2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/sched/mm.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <linux/bpf.h>
99 #include <linux/bpf_trace.h>
100 #include <net/net_namespace.h>
101 #include <net/sock.h>
102 #include <net/busy_poll.h>
103 #include <linux/rtnetlink.h>
104 #include <linux/stat.h>
106 #include <net/dst_metadata.h>
107 #include <net/pkt_sched.h>
108 #include <net/pkt_cls.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <linux/highmem.h>
112 #include <linux/init.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <net/mpls.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <trace/events/net.h>
134 #include <trace/events/skb.h>
135 #include <linux/pci.h>
136 #include <linux/inetdevice.h>
137 #include <linux/cpu_rmap.h>
138 #include <linux/static_key.h>
139 #include <linux/hashtable.h>
140 #include <linux/vmalloc.h>
141 #include <linux/if_macvlan.h>
142 #include <linux/errqueue.h>
143 #include <linux/hrtimer.h>
144 #include <linux/netfilter_ingress.h>
145 #include <linux/crash_dump.h>
146 #include <linux/sctp.h>
147 #include <net/udp_tunnel.h>
148 #include <linux/net_namespace.h>
150 #include "net-sysfs.h"
152 /* Instead of increasing this, you should create a hash table. */
153 #define MAX_GRO_SKBS 8
155 /* This should be increased if a protocol with a bigger head is added. */
156 #define GRO_MAX_HEAD (MAX_HEADER + 128)
158 static DEFINE_SPINLOCK(ptype_lock);
159 static DEFINE_SPINLOCK(offload_lock);
160 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
161 struct list_head ptype_all __read_mostly; /* Taps */
162 static struct list_head offload_base __read_mostly;
164 static int netif_rx_internal(struct sk_buff *skb);
165 static int call_netdevice_notifiers_info(unsigned long val,
166 struct netdev_notifier_info *info);
167 static struct napi_struct *napi_by_id(unsigned int napi_id);
170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
175 * Writers must hold the rtnl semaphore while they loop through the
176 * dev_base_head list, and hold dev_base_lock for writing when they do the
177 * actual updates. This allows pure readers to access the list even
178 * while a writer is preparing to update it.
180 * To put it another way, dev_base_lock is held for writing only to
181 * protect against pure readers; the rtnl semaphore provides the
182 * protection against other writers.
184 * See, for example usages, register_netdevice() and
185 * unregister_netdevice(), which must be called with the rtnl
188 DEFINE_RWLOCK(dev_base_lock);
189 EXPORT_SYMBOL(dev_base_lock);
191 static DEFINE_MUTEX(ifalias_mutex);
193 /* protects napi_hash addition/deletion and napi_gen_id */
194 static DEFINE_SPINLOCK(napi_hash_lock);
196 static unsigned int napi_gen_id = NR_CPUS;
197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
199 static seqcount_t devnet_rename_seq;
201 static inline void dev_base_seq_inc(struct net *net)
203 while (++net->dev_base_seq == 0)
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
219 static inline void rps_lock(struct softnet_data *sd)
222 spin_lock(&sd->input_pkt_queue.lock);
226 static inline void rps_unlock(struct softnet_data *sd)
229 spin_unlock(&sd->input_pkt_queue.lock);
233 /* Device list insertion */
234 static void list_netdevice(struct net_device *dev)
236 struct net *net = dev_net(dev);
240 write_lock_bh(&dev_base_lock);
241 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
243 hlist_add_head_rcu(&dev->index_hlist,
244 dev_index_hash(net, dev->ifindex));
245 write_unlock_bh(&dev_base_lock);
247 dev_base_seq_inc(net);
250 /* Device list removal
251 * caller must respect a RCU grace period before freeing/reusing dev
253 static void unlist_netdevice(struct net_device *dev)
257 /* Unlink dev from the device chain */
258 write_lock_bh(&dev_base_lock);
259 list_del_rcu(&dev->dev_list);
260 hlist_del_rcu(&dev->name_hlist);
261 hlist_del_rcu(&dev->index_hlist);
262 write_unlock_bh(&dev_base_lock);
264 dev_base_seq_inc(dev_net(dev));
271 static RAW_NOTIFIER_HEAD(netdev_chain);
274 * Device drivers call our routines to queue packets here. We empty the
275 * queue in the local softnet handler.
278 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
279 EXPORT_PER_CPU_SYMBOL(softnet_data);
281 #ifdef CONFIG_LOCKDEP
283 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
284 * according to dev->type
286 static const unsigned short netdev_lock_type[] = {
287 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
288 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
289 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
290 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
291 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
292 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
293 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
294 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
295 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
296 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
297 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
298 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
299 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
300 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
301 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
303 static const char *const netdev_lock_name[] = {
304 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
305 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
306 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
307 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
308 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
309 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
310 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
311 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
312 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
313 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
314 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
315 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
316 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
317 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
318 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
320 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
321 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
323 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
327 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
328 if (netdev_lock_type[i] == dev_type)
330 /* the last key is used by default */
331 return ARRAY_SIZE(netdev_lock_type) - 1;
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
339 i = netdev_lock_pos(dev_type);
340 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
348 i = netdev_lock_pos(dev->type);
349 lockdep_set_class_and_name(&dev->addr_list_lock,
350 &netdev_addr_lock_key[i],
351 netdev_lock_name[i]);
354 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
355 unsigned short dev_type)
358 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
363 /*******************************************************************************
365 * Protocol management and registration routines
367 *******************************************************************************/
371 * Add a protocol ID to the list. Now that the input handler is
372 * smarter we can dispense with all the messy stuff that used to be
375 * BEWARE!!! Protocol handlers, mangling input packets,
376 * MUST BE last in hash buckets and checking protocol handlers
377 * MUST start from promiscuous ptype_all chain in net_bh.
378 * It is true now, do not change it.
379 * Explanation follows: if protocol handler, mangling packet, will
380 * be the first on list, it is not able to sense, that packet
381 * is cloned and should be copied-on-write, so that it will
382 * change it and subsequent readers will get broken packet.
386 static inline struct list_head *ptype_head(const struct packet_type *pt)
388 if (pt->type == htons(ETH_P_ALL))
389 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
391 return pt->dev ? &pt->dev->ptype_specific :
392 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
396 * dev_add_pack - add packet handler
397 * @pt: packet type declaration
399 * Add a protocol handler to the networking stack. The passed &packet_type
400 * is linked into kernel lists and may not be freed until it has been
401 * removed from the kernel lists.
403 * This call does not sleep therefore it can not
404 * guarantee all CPU's that are in middle of receiving packets
405 * will see the new packet type (until the next received packet).
408 void dev_add_pack(struct packet_type *pt)
410 struct list_head *head = ptype_head(pt);
412 spin_lock(&ptype_lock);
413 list_add_rcu(&pt->list, head);
414 spin_unlock(&ptype_lock);
416 EXPORT_SYMBOL(dev_add_pack);
419 * __dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * The packet type might still be in use by receivers
428 * and must not be freed until after all the CPU's have gone
429 * through a quiescent state.
431 void __dev_remove_pack(struct packet_type *pt)
433 struct list_head *head = ptype_head(pt);
434 struct packet_type *pt1;
436 spin_lock(&ptype_lock);
438 list_for_each_entry(pt1, head, list) {
440 list_del_rcu(&pt->list);
445 pr_warn("dev_remove_pack: %p not found\n", pt);
447 spin_unlock(&ptype_lock);
449 EXPORT_SYMBOL(__dev_remove_pack);
452 * dev_remove_pack - remove packet handler
453 * @pt: packet type declaration
455 * Remove a protocol handler that was previously added to the kernel
456 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
457 * from the kernel lists and can be freed or reused once this function
460 * This call sleeps to guarantee that no CPU is looking at the packet
463 void dev_remove_pack(struct packet_type *pt)
465 __dev_remove_pack(pt);
469 EXPORT_SYMBOL(dev_remove_pack);
473 * dev_add_offload - register offload handlers
474 * @po: protocol offload declaration
476 * Add protocol offload handlers to the networking stack. The passed
477 * &proto_offload is linked into kernel lists and may not be freed until
478 * it has been removed from the kernel lists.
480 * This call does not sleep therefore it can not
481 * guarantee all CPU's that are in middle of receiving packets
482 * will see the new offload handlers (until the next received packet).
484 void dev_add_offload(struct packet_offload *po)
486 struct packet_offload *elem;
488 spin_lock(&offload_lock);
489 list_for_each_entry(elem, &offload_base, list) {
490 if (po->priority < elem->priority)
493 list_add_rcu(&po->list, elem->list.prev);
494 spin_unlock(&offload_lock);
496 EXPORT_SYMBOL(dev_add_offload);
499 * __dev_remove_offload - remove offload handler
500 * @po: packet offload declaration
502 * Remove a protocol offload handler that was previously added to the
503 * kernel offload handlers by dev_add_offload(). The passed &offload_type
504 * is removed from the kernel lists and can be freed or reused once this
507 * The packet type might still be in use by receivers
508 * and must not be freed until after all the CPU's have gone
509 * through a quiescent state.
511 static void __dev_remove_offload(struct packet_offload *po)
513 struct list_head *head = &offload_base;
514 struct packet_offload *po1;
516 spin_lock(&offload_lock);
518 list_for_each_entry(po1, head, list) {
520 list_del_rcu(&po->list);
525 pr_warn("dev_remove_offload: %p not found\n", po);
527 spin_unlock(&offload_lock);
531 * dev_remove_offload - remove packet offload handler
532 * @po: packet offload declaration
534 * Remove a packet offload handler that was previously added to the kernel
535 * offload handlers by dev_add_offload(). The passed &offload_type is
536 * removed from the kernel lists and can be freed or reused once this
539 * This call sleeps to guarantee that no CPU is looking at the packet
542 void dev_remove_offload(struct packet_offload *po)
544 __dev_remove_offload(po);
548 EXPORT_SYMBOL(dev_remove_offload);
550 /******************************************************************************
552 * Device Boot-time Settings Routines
554 ******************************************************************************/
556 /* Boot time configuration table */
557 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
560 * netdev_boot_setup_add - add new setup entry
561 * @name: name of the device
562 * @map: configured settings for the device
564 * Adds new setup entry to the dev_boot_setup list. The function
565 * returns 0 on error and 1 on success. This is a generic routine to
568 static int netdev_boot_setup_add(char *name, struct ifmap *map)
570 struct netdev_boot_setup *s;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
576 memset(s[i].name, 0, sizeof(s[i].name));
577 strlcpy(s[i].name, name, IFNAMSIZ);
578 memcpy(&s[i].map, map, sizeof(s[i].map));
583 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
587 * netdev_boot_setup_check - check boot time settings
588 * @dev: the netdevice
590 * Check boot time settings for the device.
591 * The found settings are set for the device to be used
592 * later in the device probing.
593 * Returns 0 if no settings found, 1 if they are.
595 int netdev_boot_setup_check(struct net_device *dev)
597 struct netdev_boot_setup *s = dev_boot_setup;
600 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
601 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
602 !strcmp(dev->name, s[i].name)) {
603 dev->irq = s[i].map.irq;
604 dev->base_addr = s[i].map.base_addr;
605 dev->mem_start = s[i].map.mem_start;
606 dev->mem_end = s[i].map.mem_end;
612 EXPORT_SYMBOL(netdev_boot_setup_check);
616 * netdev_boot_base - get address from boot time settings
617 * @prefix: prefix for network device
618 * @unit: id for network device
620 * Check boot time settings for the base address of device.
621 * The found settings are set for the device to be used
622 * later in the device probing.
623 * Returns 0 if no settings found.
625 unsigned long netdev_boot_base(const char *prefix, int unit)
627 const struct netdev_boot_setup *s = dev_boot_setup;
631 sprintf(name, "%s%d", prefix, unit);
634 * If device already registered then return base of 1
635 * to indicate not to probe for this interface
637 if (__dev_get_by_name(&init_net, name))
640 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
641 if (!strcmp(name, s[i].name))
642 return s[i].map.base_addr;
647 * Saves at boot time configured settings for any netdevice.
649 int __init netdev_boot_setup(char *str)
654 str = get_options(str, ARRAY_SIZE(ints), ints);
659 memset(&map, 0, sizeof(map));
663 map.base_addr = ints[2];
665 map.mem_start = ints[3];
667 map.mem_end = ints[4];
669 /* Add new entry to the list */
670 return netdev_boot_setup_add(str, &map);
673 __setup("netdev=", netdev_boot_setup);
675 /*******************************************************************************
677 * Device Interface Subroutines
679 *******************************************************************************/
682 * dev_get_iflink - get 'iflink' value of a interface
683 * @dev: targeted interface
685 * Indicates the ifindex the interface is linked to.
686 * Physical interfaces have the same 'ifindex' and 'iflink' values.
689 int dev_get_iflink(const struct net_device *dev)
691 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
692 return dev->netdev_ops->ndo_get_iflink(dev);
696 EXPORT_SYMBOL(dev_get_iflink);
699 * dev_fill_metadata_dst - Retrieve tunnel egress information.
700 * @dev: targeted interface
703 * For better visibility of tunnel traffic OVS needs to retrieve
704 * egress tunnel information for a packet. Following API allows
705 * user to get this info.
707 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
709 struct ip_tunnel_info *info;
711 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
714 info = skb_tunnel_info_unclone(skb);
717 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
720 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
722 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
725 * __dev_get_by_name - find a device by its name
726 * @net: the applicable net namespace
727 * @name: name to find
729 * Find an interface by name. Must be called under RTNL semaphore
730 * or @dev_base_lock. If the name is found a pointer to the device
731 * is returned. If the name is not found then %NULL is returned. The
732 * reference counters are not incremented so the caller must be
733 * careful with locks.
736 struct net_device *__dev_get_by_name(struct net *net, const char *name)
738 struct net_device *dev;
739 struct hlist_head *head = dev_name_hash(net, name);
741 hlist_for_each_entry(dev, head, name_hlist)
742 if (!strncmp(dev->name, name, IFNAMSIZ))
747 EXPORT_SYMBOL(__dev_get_by_name);
750 * dev_get_by_name_rcu - find a device by its name
751 * @net: the applicable net namespace
752 * @name: name to find
754 * Find an interface by name.
755 * If the name is found a pointer to the device is returned.
756 * If the name is not found then %NULL is returned.
757 * The reference counters are not incremented so the caller must be
758 * careful with locks. The caller must hold RCU lock.
761 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
763 struct net_device *dev;
764 struct hlist_head *head = dev_name_hash(net, name);
766 hlist_for_each_entry_rcu(dev, head, name_hlist)
767 if (!strncmp(dev->name, name, IFNAMSIZ))
772 EXPORT_SYMBOL(dev_get_by_name_rcu);
775 * dev_get_by_name - find a device by its name
776 * @net: the applicable net namespace
777 * @name: name to find
779 * Find an interface by name. This can be called from any
780 * context and does its own locking. The returned handle has
781 * the usage count incremented and the caller must use dev_put() to
782 * release it when it is no longer needed. %NULL is returned if no
783 * matching device is found.
786 struct net_device *dev_get_by_name(struct net *net, const char *name)
788 struct net_device *dev;
791 dev = dev_get_by_name_rcu(net, name);
797 EXPORT_SYMBOL(dev_get_by_name);
800 * __dev_get_by_index - find a device by its ifindex
801 * @net: the applicable net namespace
802 * @ifindex: index of device
804 * Search for an interface by index. Returns %NULL if the device
805 * is not found or a pointer to the device. The device has not
806 * had its reference counter increased so the caller must be careful
807 * about locking. The caller must hold either the RTNL semaphore
811 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
813 struct net_device *dev;
814 struct hlist_head *head = dev_index_hash(net, ifindex);
816 hlist_for_each_entry(dev, head, index_hlist)
817 if (dev->ifindex == ifindex)
822 EXPORT_SYMBOL(__dev_get_by_index);
825 * dev_get_by_index_rcu - find a device by its ifindex
826 * @net: the applicable net namespace
827 * @ifindex: index of device
829 * Search for an interface by index. Returns %NULL if the device
830 * is not found or a pointer to the device. The device has not
831 * had its reference counter increased so the caller must be careful
832 * about locking. The caller must hold RCU lock.
835 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
837 struct net_device *dev;
838 struct hlist_head *head = dev_index_hash(net, ifindex);
840 hlist_for_each_entry_rcu(dev, head, index_hlist)
841 if (dev->ifindex == ifindex)
846 EXPORT_SYMBOL(dev_get_by_index_rcu);
850 * dev_get_by_index - find a device by its ifindex
851 * @net: the applicable net namespace
852 * @ifindex: index of device
854 * Search for an interface by index. Returns NULL if the device
855 * is not found or a pointer to the device. The device returned has
856 * had a reference added and the pointer is safe until the user calls
857 * dev_put to indicate they have finished with it.
860 struct net_device *dev_get_by_index(struct net *net, int ifindex)
862 struct net_device *dev;
865 dev = dev_get_by_index_rcu(net, ifindex);
871 EXPORT_SYMBOL(dev_get_by_index);
874 * dev_get_by_napi_id - find a device by napi_id
875 * @napi_id: ID of the NAPI struct
877 * Search for an interface by NAPI ID. Returns %NULL if the device
878 * is not found or a pointer to the device. The device has not had
879 * its reference counter increased so the caller must be careful
880 * about locking. The caller must hold RCU lock.
883 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
885 struct napi_struct *napi;
887 WARN_ON_ONCE(!rcu_read_lock_held());
889 if (napi_id < MIN_NAPI_ID)
892 napi = napi_by_id(napi_id);
894 return napi ? napi->dev : NULL;
896 EXPORT_SYMBOL(dev_get_by_napi_id);
899 * netdev_get_name - get a netdevice name, knowing its ifindex.
900 * @net: network namespace
901 * @name: a pointer to the buffer where the name will be stored.
902 * @ifindex: the ifindex of the interface to get the name from.
904 * The use of raw_seqcount_begin() and cond_resched() before
905 * retrying is required as we want to give the writers a chance
906 * to complete when CONFIG_PREEMPT is not set.
908 int netdev_get_name(struct net *net, char *name, int ifindex)
910 struct net_device *dev;
914 seq = raw_seqcount_begin(&devnet_rename_seq);
916 dev = dev_get_by_index_rcu(net, ifindex);
922 strcpy(name, dev->name);
924 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
933 * dev_getbyhwaddr_rcu - find a device by its hardware address
934 * @net: the applicable net namespace
935 * @type: media type of device
936 * @ha: hardware address
938 * Search for an interface by MAC address. Returns NULL if the device
939 * is not found or a pointer to the device.
940 * The caller must hold RCU or RTNL.
941 * The returned device has not had its ref count increased
942 * and the caller must therefore be careful about locking
946 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
949 struct net_device *dev;
951 for_each_netdev_rcu(net, dev)
952 if (dev->type == type &&
953 !memcmp(dev->dev_addr, ha, dev->addr_len))
958 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
960 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
962 struct net_device *dev;
965 for_each_netdev(net, dev)
966 if (dev->type == type)
971 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
973 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
975 struct net_device *dev, *ret = NULL;
978 for_each_netdev_rcu(net, dev)
979 if (dev->type == type) {
987 EXPORT_SYMBOL(dev_getfirstbyhwtype);
990 * __dev_get_by_flags - find any device with given flags
991 * @net: the applicable net namespace
992 * @if_flags: IFF_* values
993 * @mask: bitmask of bits in if_flags to check
995 * Search for any interface with the given flags. Returns NULL if a device
996 * is not found or a pointer to the device. Must be called inside
997 * rtnl_lock(), and result refcount is unchanged.
1000 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1001 unsigned short mask)
1003 struct net_device *dev, *ret;
1008 for_each_netdev(net, dev) {
1009 if (((dev->flags ^ if_flags) & mask) == 0) {
1016 EXPORT_SYMBOL(__dev_get_by_flags);
1019 * dev_valid_name - check if name is okay for network device
1020 * @name: name string
1022 * Network device names need to be valid file names to
1023 * to allow sysfs to work. We also disallow any kind of
1026 bool dev_valid_name(const char *name)
1030 if (strlen(name) >= IFNAMSIZ)
1032 if (!strcmp(name, ".") || !strcmp(name, ".."))
1036 if (*name == '/' || *name == ':' || isspace(*name))
1042 EXPORT_SYMBOL(dev_valid_name);
1045 * __dev_alloc_name - allocate a name for a device
1046 * @net: network namespace to allocate the device name in
1047 * @name: name format string
1048 * @buf: scratch buffer and result name string
1050 * Passed a format string - eg "lt%d" it will try and find a suitable
1051 * id. It scans list of devices to build up a free map, then chooses
1052 * the first empty slot. The caller must hold the dev_base or rtnl lock
1053 * while allocating the name and adding the device in order to avoid
1055 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1056 * Returns the number of the unit assigned or a negative errno code.
1059 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1063 const int max_netdevices = 8*PAGE_SIZE;
1064 unsigned long *inuse;
1065 struct net_device *d;
1067 p = strnchr(name, IFNAMSIZ-1, '%');
1070 * Verify the string as this thing may have come from
1071 * the user. There must be either one "%d" and no other "%"
1074 if (p[1] != 'd' || strchr(p + 2, '%'))
1077 /* Use one page as a bit array of possible slots */
1078 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1082 for_each_netdev(net, d) {
1083 if (!sscanf(d->name, name, &i))
1085 if (i < 0 || i >= max_netdevices)
1088 /* avoid cases where sscanf is not exact inverse of printf */
1089 snprintf(buf, IFNAMSIZ, name, i);
1090 if (!strncmp(buf, d->name, IFNAMSIZ))
1094 i = find_first_zero_bit(inuse, max_netdevices);
1095 free_page((unsigned long) inuse);
1099 snprintf(buf, IFNAMSIZ, name, i);
1100 if (!__dev_get_by_name(net, buf))
1103 /* It is possible to run out of possible slots
1104 * when the name is long and there isn't enough space left
1105 * for the digits, or if all bits are used.
1111 * dev_alloc_name - allocate a name for a device
1113 * @name: name format string
1115 * Passed a format string - eg "lt%d" it will try and find a suitable
1116 * id. It scans list of devices to build up a free map, then chooses
1117 * the first empty slot. The caller must hold the dev_base or rtnl lock
1118 * while allocating the name and adding the device in order to avoid
1120 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1121 * Returns the number of the unit assigned or a negative errno code.
1124 int dev_alloc_name(struct net_device *dev, const char *name)
1130 BUG_ON(!dev_net(dev));
1132 ret = __dev_alloc_name(net, name, buf);
1134 strlcpy(dev->name, buf, IFNAMSIZ);
1137 EXPORT_SYMBOL(dev_alloc_name);
1139 static int dev_alloc_name_ns(struct net *net,
1140 struct net_device *dev,
1146 ret = __dev_alloc_name(net, name, buf);
1148 strlcpy(dev->name, buf, IFNAMSIZ);
1152 int dev_get_valid_name(struct net *net, struct net_device *dev,
1157 if (!dev_valid_name(name))
1160 if (strchr(name, '%'))
1161 return dev_alloc_name_ns(net, dev, name);
1162 else if (__dev_get_by_name(net, name))
1164 else if (dev->name != name)
1165 strlcpy(dev->name, name, IFNAMSIZ);
1169 EXPORT_SYMBOL(dev_get_valid_name);
1172 * dev_change_name - change name of a device
1174 * @newname: name (or format string) must be at least IFNAMSIZ
1176 * Change name of a device, can pass format strings "eth%d".
