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/rwsem.h>
87 #include <linux/string.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/errno.h>
92 #include <linux/interrupt.h>
93 #include <linux/if_ether.h>
94 #include <linux/netdevice.h>
95 #include <linux/etherdevice.h>
96 #include <linux/ethtool.h>
97 #include <linux/notifier.h>
98 #include <linux/skbuff.h>
99 #include <linux/bpf.h>
100 #include <linux/bpf_trace.h>
101 #include <net/net_namespace.h>
102 #include <net/sock.h>
103 #include <net/busy_poll.h>
104 #include <linux/rtnetlink.h>
105 #include <linux/stat.h>
107 #include <net/dst_metadata.h>
108 #include <net/pkt_sched.h>
109 #include <net/pkt_cls.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <linux/highmem.h>
113 #include <linux/init.h>
114 #include <linux/module.h>
115 #include <linux/netpoll.h>
116 #include <linux/rcupdate.h>
117 #include <linux/delay.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <net/mpls.h>
129 #include <linux/ipv6.h>
130 #include <linux/in.h>
131 #include <linux/jhash.h>
132 #include <linux/random.h>
133 #include <trace/events/napi.h>
134 #include <trace/events/net.h>
135 #include <trace/events/skb.h>
136 #include <linux/pci.h>
137 #include <linux/inetdevice.h>
138 #include <linux/cpu_rmap.h>
139 #include <linux/static_key.h>
140 #include <linux/hashtable.h>
141 #include <linux/vmalloc.h>
142 #include <linux/if_macvlan.h>
143 #include <linux/errqueue.h>
144 #include <linux/hrtimer.h>
145 #include <linux/netfilter_ingress.h>
146 #include <linux/crash_dump.h>
147 #include <linux/sctp.h>
148 #include <net/udp_tunnel.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 net_device *dev,
167 struct netdev_notifier_info *info);
168 static struct napi_struct *napi_by_id(unsigned int napi_id);
171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
176 * Writers must hold the rtnl semaphore while they loop through the
177 * dev_base_head list, and hold dev_base_lock for writing when they do the
178 * actual updates. This allows pure readers to access the list even
179 * while a writer is preparing to update it.
181 * To put it another way, dev_base_lock is held for writing only to
182 * protect against pure readers; the rtnl semaphore provides the
183 * protection against other writers.
185 * See, for example usages, register_netdevice() and
186 * unregister_netdevice(), which must be called with the rtnl
189 DEFINE_RWLOCK(dev_base_lock);
190 EXPORT_SYMBOL(dev_base_lock);
192 /* protects napi_hash addition/deletion and napi_gen_id */
193 static DEFINE_SPINLOCK(napi_hash_lock);
195 static unsigned int napi_gen_id = NR_CPUS;
196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
198 static DECLARE_RWSEM(devnet_rename_sem);
200 static inline void dev_base_seq_inc(struct net *net)
202 while (++net->dev_base_seq == 0)
206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
210 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
218 static inline void rps_lock(struct softnet_data *sd)
221 spin_lock(&sd->input_pkt_queue.lock);
225 static inline void rps_unlock(struct softnet_data *sd)
228 spin_unlock(&sd->input_pkt_queue.lock);
232 /* Device list insertion */
233 static void list_netdevice(struct net_device *dev)
235 struct net *net = dev_net(dev);
239 write_lock_bh(&dev_base_lock);
240 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
241 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
242 hlist_add_head_rcu(&dev->index_hlist,
243 dev_index_hash(net, dev->ifindex));
244 write_unlock_bh(&dev_base_lock);
246 dev_base_seq_inc(net);
249 /* Device list removal
250 * caller must respect a RCU grace period before freeing/reusing dev
252 static void unlist_netdevice(struct net_device *dev)
256 /* Unlink dev from the device chain */
257 write_lock_bh(&dev_base_lock);
258 list_del_rcu(&dev->dev_list);
259 hlist_del_rcu(&dev->name_hlist);
260 hlist_del_rcu(&dev->index_hlist);
261 write_unlock_bh(&dev_base_lock);
263 dev_base_seq_inc(dev_net(dev));
270 static RAW_NOTIFIER_HEAD(netdev_chain);
273 * Device drivers call our routines to queue packets here. We empty the
274 * queue in the local softnet handler.
277 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
278 EXPORT_PER_CPU_SYMBOL(softnet_data);
280 #ifdef CONFIG_LOCKDEP
282 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
283 * according to dev->type
285 static const unsigned short netdev_lock_type[] = {
286 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
287 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
288 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
289 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
290 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
291 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
292 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
293 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
294 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
295 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
296 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
297 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
298 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
299 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
300 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
302 static const char *const netdev_lock_name[] = {
303 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
304 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
305 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
306 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
307 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
308 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
309 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
310 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
311 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
312 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
313 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
314 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
315 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
316 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
317 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
319 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
320 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
322 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
326 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
327 if (netdev_lock_type[i] == dev_type)
329 /* the last key is used by default */
330 return ARRAY_SIZE(netdev_lock_type) - 1;
333 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
334 unsigned short dev_type)
338 i = netdev_lock_pos(dev_type);
339 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
340 netdev_lock_name[i]);
343 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 i = netdev_lock_pos(dev->type);
348 lockdep_set_class_and_name(&dev->addr_list_lock,
349 &netdev_addr_lock_key[i],
350 netdev_lock_name[i]);
353 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
354 unsigned short dev_type)
357 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
362 /*******************************************************************************
364 * Protocol management and registration routines
366 *******************************************************************************/
370 * Add a protocol ID to the list. Now that the input handler is
371 * smarter we can dispense with all the messy stuff that used to be
374 * BEWARE!!! Protocol handlers, mangling input packets,
375 * MUST BE last in hash buckets and checking protocol handlers
376 * MUST start from promiscuous ptype_all chain in net_bh.
377 * It is true now, do not change it.
378 * Explanation follows: if protocol handler, mangling packet, will
379 * be the first on list, it is not able to sense, that packet
380 * is cloned and should be copied-on-write, so that it will
381 * change it and subsequent readers will get broken packet.
385 static inline struct list_head *ptype_head(const struct packet_type *pt)
387 if (pt->type == htons(ETH_P_ALL))
388 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
390 return pt->dev ? &pt->dev->ptype_specific :
391 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
395 * dev_add_pack - add packet handler
396 * @pt: packet type declaration
398 * Add a protocol handler to the networking stack. The passed &packet_type
399 * is linked into kernel lists and may not be freed until it has been
400 * removed from the kernel lists.
402 * This call does not sleep therefore it can not
403 * guarantee all CPU's that are in middle of receiving packets
404 * will see the new packet type (until the next received packet).
407 void dev_add_pack(struct packet_type *pt)
409 struct list_head *head = ptype_head(pt);
411 spin_lock(&ptype_lock);
412 list_add_rcu(&pt->list, head);
413 spin_unlock(&ptype_lock);
415 EXPORT_SYMBOL(dev_add_pack);
418 * __dev_remove_pack - remove packet handler
419 * @pt: packet type declaration
421 * Remove a protocol handler that was previously added to the kernel
422 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
423 * from the kernel lists and can be freed or reused once this function
426 * The packet type might still be in use by receivers
427 * and must not be freed until after all the CPU's have gone
428 * through a quiescent state.
430 void __dev_remove_pack(struct packet_type *pt)
432 struct list_head *head = ptype_head(pt);
433 struct packet_type *pt1;
435 spin_lock(&ptype_lock);
437 list_for_each_entry(pt1, head, list) {
439 list_del_rcu(&pt->list);
444 pr_warn("dev_remove_pack: %p not found\n", pt);
446 spin_unlock(&ptype_lock);
448 EXPORT_SYMBOL(__dev_remove_pack);
451 * dev_remove_pack - remove packet handler
452 * @pt: packet type declaration
454 * Remove a protocol handler that was previously added to the kernel
455 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
456 * from the kernel lists and can be freed or reused once this function
459 * This call sleeps to guarantee that no CPU is looking at the packet
462 void dev_remove_pack(struct packet_type *pt)
464 __dev_remove_pack(pt);
468 EXPORT_SYMBOL(dev_remove_pack);
472 * dev_add_offload - register offload handlers
473 * @po: protocol offload declaration
475 * Add protocol offload handlers to the networking stack. The passed
476 * &proto_offload is linked into kernel lists and may not be freed until
477 * it has been removed from the kernel lists.
479 * This call does not sleep therefore it can not
480 * guarantee all CPU's that are in middle of receiving packets
481 * will see the new offload handlers (until the next received packet).
483 void dev_add_offload(struct packet_offload *po)
485 struct packet_offload *elem;
487 spin_lock(&offload_lock);
488 list_for_each_entry(elem, &offload_base, list) {
489 if (po->priority < elem->priority)
492 list_add_rcu(&po->list, elem->list.prev);
493 spin_unlock(&offload_lock);
495 EXPORT_SYMBOL(dev_add_offload);
498 * __dev_remove_offload - remove offload handler
499 * @po: packet offload declaration
501 * Remove a protocol offload handler that was previously added to the
502 * kernel offload handlers by dev_add_offload(). The passed &offload_type
503 * is removed from the kernel lists and can be freed or reused once this
506 * The packet type might still be in use by receivers
507 * and must not be freed until after all the CPU's have gone
508 * through a quiescent state.
510 static void __dev_remove_offload(struct packet_offload *po)
512 struct list_head *head = &offload_base;
513 struct packet_offload *po1;
515 spin_lock(&offload_lock);
517 list_for_each_entry(po1, head, list) {
519 list_del_rcu(&po->list);
524 pr_warn("dev_remove_offload: %p not found\n", po);
526 spin_unlock(&offload_lock);
530 * dev_remove_offload - remove packet offload handler
531 * @po: packet offload declaration
533 * Remove a packet offload handler that was previously added to the kernel
534 * offload handlers by dev_add_offload(). The passed &offload_type is
535 * removed from the kernel lists and can be freed or reused once this
538 * This call sleeps to guarantee that no CPU is looking at the packet
541 void dev_remove_offload(struct packet_offload *po)
543 __dev_remove_offload(po);
547 EXPORT_SYMBOL(dev_remove_offload);
549 /******************************************************************************
551 * Device Boot-time Settings Routines
553 ******************************************************************************/
555 /* Boot time configuration table */
556 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
559 * netdev_boot_setup_add - add new setup entry
560 * @name: name of the device
561 * @map: configured settings for the device
563 * Adds new setup entry to the dev_boot_setup list. The function
564 * returns 0 on error and 1 on success. This is a generic routine to
567 static int netdev_boot_setup_add(char *name, struct ifmap *map)
569 struct netdev_boot_setup *s;
573 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
574 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
575 memset(s[i].name, 0, sizeof(s[i].name));
576 strlcpy(s[i].name, name, IFNAMSIZ);
577 memcpy(&s[i].map, map, sizeof(s[i].map));
582 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
586 * netdev_boot_setup_check - check boot time settings
587 * @dev: the netdevice
589 * Check boot time settings for the device.
590 * The found settings are set for the device to be used
591 * later in the device probing.
592 * Returns 0 if no settings found, 1 if they are.
594 int netdev_boot_setup_check(struct net_device *dev)
596 struct netdev_boot_setup *s = dev_boot_setup;
599 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
600 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
601 !strcmp(dev->name, s[i].name)) {
602 dev->irq = s[i].map.irq;
603 dev->base_addr = s[i].map.base_addr;
604 dev->mem_start = s[i].map.mem_start;
605 dev->mem_end = s[i].map.mem_end;
611 EXPORT_SYMBOL(netdev_boot_setup_check);
615 * netdev_boot_base - get address from boot time settings
616 * @prefix: prefix for network device
617 * @unit: id for network device
619 * Check boot time settings for the base address of device.
620 * The found settings are set for the device to be used
621 * later in the device probing.
622 * Returns 0 if no settings found.
624 unsigned long netdev_boot_base(const char *prefix, int unit)
626 const struct netdev_boot_setup *s = dev_boot_setup;
630 sprintf(name, "%s%d", prefix, unit);
633 * If device already registered then return base of 1
634 * to indicate not to probe for this interface
636 if (__dev_get_by_name(&init_net, name))
639 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
640 if (!strcmp(name, s[i].name))
641 return s[i].map.base_addr;
646 * Saves at boot time configured settings for any netdevice.
648 int __init netdev_boot_setup(char *str)
653 str = get_options(str, ARRAY_SIZE(ints), ints);
658 memset(&map, 0, sizeof(map));
662 map.base_addr = ints[2];
664 map.mem_start = ints[3];
666 map.mem_end = ints[4];
668 /* Add new entry to the list */
669 return netdev_boot_setup_add(str, &map);
672 __setup("netdev=", netdev_boot_setup);
674 /*******************************************************************************
676 * Device Interface Subroutines
678 *******************************************************************************/
681 * dev_get_iflink - get 'iflink' value of a interface
682 * @dev: targeted interface
684 * Indicates the ifindex the interface is linked to.
685 * Physical interfaces have the same 'ifindex' and 'iflink' values.
688 int dev_get_iflink(const struct net_device *dev)
690 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
691 return dev->netdev_ops->ndo_get_iflink(dev);
695 EXPORT_SYMBOL(dev_get_iflink);
698 * dev_fill_metadata_dst - Retrieve tunnel egress information.
699 * @dev: targeted interface
702 * For better visibility of tunnel traffic OVS needs to retrieve
703 * egress tunnel information for a packet. Following API allows
704 * user to get this info.
706 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
708 struct ip_tunnel_info *info;
710 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
713 info = skb_tunnel_info_unclone(skb);
716 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
719 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
721 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
724 * __dev_get_by_name - find a device by its name
725 * @net: the applicable net namespace
726 * @name: name to find
728 * Find an interface by name. Must be called under RTNL semaphore
729 * or @dev_base_lock. If the name is found a pointer to the device
730 * is returned. If the name is not found then %NULL is returned. The
731 * reference counters are not incremented so the caller must be
732 * careful with locks.
735 struct net_device *__dev_get_by_name(struct net *net, const char *name)
737 struct net_device *dev;
738 struct hlist_head *head = dev_name_hash(net, name);
740 hlist_for_each_entry(dev, head, name_hlist)
741 if (!strncmp(dev->name, name, IFNAMSIZ))
746 EXPORT_SYMBOL(__dev_get_by_name);
749 * dev_get_by_name_rcu - find a device by its name
750 * @net: the applicable net namespace
751 * @name: name to find
753 * Find an interface by name.
754 * If the name is found a pointer to the device is returned.
755 * If the name is not found then %NULL is returned.
756 * The reference counters are not incremented so the caller must be
757 * careful with locks. The caller must hold RCU lock.
760 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
762 struct net_device *dev;
763 struct hlist_head *head = dev_name_hash(net, name);
765 hlist_for_each_entry_rcu(dev, head, name_hlist)
766 if (!strncmp(dev->name, name, IFNAMSIZ))
771 EXPORT_SYMBOL(dev_get_by_name_rcu);
774 * dev_get_by_name - find a device by its name
775 * @net: the applicable net namespace
776 * @name: name to find
778 * Find an interface by name. This can be called from any
779 * context and does its own locking. The returned handle has
780 * the usage count incremented and the caller must use dev_put() to
781 * release it when it is no longer needed. %NULL is returned if no
782 * matching device is found.
785 struct net_device *dev_get_by_name(struct net *net, const char *name)
787 struct net_device *dev;
790 dev = dev_get_by_name_rcu(net, name);
796 EXPORT_SYMBOL(dev_get_by_name);
799 * __dev_get_by_index - find a device by its ifindex
800 * @net: the applicable net namespace
801 * @ifindex: index of device
803 * Search for an interface by index. Returns %NULL if the device
804 * is not found or a pointer to the device. The device has not
805 * had its reference counter increased so the caller must be careful
806 * about locking. The caller must hold either the RTNL semaphore
810 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
812 struct net_device *dev;
813 struct hlist_head *head = dev_index_hash(net, ifindex);
815 hlist_for_each_entry(dev, head, index_hlist)
816 if (dev->ifindex == ifindex)
821 EXPORT_SYMBOL(__dev_get_by_index);
824 * dev_get_by_index_rcu - find a device by its ifindex
825 * @net: the applicable net namespace
826 * @ifindex: index of device
828 * Search for an interface by index. Returns %NULL if the device
829 * is not found or a pointer to the device. The device has not
830 * had its reference counter increased so the caller must be careful
831 * about locking. The caller must hold RCU lock.
834 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
836 struct net_device *dev;
837 struct hlist_head *head = dev_index_hash(net, ifindex);
839 hlist_for_each_entry_rcu(dev, head, index_hlist)
840 if (dev->ifindex == ifindex)
845 EXPORT_SYMBOL(dev_get_by_index_rcu);
849 * dev_get_by_index - find a device by its ifindex
850 * @net: the applicable net namespace
851 * @ifindex: index of device
853 * Search for an interface by index. Returns NULL if the device
854 * is not found or a pointer to the device. The device returned has
855 * had a reference added and the pointer is safe until the user calls
856 * dev_put to indicate they have finished with it.
859 struct net_device *dev_get_by_index(struct net *net, int ifindex)
861 struct net_device *dev;
864 dev = dev_get_by_index_rcu(net, ifindex);
870 EXPORT_SYMBOL(dev_get_by_index);
873 * dev_get_by_napi_id - find a device by napi_id
874 * @napi_id: ID of the NAPI struct
876 * Search for an interface by NAPI ID. Returns %NULL if the device
877 * is not found or a pointer to the device. The device has not had
878 * its reference counter increased so the caller must be careful
879 * about locking. The caller must hold RCU lock.
882 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
884 struct napi_struct *napi;
886 WARN_ON_ONCE(!rcu_read_lock_held());
888 if (napi_id < MIN_NAPI_ID)
891 napi = napi_by_id(napi_id);
893 return napi ? napi->dev : NULL;
895 EXPORT_SYMBOL(dev_get_by_napi_id);
898 * netdev_get_name - get a netdevice name, knowing its ifindex.
899 * @net: network namespace
900 * @name: a pointer to the buffer where the name will be stored.
901 * @ifindex: the ifindex of the interface to get the name from.
903 int netdev_get_name(struct net *net, char *name, int ifindex)
905 struct net_device *dev;
908 down_read(&devnet_rename_sem);
911 dev = dev_get_by_index_rcu(net, ifindex);
917 strcpy(name, dev->name);
922 up_read(&devnet_rename_sem);
927 * dev_getbyhwaddr_rcu - find a device by its hardware address
928 * @net: the applicable net namespace
929 * @type: media type of device
930 * @ha: hardware address
932 * Search for an interface by MAC address. Returns NULL if the device
933 * is not found or a pointer to the device.
934 * The caller must hold RCU or RTNL.
935 * The returned device has not had its ref count increased
936 * and the caller must therefore be careful about locking
940 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
943 struct net_device *dev;
945 for_each_netdev_rcu(net, dev)
946 if (dev->type == type &&
947 !memcmp(dev->dev_addr, ha, dev->addr_len))
952 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
954 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
956 struct net_device *dev;
959 for_each_netdev(net, dev)
960 if (dev->type == type)
965 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
967 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
969 struct net_device *dev, *ret = NULL;
972 for_each_netdev_rcu(net, dev)
973 if (dev->type == type) {
981 EXPORT_SYMBOL(dev_getfirstbyhwtype);
984 * __dev_get_by_flags - find any device with given flags
985 * @net: the applicable net namespace
986 * @if_flags: IFF_* values
987 * @mask: bitmask of bits in if_flags to check
989 * Search for any interface with the given flags. Returns NULL if a device
990 * is not found or a pointer to the device. Must be called inside
991 * rtnl_lock(), and result refcount is unchanged.
994 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
997 struct net_device *dev, *ret;
1002 for_each_netdev(net, dev) {
1003 if (((dev->flags ^ if_flags) & mask) == 0) {
1010 EXPORT_SYMBOL(__dev_get_by_flags);
1013 * dev_valid_name - check if name is okay for network device
1014 * @name: name string
1016 * Network device names need to be valid file names to
1017 * to allow sysfs to work. We also disallow any kind of
1020 bool dev_valid_name(const char *name)
1024 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1026 if (!strcmp(name, ".") || !strcmp(name, ".."))
1030 if (*name == '/' || *name == ':' || isspace(*name))
1036 EXPORT_SYMBOL(dev_valid_name);
1039 * __dev_alloc_name - allocate a name for a device
1040 * @net: network namespace to allocate the device name in
1041 * @name: name format string
1042 * @buf: scratch buffer and result name string
1044 * Passed a format string - eg "lt%d" it will try and find a suitable
1045 * id. It scans list of devices to build up a free map, then chooses
1046 * the first empty slot. The caller must hold the dev_base or rtnl lock
1047 * while allocating the name and adding the device in order to avoid
1049 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1050 * Returns the number of the unit assigned or a negative errno code.
1053 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1057 const int max_netdevices = 8*PAGE_SIZE;
1058 unsigned long *inuse;
1059 struct net_device *d;
1061 p = strnchr(name, IFNAMSIZ-1, '%');
1064 * Verify the string as this thing may have come from
1065 * the user. There must be either one "%d" and no other "%"
1068 if (p[1] != 'd' || strchr(p + 2, '%'))
1071 /* Use one page as a bit array of possible slots */
1072 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1076 for_each_netdev(net, d) {
1077 if (!sscanf(d->name, name, &i))
1079 if (i < 0 || i >= max_netdevices)
1082 /* avoid cases where sscanf is not exact inverse of printf */
1083 snprintf(buf, IFNAMSIZ, name, i);
1084 if (!strncmp(buf, d->name, IFNAMSIZ))
1088 i = find_first_zero_bit(inuse, max_netdevices);
1089 free_page((unsigned long) inuse);
1093 snprintf(buf, IFNAMSIZ, name, i);
1094 if (!__dev_get_by_name(net, buf))
1097 /* It is possible to run out of possible slots
1098 * when the name is long and there isn't enough space left
1099 * for the digits, or if all bits are used.
1105 * dev_alloc_name - allocate a name for a device
1107 * @name: name format string
1109 * Passed a format string - eg "lt%d" it will try and find a suitable
1110 * id. It scans list of devices to build up a free map, then chooses
1111 * the first empty slot. The caller must hold the dev_base or rtnl lock
1112 * while allocating the name and adding the device in order to avoid
1114 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1115 * Returns the number of the unit assigned or a negative errno code.
1118 int dev_alloc_name(struct net_device *dev, const char *name)
1124 BUG_ON(!dev_net(dev));
1126 ret = __dev_alloc_name(net, name, buf);
1128 strlcpy(dev->name, buf, IFNAMSIZ);
1131 EXPORT_SYMBOL(dev_alloc_name);
1133 static int dev_alloc_name_ns(struct net *net,
1134 struct net_device *dev,
1140 ret = __dev_alloc_name(net, name, buf);
1142 strlcpy(dev->name, buf, IFNAMSIZ);
1146 int dev_get_valid_name(struct net *net, struct net_device *dev,
1151 if (!dev_valid_name(name))
1154 if (strchr(name, '%'))
1155 return dev_alloc_name_ns(net, dev, name);
1156 else if (__dev_get_by_name(net, name))
1158 else if (dev->name != name)
1159 strlcpy(dev->name, name, IFNAMSIZ);
1163 EXPORT_SYMBOL(dev_get_valid_name);
1166 * dev_change_name - change name of a device
1168 * @newname: name (or format string) must be at least IFNAMSIZ
1170 * Change name of a device, can pass format strings "eth%d".
