct net *net_old = dev_net(dev);
10150 int err, new_nsid, new_ifindex;
10154 /* Don't allow namespace local devices to be moved. */
10156 if (dev->features & NETIF_F_NETNS_LOCAL)
10159 /* Ensure the device has been registrered */
10160 if (dev->reg_state != NETREG_REGISTERED)
10163 /* Get out if there is nothing todo */
10165 if (net_eq(net_old, net))
10168 /* Pick the destination device name, and ensure
10169 * we can use it in the destination network namespace.
10172 if (__dev_get_by_name(net, dev->name)) {
10173 /* We get here if we can't use the current device name */
10176 err = dev_get_valid_name(net, dev, pat);
10182 * And now a mini version of register_netdevice unregister_netdevice.
10185 /* If device is running close it first. */
10188 /* And unlink it from device chain */
10189 unlist_netdevice(dev);
10193 /* Shutdown queueing discipline. */
10196 /* Notify protocols, that we are about to destroy
10197 * this device. They should clean all the things.
10199 * Note that dev->reg_state stays at NETREG_REGISTERED.
10200 * This is wanted because this way 8021q and macvlan know
10201 * the device is just moving and can keep their slaves up.
10203 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10206 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10207 /* If there is an ifindex conflict assign a new one */
10208 if (__dev_get_by_index(net, dev->ifindex))
10209 new_ifindex = dev_new_index(net);
10211 new_ifindex = dev->ifindex;
10213 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10217 * Flush the unicast and multicast chains
10222 /* Send a netdev-removed uevent to the old namespace */
10223 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10224 netdev_adjacent_del_links(dev);
10226 /* Move per-net netdevice notifiers that are following the netdevice */
10227 move_netdevice_notifiers_dev_net(dev, net);
10229 /* Actually switch the network namespace */
10230 dev_net_set(dev, net);
10231 dev->ifindex = new_ifindex;
10233 /* Send a netdev-add uevent to the new namespace */
10234 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10235 netdev_adjacent_add_links(dev);
10237 /* Fixup kobjects */
10238 err = device_rename(&dev->dev, dev->name);
10241 /* Adapt owner in case owning user namespace of target network
10242 * namespace is different from the original one.
10244 err = netdev_change_owner(dev, net_old, net);
10247 /* Add the device back in the hashes */
10248 list_netdevice(dev);
10250 /* Notify protocols, that a new device appeared. */
10251 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10254 * Prevent userspace races by waiting until the network
10255 * device is fully setup before sending notifications.
10257 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10264 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10266 static int dev_cpu_dead(unsigned int oldcpu)
10268 struct sk_buff **list_skb;
10269 struct sk_buff *skb;
10271 struct softnet_data *sd, *oldsd, *remsd = NULL;
10273 local_irq_disable();
10274 cpu = smp_processor_id();
10275 sd = &per_cpu(softnet_data, cpu);
10276 oldsd = &per_cpu(softnet_data, oldcpu);
10278 /* Find end of our completion_queue. */
10279 list_skb = &sd->completion_queue;
10281 list_skb = &(*list_skb)->next;
10282 /* Append completion queue from offline CPU. */
10283 *list_skb = oldsd->completion_queue;
10284 oldsd->completion_queue = NULL;
10286 /* Append output queue from offline CPU. */
10287 if (oldsd->output_queue) {
10288 *sd->output_queue_tailp = oldsd->output_queue;
10289 sd->output_queue_tailp = oldsd->output_queue_tailp;
10290 oldsd->output_queue = NULL;
10291 oldsd->output_queue_tailp = &oldsd->output_queue;
10293 /* Append NAPI poll list from offline CPU, with one exception :
10294 * process_backlog() must be called by cpu owning percpu backlog.
10295 * We properly handle process_queue & input_pkt_queue later.
10297 while (!list_empty(&oldsd->poll_list)) {
10298 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10299 struct napi_struct,
10302 list_del_init(&napi->poll_list);
10303 if (napi->poll == process_backlog)
10306 ____napi_schedule(sd, napi);
10309 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10310 local_irq_enable();
10313 remsd = oldsd->rps_ipi_list;
10314 oldsd->rps_ipi_list = NULL;
10316 /* send out pending IPI's on offline CPU */
10317 net_rps_send_ipi(remsd);
10319 /* Process offline CPU's input_pkt_queue */
10320 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10322 input_queue_head_incr(oldsd);
10324 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10326 input_queue_head_incr(oldsd);
10333 * netdev_increment_features - increment feature set by one
10334 * @all: current feature set
10335 * @one: new feature set
10336 * @mask: mask feature set
10338 * Computes a new feature set after adding a device with feature set
10339 * @one to the master device with current feature set @all. Will not
10340 * enable anything that is off in @mask. Returns the new feature set.