1179 int dev_change_name(struct net_device *dev, const char *newname)
1181 unsigned char old_assign_type;
1182 char oldname[IFNAMSIZ];
1188 BUG_ON(!dev_net(dev));
1191 if (dev->flags & IFF_UP)
1194 write_seqcount_begin(&devnet_rename_seq);
1196 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1197 write_seqcount_end(&devnet_rename_seq);
1201 memcpy(oldname, dev->name, IFNAMSIZ);
1203 err = dev_get_valid_name(net, dev, newname);
1205 write_seqcount_end(&devnet_rename_seq);
1209 if (oldname[0] && !strchr(oldname, '%'))
1210 netdev_info(dev, "renamed from %s\n", oldname);
1212 old_assign_type = dev->name_assign_type;
1213 dev->name_assign_type = NET_NAME_RENAMED;
1216 ret = device_rename(&dev->dev, dev->name);
1218 memcpy(dev->name, oldname, IFNAMSIZ);
1219 dev->name_assign_type = old_assign_type;
1220 write_seqcount_end(&devnet_rename_seq);
1224 write_seqcount_end(&devnet_rename_seq);
1226 netdev_adjacent_rename_links(dev, oldname);
1228 write_lock_bh(&dev_base_lock);
1229 hlist_del_rcu(&dev->name_hlist);
1230 write_unlock_bh(&dev_base_lock);
1234 write_lock_bh(&dev_base_lock);
1235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1236 write_unlock_bh(&dev_base_lock);
1238 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1239 ret = notifier_to_errno(ret);
1242 /* err >= 0 after dev_alloc_name() or stores the first errno */
1245 write_seqcount_begin(&devnet_rename_seq);
1246 memcpy(dev->name, oldname, IFNAMSIZ);
1247 memcpy(oldname, newname, IFNAMSIZ);
1248 dev->name_assign_type = old_assign_type;
1249 old_assign_type = NET_NAME_RENAMED;
1252 pr_err("%s: name change rollback failed: %d\n",
1261 * dev_set_alias - change ifalias of a device
1263 * @alias: name up to IFALIASZ
1264 * @len: limit of bytes to copy from info
1266 * Set ifalias for a device,
1268 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1270 struct dev_ifalias *new_alias = NULL;
1272 if (len >= IFALIASZ)
1276 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1280 memcpy(new_alias->ifalias, alias, len);
1281 new_alias->ifalias[len] = 0;
1284 mutex_lock(&ifalias_mutex);
1285 rcu_swap_protected(dev->ifalias, new_alias,
1286 mutex_is_locked(&ifalias_mutex));
1287 mutex_unlock(&ifalias_mutex);
1290 kfree_rcu(new_alias, rcuhead);
1296 * dev_get_alias - get ifalias of a device
1298 * @name: buffer to store name of ifalias
1299 * @len: size of buffer
1301 * get ifalias for a device. Caller must make sure dev cannot go
1302 * away, e.g. rcu read lock or own a reference count to device.
1304 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1306 const struct dev_ifalias *alias;
1310 alias = rcu_dereference(dev->ifalias);
1312 ret = snprintf(name, len, "%s", alias->ifalias);
1319 * netdev_features_change - device changes features
1320 * @dev: device to cause notification
1322 * Called to indicate a device has changed features.
1324 void netdev_features_change(struct net_device *dev)
1326 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1328 EXPORT_SYMBOL(netdev_features_change);
1331 * netdev_state_change - device changes state
1332 * @dev: device to cause notification
1334 * Called to indicate a device has changed state. This function calls
1335 * the notifier chains for netdev_chain and sends a NEWLINK message
1336 * to the routing socket.
1338 void netdev_state_change(struct net_device *dev)
1340 if (dev->flags & IFF_UP) {
1341 struct netdev_notifier_change_info change_info = {
1345 call_netdevice_notifiers_info(NETDEV_CHANGE,
1347 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1350 EXPORT_SYMBOL(netdev_state_change);
1353 * netdev_notify_peers - notify network peers about existence of @dev
1354 * @dev: network device
1356 * Generate traffic such that interested network peers are aware of
1357 * @dev, such as by generating a gratuitous ARP. This may be used when
1358 * a device wants to inform the rest of the network about some sort of
1359 * reconfiguration such as a failover event or virtual machine
1362 void netdev_notify_peers(struct net_device *dev)
1365 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1366 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1369 EXPORT_SYMBOL(netdev_notify_peers);
1371 static int __dev_open(struct net_device *dev)
1373 const struct net_device_ops *ops = dev->netdev_ops;
1378 if (!netif_device_present(dev))
1381 /* Block netpoll from trying to do any rx path servicing.
1382 * If we don't do this there is a chance ndo_poll_controller
1383 * or ndo_poll may be running while we open the device
1385 netpoll_poll_disable(dev);
1387 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1388 ret = notifier_to_errno(ret);
1392 set_bit(__LINK_STATE_START, &dev->state);
1394 if (ops->ndo_validate_addr)
1395 ret = ops->ndo_validate_addr(dev);
1397 if (!ret && ops->ndo_open)
1398 ret = ops->ndo_open(dev);
1400 netpoll_poll_enable(dev);
1403 clear_bit(__LINK_STATE_START, &dev->state);
1405 dev->flags |= IFF_UP;
1406 dev_set_rx_mode(dev);
1408 add_device_randomness(dev->dev_addr, dev->addr_len);
1415 * dev_open - prepare an interface for use.
1416 * @dev: device to open
1418 * Takes a device from down to up state. The device's private open
1419 * function is invoked and then the multicast lists are loaded. Finally
1420 * the device is moved into the up state and a %NETDEV_UP message is
1421 * sent to the netdev notifier chain.
1423 * Calling this function on an active interface is a nop. On a failure
1424 * a negative errno code is returned.
1426 int dev_open(struct net_device *dev)
1430 if (dev->flags & IFF_UP)
1433 ret = __dev_open(dev);
1437 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1438 call_netdevice_notifiers(NETDEV_UP, dev);
1442 EXPORT_SYMBOL(dev_open);
1444 static void __dev_close_many(struct list_head *head)
1446 struct net_device *dev;
1451 list_for_each_entry(dev, head, close_list) {
1452 /* Temporarily disable netpoll until the interface is down */
1453 netpoll_poll_disable(dev);
1455 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1457 clear_bit(__LINK_STATE_START, &dev->state);
1459 /* Synchronize to scheduled poll. We cannot touch poll list, it
1460 * can be even on different cpu. So just clear netif_running().
1462 * dev->stop() will invoke napi_disable() on all of it's
1463 * napi_struct instances on this device.
1465 smp_mb__after_atomic(); /* Commit netif_running(). */
1468 dev_deactivate_many(head);
1470 list_for_each_entry(dev, head, close_list) {
1471 const struct net_device_ops *ops = dev->netdev_ops;
1474 * Call the device specific close. This cannot fail.
1475 * Only if device is UP
1477 * We allow it to be called even after a DETACH hot-plug
1483 dev->flags &= ~IFF_UP;
1484 netpoll_poll_enable(dev);
1488 static void __dev_close(struct net_device *dev)
1492 list_add(&dev->close_list, &single);
1493 __dev_close_many(&single);
1497 void dev_close_many(struct list_head *head, bool unlink)
1499 struct net_device *dev, *tmp;
1501 /* Remove the devices that don't need to be closed */
1502 list_for_each_entry_safe(dev, tmp, head, close_list)
1503 if (!(dev->flags & IFF_UP))
1504 list_del_init(&dev->close_list);
1506 __dev_close_many(head);
1508 list_for_each_entry_safe(dev, tmp, head, close_list) {
1509 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1510 call_netdevice_notifiers(NETDEV_DOWN, dev);
1512 list_del_init(&dev->close_list);
1515 EXPORT_SYMBOL(dev_close_many);
1518 * dev_close - shutdown an interface.
1519 * @dev: device to shutdown
1521 * This function moves an active device into down state. A
1522 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1523 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1526 void dev_close(struct net_device *dev)
1528 if (dev->flags & IFF_UP) {
1531 list_add(&dev->close_list, &single);
1532 dev_close_many(&single, true);
1536 EXPORT_SYMBOL(dev_close);
1540 * dev_disable_lro - disable Large Receive Offload on a device
1543 * Disable Large Receive Offload (LRO) on a net device. Must be
1544 * called under RTNL. This is needed if received packets may be
1545 * forwarded to another interface.
1547 void dev_disable_lro(struct net_device *dev)
1549 struct net_device *lower_dev;
1550 struct list_head *iter;
1552 dev->wanted_features &= ~NETIF_F_LRO;
1553 netdev_update_features(dev);
1555 if (unlikely(dev->features & NETIF_F_LRO))
1556 netdev_WARN(dev, "failed to disable LRO!\n");
1558 netdev_for_each_lower_dev(dev, lower_dev, iter)
1559 dev_disable_lro(lower_dev);
1561 EXPORT_SYMBOL(dev_disable_lro);
1563 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1564 struct net_device *dev)
1566 struct netdev_notifier_info info = {
1570 return nb->notifier_call(nb, val, &info);
1573 static int dev_boot_phase = 1;
1576 * register_netdevice_notifier - register a network notifier block
1579 * Register a notifier to be called when network device events occur.
1580 * The notifier passed is linked into the kernel structures and must
1581 * not be reused until it has been unregistered. A negative errno code
1582 * is returned on a failure.
1584 * When registered all registration and up events are replayed
1585 * to the new notifier to allow device to have a race free
1586 * view of the network device list.
1589 int register_netdevice_notifier(struct notifier_block *nb)
1591 struct net_device *dev;
1592 struct net_device *last;
1597 err = raw_notifier_chain_register(&netdev_chain, nb);
1603 for_each_netdev(net, dev) {
1604 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1605 err = notifier_to_errno(err);
1609 if (!(dev->flags & IFF_UP))
1612 call_netdevice_notifier(nb, NETDEV_UP, dev);
1623 for_each_netdev(net, dev) {
1627 if (dev->flags & IFF_UP) {
1628 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1630 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1632 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1637 raw_notifier_chain_unregister(&netdev_chain, nb);
1640 EXPORT_SYMBOL(register_netdevice_notifier);
1643 * unregister_netdevice_notifier - unregister a network notifier block
1646 * Unregister a notifier previously registered by
1647 * register_netdevice_notifier(). The notifier is unlinked into the
1648 * kernel structures and may then be reused. A negative errno code
1649 * is returned on a failure.
1651 * After unregistering unregister and down device events are synthesized
1652 * for all devices on the device list to the removed notifier to remove
1653 * the need for special case cleanup code.
1656 int unregister_netdevice_notifier(struct notifier_block *nb)
1658 struct net_device *dev;
1663 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1668 for_each_netdev(net, dev) {
1669 if (dev->flags & IFF_UP) {
1670 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1672 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1674 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1681 EXPORT_SYMBOL(unregister_netdevice_notifier);
1684 * call_netdevice_notifiers_info - call all network notifier blocks
1685 * @val: value passed unmodified to notifier function
1686 * @dev: net_device pointer passed unmodified to notifier function
1687 * @info: notifier information data
1689 * Call all network notifier blocks. Parameters and return value
1690 * are as for raw_notifier_call_chain().
1693 static int call_netdevice_notifiers_info(unsigned long val,
1694 struct netdev_notifier_info *info)
1697 return raw_notifier_call_chain(&netdev_chain, val, info);
1701 * call_netdevice_notifiers - call all network notifier blocks
1702 * @val: value passed unmodified to notifier function
1703 * @dev: net_device pointer passed unmodified to notifier function
1705 * Call all network notifier blocks. Parameters and return value
1706 * are as for raw_notifier_call_chain().
1709 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1711 struct netdev_notifier_info info = {
1715 return call_netdevice_notifiers_info(val, &info);
1717 EXPORT_SYMBOL(call_netdevice_notifiers);
1719 #ifdef CONFIG_NET_INGRESS
1720 static struct static_key ingress_needed __read_mostly;
1722 void net_inc_ingress_queue(void)
1724 static_key_slow_inc(&ingress_needed);
1726 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1728 void net_dec_ingress_queue(void)
1730 static_key_slow_dec(&ingress_needed);
1732 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1735 #ifdef CONFIG_NET_EGRESS
1736 static struct static_key egress_needed __read_mostly;
1738 void net_inc_egress_queue(void)
1740 static_key_slow_inc(&egress_needed);
1742 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1744 void net_dec_egress_queue(void)
1746 static_key_slow_dec(&egress_needed);
1748 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1751 static struct static_key netstamp_needed __read_mostly;
1752 #ifdef HAVE_JUMP_LABEL
1753 static atomic_t netstamp_needed_deferred;
1754 static atomic_t netstamp_wanted;
1755 static void netstamp_clear(struct work_struct *work)
1757 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1760 wanted = atomic_add_return(deferred, &netstamp_wanted);
1762 static_key_enable(&netstamp_needed);
1764 static_key_disable(&netstamp_needed);
1766 static DECLARE_WORK(netstamp_work, netstamp_clear);
1769 void net_enable_timestamp(void)
1771 #ifdef HAVE_JUMP_LABEL
1775 wanted = atomic_read(&netstamp_wanted);
1778 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1781 atomic_inc(&netstamp_needed_deferred);
1782 schedule_work(&netstamp_work);
1784 static_key_slow_inc(&netstamp_needed);
1787 EXPORT_SYMBOL(net_enable_timestamp);
1789 void net_disable_timestamp(void)
1791 #ifdef HAVE_JUMP_LABEL
1795 wanted = atomic_read(&netstamp_wanted);
1798 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1801 atomic_dec(&netstamp_needed_deferred);
1802 schedule_work(&netstamp_work);
1804 static_key_slow_dec(&netstamp_needed);
1807 EXPORT_SYMBOL(net_disable_timestamp);
1809 static inline void net_timestamp_set(struct sk_buff *skb)
1812 if (static_key_false(&netstamp_needed))
1813 __net_timestamp(skb);
1816 #define net_timestamp_check(COND, SKB) \
1817 if (static_key_false(&netstamp_needed)) { \
1818 if ((COND) && !(SKB)->tstamp) \
1819 __net_timestamp(SKB); \
1822 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1826 if (!(dev->flags & IFF_UP))
1829 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1830 if (skb->len <= len)
1833 /* if TSO is enabled, we don't care about the length as the packet
1834 * could be forwarded without being segmented before
1836 if (skb_is_gso(skb))
1841 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1843 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1845 int ret = ____dev_forward_skb(dev, skb);
1848 skb->protocol = eth_type_trans(skb, dev);
1849 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1854 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1857 * dev_forward_skb - loopback an skb to another netif
1859 * @dev: destination network device
1860 * @skb: buffer to forward
1863 * NET_RX_SUCCESS (no congestion)
1864 * NET_RX_DROP (packet was dropped, but freed)
1866 * dev_forward_skb can be used for injecting an skb from the
1867 * start_xmit function of one device into the receive queue
1868 * of another device.
1870 * The receiving device may be in another namespace, so
1871 * we have to clear all information in the skb that could
1872 * impact namespace isolation.
1874 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1876 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1878 EXPORT_SYMBOL_GPL(dev_forward_skb);
1880 static inline int deliver_skb(struct sk_buff *skb,
1881 struct packet_type *pt_prev,
1882 struct net_device *orig_dev)
1884 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1886 refcount_inc(&skb->users);
1887 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1890 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1891 struct packet_type **pt,
1892 struct net_device *orig_dev,
1894 struct list_head *ptype_list)
1896 struct packet_type *ptype, *pt_prev = *pt;
1898 list_for_each_entry_rcu(ptype, ptype_list, list) {
1899 if (ptype->type != type)
1902 deliver_skb(skb, pt_prev, orig_dev);
1908 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1910 if (!ptype->af_packet_priv || !skb->sk)
1913 if (ptype->id_match)
1914 return ptype->id_match(ptype, skb->sk);
1915 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1922 * Support routine. Sends outgoing frames to any network
1923 * taps currently in use.
1926 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1928 struct packet_type *ptype;
1929 struct sk_buff *skb2 = NULL;
1930 struct packet_type *pt_prev = NULL;
1931 struct list_head *ptype_list = &ptype_all;
1935 list_for_each_entry_rcu(ptype, ptype_list, list) {
1936 /* Never send packets back to the socket
1937 * they originated from - MvS (miquels@drinkel.ow.org)
1939 if (skb_loop_sk(ptype, skb))
1943 deliver_skb(skb2, pt_prev, skb->dev);
1948 /* need to clone skb, done only once */
1949 skb2 = skb_clone(skb, GFP_ATOMIC);
1953 net_timestamp_set(skb2);
1955 /* skb->nh should be correctly
1956 * set by sender, so that the second statement is
1957 * just protection against buggy protocols.
1959 skb_reset_mac_header(skb2);
1961 if (skb_network_header(skb2) < skb2->data ||
1962 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1963 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1964 ntohs(skb2->protocol),
1966 skb_reset_network_header(skb2);
1969 skb2->transport_header = skb2->network_header;
1970 skb2->pkt_type = PACKET_OUTGOING;
1974 if (ptype_list == &ptype_all) {
1975 ptype_list = &dev->ptype_all;
1980 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
1981 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1987 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1990 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1991 * @dev: Network device
1992 * @txq: number of queues available
1994 * If real_num_tx_queues is changed the tc mappings may no longer be
1995 * valid. To resolve this verify the tc mapping remains valid and if
1996 * not NULL the mapping. With no priorities mapping to this
1997 * offset/count pair it will no longer be used. In the worst case TC0
1998 * is invalid nothing can be done so disable priority mappings. If is
1999 * expected that drivers will fix this mapping if they can before
2000 * calling netif_set_real_num_tx_queues.
2002 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2005 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2007 /* If TC0 is invalidated disable TC mapping */
2008 if (tc->offset + tc->count > txq) {
2009 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2014 /* Invalidated prio to tc mappings set to TC0 */
2015 for (i = 1; i < TC_BITMASK + 1; i++) {
2016 int q = netdev_get_prio_tc_map(dev, i);
2018 tc = &dev->tc_to_txq[q];
2019 if (tc->offset + tc->count > txq) {
2020 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2022 netdev_set_prio_tc_map(dev, i, 0);
2027 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2030 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2033 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2034 if ((txq - tc->offset) < tc->count)
2043 EXPORT_SYMBOL(netdev_txq_to_tc);
2046 static DEFINE_MUTEX(xps_map_mutex);
2047 #define xmap_dereference(P) \
2048 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2050 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2053 struct xps_map *map = NULL;
2057 map = xmap_dereference(dev_maps->cpu_map[tci]);
2061 for (pos = map->len; pos--;) {
2062 if (map->queues[pos] != index)
2066 map->queues[pos] = map->queues[--map->len];
2070 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
2071 kfree_rcu(map, rcu);
2078 static bool remove_xps_queue_cpu(struct net_device *dev,
2079 struct xps_dev_maps *dev_maps,
2080 int cpu, u16 offset, u16 count)
2082 int num_tc = dev->num_tc ? : 1;
2083 bool active = false;
2086 for (tci = cpu * num_tc; num_tc--; tci++) {
2089 for (i = count, j = offset; i--; j++) {
2090 if (!remove_xps_queue(dev_maps, cpu, j))
2100 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2103 struct xps_dev_maps *dev_maps;
2105 bool active = false;
2107 mutex_lock(&xps_map_mutex);
2108 dev_maps = xmap_dereference(dev->xps_maps);
2113 for_each_possible_cpu(cpu)
2114 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2118 RCU_INIT_POINTER(dev->xps_maps, NULL);
2119 kfree_rcu(dev_maps, rcu);
2122 for (i = offset + (count - 1); count--; i--)
2123 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2127 mutex_unlock(&xps_map_mutex);
2130 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2132 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2135 static struct xps_map *expand_xps_map(struct xps_map *map,
2138 struct xps_map *new_map;
2139 int alloc_len = XPS_MIN_MAP_ALLOC;
2142 for (pos = 0; map && pos < map->len; pos++) {
2143 if (map->queues[pos] != index)
2148 /* Need to add queue to this CPU's existing map */
2150 if (pos < map->alloc_len)
2153 alloc_len = map->alloc_len * 2;
2156 /* Need to allocate new map to store queue on this CPU's map */
2157 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2162 for (i = 0; i < pos; i++)
2163 new_map->queues[i] = map->queues[i];
2164 new_map->alloc_len = alloc_len;
2170 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2173 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2174 int i, cpu, tci, numa_node_id = -2;
2175 int maps_sz, num_tc = 1, tc = 0;
2176 struct xps_map *map, *new_map;
2177 bool active = false;
2180 num_tc = dev->num_tc;
2181 tc = netdev_txq_to_tc(dev, index);
2186 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2187 if (maps_sz < L1_CACHE_BYTES)
2188 maps_sz = L1_CACHE_BYTES;
2190 mutex_lock(&xps_map_mutex);
2192 dev_maps = xmap_dereference(dev->xps_maps);
2194 /* allocate memory for queue storage */
2195 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2197 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2198 if (!new_dev_maps) {
2199 mutex_unlock(&xps_map_mutex);
2203 tci = cpu * num_tc + tc;
2204 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2207 map = expand_xps_map(map, cpu, index);
2211 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2215 goto out_no_new_maps;
2217 for_each_possible_cpu(cpu) {
2218 /* copy maps belonging to foreign traffic classes */
2219 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2220 /* fill in the new device map from the old device map */
2221 map = xmap_dereference(dev_maps->cpu_map[tci]);
2222 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2225 /* We need to explicitly update tci as prevous loop
2226 * could break out early if dev_maps is NULL.