1173 int dev_change_name(struct net_device *dev, const char *newname)
1175 unsigned char old_assign_type;
1176 char oldname[IFNAMSIZ];
1182 BUG_ON(!dev_net(dev));
1185 if (dev->flags & IFF_UP)
1188 down_write(&devnet_rename_sem);
1190 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1191 up_write(&devnet_rename_sem);
1195 memcpy(oldname, dev->name, IFNAMSIZ);
1197 err = dev_get_valid_name(net, dev, newname);
1199 up_write(&devnet_rename_sem);
1203 if (oldname[0] && !strchr(oldname, '%'))
1204 netdev_info(dev, "renamed from %s\n", oldname);
1206 old_assign_type = dev->name_assign_type;
1207 dev->name_assign_type = NET_NAME_RENAMED;
1210 ret = device_rename(&dev->dev, dev->name);
1212 memcpy(dev->name, oldname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 up_write(&devnet_rename_sem);
1218 up_write(&devnet_rename_sem);
1220 netdev_adjacent_rename_links(dev, oldname);
1222 write_lock_bh(&dev_base_lock);
1223 hlist_del_rcu(&dev->name_hlist);
1224 write_unlock_bh(&dev_base_lock);
1228 write_lock_bh(&dev_base_lock);
1229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1230 write_unlock_bh(&dev_base_lock);
1232 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1233 ret = notifier_to_errno(ret);
1236 /* err >= 0 after dev_alloc_name() or stores the first errno */
1239 down_write(&devnet_rename_sem);
1240 memcpy(dev->name, oldname, IFNAMSIZ);
1241 memcpy(oldname, newname, IFNAMSIZ);
1242 dev->name_assign_type = old_assign_type;
1243 old_assign_type = NET_NAME_RENAMED;
1246 pr_err("%s: name change rollback failed: %d\n",
1255 * dev_set_alias - change ifalias of a device
1257 * @alias: name up to IFALIASZ
1258 * @len: limit of bytes to copy from info
1260 * Set ifalias for a device,
1262 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1268 if (len >= IFALIASZ)
1272 kfree(dev->ifalias);
1273 dev->ifalias = NULL;
1277 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1280 dev->ifalias = new_ifalias;
1281 memcpy(dev->ifalias, alias, len);
1282 dev->ifalias[len] = 0;
1289 * netdev_features_change - device changes features
1290 * @dev: device to cause notification
1292 * Called to indicate a device has changed features.
1294 void netdev_features_change(struct net_device *dev)
1296 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1298 EXPORT_SYMBOL(netdev_features_change);
1301 * netdev_state_change - device changes state
1302 * @dev: device to cause notification
1304 * Called to indicate a device has changed state. This function calls
1305 * the notifier chains for netdev_chain and sends a NEWLINK message
1306 * to the routing socket.
1308 void netdev_state_change(struct net_device *dev)
1310 if (dev->flags & IFF_UP) {
1311 struct netdev_notifier_change_info change_info;
1313 change_info.flags_changed = 0;
1314 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1316 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1319 EXPORT_SYMBOL(netdev_state_change);
1322 * netdev_notify_peers - notify network peers about existence of @dev
1323 * @dev: network device
1325 * Generate traffic such that interested network peers are aware of
1326 * @dev, such as by generating a gratuitous ARP. This may be used when
1327 * a device wants to inform the rest of the network about some sort of
1328 * reconfiguration such as a failover event or virtual machine
1331 void netdev_notify_peers(struct net_device *dev)
1334 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1335 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1338 EXPORT_SYMBOL(netdev_notify_peers);
1340 static int __dev_open(struct net_device *dev)
1342 const struct net_device_ops *ops = dev->netdev_ops;
1347 if (!netif_device_present(dev))
1350 /* Block netpoll from trying to do any rx path servicing.
1351 * If we don't do this there is a chance ndo_poll_controller
1352 * or ndo_poll may be running while we open the device
1354 netpoll_poll_disable(dev);
1356 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1357 ret = notifier_to_errno(ret);
1361 set_bit(__LINK_STATE_START, &dev->state);
1363 if (ops->ndo_validate_addr)
1364 ret = ops->ndo_validate_addr(dev);
1366 if (!ret && ops->ndo_open)
1367 ret = ops->ndo_open(dev);
1369 netpoll_poll_enable(dev);
1372 clear_bit(__LINK_STATE_START, &dev->state);
1374 dev->flags |= IFF_UP;
1375 dev_set_rx_mode(dev);
1377 add_device_randomness(dev->dev_addr, dev->addr_len);
1384 * dev_open - prepare an interface for use.
1385 * @dev: device to open
1387 * Takes a device from down to up state. The device's private open
1388 * function is invoked and then the multicast lists are loaded. Finally
1389 * the device is moved into the up state and a %NETDEV_UP message is
1390 * sent to the netdev notifier chain.
1392 * Calling this function on an active interface is a nop. On a failure
1393 * a negative errno code is returned.
1395 int dev_open(struct net_device *dev)
1399 if (dev->flags & IFF_UP)
1402 ret = __dev_open(dev);
1406 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1407 call_netdevice_notifiers(NETDEV_UP, dev);
1411 EXPORT_SYMBOL(dev_open);
1413 static void __dev_close_many(struct list_head *head)
1415 struct net_device *dev;
1420 list_for_each_entry(dev, head, close_list) {
1421 /* Temporarily disable netpoll until the interface is down */
1422 netpoll_poll_disable(dev);
1424 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1426 clear_bit(__LINK_STATE_START, &dev->state);
1428 /* Synchronize to scheduled poll. We cannot touch poll list, it
1429 * can be even on different cpu. So just clear netif_running().
1431 * dev->stop() will invoke napi_disable() on all of it's
1432 * napi_struct instances on this device.
1434 smp_mb__after_atomic(); /* Commit netif_running(). */
1437 dev_deactivate_many(head);
1439 list_for_each_entry(dev, head, close_list) {
1440 const struct net_device_ops *ops = dev->netdev_ops;
1443 * Call the device specific close. This cannot fail.
1444 * Only if device is UP
1446 * We allow it to be called even after a DETACH hot-plug
1452 dev->flags &= ~IFF_UP;
1453 netpoll_poll_enable(dev);
1457 static void __dev_close(struct net_device *dev)
1461 list_add(&dev->close_list, &single);
1462 __dev_close_many(&single);
1466 void dev_close_many(struct list_head *head, bool unlink)
1468 struct net_device *dev, *tmp;
1470 /* Remove the devices that don't need to be closed */
1471 list_for_each_entry_safe(dev, tmp, head, close_list)
1472 if (!(dev->flags & IFF_UP))
1473 list_del_init(&dev->close_list);
1475 __dev_close_many(head);
1477 list_for_each_entry_safe(dev, tmp, head, close_list) {
1478 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1479 call_netdevice_notifiers(NETDEV_DOWN, dev);
1481 list_del_init(&dev->close_list);
1484 EXPORT_SYMBOL(dev_close_many);
1487 * dev_close - shutdown an interface.
1488 * @dev: device to shutdown
1490 * This function moves an active device into down state. A
1491 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1492 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1495 void dev_close(struct net_device *dev)
1497 if (dev->flags & IFF_UP) {
1500 list_add(&dev->close_list, &single);
1501 dev_close_many(&single, true);
1505 EXPORT_SYMBOL(dev_close);
1509 * dev_disable_lro - disable Large Receive Offload on a device
1512 * Disable Large Receive Offload (LRO) on a net device. Must be
1513 * called under RTNL. This is needed if received packets may be
1514 * forwarded to another interface.
1516 void dev_disable_lro(struct net_device *dev)
1518 struct net_device *lower_dev;
1519 struct list_head *iter;
1521 dev->wanted_features &= ~NETIF_F_LRO;
1522 netdev_update_features(dev);
1524 if (unlikely(dev->features & NETIF_F_LRO))
1525 netdev_WARN(dev, "failed to disable LRO!\n");
1527 netdev_for_each_lower_dev(dev, lower_dev, iter)
1528 dev_disable_lro(lower_dev);
1530 EXPORT_SYMBOL(dev_disable_lro);
1532 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1533 struct net_device *dev)
1535 struct netdev_notifier_info info;
1537 netdev_notifier_info_init(&info, dev);
1538 return nb->notifier_call(nb, val, &info);
1541 static int dev_boot_phase = 1;
1544 * register_netdevice_notifier - register a network notifier block
1547 * Register a notifier to be called when network device events occur.
1548 * The notifier passed is linked into the kernel structures and must
1549 * not be reused until it has been unregistered. A negative errno code
1550 * is returned on a failure.
1552 * When registered all registration and up events are replayed
1553 * to the new notifier to allow device to have a race free
1554 * view of the network device list.
1557 int register_netdevice_notifier(struct notifier_block *nb)
1559 struct net_device *dev;
1560 struct net_device *last;
1565 err = raw_notifier_chain_register(&netdev_chain, nb);
1571 for_each_netdev(net, dev) {
1572 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1573 err = notifier_to_errno(err);
1577 if (!(dev->flags & IFF_UP))
1580 call_netdevice_notifier(nb, NETDEV_UP, dev);
1591 for_each_netdev(net, dev) {
1595 if (dev->flags & IFF_UP) {
1596 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1598 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1600 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1605 raw_notifier_chain_unregister(&netdev_chain, nb);
1608 EXPORT_SYMBOL(register_netdevice_notifier);
1611 * unregister_netdevice_notifier - unregister a network notifier block
1614 * Unregister a notifier previously registered by
1615 * register_netdevice_notifier(). The notifier is unlinked into the
1616 * kernel structures and may then be reused. A negative errno code
1617 * is returned on a failure.
1619 * After unregistering unregister and down device events are synthesized
1620 * for all devices on the device list to the removed notifier to remove
1621 * the need for special case cleanup code.
1624 int unregister_netdevice_notifier(struct notifier_block *nb)
1626 struct net_device *dev;
1631 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1636 for_each_netdev(net, dev) {
1637 if (dev->flags & IFF_UP) {
1638 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1640 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1642 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1649 EXPORT_SYMBOL(unregister_netdevice_notifier);
1652 * call_netdevice_notifiers_info - call all network notifier blocks
1653 * @val: value passed unmodified to notifier function
1654 * @dev: net_device pointer passed unmodified to notifier function
1655 * @info: notifier information data
1657 * Call all network notifier blocks. Parameters and return value
1658 * are as for raw_notifier_call_chain().
1661 static int call_netdevice_notifiers_info(unsigned long val,
1662 struct net_device *dev,
1663 struct netdev_notifier_info *info)
1666 netdev_notifier_info_init(info, dev);
1667 return raw_notifier_call_chain(&netdev_chain, val, info);
1671 * call_netdevice_notifiers - call all network notifier blocks
1672 * @val: value passed unmodified to notifier function
1673 * @dev: net_device pointer passed unmodified to notifier function
1675 * Call all network notifier blocks. Parameters and return value
1676 * are as for raw_notifier_call_chain().
1679 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1681 struct netdev_notifier_info info;
1683 return call_netdevice_notifiers_info(val, dev, &info);
1685 EXPORT_SYMBOL(call_netdevice_notifiers);
1688 * call_netdevice_notifiers_mtu - call all network notifier blocks
1689 * @val: value passed unmodified to notifier function
1690 * @dev: net_device pointer passed unmodified to notifier function
1691 * @arg: additional u32 argument passed to the notifier function
1693 * Call all network notifier blocks. Parameters and return value
1694 * are as for raw_notifier_call_chain().
1696 static int call_netdevice_notifiers_mtu(unsigned long val,
1697 struct net_device *dev, u32 arg)
1699 struct netdev_notifier_info_ext info = {
1704 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1706 return call_netdevice_notifiers_info(val, dev, &info.info);
1709 #ifdef CONFIG_NET_INGRESS
1710 static struct static_key ingress_needed __read_mostly;
1712 void net_inc_ingress_queue(void)
1714 static_key_slow_inc(&ingress_needed);
1716 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1718 void net_dec_ingress_queue(void)
1720 static_key_slow_dec(&ingress_needed);
1722 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1725 #ifdef CONFIG_NET_EGRESS
1726 static struct static_key egress_needed __read_mostly;
1728 void net_inc_egress_queue(void)
1730 static_key_slow_inc(&egress_needed);
1732 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1734 void net_dec_egress_queue(void)
1736 static_key_slow_dec(&egress_needed);
1738 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1741 static struct static_key netstamp_needed __read_mostly;
1742 #ifdef HAVE_JUMP_LABEL
1743 static atomic_t netstamp_needed_deferred;
1744 static atomic_t netstamp_wanted;
1745 static void netstamp_clear(struct work_struct *work)
1747 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1750 wanted = atomic_add_return(deferred, &netstamp_wanted);
1752 static_key_enable(&netstamp_needed);
1754 static_key_disable(&netstamp_needed);
1756 static DECLARE_WORK(netstamp_work, netstamp_clear);
1759 void net_enable_timestamp(void)
1761 #ifdef HAVE_JUMP_LABEL
1765 wanted = atomic_read(&netstamp_wanted);
1768 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1771 atomic_inc(&netstamp_needed_deferred);
1772 schedule_work(&netstamp_work);
1774 static_key_slow_inc(&netstamp_needed);
1777 EXPORT_SYMBOL(net_enable_timestamp);
1779 void net_disable_timestamp(void)
1781 #ifdef HAVE_JUMP_LABEL
1785 wanted = atomic_read(&netstamp_wanted);
1788 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1791 atomic_dec(&netstamp_needed_deferred);
1792 schedule_work(&netstamp_work);
1794 static_key_slow_dec(&netstamp_needed);
1797 EXPORT_SYMBOL(net_disable_timestamp);
1799 static inline void net_timestamp_set(struct sk_buff *skb)
1802 if (static_key_false(&netstamp_needed))
1803 __net_timestamp(skb);
1806 #define net_timestamp_check(COND, SKB) \
1807 if (static_key_false(&netstamp_needed)) { \
1808 if ((COND) && !(SKB)->tstamp) \
1809 __net_timestamp(SKB); \
1812 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1816 if (!(dev->flags & IFF_UP))
1819 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1820 if (skb->len <= len)
1823 /* if TSO is enabled, we don't care about the length as the packet
1824 * could be forwarded without being segmented before
1826 if (skb_is_gso(skb))
1831 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1833 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1835 int ret = ____dev_forward_skb(dev, skb);
1838 skb->protocol = eth_type_trans(skb, dev);
1839 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1844 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1847 * dev_forward_skb - loopback an skb to another netif
1849 * @dev: destination network device
1850 * @skb: buffer to forward
1853 * NET_RX_SUCCESS (no congestion)
1854 * NET_RX_DROP (packet was dropped, but freed)
1856 * dev_forward_skb can be used for injecting an skb from the
1857 * start_xmit function of one device into the receive queue
1858 * of another device.
1860 * The receiving device may be in another namespace, so
1861 * we have to clear all information in the skb that could
1862 * impact namespace isolation.
1864 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1866 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1868 EXPORT_SYMBOL_GPL(dev_forward_skb);
1870 static inline int deliver_skb(struct sk_buff *skb,
1871 struct packet_type *pt_prev,
1872 struct net_device *orig_dev)
1874 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1876 refcount_inc(&skb->users);
1877 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1880 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1881 struct packet_type **pt,
1882 struct net_device *orig_dev,
1884 struct list_head *ptype_list)
1886 struct packet_type *ptype, *pt_prev = *pt;
1888 list_for_each_entry_rcu(ptype, ptype_list, list) {
1889 if (ptype->type != type)
1892 deliver_skb(skb, pt_prev, orig_dev);
1898 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1900 if (!ptype->af_packet_priv || !skb->sk)
1903 if (ptype->id_match)
1904 return ptype->id_match(ptype, skb->sk);
1905 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1912 * Support routine. Sends outgoing frames to any network
1913 * taps currently in use.
1916 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1918 struct packet_type *ptype;
1919 struct sk_buff *skb2 = NULL;
1920 struct packet_type *pt_prev = NULL;
1921 struct list_head *ptype_list = &ptype_all;
1925 list_for_each_entry_rcu(ptype, ptype_list, list) {
1926 /* Never send packets back to the socket
1927 * they originated from - MvS (miquels@drinkel.ow.org)
1929 if (skb_loop_sk(ptype, skb))
1933 deliver_skb(skb2, pt_prev, skb->dev);
1938 /* need to clone skb, done only once */
1939 skb2 = skb_clone(skb, GFP_ATOMIC);
1943 net_timestamp_set(skb2);
1945 /* skb->nh should be correctly
1946 * set by sender, so that the second statement is
1947 * just protection against buggy protocols.
1949 skb_reset_mac_header(skb2);
1951 if (skb_network_header(skb2) < skb2->data ||
1952 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1953 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1954 ntohs(skb2->protocol),
1956 skb_reset_network_header(skb2);
1959 skb2->transport_header = skb2->network_header;
1960 skb2->pkt_type = PACKET_OUTGOING;
1964 if (ptype_list == &ptype_all) {
1965 ptype_list = &dev->ptype_all;
1970 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
1971 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1977 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1980 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1981 * @dev: Network device
1982 * @txq: number of queues available
1984 * If real_num_tx_queues is changed the tc mappings may no longer be
1985 * valid. To resolve this verify the tc mapping remains valid and if
1986 * not NULL the mapping. With no priorities mapping to this
1987 * offset/count pair it will no longer be used. In the worst case TC0
1988 * is invalid nothing can be done so disable priority mappings. If is
1989 * expected that drivers will fix this mapping if they can before
1990 * calling netif_set_real_num_tx_queues.
1992 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1995 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1997 /* If TC0 is invalidated disable TC mapping */
1998 if (tc->offset + tc->count > txq) {
1999 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2004 /* Invalidated prio to tc mappings set to TC0 */
2005 for (i = 1; i < TC_BITMASK + 1; i++) {
2006 int q = netdev_get_prio_tc_map(dev, i);
2008 tc = &dev->tc_to_txq[q];
2009 if (tc->offset + tc->count > txq) {
2010 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2012 netdev_set_prio_tc_map(dev, i, 0);
2017 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2020 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2023 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2024 if ((txq - tc->offset) < tc->count)
2035 static DEFINE_MUTEX(xps_map_mutex);
2036 #define xmap_dereference(P) \
2037 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2039 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2042 struct xps_map *map = NULL;
2046 map = xmap_dereference(dev_maps->cpu_map[tci]);
2050 for (pos = map->len; pos--;) {
2051 if (map->queues[pos] != index)
2055 map->queues[pos] = map->queues[--map->len];
2059 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
2060 kfree_rcu(map, rcu);
2067 static bool remove_xps_queue_cpu(struct net_device *dev,
2068 struct xps_dev_maps *dev_maps,
2069 int cpu, u16 offset, u16 count)
2071 int num_tc = dev->num_tc ? : 1;
2072 bool active = false;
2075 for (tci = cpu * num_tc; num_tc--; tci++) {
2078 for (i = count, j = offset; i--; j++) {
2079 if (!remove_xps_queue(dev_maps, tci, j))
2089 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2092 struct xps_dev_maps *dev_maps;
2094 bool active = false;
2096 mutex_lock(&xps_map_mutex);
2097 dev_maps = xmap_dereference(dev->xps_maps);
2102 for_each_possible_cpu(cpu)
2103 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2107 RCU_INIT_POINTER(dev->xps_maps, NULL);
2108 kfree_rcu(dev_maps, rcu);
2111 for (i = offset + (count - 1); count--; i--)
2112 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2116 mutex_unlock(&xps_map_mutex);
2119 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2121 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2124 static struct xps_map *expand_xps_map(struct xps_map *map,
2127 struct xps_map *new_map;
2128 int alloc_len = XPS_MIN_MAP_ALLOC;
2131 for (pos = 0; map && pos < map->len; pos++) {
2132 if (map->queues[pos] != index)
2137 /* Need to add queue to this CPU's existing map */
2139 if (pos < map->alloc_len)
2142 alloc_len = map->alloc_len * 2;
2145 /* Need to allocate new map to store queue on this CPU's map */
2146 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2151 for (i = 0; i < pos; i++)
2152 new_map->queues[i] = map->queues[i];
2153 new_map->alloc_len = alloc_len;
2159 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2162 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2163 int i, cpu, tci, numa_node_id = -2;
2164 int maps_sz, num_tc = 1, tc = 0;
2165 struct xps_map *map, *new_map;
2166 bool active = false;
2169 num_tc = dev->num_tc;
2170 tc = netdev_txq_to_tc(dev, index);
2175 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2176 if (maps_sz < L1_CACHE_BYTES)
2177 maps_sz = L1_CACHE_BYTES;
2179 mutex_lock(&xps_map_mutex);
2181 dev_maps = xmap_dereference(dev->xps_maps);
2183 /* allocate memory for queue storage */
2184 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2186 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2187 if (!new_dev_maps) {
2188 mutex_unlock(&xps_map_mutex);
2192 tci = cpu * num_tc + tc;
2193 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2196 map = expand_xps_map(map, cpu, index);
2200 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2204 goto out_no_new_maps;
2206 for_each_possible_cpu(cpu) {
2207 /* copy maps belonging to foreign traffic classes */
2208 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2209 /* fill in the new device map from the old device map */
2210 map = xmap_dereference(dev_maps->cpu_map[tci]);
2211 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2214 /* We need to explicitly update tci as prevous loop
2215 * could break out early if dev_maps is NULL.