10342 netdev_features_t netdev_increment_features(netdev_features_t all,
10343 netdev_features_t one, netdev_features_t mask)
10345 if (mask & NETIF_F_HW_CSUM)
10346 mask |= NETIF_F_CSUM_MASK;
10347 mask |= NETIF_F_VLAN_CHALLENGED;
10349 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10350 all &= one | ~NETIF_F_ALL_FOR_ALL;
10352 /* If one device supports hw checksumming, set for all. */
10353 if (all & NETIF_F_HW_CSUM)
10354 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10358 EXPORT_SYMBOL(netdev_increment_features);
10360 static struct hlist_head * __net_init netdev_create_hash(void)
10363 struct hlist_head *hash;
10365 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10367 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10368 INIT_HLIST_HEAD(&hash[i]);
10373 /* Initialize per network namespace state */
10374 static int __net_init netdev_init(struct net *net)
10376 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10377 8 * sizeof_field(struct napi_struct, gro_bitmask));
10379 if (net != &init_net)
10380 INIT_LIST_HEAD(&net->dev_base_head);
10382 net->dev_name_head = netdev_create_hash();
10383 if (net->dev_name_head == NULL)
10386 net->dev_index_head = netdev_create_hash();
10387 if (net->dev_index_head == NULL)
10390 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10395 kfree(net->dev_name_head);
10401 * netdev_drivername - network driver for the device
10402 * @dev: network device
10404 * Determine network driver for device.
10406 const char *netdev_drivername(const struct net_device *dev)
10408 const struct device_driver *driver;
10409 const struct device *parent;
10410 const char *empty = "";
10412 parent = dev->dev.parent;
10416 driver = parent->driver;
10417 if (driver && driver->name)
10418 return driver->name;
10422 static void __netdev_printk(const char *level, const struct net_device *dev,
10423 struct va_format *vaf)
10425 if (dev && dev->dev.parent) {
10426 dev_printk_emit(level[1] - '0',
10429 dev_driver_string(dev->dev.parent),
10430 dev_name(dev->dev.parent),
10431 netdev_name(dev), netdev_reg_state(dev),
10434 printk("%s%s%s: %pV",
10435 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10437 printk("%s(NULL net_device): %pV", level, vaf);
10441 void netdev_printk(const char *level, const struct net_device *dev,
10442 const char *format, ...)
10444 struct va_format vaf;
10447 va_start(args, format);
10452 __netdev_printk(level, dev, &vaf);
10456 EXPORT_SYMBOL(netdev_printk);
10458 #define define_netdev_printk_level(func, level) \
10459 void func(const struct net_device *dev, const char *fmt, ...) \
10461 struct va_format vaf; \
10464 va_start(args, fmt); \
10469 __netdev_printk(level, dev, &vaf); \
10473 EXPORT_SYMBOL(func);
10475 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10476 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10477 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10478 define_netdev_printk_level(netdev_err, KERN_ERR);
10479 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10480 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10481 define_netdev_printk_level(netdev_info, KERN_INFO);
10483 static void __net_exit netdev_exit(struct net *net)
10485 kfree(net->dev_name_head);
10486 kfree(net->dev_index_head);
10487 if (net != &init_net)
10488 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10491 static struct pernet_operations __net_initdata netdev_net_ops = {
10492 .init = netdev_init,
10493 .exit = netdev_exit,
10496 static void __net_exit default_device_exit(struct net *net)
10498 struct net_device *dev, *aux;
10500 * Push all migratable network devices back to the
10501 * initial network namespace
10504 for_each_netdev_safe(net, dev, aux) {
10506 char fb_name[IFNAMSIZ];
10508 /* Ignore unmoveable devices (i.e. loopback) */
10509 if (dev->features & NETIF_F_NETNS_LOCAL)
10512 /* Leave virtual devices for the generic cleanup */
10513 if (dev->rtnl_link_ops)
10516 /* Push remaining network devices to init_net */
10517 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10518 if (__dev_get_by_name(&init_net, fb_name))
10519 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10520 err = dev_change_net_namespace(dev, &init_net, fb_name);
10522 pr_emerg("%s: failed to move %s to init_net: %d\n",
10523 __func__, dev->name, err);
10530 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10532 /* Return with the rtnl_lock held when there are no network
10533 * devices unregistering in any network namespace in net_list.