2228 tci = cpu * num_tc + tc;
2230 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2231 /* add queue to CPU maps */
2234 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2235 while ((pos < map->len) && (map->queues[pos] != index))
2238 if (pos == map->len)
2239 map->queues[map->len++] = index;
2241 if (numa_node_id == -2)
2242 numa_node_id = cpu_to_node(cpu);
2243 else if (numa_node_id != cpu_to_node(cpu))
2246 } else if (dev_maps) {
2247 /* fill in the new device map from the old device map */
2248 map = xmap_dereference(dev_maps->cpu_map[tci]);
2249 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2252 /* copy maps belonging to foreign traffic classes */
2253 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2254 /* fill in the new device map from the old device map */
2255 map = xmap_dereference(dev_maps->cpu_map[tci]);
2256 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2260 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2262 /* Cleanup old maps */
2264 goto out_no_old_maps;
2266 for_each_possible_cpu(cpu) {
2267 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2268 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2269 map = xmap_dereference(dev_maps->cpu_map[tci]);
2270 if (map && map != new_map)
2271 kfree_rcu(map, rcu);
2275 kfree_rcu(dev_maps, rcu);
2278 dev_maps = new_dev_maps;
2282 /* update Tx queue numa node */
2283 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2284 (numa_node_id >= 0) ? numa_node_id :
2290 /* removes queue from unused CPUs */
2291 for_each_possible_cpu(cpu) {
2292 for (i = tc, tci = cpu * num_tc; i--; tci++)
2293 active |= remove_xps_queue(dev_maps, tci, index);
2294 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2295 active |= remove_xps_queue(dev_maps, tci, index);
2296 for (i = num_tc - tc, tci++; --i; tci++)
2297 active |= remove_xps_queue(dev_maps, tci, index);
2300 /* free map if not active */
2302 RCU_INIT_POINTER(dev->xps_maps, NULL);
2303 kfree_rcu(dev_maps, rcu);
2307 mutex_unlock(&xps_map_mutex);
2311 /* remove any maps that we added */
2312 for_each_possible_cpu(cpu) {
2313 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2314 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2316 xmap_dereference(dev_maps->cpu_map[tci]) :
2318 if (new_map && new_map != map)
2323 mutex_unlock(&xps_map_mutex);
2325 kfree(new_dev_maps);
2328 EXPORT_SYMBOL(netif_set_xps_queue);
2331 void netdev_reset_tc(struct net_device *dev)
2334 netif_reset_xps_queues_gt(dev, 0);
2337 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2338 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2340 EXPORT_SYMBOL(netdev_reset_tc);
2342 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2344 if (tc >= dev->num_tc)
2348 netif_reset_xps_queues(dev, offset, count);
2350 dev->tc_to_txq[tc].count = count;
2351 dev->tc_to_txq[tc].offset = offset;
2354 EXPORT_SYMBOL(netdev_set_tc_queue);
2356 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2358 if (num_tc > TC_MAX_QUEUE)
2362 netif_reset_xps_queues_gt(dev, 0);
2364 dev->num_tc = num_tc;
2367 EXPORT_SYMBOL(netdev_set_num_tc);
2370 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2371 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2373 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2377 if (txq < 1 || txq > dev->num_tx_queues)
2380 if (dev->reg_state == NETREG_REGISTERED ||
2381 dev->reg_state == NETREG_UNREGISTERING) {
2384 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2390 netif_setup_tc(dev, txq);
2392 if (txq < dev->real_num_tx_queues) {
2393 qdisc_reset_all_tx_gt(dev, txq);
2395 netif_reset_xps_queues_gt(dev, txq);
2400 dev->real_num_tx_queues = txq;
2403 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2407 * netif_set_real_num_rx_queues - set actual number of RX queues used
2408 * @dev: Network device
2409 * @rxq: Actual number of RX queues
2411 * This must be called either with the rtnl_lock held or before
2412 * registration of the net device. Returns 0 on success, or a
2413 * negative error code. If called before registration, it always
2416 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2420 if (rxq < 1 || rxq > dev->num_rx_queues)
2423 if (dev->reg_state == NETREG_REGISTERED) {
2426 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2432 dev->real_num_rx_queues = rxq;
2435 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2439 * netif_get_num_default_rss_queues - default number of RSS queues
2441 * This routine should set an upper limit on the number of RSS queues
2442 * used by default by multiqueue devices.
2444 int netif_get_num_default_rss_queues(void)
2446 return is_kdump_kernel() ?
2447 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2449 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2451 static void __netif_reschedule(struct Qdisc *q)
2453 struct softnet_data *sd;
2454 unsigned long flags;
2456 local_irq_save(flags);
2457 sd = this_cpu_ptr(&softnet_data);
2458 q->next_sched = NULL;
2459 *sd->output_queue_tailp = q;
2460 sd->output_queue_tailp = &q->next_sched;
2461 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2462 local_irq_restore(flags);
2465 void __netif_schedule(struct Qdisc *q)
2467 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2468 __netif_reschedule(q);
2470 EXPORT_SYMBOL(__netif_schedule);
2472 struct dev_kfree_skb_cb {
2473 enum skb_free_reason reason;
2476 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2478 return (struct dev_kfree_skb_cb *)skb->cb;
2481 void netif_schedule_queue(struct netdev_queue *txq)
2484 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2485 struct Qdisc *q = rcu_dereference(txq->qdisc);
2487 __netif_schedule(q);
2491 EXPORT_SYMBOL(netif_schedule_queue);
2493 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2495 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2499 q = rcu_dereference(dev_queue->qdisc);
2500 __netif_schedule(q);
2504 EXPORT_SYMBOL(netif_tx_wake_queue);
2506 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2508 unsigned long flags;
2513 if (likely(refcount_read(&skb->users) == 1)) {
2515 refcount_set(&skb->users, 0);
2516 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2519 get_kfree_skb_cb(skb)->reason = reason;
2520 local_irq_save(flags);
2521 skb->next = __this_cpu_read(softnet_data.completion_queue);
2522 __this_cpu_write(softnet_data.completion_queue, skb);
2523 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2524 local_irq_restore(flags);
2526 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2528 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2530 if (in_irq() || irqs_disabled())
2531 __dev_kfree_skb_irq(skb, reason);
2535 EXPORT_SYMBOL(__dev_kfree_skb_any);
2539 * netif_device_detach - mark device as removed
2540 * @dev: network device
2542 * Mark device as removed from system and therefore no longer available.
2544 void netif_device_detach(struct net_device *dev)
2546 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2547 netif_running(dev)) {
2548 netif_tx_stop_all_queues(dev);
2551 EXPORT_SYMBOL(netif_device_detach);
2554 * netif_device_attach - mark device as attached
2555 * @dev: network device
2557 * Mark device as attached from system and restart if needed.
2559 void netif_device_attach(struct net_device *dev)
2561 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2562 netif_running(dev)) {
2563 netif_tx_wake_all_queues(dev);
2564 __netdev_watchdog_up(dev);
2567 EXPORT_SYMBOL(netif_device_attach);
2570 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2571 * to be used as a distribution range.
2573 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2574 unsigned int num_tx_queues)
2578 u16 qcount = num_tx_queues;
2580 if (skb_rx_queue_recorded(skb)) {
2581 hash = skb_get_rx_queue(skb);
2582 while (unlikely(hash >= num_tx_queues))
2583 hash -= num_tx_queues;
2588 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2590 qoffset = dev->tc_to_txq[tc].offset;
2591 qcount = dev->tc_to_txq[tc].count;
2594 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2596 EXPORT_SYMBOL(__skb_tx_hash);
2598 static void skb_warn_bad_offload(const struct sk_buff *skb)
2600 static const netdev_features_t null_features;
2601 struct net_device *dev = skb->dev;
2602 const char *name = "";
2604 if (!net_ratelimit())
2608 if (dev->dev.parent)
2609 name = dev_driver_string(dev->dev.parent);
2611 name = netdev_name(dev);
2613 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2614 "gso_type=%d ip_summed=%d\n",
2615 name, dev ? &dev->features : &null_features,
2616 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2617 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2618 skb_shinfo(skb)->gso_type, skb->ip_summed);
2622 * Invalidate hardware checksum when packet is to be mangled, and
2623 * complete checksum manually on outgoing path.
2625 int skb_checksum_help(struct sk_buff *skb)
2628 int ret = 0, offset;
2630 if (skb->ip_summed == CHECKSUM_COMPLETE)
2631 goto out_set_summed;
2633 if (unlikely(skb_shinfo(skb)->gso_size)) {
2634 skb_warn_bad_offload(skb);
2638 /* Before computing a checksum, we should make sure no frag could
2639 * be modified by an external entity : checksum could be wrong.
2641 if (skb_has_shared_frag(skb)) {
2642 ret = __skb_linearize(skb);
2647 offset = skb_checksum_start_offset(skb);
2648 BUG_ON(offset >= skb_headlen(skb));
2649 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2651 offset += skb->csum_offset;
2652 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2654 if (skb_cloned(skb) &&
2655 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2656 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2661 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2663 skb->ip_summed = CHECKSUM_NONE;
2667 EXPORT_SYMBOL(skb_checksum_help);
2669 int skb_crc32c_csum_help(struct sk_buff *skb)
2672 int ret = 0, offset, start;
2674 if (skb->ip_summed != CHECKSUM_PARTIAL)
2677 if (unlikely(skb_is_gso(skb)))
2680 /* Before computing a checksum, we should make sure no frag could
2681 * be modified by an external entity : checksum could be wrong.
2683 if (unlikely(skb_has_shared_frag(skb))) {
2684 ret = __skb_linearize(skb);
2688 start = skb_checksum_start_offset(skb);
2689 offset = start + offsetof(struct sctphdr, checksum);
2690 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2694 if (skb_cloned(skb) &&
2695 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2696 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2700 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2701 skb->len - start, ~(__u32)0,
2703 *(__le32 *)(skb->data + offset) = crc32c_csum;
2704 skb->ip_summed = CHECKSUM_NONE;
2705 skb->csum_not_inet = 0;
2710 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2712 __be16 type = skb->protocol;
2714 /* Tunnel gso handlers can set protocol to ethernet. */
2715 if (type == htons(ETH_P_TEB)) {
2718 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2721 eth = (struct ethhdr *)skb_mac_header(skb);
2722 type = eth->h_proto;
2725 return __vlan_get_protocol(skb, type, depth);
2729 * skb_mac_gso_segment - mac layer segmentation handler.
2730 * @skb: buffer to segment
2731 * @features: features for the output path (see dev->features)
2733 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2734 netdev_features_t features)
2736 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2737 struct packet_offload *ptype;
2738 int vlan_depth = skb->mac_len;
2739 __be16 type = skb_network_protocol(skb, &vlan_depth);
2741 if (unlikely(!type))
2742 return ERR_PTR(-EINVAL);
2744 __skb_pull(skb, vlan_depth);
2747 list_for_each_entry_rcu(ptype, &offload_base, list) {
2748 if (ptype->type == type && ptype->callbacks.gso_segment) {
2749 segs = ptype->callbacks.gso_segment(skb, features);
2755 __skb_push(skb, skb->data - skb_mac_header(skb));
2759 EXPORT_SYMBOL(skb_mac_gso_segment);
2762 /* openvswitch calls this on rx path, so we need a different check.
2764 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2767 return skb->ip_summed != CHECKSUM_PARTIAL;
2769 return skb->ip_summed == CHECKSUM_NONE;
2773 * __skb_gso_segment - Perform segmentation on skb.
2774 * @skb: buffer to segment
2775 * @features: features for the output path (see dev->features)
2776 * @tx_path: whether it is called in TX path
2778 * This function segments the given skb and returns a list of segments.
2780 * It may return NULL if the skb requires no segmentation. This is
2781 * only possible when GSO is used for verifying header integrity.
2783 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2785 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2786 netdev_features_t features, bool tx_path)
2788 struct sk_buff *segs;
2790 if (unlikely(skb_needs_check(skb, tx_path))) {
2793 /* We're going to init ->check field in TCP or UDP header */
2794 err = skb_cow_head(skb, 0);
2796 return ERR_PTR(err);
2799 /* Only report GSO partial support if it will enable us to
2800 * support segmentation on this frame without needing additional
2803 if (features & NETIF_F_GSO_PARTIAL) {
2804 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2805 struct net_device *dev = skb->dev;
2807 partial_features |= dev->features & dev->gso_partial_features;
2808 if (!skb_gso_ok(skb, features | partial_features))
2809 features &= ~NETIF_F_GSO_PARTIAL;
2812 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2813 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2815 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2816 SKB_GSO_CB(skb)->encap_level = 0;
2818 skb_reset_mac_header(skb);
2819 skb_reset_mac_len(skb);
2821 segs = skb_mac_gso_segment(skb, features);
2823 if (unlikely(skb_needs_check(skb, tx_path)))
2824 skb_warn_bad_offload(skb);
2828 EXPORT_SYMBOL(__skb_gso_segment);
2830 /* Take action when hardware reception checksum errors are detected. */
2832 void netdev_rx_csum_fault(struct net_device *dev)
2834 if (net_ratelimit()) {
2835 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2839 EXPORT_SYMBOL(netdev_rx_csum_fault);
2842 /* Actually, we should eliminate this check as soon as we know, that:
2843 * 1. IOMMU is present and allows to map all the memory.
2844 * 2. No high memory really exists on this machine.
2847 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2849 #ifdef CONFIG_HIGHMEM
2852 if (!(dev->features & NETIF_F_HIGHDMA)) {
2853 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2854 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2856 if (PageHighMem(skb_frag_page(frag)))
2861 if (PCI_DMA_BUS_IS_PHYS) {
2862 struct device *pdev = dev->dev.parent;
2866 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2867 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2868 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2870 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2878 /* If MPLS offload request, verify we are testing hardware MPLS features
2879 * instead of standard features for the netdev.
2881 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2882 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2883 netdev_features_t features,
2886 if (eth_p_mpls(type))
2887 features &= skb->dev->mpls_features;
2892 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2893 netdev_features_t features,
2900 static netdev_features_t harmonize_features(struct sk_buff *skb,
2901 netdev_features_t features)
2906 type = skb_network_protocol(skb, &tmp);
2907 features = net_mpls_features(skb, features, type);
2909 if (skb->ip_summed != CHECKSUM_NONE &&
2910 !can_checksum_protocol(features, type)) {
2911 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2913 if (illegal_highdma(skb->dev, skb))
2914 features &= ~NETIF_F_SG;
2919 netdev_features_t passthru_features_check(struct sk_buff *skb,
2920 struct net_device *dev,
2921 netdev_features_t features)
2925 EXPORT_SYMBOL(passthru_features_check);
2927 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2928 struct net_device *dev,
2929 netdev_features_t features)
2931 return vlan_features_check(skb, features);
2934 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2935 struct net_device *dev,
2936 netdev_features_t features)
2938 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2940 if (gso_segs > dev->gso_max_segs)
2941 return features & ~NETIF_F_GSO_MASK;
2943 /* Support for GSO partial features requires software
2944 * intervention before we can actually process the packets
2945 * so we need to strip support for any partial features now
2946 * and we can pull them back in after we have partially
2947 * segmented the frame.
2949 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2950 features &= ~dev->gso_partial_features;
2952 /* Make sure to clear the IPv4 ID mangling feature if the
2953 * IPv4 header has the potential to be fragmented.
2955 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2956 struct iphdr *iph = skb->encapsulation ?
2957 inner_ip_hdr(skb) : ip_hdr(skb);
2959 if (!(iph->frag_off & htons(IP_DF)))
2960 features &= ~NETIF_F_TSO_MANGLEID;
2966 netdev_features_t netif_skb_features(struct sk_buff *skb)
2968 struct net_device *dev = skb->dev;
2969 netdev_features_t features = dev->features;
2971 if (skb_is_gso(skb))
2972 features = gso_features_check(skb, dev, features);
2974 /* If encapsulation offload request, verify we are testing
2975 * hardware encapsulation features instead of standard
2976 * features for the netdev
2978 if (skb->encapsulation)
2979 features &= dev->hw_enc_features;
2981 if (skb_vlan_tagged(skb))
2982 features = netdev_intersect_features(features,
2983 dev->vlan_features |
2984 NETIF_F_HW_VLAN_CTAG_TX |
2985 NETIF_F_HW_VLAN_STAG_TX);
2987 if (dev->netdev_ops->ndo_features_check)
2988 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2991 features &= dflt_features_check(skb, dev, features);
2993 return harmonize_features(skb, features);
2995 EXPORT_SYMBOL(netif_skb_features);
2997 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2998 struct netdev_queue *txq, bool more)
3003 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3004 dev_queue_xmit_nit(skb, dev);
3007 trace_net_dev_start_xmit(skb, dev);
3008 rc = netdev_start_xmit(skb, dev, txq, more);
3009 trace_net_dev_xmit(skb, rc, dev, len);
3014 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3015 struct netdev_queue *txq, int *ret)
3017 struct sk_buff *skb = first;
3018 int rc = NETDEV_TX_OK;
3021 struct sk_buff *next = skb->next;
3024 rc = xmit_one(skb, dev, txq, next != NULL);
3025 if (unlikely(!dev_xmit_complete(rc))) {
3031 if (netif_xmit_stopped(txq) && skb) {
3032 rc = NETDEV_TX_BUSY;
3042 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3043 netdev_features_t features)
3045 if (skb_vlan_tag_present(skb) &&
3046 !vlan_hw_offload_capable(features, skb->vlan_proto))
3047 skb = __vlan_hwaccel_push_inside(skb);
3051 int skb_csum_hwoffload_help(struct sk_buff *skb,
3052 const netdev_features_t features)
3054 if (unlikely(skb->csum_not_inet))
3055 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3056 skb_crc32c_csum_help(skb);
3058 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3060 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3062 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
3064 netdev_features_t features;
3066 features = netif_skb_features(skb);
3067 skb = validate_xmit_vlan(skb, features);
3071 if (netif_needs_gso(skb, features)) {
3072 struct sk_buff *segs;
3074 segs = skb_gso_segment(skb, features);
3082 if (skb_needs_linearize(skb, features) &&
3083 __skb_linearize(skb))
3086 if (validate_xmit_xfrm(skb, features))
3089 /* If packet is not checksummed and device does not
3090 * support checksumming for this protocol, complete
3091 * checksumming here.
3093 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3094 if (skb->encapsulation)
3095 skb_set_inner_transport_header(skb,
3096 skb_checksum_start_offset(skb));
3098 skb_set_transport_header(skb,
3099 skb_checksum_start_offset(skb));
3100 if (skb_csum_hwoffload_help(skb, features))
3110 atomic_long_inc(&dev->tx_dropped);
3114 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3116 struct sk_buff *next, *head = NULL, *tail;
3118 for (; skb != NULL; skb = next) {
3122 /* in case skb wont be segmented, point to itself */
3125 skb = validate_xmit_skb(skb, dev);
3133 /* If skb was segmented, skb->prev points to
3134 * the last segment. If not, it still contains skb.
3140 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3142 static void qdisc_pkt_len_init(struct sk_buff *skb)
3144 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3146 qdisc_skb_cb(skb)->pkt_len = skb->len;
3148 /* To get more precise estimation of bytes sent on wire,
3149 * we add to pkt_len the headers size of all segments
3151 if (shinfo->gso_size) {
3152 unsigned int hdr_len;
3153 u16 gso_segs = shinfo->gso_segs;
3155 /* mac layer + network layer */
3156 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3158 /* + transport layer */
3159 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3160 hdr_len += tcp_hdrlen(skb);
3162 hdr_len += sizeof(struct udphdr);
3164 if (shinfo->gso_type & SKB_GSO_DODGY)
3165 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3168 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3172 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3173 struct net_device *dev,
3174 struct netdev_queue *txq)
3176 spinlock_t *root_lock = qdisc_lock(q);
3177 struct sk_buff *to_free = NULL;
3181 qdisc_calculate_pkt_len(skb, q);
3183 * Heuristic to force contended enqueues to serialize on a
3184 * separate lock before trying to get qdisc main lock.
3185 * This permits qdisc->running owner to get the lock more
3186 * often and dequeue packets faster.
3188 contended = qdisc_is_running(q);
3189 if (unlikely(contended))
3190 spin_lock(&q->busylock);
3192 spin_lock(root_lock);
3193 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3194 __qdisc_drop(skb, &to_free);
3196 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3197 qdisc_run_begin(q)) {
3199 * This is a work-conserving queue; there are no old skbs
3200 * waiting to be sent out; and the qdisc is not running -
3201 * xmit the skb directly.
3204 qdisc_bstats_update(q, skb);
3206 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3207 if (unlikely(contended)) {
3208 spin_unlock(&q->busylock);
3215 rc = NET_XMIT_SUCCESS;
3217 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3218 if (qdisc_run_begin(q)) {
3219 if (unlikely(contended)) {
3220 spin_unlock(&q->busylock);
3226 spin_unlock(root_lock);
3227 if (unlikely(to_free))
3228 kfree_skb_list(to_free);
3229 if (unlikely(contended))
3230 spin_unlock(&q->busylock);
3234 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3235 static void skb_update_prio(struct sk_buff *skb)
3237 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3239 if (!skb->priority && skb->sk && map) {
3240 unsigned int prioidx =
3241 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3243 if (prioidx < map->priomap_len)
3244 skb->priority = map->priomap[prioidx];
3248 #define skb_update_prio(skb)
3251 DEFINE_PER_CPU(int, xmit_recursion);
3252 EXPORT_SYMBOL(xmit_recursion);
3255 * dev_loopback_xmit - loop back @skb
3256 * @net: network namespace this loopback is happening in
3257 * @sk: sk needed to be a netfilter okfn
3258 * @skb: buffer to transmit
3260 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3262 skb_reset_mac_header(skb);
3263 __skb_pull(skb, skb_network_offset(skb));
3264 skb->pkt_type = PACKET_LOOPBACK;
3265 skb->ip_summed = CHECKSUM_UNNECESSARY;
3266 WARN_ON(!skb_dst(skb));
3271 EXPORT_SYMBOL(dev_loopback_xmit);
3273 #ifdef CONFIG_NET_EGRESS
3274 static struct sk_buff *
3275 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3277 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3278 struct tcf_result cl_res;
3283 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3284 mini_qdisc_bstats_cpu_update(miniq, skb);
3286 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3288 case TC_ACT_RECLASSIFY:
3289 skb->tc_index = TC_H_MIN(cl_res.classid);
3292 mini_qdisc_qstats_cpu_drop(miniq);
3293 *ret = NET_XMIT_DROP;
3299 *ret = NET_XMIT_SUCCESS;
3302 case TC_ACT_REDIRECT:
3303 /* No need to push/pop skb's mac_header here on egress! */
3304 skb_do_redirect(skb);
3305 *ret = NET_XMIT_SUCCESS;
3313 #endif /* CONFIG_NET_EGRESS */
3315 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3318 struct xps_dev_maps *dev_maps;
3319 struct xps_map *map;
3320 int queue_index = -1;
3323 dev_maps = rcu_dereference(dev->xps_maps);
3325 unsigned int tci = skb->sender_cpu - 1;
3329 tci += netdev_get_prio_tc_map(dev, skb->priority);
3332 map = rcu_dereference(dev_maps->cpu_map[tci]);
3335 queue_index = map->queues[0];
3337 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3339 if (unlikely(queue_index >= dev->real_num_tx_queues))
3351 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3353 struct sock *sk = skb->sk;
3354 int queue_index = sk_tx_queue_get(sk);
3356 if (queue_index < 0 || skb->ooo_okay ||
3357 queue_index >= dev->real_num_tx_queues) {
3358 int new_index = get_xps_queue(dev, skb);
3361 new_index = skb_tx_hash(dev, skb);
3363 if (queue_index != new_index && sk &&
3365 rcu_access_pointer(sk->sk_dst_cache))
3366 sk_tx_queue_set(sk, new_index);
3368 queue_index = new_index;
3374 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3375 struct sk_buff *skb,
3378 int queue_index = 0;
3381 u32 sender_cpu = skb->sender_cpu - 1;
3383 if (sender_cpu >= (u32)NR_CPUS)
3384 skb->sender_cpu = raw_smp_processor_id() + 1;
3387 if (dev->real_num_tx_queues != 1) {
3388 const struct net_device_ops *ops = dev->netdev_ops;
3390 if (ops->ndo_select_queue)
3391 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3394 queue_index = __netdev_pick_tx(dev, skb);
3397 queue_index = netdev_cap_txqueue(dev, queue_index);
3400 skb_set_queue_mapping(skb, queue_index);
3401 return netdev_get_tx_queue(dev, queue_index);
3405 * __dev_queue_xmit - transmit a buffer
3406 * @skb: buffer to transmit
3407 * @accel_priv: private data used for L2 forwarding offload
3409 * Queue a buffer for transmission to a network device. The caller must
3410 * have set the device and priority and built the buffer before calling
3411 * this function. The function can be called from an interrupt.
3413 * A negative errno code is returned on a failure. A success does not
3414 * guarantee the frame will be transmitted as it may be dropped due
3415 * to congestion or traffic shaping.