2217 tci = cpu * num_tc + tc;
2219 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2220 /* add queue to CPU maps */
2223 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2224 while ((pos < map->len) && (map->queues[pos] != index))
2227 if (pos == map->len)
2228 map->queues[map->len++] = index;
2230 if (numa_node_id == -2)
2231 numa_node_id = cpu_to_node(cpu);
2232 else if (numa_node_id != cpu_to_node(cpu))
2235 } else if (dev_maps) {
2236 /* fill in the new device map from the old device map */
2237 map = xmap_dereference(dev_maps->cpu_map[tci]);
2238 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2241 /* copy maps belonging to foreign traffic classes */
2242 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2243 /* fill in the new device map from the old device map */
2244 map = xmap_dereference(dev_maps->cpu_map[tci]);
2245 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2249 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2251 /* Cleanup old maps */
2253 goto out_no_old_maps;
2255 for_each_possible_cpu(cpu) {
2256 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2257 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2258 map = xmap_dereference(dev_maps->cpu_map[tci]);
2259 if (map && map != new_map)
2260 kfree_rcu(map, rcu);
2264 kfree_rcu(dev_maps, rcu);
2267 dev_maps = new_dev_maps;
2271 /* update Tx queue numa node */
2272 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2273 (numa_node_id >= 0) ? numa_node_id :
2279 /* removes queue from unused CPUs */
2280 for_each_possible_cpu(cpu) {
2281 for (i = tc, tci = cpu * num_tc; i--; tci++)
2282 active |= remove_xps_queue(dev_maps, tci, index);
2283 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2284 active |= remove_xps_queue(dev_maps, tci, index);
2285 for (i = num_tc - tc, tci++; --i; tci++)
2286 active |= remove_xps_queue(dev_maps, tci, index);
2289 /* free map if not active */
2291 RCU_INIT_POINTER(dev->xps_maps, NULL);
2292 kfree_rcu(dev_maps, rcu);
2296 mutex_unlock(&xps_map_mutex);
2300 /* remove any maps that we added */
2301 for_each_possible_cpu(cpu) {
2302 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2303 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2305 xmap_dereference(dev_maps->cpu_map[tci]) :
2307 if (new_map && new_map != map)
2312 mutex_unlock(&xps_map_mutex);
2314 kfree(new_dev_maps);
2317 EXPORT_SYMBOL(netif_set_xps_queue);
2320 void netdev_reset_tc(struct net_device *dev)
2323 netif_reset_xps_queues_gt(dev, 0);
2326 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2327 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2329 EXPORT_SYMBOL(netdev_reset_tc);
2331 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2333 if (tc >= dev->num_tc)
2337 netif_reset_xps_queues(dev, offset, count);
2339 dev->tc_to_txq[tc].count = count;
2340 dev->tc_to_txq[tc].offset = offset;
2343 EXPORT_SYMBOL(netdev_set_tc_queue);
2345 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2347 if (num_tc > TC_MAX_QUEUE)
2351 netif_reset_xps_queues_gt(dev, 0);
2353 dev->num_tc = num_tc;
2356 EXPORT_SYMBOL(netdev_set_num_tc);
2359 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2360 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2362 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2367 disabling = txq < dev->real_num_tx_queues;
2369 if (txq < 1 || txq > dev->num_tx_queues)
2372 if (dev->reg_state == NETREG_REGISTERED ||
2373 dev->reg_state == NETREG_UNREGISTERING) {
2376 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2382 netif_setup_tc(dev, txq);
2384 dev->real_num_tx_queues = txq;
2388 qdisc_reset_all_tx_gt(dev, txq);
2390 netif_reset_xps_queues_gt(dev, txq);
2394 dev->real_num_tx_queues = txq;
2399 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2403 * netif_set_real_num_rx_queues - set actual number of RX queues used
2404 * @dev: Network device
2405 * @rxq: Actual number of RX queues
2407 * This must be called either with the rtnl_lock held or before
2408 * registration of the net device. Returns 0 on success, or a
2409 * negative error code. If called before registration, it always
2412 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2416 if (rxq < 1 || rxq > dev->num_rx_queues)
2419 if (dev->reg_state == NETREG_REGISTERED) {
2422 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2428 dev->real_num_rx_queues = rxq;
2431 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2435 * netif_get_num_default_rss_queues - default number of RSS queues
2437 * This routine should set an upper limit on the number of RSS queues
2438 * used by default by multiqueue devices.
2440 int netif_get_num_default_rss_queues(void)
2442 return is_kdump_kernel() ?
2443 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2445 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2447 static void __netif_reschedule(struct Qdisc *q)
2449 struct softnet_data *sd;
2450 unsigned long flags;
2452 local_irq_save(flags);
2453 sd = this_cpu_ptr(&softnet_data);
2454 q->next_sched = NULL;
2455 *sd->output_queue_tailp = q;
2456 sd->output_queue_tailp = &q->next_sched;
2457 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2458 local_irq_restore(flags);
2461 void __netif_schedule(struct Qdisc *q)
2463 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2464 __netif_reschedule(q);
2466 EXPORT_SYMBOL(__netif_schedule);
2468 struct dev_kfree_skb_cb {
2469 enum skb_free_reason reason;
2472 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2474 return (struct dev_kfree_skb_cb *)skb->cb;
2477 void netif_schedule_queue(struct netdev_queue *txq)
2480 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2481 struct Qdisc *q = rcu_dereference(txq->qdisc);
2483 __netif_schedule(q);
2487 EXPORT_SYMBOL(netif_schedule_queue);
2489 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2491 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2495 q = rcu_dereference(dev_queue->qdisc);
2496 __netif_schedule(q);
2500 EXPORT_SYMBOL(netif_tx_wake_queue);
2502 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2504 unsigned long flags;
2509 if (likely(refcount_read(&skb->users) == 1)) {
2511 refcount_set(&skb->users, 0);
2512 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2515 get_kfree_skb_cb(skb)->reason = reason;
2516 local_irq_save(flags);
2517 skb->next = __this_cpu_read(softnet_data.completion_queue);
2518 __this_cpu_write(softnet_data.completion_queue, skb);
2519 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2520 local_irq_restore(flags);
2522 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2524 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2526 if (in_irq() || irqs_disabled())
2527 __dev_kfree_skb_irq(skb, reason);
2531 EXPORT_SYMBOL(__dev_kfree_skb_any);
2535 * netif_device_detach - mark device as removed
2536 * @dev: network device
2538 * Mark device as removed from system and therefore no longer available.
2540 void netif_device_detach(struct net_device *dev)
2542 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2543 netif_running(dev)) {
2544 netif_tx_stop_all_queues(dev);
2547 EXPORT_SYMBOL(netif_device_detach);
2550 * netif_device_attach - mark device as attached
2551 * @dev: network device
2553 * Mark device as attached from system and restart if needed.
2555 void netif_device_attach(struct net_device *dev)
2557 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2558 netif_running(dev)) {
2559 netif_tx_wake_all_queues(dev);
2560 __netdev_watchdog_up(dev);
2563 EXPORT_SYMBOL(netif_device_attach);
2566 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2567 * to be used as a distribution range.
2569 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2570 unsigned int num_tx_queues)
2574 u16 qcount = num_tx_queues;
2576 if (skb_rx_queue_recorded(skb)) {
2577 hash = skb_get_rx_queue(skb);
2578 while (unlikely(hash >= num_tx_queues))
2579 hash -= num_tx_queues;
2584 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2586 qoffset = dev->tc_to_txq[tc].offset;
2587 qcount = dev->tc_to_txq[tc].count;
2590 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2592 EXPORT_SYMBOL(__skb_tx_hash);
2594 static void skb_warn_bad_offload(const struct sk_buff *skb)
2596 static const netdev_features_t null_features;
2597 struct net_device *dev = skb->dev;
2598 const char *name = "";
2600 if (!net_ratelimit())
2604 if (dev->dev.parent)
2605 name = dev_driver_string(dev->dev.parent);
2607 name = netdev_name(dev);
2609 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2610 "gso_type=%d ip_summed=%d\n",
2611 name, dev ? &dev->features : &null_features,
2612 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2613 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2614 skb_shinfo(skb)->gso_type, skb->ip_summed);
2618 * Invalidate hardware checksum when packet is to be mangled, and
2619 * complete checksum manually on outgoing path.
2621 int skb_checksum_help(struct sk_buff *skb)
2624 int ret = 0, offset;
2626 if (skb->ip_summed == CHECKSUM_COMPLETE)
2627 goto out_set_summed;
2629 if (unlikely(skb_shinfo(skb)->gso_size)) {
2630 skb_warn_bad_offload(skb);
2634 /* Before computing a checksum, we should make sure no frag could
2635 * be modified by an external entity : checksum could be wrong.
2637 if (skb_has_shared_frag(skb)) {
2638 ret = __skb_linearize(skb);
2643 offset = skb_checksum_start_offset(skb);
2644 BUG_ON(offset >= skb_headlen(skb));
2645 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2647 offset += skb->csum_offset;
2648 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2650 if (skb_cloned(skb) &&
2651 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2652 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2657 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2659 skb->ip_summed = CHECKSUM_NONE;
2663 EXPORT_SYMBOL(skb_checksum_help);
2665 int skb_crc32c_csum_help(struct sk_buff *skb)
2668 int ret = 0, offset, start;
2670 if (skb->ip_summed != CHECKSUM_PARTIAL)
2673 if (unlikely(skb_is_gso(skb)))
2676 /* Before computing a checksum, we should make sure no frag could
2677 * be modified by an external entity : checksum could be wrong.
2679 if (unlikely(skb_has_shared_frag(skb))) {
2680 ret = __skb_linearize(skb);
2684 start = skb_checksum_start_offset(skb);
2685 offset = start + offsetof(struct sctphdr, checksum);
2686 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2690 if (skb_cloned(skb) &&
2691 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2692 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2696 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2697 skb->len - start, ~(__u32)0,
2699 *(__le32 *)(skb->data + offset) = crc32c_csum;
2700 skb->ip_summed = CHECKSUM_NONE;
2701 skb->csum_not_inet = 0;
2706 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2708 __be16 type = skb->protocol;
2710 /* Tunnel gso handlers can set protocol to ethernet. */
2711 if (type == htons(ETH_P_TEB)) {
2714 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2717 eth = (struct ethhdr *)skb->data;
2718 type = eth->h_proto;
2721 return __vlan_get_protocol(skb, type, depth);
2725 * skb_mac_gso_segment - mac layer segmentation handler.
2726 * @skb: buffer to segment
2727 * @features: features for the output path (see dev->features)
2729 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2730 netdev_features_t features)
2732 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2733 struct packet_offload *ptype;
2734 int vlan_depth = skb->mac_len;
2735 __be16 type = skb_network_protocol(skb, &vlan_depth);
2737 if (unlikely(!type))
2738 return ERR_PTR(-EINVAL);
2740 __skb_pull(skb, vlan_depth);
2743 list_for_each_entry_rcu(ptype, &offload_base, list) {
2744 if (ptype->type == type && ptype->callbacks.gso_segment) {
2745 segs = ptype->callbacks.gso_segment(skb, features);
2751 __skb_push(skb, skb->data - skb_mac_header(skb));
2755 EXPORT_SYMBOL(skb_mac_gso_segment);
2758 /* openvswitch calls this on rx path, so we need a different check.
2760 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2763 return skb->ip_summed != CHECKSUM_PARTIAL &&
2764 skb->ip_summed != CHECKSUM_UNNECESSARY;
2766 return skb->ip_summed == CHECKSUM_NONE;
2770 * __skb_gso_segment - Perform segmentation on skb.
2771 * @skb: buffer to segment
2772 * @features: features for the output path (see dev->features)
2773 * @tx_path: whether it is called in TX path
2775 * This function segments the given skb and returns a list of segments.
2777 * It may return NULL if the skb requires no segmentation. This is
2778 * only possible when GSO is used for verifying header integrity.
2780 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2782 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2783 netdev_features_t features, bool tx_path)
2785 struct sk_buff *segs;
2787 if (unlikely(skb_needs_check(skb, tx_path))) {
2790 /* We're going to init ->check field in TCP or UDP header */
2791 err = skb_cow_head(skb, 0);
2793 return ERR_PTR(err);
2796 /* Only report GSO partial support if it will enable us to
2797 * support segmentation on this frame without needing additional
2800 if (features & NETIF_F_GSO_PARTIAL) {
2801 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2802 struct net_device *dev = skb->dev;
2804 partial_features |= dev->features & dev->gso_partial_features;
2805 if (!skb_gso_ok(skb, features | partial_features))
2806 features &= ~NETIF_F_GSO_PARTIAL;
2809 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2810 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2812 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2813 SKB_GSO_CB(skb)->encap_level = 0;
2815 skb_reset_mac_header(skb);
2816 skb_reset_mac_len(skb);
2818 segs = skb_mac_gso_segment(skb, features);
2820 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
2821 skb_warn_bad_offload(skb);
2825 EXPORT_SYMBOL(__skb_gso_segment);
2827 /* Take action when hardware reception checksum errors are detected. */
2829 void netdev_rx_csum_fault(struct net_device *dev)
2831 if (net_ratelimit()) {
2832 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2836 EXPORT_SYMBOL(netdev_rx_csum_fault);
2839 /* Actually, we should eliminate this check as soon as we know, that:
2840 * 1. IOMMU is present and allows to map all the memory.
2841 * 2. No high memory really exists on this machine.
2844 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2846 #ifdef CONFIG_HIGHMEM
2849 if (!(dev->features & NETIF_F_HIGHDMA)) {
2850 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2851 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2853 if (PageHighMem(skb_frag_page(frag)))
2858 if (PCI_DMA_BUS_IS_PHYS) {
2859 struct device *pdev = dev->dev.parent;
2863 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2864 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2865 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2867 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2875 /* If MPLS offload request, verify we are testing hardware MPLS features
2876 * instead of standard features for the netdev.
2878 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2879 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2880 netdev_features_t features,
2883 if (eth_p_mpls(type))
2884 features &= skb->dev->mpls_features;
2889 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2890 netdev_features_t features,
2897 static netdev_features_t harmonize_features(struct sk_buff *skb,
2898 netdev_features_t features)
2903 type = skb_network_protocol(skb, &tmp);
2904 features = net_mpls_features(skb, features, type);
2906 if (skb->ip_summed != CHECKSUM_NONE &&
2907 !can_checksum_protocol(features, type)) {
2908 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2910 if (illegal_highdma(skb->dev, skb))
2911 features &= ~NETIF_F_SG;
2916 netdev_features_t passthru_features_check(struct sk_buff *skb,
2917 struct net_device *dev,
2918 netdev_features_t features)
2922 EXPORT_SYMBOL(passthru_features_check);
2924 static netdev_features_t dflt_features_check(struct sk_buff *skb,
2925 struct net_device *dev,
2926 netdev_features_t features)
2928 return vlan_features_check(skb, features);
2931 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2932 struct net_device *dev,
2933 netdev_features_t features)
2935 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2937 if (gso_segs > dev->gso_max_segs)
2938 return features & ~NETIF_F_GSO_MASK;
2940 /* Support for GSO partial features requires software
2941 * intervention before we can actually process the packets
2942 * so we need to strip support for any partial features now
2943 * and we can pull them back in after we have partially
2944 * segmented the frame.
2946 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2947 features &= ~dev->gso_partial_features;
2949 /* Make sure to clear the IPv4 ID mangling feature if the
2950 * IPv4 header has the potential to be fragmented.
2952 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2953 struct iphdr *iph = skb->encapsulation ?
2954 inner_ip_hdr(skb) : ip_hdr(skb);
2956 if (!(iph->frag_off & htons(IP_DF)))
2957 features &= ~NETIF_F_TSO_MANGLEID;
2963 netdev_features_t netif_skb_features(struct sk_buff *skb)
2965 struct net_device *dev = skb->dev;
2966 netdev_features_t features = dev->features;
2968 if (skb_is_gso(skb))
2969 features = gso_features_check(skb, dev, features);
2971 /* If encapsulation offload request, verify we are testing
2972 * hardware encapsulation features instead of standard
2973 * features for the netdev
2975 if (skb->encapsulation)
2976 features &= dev->hw_enc_features;
2978 if (skb_vlan_tagged(skb))
2979 features = netdev_intersect_features(features,
2980 dev->vlan_features |
2981 NETIF_F_HW_VLAN_CTAG_TX |
2982 NETIF_F_HW_VLAN_STAG_TX);
2984 if (dev->netdev_ops->ndo_features_check)
2985 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2988 features &= dflt_features_check(skb, dev, features);
2990 return harmonize_features(skb, features);
2992 EXPORT_SYMBOL(netif_skb_features);
2994 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2995 struct netdev_queue *txq, bool more)
3000 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3001 dev_queue_xmit_nit(skb, dev);
3004 trace_net_dev_start_xmit(skb, dev);
3005 rc = netdev_start_xmit(skb, dev, txq, more);
3006 trace_net_dev_xmit(skb, rc, dev, len);
3011 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3012 struct netdev_queue *txq, int *ret)
3014 struct sk_buff *skb = first;
3015 int rc = NETDEV_TX_OK;
3018 struct sk_buff *next = skb->next;
3021 rc = xmit_one(skb, dev, txq, next != NULL);
3022 if (unlikely(!dev_xmit_complete(rc))) {
3028 if (netif_tx_queue_stopped(txq) && skb) {
3029 rc = NETDEV_TX_BUSY;
3039 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3040 netdev_features_t features)
3042 if (skb_vlan_tag_present(skb) &&
3043 !vlan_hw_offload_capable(features, skb->vlan_proto))
3044 skb = __vlan_hwaccel_push_inside(skb);
3048 int skb_csum_hwoffload_help(struct sk_buff *skb,
3049 const netdev_features_t features)
3051 if (unlikely(skb->csum_not_inet))
3052 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3053 skb_crc32c_csum_help(skb);
3055 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3057 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3059 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
3061 netdev_features_t features;
3063 features = netif_skb_features(skb);
3064 skb = validate_xmit_vlan(skb, features);
3068 if (netif_needs_gso(skb, features)) {
3069 struct sk_buff *segs;
3071 segs = skb_gso_segment(skb, features);
3079 if (skb_needs_linearize(skb, features) &&
3080 __skb_linearize(skb))
3083 if (validate_xmit_xfrm(skb, features))
3086 /* If packet is not checksummed and device does not
3087 * support checksumming for this protocol, complete
3088 * checksumming here.
3090 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3091 if (skb->encapsulation)
3092 skb_set_inner_transport_header(skb,
3093 skb_checksum_start_offset(skb));
3095 skb_set_transport_header(skb,
3096 skb_checksum_start_offset(skb));
3097 if (skb_csum_hwoffload_help(skb, features))
3107 atomic_long_inc(&dev->tx_dropped);
3111 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3113 struct sk_buff *next, *head = NULL, *tail;
3115 for (; skb != NULL; skb = next) {
3119 /* in case skb wont be segmented, point to itself */
3122 skb = validate_xmit_skb(skb, dev);
3130 /* If skb was segmented, skb->prev points to
3131 * the last segment. If not, it still contains skb.
3137 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3139 static void qdisc_pkt_len_init(struct sk_buff *skb)
3141 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3143 qdisc_skb_cb(skb)->pkt_len = skb->len;
3145 /* To get more precise estimation of bytes sent on wire,
3146 * we add to pkt_len the headers size of all segments
3148 if (shinfo->gso_size) {
3149 unsigned int hdr_len;
3150 u16 gso_segs = shinfo->gso_segs;
3152 /* mac layer + network layer */
3153 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3155 /* + transport layer */
3156 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3157 const struct tcphdr *th;
3158 struct tcphdr _tcphdr;
3160 th = skb_header_pointer(skb, skb_transport_offset(skb),
3161 sizeof(_tcphdr), &_tcphdr);
3163 hdr_len += __tcp_hdrlen(th);
3165 struct udphdr _udphdr;
3167 if (skb_header_pointer(skb, skb_transport_offset(skb),
3168 sizeof(_udphdr), &_udphdr))
3169 hdr_len += sizeof(struct udphdr);
3172 if (shinfo->gso_type & SKB_GSO_DODGY)
3173 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3176 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3180 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3181 struct net_device *dev,
3182 struct netdev_queue *txq)
3184 spinlock_t *root_lock = qdisc_lock(q);
3185 struct sk_buff *to_free = NULL;
3189 qdisc_calculate_pkt_len(skb, q);
3191 * Heuristic to force contended enqueues to serialize on a
3192 * separate lock before trying to get qdisc main lock.
3193 * This permits qdisc->running owner to get the lock more
3194 * often and dequeue packets faster.
3196 contended = qdisc_is_running(q);
3197 if (unlikely(contended))
3198 spin_lock(&q->busylock);
3200 spin_lock(root_lock);
3201 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3202 __qdisc_drop(skb, &to_free);
3204 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3205 qdisc_run_begin(q)) {
3207 * This is a work-conserving queue; there are no old skbs
3208 * waiting to be sent out; and the qdisc is not running -
3209 * xmit the skb directly.
3212 qdisc_bstats_update(q, skb);
3214 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3215 if (unlikely(contended)) {
3216 spin_unlock(&q->busylock);
3223 rc = NET_XMIT_SUCCESS;
3225 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3226 if (qdisc_run_begin(q)) {
3227 if (unlikely(contended)) {
3228 spin_unlock(&q->busylock);
3234 spin_unlock(root_lock);
3235 if (unlikely(to_free))
3236 kfree_skb_list(to_free);
3237 if (unlikely(contended))
3238 spin_unlock(&q->busylock);
3242 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3243 static void skb_update_prio(struct sk_buff *skb)
3245 const struct netprio_map *map;
3246 const struct sock *sk;
3247 unsigned int prioidx;
3251 map = rcu_dereference_bh(skb->dev->priomap);
3254 sk = skb_to_full_sk(skb);
3258 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3260 if (prioidx < map->priomap_len)
3261 skb->priority = map->priomap[prioidx];
3264 #define skb_update_prio(skb)
3267 DEFINE_PER_CPU(int, xmit_recursion);
3268 EXPORT_SYMBOL(xmit_recursion);
3271 * dev_loopback_xmit - loop back @skb
3272 * @net: network namespace this loopback is happening in
3273 * @sk: sk needed to be a netfilter okfn
3274 * @skb: buffer to transmit
3276 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3278 skb_reset_mac_header(skb);
3279 __skb_pull(skb, skb_network_offset(skb));
3280 skb->pkt_type = PACKET_LOOPBACK;
3281 skb->ip_summed = CHECKSUM_UNNECESSARY;
3282 WARN_ON(!skb_dst(skb));
3287 EXPORT_SYMBOL(dev_loopback_xmit);
3289 #ifdef CONFIG_NET_EGRESS
3290 static struct sk_buff *
3291 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3293 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3294 struct tcf_result cl_res;
3299 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3300 qdisc_bstats_cpu_update(cl->q, skb);
3302 switch (tcf_classify(skb, cl, &cl_res, false)) {
3304 case TC_ACT_RECLASSIFY:
3305 skb->tc_index = TC_H_MIN(cl_res.classid);
3308 qdisc_qstats_cpu_drop(cl->q);
3309 *ret = NET_XMIT_DROP;
3315 *ret = NET_XMIT_SUCCESS;
3318 case TC_ACT_REDIRECT:
3319 /* No need to push/pop skb's mac_header here on egress! */
3320 skb_do_redirect(skb);
3321 *ret = NET_XMIT_SUCCESS;
3329 #endif /* CONFIG_NET_EGRESS */
3331 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3334 struct xps_dev_maps *dev_maps;
3335 struct xps_map *map;
3336 int queue_index = -1;
3339 dev_maps = rcu_dereference(dev->xps_maps);
3341 unsigned int tci = skb->sender_cpu - 1;
3345 tci += netdev_get_prio_tc_map(dev, skb->priority);
3348 map = rcu_dereference(dev_maps->cpu_map[tci]);
3351 queue_index = map->queues[0];
3353 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3355 if (unlikely(queue_index >= dev->real_num_tx_queues))
3367 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3369 struct sock *sk = skb->sk;
3370 int queue_index = sk_tx_queue_get(sk);
3372 if (queue_index < 0 || skb->ooo_okay ||
3373 queue_index >= dev->real_num_tx_queues) {
3374 int new_index = get_xps_queue(dev, skb);
3377 new_index = skb_tx_hash(dev, skb);
3379 if (queue_index != new_index && sk &&
3381 rcu_access_pointer(sk->sk_dst_cache))
3382 sk_tx_queue_set(sk, new_index);
3384 queue_index = new_index;
3390 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3391 struct sk_buff *skb,
3394 int queue_index = 0;
3397 u32 sender_cpu = skb->sender_cpu - 1;
3399 if (sender_cpu >= (u32)NR_CPUS)
3400 skb->sender_cpu = raw_smp_processor_id() + 1;
3403 if (dev->real_num_tx_queues != 1) {
3404 const struct net_device_ops *ops = dev->netdev_ops;
3406 if (ops->ndo_select_queue)
3407 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3410 queue_index = __netdev_pick_tx(dev, skb);
3413 queue_index = netdev_cap_txqueue(dev, queue_index);
3416 skb_set_queue_mapping(skb, queue_index);
3417 return netdev_get_tx_queue(dev, queue_index);
3421 * __dev_queue_xmit - transmit a buffer
3422 * @skb: buffer to transmit
3423 * @accel_priv: private data used for L2 forwarding offload
3425 * Queue a buffer for transmission to a network device. The caller must
3426 * have set the device and priority and built the buffer before calling
3427 * this function. The function can be called from an interrupt.