10536 bool unregistering;
10537 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10539 add_wait_queue(&netdev_unregistering_wq, &wait);
10541 unregistering = false;
10543 list_for_each_entry(net, net_list, exit_list) {
10544 if (net->dev_unreg_count > 0) {
10545 unregistering = true;
10549 if (!unregistering)
10553 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10555 remove_wait_queue(&netdev_unregistering_wq, &wait);
10558 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10560 /* At exit all network devices most be removed from a network
10561 * namespace. Do this in the reverse order of registration.
10562 * Do this across as many network namespaces as possible to
10563 * improve batching efficiency.
10565 struct net_device *dev;
10567 LIST_HEAD(dev_kill_list);
10569 /* To prevent network device cleanup code from dereferencing
10570 * loopback devices or network devices that have been freed
10571 * wait here for all pending unregistrations to complete,
10572 * before unregistring the loopback device and allowing the
10573 * network namespace be freed.
10575 * The netdev todo list containing all network devices
10576 * unregistrations that happen in default_device_exit_batch
10577 * will run in the rtnl_unlock() at the end of
10578 * default_device_exit_batch.
10580 rtnl_lock_unregistering(net_list);
10581 list_for_each_entry(net, net_list, exit_list) {
10582 for_each_netdev_reverse(net, dev) {
10583 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10584 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10586 unregister_netdevice_queue(dev, &dev_kill_list);
10589 unregister_netdevice_many(&dev_kill_list);
10593 static struct pernet_operations __net_initdata default_device_ops = {
10594 .exit = default_device_exit,
10595 .exit_batch = default_device_exit_batch,
10599 * Initialize the DEV module. At boot time this walks the device list and
10600 * unhooks any devices that fail to initialise (normally hardware not
10601 * present) and leaves us with a valid list of present and active devices.
10606 * This is called single threaded during boot, so no need
10607 * to take the rtnl semaphore.
10609 static int __init net_dev_init(void)
10611 int i, rc = -ENOMEM;
10613 BUG_ON(!dev_boot_phase);
10615 if (dev_proc_init())
10618 if (netdev_kobject_init())
10621 INIT_LIST_HEAD(&ptype_all);
10622 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10623 INIT_LIST_HEAD(&ptype_base[i]);
10625 INIT_LIST_HEAD(&offload_base);
10627 if (register_pernet_subsys(&netdev_net_ops))
10631 * Initialise the packet receive queues.
10634 for_each_possible_cpu(i) {
10635 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10636 struct softnet_data *sd = &per_cpu(softnet_data, i);
10638 INIT_WORK(flush, flush_backlog);
10640 skb_queue_head_init(&sd->input_pkt_queue);
10641 skb_queue_head_init(&sd->process_queue);
10642 #ifdef CONFIG_XFRM_OFFLOAD
10643 skb_queue_head_init(&sd->xfrm_backlog);
10645 INIT_LIST_HEAD(&sd->poll_list);
10646 sd->output_queue_tailp = &sd->output_queue;
10648 sd->csd.func = rps_trigger_softirq;
10653 init_gro_hash(&sd->backlog);
10654 sd->backlog.poll = process_backlog;
10655 sd->backlog.weight = weight_p;
10658 dev_boot_phase = 0;
10660 /* The loopback device is special if any other network devices
10661 * is present in a network namespace the loopback device must
10662 * be present. Since we now dynamically allocate and free the
10663 * loopback device ensure this invariant is maintained by
10664 * keeping the loopback device as the first device on the
10665 * list of network devices. Ensuring the loopback devices
10666 * is the first device that appears and the last network device
10669 if (register_pernet_device(&loopback_net_ops))
10672 if (register_pernet_device(&default_device_ops))
10675 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10676 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10678 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10679 NULL, dev_cpu_dead);
10686 subsys_initcall(net_dev_init);