3417 * -----------------------------------------------------------------------------------
3418 * I notice this method can also return errors from the queue disciplines,
3419 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3422 * Regardless of the return value, the skb is consumed, so it is currently
3423 * difficult to retry a send to this method. (You can bump the ref count
3424 * before sending to hold a reference for retry if you are careful.)
3426 * When calling this method, interrupts MUST be enabled. This is because
3427 * the BH enable code must have IRQs enabled so that it will not deadlock.
3430 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3432 struct net_device *dev = skb->dev;
3433 struct netdev_queue *txq;
3437 skb_reset_mac_header(skb);
3439 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3440 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3442 /* Disable soft irqs for various locks below. Also
3443 * stops preemption for RCU.
3447 skb_update_prio(skb);
3449 qdisc_pkt_len_init(skb);
3450 #ifdef CONFIG_NET_CLS_ACT
3451 skb->tc_at_ingress = 0;
3452 # ifdef CONFIG_NET_EGRESS
3453 if (static_key_false(&egress_needed)) {
3454 skb = sch_handle_egress(skb, &rc, dev);
3460 /* If device/qdisc don't need skb->dst, release it right now while
3461 * its hot in this cpu cache.
3463 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3468 txq = netdev_pick_tx(dev, skb, accel_priv);
3469 q = rcu_dereference_bh(txq->qdisc);
3471 trace_net_dev_queue(skb);
3473 rc = __dev_xmit_skb(skb, q, dev, txq);
3477 /* The device has no queue. Common case for software devices:
3478 * loopback, all the sorts of tunnels...
3480 * Really, it is unlikely that netif_tx_lock protection is necessary
3481 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3483 * However, it is possible, that they rely on protection
3486 * Check this and shot the lock. It is not prone from deadlocks.
3487 *Either shot noqueue qdisc, it is even simpler 8)
3489 if (dev->flags & IFF_UP) {
3490 int cpu = smp_processor_id(); /* ok because BHs are off */
3492 if (txq->xmit_lock_owner != cpu) {
3493 if (unlikely(__this_cpu_read(xmit_recursion) >
3494 XMIT_RECURSION_LIMIT))
3495 goto recursion_alert;
3497 skb = validate_xmit_skb(skb, dev);
3501 HARD_TX_LOCK(dev, txq, cpu);
3503 if (!netif_xmit_stopped(txq)) {
3504 __this_cpu_inc(xmit_recursion);
3505 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3506 __this_cpu_dec(xmit_recursion);
3507 if (dev_xmit_complete(rc)) {
3508 HARD_TX_UNLOCK(dev, txq);
3512 HARD_TX_UNLOCK(dev, txq);
3513 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3516 /* Recursion is detected! It is possible,
3520 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3526 rcu_read_unlock_bh();
3528 atomic_long_inc(&dev->tx_dropped);
3529 kfree_skb_list(skb);
3532 rcu_read_unlock_bh();
3536 int dev_queue_xmit(struct sk_buff *skb)
3538 return __dev_queue_xmit(skb, NULL);
3540 EXPORT_SYMBOL(dev_queue_xmit);
3542 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3544 return __dev_queue_xmit(skb, accel_priv);
3546 EXPORT_SYMBOL(dev_queue_xmit_accel);
3549 /*************************************************************************
3551 *************************************************************************/
3553 int netdev_max_backlog __read_mostly = 1000;
3554 EXPORT_SYMBOL(netdev_max_backlog);
3556 int netdev_tstamp_prequeue __read_mostly = 1;
3557 int netdev_budget __read_mostly = 300;
3558 unsigned int __read_mostly netdev_budget_usecs = 2000;
3559 int weight_p __read_mostly = 64; /* old backlog weight */
3560 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3561 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3562 int dev_rx_weight __read_mostly = 64;
3563 int dev_tx_weight __read_mostly = 64;
3565 /* Called with irq disabled */
3566 static inline void ____napi_schedule(struct softnet_data *sd,
3567 struct napi_struct *napi)
3569 list_add_tail(&napi->poll_list, &sd->poll_list);
3570 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3575 /* One global table that all flow-based protocols share. */
3576 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3577 EXPORT_SYMBOL(rps_sock_flow_table);
3578 u32 rps_cpu_mask __read_mostly;
3579 EXPORT_SYMBOL(rps_cpu_mask);
3581 struct static_key rps_needed __read_mostly;
3582 EXPORT_SYMBOL(rps_needed);
3583 struct static_key rfs_needed __read_mostly;
3584 EXPORT_SYMBOL(rfs_needed);
3586 static struct rps_dev_flow *
3587 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3588 struct rps_dev_flow *rflow, u16 next_cpu)
3590 if (next_cpu < nr_cpu_ids) {
3591 #ifdef CONFIG_RFS_ACCEL
3592 struct netdev_rx_queue *rxqueue;
3593 struct rps_dev_flow_table *flow_table;
3594 struct rps_dev_flow *old_rflow;
3599 /* Should we steer this flow to a different hardware queue? */
3600 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3601 !(dev->features & NETIF_F_NTUPLE))
3603 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3604 if (rxq_index == skb_get_rx_queue(skb))
3607 rxqueue = dev->_rx + rxq_index;
3608 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3611 flow_id = skb_get_hash(skb) & flow_table->mask;
3612 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3613 rxq_index, flow_id);
3617 rflow = &flow_table->flows[flow_id];
3619 if (old_rflow->filter == rflow->filter)
3620 old_rflow->filter = RPS_NO_FILTER;
3624 per_cpu(softnet_data, next_cpu).input_queue_head;
3627 rflow->cpu = next_cpu;
3632 * get_rps_cpu is called from netif_receive_skb and returns the target
3633 * CPU from the RPS map of the receiving queue for a given skb.
3634 * rcu_read_lock must be held on entry.
3636 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3637 struct rps_dev_flow **rflowp)
3639 const struct rps_sock_flow_table *sock_flow_table;
3640 struct netdev_rx_queue *rxqueue = dev->_rx;
3641 struct rps_dev_flow_table *flow_table;
3642 struct rps_map *map;
3647 if (skb_rx_queue_recorded(skb)) {
3648 u16 index = skb_get_rx_queue(skb);
3650 if (unlikely(index >= dev->real_num_rx_queues)) {
3651 WARN_ONCE(dev->real_num_rx_queues > 1,
3652 "%s received packet on queue %u, but number "
3653 "of RX queues is %u\n",
3654 dev->name, index, dev->real_num_rx_queues);
3660 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3662 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3663 map = rcu_dereference(rxqueue->rps_map);
3664 if (!flow_table && !map)
3667 skb_reset_network_header(skb);
3668 hash = skb_get_hash(skb);
3672 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3673 if (flow_table && sock_flow_table) {
3674 struct rps_dev_flow *rflow;
3678 /* First check into global flow table if there is a match */
3679 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3680 if ((ident ^ hash) & ~rps_cpu_mask)
3683 next_cpu = ident & rps_cpu_mask;
3685 /* OK, now we know there is a match,
3686 * we can look at the local (per receive queue) flow table
3688 rflow = &flow_table->flows[hash & flow_table->mask];
3692 * If the desired CPU (where last recvmsg was done) is
3693 * different from current CPU (one in the rx-queue flow
3694 * table entry), switch if one of the following holds:
3695 * - Current CPU is unset (>= nr_cpu_ids).
3696 * - Current CPU is offline.
3697 * - The current CPU's queue tail has advanced beyond the
3698 * last packet that was enqueued using this table entry.
3699 * This guarantees that all previous packets for the flow
3700 * have been dequeued, thus preserving in order delivery.
3702 if (unlikely(tcpu != next_cpu) &&
3703 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3704 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3705 rflow->last_qtail)) >= 0)) {
3707 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3710 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3720 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3721 if (cpu_online(tcpu)) {
3731 #ifdef CONFIG_RFS_ACCEL
3734 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3735 * @dev: Device on which the filter was set
3736 * @rxq_index: RX queue index
3737 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3738 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3740 * Drivers that implement ndo_rx_flow_steer() should periodically call
3741 * this function for each installed filter and remove the filters for
3742 * which it returns %true.
3744 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3745 u32 flow_id, u16 filter_id)
3747 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3748 struct rps_dev_flow_table *flow_table;
3749 struct rps_dev_flow *rflow;
3754 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3755 if (flow_table && flow_id <= flow_table->mask) {
3756 rflow = &flow_table->flows[flow_id];
3757 cpu = ACCESS_ONCE(rflow->cpu);
3758 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3759 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3760 rflow->last_qtail) <
3761 (int)(10 * flow_table->mask)))
3767 EXPORT_SYMBOL(rps_may_expire_flow);
3769 #endif /* CONFIG_RFS_ACCEL */
3771 /* Called from hardirq (IPI) context */
3772 static void rps_trigger_softirq(void *data)
3774 struct softnet_data *sd = data;
3776 ____napi_schedule(sd, &sd->backlog);
3780 #endif /* CONFIG_RPS */
3783 * Check if this softnet_data structure is another cpu one
3784 * If yes, queue it to our IPI list and return 1
3787 static int rps_ipi_queued(struct softnet_data *sd)
3790 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3793 sd->rps_ipi_next = mysd->rps_ipi_list;
3794 mysd->rps_ipi_list = sd;
3796 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3799 #endif /* CONFIG_RPS */
3803 #ifdef CONFIG_NET_FLOW_LIMIT
3804 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3807 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3809 #ifdef CONFIG_NET_FLOW_LIMIT
3810 struct sd_flow_limit *fl;
3811 struct softnet_data *sd;
3812 unsigned int old_flow, new_flow;
3814 if (qlen < (netdev_max_backlog >> 1))
3817 sd = this_cpu_ptr(&softnet_data);
3820 fl = rcu_dereference(sd->flow_limit);
3822 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3823 old_flow = fl->history[fl->history_head];
3824 fl->history[fl->history_head] = new_flow;
3827 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3829 if (likely(fl->buckets[old_flow]))
3830 fl->buckets[old_flow]--;
3832 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3844 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3845 * queue (may be a remote CPU queue).
3847 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3848 unsigned int *qtail)
3850 struct softnet_data *sd;
3851 unsigned long flags;
3854 sd = &per_cpu(softnet_data, cpu);
3856 local_irq_save(flags);
3859 if (!netif_running(skb->dev))
3861 qlen = skb_queue_len(&sd->input_pkt_queue);
3862 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3865 __skb_queue_tail(&sd->input_pkt_queue, skb);
3866 input_queue_tail_incr_save(sd, qtail);
3868 local_irq_restore(flags);
3869 return NET_RX_SUCCESS;
3872 /* Schedule NAPI for backlog device
3873 * We can use non atomic operation since we own the queue lock
3875 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3876 if (!rps_ipi_queued(sd))
3877 ____napi_schedule(sd, &sd->backlog);
3886 local_irq_restore(flags);
3888 atomic_long_inc(&skb->dev->rx_dropped);
3893 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
3894 struct bpf_prog *xdp_prog)
3896 u32 metalen, act = XDP_DROP;
3897 struct xdp_buff xdp;
3902 /* Reinjected packets coming from act_mirred or similar should
3903 * not get XDP generic processing.
3905 if (skb_cloned(skb))
3908 /* XDP packets must be linear and must have sufficient headroom
3909 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
3910 * native XDP provides, thus we need to do it here as well.
3912 if (skb_is_nonlinear(skb) ||
3913 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
3914 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
3915 int troom = skb->tail + skb->data_len - skb->end;
3917 /* In case we have to go down the path and also linearize,
3918 * then lets do the pskb_expand_head() work just once here.
3920 if (pskb_expand_head(skb,
3921 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
3922 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
3924 if (troom > 0 && __skb_linearize(skb))
3928 /* The XDP program wants to see the packet starting at the MAC
3931 mac_len = skb->data - skb_mac_header(skb);
3932 hlen = skb_headlen(skb) + mac_len;
3933 xdp.data = skb->data - mac_len;
3934 xdp.data_meta = xdp.data;
3935 xdp.data_end = xdp.data + hlen;
3936 xdp.data_hard_start = skb->data - skb_headroom(skb);
3937 orig_data = xdp.data;
3939 act = bpf_prog_run_xdp(xdp_prog, &xdp);
3941 off = xdp.data - orig_data;
3943 __skb_pull(skb, off);
3945 __skb_push(skb, -off);
3946 skb->mac_header += off;
3951 __skb_push(skb, mac_len);
3954 metalen = xdp.data - xdp.data_meta;
3956 skb_metadata_set(skb, metalen);
3959 bpf_warn_invalid_xdp_action(act);
3962 trace_xdp_exception(skb->dev, xdp_prog, act);
3973 /* When doing generic XDP we have to bypass the qdisc layer and the
3974 * network taps in order to match in-driver-XDP behavior.
3976 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
3978 struct net_device *dev = skb->dev;
3979 struct netdev_queue *txq;
3980 bool free_skb = true;
3983 txq = netdev_pick_tx(dev, skb, NULL);
3984 cpu = smp_processor_id();
3985 HARD_TX_LOCK(dev, txq, cpu);
3986 if (!netif_xmit_stopped(txq)) {
3987 rc = netdev_start_xmit(skb, dev, txq, 0);
3988 if (dev_xmit_complete(rc))
3991 HARD_TX_UNLOCK(dev, txq);
3993 trace_xdp_exception(dev, xdp_prog, XDP_TX);
3997 EXPORT_SYMBOL_GPL(generic_xdp_tx);
3999 static struct static_key generic_xdp_needed __read_mostly;
4001 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4004 u32 act = netif_receive_generic_xdp(skb, xdp_prog);
4007 if (act != XDP_PASS) {
4010 err = xdp_do_generic_redirect(skb->dev, skb,
4014 /* fallthru to submit skb */
4016 generic_xdp_tx(skb, xdp_prog);
4027 EXPORT_SYMBOL_GPL(do_xdp_generic);
4029 static int netif_rx_internal(struct sk_buff *skb)
4033 net_timestamp_check(netdev_tstamp_prequeue, skb);
4035 trace_netif_rx(skb);
4037 if (static_key_false(&generic_xdp_needed)) {
4042 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4046 /* Consider XDP consuming the packet a success from
4047 * the netdev point of view we do not want to count
4050 if (ret != XDP_PASS)
4051 return NET_RX_SUCCESS;
4055 if (static_key_false(&rps_needed)) {
4056 struct rps_dev_flow voidflow, *rflow = &voidflow;
4062 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4064 cpu = smp_processor_id();
4066 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4075 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4082 * netif_rx - post buffer to the network code
4083 * @skb: buffer to post
4085 * This function receives a packet from a device driver and queues it for
4086 * the upper (protocol) levels to process. It always succeeds. The buffer
4087 * may be dropped during processing for congestion control or by the
4091 * NET_RX_SUCCESS (no congestion)
4092 * NET_RX_DROP (packet was dropped)
4096 int netif_rx(struct sk_buff *skb)
4098 trace_netif_rx_entry(skb);
4100 return netif_rx_internal(skb);
4102 EXPORT_SYMBOL(netif_rx);
4104 int netif_rx_ni(struct sk_buff *skb)
4108 trace_netif_rx_ni_entry(skb);
4111 err = netif_rx_internal(skb);
4112 if (local_softirq_pending())
4118 EXPORT_SYMBOL(netif_rx_ni);
4120 static __latent_entropy void net_tx_action(struct softirq_action *h)
4122 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4124 if (sd->completion_queue) {
4125 struct sk_buff *clist;
4127 local_irq_disable();
4128 clist = sd->completion_queue;
4129 sd->completion_queue = NULL;
4133 struct sk_buff *skb = clist;
4135 clist = clist->next;
4137 WARN_ON(refcount_read(&skb->users));
4138 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4139 trace_consume_skb(skb);
4141 trace_kfree_skb(skb, net_tx_action);
4143 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4146 __kfree_skb_defer(skb);
4149 __kfree_skb_flush();
4152 if (sd->output_queue) {
4155 local_irq_disable();
4156 head = sd->output_queue;
4157 sd->output_queue = NULL;
4158 sd->output_queue_tailp = &sd->output_queue;
4162 struct Qdisc *q = head;
4163 spinlock_t *root_lock;
4165 head = head->next_sched;
4167 root_lock = qdisc_lock(q);
4168 spin_lock(root_lock);
4169 /* We need to make sure head->next_sched is read
4170 * before clearing __QDISC_STATE_SCHED
4172 smp_mb__before_atomic();
4173 clear_bit(__QDISC_STATE_SCHED, &q->state);
4175 spin_unlock(root_lock);
4180 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4181 /* This hook is defined here for ATM LANE */
4182 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4183 unsigned char *addr) __read_mostly;
4184 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4187 static inline struct sk_buff *
4188 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4189 struct net_device *orig_dev)
4191 #ifdef CONFIG_NET_CLS_ACT
4192 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4193 struct tcf_result cl_res;
4195 /* If there's at least one ingress present somewhere (so
4196 * we get here via enabled static key), remaining devices
4197 * that are not configured with an ingress qdisc will bail
4204 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4208 qdisc_skb_cb(skb)->pkt_len = skb->len;
4209 skb->tc_at_ingress = 1;
4210 mini_qdisc_bstats_cpu_update(miniq, skb);
4212 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4214 case TC_ACT_RECLASSIFY:
4215 skb->tc_index = TC_H_MIN(cl_res.classid);
4218 mini_qdisc_qstats_cpu_drop(miniq);
4226 case TC_ACT_REDIRECT:
4227 /* skb_mac_header check was done by cls/act_bpf, so
4228 * we can safely push the L2 header back before
4229 * redirecting to another netdev
4231 __skb_push(skb, skb->mac_len);
4232 skb_do_redirect(skb);
4237 #endif /* CONFIG_NET_CLS_ACT */
4242 * netdev_is_rx_handler_busy - check if receive handler is registered
4243 * @dev: device to check
4245 * Check if a receive handler is already registered for a given device.
4246 * Return true if there one.
4248 * The caller must hold the rtnl_mutex.
4250 bool netdev_is_rx_handler_busy(struct net_device *dev)
4253 return dev && rtnl_dereference(dev->rx_handler);
4255 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4258 * netdev_rx_handler_register - register receive handler
4259 * @dev: device to register a handler for
4260 * @rx_handler: receive handler to register
4261 * @rx_handler_data: data pointer that is used by rx handler
4263 * Register a receive handler for a device. This handler will then be
4264 * called from __netif_receive_skb. A negative errno code is returned
4267 * The caller must hold the rtnl_mutex.
4269 * For a general description of rx_handler, see enum rx_handler_result.
4271 int netdev_rx_handler_register(struct net_device *dev,
4272 rx_handler_func_t *rx_handler,
4273 void *rx_handler_data)
4275 if (netdev_is_rx_handler_busy(dev))
4278 /* Note: rx_handler_data must be set before rx_handler */
4279 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4280 rcu_assign_pointer(dev->rx_handler, rx_handler);
4284 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4287 * netdev_rx_handler_unregister - unregister receive handler
4288 * @dev: device to unregister a handler from
4290 * Unregister a receive handler from a device.
4292 * The caller must hold the rtnl_mutex.
4294 void netdev_rx_handler_unregister(struct net_device *dev)
4298 RCU_INIT_POINTER(dev->rx_handler, NULL);
4299 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4300 * section has a guarantee to see a non NULL rx_handler_data
4304 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4306 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4309 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4310 * the special handling of PFMEMALLOC skbs.
4312 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4314 switch (skb->protocol) {
4315 case htons(ETH_P_ARP):
4316 case htons(ETH_P_IP):
4317 case htons(ETH_P_IPV6):
4318 case htons(ETH_P_8021Q):
4319 case htons(ETH_P_8021AD):
4326 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4327 int *ret, struct net_device *orig_dev)
4329 #ifdef CONFIG_NETFILTER_INGRESS
4330 if (nf_hook_ingress_active(skb)) {
4334 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4339 ingress_retval = nf_hook_ingress(skb);
4341 return ingress_retval;
4343 #endif /* CONFIG_NETFILTER_INGRESS */
4347 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4349 struct packet_type *ptype, *pt_prev;
4350 rx_handler_func_t *rx_handler;
4351 struct net_device *orig_dev;
4352 bool deliver_exact = false;
4353 int ret = NET_RX_DROP;
4356 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4358 trace_netif_receive_skb(skb);
4360 orig_dev = skb->dev;
4362 skb_reset_network_header(skb);
4363 if (!skb_transport_header_was_set(skb))
4364 skb_reset_transport_header(skb);
4365 skb_reset_mac_len(skb);
4370 skb->skb_iif = skb->dev->ifindex;
4372 __this_cpu_inc(softnet_data.processed);
4374 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4375 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4376 skb = skb_vlan_untag(skb);
4381 if (skb_skip_tc_classify(skb))
4387 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4389 ret = deliver_skb(skb, pt_prev, orig_dev);
4393 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4395 ret = deliver_skb(skb, pt_prev, orig_dev);
4400 #ifdef CONFIG_NET_INGRESS
4401 if (static_key_false(&ingress_needed)) {
4402 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4406 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4412 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4415 if (skb_vlan_tag_present(skb)) {
4417 ret = deliver_skb(skb, pt_prev, orig_dev);
4420 if (vlan_do_receive(&skb))
4422 else if (unlikely(!skb))
4426 rx_handler = rcu_dereference(skb->dev->rx_handler);
4429 ret = deliver_skb(skb, pt_prev, orig_dev);
4432 switch (rx_handler(&skb)) {
4433 case RX_HANDLER_CONSUMED:
4434 ret = NET_RX_SUCCESS;
4436 case RX_HANDLER_ANOTHER:
4438 case RX_HANDLER_EXACT:
4439 deliver_exact = true;
4440 case RX_HANDLER_PASS:
4447 if (unlikely(skb_vlan_tag_present(skb))) {
4448 if (skb_vlan_tag_get_id(skb))
4449 skb->pkt_type = PACKET_OTHERHOST;
4450 /* Note: we might in the future use prio bits
4451 * and set skb->priority like in vlan_do_receive()
4452 * For the time being, just ignore Priority Code Point
4457 type = skb->protocol;
4459 /* deliver only exact match when indicated */
4460 if (likely(!deliver_exact)) {
4461 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4462 &ptype_base[ntohs(type) &
4466 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4467 &orig_dev->ptype_specific);
4469 if (unlikely(skb->dev != orig_dev)) {
4470 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4471 &skb->dev->ptype_specific);
4475 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4478 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4482 atomic_long_inc(&skb->dev->rx_dropped);
4484 atomic_long_inc(&skb->dev->rx_nohandler);
4486 /* Jamal, now you will not able to escape explaining
4487 * me how you were going to use this. :-)
4497 * netif_receive_skb_core - special purpose version of netif_receive_skb
4498 * @skb: buffer to process
4500 * More direct receive version of netif_receive_skb(). It should
4501 * only be used by callers that have a need to skip RPS and Generic XDP.
4502 * Caller must also take care of handling if (page_is_)pfmemalloc.
4504 * This function may only be called from softirq context and interrupts
4505 * should be enabled.