3429 * A negative errno code is returned on a failure. A success does not
3430 * guarantee the frame will be transmitted as it may be dropped due
3431 * to congestion or traffic shaping.
3433 * -----------------------------------------------------------------------------------
3434 * I notice this method can also return errors from the queue disciplines,
3435 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3438 * Regardless of the return value, the skb is consumed, so it is currently
3439 * difficult to retry a send to this method. (You can bump the ref count
3440 * before sending to hold a reference for retry if you are careful.)
3442 * When calling this method, interrupts MUST be enabled. This is because
3443 * the BH enable code must have IRQs enabled so that it will not deadlock.
3446 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3448 struct net_device *dev = skb->dev;
3449 struct netdev_queue *txq;
3453 skb_reset_mac_header(skb);
3455 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3456 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3458 /* Disable soft irqs for various locks below. Also
3459 * stops preemption for RCU.
3463 skb_update_prio(skb);
3465 qdisc_pkt_len_init(skb);
3466 #ifdef CONFIG_NET_CLS_ACT
3467 skb->tc_at_ingress = 0;
3468 # ifdef CONFIG_NET_EGRESS
3469 if (static_key_false(&egress_needed)) {
3470 skb = sch_handle_egress(skb, &rc, dev);
3476 /* If device/qdisc don't need skb->dst, release it right now while
3477 * its hot in this cpu cache.
3479 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3484 txq = netdev_pick_tx(dev, skb, accel_priv);
3485 q = rcu_dereference_bh(txq->qdisc);
3487 trace_net_dev_queue(skb);
3489 rc = __dev_xmit_skb(skb, q, dev, txq);
3493 /* The device has no queue. Common case for software devices:
3494 * loopback, all the sorts of tunnels...
3496 * Really, it is unlikely that netif_tx_lock protection is necessary
3497 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3499 * However, it is possible, that they rely on protection
3502 * Check this and shot the lock. It is not prone from deadlocks.
3503 *Either shot noqueue qdisc, it is even simpler 8)
3505 if (dev->flags & IFF_UP) {
3506 int cpu = smp_processor_id(); /* ok because BHs are off */
3508 if (txq->xmit_lock_owner != cpu) {
3509 if (unlikely(__this_cpu_read(xmit_recursion) >
3510 XMIT_RECURSION_LIMIT))
3511 goto recursion_alert;
3513 skb = validate_xmit_skb(skb, dev);
3517 HARD_TX_LOCK(dev, txq, cpu);
3519 if (!netif_xmit_stopped(txq)) {
3520 __this_cpu_inc(xmit_recursion);
3521 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3522 __this_cpu_dec(xmit_recursion);
3523 if (dev_xmit_complete(rc)) {
3524 HARD_TX_UNLOCK(dev, txq);
3528 HARD_TX_UNLOCK(dev, txq);
3529 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3532 /* Recursion is detected! It is possible,
3536 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3542 rcu_read_unlock_bh();
3544 atomic_long_inc(&dev->tx_dropped);
3545 kfree_skb_list(skb);
3548 rcu_read_unlock_bh();
3552 int dev_queue_xmit(struct sk_buff *skb)
3554 return __dev_queue_xmit(skb, NULL);
3556 EXPORT_SYMBOL(dev_queue_xmit);
3558 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3560 return __dev_queue_xmit(skb, accel_priv);
3562 EXPORT_SYMBOL(dev_queue_xmit_accel);
3565 /*************************************************************************
3567 *************************************************************************/
3569 int netdev_max_backlog __read_mostly = 1000;
3570 EXPORT_SYMBOL(netdev_max_backlog);
3572 int netdev_tstamp_prequeue __read_mostly = 1;
3573 int netdev_budget __read_mostly = 300;
3574 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
3575 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
3576 int weight_p __read_mostly = 64; /* old backlog weight */
3577 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3578 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3579 int dev_rx_weight __read_mostly = 64;
3580 int dev_tx_weight __read_mostly = 64;
3582 /* Called with irq disabled */
3583 static inline void ____napi_schedule(struct softnet_data *sd,
3584 struct napi_struct *napi)
3586 list_add_tail(&napi->poll_list, &sd->poll_list);
3587 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3592 /* One global table that all flow-based protocols share. */
3593 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3594 EXPORT_SYMBOL(rps_sock_flow_table);
3595 u32 rps_cpu_mask __read_mostly;
3596 EXPORT_SYMBOL(rps_cpu_mask);
3598 struct static_key rps_needed __read_mostly;
3599 EXPORT_SYMBOL(rps_needed);
3600 struct static_key rfs_needed __read_mostly;
3601 EXPORT_SYMBOL(rfs_needed);
3603 static struct rps_dev_flow *
3604 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3605 struct rps_dev_flow *rflow, u16 next_cpu)
3607 if (next_cpu < nr_cpu_ids) {
3608 #ifdef CONFIG_RFS_ACCEL
3609 struct netdev_rx_queue *rxqueue;
3610 struct rps_dev_flow_table *flow_table;
3611 struct rps_dev_flow *old_rflow;
3616 /* Should we steer this flow to a different hardware queue? */
3617 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3618 !(dev->features & NETIF_F_NTUPLE))
3620 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3621 if (rxq_index == skb_get_rx_queue(skb))
3624 rxqueue = dev->_rx + rxq_index;
3625 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3628 flow_id = skb_get_hash(skb) & flow_table->mask;
3629 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3630 rxq_index, flow_id);
3634 rflow = &flow_table->flows[flow_id];
3636 if (old_rflow->filter == rflow->filter)
3637 old_rflow->filter = RPS_NO_FILTER;
3641 per_cpu(softnet_data, next_cpu).input_queue_head;
3644 rflow->cpu = next_cpu;
3649 * get_rps_cpu is called from netif_receive_skb and returns the target
3650 * CPU from the RPS map of the receiving queue for a given skb.
3651 * rcu_read_lock must be held on entry.
3653 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3654 struct rps_dev_flow **rflowp)
3656 const struct rps_sock_flow_table *sock_flow_table;
3657 struct netdev_rx_queue *rxqueue = dev->_rx;
3658 struct rps_dev_flow_table *flow_table;
3659 struct rps_map *map;
3664 if (skb_rx_queue_recorded(skb)) {
3665 u16 index = skb_get_rx_queue(skb);
3667 if (unlikely(index >= dev->real_num_rx_queues)) {
3668 WARN_ONCE(dev->real_num_rx_queues > 1,
3669 "%s received packet on queue %u, but number "
3670 "of RX queues is %u\n",
3671 dev->name, index, dev->real_num_rx_queues);
3677 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3679 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3680 map = rcu_dereference(rxqueue->rps_map);
3681 if (!flow_table && !map)
3684 skb_reset_network_header(skb);
3685 hash = skb_get_hash(skb);
3689 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3690 if (flow_table && sock_flow_table) {
3691 struct rps_dev_flow *rflow;
3695 /* First check into global flow table if there is a match */
3696 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3697 if ((ident ^ hash) & ~rps_cpu_mask)
3700 next_cpu = ident & rps_cpu_mask;
3702 /* OK, now we know there is a match,
3703 * we can look at the local (per receive queue) flow table
3705 rflow = &flow_table->flows[hash & flow_table->mask];
3709 * If the desired CPU (where last recvmsg was done) is
3710 * different from current CPU (one in the rx-queue flow
3711 * table entry), switch if one of the following holds:
3712 * - Current CPU is unset (>= nr_cpu_ids).
3713 * - Current CPU is offline.
3714 * - The current CPU's queue tail has advanced beyond the
3715 * last packet that was enqueued using this table entry.
3716 * This guarantees that all previous packets for the flow
3717 * have been dequeued, thus preserving in order delivery.
3719 if (unlikely(tcpu != next_cpu) &&
3720 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3721 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3722 rflow->last_qtail)) >= 0)) {
3724 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3727 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3737 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3738 if (cpu_online(tcpu)) {
3748 #ifdef CONFIG_RFS_ACCEL
3751 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3752 * @dev: Device on which the filter was set
3753 * @rxq_index: RX queue index
3754 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3755 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3757 * Drivers that implement ndo_rx_flow_steer() should periodically call
3758 * this function for each installed filter and remove the filters for
3759 * which it returns %true.
3761 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3762 u32 flow_id, u16 filter_id)
3764 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3765 struct rps_dev_flow_table *flow_table;
3766 struct rps_dev_flow *rflow;
3771 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3772 if (flow_table && flow_id <= flow_table->mask) {
3773 rflow = &flow_table->flows[flow_id];
3774 cpu = ACCESS_ONCE(rflow->cpu);
3775 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3776 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3777 rflow->last_qtail) <
3778 (int)(10 * flow_table->mask)))
3784 EXPORT_SYMBOL(rps_may_expire_flow);
3786 #endif /* CONFIG_RFS_ACCEL */
3788 /* Called from hardirq (IPI) context */
3789 static void rps_trigger_softirq(void *data)
3791 struct softnet_data *sd = data;
3793 ____napi_schedule(sd, &sd->backlog);
3797 #endif /* CONFIG_RPS */
3800 * Check if this softnet_data structure is another cpu one
3801 * If yes, queue it to our IPI list and return 1
3804 static int rps_ipi_queued(struct softnet_data *sd)
3807 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3810 sd->rps_ipi_next = mysd->rps_ipi_list;
3811 mysd->rps_ipi_list = sd;
3813 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3816 #endif /* CONFIG_RPS */
3820 #ifdef CONFIG_NET_FLOW_LIMIT
3821 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3824 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3826 #ifdef CONFIG_NET_FLOW_LIMIT
3827 struct sd_flow_limit *fl;
3828 struct softnet_data *sd;
3829 unsigned int old_flow, new_flow;
3831 if (qlen < (netdev_max_backlog >> 1))
3834 sd = this_cpu_ptr(&softnet_data);
3837 fl = rcu_dereference(sd->flow_limit);
3839 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3840 old_flow = fl->history[fl->history_head];
3841 fl->history[fl->history_head] = new_flow;
3844 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3846 if (likely(fl->buckets[old_flow]))
3847 fl->buckets[old_flow]--;
3849 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3861 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3862 * queue (may be a remote CPU queue).
3864 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3865 unsigned int *qtail)
3867 struct softnet_data *sd;
3868 unsigned long flags;
3871 sd = &per_cpu(softnet_data, cpu);
3873 local_irq_save(flags);
3876 if (!netif_running(skb->dev))
3878 qlen = skb_queue_len(&sd->input_pkt_queue);
3879 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3882 __skb_queue_tail(&sd->input_pkt_queue, skb);
3883 input_queue_tail_incr_save(sd, qtail);
3885 local_irq_restore(flags);
3886 return NET_RX_SUCCESS;
3889 /* Schedule NAPI for backlog device
3890 * We can use non atomic operation since we own the queue lock
3892 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3893 if (!rps_ipi_queued(sd))
3894 ____napi_schedule(sd, &sd->backlog);
3903 local_irq_restore(flags);
3905 atomic_long_inc(&skb->dev->rx_dropped);
3910 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
3911 struct bpf_prog *xdp_prog)
3913 struct xdp_buff xdp;
3919 /* Reinjected packets coming from act_mirred or similar should
3920 * not get XDP generic processing.
3922 if (skb_cloned(skb))
3925 if (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_end = xdp.data + hlen;
3935 xdp.data_hard_start = skb->data - skb_headroom(skb);
3936 orig_data = xdp.data;
3938 act = bpf_prog_run_xdp(xdp_prog, &xdp);
3940 off = xdp.data - orig_data;
3942 __skb_pull(skb, off);
3944 __skb_push(skb, -off);
3945 skb->mac_header += off;
3950 __skb_push(skb, mac_len);
3956 bpf_warn_invalid_xdp_action(act);
3959 trace_xdp_exception(skb->dev, xdp_prog, act);
3970 /* When doing generic XDP we have to bypass the qdisc layer and the
3971 * network taps in order to match in-driver-XDP behavior.
3973 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
3975 struct net_device *dev = skb->dev;
3976 struct netdev_queue *txq;
3977 bool free_skb = true;
3980 txq = netdev_pick_tx(dev, skb, NULL);
3981 cpu = smp_processor_id();
3982 HARD_TX_LOCK(dev, txq, cpu);
3983 if (!netif_xmit_stopped(txq)) {
3984 rc = netdev_start_xmit(skb, dev, txq, 0);
3985 if (dev_xmit_complete(rc))
3988 HARD_TX_UNLOCK(dev, txq);
3990 trace_xdp_exception(dev, xdp_prog, XDP_TX);
3994 EXPORT_SYMBOL_GPL(generic_xdp_tx);
3996 static struct static_key generic_xdp_needed __read_mostly;
3998 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4001 u32 act = netif_receive_generic_xdp(skb, xdp_prog);
4004 if (act != XDP_PASS) {
4007 err = xdp_do_generic_redirect(skb->dev, skb,
4011 /* fallthru to submit skb */
4013 generic_xdp_tx(skb, xdp_prog);
4024 EXPORT_SYMBOL_GPL(do_xdp_generic);
4026 static int netif_rx_internal(struct sk_buff *skb)
4030 net_timestamp_check(netdev_tstamp_prequeue, skb);
4032 trace_netif_rx(skb);
4034 if (static_key_false(&generic_xdp_needed)) {
4039 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4043 /* Consider XDP consuming the packet a success from
4044 * the netdev point of view we do not want to count
4047 if (ret != XDP_PASS)
4048 return NET_RX_SUCCESS;
4052 if (static_key_false(&rps_needed)) {
4053 struct rps_dev_flow voidflow, *rflow = &voidflow;
4059 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4061 cpu = smp_processor_id();
4063 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4072 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4079 * netif_rx - post buffer to the network code
4080 * @skb: buffer to post
4082 * This function receives a packet from a device driver and queues it for
4083 * the upper (protocol) levels to process. It always succeeds. The buffer
4084 * may be dropped during processing for congestion control or by the
4088 * NET_RX_SUCCESS (no congestion)
4089 * NET_RX_DROP (packet was dropped)
4093 int netif_rx(struct sk_buff *skb)
4095 trace_netif_rx_entry(skb);
4097 return netif_rx_internal(skb);
4099 EXPORT_SYMBOL(netif_rx);
4101 int netif_rx_ni(struct sk_buff *skb)
4105 trace_netif_rx_ni_entry(skb);
4108 err = netif_rx_internal(skb);
4109 if (local_softirq_pending())
4115 EXPORT_SYMBOL(netif_rx_ni);
4117 static __latent_entropy void net_tx_action(struct softirq_action *h)
4119 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4121 if (sd->completion_queue) {
4122 struct sk_buff *clist;
4124 local_irq_disable();
4125 clist = sd->completion_queue;
4126 sd->completion_queue = NULL;
4130 struct sk_buff *skb = clist;
4132 clist = clist->next;
4134 WARN_ON(refcount_read(&skb->users));
4135 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4136 trace_consume_skb(skb);
4138 trace_kfree_skb(skb, net_tx_action);
4140 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4143 __kfree_skb_defer(skb);
4146 __kfree_skb_flush();
4149 if (sd->output_queue) {
4152 local_irq_disable();
4153 head = sd->output_queue;
4154 sd->output_queue = NULL;
4155 sd->output_queue_tailp = &sd->output_queue;
4159 struct Qdisc *q = head;
4160 spinlock_t *root_lock;
4162 head = head->next_sched;
4164 root_lock = qdisc_lock(q);
4165 spin_lock(root_lock);
4166 /* We need to make sure head->next_sched is read
4167 * before clearing __QDISC_STATE_SCHED
4169 smp_mb__before_atomic();
4170 clear_bit(__QDISC_STATE_SCHED, &q->state);
4172 spin_unlock(root_lock);
4177 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4178 /* This hook is defined here for ATM LANE */
4179 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4180 unsigned char *addr) __read_mostly;
4181 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4184 static inline struct sk_buff *
4185 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4186 struct net_device *orig_dev)
4188 #ifdef CONFIG_NET_CLS_ACT
4189 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
4190 struct tcf_result cl_res;
4192 /* If there's at least one ingress present somewhere (so
4193 * we get here via enabled static key), remaining devices
4194 * that are not configured with an ingress qdisc will bail
4200 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4204 qdisc_skb_cb(skb)->pkt_len = skb->len;
4205 skb->tc_at_ingress = 1;
4206 qdisc_bstats_cpu_update(cl->q, skb);
4208 switch (tcf_classify(skb, cl, &cl_res, false)) {
4210 case TC_ACT_RECLASSIFY:
4211 skb->tc_index = TC_H_MIN(cl_res.classid);
4214 qdisc_qstats_cpu_drop(cl->q);
4222 case TC_ACT_REDIRECT:
4223 /* skb_mac_header check was done by cls/act_bpf, so
4224 * we can safely push the L2 header back before
4225 * redirecting to another netdev
4227 __skb_push(skb, skb->mac_len);
4228 skb_do_redirect(skb);
4233 #endif /* CONFIG_NET_CLS_ACT */
4238 * netdev_is_rx_handler_busy - check if receive handler is registered
4239 * @dev: device to check
4241 * Check if a receive handler is already registered for a given device.
4242 * Return true if there one.
4244 * The caller must hold the rtnl_mutex.
4246 bool netdev_is_rx_handler_busy(struct net_device *dev)
4249 return dev && rtnl_dereference(dev->rx_handler);
4251 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4254 * netdev_rx_handler_register - register receive handler
4255 * @dev: device to register a handler for
4256 * @rx_handler: receive handler to register
4257 * @rx_handler_data: data pointer that is used by rx handler
4259 * Register a receive handler for a device. This handler will then be
4260 * called from __netif_receive_skb. A negative errno code is returned
4263 * The caller must hold the rtnl_mutex.
4265 * For a general description of rx_handler, see enum rx_handler_result.
4267 int netdev_rx_handler_register(struct net_device *dev,
4268 rx_handler_func_t *rx_handler,
4269 void *rx_handler_data)
4271 if (netdev_is_rx_handler_busy(dev))
4274 /* Note: rx_handler_data must be set before rx_handler */
4275 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4276 rcu_assign_pointer(dev->rx_handler, rx_handler);
4280 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4283 * netdev_rx_handler_unregister - unregister receive handler
4284 * @dev: device to unregister a handler from
4286 * Unregister a receive handler from a device.
4288 * The caller must hold the rtnl_mutex.
4290 void netdev_rx_handler_unregister(struct net_device *dev)
4294 RCU_INIT_POINTER(dev->rx_handler, NULL);
4295 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4296 * section has a guarantee to see a non NULL rx_handler_data
4300 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4302 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4305 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4306 * the special handling of PFMEMALLOC skbs.
4308 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4310 switch (skb->protocol) {
4311 case htons(ETH_P_ARP):
4312 case htons(ETH_P_IP):
4313 case htons(ETH_P_IPV6):
4314 case htons(ETH_P_8021Q):
4315 case htons(ETH_P_8021AD):
4322 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4323 int *ret, struct net_device *orig_dev)
4325 #ifdef CONFIG_NETFILTER_INGRESS
4326 if (nf_hook_ingress_active(skb)) {
4330 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4335 ingress_retval = nf_hook_ingress(skb);
4337 return ingress_retval;
4339 #endif /* CONFIG_NETFILTER_INGRESS */
4343 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4345 struct packet_type *ptype, *pt_prev;
4346 rx_handler_func_t *rx_handler;
4347 struct net_device *orig_dev;
4348 bool deliver_exact = false;
4349 int ret = NET_RX_DROP;
4352 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4354 trace_netif_receive_skb(skb);
4356 orig_dev = skb->dev;
4358 skb_reset_network_header(skb);
4359 if (!skb_transport_header_was_set(skb))
4360 skb_reset_transport_header(skb);
4361 skb_reset_mac_len(skb);
4366 skb->skb_iif = skb->dev->ifindex;
4368 __this_cpu_inc(softnet_data.processed);
4370 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4371 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4372 skb = skb_vlan_untag(skb);
4377 if (skb_skip_tc_classify(skb))
4383 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4385 ret = deliver_skb(skb, pt_prev, orig_dev);
4389 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4391 ret = deliver_skb(skb, pt_prev, orig_dev);
4396 #ifdef CONFIG_NET_INGRESS
4397 if (static_key_false(&ingress_needed)) {
4398 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4402 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4408 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4411 if (skb_vlan_tag_present(skb)) {
4413 ret = deliver_skb(skb, pt_prev, orig_dev);
4416 if (vlan_do_receive(&skb))
4418 else if (unlikely(!skb))
4422 rx_handler = rcu_dereference(skb->dev->rx_handler);
4425 ret = deliver_skb(skb, pt_prev, orig_dev);
4428 switch (rx_handler(&skb)) {
4429 case RX_HANDLER_CONSUMED:
4430 ret = NET_RX_SUCCESS;
4432 case RX_HANDLER_ANOTHER:
4434 case RX_HANDLER_EXACT:
4435 deliver_exact = true;
4436 case RX_HANDLER_PASS:
4443 if (unlikely(skb_vlan_tag_present(skb))) {
4444 if (skb_vlan_tag_get_id(skb))
4445 skb->pkt_type = PACKET_OTHERHOST;
4446 /* Note: we might in the future use prio bits
4447 * and set skb->priority like in vlan_do_receive()
4448 * For the time being, just ignore Priority Code Point
4453 type = skb->protocol;
4455 /* deliver only exact match when indicated */
4456 if (likely(!deliver_exact)) {
4457 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4458 &ptype_base[ntohs(type) &
4462 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4463 &orig_dev->ptype_specific);
4465 if (unlikely(skb->dev != orig_dev)) {
4466 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4467 &skb->dev->ptype_specific);
4471 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4474 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4478 atomic_long_inc(&skb->dev->rx_dropped);
4480 atomic_long_inc(&skb->dev->rx_nohandler);
4482 /* Jamal, now you will not able to escape explaining
4483 * me how you were going to use this. :-)
4492 static int __netif_receive_skb(struct sk_buff *skb)
4496 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4497 unsigned int noreclaim_flag;
4500 * PFMEMALLOC skbs are special, they should
4501 * - be delivered to SOCK_MEMALLOC sockets only
4502 * - stay away from userspace
4503 * - have bounded memory usage
4505 * Use PF_MEMALLOC as this saves us from propagating the allocation
4506 * context down to all allocation sites.