4507 * Return values (usually ignored):
4508 * NET_RX_SUCCESS: no congestion
4509 * NET_RX_DROP: packet was dropped
4511 int netif_receive_skb_core(struct sk_buff *skb)
4516 ret = __netif_receive_skb_core(skb, false);
4521 EXPORT_SYMBOL(netif_receive_skb_core);
4523 static int __netif_receive_skb(struct sk_buff *skb)
4527 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4528 unsigned int noreclaim_flag;
4531 * PFMEMALLOC skbs are special, they should
4532 * - be delivered to SOCK_MEMALLOC sockets only
4533 * - stay away from userspace
4534 * - have bounded memory usage
4536 * Use PF_MEMALLOC as this saves us from propagating the allocation
4537 * context down to all allocation sites.
4539 noreclaim_flag = memalloc_noreclaim_save();
4540 ret = __netif_receive_skb_core(skb, true);
4541 memalloc_noreclaim_restore(noreclaim_flag);
4543 ret = __netif_receive_skb_core(skb, false);
4548 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
4550 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
4551 struct bpf_prog *new = xdp->prog;
4554 switch (xdp->command) {
4555 case XDP_SETUP_PROG:
4556 rcu_assign_pointer(dev->xdp_prog, new);
4561 static_key_slow_dec(&generic_xdp_needed);
4562 } else if (new && !old) {
4563 static_key_slow_inc(&generic_xdp_needed);
4564 dev_disable_lro(dev);
4568 case XDP_QUERY_PROG:
4569 xdp->prog_attached = !!old;
4570 xdp->prog_id = old ? old->aux->id : 0;
4581 static int netif_receive_skb_internal(struct sk_buff *skb)
4585 net_timestamp_check(netdev_tstamp_prequeue, skb);
4587 if (skb_defer_rx_timestamp(skb))
4588 return NET_RX_SUCCESS;
4590 if (static_key_false(&generic_xdp_needed)) {
4595 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4599 if (ret != XDP_PASS)
4605 if (static_key_false(&rps_needed)) {
4606 struct rps_dev_flow voidflow, *rflow = &voidflow;
4607 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4610 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4616 ret = __netif_receive_skb(skb);
4622 * netif_receive_skb - process receive buffer from network
4623 * @skb: buffer to process
4625 * netif_receive_skb() is the main receive data processing function.
4626 * It always succeeds. The buffer may be dropped during processing
4627 * for congestion control or by the protocol layers.
4629 * This function may only be called from softirq context and interrupts
4630 * should be enabled.
4632 * Return values (usually ignored):
4633 * NET_RX_SUCCESS: no congestion
4634 * NET_RX_DROP: packet was dropped
4636 int netif_receive_skb(struct sk_buff *skb)
4638 trace_netif_receive_skb_entry(skb);
4640 return netif_receive_skb_internal(skb);
4642 EXPORT_SYMBOL(netif_receive_skb);
4644 DEFINE_PER_CPU(struct work_struct, flush_works);
4646 /* Network device is going away, flush any packets still pending */
4647 static void flush_backlog(struct work_struct *work)
4649 struct sk_buff *skb, *tmp;
4650 struct softnet_data *sd;
4653 sd = this_cpu_ptr(&softnet_data);
4655 local_irq_disable();
4657 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4658 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4659 __skb_unlink(skb, &sd->input_pkt_queue);
4661 input_queue_head_incr(sd);
4667 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4668 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4669 __skb_unlink(skb, &sd->process_queue);
4671 input_queue_head_incr(sd);
4677 static void flush_all_backlogs(void)
4683 for_each_online_cpu(cpu)
4684 queue_work_on(cpu, system_highpri_wq,
4685 per_cpu_ptr(&flush_works, cpu));
4687 for_each_online_cpu(cpu)
4688 flush_work(per_cpu_ptr(&flush_works, cpu));
4693 static int napi_gro_complete(struct sk_buff *skb)
4695 struct packet_offload *ptype;
4696 __be16 type = skb->protocol;
4697 struct list_head *head = &offload_base;
4700 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4702 if (NAPI_GRO_CB(skb)->count == 1) {
4703 skb_shinfo(skb)->gso_size = 0;
4708 list_for_each_entry_rcu(ptype, head, list) {
4709 if (ptype->type != type || !ptype->callbacks.gro_complete)
4712 err = ptype->callbacks.gro_complete(skb, 0);
4718 WARN_ON(&ptype->list == head);
4720 return NET_RX_SUCCESS;
4724 return netif_receive_skb_internal(skb);
4727 /* napi->gro_list contains packets ordered by age.
4728 * youngest packets at the head of it.
4729 * Complete skbs in reverse order to reduce latencies.
4731 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4733 struct sk_buff *skb, *prev = NULL;
4735 /* scan list and build reverse chain */
4736 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4741 for (skb = prev; skb; skb = prev) {
4744 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4748 napi_gro_complete(skb);
4752 napi->gro_list = NULL;
4754 EXPORT_SYMBOL(napi_gro_flush);
4756 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4759 unsigned int maclen = skb->dev->hard_header_len;
4760 u32 hash = skb_get_hash_raw(skb);
4762 for (p = napi->gro_list; p; p = p->next) {
4763 unsigned long diffs;
4765 NAPI_GRO_CB(p)->flush = 0;
4767 if (hash != skb_get_hash_raw(p)) {
4768 NAPI_GRO_CB(p)->same_flow = 0;
4772 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4773 diffs |= p->vlan_tci ^ skb->vlan_tci;
4774 diffs |= skb_metadata_dst_cmp(p, skb);
4775 diffs |= skb_metadata_differs(p, skb);
4776 if (maclen == ETH_HLEN)
4777 diffs |= compare_ether_header(skb_mac_header(p),
4778 skb_mac_header(skb));
4780 diffs = memcmp(skb_mac_header(p),
4781 skb_mac_header(skb),
4783 NAPI_GRO_CB(p)->same_flow = !diffs;
4787 static void skb_gro_reset_offset(struct sk_buff *skb)
4789 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4790 const skb_frag_t *frag0 = &pinfo->frags[0];
4792 NAPI_GRO_CB(skb)->data_offset = 0;
4793 NAPI_GRO_CB(skb)->frag0 = NULL;
4794 NAPI_GRO_CB(skb)->frag0_len = 0;
4796 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4798 !PageHighMem(skb_frag_page(frag0))) {
4799 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4800 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4801 skb_frag_size(frag0),
4802 skb->end - skb->tail);
4806 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4808 struct skb_shared_info *pinfo = skb_shinfo(skb);
4810 BUG_ON(skb->end - skb->tail < grow);
4812 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4814 skb->data_len -= grow;
4817 pinfo->frags[0].page_offset += grow;
4818 skb_frag_size_sub(&pinfo->frags[0], grow);
4820 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4821 skb_frag_unref(skb, 0);
4822 memmove(pinfo->frags, pinfo->frags + 1,
4823 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4827 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4829 struct sk_buff **pp = NULL;
4830 struct packet_offload *ptype;
4831 __be16 type = skb->protocol;
4832 struct list_head *head = &offload_base;
4834 enum gro_result ret;
4837 if (netif_elide_gro(skb->dev))
4840 gro_list_prepare(napi, skb);
4843 list_for_each_entry_rcu(ptype, head, list) {
4844 if (ptype->type != type || !ptype->callbacks.gro_receive)
4847 skb_set_network_header(skb, skb_gro_offset(skb));
4848 skb_reset_mac_len(skb);
4849 NAPI_GRO_CB(skb)->same_flow = 0;
4850 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4851 NAPI_GRO_CB(skb)->free = 0;
4852 NAPI_GRO_CB(skb)->encap_mark = 0;
4853 NAPI_GRO_CB(skb)->recursion_counter = 0;
4854 NAPI_GRO_CB(skb)->is_fou = 0;
4855 NAPI_GRO_CB(skb)->is_atomic = 1;
4856 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4858 /* Setup for GRO checksum validation */
4859 switch (skb->ip_summed) {
4860 case CHECKSUM_COMPLETE:
4861 NAPI_GRO_CB(skb)->csum = skb->csum;
4862 NAPI_GRO_CB(skb)->csum_valid = 1;
4863 NAPI_GRO_CB(skb)->csum_cnt = 0;
4865 case CHECKSUM_UNNECESSARY:
4866 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4867 NAPI_GRO_CB(skb)->csum_valid = 0;
4870 NAPI_GRO_CB(skb)->csum_cnt = 0;
4871 NAPI_GRO_CB(skb)->csum_valid = 0;
4874 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4879 if (&ptype->list == head)
4882 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
4887 same_flow = NAPI_GRO_CB(skb)->same_flow;
4888 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4891 struct sk_buff *nskb = *pp;
4895 napi_gro_complete(nskb);
4902 if (NAPI_GRO_CB(skb)->flush)
4905 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4906 struct sk_buff *nskb = napi->gro_list;
4908 /* locate the end of the list to select the 'oldest' flow */
4909 while (nskb->next) {
4915 napi_gro_complete(nskb);
4919 NAPI_GRO_CB(skb)->count = 1;
4920 NAPI_GRO_CB(skb)->age = jiffies;
4921 NAPI_GRO_CB(skb)->last = skb;
4922 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4923 skb->next = napi->gro_list;
4924 napi->gro_list = skb;
4928 grow = skb_gro_offset(skb) - skb_headlen(skb);
4930 gro_pull_from_frag0(skb, grow);
4939 struct packet_offload *gro_find_receive_by_type(__be16 type)
4941 struct list_head *offload_head = &offload_base;
4942 struct packet_offload *ptype;
4944 list_for_each_entry_rcu(ptype, offload_head, list) {
4945 if (ptype->type != type || !ptype->callbacks.gro_receive)
4951 EXPORT_SYMBOL(gro_find_receive_by_type);
4953 struct packet_offload *gro_find_complete_by_type(__be16 type)
4955 struct list_head *offload_head = &offload_base;
4956 struct packet_offload *ptype;
4958 list_for_each_entry_rcu(ptype, offload_head, list) {
4959 if (ptype->type != type || !ptype->callbacks.gro_complete)
4965 EXPORT_SYMBOL(gro_find_complete_by_type);
4967 static void napi_skb_free_stolen_head(struct sk_buff *skb)
4971 kmem_cache_free(skbuff_head_cache, skb);
4974 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4978 if (netif_receive_skb_internal(skb))
4986 case GRO_MERGED_FREE:
4987 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4988 napi_skb_free_stolen_head(skb);
5002 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5004 skb_mark_napi_id(skb, napi);
5005 trace_napi_gro_receive_entry(skb);
5007 skb_gro_reset_offset(skb);
5009 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
5011 EXPORT_SYMBOL(napi_gro_receive);
5013 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5015 if (unlikely(skb->pfmemalloc)) {
5019 __skb_pull(skb, skb_headlen(skb));
5020 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5021 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5023 skb->dev = napi->dev;
5025 skb->encapsulation = 0;
5026 skb_shinfo(skb)->gso_type = 0;
5027 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5033 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5035 struct sk_buff *skb = napi->skb;
5038 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5041 skb_mark_napi_id(skb, napi);
5046 EXPORT_SYMBOL(napi_get_frags);
5048 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5049 struct sk_buff *skb,
5055 __skb_push(skb, ETH_HLEN);
5056 skb->protocol = eth_type_trans(skb, skb->dev);
5057 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5062 napi_reuse_skb(napi, skb);
5065 case GRO_MERGED_FREE:
5066 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5067 napi_skb_free_stolen_head(skb);
5069 napi_reuse_skb(napi, skb);
5080 /* Upper GRO stack assumes network header starts at gro_offset=0
5081 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5082 * We copy ethernet header into skb->data to have a common layout.
5084 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5086 struct sk_buff *skb = napi->skb;
5087 const struct ethhdr *eth;
5088 unsigned int hlen = sizeof(*eth);
5092 skb_reset_mac_header(skb);
5093 skb_gro_reset_offset(skb);
5095 eth = skb_gro_header_fast(skb, 0);
5096 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5097 eth = skb_gro_header_slow(skb, hlen, 0);
5098 if (unlikely(!eth)) {
5099 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5100 __func__, napi->dev->name);
5101 napi_reuse_skb(napi, skb);
5105 gro_pull_from_frag0(skb, hlen);
5106 NAPI_GRO_CB(skb)->frag0 += hlen;
5107 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5109 __skb_pull(skb, hlen);
5112 * This works because the only protocols we care about don't require
5114 * We'll fix it up properly in napi_frags_finish()
5116 skb->protocol = eth->h_proto;
5121 gro_result_t napi_gro_frags(struct napi_struct *napi)
5123 struct sk_buff *skb = napi_frags_skb(napi);
5128 trace_napi_gro_frags_entry(skb);
5130 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5132 EXPORT_SYMBOL(napi_gro_frags);
5134 /* Compute the checksum from gro_offset and return the folded value
5135 * after adding in any pseudo checksum.
5137 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5142 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5144 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5145 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5147 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5148 !skb->csum_complete_sw)
5149 netdev_rx_csum_fault(skb->dev);
5152 NAPI_GRO_CB(skb)->csum = wsum;
5153 NAPI_GRO_CB(skb)->csum_valid = 1;
5157 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5159 static void net_rps_send_ipi(struct softnet_data *remsd)
5163 struct softnet_data *next = remsd->rps_ipi_next;
5165 if (cpu_online(remsd->cpu))
5166 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5173 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5174 * Note: called with local irq disabled, but exits with local irq enabled.
5176 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5179 struct softnet_data *remsd = sd->rps_ipi_list;
5182 sd->rps_ipi_list = NULL;
5186 /* Send pending IPI's to kick RPS processing on remote cpus. */
5187 net_rps_send_ipi(remsd);
5193 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5196 return sd->rps_ipi_list != NULL;
5202 static int process_backlog(struct napi_struct *napi, int quota)
5204 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5208 /* Check if we have pending ipi, its better to send them now,
5209 * not waiting net_rx_action() end.
5211 if (sd_has_rps_ipi_waiting(sd)) {
5212 local_irq_disable();
5213 net_rps_action_and_irq_enable(sd);
5216 napi->weight = dev_rx_weight;
5218 struct sk_buff *skb;
5220 while ((skb = __skb_dequeue(&sd->process_queue))) {
5222 __netif_receive_skb(skb);
5224 input_queue_head_incr(sd);
5225 if (++work >= quota)
5230 local_irq_disable();
5232 if (skb_queue_empty(&sd->input_pkt_queue)) {
5234 * Inline a custom version of __napi_complete().
5235 * only current cpu owns and manipulates this napi,
5236 * and NAPI_STATE_SCHED is the only possible flag set
5238 * We can use a plain write instead of clear_bit(),
5239 * and we dont need an smp_mb() memory barrier.
5244 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5245 &sd->process_queue);
5255 * __napi_schedule - schedule for receive
5256 * @n: entry to schedule
5258 * The entry's receive function will be scheduled to run.
5259 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5261 void __napi_schedule(struct napi_struct *n)
5263 unsigned long flags;
5265 local_irq_save(flags);
5266 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5267 local_irq_restore(flags);
5269 EXPORT_SYMBOL(__napi_schedule);
5272 * napi_schedule_prep - check if napi can be scheduled
5275 * Test if NAPI routine is already running, and if not mark
5276 * it as running. This is used as a condition variable
5277 * insure only one NAPI poll instance runs. We also make
5278 * sure there is no pending NAPI disable.
5280 bool napi_schedule_prep(struct napi_struct *n)
5282 unsigned long val, new;
5285 val = READ_ONCE(n->state);
5286 if (unlikely(val & NAPIF_STATE_DISABLE))
5288 new = val | NAPIF_STATE_SCHED;
5290 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5291 * This was suggested by Alexander Duyck, as compiler
5292 * emits better code than :
5293 * if (val & NAPIF_STATE_SCHED)
5294 * new |= NAPIF_STATE_MISSED;
5296 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5298 } while (cmpxchg(&n->state, val, new) != val);
5300 return !(val & NAPIF_STATE_SCHED);
5302 EXPORT_SYMBOL(napi_schedule_prep);
5305 * __napi_schedule_irqoff - schedule for receive
5306 * @n: entry to schedule
5308 * Variant of __napi_schedule() assuming hard irqs are masked
5310 void __napi_schedule_irqoff(struct napi_struct *n)
5312 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5314 EXPORT_SYMBOL(__napi_schedule_irqoff);
5316 bool napi_complete_done(struct napi_struct *n, int work_done)
5318 unsigned long flags, val, new;
5321 * 1) Don't let napi dequeue from the cpu poll list
5322 * just in case its running on a different cpu.
5323 * 2) If we are busy polling, do nothing here, we have
5324 * the guarantee we will be called later.
5326 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5327 NAPIF_STATE_IN_BUSY_POLL)))
5331 unsigned long timeout = 0;
5334 timeout = n->dev->gro_flush_timeout;
5337 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5338 HRTIMER_MODE_REL_PINNED);
5340 napi_gro_flush(n, false);
5342 if (unlikely(!list_empty(&n->poll_list))) {
5343 /* If n->poll_list is not empty, we need to mask irqs */
5344 local_irq_save(flags);
5345 list_del_init(&n->poll_list);
5346 local_irq_restore(flags);
5350 val = READ_ONCE(n->state);
5352 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5354 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
5356 /* If STATE_MISSED was set, leave STATE_SCHED set,
5357 * because we will call napi->poll() one more time.
5358 * This C code was suggested by Alexander Duyck to help gcc.
5360 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5362 } while (cmpxchg(&n->state, val, new) != val);
5364 if (unlikely(val & NAPIF_STATE_MISSED)) {
5371 EXPORT_SYMBOL(napi_complete_done);
5373 /* must be called under rcu_read_lock(), as we dont take a reference */
5374 static struct napi_struct *napi_by_id(unsigned int napi_id)
5376 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5377 struct napi_struct *napi;
5379 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5380 if (napi->napi_id == napi_id)
5386 #if defined(CONFIG_NET_RX_BUSY_POLL)
5388 #define BUSY_POLL_BUDGET 8
5390 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
5394 /* Busy polling means there is a high chance device driver hard irq
5395 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5396 * set in napi_schedule_prep().
5397 * Since we are about to call napi->poll() once more, we can safely
5398 * clear NAPI_STATE_MISSED.
5400 * Note: x86 could use a single "lock and ..." instruction
5401 * to perform these two clear_bit()
5403 clear_bit(NAPI_STATE_MISSED, &napi->state);
5404 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
5408 /* All we really want here is to re-enable device interrupts.
5409 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5411 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5412 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5413 netpoll_poll_unlock(have_poll_lock);
5414 if (rc == BUSY_POLL_BUDGET)
5415 __napi_schedule(napi);
5419 void napi_busy_loop(unsigned int napi_id,
5420 bool (*loop_end)(void *, unsigned long),
5423 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
5424 int (*napi_poll)(struct napi_struct *napi, int budget);
5425 void *have_poll_lock = NULL;
5426 struct napi_struct *napi;
5433 napi = napi_by_id(napi_id);
5443 unsigned long val = READ_ONCE(napi->state);
5445 /* If multiple threads are competing for this napi,
5446 * we avoid dirtying napi->state as much as we can.
5448 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5449 NAPIF_STATE_IN_BUSY_POLL))
5451 if (cmpxchg(&napi->state, val,
5452 val | NAPIF_STATE_IN_BUSY_POLL |
5453 NAPIF_STATE_SCHED) != val)
5455 have_poll_lock = netpoll_poll_lock(napi);
5456 napi_poll = napi->poll;
5458 work = napi_poll(napi, BUSY_POLL_BUDGET);
5459 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
5462 __NET_ADD_STATS(dev_net(napi->dev),
5463 LINUX_MIB_BUSYPOLLRXPACKETS, work);
5466 if (!loop_end || loop_end(loop_end_arg, start_time))
5469 if (unlikely(need_resched())) {
5471 busy_poll_stop(napi, have_poll_lock);
5475 if (loop_end(loop_end_arg, start_time))
5482 busy_poll_stop(napi, have_poll_lock);
5487 EXPORT_SYMBOL(napi_busy_loop);
5489 #endif /* CONFIG_NET_RX_BUSY_POLL */
5491 static void napi_hash_add(struct napi_struct *napi)
5493 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5494 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5497 spin_lock(&napi_hash_lock);
5499 /* 0..NR_CPUS range is reserved for sender_cpu use */
5501 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
5502 napi_gen_id = MIN_NAPI_ID;
5503 } while (napi_by_id(napi_gen_id));
5504 napi->napi_id = napi_gen_id;
5506 hlist_add_head_rcu(&napi->napi_hash_node,
5507 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5509 spin_unlock(&napi_hash_lock);
5512 /* Warning : caller is responsible to make sure rcu grace period
5513 * is respected before freeing memory containing @napi
5515 bool napi_hash_del(struct napi_struct *napi)
5517 bool rcu_sync_needed = false;
5519 spin_lock(&napi_hash_lock);
5521 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5522 rcu_sync_needed = true;
5523 hlist_del_rcu(&napi->napi_hash_node);
5525 spin_unlock(&napi_hash_lock);
5526 return rcu_sync_needed;
5528 EXPORT_SYMBOL_GPL(napi_hash_del);
5530 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5532 struct napi_struct *napi;
5534 napi = container_of(timer, struct napi_struct, timer);
5536 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5537 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5539 if (napi->gro_list && !napi_disable_pending(napi) &&
5540 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
5541 __napi_schedule_irqoff(napi);
5543 return HRTIMER_NORESTART;
5546 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5547 int (*poll)(struct napi_struct *, int), int weight)
5549 INIT_LIST_HEAD(&napi->poll_list);
5550 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5551 napi->timer.function = napi_watchdog;
5552 napi->gro_count = 0;
5553 napi->gro_list = NULL;
5556 if (weight > NAPI_POLL_WEIGHT)
5557 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5559 napi->weight = weight;
5560 list_add(&napi->dev_list, &dev->napi_list);
5562 #ifdef CONFIG_NETPOLL
5563 napi->poll_owner = -1;
5565 set_bit(NAPI_STATE_SCHED, &napi->state);
5566 napi_hash_add(napi);
5568 EXPORT_SYMBOL(netif_napi_add);
5570 void napi_disable(struct napi_struct *n)
5573 set_bit(NAPI_STATE_DISABLE, &n->state);
5575 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5577 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5580 hrtimer_cancel(&n->timer);
5582 clear_bit(NAPI_STATE_DISABLE, &n->state);
5584 EXPORT_SYMBOL(napi_disable);
5586 /* Must be called in process context */
5587 void netif_napi_del(struct napi_struct *napi)
5590 if (napi_hash_del(napi))
5592 list_del_init(&napi->dev_list);
5593 napi_free_frags(napi);
5595 kfree_skb_list(napi->gro_list);
5596 napi->gro_list = NULL;
5597 napi->gro_count = 0;
5599 EXPORT_SYMBOL(netif_napi_del);
5601 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5606 list_del_init(&n->poll_list);
5608 have = netpoll_poll_lock(n);
5612 /* This NAPI_STATE_SCHED test is for avoiding a race
5613 * with netpoll's poll_napi(). Only the entity which
5614 * obtains the lock and sees NAPI_STATE_SCHED set will
5615 * actually make the ->poll() call. Therefore we avoid
5616 * accidentally calling ->poll() when NAPI is not scheduled.
5619 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5620 work = n->poll(n, weight);
5621 trace_napi_poll(n, work, weight);
5624 WARN_ON_ONCE(work > weight);
5626 if (likely(work < weight))
5629 /* Drivers must not modify the NAPI state if they
5630 * consume the entire weight. In such cases this code
5631 * still "owns" the NAPI instance and therefore can
5632 * move the instance around on the list at-will.
5634 if (unlikely(napi_disable_pending(n))) {
5640 /* flush too old packets
5641 * If HZ < 1000, flush all packets.
5643 napi_gro_flush(n, HZ >= 1000);
5646 /* Some drivers may have called napi_schedule
5647 * prior to exhausting their budget.
5649 if (unlikely(!list_empty(&n->poll_list))) {
5650 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5651 n->dev ? n->dev->name : "backlog");
5655 list_add_tail(&n->poll_list, repoll);
5658 netpoll_poll_unlock(have);
5663 static __latent_entropy void net_rx_action(struct softirq_action *h)
5665 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5666 unsigned long time_limit = jiffies +
5667 usecs_to_jiffies(netdev_budget_usecs);
5668 int budget = netdev_budget;
5672 local_irq_disable();
5673 list_splice_init(&sd->poll_list, &list);
5677 struct napi_struct *n;
5679 if (list_empty(&list)) {
5680 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5685 n = list_first_entry(&list, struct napi_struct, poll_list);
5686 budget -= napi_poll(n, &repoll);
5688 /* If softirq window is exhausted then punt.
5689 * Allow this to run for 2 jiffies since which will allow
5690 * an average latency of 1.5/HZ.
5692 if (unlikely(budget <= 0 ||
5693 time_after_eq(jiffies, time_limit))) {
5699 local_irq_disable();
5701 list_splice_tail_init(&sd->poll_list, &list);
5702 list_splice_tail(&repoll, &list);
5703 list_splice(&list, &sd->poll_list);
5704 if (!list_empty(&sd->poll_list))
5705 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5707 net_rps_action_and_irq_enable(sd);
5709 __kfree_skb_flush();
5712 struct netdev_adjacent {
5713 struct net_device *dev;
5715 /* upper master flag, there can only be one master device per list */
5718 /* counter for the number of times this device was added to us */
5721 /* private field for the users */
5724 struct list_head list;
5725 struct rcu_head rcu;
5728 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5729 struct list_head *adj_list)
5731 struct netdev_adjacent *adj;
5733 list_for_each_entry(adj, adj_list, list) {
5734 if (adj->dev == adj_dev)
5740 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5742 struct net_device *dev = data;
5744 return upper_dev == dev;
5748 * netdev_has_upper_dev - Check if device is linked to an upper device
5750 * @upper_dev: upper device to check
5752 * Find out if a device is linked to specified upper device and return true
5753 * in case it is. Note that this checks only immediate upper device,
5754 * not through a complete stack of devices. The caller must hold the RTNL lock.
5756 bool netdev_has_upper_dev(struct net_device *dev,
5757 struct net_device *upper_dev)
5761 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5764 EXPORT_SYMBOL(netdev_has_upper_dev);
5767 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5769 * @upper_dev: upper device to check
5771 * Find out if a device is linked to specified upper device and return true
5772 * in case it is. Note that this checks the entire upper device chain.
5773 * The caller must hold rcu lock.
5776 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5777 struct net_device *upper_dev)
5779 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5782 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5785 * netdev_has_any_upper_dev - Check if device is linked to some device
5788 * Find out if a device is linked to an upper device and return true in case
5789 * it is. The caller must hold the RTNL lock.
5791 bool netdev_has_any_upper_dev(struct net_device *dev)
5795 return !list_empty(&dev->adj_list.upper);
5797 EXPORT_SYMBOL(netdev_has_any_upper_dev);
5800 * netdev_master_upper_dev_get - Get master upper device
5803 * Find a master upper device and return pointer to it or NULL in case
5804 * it's not there. The caller must hold the RTNL lock.
5806 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5808 struct netdev_adjacent *upper;
5812 if (list_empty(&dev->adj_list.upper))
5815 upper = list_first_entry(&dev->adj_list.upper,
5816 struct netdev_adjacent, list);
5817 if (likely(upper->master))
5821 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5824 * netdev_has_any_lower_dev - Check if device is linked to some device
5827 * Find out if a device is linked to a lower device and return true in case
5828 * it is. The caller must hold the RTNL lock.
5830 static bool netdev_has_any_lower_dev(struct net_device *dev)
5834 return !list_empty(&dev->adj_list.lower);
5837 void *netdev_adjacent_get_private(struct list_head *adj_list)
5839 struct netdev_adjacent *adj;
5841 adj = list_entry(adj_list, struct netdev_adjacent, list);
5843 return adj->private;
5845 EXPORT_SYMBOL(netdev_adjacent_get_private);
5848 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5850 * @iter: list_head ** of the current position
5852 * Gets the next device from the dev's upper list, starting from iter
5853 * position. The caller must hold RCU read lock.
5855 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5856 struct list_head **iter)
5858 struct netdev_adjacent *upper;
5860 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5862 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5864 if (&upper->list == &dev->adj_list.upper)
5867 *iter = &upper->list;
5871 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5873 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5874 struct list_head **iter)
5876 struct netdev_adjacent *upper;
5878 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5880 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5882 if (&upper->list == &dev->adj_list.upper)
5885 *iter = &upper->list;
5890 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5891 int (*fn)(struct net_device *dev,
5895 struct net_device *udev;
5896 struct list_head *iter;
5899 for (iter = &dev->adj_list.upper,
5900 udev = netdev_next_upper_dev_rcu(dev, &iter);
5902 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5903 /* first is the upper device itself */
5904 ret = fn(udev, data);
5908 /* then look at all of its upper devices */
5909 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5916 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5919 * netdev_lower_get_next_private - Get the next ->private from the
5920 * lower neighbour list
5922 * @iter: list_head ** of the current position
5924 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5925 * list, starting from iter position. The caller must hold either hold the
5926 * RTNL lock or its own locking that guarantees that the neighbour lower
5927 * list will remain unchanged.
5929 void *netdev_lower_get_next_private(struct net_device *dev,
5930 struct list_head **iter)
5932 struct netdev_adjacent *lower;
5934 lower = list_entry(*iter, struct netdev_adjacent, list);
5936 if (&lower->list == &dev->adj_list.lower)
5939 *iter = lower->list.next;
5941 return lower->private;
5943 EXPORT_SYMBOL(netdev_lower_get_next_private);
5946 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5947 * lower neighbour list, RCU
5950 * @iter: list_head ** of the current position
5952 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5953 * list, starting from iter position. The caller must hold RCU read lock.
5955 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5956 struct list_head **iter)
5958 struct netdev_adjacent *lower;
5960 WARN_ON_ONCE(!rcu_read_lock_held());
5962 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5964 if (&lower->list == &dev->adj_list.lower)
5967 *iter = &lower->list;
5969 return lower->private;
5971 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5974 * netdev_lower_get_next - Get the next device from the lower neighbour
5977 * @iter: list_head ** of the current position
5979 * Gets the next netdev_adjacent from the dev's lower neighbour
5980 * list, starting from iter position. The caller must hold RTNL lock or
5981 * its own locking that guarantees that the neighbour lower
5982 * list will remain unchanged.
5984 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5986 struct netdev_adjacent *lower;
5988 lower = list_entry(*iter, struct netdev_adjacent, list);
5990 if (&lower->list == &dev->adj_list.lower)
5993 *iter = lower->list.next;
5997 EXPORT_SYMBOL(netdev_lower_get_next);
5999 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6000 struct list_head **iter)
6002 struct netdev_adjacent *lower;
6004 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6006 if (&lower->list == &dev->adj_list.lower)
6009 *iter = &lower->list;
6014 int netdev_walk_all_lower_dev(struct net_device *dev,
6015 int (*fn)(struct net_device *dev,
6019 struct net_device *ldev;
6020 struct list_head *iter;
6023 for (iter = &dev->adj_list.lower,
6024 ldev = netdev_next_lower_dev(dev, &iter);
6026 ldev = netdev_next_lower_dev(dev, &iter)) {
6027 /* first is the lower device itself */
6028 ret = fn(ldev, data);
6032 /* then look at all of its lower devices */
6033 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6040 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6042 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6043 struct list_head **iter)
6045 struct netdev_adjacent *lower;
6047 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6048 if (&lower->list == &dev->adj_list.lower)
6051 *iter = &lower->list;
6056 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6057 int (*fn)(struct net_device *dev,
6061 struct net_device *ldev;
6062 struct list_head *iter;
6065 for (iter = &dev->adj_list.lower,
6066 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6068 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6069 /* first is the lower device itself */
6070 ret = fn(ldev, data);
6074 /* then look at all of its lower devices */
6075 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6082 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6085 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6086 * lower neighbour list, RCU
6090 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6091 * list. The caller must hold RCU read lock.
6093 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6095 struct netdev_adjacent *lower;
6097 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6098 struct netdev_adjacent, list);
6100 return lower->private;
6103 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6106 * netdev_master_upper_dev_get_rcu - Get master upper device
6109 * Find a master upper device and return pointer to it or NULL in case
6110 * it's not there. The caller must hold the RCU read lock.
6112 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6114 struct netdev_adjacent *upper;
6116 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6117 struct netdev_adjacent, list);
6118 if (upper && likely(upper->master))
6122 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6124 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6125 struct net_device *adj_dev,
6126 struct list_head *dev_list)
6128 char linkname[IFNAMSIZ+7];
6130 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6131 "upper_%s" : "lower_%s", adj_dev->name);
6132 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6135 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6137 struct list_head *dev_list)
6139 char linkname[IFNAMSIZ+7];
6141 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6142 "upper_%s" : "lower_%s", name);
6143 sysfs_remove_link(&(dev->dev.kobj), linkname);
6146 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6147 struct net_device *adj_dev,
6148 struct list_head *dev_list)
6150 return (dev_list == &dev->adj_list.upper ||
6151 dev_list == &dev->adj_list.lower) &&
6152 net_eq(dev_net(dev), dev_net(adj_dev));
6155 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6156 struct net_device *adj_dev,
6157 struct list_head *dev_list,
6158 void *private, bool master)
6160 struct netdev_adjacent *adj;
6163 adj = __netdev_find_adj(adj_dev, dev_list);
6167 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6168 dev->name, adj_dev->name, adj->ref_nr);
6173 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6178 adj->master = master;
6180 adj->private = private;
6183 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6184 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6186 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6187 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6192 /* Ensure that master link is always the first item in list. */
6194 ret = sysfs_create_link(&(dev->dev.kobj),
6195 &(adj_dev->dev.kobj), "master");
6197 goto remove_symlinks;
6199 list_add_rcu(&adj->list, dev_list);
6201 list_add_tail_rcu(&adj->list, dev_list);
6207 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6208 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6216 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6217 struct net_device *adj_dev,
6219 struct list_head *dev_list)
6221 struct netdev_adjacent *adj;
6223 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6224 dev->name, adj_dev->name, ref_nr);
6226 adj = __netdev_find_adj(adj_dev, dev_list);
6229 pr_err("Adjacency does not exist for device %s from %s\n",
6230 dev->name, adj_dev->name);
6235 if (adj->ref_nr > ref_nr) {
6236 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6237 dev->name, adj_dev->name, ref_nr,
6238 adj->ref_nr - ref_nr);
6239 adj->ref_nr -= ref_nr;
6244 sysfs_remove_link(&(dev->dev.kobj), "master");
6246 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6247 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6249 list_del_rcu(&adj->list);
6250 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6251 adj_dev->name, dev->name, adj_dev->name);
6253 kfree_rcu(adj, rcu);
6256 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6257 struct net_device *upper_dev,
6258 struct list_head *up_list,
6259 struct list_head *down_list,
6260 void *private, bool master)
6264 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6269 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6272 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6279 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6280 struct net_device *upper_dev,
6282 struct list_head *up_list,
6283 struct list_head *down_list)
6285 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6286 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6289 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6290 struct net_device *upper_dev,
6291 void *private, bool master)
6293 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6294 &dev->adj_list.upper,
6295 &upper_dev->adj_list.lower,
6299 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6300 struct net_device *upper_dev)
6302 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
6303 &dev->adj_list.upper,
6304 &upper_dev->adj_list.lower);
6307 static int __netdev_upper_dev_link(struct net_device *dev,
6308 struct net_device *upper_dev, bool master,
6309 void *upper_priv, void *upper_info,
6310 struct netlink_ext_ack *extack)
6312 struct netdev_notifier_changeupper_info changeupper_info = {
6317 .upper_dev = upper_dev,
6320 .upper_info = upper_info,
6326 if (dev == upper_dev)
6329 /* To prevent loops, check if dev is not upper device to upper_dev. */
6330 if (netdev_has_upper_dev(upper_dev, dev))
6333 if (netdev_has_upper_dev(dev, upper_dev))
6336 if (master && netdev_master_upper_dev_get(dev))
6339 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
6340 &changeupper_info.info);
6341 ret = notifier_to_errno(ret);
6345 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
6350 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
6351 &changeupper_info.info);
6352 ret = notifier_to_errno(ret);
6359 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6365 * netdev_upper_dev_link - Add a link to the upper device
6367 * @upper_dev: new upper device
6369 * Adds a link to device which is upper to this one. The caller must hold
6370 * the RTNL lock. On a failure a negative errno code is returned.
6371 * On success the reference counts are adjusted and the function
6374 int netdev_upper_dev_link(struct net_device *dev,
6375 struct net_device *upper_dev,
6376 struct netlink_ext_ack *extack)
6378 return __netdev_upper_dev_link(dev, upper_dev, false,
6379 NULL, NULL, extack);
6381 EXPORT_SYMBOL(netdev_upper_dev_link);
6384 * netdev_master_upper_dev_link - Add a master link to the upper device
6386 * @upper_dev: new upper device
6387 * @upper_priv: upper device private
6388 * @upper_info: upper info to be passed down via notifier
6390 * Adds a link to device which is upper to this one. In this case, only
6391 * one master upper device can be linked, although other non-master devices
6392 * might be linked as well. The caller must hold the RTNL lock.
6393 * On a failure a negative errno code is returned. On success the reference
6394 * counts are adjusted and the function returns zero.
6396 int netdev_master_upper_dev_link(struct net_device *dev,
6397 struct net_device *upper_dev,
6398 void *upper_priv, void *upper_info,
6399 struct netlink_ext_ack *extack)
6401 return __netdev_upper_dev_link(dev, upper_dev, true,
6402 upper_priv, upper_info, extack);
6404 EXPORT_SYMBOL(netdev_master_upper_dev_link);
6407 * netdev_upper_dev_unlink - Removes a link to upper device
6409 * @upper_dev: new upper device
6411 * Removes a link to device which is upper to this one. The caller must hold
6414 void netdev_upper_dev_unlink(struct net_device *dev,
6415 struct net_device *upper_dev)
6417 struct netdev_notifier_changeupper_info changeupper_info = {
6421 .upper_dev = upper_dev,
6427 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6429 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
6430 &changeupper_info.info);
6432 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6434 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
6435 &changeupper_info.info);
6437 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6440 * netdev_bonding_info_change - Dispatch event about slave change
6442 * @bonding_info: info to dispatch
6444 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6445 * The caller must hold the RTNL lock.
6447 void netdev_bonding_info_change(struct net_device *dev,
6448 struct netdev_bonding_info *bonding_info)
6450 struct netdev_notifier_bonding_info info = {
6454 memcpy(&info.bonding_info, bonding_info,
6455 sizeof(struct netdev_bonding_info));
6456 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
6459 EXPORT_SYMBOL(netdev_bonding_info_change);
6461 static void netdev_adjacent_add_links(struct net_device *dev)
6463 struct netdev_adjacent *iter;
6465 struct net *net = dev_net(dev);
6467 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6468 if (!net_eq(net, dev_net(iter->dev)))
6470 netdev_adjacent_sysfs_add(iter->dev, dev,
6471 &iter->dev->adj_list.lower);
6472 netdev_adjacent_sysfs_add(dev, iter->dev,
6473 &dev->adj_list.upper);
6476 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6477 if (!net_eq(net, dev_net(iter->dev)))
6479 netdev_adjacent_sysfs_add(iter->dev, dev,
6480 &iter->dev->adj_list.upper);
6481 netdev_adjacent_sysfs_add(dev, iter->dev,
6482 &dev->adj_list.lower);
6486 static void netdev_adjacent_del_links(struct net_device *dev)
6488 struct netdev_adjacent *iter;
6490 struct net *net = dev_net(dev);
6492 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6493 if (!net_eq(net, dev_net(iter->dev)))
6495 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6496 &iter->dev->adj_list.lower);
6497 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6498 &dev->adj_list.upper);
6501 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6502 if (!net_eq(net, dev_net(iter->dev)))
6504 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6505 &iter->dev->adj_list.upper);
6506 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6507 &dev->adj_list.lower);
6511 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6513 struct netdev_adjacent *iter;
6515 struct net *net = dev_net(dev);
6517 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6518 if (!net_eq(net, dev_net(iter->dev)))
6520 netdev_adjacent_sysfs_del(iter->dev, oldname,
6521 &iter->dev->adj_list.lower);
6522 netdev_adjacent_sysfs_add(iter->dev, dev,
6523 &iter->dev->adj_list.lower);
6526 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6527 if (!net_eq(net, dev_net(iter->dev)))
6529 netdev_adjacent_sysfs_del(iter->dev, oldname,
6530 &iter->dev->adj_list.upper);
6531 netdev_adjacent_sysfs_add(iter->dev, dev,
6532 &iter->dev->adj_list.upper);
6536 void *netdev_lower_dev_get_private(struct net_device *dev,
6537 struct net_device *lower_dev)
6539 struct netdev_adjacent *lower;
6543 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6547 return lower->private;
6549 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6552 int dev_get_nest_level(struct net_device *dev)
6554 struct net_device *lower = NULL;
6555 struct list_head *iter;
6561 netdev_for_each_lower_dev(dev, lower, iter) {
6562 nest = dev_get_nest_level(lower);
6563 if (max_nest < nest)
6567 return max_nest + 1;
6569 EXPORT_SYMBOL(dev_get_nest_level);
6572 * netdev_lower_change - Dispatch event about lower device state change
6573 * @lower_dev: device
6574 * @lower_state_info: state to dispatch
6576 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6577 * The caller must hold the RTNL lock.
6579 void netdev_lower_state_changed(struct net_device *lower_dev,
6580 void *lower_state_info)
6582 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
6583 .info.dev = lower_dev,
6587 changelowerstate_info.lower_state_info = lower_state_info;
6588 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
6589 &changelowerstate_info.info);
6591 EXPORT_SYMBOL(netdev_lower_state_changed);
6593 static void dev_change_rx_flags(struct net_device *dev, int flags)
6595 const struct net_device_ops *ops = dev->netdev_ops;
6597 if (ops->ndo_change_rx_flags)
6598 ops->ndo_change_rx_flags(dev, flags);
6601 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6603 unsigned int old_flags = dev->flags;
6609 dev->flags |= IFF_PROMISC;
6610 dev->promiscuity += inc;
6611 if (dev->promiscuity == 0) {
6614 * If inc causes overflow, untouch promisc and return error.
6617 dev->flags &= ~IFF_PROMISC;
6619 dev->promiscuity -= inc;
6620 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6625 if (dev->flags != old_flags) {
6626 pr_info("device %s %s promiscuous mode\n",
6628 dev->flags & IFF_PROMISC ? "entered" : "left");
6629 if (audit_enabled) {
6630 current_uid_gid(&uid, &gid);
6631 audit_log(current->audit_context, GFP_ATOMIC,
6632 AUDIT_ANOM_PROMISCUOUS,
6633 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6634 dev->name, (dev->flags & IFF_PROMISC),
6635 (old_flags & IFF_PROMISC),
6636 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6637 from_kuid(&init_user_ns, uid),
6638 from_kgid(&init_user_ns, gid),
6639 audit_get_sessionid(current));
6642 dev_change_rx_flags(dev, IFF_PROMISC);
6645 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6650 * dev_set_promiscuity - update promiscuity count on a device
6654 * Add or remove promiscuity from a device. While the count in the device
6655 * remains above zero the interface remains promiscuous. Once it hits zero
6656 * the device reverts back to normal filtering operation. A negative inc
6657 * value is used to drop promiscuity on the device.
6658 * Return 0 if successful or a negative errno code on error.
6660 int dev_set_promiscuity(struct net_device *dev, int inc)
6662 unsigned int old_flags = dev->flags;
6665 err = __dev_set_promiscuity(dev, inc, true);
6668 if (dev->flags != old_flags)
6669 dev_set_rx_mode(dev);
6672 EXPORT_SYMBOL(dev_set_promiscuity);
6674 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6676 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6680 dev->flags |= IFF_ALLMULTI;
6681 dev->allmulti += inc;
6682 if (dev->allmulti == 0) {
6685 * If inc causes overflow, untouch allmulti and return error.
6688 dev->flags &= ~IFF_ALLMULTI;
6690 dev->allmulti -= inc;
6691 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6696 if (dev->flags ^ old_flags) {
6697 dev_change_rx_flags(dev, IFF_ALLMULTI);
6698 dev_set_rx_mode(dev);
6700 __dev_notify_flags(dev, old_flags,
6701 dev->gflags ^ old_gflags);
6707 * dev_set_allmulti - update allmulti count on a device
6711 * Add or remove reception of all multicast frames to a device. While the
6712 * count in the device remains above zero the interface remains listening
6713 * to all interfaces. Once it hits zero the device reverts back to normal
6714 * filtering operation. A negative @inc value is used to drop the counter
6715 * when releasing a resource needing all multicasts.
6716 * Return 0 if successful or a negative errno code on error.
6719 int dev_set_allmulti(struct net_device *dev, int inc)
6721 return __dev_set_allmulti(dev, inc, true);
6723 EXPORT_SYMBOL(dev_set_allmulti);
6726 * Upload unicast and multicast address lists to device and
6727 * configure RX filtering. When the device doesn't support unicast
6728 * filtering it is put in promiscuous mode while unicast addresses
6731 void __dev_set_rx_mode(struct net_device *dev)
6733 const struct net_device_ops *ops = dev->netdev_ops;
6735 /* dev_open will call this function so the list will stay sane. */
6736 if (!(dev->flags&IFF_UP))
6739 if (!netif_device_present(dev))
6742 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6743 /* Unicast addresses changes may only happen under the rtnl,
6744 * therefore calling __dev_set_promiscuity here is safe.
6746 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6747 __dev_set_promiscuity(dev, 1, false);
6748 dev->uc_promisc = true;
6749 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6750 __dev_set_promiscuity(dev, -1, false);
6751 dev->uc_promisc = false;
6755 if (ops->ndo_set_rx_mode)
6756 ops->ndo_set_rx_mode(dev);
6759 void dev_set_rx_mode(struct net_device *dev)
6761 netif_addr_lock_bh(dev);
6762 __dev_set_rx_mode(dev);
6763 netif_addr_unlock_bh(dev);
6767 * dev_get_flags - get flags reported to userspace
6770 * Get the combination of flag bits exported through APIs to userspace.
6772 unsigned int dev_get_flags(const struct net_device *dev)
6776 flags = (dev->flags & ~(IFF_PROMISC |
6781 (dev->gflags & (IFF_PROMISC |
6784 if (netif_running(dev)) {
6785 if (netif_oper_up(dev))
6786 flags |= IFF_RUNNING;
6787 if (netif_carrier_ok(dev))
6788 flags |= IFF_LOWER_UP;
6789 if (netif_dormant(dev))
6790 flags |= IFF_DORMANT;
6795 EXPORT_SYMBOL(dev_get_flags);
6797 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6799 unsigned int old_flags = dev->flags;
6805 * Set the flags on our device.