4508 noreclaim_flag = memalloc_noreclaim_save();
4509 ret = __netif_receive_skb_core(skb, true);
4510 memalloc_noreclaim_restore(noreclaim_flag);
4512 ret = __netif_receive_skb_core(skb, false);
4517 static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp)
4519 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
4520 struct bpf_prog *new = xdp->prog;
4523 switch (xdp->command) {
4524 case XDP_SETUP_PROG:
4525 rcu_assign_pointer(dev->xdp_prog, new);
4530 static_key_slow_dec(&generic_xdp_needed);
4531 } else if (new && !old) {
4532 static_key_slow_inc(&generic_xdp_needed);
4533 dev_disable_lro(dev);
4537 case XDP_QUERY_PROG:
4538 xdp->prog_attached = !!old;
4539 xdp->prog_id = old ? old->aux->id : 0;
4550 static int netif_receive_skb_internal(struct sk_buff *skb)
4554 net_timestamp_check(netdev_tstamp_prequeue, skb);
4556 if (skb_defer_rx_timestamp(skb))
4557 return NET_RX_SUCCESS;
4559 if (static_key_false(&generic_xdp_needed)) {
4564 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4568 if (ret != XDP_PASS)
4574 if (static_key_false(&rps_needed)) {
4575 struct rps_dev_flow voidflow, *rflow = &voidflow;
4576 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4579 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4585 ret = __netif_receive_skb(skb);
4591 * netif_receive_skb - process receive buffer from network
4592 * @skb: buffer to process
4594 * netif_receive_skb() is the main receive data processing function.
4595 * It always succeeds. The buffer may be dropped during processing
4596 * for congestion control or by the protocol layers.
4598 * This function may only be called from softirq context and interrupts
4599 * should be enabled.
4601 * Return values (usually ignored):
4602 * NET_RX_SUCCESS: no congestion
4603 * NET_RX_DROP: packet was dropped
4605 int netif_receive_skb(struct sk_buff *skb)
4607 trace_netif_receive_skb_entry(skb);
4609 return netif_receive_skb_internal(skb);
4611 EXPORT_SYMBOL(netif_receive_skb);
4613 DEFINE_PER_CPU(struct work_struct, flush_works);
4615 /* Network device is going away, flush any packets still pending */
4616 static void flush_backlog(struct work_struct *work)
4618 struct sk_buff *skb, *tmp;
4619 struct softnet_data *sd;
4622 sd = this_cpu_ptr(&softnet_data);
4624 local_irq_disable();
4626 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4627 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4628 __skb_unlink(skb, &sd->input_pkt_queue);
4629 dev_kfree_skb_irq(skb);
4630 input_queue_head_incr(sd);
4636 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4637 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4638 __skb_unlink(skb, &sd->process_queue);
4640 input_queue_head_incr(sd);
4646 static void flush_all_backlogs(void)
4652 for_each_online_cpu(cpu)
4653 queue_work_on(cpu, system_highpri_wq,
4654 per_cpu_ptr(&flush_works, cpu));
4656 for_each_online_cpu(cpu)
4657 flush_work(per_cpu_ptr(&flush_works, cpu));
4662 static int napi_gro_complete(struct sk_buff *skb)
4664 struct packet_offload *ptype;
4665 __be16 type = skb->protocol;
4666 struct list_head *head = &offload_base;
4669 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4671 if (NAPI_GRO_CB(skb)->count == 1) {
4672 skb_shinfo(skb)->gso_size = 0;
4677 list_for_each_entry_rcu(ptype, head, list) {
4678 if (ptype->type != type || !ptype->callbacks.gro_complete)
4681 err = ptype->callbacks.gro_complete(skb, 0);
4687 WARN_ON(&ptype->list == head);
4689 return NET_RX_SUCCESS;
4693 return netif_receive_skb_internal(skb);
4696 /* napi->gro_list contains packets ordered by age.
4697 * youngest packets at the head of it.
4698 * Complete skbs in reverse order to reduce latencies.
4700 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4702 struct sk_buff *skb, *prev = NULL;
4704 /* scan list and build reverse chain */
4705 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4710 for (skb = prev; skb; skb = prev) {
4713 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4717 napi_gro_complete(skb);
4721 napi->gro_list = NULL;
4723 EXPORT_SYMBOL(napi_gro_flush);
4725 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4728 unsigned int maclen = skb->dev->hard_header_len;
4729 u32 hash = skb_get_hash_raw(skb);
4731 for (p = napi->gro_list; p; p = p->next) {
4732 unsigned long diffs;
4734 NAPI_GRO_CB(p)->flush = 0;
4736 if (hash != skb_get_hash_raw(p)) {
4737 NAPI_GRO_CB(p)->same_flow = 0;
4741 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4742 diffs |= p->vlan_tci ^ skb->vlan_tci;
4743 diffs |= skb_metadata_dst_cmp(p, skb);
4744 if (maclen == ETH_HLEN)
4745 diffs |= compare_ether_header(skb_mac_header(p),
4746 skb_mac_header(skb));
4748 diffs = memcmp(skb_mac_header(p),
4749 skb_mac_header(skb),
4751 NAPI_GRO_CB(p)->same_flow = !diffs;
4755 static void skb_gro_reset_offset(struct sk_buff *skb)
4757 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4758 const skb_frag_t *frag0 = &pinfo->frags[0];
4760 NAPI_GRO_CB(skb)->data_offset = 0;
4761 NAPI_GRO_CB(skb)->frag0 = NULL;
4762 NAPI_GRO_CB(skb)->frag0_len = 0;
4764 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4766 !PageHighMem(skb_frag_page(frag0)) &&
4767 (!NET_IP_ALIGN || !(skb_frag_off(frag0) & 3))) {
4768 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4769 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4770 skb_frag_size(frag0),
4771 skb->end - skb->tail);
4775 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4777 struct skb_shared_info *pinfo = skb_shinfo(skb);
4779 BUG_ON(skb->end - skb->tail < grow);
4781 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4783 skb->data_len -= grow;
4786 pinfo->frags[0].page_offset += grow;
4787 skb_frag_size_sub(&pinfo->frags[0], grow);
4789 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4790 skb_frag_unref(skb, 0);
4791 memmove(pinfo->frags, pinfo->frags + 1,
4792 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4796 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4798 struct sk_buff **pp = NULL;
4799 struct packet_offload *ptype;
4800 __be16 type = skb->protocol;
4801 struct list_head *head = &offload_base;
4803 enum gro_result ret;
4806 if (netif_elide_gro(skb->dev))
4809 gro_list_prepare(napi, skb);
4812 list_for_each_entry_rcu(ptype, head, list) {
4813 if (ptype->type != type || !ptype->callbacks.gro_receive)
4816 skb_set_network_header(skb, skb_gro_offset(skb));
4817 skb_reset_mac_len(skb);
4818 NAPI_GRO_CB(skb)->same_flow = 0;
4819 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4820 NAPI_GRO_CB(skb)->free = 0;
4821 NAPI_GRO_CB(skb)->encap_mark = 0;
4822 NAPI_GRO_CB(skb)->recursion_counter = 0;
4823 NAPI_GRO_CB(skb)->is_fou = 0;
4824 NAPI_GRO_CB(skb)->is_atomic = 1;
4825 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4827 /* Setup for GRO checksum validation */
4828 switch (skb->ip_summed) {
4829 case CHECKSUM_COMPLETE:
4830 NAPI_GRO_CB(skb)->csum = skb->csum;
4831 NAPI_GRO_CB(skb)->csum_valid = 1;
4832 NAPI_GRO_CB(skb)->csum_cnt = 0;
4834 case CHECKSUM_UNNECESSARY:
4835 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4836 NAPI_GRO_CB(skb)->csum_valid = 0;
4839 NAPI_GRO_CB(skb)->csum_cnt = 0;
4840 NAPI_GRO_CB(skb)->csum_valid = 0;
4843 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4848 if (&ptype->list == head)
4851 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
4856 same_flow = NAPI_GRO_CB(skb)->same_flow;
4857 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4860 struct sk_buff *nskb = *pp;
4864 napi_gro_complete(nskb);
4871 if (NAPI_GRO_CB(skb)->flush)
4874 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4875 struct sk_buff *nskb = napi->gro_list;
4877 /* locate the end of the list to select the 'oldest' flow */
4878 while (nskb->next) {
4884 napi_gro_complete(nskb);
4888 NAPI_GRO_CB(skb)->count = 1;
4889 NAPI_GRO_CB(skb)->age = jiffies;
4890 NAPI_GRO_CB(skb)->last = skb;
4891 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4892 skb->next = napi->gro_list;
4893 napi->gro_list = skb;
4897 grow = skb_gro_offset(skb) - skb_headlen(skb);
4899 gro_pull_from_frag0(skb, grow);
4908 struct packet_offload *gro_find_receive_by_type(__be16 type)
4910 struct list_head *offload_head = &offload_base;
4911 struct packet_offload *ptype;
4913 list_for_each_entry_rcu(ptype, offload_head, list) {
4914 if (ptype->type != type || !ptype->callbacks.gro_receive)
4920 EXPORT_SYMBOL(gro_find_receive_by_type);
4922 struct packet_offload *gro_find_complete_by_type(__be16 type)
4924 struct list_head *offload_head = &offload_base;
4925 struct packet_offload *ptype;
4927 list_for_each_entry_rcu(ptype, offload_head, list) {
4928 if (ptype->type != type || !ptype->callbacks.gro_complete)
4934 EXPORT_SYMBOL(gro_find_complete_by_type);
4936 static void napi_skb_free_stolen_head(struct sk_buff *skb)
4940 kmem_cache_free(skbuff_head_cache, skb);
4943 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4947 if (netif_receive_skb_internal(skb))
4955 case GRO_MERGED_FREE:
4956 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4957 napi_skb_free_stolen_head(skb);
4971 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4973 skb_mark_napi_id(skb, napi);
4974 trace_napi_gro_receive_entry(skb);
4976 skb_gro_reset_offset(skb);
4978 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4980 EXPORT_SYMBOL(napi_gro_receive);
4982 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4984 if (unlikely(skb->pfmemalloc)) {
4988 __skb_pull(skb, skb_headlen(skb));
4989 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4990 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4992 skb->dev = napi->dev;
4995 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
4996 skb->pkt_type = PACKET_HOST;
4998 skb->encapsulation = 0;
4999 skb_shinfo(skb)->gso_type = 0;
5000 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5006 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5008 struct sk_buff *skb = napi->skb;
5011 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5014 skb_mark_napi_id(skb, napi);
5019 EXPORT_SYMBOL(napi_get_frags);
5021 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5022 struct sk_buff *skb,
5028 __skb_push(skb, ETH_HLEN);
5029 skb->protocol = eth_type_trans(skb, skb->dev);
5030 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5035 napi_reuse_skb(napi, skb);
5038 case GRO_MERGED_FREE:
5039 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5040 napi_skb_free_stolen_head(skb);
5042 napi_reuse_skb(napi, skb);
5053 /* Upper GRO stack assumes network header starts at gro_offset=0
5054 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5055 * We copy ethernet header into skb->data to have a common layout.
5057 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5059 struct sk_buff *skb = napi->skb;
5060 const struct ethhdr *eth;
5061 unsigned int hlen = sizeof(*eth);
5065 skb_reset_mac_header(skb);
5066 skb_gro_reset_offset(skb);
5068 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5069 eth = skb_gro_header_slow(skb, hlen, 0);
5070 if (unlikely(!eth)) {
5071 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5072 __func__, napi->dev->name);
5073 napi_reuse_skb(napi, skb);
5077 eth = (const struct ethhdr *)skb->data;
5078 gro_pull_from_frag0(skb, hlen);
5079 NAPI_GRO_CB(skb)->frag0 += hlen;
5080 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5082 __skb_pull(skb, hlen);
5085 * This works because the only protocols we care about don't require
5087 * We'll fix it up properly in napi_frags_finish()
5089 skb->protocol = eth->h_proto;
5094 gro_result_t napi_gro_frags(struct napi_struct *napi)
5096 struct sk_buff *skb = napi_frags_skb(napi);
5101 trace_napi_gro_frags_entry(skb);
5103 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5105 EXPORT_SYMBOL(napi_gro_frags);
5107 /* Compute the checksum from gro_offset and return the folded value
5108 * after adding in any pseudo checksum.
5110 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5115 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5117 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5118 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5120 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5121 !skb->csum_complete_sw)
5122 netdev_rx_csum_fault(skb->dev);
5125 NAPI_GRO_CB(skb)->csum = wsum;
5126 NAPI_GRO_CB(skb)->csum_valid = 1;
5130 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5132 static void net_rps_send_ipi(struct softnet_data *remsd)
5136 struct softnet_data *next = remsd->rps_ipi_next;
5138 if (cpu_online(remsd->cpu))
5139 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5146 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5147 * Note: called with local irq disabled, but exits with local irq enabled.
5149 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5152 struct softnet_data *remsd = sd->rps_ipi_list;
5155 sd->rps_ipi_list = NULL;
5159 /* Send pending IPI's to kick RPS processing on remote cpus. */
5160 net_rps_send_ipi(remsd);
5166 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5169 return sd->rps_ipi_list != NULL;
5175 static int process_backlog(struct napi_struct *napi, int quota)
5177 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5181 /* Check if we have pending ipi, its better to send them now,
5182 * not waiting net_rx_action() end.
5184 if (sd_has_rps_ipi_waiting(sd)) {
5185 local_irq_disable();
5186 net_rps_action_and_irq_enable(sd);
5189 napi->weight = dev_rx_weight;
5191 struct sk_buff *skb;
5193 while ((skb = __skb_dequeue(&sd->process_queue))) {
5195 __netif_receive_skb(skb);
5197 input_queue_head_incr(sd);
5198 if (++work >= quota)
5203 local_irq_disable();
5205 if (skb_queue_empty(&sd->input_pkt_queue)) {
5207 * Inline a custom version of __napi_complete().
5208 * only current cpu owns and manipulates this napi,
5209 * and NAPI_STATE_SCHED is the only possible flag set
5211 * We can use a plain write instead of clear_bit(),
5212 * and we dont need an smp_mb() memory barrier.
5217 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5218 &sd->process_queue);
5228 * __napi_schedule - schedule for receive
5229 * @n: entry to schedule
5231 * The entry's receive function will be scheduled to run.
5232 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5234 void __napi_schedule(struct napi_struct *n)
5236 unsigned long flags;
5238 local_irq_save(flags);
5239 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5240 local_irq_restore(flags);
5242 EXPORT_SYMBOL(__napi_schedule);
5245 * napi_schedule_prep - check if napi can be scheduled
5248 * Test if NAPI routine is already running, and if not mark
5249 * it as running. This is used as a condition variable
5250 * insure only one NAPI poll instance runs. We also make
5251 * sure there is no pending NAPI disable.
5253 bool napi_schedule_prep(struct napi_struct *n)
5255 unsigned long val, new;
5258 val = READ_ONCE(n->state);
5259 if (unlikely(val & NAPIF_STATE_DISABLE))
5261 new = val | NAPIF_STATE_SCHED;
5263 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5264 * This was suggested by Alexander Duyck, as compiler
5265 * emits better code than :
5266 * if (val & NAPIF_STATE_SCHED)
5267 * new |= NAPIF_STATE_MISSED;
5269 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5271 } while (cmpxchg(&n->state, val, new) != val);
5273 return !(val & NAPIF_STATE_SCHED);
5275 EXPORT_SYMBOL(napi_schedule_prep);
5278 * __napi_schedule_irqoff - schedule for receive
5279 * @n: entry to schedule
5281 * Variant of __napi_schedule() assuming hard irqs are masked.
5283 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5284 * because the interrupt disabled assumption might not be true
5285 * due to force-threaded interrupts and spinlock substitution.
5287 void __napi_schedule_irqoff(struct napi_struct *n)
5289 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5290 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5294 EXPORT_SYMBOL(__napi_schedule_irqoff);
5296 bool napi_complete_done(struct napi_struct *n, int work_done)
5298 unsigned long flags, val, new;
5301 * 1) Don't let napi dequeue from the cpu poll list
5302 * just in case its running on a different cpu.
5303 * 2) If we are busy polling, do nothing here, we have
5304 * the guarantee we will be called later.
5306 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5307 NAPIF_STATE_IN_BUSY_POLL)))
5311 unsigned long timeout = 0;
5314 timeout = n->dev->gro_flush_timeout;
5316 /* When the NAPI instance uses a timeout and keeps postponing
5317 * it, we need to bound somehow the time packets are kept in
5320 napi_gro_flush(n, !!timeout);
5322 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5323 HRTIMER_MODE_REL_PINNED);
5325 if (unlikely(!list_empty(&n->poll_list))) {
5326 /* If n->poll_list is not empty, we need to mask irqs */
5327 local_irq_save(flags);
5328 list_del_init(&n->poll_list);
5329 local_irq_restore(flags);
5333 val = READ_ONCE(n->state);
5335 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5337 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
5339 /* If STATE_MISSED was set, leave STATE_SCHED set,
5340 * because we will call napi->poll() one more time.
5341 * This C code was suggested by Alexander Duyck to help gcc.
5343 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5345 } while (cmpxchg(&n->state, val, new) != val);
5347 if (unlikely(val & NAPIF_STATE_MISSED)) {
5354 EXPORT_SYMBOL(napi_complete_done);
5356 /* must be called under rcu_read_lock(), as we dont take a reference */
5357 static struct napi_struct *napi_by_id(unsigned int napi_id)
5359 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5360 struct napi_struct *napi;
5362 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5363 if (napi->napi_id == napi_id)
5369 #if defined(CONFIG_NET_RX_BUSY_POLL)
5371 #define BUSY_POLL_BUDGET 8
5373 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
5377 /* Busy polling means there is a high chance device driver hard irq
5378 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5379 * set in napi_schedule_prep().
5380 * Since we are about to call napi->poll() once more, we can safely
5381 * clear NAPI_STATE_MISSED.
5383 * Note: x86 could use a single "lock and ..." instruction
5384 * to perform these two clear_bit()
5386 clear_bit(NAPI_STATE_MISSED, &napi->state);
5387 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
5391 /* All we really want here is to re-enable device interrupts.
5392 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5394 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5395 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5396 netpoll_poll_unlock(have_poll_lock);
5397 if (rc == BUSY_POLL_BUDGET)
5398 __napi_schedule(napi);
5402 void napi_busy_loop(unsigned int napi_id,
5403 bool (*loop_end)(void *, unsigned long),
5406 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
5407 int (*napi_poll)(struct napi_struct *napi, int budget);
5408 void *have_poll_lock = NULL;
5409 struct napi_struct *napi;
5416 napi = napi_by_id(napi_id);
5426 unsigned long val = READ_ONCE(napi->state);
5428 /* If multiple threads are competing for this napi,
5429 * we avoid dirtying napi->state as much as we can.
5431 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5432 NAPIF_STATE_IN_BUSY_POLL))
5434 if (cmpxchg(&napi->state, val,
5435 val | NAPIF_STATE_IN_BUSY_POLL |
5436 NAPIF_STATE_SCHED) != val)
5438 have_poll_lock = netpoll_poll_lock(napi);
5439 napi_poll = napi->poll;
5441 work = napi_poll(napi, BUSY_POLL_BUDGET);
5442 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
5445 __NET_ADD_STATS(dev_net(napi->dev),
5446 LINUX_MIB_BUSYPOLLRXPACKETS, work);
5449 if (!loop_end || loop_end(loop_end_arg, start_time))
5452 if (unlikely(need_resched())) {
5454 busy_poll_stop(napi, have_poll_lock);
5458 if (loop_end(loop_end_arg, start_time))
5465 busy_poll_stop(napi, have_poll_lock);
5470 EXPORT_SYMBOL(napi_busy_loop);
5472 #endif /* CONFIG_NET_RX_BUSY_POLL */
5474 static void napi_hash_add(struct napi_struct *napi)
5476 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5477 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5480 spin_lock(&napi_hash_lock);
5482 /* 0..NR_CPUS range is reserved for sender_cpu use */
5484 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
5485 napi_gen_id = MIN_NAPI_ID;
5486 } while (napi_by_id(napi_gen_id));
5487 napi->napi_id = napi_gen_id;
5489 hlist_add_head_rcu(&napi->napi_hash_node,
5490 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5492 spin_unlock(&napi_hash_lock);
5495 /* Warning : caller is responsible to make sure rcu grace period
5496 * is respected before freeing memory containing @napi
5498 bool napi_hash_del(struct napi_struct *napi)
5500 bool rcu_sync_needed = false;
5502 spin_lock(&napi_hash_lock);
5504 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5505 rcu_sync_needed = true;
5506 hlist_del_rcu(&napi->napi_hash_node);
5508 spin_unlock(&napi_hash_lock);
5509 return rcu_sync_needed;
5511 EXPORT_SYMBOL_GPL(napi_hash_del);
5513 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5515 struct napi_struct *napi;
5517 napi = container_of(timer, struct napi_struct, timer);
5519 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5520 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5522 if (napi->gro_list && !napi_disable_pending(napi) &&
5523 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
5524 __napi_schedule_irqoff(napi);
5526 return HRTIMER_NORESTART;
5529 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5530 int (*poll)(struct napi_struct *, int), int weight)
5532 INIT_LIST_HEAD(&napi->poll_list);
5533 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5534 napi->timer.function = napi_watchdog;
5535 napi->gro_count = 0;
5536 napi->gro_list = NULL;
5539 if (weight > NAPI_POLL_WEIGHT)
5540 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5542 napi->weight = weight;
5544 #ifdef CONFIG_NETPOLL
5545 napi->poll_owner = -1;
5547 set_bit(NAPI_STATE_SCHED, &napi->state);
5548 set_bit(NAPI_STATE_NPSVC, &napi->state);
5549 list_add_rcu(&napi->dev_list, &dev->napi_list);
5550 napi_hash_add(napi);
5552 EXPORT_SYMBOL(netif_napi_add);
5554 void napi_disable(struct napi_struct *n)
5557 set_bit(NAPI_STATE_DISABLE, &n->state);
5559 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5561 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5564 hrtimer_cancel(&n->timer);
5566 clear_bit(NAPI_STATE_DISABLE, &n->state);
5568 EXPORT_SYMBOL(napi_disable);
5570 /* Must be called in process context */
5571 void netif_napi_del(struct napi_struct *napi)
5574 if (napi_hash_del(napi))
5576 list_del_init(&napi->dev_list);
5577 napi_free_frags(napi);
5579 kfree_skb_list(napi->gro_list);
5580 napi->gro_list = NULL;
5581 napi->gro_count = 0;
5583 EXPORT_SYMBOL(netif_napi_del);
5585 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5590 list_del_init(&n->poll_list);
5592 have = netpoll_poll_lock(n);
5596 /* This NAPI_STATE_SCHED test is for avoiding a race
5597 * with netpoll's poll_napi(). Only the entity which
5598 * obtains the lock and sees NAPI_STATE_SCHED set will
5599 * actually make the ->poll() call. Therefore we avoid
5600 * accidentally calling ->poll() when NAPI is not scheduled.
5603 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5604 work = n->poll(n, weight);
5605 trace_napi_poll(n, work, weight);
5608 WARN_ON_ONCE(work > weight);
5610 if (likely(work < weight))
5613 /* Drivers must not modify the NAPI state if they
5614 * consume the entire weight. In such cases this code
5615 * still "owns" the NAPI instance and therefore can
5616 * move the instance around on the list at-will.
5618 if (unlikely(napi_disable_pending(n))) {
5624 /* flush too old packets
5625 * If HZ < 1000, flush all packets.
5627 napi_gro_flush(n, HZ >= 1000);
5630 /* Some drivers may have called napi_schedule
5631 * prior to exhausting their budget.