6808 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6809 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6811 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6815 * Load in the correct multicast list now the flags have changed.
6818 if ((old_flags ^ flags) & IFF_MULTICAST)
6819 dev_change_rx_flags(dev, IFF_MULTICAST);
6821 dev_set_rx_mode(dev);
6824 * Have we downed the interface. We handle IFF_UP ourselves
6825 * according to user attempts to set it, rather than blindly
6830 if ((old_flags ^ flags) & IFF_UP) {
6831 if (old_flags & IFF_UP)
6834 ret = __dev_open(dev);
6837 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6838 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6839 unsigned int old_flags = dev->flags;
6841 dev->gflags ^= IFF_PROMISC;
6843 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6844 if (dev->flags != old_flags)
6845 dev_set_rx_mode(dev);
6848 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6849 * is important. Some (broken) drivers set IFF_PROMISC, when
6850 * IFF_ALLMULTI is requested not asking us and not reporting.
6852 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6853 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6855 dev->gflags ^= IFF_ALLMULTI;
6856 __dev_set_allmulti(dev, inc, false);
6862 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6863 unsigned int gchanges)
6865 unsigned int changes = dev->flags ^ old_flags;
6868 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6870 if (changes & IFF_UP) {
6871 if (dev->flags & IFF_UP)
6872 call_netdevice_notifiers(NETDEV_UP, dev);
6874 call_netdevice_notifiers(NETDEV_DOWN, dev);
6877 if (dev->flags & IFF_UP &&
6878 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6879 struct netdev_notifier_change_info change_info = {
6883 .flags_changed = changes,
6886 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
6891 * dev_change_flags - change device settings
6893 * @flags: device state flags
6895 * Change settings on device based state flags. The flags are
6896 * in the userspace exported format.
6898 int dev_change_flags(struct net_device *dev, unsigned int flags)
6901 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6903 ret = __dev_change_flags(dev, flags);
6907 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6908 __dev_notify_flags(dev, old_flags, changes);
6911 EXPORT_SYMBOL(dev_change_flags);
6913 int __dev_set_mtu(struct net_device *dev, int new_mtu)
6915 const struct net_device_ops *ops = dev->netdev_ops;
6917 if (ops->ndo_change_mtu)
6918 return ops->ndo_change_mtu(dev, new_mtu);
6923 EXPORT_SYMBOL(__dev_set_mtu);
6926 * dev_set_mtu - Change maximum transfer unit
6928 * @new_mtu: new transfer unit
6930 * Change the maximum transfer size of the network device.
6932 int dev_set_mtu(struct net_device *dev, int new_mtu)
6936 if (new_mtu == dev->mtu)
6939 /* MTU must be positive, and in range */
6940 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6941 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6942 dev->name, new_mtu, dev->min_mtu);
6946 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6947 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6948 dev->name, new_mtu, dev->max_mtu);
6952 if (!netif_device_present(dev))
6955 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6956 err = notifier_to_errno(err);
6960 orig_mtu = dev->mtu;
6961 err = __dev_set_mtu(dev, new_mtu);
6964 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6965 err = notifier_to_errno(err);
6967 /* setting mtu back and notifying everyone again,
6968 * so that they have a chance to revert changes.
6970 __dev_set_mtu(dev, orig_mtu);
6971 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6976 EXPORT_SYMBOL(dev_set_mtu);
6979 * dev_set_group - Change group this device belongs to
6981 * @new_group: group this device should belong to
6983 void dev_set_group(struct net_device *dev, int new_group)
6985 dev->group = new_group;
6987 EXPORT_SYMBOL(dev_set_group);
6990 * dev_set_mac_address - Change Media Access Control Address
6994 * Change the hardware (MAC) address of the device
6996 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6998 const struct net_device_ops *ops = dev->netdev_ops;
7001 if (!ops->ndo_set_mac_address)
7003 if (sa->sa_family != dev->type)
7005 if (!netif_device_present(dev))
7007 err = ops->ndo_set_mac_address(dev, sa);
7010 dev->addr_assign_type = NET_ADDR_SET;
7011 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7012 add_device_randomness(dev->dev_addr, dev->addr_len);
7015 EXPORT_SYMBOL(dev_set_mac_address);
7018 * dev_change_carrier - Change device carrier
7020 * @new_carrier: new value
7022 * Change device carrier
7024 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7026 const struct net_device_ops *ops = dev->netdev_ops;
7028 if (!ops->ndo_change_carrier)
7030 if (!netif_device_present(dev))
7032 return ops->ndo_change_carrier(dev, new_carrier);
7034 EXPORT_SYMBOL(dev_change_carrier);
7037 * dev_get_phys_port_id - Get device physical port ID
7041 * Get device physical port ID
7043 int dev_get_phys_port_id(struct net_device *dev,
7044 struct netdev_phys_item_id *ppid)
7046 const struct net_device_ops *ops = dev->netdev_ops;
7048 if (!ops->ndo_get_phys_port_id)
7050 return ops->ndo_get_phys_port_id(dev, ppid);
7052 EXPORT_SYMBOL(dev_get_phys_port_id);
7055 * dev_get_phys_port_name - Get device physical port name
7058 * @len: limit of bytes to copy to name
7060 * Get device physical port name
7062 int dev_get_phys_port_name(struct net_device *dev,
7063 char *name, size_t len)
7065 const struct net_device_ops *ops = dev->netdev_ops;
7067 if (!ops->ndo_get_phys_port_name)
7069 return ops->ndo_get_phys_port_name(dev, name, len);
7071 EXPORT_SYMBOL(dev_get_phys_port_name);
7074 * dev_change_proto_down - update protocol port state information
7076 * @proto_down: new value
7078 * This info can be used by switch drivers to set the phys state of the
7081 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7083 const struct net_device_ops *ops = dev->netdev_ops;
7085 if (!ops->ndo_change_proto_down)
7087 if (!netif_device_present(dev))
7089 return ops->ndo_change_proto_down(dev, proto_down);
7091 EXPORT_SYMBOL(dev_change_proto_down);
7093 u8 __dev_xdp_attached(struct net_device *dev, bpf_op_t bpf_op, u32 *prog_id)
7095 struct netdev_bpf xdp;
7097 memset(&xdp, 0, sizeof(xdp));
7098 xdp.command = XDP_QUERY_PROG;
7100 /* Query must always succeed. */
7101 WARN_ON(bpf_op(dev, &xdp) < 0);
7103 *prog_id = xdp.prog_id;
7105 return xdp.prog_attached;
7108 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
7109 struct netlink_ext_ack *extack, u32 flags,
7110 struct bpf_prog *prog)
7112 struct netdev_bpf xdp;
7114 memset(&xdp, 0, sizeof(xdp));
7115 if (flags & XDP_FLAGS_HW_MODE)
7116 xdp.command = XDP_SETUP_PROG_HW;
7118 xdp.command = XDP_SETUP_PROG;
7119 xdp.extack = extack;
7123 return bpf_op(dev, &xdp);
7127 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
7129 * @extack: netlink extended ack
7130 * @fd: new program fd or negative value to clear
7131 * @flags: xdp-related flags
7133 * Set or clear a bpf program for a device
7135 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
7138 const struct net_device_ops *ops = dev->netdev_ops;
7139 struct bpf_prog *prog = NULL;
7140 bpf_op_t bpf_op, bpf_chk;
7145 bpf_op = bpf_chk = ops->ndo_bpf;
7146 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7148 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
7149 bpf_op = generic_xdp_install;
7150 if (bpf_op == bpf_chk)
7151 bpf_chk = generic_xdp_install;
7154 if (bpf_chk && __dev_xdp_attached(dev, bpf_chk, NULL))
7156 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7157 __dev_xdp_attached(dev, bpf_op, NULL))
7160 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
7162 return PTR_ERR(prog);
7165 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
7166 if (err < 0 && prog)
7173 * dev_new_index - allocate an ifindex
7174 * @net: the applicable net namespace
7176 * Returns a suitable unique value for a new device interface
7177 * number. The caller must hold the rtnl semaphore or the
7178 * dev_base_lock to be sure it remains unique.
7180 static int dev_new_index(struct net *net)
7182 int ifindex = net->ifindex;
7187 if (!__dev_get_by_index(net, ifindex))
7188 return net->ifindex = ifindex;
7192 /* Delayed registration/unregisteration */
7193 static LIST_HEAD(net_todo_list);
7194 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7196 static void net_set_todo(struct net_device *dev)
7198 list_add_tail(&dev->todo_list, &net_todo_list);
7199 dev_net(dev)->dev_unreg_count++;
7202 static void rollback_registered_many(struct list_head *head)
7204 struct net_device *dev, *tmp;
7205 LIST_HEAD(close_head);
7207 BUG_ON(dev_boot_phase);
7210 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
7211 /* Some devices call without registering
7212 * for initialization unwind. Remove those
7213 * devices and proceed with the remaining.
7215 if (dev->reg_state == NETREG_UNINITIALIZED) {
7216 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7220 list_del(&dev->unreg_list);
7223 dev->dismantle = true;
7224 BUG_ON(dev->reg_state != NETREG_REGISTERED);
7227 /* If device is running, close it first. */
7228 list_for_each_entry(dev, head, unreg_list)
7229 list_add_tail(&dev->close_list, &close_head);
7230 dev_close_many(&close_head, true);
7232 list_for_each_entry(dev, head, unreg_list) {
7233 /* And unlink it from device chain. */
7234 unlist_netdevice(dev);
7236 dev->reg_state = NETREG_UNREGISTERING;
7238 flush_all_backlogs();
7242 list_for_each_entry(dev, head, unreg_list) {
7243 struct sk_buff *skb = NULL;
7245 /* Shutdown queueing discipline. */
7249 /* Notify protocols, that we are about to destroy
7250 * this device. They should clean all the things.
7252 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7254 if (!dev->rtnl_link_ops ||
7255 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7256 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
7260 * Flush the unicast and multicast chains
7265 if (dev->netdev_ops->ndo_uninit)
7266 dev->netdev_ops->ndo_uninit(dev);
7269 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
7271 /* Notifier chain MUST detach us all upper devices. */
7272 WARN_ON(netdev_has_any_upper_dev(dev));
7273 WARN_ON(netdev_has_any_lower_dev(dev));
7275 /* Remove entries from kobject tree */
7276 netdev_unregister_kobject(dev);
7278 /* Remove XPS queueing entries */
7279 netif_reset_xps_queues_gt(dev, 0);
7285 list_for_each_entry(dev, head, unreg_list)
7289 static void rollback_registered(struct net_device *dev)
7293 list_add(&dev->unreg_list, &single);
7294 rollback_registered_many(&single);
7298 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
7299 struct net_device *upper, netdev_features_t features)
7301 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7302 netdev_features_t feature;
7305 for_each_netdev_feature(&upper_disables, feature_bit) {
7306 feature = __NETIF_F_BIT(feature_bit);
7307 if (!(upper->wanted_features & feature)
7308 && (features & feature)) {
7309 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
7310 &feature, upper->name);
7311 features &= ~feature;
7318 static void netdev_sync_lower_features(struct net_device *upper,
7319 struct net_device *lower, netdev_features_t features)
7321 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7322 netdev_features_t feature;
7325 for_each_netdev_feature(&upper_disables, feature_bit) {
7326 feature = __NETIF_F_BIT(feature_bit);
7327 if (!(features & feature) && (lower->features & feature)) {
7328 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
7329 &feature, lower->name);
7330 lower->wanted_features &= ~feature;
7331 netdev_update_features(lower);
7333 if (unlikely(lower->features & feature))
7334 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
7335 &feature, lower->name);
7340 static netdev_features_t netdev_fix_features(struct net_device *dev,
7341 netdev_features_t features)
7343 /* Fix illegal checksum combinations */
7344 if ((features & NETIF_F_HW_CSUM) &&
7345 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
7346 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
7347 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
7350 /* TSO requires that SG is present as well. */
7351 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
7352 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
7353 features &= ~NETIF_F_ALL_TSO;
7356 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
7357 !(features & NETIF_F_IP_CSUM)) {
7358 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
7359 features &= ~NETIF_F_TSO;
7360 features &= ~NETIF_F_TSO_ECN;
7363 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
7364 !(features & NETIF_F_IPV6_CSUM)) {
7365 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
7366 features &= ~NETIF_F_TSO6;
7369 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
7370 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
7371 features &= ~NETIF_F_TSO_MANGLEID;
7373 /* TSO ECN requires that TSO is present as well. */
7374 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
7375 features &= ~NETIF_F_TSO_ECN;
7377 /* Software GSO depends on SG. */
7378 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
7379 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
7380 features &= ~NETIF_F_GSO;
7383 /* GSO partial features require GSO partial be set */
7384 if ((features & dev->gso_partial_features) &&
7385 !(features & NETIF_F_GSO_PARTIAL)) {
7387 "Dropping partially supported GSO features since no GSO partial.\n");
7388 features &= ~dev->gso_partial_features;
7394 int __netdev_update_features(struct net_device *dev)
7396 struct net_device *upper, *lower;
7397 netdev_features_t features;
7398 struct list_head *iter;
7403 features = netdev_get_wanted_features(dev);
7405 if (dev->netdev_ops->ndo_fix_features)
7406 features = dev->netdev_ops->ndo_fix_features(dev, features);
7408 /* driver might be less strict about feature dependencies */
7409 features = netdev_fix_features(dev, features);
7411 /* some features can't be enabled if they're off an an upper device */
7412 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7413 features = netdev_sync_upper_features(dev, upper, features);
7415 if (dev->features == features)
7418 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7419 &dev->features, &features);
7421 if (dev->netdev_ops->ndo_set_features)
7422 err = dev->netdev_ops->ndo_set_features(dev, features);
7426 if (unlikely(err < 0)) {
7428 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7429 err, &features, &dev->features);
7430 /* return non-0 since some features might have changed and
7431 * it's better to fire a spurious notification than miss it
7437 /* some features must be disabled on lower devices when disabled
7438 * on an upper device (think: bonding master or bridge)
7440 netdev_for_each_lower_dev(dev, lower, iter)
7441 netdev_sync_lower_features(dev, lower, features);
7444 netdev_features_t diff = features ^ dev->features;
7446 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
7447 /* udp_tunnel_{get,drop}_rx_info both need
7448 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
7449 * device, or they won't do anything.
7450 * Thus we need to update dev->features
7451 * *before* calling udp_tunnel_get_rx_info,
7452 * but *after* calling udp_tunnel_drop_rx_info.
7454 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
7455 dev->features = features;
7456 udp_tunnel_get_rx_info(dev);
7458 udp_tunnel_drop_rx_info(dev);
7462 dev->features = features;
7465 return err < 0 ? 0 : 1;
7469 * netdev_update_features - recalculate device features
7470 * @dev: the device to check
7472 * Recalculate dev->features set and send notifications if it
7473 * has changed. Should be called after driver or hardware dependent
7474 * conditions might have changed that influence the features.
7476 void netdev_update_features(struct net_device *dev)
7478 if (__netdev_update_features(dev))
7479 netdev_features_change(dev);
7481 EXPORT_SYMBOL(netdev_update_features);
7484 * netdev_change_features - recalculate device features
7485 * @dev: the device to check
7487 * Recalculate dev->features set and send notifications even
7488 * if they have not changed. Should be called instead of
7489 * netdev_update_features() if also dev->vlan_features might
7490 * have changed to allow the changes to be propagated to stacked
7493 void netdev_change_features(struct net_device *dev)
7495 __netdev_update_features(dev);
7496 netdev_features_change(dev);
7498 EXPORT_SYMBOL(netdev_change_features);
7501 * netif_stacked_transfer_operstate - transfer operstate
7502 * @rootdev: the root or lower level device to transfer state from
7503 * @dev: the device to transfer operstate to
7505 * Transfer operational state from root to device. This is normally
7506 * called when a stacking relationship exists between the root
7507 * device and the device(a leaf device).
7509 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7510 struct net_device *dev)
7512 if (rootdev->operstate == IF_OPER_DORMANT)
7513 netif_dormant_on(dev);
7515 netif_dormant_off(dev);
7517 if (netif_carrier_ok(rootdev))
7518 netif_carrier_on(dev);
7520 netif_carrier_off(dev);
7522 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7525 static int netif_alloc_rx_queues(struct net_device *dev)
7527 unsigned int i, count = dev->num_rx_queues;
7528 struct netdev_rx_queue *rx;
7529 size_t sz = count * sizeof(*rx);
7533 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7539 for (i = 0; i < count; i++)
7545 static void netdev_init_one_queue(struct net_device *dev,
7546 struct netdev_queue *queue, void *_unused)
7548 /* Initialize queue lock */
7549 spin_lock_init(&queue->_xmit_lock);
7550 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7551 queue->xmit_lock_owner = -1;
7552 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7555 dql_init(&queue->dql, HZ);
7559 static void netif_free_tx_queues(struct net_device *dev)
7564 static int netif_alloc_netdev_queues(struct net_device *dev)
7566 unsigned int count = dev->num_tx_queues;
7567 struct netdev_queue *tx;
7568 size_t sz = count * sizeof(*tx);
7570 if (count < 1 || count > 0xffff)
7573 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7579 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7580 spin_lock_init(&dev->tx_global_lock);
7585 void netif_tx_stop_all_queues(struct net_device *dev)
7589 for (i = 0; i < dev->num_tx_queues; i++) {
7590 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7592 netif_tx_stop_queue(txq);
7595 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7598 * register_netdevice - register a network device
7599 * @dev: device to register
7601 * Take a completed network device structure and add it to the kernel
7602 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7603 * chain. 0 is returned on success. A negative errno code is returned
7604 * on a failure to set up the device, or if the name is a duplicate.
7606 * Callers must hold the rtnl semaphore. You may want
7607 * register_netdev() instead of this.
7610 * The locking appears insufficient to guarantee two parallel registers
7611 * will not get the same name.
7614 int register_netdevice(struct net_device *dev)
7617 struct net *net = dev_net(dev);
7619 BUG_ON(dev_boot_phase);
7624 /* When net_device's are persistent, this will be fatal. */
7625 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7628 spin_lock_init(&dev->addr_list_lock);
7629 netdev_set_addr_lockdep_class(dev);
7631 ret = dev_get_valid_name(net, dev, dev->name);
7635 /* Init, if this function is available */
7636 if (dev->netdev_ops->ndo_init) {
7637 ret = dev->netdev_ops->ndo_init(dev);
7645 if (((dev->hw_features | dev->features) &
7646 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7647 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7648 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7649 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7656 dev->ifindex = dev_new_index(net);
7657 else if (__dev_get_by_index(net, dev->ifindex))
7660 /* Transfer changeable features to wanted_features and enable
7661 * software offloads (GSO and GRO).
7663 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7664 dev->features |= NETIF_F_SOFT_FEATURES;
7666 if (dev->netdev_ops->ndo_udp_tunnel_add) {
7667 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7668 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7671 dev->wanted_features = dev->features & dev->hw_features;
7673 if (!(dev->flags & IFF_LOOPBACK))
7674 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7676 /* If IPv4 TCP segmentation offload is supported we should also
7677 * allow the device to enable segmenting the frame with the option
7678 * of ignoring a static IP ID value. This doesn't enable the
7679 * feature itself but allows the user to enable it later.
7681 if (dev->hw_features & NETIF_F_TSO)
7682 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7683 if (dev->vlan_features & NETIF_F_TSO)
7684 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7685 if (dev->mpls_features & NETIF_F_TSO)
7686 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7687 if (dev->hw_enc_features & NETIF_F_TSO)
7688 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7690 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7692 dev->vlan_features |= NETIF_F_HIGHDMA;
7694 /* Make NETIF_F_SG inheritable to tunnel devices.
7696 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7698 /* Make NETIF_F_SG inheritable to MPLS.
7700 dev->mpls_features |= NETIF_F_SG;
7702 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7703 ret = notifier_to_errno(ret);
7707 ret = netdev_register_kobject(dev);
7710 dev->reg_state = NETREG_REGISTERED;
7712 __netdev_update_features(dev);
7715 * Default initial state at registry is that the
7716 * device is present.
7719 set_bit(__LINK_STATE_PRESENT, &dev->state);
7721 linkwatch_init_dev(dev);
7723 dev_init_scheduler(dev);
7725 list_netdevice(dev);
7726 add_device_randomness(dev->dev_addr, dev->addr_len);
7728 /* If the device has permanent device address, driver should
7729 * set dev_addr and also addr_assign_type should be set to
7730 * NET_ADDR_PERM (default value).
7732 if (dev->addr_assign_type == NET_ADDR_PERM)
7733 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7735 /* Notify protocols, that a new device appeared. */
7736 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7737 ret = notifier_to_errno(ret);
7739 rollback_registered(dev);
7740 dev->reg_state = NETREG_UNREGISTERED;
7743 * Prevent userspace races by waiting until the network
7744 * device is fully setup before sending notifications.
7746 if (!dev->rtnl_link_ops ||
7747 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7748 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7754 if (dev->netdev_ops->ndo_uninit)
7755 dev->netdev_ops->ndo_uninit(dev);
7756 if (dev->priv_destructor)
7757 dev->priv_destructor(dev);
7760 EXPORT_SYMBOL(register_netdevice);
7763 * init_dummy_netdev - init a dummy network device for NAPI
7764 * @dev: device to init
7766 * This takes a network device structure and initialize the minimum
7767 * amount of fields so it can be used to schedule NAPI polls without
7768 * registering a full blown interface. This is to be used by drivers
7769 * that need to tie several hardware interfaces to a single NAPI
7770 * poll scheduler due to HW limitations.
7772 int init_dummy_netdev(struct net_device *dev)
7774 /* Clear everything. Note we don't initialize spinlocks
7775 * are they aren't supposed to be taken by any of the
7776 * NAPI code and this dummy netdev is supposed to be
7777 * only ever used for NAPI polls
7779 memset(dev, 0, sizeof(struct net_device));
7781 /* make sure we BUG if trying to hit standard
7782 * register/unregister code path
7784 dev->reg_state = NETREG_DUMMY;
7786 /* NAPI wants this */
7787 INIT_LIST_HEAD(&dev->napi_list);
7789 /* a dummy interface is started by default */
7790 set_bit(__LINK_STATE_PRESENT, &dev->state);
7791 set_bit(__LINK_STATE_START, &dev->state);
7793 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7794 * because users of this 'device' dont need to change
7800 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7804 * register_netdev - register a network device
7805 * @dev: device to register
7807 * Take a completed network device structure and add it to the kernel
7808 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7809 * chain. 0 is returned on success. A negative errno code is returned
7810 * on a failure to set up the device, or if the name is a duplicate.
7812 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7813 * and expands the device name if you passed a format string to
7816 int register_netdev(struct net_device *dev)
7821 err = register_netdevice(dev);
7825 EXPORT_SYMBOL(register_netdev);
7827 int netdev_refcnt_read(const struct net_device *dev)
7831 for_each_possible_cpu(i)
7832 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7835 EXPORT_SYMBOL(netdev_refcnt_read);
7838 * netdev_wait_allrefs - wait until all references are gone.
7839 * @dev: target net_device
7841 * This is called when unregistering network devices.
7843 * Any protocol or device that holds a reference should register
7844 * for netdevice notification, and cleanup and put back the
7845 * reference if they receive an UNREGISTER event.
7846 * We can get stuck here if buggy protocols don't correctly
7849 static void netdev_wait_allrefs(struct net_device *dev)
7851 unsigned long rebroadcast_time, warning_time;
7854 linkwatch_forget_dev(dev);
7856 rebroadcast_time = warning_time = jiffies;
7857 refcnt = netdev_refcnt_read(dev);
7859 while (refcnt != 0) {
7860 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7863 /* Rebroadcast unregister notification */
7864 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7870 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7871 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7873 /* We must not have linkwatch events
7874 * pending on unregister. If this
7875 * happens, we simply run the queue
7876 * unscheduled, resulting in a noop
7879 linkwatch_run_queue();
7884 rebroadcast_time = jiffies;
7889 refcnt = netdev_refcnt_read(dev);
7891 if (time_after(jiffies, warning_time + 10 * HZ)) {
7892 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7894 warning_time = jiffies;
7903 * register_netdevice(x1);
7904 * register_netdevice(x2);
7906 * unregister_netdevice(y1);
7907 * unregister_netdevice(y2);
7913 * We are invoked by rtnl_unlock().
7914 * This allows us to deal with problems:
7915 * 1) We can delete sysfs objects which invoke hotplug
7916 * without deadlocking with linkwatch via keventd.
7917 * 2) Since we run with the RTNL semaphore not held, we can sleep
7918 * safely in order to wait for the netdev refcnt to drop to zero.
7920 * We must not return until all unregister events added during
7921 * the interval the lock was held have been completed.
7923 void netdev_run_todo(void)
7925 struct list_head list;
7927 /* Snapshot list, allow later requests */
7928 list_replace_init(&net_todo_list, &list);
7933 /* Wait for rcu callbacks to finish before next phase */
7934 if (!list_empty(&list))
7937 while (!list_empty(&list)) {
7938 struct net_device *dev
7939 = list_first_entry(&list, struct net_device, todo_list);
7940 list_del(&dev->todo_list);
7943 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7946 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7947 pr_err("network todo '%s' but state %d\n",
7948 dev->name, dev->reg_state);
7953 dev->reg_state = NETREG_UNREGISTERED;
7955 netdev_wait_allrefs(dev);
7958 BUG_ON(netdev_refcnt_read(dev));
7959 BUG_ON(!list_empty(&dev->ptype_all));
7960 BUG_ON(!list_empty(&dev->ptype_specific));
7961 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7962 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7963 WARN_ON(dev->dn_ptr);
7965 if (dev->priv_destructor)
7966 dev->priv_destructor(dev);
7967 if (dev->needs_free_netdev)
7970 /* Report a network device has been unregistered */
7972 dev_net(dev)->dev_unreg_count--;
7974 wake_up(&netdev_unregistering_wq);
7976 /* Free network device */
7977 kobject_put(&dev->dev.kobj);
7981 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7982 * all the same fields in the same order as net_device_stats, with only
7983 * the type differing, but rtnl_link_stats64 may have additional fields
7984 * at the end for newer counters.
7986 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7987 const struct net_device_stats *netdev_stats)
7989 #if BITS_PER_LONG == 64
7990 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7991 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
7992 /* zero out counters that only exist in rtnl_link_stats64 */
7993 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7994 sizeof(*stats64) - sizeof(*netdev_stats));
7996 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7997 const unsigned long *src = (const unsigned long *)netdev_stats;
7998 u64 *dst = (u64 *)stats64;
8000 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
8001 for (i = 0; i < n; i++)
8003 /* zero out counters that only exist in rtnl_link_stats64 */
8004 memset((char *)stats64 + n * sizeof(u64), 0,
8005 sizeof(*stats64) - n * sizeof(u64));
8008 EXPORT_SYMBOL(netdev_stats_to_stats64);
8011 * dev_get_stats - get network device statistics
8012 * @dev: device to get statistics from
8013 * @storage: place to store stats
8015 * Get network statistics from device. Return @storage.
8016 * The device driver may provide its own method by setting
8017 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8018 * otherwise the internal statistics structure is used.
8020 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
8021 struct rtnl_link_stats64 *storage)
8023 const struct net_device_ops *ops = dev->netdev_ops;
8025 if (ops->ndo_get_stats64) {
8026 memset(storage, 0, sizeof(*storage));
8027 ops->ndo_get_stats64(dev, storage);
8028 } else if (ops->ndo_get_stats) {
8029 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
8031 netdev_stats_to_stats64(storage, &dev->stats);
8033 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
8034 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
8035 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
8038 EXPORT_SYMBOL(dev_get_stats);
8040 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
8042 struct netdev_queue *queue = dev_ingress_queue(dev);
8044 #ifdef CONFIG_NET_CLS_ACT
8047 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
8050 netdev_init_one_queue(dev, queue, NULL);
8051 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
8052 queue->qdisc_sleeping = &noop_qdisc;
8053 rcu_assign_pointer(dev->ingress_queue, queue);
8058 static const struct ethtool_ops default_ethtool_ops;
8060 void netdev_set_default_ethtool_ops(struct net_device *dev,
8061 const struct ethtool_ops *ops)
8063 if (dev->ethtool_ops == &default_ethtool_ops)
8064 dev->ethtool_ops = ops;
8066 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
8068 void netdev_freemem(struct net_device *dev)
8070 char *addr = (char *)dev - dev->padded;
8076 * alloc_netdev_mqs - allocate network device
8077 * @sizeof_priv: size of private data to allocate space for
8078 * @name: device name format string
8079 * @name_assign_type: origin of device name
8080 * @setup: callback to initialize device
8081 * @txqs: the number of TX subqueues to allocate
8082 * @rxqs: the number of RX subqueues to allocate
8084 * Allocates a struct net_device with private data area for driver use
8085 * and performs basic initialization. Also allocates subqueue structs
8086 * for each queue on the device.
8088 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
8089 unsigned char name_assign_type,
8090 void (*setup)(struct net_device *),
8091 unsigned int txqs, unsigned int rxqs)
8093 struct net_device *dev;
8094 unsigned int alloc_size;
8095 struct net_device *p;
8097 BUG_ON(strlen(name) >= sizeof(dev->name));
8100 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8106 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
8111 alloc_size = sizeof(struct net_device);
8113 /* ensure 32-byte alignment of private area */
8114 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
8115 alloc_size += sizeof_priv;
8117 /* ensure 32-byte alignment of whole construct */
8118 alloc_size += NETDEV_ALIGN - 1;
8120 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8124 dev = PTR_ALIGN(p, NETDEV_ALIGN);
8125 dev->padded = (char *)dev - (char *)p;
8127 dev->pcpu_refcnt = alloc_percpu(int);
8128 if (!dev->pcpu_refcnt)
8131 if (dev_addr_init(dev))
8137 dev_net_set(dev, &init_net);
8139 dev->gso_max_size = GSO_MAX_SIZE;
8140 dev->gso_max_segs = GSO_MAX_SEGS;
8142 INIT_LIST_HEAD(&dev->napi_list);
8143 INIT_LIST_HEAD(&dev->unreg_list);
8144 INIT_LIST_HEAD(&dev->close_list);
8145 INIT_LIST_HEAD(&dev->link_watch_list);
8146 INIT_LIST_HEAD(&dev->adj_list.upper);
8147 INIT_LIST_HEAD(&dev->adj_list.lower);
8148 INIT_LIST_HEAD(&dev->ptype_all);
8149 INIT_LIST_HEAD(&dev->ptype_specific);
8150 #ifdef CONFIG_NET_SCHED
8151 hash_init(dev->qdisc_hash);
8153 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
8156 if (!dev->tx_queue_len) {
8157 dev->priv_flags |= IFF_NO_QUEUE;
8158 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
8161 dev->num_tx_queues = txqs;
8162 dev->real_num_tx_queues = txqs;
8163 if (netif_alloc_netdev_queues(dev))
8167 dev->num_rx_queues = rxqs;
8168 dev->real_num_rx_queues = rxqs;
8169 if (netif_alloc_rx_queues(dev))
8173 strcpy(dev->name, name);
8174 dev->name_assign_type = name_assign_type;
8175 dev->group = INIT_NETDEV_GROUP;
8176 if (!dev->ethtool_ops)
8177 dev->ethtool_ops = &default_ethtool_ops;
8179 nf_hook_ingress_init(dev);
8188 free_percpu(dev->pcpu_refcnt);
8190 netdev_freemem(dev);
8193 EXPORT_SYMBOL(alloc_netdev_mqs);
8196 * free_netdev - free network device
8199 * This function does the last stage of destroying an allocated device
8200 * interface. The reference to the device object is released. If this
8201 * is the last reference then it will be freed.Must be called in process
8204 void free_netdev(struct net_device *dev)
8206 struct napi_struct *p, *n;
8207 struct bpf_prog *prog;
8210 netif_free_tx_queues(dev);
8215 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
8217 /* Flush device addresses */
8218 dev_addr_flush(dev);
8220 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
8223 free_percpu(dev->pcpu_refcnt);
8224 dev->pcpu_refcnt = NULL;
8226 prog = rcu_dereference_protected(dev->xdp_prog, 1);
8229 static_key_slow_dec(&generic_xdp_needed);
8232 /* Compatibility with error handling in drivers */
8233 if (dev->reg_state == NETREG_UNINITIALIZED) {
8234 netdev_freemem(dev);
8238 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
8239 dev->reg_state = NETREG_RELEASED;
8241 /* will free via device release */
8242 put_device(&dev->dev);
8244 EXPORT_SYMBOL(free_netdev);
8247 * synchronize_net - Synchronize with packet receive processing
8249 * Wait for packets currently being received to be done.
8250 * Does not block later packets from starting.
8252 void synchronize_net(void)
8255 if (rtnl_is_locked())
8256 synchronize_rcu_expedited();
8260 EXPORT_SYMBOL(synchronize_net);
8263 * unregister_netdevice_queue - remove device from the kernel
8267 * This function shuts down a device interface and removes it
8268 * from the kernel tables.
8269 * If head not NULL, device is queued to be unregistered later.
8271 * Callers must hold the rtnl semaphore. You may want
8272 * unregister_netdev() instead of this.
8275 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
8280 list_move_tail(&dev->unreg_list, head);
8282 rollback_registered(dev);
8283 /* Finish processing unregister after unlock */
8287 EXPORT_SYMBOL(unregister_netdevice_queue);
8290 * unregister_netdevice_many - unregister many devices
8291 * @head: list of devices
8293 * Note: As most callers use a stack allocated list_head,
8294 * we force a list_del() to make sure stack wont be corrupted later.
8296 void unregister_netdevice_many(struct list_head *head)
8298 struct net_device *dev;
8300 if (!list_empty(head)) {
8301 rollback_registered_many(head);
8302 list_for_each_entry(dev, head, unreg_list)
8307 EXPORT_SYMBOL(unregister_netdevice_many);
8310 * unregister_netdev - remove device from the kernel
8313 * This function shuts down a device interface and removes it
8314 * from the kernel tables.
8316 * This is just a wrapper for unregister_netdevice that takes
8317 * the rtnl semaphore. In general you want to use this and not
8318 * unregister_netdevice.
8320 void unregister_netdev(struct net_device *dev)
8323 unregister_netdevice(dev);
8326 EXPORT_SYMBOL(unregister_netdev);
8329 * dev_change_net_namespace - move device to different nethost namespace
8331 * @net: network namespace
8332 * @pat: If not NULL name pattern to try if the current device name
8333 * is already taken in the destination network namespace.
8335 * This function shuts down a device interface and moves it
8336 * to a new network namespace. On success 0 is returned, on
8337 * a failure a netagive errno code is returned.
8339 * Callers must hold the rtnl semaphore.
8342 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
8348 /* Don't allow namespace local devices to be moved. */
8350 if (dev->features & NETIF_F_NETNS_LOCAL)
8353 /* Ensure the device has been registrered */
8354 if (dev->reg_state != NETREG_REGISTERED)
8357 /* Get out if there is nothing todo */
8359 if (net_eq(dev_net(dev), net))
8362 /* Pick the destination device name, and ensure
8363 * we can use it in the destination network namespace.
8366 if (__dev_get_by_name(net, dev->name)) {
8367 /* We get here if we can't use the current device name */
8370 if (dev_get_valid_name(net, dev, pat) < 0)
8375 * And now a mini version of register_netdevice unregister_netdevice.
8378 /* If device is running close it first. */
8381 /* And unlink it from device chain */
8383 unlist_netdevice(dev);
8387 /* Shutdown queueing discipline. */
8390 /* Notify protocols, that we are about to destroy
8391 * this device. They should clean all the things.
8393 * Note that dev->reg_state stays at NETREG_REGISTERED.
8394 * This is wanted because this way 8021q and macvlan know
8395 * the device is just moving and can keep their slaves up.
8397 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8399 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
8400 if (dev->rtnl_link_ops && dev->rtnl_link_ops->get_link_net)
8401 new_nsid = peernet2id_alloc(dev_net(dev), net);
8403 new_nsid = peernet2id(dev_net(dev), net);
8404 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid);
8407 * Flush the unicast and multicast chains
8412 /* Send a netdev-removed uevent to the old namespace */
8413 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
8414 netdev_adjacent_del_links(dev);
8416 /* Actually switch the network namespace */
8417 dev_net_set(dev, net);
8419 /* If there is an ifindex conflict assign a new one */
8420 if (__dev_get_by_index(net, dev->ifindex))
8421 dev->ifindex = dev_new_index(net);
8423 /* Send a netdev-add uevent to the new namespace */
8424 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
8425 netdev_adjacent_add_links(dev);
8427 /* Fixup kobjects */
8428 err = device_rename(&dev->dev, dev->name);
8431 /* Add the device back in the hashes */
8432 list_netdevice(dev);
8434 /* Notify protocols, that a new device appeared. */
8435 call_netdevice_notifiers(NETDEV_REGISTER, dev);
8438 * Prevent userspace races by waiting until the network
8439 * device is fully setup before sending notifications.
8441 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8448 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8450 static int dev_cpu_dead(unsigned int oldcpu)
8452 struct sk_buff **list_skb;
8453 struct sk_buff *skb;
8455 struct softnet_data *sd, *oldsd, *remsd = NULL;
8457 local_irq_disable();
8458 cpu = smp_processor_id();
8459 sd = &per_cpu(softnet_data, cpu);
8460 oldsd = &per_cpu(softnet_data, oldcpu);
8462 /* Find end of our completion_queue. */
8463 list_skb = &sd->completion_queue;
8465 list_skb = &(*list_skb)->next;
8466 /* Append completion queue from offline CPU. */
8467 *list_skb = oldsd->completion_queue;
8468 oldsd->completion_queue = NULL;
8470 /* Append output queue from offline CPU. */
8471 if (oldsd->output_queue) {
8472 *sd->output_queue_tailp = oldsd->output_queue;
8473 sd->output_queue_tailp = oldsd->output_queue_tailp;
8474 oldsd->output_queue = NULL;
8475 oldsd->output_queue_tailp = &oldsd->output_queue;
8477 /* Append NAPI poll list from offline CPU, with one exception :
8478 * process_backlog() must be called by cpu owning percpu backlog.
8479 * We properly handle process_queue & input_pkt_queue later.
8481 while (!list_empty(&oldsd->poll_list)) {
8482 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8486 list_del_init(&napi->poll_list);
8487 if (napi->poll == process_backlog)
8490 ____napi_schedule(sd, napi);
8493 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8497 remsd = oldsd->rps_ipi_list;
8498 oldsd->rps_ipi_list = NULL;
8500 /* send out pending IPI's on offline CPU */
8501 net_rps_send_ipi(remsd);
8503 /* Process offline CPU's input_pkt_queue */
8504 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8506 input_queue_head_incr(oldsd);
8508 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8510 input_queue_head_incr(oldsd);
8517 * netdev_increment_features - increment feature set by one
8518 * @all: current feature set
8519 * @one: new feature set
8520 * @mask: mask feature set
8522 * Computes a new feature set after adding a device with feature set
8523 * @one to the master device with current feature set @all. Will not
8524 * enable anything that is off in @mask. Returns the new feature set.
8526 netdev_features_t netdev_increment_features(netdev_features_t all,
8527 netdev_features_t one, netdev_features_t mask)
8529 if (mask & NETIF_F_HW_CSUM)
8530 mask |= NETIF_F_CSUM_MASK;
8531 mask |= NETIF_F_VLAN_CHALLENGED;
8533 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8534 all &= one | ~NETIF_F_ALL_FOR_ALL;
8536 /* If one device supports hw checksumming, set for all. */
8537 if (all & NETIF_F_HW_CSUM)
8538 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8542 EXPORT_SYMBOL(netdev_increment_features);
8544 static struct hlist_head * __net_init netdev_create_hash(void)
8547 struct hlist_head *hash;
8549 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8551 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8552 INIT_HLIST_HEAD(&hash[i]);
8557 /* Initialize per network namespace state */
8558 static int __net_init netdev_init(struct net *net)
8560 if (net != &init_net)
8561 INIT_LIST_HEAD(&net->dev_base_head);
8563 net->dev_name_head = netdev_create_hash();
8564 if (net->dev_name_head == NULL)
8567 net->dev_index_head = netdev_create_hash();
8568 if (net->dev_index_head == NULL)
8574 kfree(net->dev_name_head);
8580 * netdev_drivername - network driver for the device
8581 * @dev: network device
8583 * Determine network driver for device.
8585 const char *netdev_drivername(const struct net_device *dev)
8587 const struct device_driver *driver;
8588 const struct device *parent;
8589 const char *empty = "";
8591 parent = dev->dev.parent;
8595 driver = parent->driver;
8596 if (driver && driver->name)
8597 return driver->name;
8601 static void __netdev_printk(const char *level, const struct net_device *dev,
8602 struct va_format *vaf)
8604 if (dev && dev->dev.parent) {
8605 dev_printk_emit(level[1] - '0',
8608 dev_driver_string(dev->dev.parent),
8609 dev_name(dev->dev.parent),
8610 netdev_name(dev), netdev_reg_state(dev),
8613 printk("%s%s%s: %pV",
8614 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8616 printk("%s(NULL net_device): %pV", level, vaf);
8620 void netdev_printk(const char *level, const struct net_device *dev,
8621 const char *format, ...)
8623 struct va_format vaf;
8626 va_start(args, format);
8631 __netdev_printk(level, dev, &vaf);
8635 EXPORT_SYMBOL(netdev_printk);
8637 #define define_netdev_printk_level(func, level) \
8638 void func(const struct net_device *dev, const char *fmt, ...) \
8640 struct va_format vaf; \
8643 va_start(args, fmt); \
8648 __netdev_printk(level, dev, &vaf); \
8652 EXPORT_SYMBOL(func);
8654 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8655 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8656 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8657 define_netdev_printk_level(netdev_err, KERN_ERR);
8658 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8659 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8660 define_netdev_printk_level(netdev_info, KERN_INFO);
8662 static void __net_exit netdev_exit(struct net *net)
8664 kfree(net->dev_name_head);
8665 kfree(net->dev_index_head);
8668 static struct pernet_operations __net_initdata netdev_net_ops = {
8669 .init = netdev_init,
8670 .exit = netdev_exit,
8673 static void __net_exit default_device_exit(struct net *net)
8675 struct net_device *dev, *aux;
8677 * Push all migratable network devices back to the
8678 * initial network namespace
8681 for_each_netdev_safe(net, dev, aux) {
8683 char fb_name[IFNAMSIZ];
8685 /* Ignore unmoveable devices (i.e. loopback) */
8686 if (dev->features & NETIF_F_NETNS_LOCAL)
8689 /* Leave virtual devices for the generic cleanup */
8690 if (dev->rtnl_link_ops)
8693 /* Push remaining network devices to init_net */
8694 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8695 err = dev_change_net_namespace(dev, &init_net, fb_name);
8697 pr_emerg("%s: failed to move %s to init_net: %d\n",
8698 __func__, dev->name, err);
8705 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8707 /* Return with the rtnl_lock held when there are no network
8708 * devices unregistering in any network namespace in net_list.
8712 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8714 add_wait_queue(&netdev_unregistering_wq, &wait);
8716 unregistering = false;
8718 list_for_each_entry(net, net_list, exit_list) {
8719 if (net->dev_unreg_count > 0) {
8720 unregistering = true;
8728 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8730 remove_wait_queue(&netdev_unregistering_wq, &wait);
8733 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8735 /* At exit all network devices most be removed from a network
8736 * namespace. Do this in the reverse order of registration.
8737 * Do this across as many network namespaces as possible to
8738 * improve batching efficiency.
8740 struct net_device *dev;
8742 LIST_HEAD(dev_kill_list);
8744 /* To prevent network device cleanup code from dereferencing
8745 * loopback devices or network devices that have been freed
8746 * wait here for all pending unregistrations to complete,
8747 * before unregistring the loopback device and allowing the
8748 * network namespace be freed.
8750 * The netdev todo list containing all network devices
8751 * unregistrations that happen in default_device_exit_batch
8752 * will run in the rtnl_unlock() at the end of
8753 * default_device_exit_batch.
8755 rtnl_lock_unregistering(net_list);
8756 list_for_each_entry(net, net_list, exit_list) {
8757 for_each_netdev_reverse(net, dev) {
8758 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8759 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8761 unregister_netdevice_queue(dev, &dev_kill_list);
8764 unregister_netdevice_many(&dev_kill_list);
8768 static struct pernet_operations __net_initdata default_device_ops = {
8769 .exit = default_device_exit,
8770 .exit_batch = default_device_exit_batch,
8774 * Initialize the DEV module. At boot time this walks the device list and
8775 * unhooks any devices that fail to initialise (normally hardware not
8776 * present) and leaves us with a valid list of present and active devices.
8781 * This is called single threaded during boot, so no need
8782 * to take the rtnl semaphore.
8784 static int __init net_dev_init(void)
8786 int i, rc = -ENOMEM;
8788 BUG_ON(!dev_boot_phase);
8790 if (dev_proc_init())
8793 if (netdev_kobject_init())
8796 INIT_LIST_HEAD(&ptype_all);
8797 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8798 INIT_LIST_HEAD(&ptype_base[i]);
8800 INIT_LIST_HEAD(&offload_base);
8802 if (register_pernet_subsys(&netdev_net_ops))
8806 * Initialise the packet receive queues.
8809 for_each_possible_cpu(i) {
8810 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8811 struct softnet_data *sd = &per_cpu(softnet_data, i);
8813 INIT_WORK(flush, flush_backlog);
8815 skb_queue_head_init(&sd->input_pkt_queue);
8816 skb_queue_head_init(&sd->process_queue);
8817 INIT_LIST_HEAD(&sd->poll_list);
8818 sd->output_queue_tailp = &sd->output_queue;
8820 sd->csd.func = rps_trigger_softirq;
8825 sd->backlog.poll = process_backlog;
8826 sd->backlog.weight = weight_p;
8831 /* The loopback device is special if any other network devices
8832 * is present in a network namespace the loopback device must
8833 * be present. Since we now dynamically allocate and free the
8834 * loopback device ensure this invariant is maintained by
8835 * keeping the loopback device as the first device on the
8836 * list of network devices. Ensuring the loopback devices
8837 * is the first device that appears and the last network device
8840 if (register_pernet_device(&loopback_net_ops))
8843 if (register_pernet_device(&default_device_ops))
8846 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8847 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8849 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8850 NULL, dev_cpu_dead);
8857 subsys_initcall(net_dev_init);