5633 if (unlikely(!list_empty(&n->poll_list))) {
5634 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5635 n->dev ? n->dev->name : "backlog");
5639 list_add_tail(&n->poll_list, repoll);
5642 netpoll_poll_unlock(have);
5647 static __latent_entropy void net_rx_action(struct softirq_action *h)
5649 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5650 unsigned long time_limit = jiffies +
5651 usecs_to_jiffies(netdev_budget_usecs);
5652 int budget = netdev_budget;
5656 local_irq_disable();
5657 list_splice_init(&sd->poll_list, &list);
5661 struct napi_struct *n;
5663 if (list_empty(&list)) {
5664 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5669 n = list_first_entry(&list, struct napi_struct, poll_list);
5670 budget -= napi_poll(n, &repoll);
5672 /* If softirq window is exhausted then punt.
5673 * Allow this to run for 2 jiffies since which will allow
5674 * an average latency of 1.5/HZ.
5676 if (unlikely(budget <= 0 ||
5677 time_after_eq(jiffies, time_limit))) {
5683 local_irq_disable();
5685 list_splice_tail_init(&sd->poll_list, &list);
5686 list_splice_tail(&repoll, &list);
5687 list_splice(&list, &sd->poll_list);
5688 if (!list_empty(&sd->poll_list))
5689 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5691 net_rps_action_and_irq_enable(sd);
5693 __kfree_skb_flush();
5696 struct netdev_adjacent {
5697 struct net_device *dev;
5699 /* upper master flag, there can only be one master device per list */
5702 /* counter for the number of times this device was added to us */
5705 /* private field for the users */
5708 struct list_head list;
5709 struct rcu_head rcu;
5712 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5713 struct list_head *adj_list)
5715 struct netdev_adjacent *adj;
5717 list_for_each_entry(adj, adj_list, list) {
5718 if (adj->dev == adj_dev)
5724 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5726 struct net_device *dev = data;
5728 return upper_dev == dev;
5732 * netdev_has_upper_dev - Check if device is linked to an upper device
5734 * @upper_dev: upper device to check
5736 * Find out if a device is linked to specified upper device and return true
5737 * in case it is. Note that this checks only immediate upper device,
5738 * not through a complete stack of devices. The caller must hold the RTNL lock.
5740 bool netdev_has_upper_dev(struct net_device *dev,
5741 struct net_device *upper_dev)
5745 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5748 EXPORT_SYMBOL(netdev_has_upper_dev);
5751 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5753 * @upper_dev: upper device to check
5755 * Find out if a device is linked to specified upper device and return true
5756 * in case it is. Note that this checks the entire upper device chain.
5757 * The caller must hold rcu lock.
5760 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5761 struct net_device *upper_dev)
5763 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5766 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5769 * netdev_has_any_upper_dev - Check if device is linked to some device
5772 * Find out if a device is linked to an upper device and return true in case
5773 * it is. The caller must hold the RTNL lock.
5775 bool netdev_has_any_upper_dev(struct net_device *dev)
5779 return !list_empty(&dev->adj_list.upper);
5781 EXPORT_SYMBOL(netdev_has_any_upper_dev);
5784 * netdev_master_upper_dev_get - Get master upper device
5787 * Find a master upper device and return pointer to it or NULL in case
5788 * it's not there. The caller must hold the RTNL lock.
5790 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5792 struct netdev_adjacent *upper;
5796 if (list_empty(&dev->adj_list.upper))
5799 upper = list_first_entry(&dev->adj_list.upper,
5800 struct netdev_adjacent, list);
5801 if (likely(upper->master))
5805 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5808 * netdev_has_any_lower_dev - Check if device is linked to some device
5811 * Find out if a device is linked to a lower device and return true in case
5812 * it is. The caller must hold the RTNL lock.
5814 static bool netdev_has_any_lower_dev(struct net_device *dev)
5818 return !list_empty(&dev->adj_list.lower);
5821 void *netdev_adjacent_get_private(struct list_head *adj_list)
5823 struct netdev_adjacent *adj;
5825 adj = list_entry(adj_list, struct netdev_adjacent, list);
5827 return adj->private;
5829 EXPORT_SYMBOL(netdev_adjacent_get_private);
5832 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5834 * @iter: list_head ** of the current position
5836 * Gets the next device from the dev's upper list, starting from iter
5837 * position. The caller must hold RCU read lock.
5839 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5840 struct list_head **iter)
5842 struct netdev_adjacent *upper;
5844 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5846 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5848 if (&upper->list == &dev->adj_list.upper)
5851 *iter = &upper->list;
5855 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5857 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5858 struct list_head **iter)
5860 struct netdev_adjacent *upper;
5862 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5864 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5866 if (&upper->list == &dev->adj_list.upper)
5869 *iter = &upper->list;
5874 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5875 int (*fn)(struct net_device *dev,
5879 struct net_device *udev;
5880 struct list_head *iter;
5883 for (iter = &dev->adj_list.upper,
5884 udev = netdev_next_upper_dev_rcu(dev, &iter);
5886 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5887 /* first is the upper device itself */
5888 ret = fn(udev, data);
5892 /* then look at all of its upper devices */
5893 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5900 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5903 * netdev_lower_get_next_private - Get the next ->private from the
5904 * lower neighbour list
5906 * @iter: list_head ** of the current position
5908 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5909 * list, starting from iter position. The caller must hold either hold the
5910 * RTNL lock or its own locking that guarantees that the neighbour lower
5911 * list will remain unchanged.
5913 void *netdev_lower_get_next_private(struct net_device *dev,
5914 struct list_head **iter)
5916 struct netdev_adjacent *lower;
5918 lower = list_entry(*iter, struct netdev_adjacent, list);
5920 if (&lower->list == &dev->adj_list.lower)
5923 *iter = lower->list.next;
5925 return lower->private;
5927 EXPORT_SYMBOL(netdev_lower_get_next_private);
5930 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5931 * lower neighbour list, RCU
5934 * @iter: list_head ** of the current position
5936 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5937 * list, starting from iter position. The caller must hold RCU read lock.
5939 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5940 struct list_head **iter)
5942 struct netdev_adjacent *lower;
5944 WARN_ON_ONCE(!rcu_read_lock_held());
5946 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5948 if (&lower->list == &dev->adj_list.lower)
5951 *iter = &lower->list;
5953 return lower->private;
5955 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5958 * netdev_lower_get_next - Get the next device from the lower neighbour
5961 * @iter: list_head ** of the current position
5963 * Gets the next netdev_adjacent from the dev's lower neighbour
5964 * list, starting from iter position. The caller must hold RTNL lock or
5965 * its own locking that guarantees that the neighbour lower
5966 * list will remain unchanged.
5968 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5970 struct netdev_adjacent *lower;
5972 lower = list_entry(*iter, struct netdev_adjacent, list);
5974 if (&lower->list == &dev->adj_list.lower)
5977 *iter = lower->list.next;
5981 EXPORT_SYMBOL(netdev_lower_get_next);
5983 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5984 struct list_head **iter)
5986 struct netdev_adjacent *lower;
5988 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5990 if (&lower->list == &dev->adj_list.lower)
5993 *iter = &lower->list;
5998 int netdev_walk_all_lower_dev(struct net_device *dev,
5999 int (*fn)(struct net_device *dev,
6003 struct net_device *ldev;
6004 struct list_head *iter;
6007 for (iter = &dev->adj_list.lower,
6008 ldev = netdev_next_lower_dev(dev, &iter);
6010 ldev = netdev_next_lower_dev(dev, &iter)) {
6011 /* first is the lower device itself */
6012 ret = fn(ldev, data);
6016 /* then look at all of its lower devices */
6017 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6024 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6026 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6027 struct list_head **iter)
6029 struct netdev_adjacent *lower;
6031 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6032 if (&lower->list == &dev->adj_list.lower)
6035 *iter = &lower->list;
6040 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6041 int (*fn)(struct net_device *dev,
6045 struct net_device *ldev;
6046 struct list_head *iter;
6049 for (iter = &dev->adj_list.lower,
6050 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6052 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6053 /* first is the lower device itself */
6054 ret = fn(ldev, data);
6058 /* then look at all of its lower devices */
6059 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6066 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6069 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6070 * lower neighbour list, RCU
6074 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6075 * list. The caller must hold RCU read lock.
6077 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6079 struct netdev_adjacent *lower;
6081 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6082 struct netdev_adjacent, list);
6084 return lower->private;
6087 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6090 * netdev_master_upper_dev_get_rcu - Get master upper device
6093 * Find a master upper device and return pointer to it or NULL in case
6094 * it's not there. The caller must hold the RCU read lock.
6096 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6098 struct netdev_adjacent *upper;
6100 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6101 struct netdev_adjacent, list);
6102 if (upper && likely(upper->master))
6106 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6108 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6109 struct net_device *adj_dev,
6110 struct list_head *dev_list)
6112 char linkname[IFNAMSIZ+7];
6114 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6115 "upper_%s" : "lower_%s", adj_dev->name);
6116 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6119 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6121 struct list_head *dev_list)
6123 char linkname[IFNAMSIZ+7];
6125 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6126 "upper_%s" : "lower_%s", name);
6127 sysfs_remove_link(&(dev->dev.kobj), linkname);
6130 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6131 struct net_device *adj_dev,
6132 struct list_head *dev_list)
6134 return (dev_list == &dev->adj_list.upper ||
6135 dev_list == &dev->adj_list.lower) &&
6136 net_eq(dev_net(dev), dev_net(adj_dev));
6139 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6140 struct net_device *adj_dev,
6141 struct list_head *dev_list,
6142 void *private, bool master)
6144 struct netdev_adjacent *adj;
6147 adj = __netdev_find_adj(adj_dev, dev_list);
6151 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6152 dev->name, adj_dev->name, adj->ref_nr);
6157 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6162 adj->master = master;
6164 adj->private = private;
6167 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6168 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6170 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6171 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6176 /* Ensure that master link is always the first item in list. */
6178 ret = sysfs_create_link(&(dev->dev.kobj),
6179 &(adj_dev->dev.kobj), "master");
6181 goto remove_symlinks;
6183 list_add_rcu(&adj->list, dev_list);
6185 list_add_tail_rcu(&adj->list, dev_list);
6191 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6192 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6200 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6201 struct net_device *adj_dev,
6203 struct list_head *dev_list)
6205 struct netdev_adjacent *adj;
6207 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6208 dev->name, adj_dev->name, ref_nr);
6210 adj = __netdev_find_adj(adj_dev, dev_list);
6213 pr_err("Adjacency does not exist for device %s from %s\n",
6214 dev->name, adj_dev->name);
6219 if (adj->ref_nr > ref_nr) {
6220 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6221 dev->name, adj_dev->name, ref_nr,
6222 adj->ref_nr - ref_nr);
6223 adj->ref_nr -= ref_nr;
6228 sysfs_remove_link(&(dev->dev.kobj), "master");
6230 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6231 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6233 list_del_rcu(&adj->list);
6234 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6235 adj_dev->name, dev->name, adj_dev->name);
6237 kfree_rcu(adj, rcu);
6240 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6241 struct net_device *upper_dev,
6242 struct list_head *up_list,
6243 struct list_head *down_list,
6244 void *private, bool master)
6248 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6253 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6256 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6263 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6264 struct net_device *upper_dev,
6266 struct list_head *up_list,
6267 struct list_head *down_list)
6269 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6270 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6273 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6274 struct net_device *upper_dev,
6275 void *private, bool master)
6277 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6278 &dev->adj_list.upper,
6279 &upper_dev->adj_list.lower,
6283 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6284 struct net_device *upper_dev)
6286 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
6287 &dev->adj_list.upper,
6288 &upper_dev->adj_list.lower);
6291 static int __netdev_upper_dev_link(struct net_device *dev,
6292 struct net_device *upper_dev, bool master,
6293 void *upper_priv, void *upper_info)
6295 struct netdev_notifier_changeupper_info changeupper_info;
6300 if (dev == upper_dev)
6303 /* To prevent loops, check if dev is not upper device to upper_dev. */
6304 if (netdev_has_upper_dev(upper_dev, dev))
6307 if (netdev_has_upper_dev(dev, upper_dev))
6310 if (master && netdev_master_upper_dev_get(dev))
6313 changeupper_info.upper_dev = upper_dev;
6314 changeupper_info.master = master;
6315 changeupper_info.linking = true;
6316 changeupper_info.upper_info = upper_info;
6318 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6319 &changeupper_info.info);
6320 ret = notifier_to_errno(ret);
6324 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
6329 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6330 &changeupper_info.info);
6331 ret = notifier_to_errno(ret);
6338 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6344 * netdev_upper_dev_link - Add a link to the upper device
6346 * @upper_dev: new upper device
6348 * Adds a link to device which is upper to this one. The caller must hold
6349 * the RTNL lock. On a failure a negative errno code is returned.
6350 * On success the reference counts are adjusted and the function
6353 int netdev_upper_dev_link(struct net_device *dev,
6354 struct net_device *upper_dev)
6356 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
6358 EXPORT_SYMBOL(netdev_upper_dev_link);
6361 * netdev_master_upper_dev_link - Add a master link to the upper device
6363 * @upper_dev: new upper device
6364 * @upper_priv: upper device private
6365 * @upper_info: upper info to be passed down via notifier
6367 * Adds a link to device which is upper to this one. In this case, only
6368 * one master upper device can be linked, although other non-master devices
6369 * might be linked as well. The caller must hold the RTNL lock.
6370 * On a failure a negative errno code is returned. On success the reference
6371 * counts are adjusted and the function returns zero.
6373 int netdev_master_upper_dev_link(struct net_device *dev,
6374 struct net_device *upper_dev,
6375 void *upper_priv, void *upper_info)
6377 return __netdev_upper_dev_link(dev, upper_dev, true,
6378 upper_priv, upper_info);
6380 EXPORT_SYMBOL(netdev_master_upper_dev_link);
6383 * netdev_upper_dev_unlink - Removes a link to upper device
6385 * @upper_dev: new upper device
6387 * Removes a link to device which is upper to this one. The caller must hold
6390 void netdev_upper_dev_unlink(struct net_device *dev,
6391 struct net_device *upper_dev)
6393 struct netdev_notifier_changeupper_info changeupper_info;
6397 changeupper_info.upper_dev = upper_dev;
6398 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6399 changeupper_info.linking = false;
6401 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6402 &changeupper_info.info);
6404 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6406 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6407 &changeupper_info.info);
6409 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6412 * netdev_bonding_info_change - Dispatch event about slave change
6414 * @bonding_info: info to dispatch
6416 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6417 * The caller must hold the RTNL lock.
6419 void netdev_bonding_info_change(struct net_device *dev,
6420 struct netdev_bonding_info *bonding_info)
6422 struct netdev_notifier_bonding_info info;
6424 memcpy(&info.bonding_info, bonding_info,
6425 sizeof(struct netdev_bonding_info));
6426 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6429 EXPORT_SYMBOL(netdev_bonding_info_change);
6431 static void netdev_adjacent_add_links(struct net_device *dev)
6433 struct netdev_adjacent *iter;
6435 struct net *net = dev_net(dev);
6437 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6438 if (!net_eq(net, dev_net(iter->dev)))
6440 netdev_adjacent_sysfs_add(iter->dev, dev,
6441 &iter->dev->adj_list.lower);
6442 netdev_adjacent_sysfs_add(dev, iter->dev,
6443 &dev->adj_list.upper);
6446 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6447 if (!net_eq(net, dev_net(iter->dev)))
6449 netdev_adjacent_sysfs_add(iter->dev, dev,
6450 &iter->dev->adj_list.upper);
6451 netdev_adjacent_sysfs_add(dev, iter->dev,
6452 &dev->adj_list.lower);
6456 static void netdev_adjacent_del_links(struct net_device *dev)
6458 struct netdev_adjacent *iter;
6460 struct net *net = dev_net(dev);
6462 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6463 if (!net_eq(net, dev_net(iter->dev)))
6465 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6466 &iter->dev->adj_list.lower);
6467 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6468 &dev->adj_list.upper);
6471 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6472 if (!net_eq(net, dev_net(iter->dev)))
6474 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6475 &iter->dev->adj_list.upper);
6476 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6477 &dev->adj_list.lower);
6481 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6483 struct netdev_adjacent *iter;
6485 struct net *net = dev_net(dev);
6487 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6488 if (!net_eq(net, dev_net(iter->dev)))
6490 netdev_adjacent_sysfs_del(iter->dev, oldname,
6491 &iter->dev->adj_list.lower);
6492 netdev_adjacent_sysfs_add(iter->dev, dev,
6493 &iter->dev->adj_list.lower);
6496 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6497 if (!net_eq(net, dev_net(iter->dev)))
6499 netdev_adjacent_sysfs_del(iter->dev, oldname,
6500 &iter->dev->adj_list.upper);
6501 netdev_adjacent_sysfs_add(iter->dev, dev,
6502 &iter->dev->adj_list.upper);
6506 void *netdev_lower_dev_get_private(struct net_device *dev,
6507 struct net_device *lower_dev)
6509 struct netdev_adjacent *lower;
6513 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6517 return lower->private;
6519 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6522 int dev_get_nest_level(struct net_device *dev)
6524 struct net_device *lower = NULL;
6525 struct list_head *iter;
6531 netdev_for_each_lower_dev(dev, lower, iter) {
6532 nest = dev_get_nest_level(lower);
6533 if (max_nest < nest)
6537 return max_nest + 1;
6539 EXPORT_SYMBOL(dev_get_nest_level);
6542 * netdev_lower_change - Dispatch event about lower device state change
6543 * @lower_dev: device
6544 * @lower_state_info: state to dispatch
6546 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6547 * The caller must hold the RTNL lock.
6549 void netdev_lower_state_changed(struct net_device *lower_dev,
6550 void *lower_state_info)
6552 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6555 changelowerstate_info.lower_state_info = lower_state_info;
6556 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6557 &changelowerstate_info.info);
6559 EXPORT_SYMBOL(netdev_lower_state_changed);
6561 static void dev_change_rx_flags(struct net_device *dev, int flags)
6563 const struct net_device_ops *ops = dev->netdev_ops;
6565 if (ops->ndo_change_rx_flags)
6566 ops->ndo_change_rx_flags(dev, flags);
6569 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6571 unsigned int old_flags = dev->flags;
6577 dev->flags |= IFF_PROMISC;
6578 dev->promiscuity += inc;
6579 if (dev->promiscuity == 0) {
6582 * If inc causes overflow, untouch promisc and return error.
6585 dev->flags &= ~IFF_PROMISC;
6587 dev->promiscuity -= inc;
6588 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6593 if (dev->flags != old_flags) {
6594 pr_info("device %s %s promiscuous mode\n",
6596 dev->flags & IFF_PROMISC ? "entered" : "left");
6597 if (audit_enabled) {
6598 current_uid_gid(&uid, &gid);
6599 audit_log(current->audit_context, GFP_ATOMIC,
6600 AUDIT_ANOM_PROMISCUOUS,
6601 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6602 dev->name, (dev->flags & IFF_PROMISC),
6603 (old_flags & IFF_PROMISC),
6604 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6605 from_kuid(&init_user_ns, uid),
6606 from_kgid(&init_user_ns, gid),
6607 audit_get_sessionid(current));
6610 dev_change_rx_flags(dev, IFF_PROMISC);
6613 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6618 * dev_set_promiscuity - update promiscuity count on a device
6622 * Add or remove promiscuity from a device. While the count in the device
6623 * remains above zero the interface remains promiscuous. Once it hits zero
6624 * the device reverts back to normal filtering operation. A negative inc
6625 * value is used to drop promiscuity on the device.
6626 * Return 0 if successful or a negative errno code on error.
6628 int dev_set_promiscuity(struct net_device *dev, int inc)
6630 unsigned int old_flags = dev->flags;
6633 err = __dev_set_promiscuity(dev, inc, true);
6636 if (dev->flags != old_flags)
6637 dev_set_rx_mode(dev);
6640 EXPORT_SYMBOL(dev_set_promiscuity);
6642 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6644 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6648 dev->flags |= IFF_ALLMULTI;
6649 dev->allmulti += inc;
6650 if (dev->allmulti == 0) {
6653 * If inc causes overflow, untouch allmulti and return error.
6656 dev->flags &= ~IFF_ALLMULTI;
6658 dev->allmulti -= inc;
6659 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6664 if (dev->flags ^ old_flags) {
6665 dev_change_rx_flags(dev, IFF_ALLMULTI);
6666 dev_set_rx_mode(dev);
6668 __dev_notify_flags(dev, old_flags,
6669 dev->gflags ^ old_gflags);
6675 * dev_set_allmulti - update allmulti count on a device
6679 * Add or remove reception of all multicast frames to a device. While the
6680 * count in the device remains above zero the interface remains listening
6681 * to all interfaces. Once it hits zero the device reverts back to normal
6682 * filtering operation. A negative @inc value is used to drop the counter
6683 * when releasing a resource needing all multicasts.
6684 * Return 0 if successful or a negative errno code on error.
6687 int dev_set_allmulti(struct net_device *dev, int inc)
6689 return __dev_set_allmulti(dev, inc, true);
6691 EXPORT_SYMBOL(dev_set_allmulti);
6694 * Upload unicast and multicast address lists to device and
6695 * configure RX filtering. When the device doesn't support unicast
6696 * filtering it is put in promiscuous mode while unicast addresses
6699 void __dev_set_rx_mode(struct net_device *dev)
6701 const struct net_device_ops *ops = dev->netdev_ops;
6703 /* dev_open will call this function so the list will stay sane. */
6704 if (!(dev->flags&IFF_UP))
6707 if (!netif_device_present(dev))
6710 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6711 /* Unicast addresses changes may only happen under the rtnl,
6712 * therefore calling __dev_set_promiscuity here is safe.
6714 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6715 __dev_set_promiscuity(dev, 1, false);
6716 dev->uc_promisc = true;
6717 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6718 __dev_set_promiscuity(dev, -1, false);
6719 dev->uc_promisc = false;
6723 if (ops->ndo_set_rx_mode)
6724 ops->ndo_set_rx_mode(dev);
6727 void dev_set_rx_mode(struct net_device *dev)
6729 netif_addr_lock_bh(dev);
6730 __dev_set_rx_mode(dev);
6731 netif_addr_unlock_bh(dev);
6735 * dev_get_flags - get flags reported to userspace
6738 * Get the combination of flag bits exported through APIs to userspace.
6740 unsigned int dev_get_flags(const struct net_device *dev)
6744 flags = (dev->flags & ~(IFF_PROMISC |
6749 (dev->gflags & (IFF_PROMISC |
6752 if (netif_running(dev)) {
6753 if (netif_oper_up(dev))
6754 flags |= IFF_RUNNING;
6755 if (netif_carrier_ok(dev))
6756 flags |= IFF_LOWER_UP;
6757 if (netif_dormant(dev))
6758 flags |= IFF_DORMANT;
6763 EXPORT_SYMBOL(dev_get_flags);
6765 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6767 unsigned int old_flags = dev->flags;
6773 * Set the flags on our device.
6776 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6777 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6779 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6783 * Load in the correct multicast list now the flags have changed.
6786 if ((old_flags ^ flags) & IFF_MULTICAST)
6787 dev_change_rx_flags(dev, IFF_MULTICAST);
6789 dev_set_rx_mode(dev);
6792 * Have we downed the interface. We handle IFF_UP ourselves
6793 * according to user attempts to set it, rather than blindly
6798 if ((old_flags ^ flags) & IFF_UP) {
6799 if (old_flags & IFF_UP)
6802 ret = __dev_open(dev);
6805 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6806 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6807 unsigned int old_flags = dev->flags;
6809 dev->gflags ^= IFF_PROMISC;
6811 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6812 if (dev->flags != old_flags)
6813 dev_set_rx_mode(dev);
6816 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6817 * is important. Some (broken) drivers set IFF_PROMISC, when
6818 * IFF_ALLMULTI is requested not asking us and not reporting.
6820 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6821 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6823 dev->gflags ^= IFF_ALLMULTI;
6824 __dev_set_allmulti(dev, inc, false);
6830 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6831 unsigned int gchanges)
6833 unsigned int changes = dev->flags ^ old_flags;
6836 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6838 if (changes & IFF_UP) {
6839 if (dev->flags & IFF_UP)
6840 call_netdevice_notifiers(NETDEV_UP, dev);
6842 call_netdevice_notifiers(NETDEV_DOWN, dev);
6845 if (dev->flags & IFF_UP &&
6846 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6847 struct netdev_notifier_change_info change_info;
6849 change_info.flags_changed = changes;
6850 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6856 * dev_change_flags - change device settings
6858 * @flags: device state flags
6860 * Change settings on device based state flags. The flags are
6861 * in the userspace exported format.
6863 int dev_change_flags(struct net_device *dev, unsigned int flags)
6866 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6868 ret = __dev_change_flags(dev, flags);
6872 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6873 __dev_notify_flags(dev, old_flags, changes);
6876 EXPORT_SYMBOL(dev_change_flags);
6878 int __dev_set_mtu(struct net_device *dev, int new_mtu)
6880 const struct net_device_ops *ops = dev->netdev_ops;
6882 if (ops->ndo_change_mtu)
6883 return ops->ndo_change_mtu(dev, new_mtu);
6885 /* Pairs with all the lockless reads of dev->mtu in the stack */
6886 WRITE_ONCE(dev->mtu, new_mtu);
6889 EXPORT_SYMBOL(__dev_set_mtu);
6892 * dev_set_mtu - Change maximum transfer unit
6894 * @new_mtu: new transfer unit
6896 * Change the maximum transfer size of the network device.
6898 int dev_set_mtu(struct net_device *dev, int new_mtu)
6902 if (new_mtu == dev->mtu)
6905 err = dev_validate_mtu(dev, new_mtu);
6909 if (!netif_device_present(dev))
6912 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6913 err = notifier_to_errno(err);
6917 orig_mtu = dev->mtu;
6918 err = __dev_set_mtu(dev, new_mtu);
6921 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
6923 err = notifier_to_errno(err);
6925 /* setting mtu back and notifying everyone again,
6926 * so that they have a chance to revert changes.
6928 __dev_set_mtu(dev, orig_mtu);
6929 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
6935 EXPORT_SYMBOL(dev_set_mtu);
6938 * dev_set_group - Change group this device belongs to
6940 * @new_group: group this device should belong to
6942 void dev_set_group(struct net_device *dev, int new_group)
6944 dev->group = new_group;
6946 EXPORT_SYMBOL(dev_set_group);
6949 * dev_set_mac_address - Change Media Access Control Address
6953 * Change the hardware (MAC) address of the device
6955 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6957 const struct net_device_ops *ops = dev->netdev_ops;
6960 if (!ops->ndo_set_mac_address)
6962 if (sa->sa_family != dev->type)
6964 if (!netif_device_present(dev))
6966 err = ops->ndo_set_mac_address(dev, sa);
6969 dev->addr_assign_type = NET_ADDR_SET;
6970 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6971 add_device_randomness(dev->dev_addr, dev->addr_len);
6974 EXPORT_SYMBOL(dev_set_mac_address);
6977 * dev_change_carrier - Change device carrier
6979 * @new_carrier: new value
6981 * Change device carrier
6983 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6985 const struct net_device_ops *ops = dev->netdev_ops;
6987 if (!ops->ndo_change_carrier)
6989 if (!netif_device_present(dev))
6991 return ops->ndo_change_carrier(dev, new_carrier);
6993 EXPORT_SYMBOL(dev_change_carrier);
6996 * dev_get_phys_port_id - Get device physical port ID
7000 * Get device physical port ID
7002 int dev_get_phys_port_id(struct net_device *dev,
7003 struct netdev_phys_item_id *ppid)
7005 const struct net_device_ops *ops = dev->netdev_ops;
7007 if (!ops->ndo_get_phys_port_id)
7009 return ops->ndo_get_phys_port_id(dev, ppid);
7011 EXPORT_SYMBOL(dev_get_phys_port_id);
7014 * dev_get_phys_port_name - Get device physical port name
7017 * @len: limit of bytes to copy to name
7019 * Get device physical port name
7021 int dev_get_phys_port_name(struct net_device *dev,
7022 char *name, size_t len)
7024 const struct net_device_ops *ops = dev->netdev_ops;
7026 if (!ops->ndo_get_phys_port_name)
7028 return ops->ndo_get_phys_port_name(dev, name, len);
7030 EXPORT_SYMBOL(dev_get_phys_port_name);
7033 * dev_change_proto_down - update protocol port state information
7035 * @proto_down: new value
7037 * This info can be used by switch drivers to set the phys state of the
7040 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7042 const struct net_device_ops *ops = dev->netdev_ops;
7044 if (!ops->ndo_change_proto_down)
7046 if (!netif_device_present(dev))
7048 return ops->ndo_change_proto_down(dev, proto_down);
7050 EXPORT_SYMBOL(dev_change_proto_down);
7052 u8 __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op, u32 *prog_id)
7054 struct netdev_xdp xdp;
7056 memset(&xdp, 0, sizeof(xdp));
7057 xdp.command = XDP_QUERY_PROG;
7059 /* Query must always succeed. */
7060 WARN_ON(xdp_op(dev, &xdp) < 0);
7062 *prog_id = xdp.prog_id;
7064 return xdp.prog_attached;
7067 static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op,
7068 struct netlink_ext_ack *extack, u32 flags,
7069 struct bpf_prog *prog)
7071 struct netdev_xdp xdp;
7073 memset(&xdp, 0, sizeof(xdp));
7074 if (flags & XDP_FLAGS_HW_MODE)
7075 xdp.command = XDP_SETUP_PROG_HW;
7077 xdp.command = XDP_SETUP_PROG;
7078 xdp.extack = extack;
7082 return xdp_op(dev, &xdp);
7086 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
7088 * @extack: netlink extended ack
7089 * @fd: new program fd or negative value to clear
7090 * @flags: xdp-related flags
7092 * Set or clear a bpf program for a device
7094 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
7097 const struct net_device_ops *ops = dev->netdev_ops;
7098 struct bpf_prog *prog = NULL;
7099 xdp_op_t xdp_op, xdp_chk;
7104 xdp_op = xdp_chk = ops->ndo_xdp;
7105 if (!xdp_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7107 if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE))
7108 xdp_op = generic_xdp_install;
7109 if (xdp_op == xdp_chk)
7110 xdp_chk = generic_xdp_install;
7113 if (xdp_chk && __dev_xdp_attached(dev, xdp_chk, NULL))
7115 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7116 __dev_xdp_attached(dev, xdp_op, NULL))
7119 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
7121 return PTR_ERR(prog);
7124 err = dev_xdp_install(dev, xdp_op, extack, flags, prog);
7125 if (err < 0 && prog)
7132 * dev_new_index - allocate an ifindex
7133 * @net: the applicable net namespace
7135 * Returns a suitable unique value for a new device interface
7136 * number. The caller must hold the rtnl semaphore or the
7137 * dev_base_lock to be sure it remains unique.
7139 static int dev_new_index(struct net *net)
7141 int ifindex = net->ifindex;
7146 if (!__dev_get_by_index(net, ifindex))
7147 return net->ifindex = ifindex;
7151 /* Delayed registration/unregisteration */
7152 static LIST_HEAD(net_todo_list);
7153 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7155 static void net_set_todo(struct net_device *dev)
7157 list_add_tail(&dev->todo_list, &net_todo_list);
7158 dev_net(dev)->dev_unreg_count++;
7161 static void rollback_registered_many(struct list_head *head)
7163 struct net_device *dev, *tmp;
7164 LIST_HEAD(close_head);
7166 BUG_ON(dev_boot_phase);
7169 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
7170 /* Some devices call without registering
7171 * for initialization unwind. Remove those
7172 * devices and proceed with the remaining.
7174 if (dev->reg_state == NETREG_UNINITIALIZED) {
7175 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7179 list_del(&dev->unreg_list);
7182 dev->dismantle = true;
7183 BUG_ON(dev->reg_state != NETREG_REGISTERED);
7186 /* If device is running, close it first. */
7187 list_for_each_entry(dev, head, unreg_list)
7188 list_add_tail(&dev->close_list, &close_head);
7189 dev_close_many(&close_head, true);
7191 list_for_each_entry(dev, head, unreg_list) {
7192 /* And unlink it from device chain. */
7193 unlist_netdevice(dev);
7195 dev->reg_state = NETREG_UNREGISTERING;
7197 flush_all_backlogs();
7201 list_for_each_entry(dev, head, unreg_list) {
7202 struct sk_buff *skb = NULL;
7204 /* Shutdown queueing discipline. */
7208 /* Notify protocols, that we are about to destroy
7209 * this device. They should clean all the things.
7211 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7213 if (!dev->rtnl_link_ops ||
7214 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7215 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
7219 * Flush the unicast and multicast chains
7224 if (dev->netdev_ops->ndo_uninit)
7225 dev->netdev_ops->ndo_uninit(dev);
7228 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
7230 /* Notifier chain MUST detach us all upper devices. */
7231 WARN_ON(netdev_has_any_upper_dev(dev));
7232 WARN_ON(netdev_has_any_lower_dev(dev));
7234 /* Remove entries from kobject tree */
7235 netdev_unregister_kobject(dev);
7237 /* Remove XPS queueing entries */
7238 netif_reset_xps_queues_gt(dev, 0);
7244 list_for_each_entry(dev, head, unreg_list)
7248 static void rollback_registered(struct net_device *dev)
7252 list_add(&dev->unreg_list, &single);
7253 rollback_registered_many(&single);
7257 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
7258 struct net_device *upper, netdev_features_t features)
7260 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7261 netdev_features_t feature;
7264 for_each_netdev_feature(upper_disables, feature_bit) {
7265 feature = __NETIF_F_BIT(feature_bit);
7266 if (!(upper->wanted_features & feature)
7267 && (features & feature)) {
7268 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
7269 &feature, upper->name);
7270 features &= ~feature;
7277 static void netdev_sync_lower_features(struct net_device *upper,
7278 struct net_device *lower, netdev_features_t features)
7280 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7281 netdev_features_t feature;
7284 for_each_netdev_feature(upper_disables, feature_bit) {
7285 feature = __NETIF_F_BIT(feature_bit);
7286 if (!(features & feature) && (lower->features & feature)) {
7287 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
7288 &feature, lower->name);
7289 lower->wanted_features &= ~feature;
7290 __netdev_update_features(lower);
7292 if (unlikely(lower->features & feature))
7293 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
7294 &feature, lower->name);
7296 netdev_features_change(lower);
7301 static netdev_features_t netdev_fix_features(struct net_device *dev,
7302 netdev_features_t features)
7304 /* Fix illegal checksum combinations */
7305 if ((features & NETIF_F_HW_CSUM) &&
7306 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
7307 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
7308 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
7311 /* TSO requires that SG is present as well. */
7312 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
7313 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
7314 features &= ~NETIF_F_ALL_TSO;
7317 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
7318 !(features & NETIF_F_IP_CSUM)) {
7319 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
7320 features &= ~NETIF_F_TSO;
7321 features &= ~NETIF_F_TSO_ECN;
7324 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
7325 !(features & NETIF_F_IPV6_CSUM)) {
7326 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
7327 features &= ~NETIF_F_TSO6;
7330 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
7331 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
7332 features &= ~NETIF_F_TSO_MANGLEID;
7334 /* TSO ECN requires that TSO is present as well. */
7335 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
7336 features &= ~NETIF_F_TSO_ECN;
7338 /* Software GSO depends on SG. */
7339 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
7340 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
7341 features &= ~NETIF_F_GSO;
7344 /* GSO partial features require GSO partial be set */
7345 if ((features & dev->gso_partial_features) &&
7346 !(features & NETIF_F_GSO_PARTIAL)) {
7348 "Dropping partially supported GSO features since no GSO partial.\n");
7349 features &= ~dev->gso_partial_features;
7355 int __netdev_update_features(struct net_device *dev)
7357 struct net_device *upper, *lower;
7358 netdev_features_t features;
7359 struct list_head *iter;
7364 features = netdev_get_wanted_features(dev);
7366 if (dev->netdev_ops->ndo_fix_features)
7367 features = dev->netdev_ops->ndo_fix_features(dev, features);
7369 /* driver might be less strict about feature dependencies */
7370 features = netdev_fix_features(dev, features);
7372 /* some features can't be enabled if they're off an an upper device */
7373 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7374 features = netdev_sync_upper_features(dev, upper, features);
7376 if (dev->features == features)
7379 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7380 &dev->features, &features);
7382 if (dev->netdev_ops->ndo_set_features)
7383 err = dev->netdev_ops->ndo_set_features(dev, features);
7387 if (unlikely(err < 0)) {
7389 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7390 err, &features, &dev->features);
7391 /* return non-0 since some features might have changed and
7392 * it's better to fire a spurious notification than miss it
7398 /* some features must be disabled on lower devices when disabled
7399 * on an upper device (think: bonding master or bridge)
7401 netdev_for_each_lower_dev(dev, lower, iter)
7402 netdev_sync_lower_features(dev, lower, features);
7405 netdev_features_t diff = features ^ dev->features;
7407 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
7408 /* udp_tunnel_{get,drop}_rx_info both need
7409 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
7410 * device, or they won't do anything.
7411 * Thus we need to update dev->features
7412 * *before* calling udp_tunnel_get_rx_info,
7413 * but *after* calling udp_tunnel_drop_rx_info.
7415 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
7416 dev->features = features;
7417 udp_tunnel_get_rx_info(dev);
7419 udp_tunnel_drop_rx_info(dev);
7423 dev->features = features;
7426 return err < 0 ? 0 : 1;
7430 * netdev_update_features - recalculate device features
7431 * @dev: the device to check
7433 * Recalculate dev->features set and send notifications if it
7434 * has changed. Should be called after driver or hardware dependent
7435 * conditions might have changed that influence the features.
7437 void netdev_update_features(struct net_device *dev)
7439 if (__netdev_update_features(dev))
7440 netdev_features_change(dev);
7442 EXPORT_SYMBOL(netdev_update_features);
7445 * netdev_change_features - recalculate device features
7446 * @dev: the device to check
7448 * Recalculate dev->features set and send notifications even
7449 * if they have not changed. Should be called instead of
7450 * netdev_update_features() if also dev->vlan_features might
7451 * have changed to allow the changes to be propagated to stacked
7454 void netdev_change_features(struct net_device *dev)
7456 __netdev_update_features(dev);
7457 netdev_features_change(dev);
7459 EXPORT_SYMBOL(netdev_change_features);
7462 * netif_stacked_transfer_operstate - transfer operstate
7463 * @rootdev: the root or lower level device to transfer state from
7464 * @dev: the device to transfer operstate to
7466 * Transfer operational state from root to device. This is normally
7467 * called when a stacking relationship exists between the root
7468 * device and the device(a leaf device).
7470 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7471 struct net_device *dev)
7473 if (rootdev->operstate == IF_OPER_DORMANT)
7474 netif_dormant_on(dev);
7476 netif_dormant_off(dev);
7478 if (netif_carrier_ok(rootdev))
7479 netif_carrier_on(dev);
7481 netif_carrier_off(dev);
7483 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7486 static int netif_alloc_rx_queues(struct net_device *dev)
7488 unsigned int i, count = dev->num_rx_queues;
7489 struct netdev_rx_queue *rx;
7490 size_t sz = count * sizeof(*rx);
7494 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7500 for (i = 0; i < count; i++)
7506 static void netdev_init_one_queue(struct net_device *dev,
7507 struct netdev_queue *queue, void *_unused)
7509 /* Initialize queue lock */
7510 spin_lock_init(&queue->_xmit_lock);
7511 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7512 queue->xmit_lock_owner = -1;
7513 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7516 dql_init(&queue->dql, HZ);
7520 static void netif_free_tx_queues(struct net_device *dev)
7525 static int netif_alloc_netdev_queues(struct net_device *dev)
7527 unsigned int count = dev->num_tx_queues;
7528 struct netdev_queue *tx;
7529 size_t sz = count * sizeof(*tx);
7531 if (count < 1 || count > 0xffff)
7534 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7540 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7541 spin_lock_init(&dev->tx_global_lock);
7546 void netif_tx_stop_all_queues(struct net_device *dev)
7550 for (i = 0; i < dev->num_tx_queues; i++) {
7551 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7553 netif_tx_stop_queue(txq);
7556 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7559 * register_netdevice - register a network device
7560 * @dev: device to register
7562 * Take a completed network device structure and add it to the kernel
7563 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7564 * chain. 0 is returned on success. A negative errno code is returned
7565 * on a failure to set up the device, or if the name is a duplicate.
7567 * Callers must hold the rtnl semaphore. You may want
7568 * register_netdev() instead of this.
7571 * The locking appears insufficient to guarantee two parallel registers
7572 * will not get the same name.
7575 int register_netdevice(struct net_device *dev)
7578 struct net *net = dev_net(dev);
7580 BUG_ON(dev_boot_phase);
7585 /* When net_device's are persistent, this will be fatal. */
7586 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7589 spin_lock_init(&dev->addr_list_lock);
7590 netdev_set_addr_lockdep_class(dev);
7592 ret = dev_get_valid_name(net, dev, dev->name);
7596 /* Init, if this function is available */
7597 if (dev->netdev_ops->ndo_init) {
7598 ret = dev->netdev_ops->ndo_init(dev);
7606 if (((dev->hw_features | dev->features) &
7607 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7608 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7609 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7610 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7617 dev->ifindex = dev_new_index(net);
7618 else if (__dev_get_by_index(net, dev->ifindex))
7621 /* Transfer changeable features to wanted_features and enable
7622 * software offloads (GSO and GRO).
7624 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7625 dev->features |= NETIF_F_SOFT_FEATURES;
7627 if (dev->netdev_ops->ndo_udp_tunnel_add) {
7628 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7629 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7632 dev->wanted_features = dev->features & dev->hw_features;
7634 if (!(dev->flags & IFF_LOOPBACK))
7635 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7637 /* If IPv4 TCP segmentation offload is supported we should also
7638 * allow the device to enable segmenting the frame with the option
7639 * of ignoring a static IP ID value. This doesn't enable the
7640 * feature itself but allows the user to enable it later.
7642 if (dev->hw_features & NETIF_F_TSO)
7643 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7644 if (dev->vlan_features & NETIF_F_TSO)
7645 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7646 if (dev->mpls_features & NETIF_F_TSO)
7647 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7648 if (dev->hw_enc_features & NETIF_F_TSO)
7649 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7651 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7653 dev->vlan_features |= NETIF_F_HIGHDMA;
7655 /* Make NETIF_F_SG inheritable to tunnel devices.
7657 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7659 /* Make NETIF_F_SG inheritable to MPLS.
7661 dev->mpls_features |= NETIF_F_SG;
7663 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7664 ret = notifier_to_errno(ret);
7668 ret = netdev_register_kobject(dev);
7670 dev->reg_state = NETREG_UNREGISTERED;
7673 dev->reg_state = NETREG_REGISTERED;
7675 __netdev_update_features(dev);
7678 * Default initial state at registry is that the
7679 * device is present.
7682 set_bit(__LINK_STATE_PRESENT, &dev->state);
7684 linkwatch_init_dev(dev);
7686 dev_init_scheduler(dev);
7688 list_netdevice(dev);
7689 add_device_randomness(dev->dev_addr, dev->addr_len);
7691 /* If the device has permanent device address, driver should
7692 * set dev_addr and also addr_assign_type should be set to
7693 * NET_ADDR_PERM (default value).
7695 if (dev->addr_assign_type == NET_ADDR_PERM)
7696 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7698 /* Notify protocols, that a new device appeared. */
7699 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7700 ret = notifier_to_errno(ret);
7702 rollback_registered(dev);
7705 dev->reg_state = NETREG_UNREGISTERED;
7706 /* We should put the kobject that hold in
7707 * netdev_unregister_kobject(), otherwise
7708 * the net device cannot be freed when
7709 * driver calls free_netdev(), because the
7710 * kobject is being hold.
7712 kobject_put(&dev->dev.kobj);
7715 * Prevent userspace races by waiting until the network
7716 * device is fully setup before sending notifications.
7718 if (!dev->rtnl_link_ops ||
7719 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7720 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7726 if (dev->netdev_ops->ndo_uninit)
7727 dev->netdev_ops->ndo_uninit(dev);
7728 if (dev->priv_destructor)
7729 dev->priv_destructor(dev);
7732 EXPORT_SYMBOL(register_netdevice);
7735 * init_dummy_netdev - init a dummy network device for NAPI
7736 * @dev: device to init
7738 * This takes a network device structure and initialize the minimum
7739 * amount of fields so it can be used to schedule NAPI polls without
7740 * registering a full blown interface. This is to be used by drivers
7741 * that need to tie several hardware interfaces to a single NAPI
7742 * poll scheduler due to HW limitations.
7744 int init_dummy_netdev(struct net_device *dev)
7746 /* Clear everything. Note we don't initialize spinlocks
7747 * are they aren't supposed to be taken by any of the
7748 * NAPI code and this dummy netdev is supposed to be
7749 * only ever used for NAPI polls
7751 memset(dev, 0, sizeof(struct net_device));
7753 /* make sure we BUG if trying to hit standard
7754 * register/unregister code path
7756 dev->reg_state = NETREG_DUMMY;
7758 /* NAPI wants this */
7759 INIT_LIST_HEAD(&dev->napi_list);
7761 /* a dummy interface is started by default */
7762 set_bit(__LINK_STATE_PRESENT, &dev->state);
7763 set_bit(__LINK_STATE_START, &dev->state);
7765 /* napi_busy_loop stats accounting wants this */
7766 dev_net_set(dev, &init_net);
7768 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7769 * because users of this 'device' dont need to change
7775 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7778 int dev_validate_mtu(struct net_device *dev, int new_mtu)
7780 /* MTU must be positive, and in range */
7781 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7782 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
7783 dev->name, new_mtu, dev->min_mtu);
7787 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7788 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
7789 dev->name, new_mtu, dev->max_mtu);
7796 * register_netdev - register a network device
7797 * @dev: device to register
7799 * Take a completed network device structure and add it to the kernel
7800 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7801 * chain. 0 is returned on success. A negative errno code is returned
7802 * on a failure to set up the device, or if the name is a duplicate.
7804 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7805 * and expands the device name if you passed a format string to
7808 int register_netdev(struct net_device *dev)
7813 err = register_netdevice(dev);
7817 EXPORT_SYMBOL(register_netdev);
7819 int netdev_refcnt_read(const struct net_device *dev)
7823 for_each_possible_cpu(i)
7824 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7827 EXPORT_SYMBOL(netdev_refcnt_read);
7830 * netdev_wait_allrefs - wait until all references are gone.
7831 * @dev: target net_device
7833 * This is called when unregistering network devices.
7835 * Any protocol or device that holds a reference should register
7836 * for netdevice notification, and cleanup and put back the
7837 * reference if they receive an UNREGISTER event.
7838 * We can get stuck here if buggy protocols don't correctly
7841 static void netdev_wait_allrefs(struct net_device *dev)
7843 unsigned long rebroadcast_time, warning_time;
7846 linkwatch_forget_dev(dev);
7848 rebroadcast_time = warning_time = jiffies;
7849 refcnt = netdev_refcnt_read(dev);
7851 while (refcnt != 0) {
7852 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7855 /* Rebroadcast unregister notification */
7856 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7862 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7863 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7865 /* We must not have linkwatch events
7866 * pending on unregister. If this
7867 * happens, we simply run the queue
7868 * unscheduled, resulting in a noop
7871 linkwatch_run_queue();
7876 rebroadcast_time = jiffies;
7881 refcnt = netdev_refcnt_read(dev);
7883 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
7884 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7886 warning_time = jiffies;
7895 * register_netdevice(x1);
7896 * register_netdevice(x2);
7898 * unregister_netdevice(y1);
7899 * unregister_netdevice(y2);
7905 * We are invoked by rtnl_unlock().
7906 * This allows us to deal with problems:
7907 * 1) We can delete sysfs objects which invoke hotplug
7908 * without deadlocking with linkwatch via keventd.
7909 * 2) Since we run with the RTNL semaphore not held, we can sleep
7910 * safely in order to wait for the netdev refcnt to drop to zero.
7912 * We must not return until all unregister events added during
7913 * the interval the lock was held have been completed.
7915 void netdev_run_todo(void)
7917 struct list_head list;
7919 /* Snapshot list, allow later requests */
7920 list_replace_init(&net_todo_list, &list);
7925 /* Wait for rcu callbacks to finish before next phase */
7926 if (!list_empty(&list))
7929 while (!list_empty(&list)) {
7930 struct net_device *dev
7931 = list_first_entry(&list, struct net_device, todo_list);
7932 list_del(&dev->todo_list);
7935 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7938 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7939 pr_err("network todo '%s' but state %d\n",
7940 dev->name, dev->reg_state);
7945 dev->reg_state = NETREG_UNREGISTERED;
7947 netdev_wait_allrefs(dev);
7950 BUG_ON(netdev_refcnt_read(dev));
7951 BUG_ON(!list_empty(&dev->ptype_all));
7952 BUG_ON(!list_empty(&dev->ptype_specific));
7953 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7954 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7955 WARN_ON(dev->dn_ptr);
7957 if (dev->priv_destructor)
7958 dev->priv_destructor(dev);
7959 if (dev->needs_free_netdev)
7962 /* Report a network device has been unregistered */
7964 dev_net(dev)->dev_unreg_count--;
7966 wake_up(&netdev_unregistering_wq);
7968 /* Free network device */
7969 kobject_put(&dev->dev.kobj);
7973 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7974 * all the same fields in the same order as net_device_stats, with only
7975 * the type differing, but rtnl_link_stats64 may have additional fields
7976 * at the end for newer counters.
7978 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7979 const struct net_device_stats *netdev_stats)
7981 #if BITS_PER_LONG == 64
7982 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7983 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
7984 /* zero out counters that only exist in rtnl_link_stats64 */
7985 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7986 sizeof(*stats64) - sizeof(*netdev_stats));
7988 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7989 const unsigned long *src = (const unsigned long *)netdev_stats;
7990 u64 *dst = (u64 *)stats64;
7992 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7993 for (i = 0; i < n; i++)
7995 /* zero out counters that only exist in rtnl_link_stats64 */
7996 memset((char *)stats64 + n * sizeof(u64), 0,
7997 sizeof(*stats64) - n * sizeof(u64));
8000 EXPORT_SYMBOL(netdev_stats_to_stats64);
8003 * dev_get_stats - get network device statistics
8004 * @dev: device to get statistics from
8005 * @storage: place to store stats
8007 * Get network statistics from device. Return @storage.
8008 * The device driver may provide its own method by setting
8009 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8010 * otherwise the internal statistics structure is used.
8012 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
8013 struct rtnl_link_stats64 *storage)
8015 const struct net_device_ops *ops = dev->netdev_ops;
8017 if (ops->ndo_get_stats64) {
8018 memset(storage, 0, sizeof(*storage));
8019 ops->ndo_get_stats64(dev, storage);
8020 } else if (ops->ndo_get_stats) {
8021 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
8023 netdev_stats_to_stats64(storage, &dev->stats);
8025 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
8026 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
8027 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
8030 EXPORT_SYMBOL(dev_get_stats);
8032 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
8034 struct netdev_queue *queue = dev_ingress_queue(dev);
8036 #ifdef CONFIG_NET_CLS_ACT
8039 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
8042 netdev_init_one_queue(dev, queue, NULL);
8043 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
8044 queue->qdisc_sleeping = &noop_qdisc;
8045 rcu_assign_pointer(dev->ingress_queue, queue);
8050 static const struct ethtool_ops default_ethtool_ops;
8052 void netdev_set_default_ethtool_ops(struct net_device *dev,
8053 const struct ethtool_ops *ops)
8055 if (dev->ethtool_ops == &default_ethtool_ops)
8056 dev->ethtool_ops = ops;
8058 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
8060 void netdev_freemem(struct net_device *dev)
8062 char *addr = (char *)dev - dev->padded;
8068 * alloc_netdev_mqs - allocate network device
8069 * @sizeof_priv: size of private data to allocate space for
8070 * @name: device name format string
8071 * @name_assign_type: origin of device name
8072 * @setup: callback to initialize device
8073 * @txqs: the number of TX subqueues to allocate
8074 * @rxqs: the number of RX subqueues to allocate
8076 * Allocates a struct net_device with private data area for driver use
8077 * and performs basic initialization. Also allocates subqueue structs
8078 * for each queue on the device.
8080 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
8081 unsigned char name_assign_type,
8082 void (*setup)(struct net_device *),
8083 unsigned int txqs, unsigned int rxqs)
8085 struct net_device *dev;
8087 struct net_device *p;
8089 BUG_ON(strlen(name) >= sizeof(dev->name));
8092 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8098 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
8103 alloc_size = sizeof(struct net_device);
8105 /* ensure 32-byte alignment of private area */
8106 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
8107 alloc_size += sizeof_priv;
8109 /* ensure 32-byte alignment of whole construct */
8110 alloc_size += NETDEV_ALIGN - 1;
8112 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8116 dev = PTR_ALIGN(p, NETDEV_ALIGN);
8117 dev->padded = (char *)dev - (char *)p;
8119 dev->pcpu_refcnt = alloc_percpu(int);
8120 if (!dev->pcpu_refcnt)
8123 if (dev_addr_init(dev))
8129 dev_net_set(dev, &init_net);
8131 dev->gso_max_size = GSO_MAX_SIZE;
8132 dev->gso_max_segs = GSO_MAX_SEGS;
8134 INIT_LIST_HEAD(&dev->napi_list);
8135 INIT_LIST_HEAD(&dev->unreg_list);
8136 INIT_LIST_HEAD(&dev->close_list);
8137 INIT_LIST_HEAD(&dev->link_watch_list);
8138 INIT_LIST_HEAD(&dev->adj_list.upper);
8139 INIT_LIST_HEAD(&dev->adj_list.lower);
8140 INIT_LIST_HEAD(&dev->ptype_all);
8141 INIT_LIST_HEAD(&dev->ptype_specific);
8142 #ifdef CONFIG_NET_SCHED
8143 hash_init(dev->qdisc_hash);
8145 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
8148 if (!dev->tx_queue_len) {
8149 dev->priv_flags |= IFF_NO_QUEUE;
8150 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
8153 dev->num_tx_queues = txqs;
8154 dev->real_num_tx_queues = txqs;
8155 if (netif_alloc_netdev_queues(dev))
8159 dev->num_rx_queues = rxqs;
8160 dev->real_num_rx_queues = rxqs;
8161 if (netif_alloc_rx_queues(dev))
8165 strcpy(dev->name, name);
8166 dev->name_assign_type = name_assign_type;
8167 dev->group = INIT_NETDEV_GROUP;
8168 if (!dev->ethtool_ops)
8169 dev->ethtool_ops = &default_ethtool_ops;
8171 nf_hook_ingress_init(dev);
8180 free_percpu(dev->pcpu_refcnt);
8182 netdev_freemem(dev);
8185 EXPORT_SYMBOL(alloc_netdev_mqs);
8188 * free_netdev - free network device
8191 * This function does the last stage of destroying an allocated device
8192 * interface. The reference to the device object is released. If this
8193 * is the last reference then it will be freed.Must be called in process
8196 void free_netdev(struct net_device *dev)
8198 struct napi_struct *p, *n;
8199 struct bpf_prog *prog;
8202 netif_free_tx_queues(dev);
8207 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
8209 /* Flush device addresses */
8210 dev_addr_flush(dev);
8212 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
8215 free_percpu(dev->pcpu_refcnt);
8216 dev->pcpu_refcnt = NULL;
8218 prog = rcu_dereference_protected(dev->xdp_prog, 1);
8221 static_key_slow_dec(&generic_xdp_needed);
8224 /* Compatibility with error handling in drivers */
8225 if (dev->reg_state == NETREG_UNINITIALIZED) {
8226 netdev_freemem(dev);
8230 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
8231 dev->reg_state = NETREG_RELEASED;
8233 /* will free via device release */
8234 put_device(&dev->dev);
8236 EXPORT_SYMBOL(free_netdev);
8239 * synchronize_net - Synchronize with packet receive processing
8241 * Wait for packets currently being received to be done.
8242 * Does not block later packets from starting.
8244 void synchronize_net(void)
8247 if (rtnl_is_locked())
8248 synchronize_rcu_expedited();
8252 EXPORT_SYMBOL(synchronize_net);
8255 * unregister_netdevice_queue - remove device from the kernel
8259 * This function shuts down a device interface and removes it
8260 * from the kernel tables.
8261 * If head not NULL, device is queued to be unregistered later.
8263 * Callers must hold the rtnl semaphore. You may want
8264 * unregister_netdev() instead of this.
8267 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
8272 list_move_tail(&dev->unreg_list, head);
8274 rollback_registered(dev);
8275 /* Finish processing unregister after unlock */
8279 EXPORT_SYMBOL(unregister_netdevice_queue);
8282 * unregister_netdevice_many - unregister many devices
8283 * @head: list of devices
8285 * Note: As most callers use a stack allocated list_head,
8286 * we force a list_del() to make sure stack wont be corrupted later.
8288 void unregister_netdevice_many(struct list_head *head)
8290 struct net_device *dev;
8292 if (!list_empty(head)) {
8293 rollback_registered_many(head);
8294 list_for_each_entry(dev, head, unreg_list)
8299 EXPORT_SYMBOL(unregister_netdevice_many);
8302 * unregister_netdev - remove device from the kernel
8305 * This function shuts down a device interface and removes it
8306 * from the kernel tables.
8308 * This is just a wrapper for unregister_netdevice that takes
8309 * the rtnl semaphore. In general you want to use this and not
8310 * unregister_netdevice.
8312 void unregister_netdev(struct net_device *dev)
8315 unregister_netdevice(dev);
8318 EXPORT_SYMBOL(unregister_netdev);
8321 * dev_change_net_namespace - move device to different nethost namespace
8323 * @net: network namespace
8324 * @pat: If not NULL name pattern to try if the current device name
8325 * is already taken in the destination network namespace.
8327 * This function shuts down a device interface and moves it
8328 * to a new network namespace. On success 0 is returned, on
8329 * a failure a netagive errno code is returned.
8331 * Callers must hold the rtnl semaphore.
8334 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
8340 /* Don't allow namespace local devices to be moved. */
8342 if (dev->features & NETIF_F_NETNS_LOCAL)
8345 /* Ensure the device has been registrered */
8346 if (dev->reg_state != NETREG_REGISTERED)
8349 /* Get out if there is nothing todo */
8351 if (net_eq(dev_net(dev), net))
8354 /* Pick the destination device name, and ensure
8355 * we can use it in the destination network namespace.
8358 if (__dev_get_by_name(net, dev->name)) {
8359 /* We get here if we can't use the current device name */
8362 err = dev_get_valid_name(net, dev, pat);
8368 * And now a mini version of register_netdevice unregister_netdevice.
8371 /* If device is running close it first. */
8374 /* And unlink it from device chain */
8375 unlist_netdevice(dev);
8379 /* Shutdown queueing discipline. */
8382 /* Notify protocols, that we are about to destroy
8383 * this device. They should clean all the things.
8385 * Note that dev->reg_state stays at NETREG_REGISTERED.
8386 * This is wanted because this way 8021q and macvlan know
8387 * the device is just moving and can keep their slaves up.
8389 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8391 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
8392 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
8395 * Flush the unicast and multicast chains
8400 /* Send a netdev-removed uevent to the old namespace */
8401 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
8402 netdev_adjacent_del_links(dev);
8404 /* Actually switch the network namespace */
8405 dev_net_set(dev, net);
8407 /* If there is an ifindex conflict assign a new one */
8408 if (__dev_get_by_index(net, dev->ifindex))
8409 dev->ifindex = dev_new_index(net);
8411 /* Send a netdev-add uevent to the new namespace */
8412 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
8413 netdev_adjacent_add_links(dev);
8415 /* Fixup kobjects */
8416 err = device_rename(&dev->dev, dev->name);
8419 /* Add the device back in the hashes */
8420 list_netdevice(dev);
8422 /* Notify protocols, that a new device appeared. */
8423 call_netdevice_notifiers(NETDEV_REGISTER, dev);
8426 * Prevent userspace races by waiting until the network
8427 * device is fully setup before sending notifications.
8429 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8436 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8438 static int dev_cpu_dead(unsigned int oldcpu)
8440 struct sk_buff **list_skb;
8441 struct sk_buff *skb;
8443 struct softnet_data *sd, *oldsd, *remsd = NULL;
8445 local_irq_disable();
8446 cpu = smp_processor_id();
8447 sd = &per_cpu(softnet_data, cpu);
8448 oldsd = &per_cpu(softnet_data, oldcpu);
8450 /* Find end of our completion_queue. */
8451 list_skb = &sd->completion_queue;
8453 list_skb = &(*list_skb)->next;
8454 /* Append completion queue from offline CPU. */
8455 *list_skb = oldsd->completion_queue;
8456 oldsd->completion_queue = NULL;
8458 /* Append output queue from offline CPU. */
8459 if (oldsd->output_queue) {
8460 *sd->output_queue_tailp = oldsd->output_queue;
8461 sd->output_queue_tailp = oldsd->output_queue_tailp;
8462 oldsd->output_queue = NULL;
8463 oldsd->output_queue_tailp = &oldsd->output_queue;
8465 /* Append NAPI poll list from offline CPU, with one exception :
8466 * process_backlog() must be called by cpu owning percpu backlog.
8467 * We properly handle process_queue & input_pkt_queue later.
8469 while (!list_empty(&oldsd->poll_list)) {
8470 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8474 list_del_init(&napi->poll_list);
8475 if (napi->poll == process_backlog)
8478 ____napi_schedule(sd, napi);
8481 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8485 remsd = oldsd->rps_ipi_list;
8486 oldsd->rps_ipi_list = NULL;
8488 /* send out pending IPI's on offline CPU */
8489 net_rps_send_ipi(remsd);
8491 /* Process offline CPU's input_pkt_queue */
8492 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8494 input_queue_head_incr(oldsd);
8496 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8498 input_queue_head_incr(oldsd);
8505 * netdev_increment_features - increment feature set by one
8506 * @all: current feature set
8507 * @one: new feature set
8508 * @mask: mask feature set
8510 * Computes a new feature set after adding a device with feature set
8511 * @one to the master device with current feature set @all. Will not
8512 * enable anything that is off in @mask. Returns the new feature set.
8514 netdev_features_t netdev_increment_features(netdev_features_t all,
8515 netdev_features_t one, netdev_features_t mask)
8517 if (mask & NETIF_F_HW_CSUM)
8518 mask |= NETIF_F_CSUM_MASK;
8519 mask |= NETIF_F_VLAN_CHALLENGED;
8521 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8522 all &= one | ~NETIF_F_ALL_FOR_ALL;
8524 /* If one device supports hw checksumming, set for all. */
8525 if (all & NETIF_F_HW_CSUM)
8526 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8530 EXPORT_SYMBOL(netdev_increment_features);
8532 static struct hlist_head * __net_init netdev_create_hash(void)
8535 struct hlist_head *hash;
8537 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8539 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8540 INIT_HLIST_HEAD(&hash[i]);
8545 /* Initialize per network namespace state */
8546 static int __net_init netdev_init(struct net *net)
8548 if (net != &init_net)
8549 INIT_LIST_HEAD(&net->dev_base_head);
8551 net->dev_name_head = netdev_create_hash();
8552 if (net->dev_name_head == NULL)
8555 net->dev_index_head = netdev_create_hash();
8556 if (net->dev_index_head == NULL)
8562 kfree(net->dev_name_head);
8568 * netdev_drivername - network driver for the device
8569 * @dev: network device
8571 * Determine network driver for device.
8573 const char *netdev_drivername(const struct net_device *dev)
8575 const struct device_driver *driver;
8576 const struct device *parent;
8577 const char *empty = "";
8579 parent = dev->dev.parent;
8583 driver = parent->driver;
8584 if (driver && driver->name)
8585 return driver->name;
8589 static void __netdev_printk(const char *level, const struct net_device *dev,
8590 struct va_format *vaf)
8592 if (dev && dev->dev.parent) {
8593 dev_printk_emit(level[1] - '0',
8596 dev_driver_string(dev->dev.parent),
8597 dev_name(dev->dev.parent),
8598 netdev_name(dev), netdev_reg_state(dev),
8601 printk("%s%s%s: %pV",
8602 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8604 printk("%s(NULL net_device): %pV", level, vaf);
8608 void netdev_printk(const char *level, const struct net_device *dev,
8609 const char *format, ...)
8611 struct va_format vaf;
8614 va_start(args, format);
8619 __netdev_printk(level, dev, &vaf);
8623 EXPORT_SYMBOL(netdev_printk);
8625 #define define_netdev_printk_level(func, level) \
8626 void func(const struct net_device *dev, const char *fmt, ...) \
8628 struct va_format vaf; \
8631 va_start(args, fmt); \
8636 __netdev_printk(level, dev, &vaf); \
8640 EXPORT_SYMBOL(func);
8642 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8643 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8644 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8645 define_netdev_printk_level(netdev_err, KERN_ERR);
8646 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8647 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8648 define_netdev_printk_level(netdev_info, KERN_INFO);
8650 static void __net_exit netdev_exit(struct net *net)
8652 kfree(net->dev_name_head);
8653 kfree(net->dev_index_head);
8656 static struct pernet_operations __net_initdata netdev_net_ops = {
8657 .init = netdev_init,
8658 .exit = netdev_exit,
8661 static void __net_exit default_device_exit(struct net *net)
8663 struct net_device *dev, *aux;
8665 * Push all migratable network devices back to the
8666 * initial network namespace
8669 for_each_netdev_safe(net, dev, aux) {
8671 char fb_name[IFNAMSIZ];
8673 /* Ignore unmoveable devices (i.e. loopback) */
8674 if (dev->features & NETIF_F_NETNS_LOCAL)
8677 /* Leave virtual devices for the generic cleanup */
8678 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
8681 /* Push remaining network devices to init_net */
8682 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8683 if (__dev_get_by_name(&init_net, fb_name))
8684 snprintf(fb_name, IFNAMSIZ, "dev%%d");
8685 err = dev_change_net_namespace(dev, &init_net, fb_name);
8687 pr_emerg("%s: failed to move %s to init_net: %d\n",
8688 __func__, dev->name, err);
8695 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8697 /* Return with the rtnl_lock held when there are no network
8698 * devices unregistering in any network namespace in net_list.
8702 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8704 add_wait_queue(&netdev_unregistering_wq, &wait);
8706 unregistering = false;
8708 list_for_each_entry(net, net_list, exit_list) {
8709 if (net->dev_unreg_count > 0) {
8710 unregistering = true;
8718 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8720 remove_wait_queue(&netdev_unregistering_wq, &wait);
8723 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8725 /* At exit all network devices most be removed from a network
8726 * namespace. Do this in the reverse order of registration.
8727 * Do this across as many network namespaces as possible to
8728 * improve batching efficiency.
8730 struct net_device *dev;
8732 LIST_HEAD(dev_kill_list);
8734 /* To prevent network device cleanup code from dereferencing
8735 * loopback devices or network devices that have been freed
8736 * wait here for all pending unregistrations to complete,
8737 * before unregistring the loopback device and allowing the
8738 * network namespace be freed.
8740 * The netdev todo list containing all network devices
8741 * unregistrations that happen in default_device_exit_batch
8742 * will run in the rtnl_unlock() at the end of
8743 * default_device_exit_batch.
8745 rtnl_lock_unregistering(net_list);
8746 list_for_each_entry(net, net_list, exit_list) {
8747 for_each_netdev_reverse(net, dev) {
8748 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8749 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8751 unregister_netdevice_queue(dev, &dev_kill_list);
8754 unregister_netdevice_many(&dev_kill_list);
8758 static struct pernet_operations __net_initdata default_device_ops = {
8759 .exit = default_device_exit,
8760 .exit_batch = default_device_exit_batch,
8764 * Initialize the DEV module. At boot time this walks the device list and
8765 * unhooks any devices that fail to initialise (normally hardware not
8766 * present) and leaves us with a valid list of present and active devices.
8771 * This is called single threaded during boot, so no need
8772 * to take the rtnl semaphore.
8774 static int __init net_dev_init(void)
8776 int i, rc = -ENOMEM;
8778 BUG_ON(!dev_boot_phase);
8780 if (dev_proc_init())
8783 if (netdev_kobject_init())
8786 INIT_LIST_HEAD(&ptype_all);
8787 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8788 INIT_LIST_HEAD(&ptype_base[i]);
8790 INIT_LIST_HEAD(&offload_base);
8792 if (register_pernet_subsys(&netdev_net_ops))
8796 * Initialise the packet receive queues.
8799 for_each_possible_cpu(i) {
8800 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8801 struct softnet_data *sd = &per_cpu(softnet_data, i);
8803 INIT_WORK(flush, flush_backlog);
8805 skb_queue_head_init(&sd->input_pkt_queue);
8806 skb_queue_head_init(&sd->process_queue);
8807 INIT_LIST_HEAD(&sd->poll_list);
8808 sd->output_queue_tailp = &sd->output_queue;
8810 sd->csd.func = rps_trigger_softirq;
8815 sd->backlog.poll = process_backlog;
8816 sd->backlog.weight = weight_p;
8821 /* The loopback device is special if any other network devices
8822 * is present in a network namespace the loopback device must
8823 * be present. Since we now dynamically allocate and free the
8824 * loopback device ensure this invariant is maintained by
8825 * keeping the loopback device as the first device on the
8826 * list of network devices. Ensuring the loopback devices
8827 * is the first device that appears and the last network device
8830 if (register_pernet_device(&loopback_net_ops))
8833 if (register_pernet_device(&default_device_ops))
8836 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8837 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8839 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8840 NULL, dev_cpu_dead);
8847 subsys_initcall(net_dev_init);