2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/sched/mm.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
105 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <linux/pci.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_ingress.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
149 #include "net-sysfs.h"
151 #define MAX_GRO_SKBS 8
153 /* This should be increased if a protocol with a bigger head is added. */
154 #define GRO_MAX_HEAD (MAX_HEADER + 128)
156 static DEFINE_SPINLOCK(ptype_lock);
157 static DEFINE_SPINLOCK(offload_lock);
158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
159 struct list_head ptype_all __read_mostly; /* Taps */
160 static struct list_head offload_base __read_mostly;
162 static int netif_rx_internal(struct sk_buff *skb);
163 static int call_netdevice_notifiers_info(unsigned long val,
164 struct netdev_notifier_info *info);
165 static struct napi_struct *napi_by_id(unsigned int napi_id);
168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base_head list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 DEFINE_RWLOCK(dev_base_lock);
187 EXPORT_SYMBOL(dev_base_lock);
189 static DEFINE_MUTEX(ifalias_mutex);
191 /* protects napi_hash addition/deletion and napi_gen_id */
192 static DEFINE_SPINLOCK(napi_hash_lock);
194 static unsigned int napi_gen_id = NR_CPUS;
195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
197 static seqcount_t devnet_rename_seq;
199 static inline void dev_base_seq_inc(struct net *net)
201 while (++net->dev_base_seq == 0)
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217 static inline void rps_lock(struct softnet_data *sd)
220 spin_lock(&sd->input_pkt_queue.lock);
224 static inline void rps_unlock(struct softnet_data *sd)
227 spin_unlock(&sd->input_pkt_queue.lock);
231 /* Device list insertion */
232 static void list_netdevice(struct net_device *dev)
234 struct net *net = dev_net(dev);
238 write_lock_bh(&dev_base_lock);
239 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
240 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
241 hlist_add_head_rcu(&dev->index_hlist,
242 dev_index_hash(net, dev->ifindex));
243 write_unlock_bh(&dev_base_lock);
245 dev_base_seq_inc(net);
248 /* Device list removal
249 * caller must respect a RCU grace period before freeing/reusing dev
251 static void unlist_netdevice(struct net_device *dev)
255 /* Unlink dev from the device chain */
256 write_lock_bh(&dev_base_lock);
257 list_del_rcu(&dev->dev_list);
258 hlist_del_rcu(&dev->name_hlist);
259 hlist_del_rcu(&dev->index_hlist);
260 write_unlock_bh(&dev_base_lock);
262 dev_base_seq_inc(dev_net(dev));
269 static RAW_NOTIFIER_HEAD(netdev_chain);
272 * Device drivers call our routines to queue packets here. We empty the
273 * queue in the local softnet handler.
276 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
277 EXPORT_PER_CPU_SYMBOL(softnet_data);
279 #ifdef CONFIG_LOCKDEP
281 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
282 * according to dev->type
284 static const unsigned short netdev_lock_type[] = {
285 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
286 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
287 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
288 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
289 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
290 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
291 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
292 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
293 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
294 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
295 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
296 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
297 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
298 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
299 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
301 static const char *const netdev_lock_name[] = {
302 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
303 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
304 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
305 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
306 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
307 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
308 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
309 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
310 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
311 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
312 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
313 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
314 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
315 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
316 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
318 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
319 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
321 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
325 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
326 if (netdev_lock_type[i] == dev_type)
328 /* the last key is used by default */
329 return ARRAY_SIZE(netdev_lock_type) - 1;
332 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
333 unsigned short dev_type)
337 i = netdev_lock_pos(dev_type);
338 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
339 netdev_lock_name[i]);
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
346 i = netdev_lock_pos(dev->type);
347 lockdep_set_class_and_name(&dev->addr_list_lock,
348 &netdev_addr_lock_key[i],
349 netdev_lock_name[i]);
352 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
353 unsigned short dev_type)
356 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
361 /*******************************************************************************
363 * Protocol management and registration routines
365 *******************************************************************************/
369 * Add a protocol ID to the list. Now that the input handler is
370 * smarter we can dispense with all the messy stuff that used to be
373 * BEWARE!!! Protocol handlers, mangling input packets,
374 * MUST BE last in hash buckets and checking protocol handlers
375 * MUST start from promiscuous ptype_all chain in net_bh.
376 * It is true now, do not change it.
377 * Explanation follows: if protocol handler, mangling packet, will
378 * be the first on list, it is not able to sense, that packet
379 * is cloned and should be copied-on-write, so that it will
380 * change it and subsequent readers will get broken packet.
384 static inline struct list_head *ptype_head(const struct packet_type *pt)
386 if (pt->type == htons(ETH_P_ALL))
387 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
389 return pt->dev ? &pt->dev->ptype_specific :
390 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
394 * dev_add_pack - add packet handler
395 * @pt: packet type declaration
397 * Add a protocol handler to the networking stack. The passed &packet_type
398 * is linked into kernel lists and may not be freed until it has been
399 * removed from the kernel lists.
401 * This call does not sleep therefore it can not
402 * guarantee all CPU's that are in middle of receiving packets
403 * will see the new packet type (until the next received packet).
406 void dev_add_pack(struct packet_type *pt)
408 struct list_head *head = ptype_head(pt);
410 spin_lock(&ptype_lock);
411 list_add_rcu(&pt->list, head);
412 spin_unlock(&ptype_lock);
414 EXPORT_SYMBOL(dev_add_pack);
417 * __dev_remove_pack - remove packet handler
418 * @pt: packet type declaration
420 * Remove a protocol handler that was previously added to the kernel
421 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
422 * from the kernel lists and can be freed or reused once this function
425 * The packet type might still be in use by receivers
426 * and must not be freed until after all the CPU's have gone
427 * through a quiescent state.
429 void __dev_remove_pack(struct packet_type *pt)
431 struct list_head *head = ptype_head(pt);
432 struct packet_type *pt1;
434 spin_lock(&ptype_lock);
436 list_for_each_entry(pt1, head, list) {
438 list_del_rcu(&pt->list);
443 pr_warn("dev_remove_pack: %p not found\n", pt);
445 spin_unlock(&ptype_lock);
447 EXPORT_SYMBOL(__dev_remove_pack);
450 * dev_remove_pack - remove packet handler
451 * @pt: packet type declaration
453 * Remove a protocol handler that was previously added to the kernel
454 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
455 * from the kernel lists and can be freed or reused once this function
458 * This call sleeps to guarantee that no CPU is looking at the packet
461 void dev_remove_pack(struct packet_type *pt)
463 __dev_remove_pack(pt);
467 EXPORT_SYMBOL(dev_remove_pack);
471 * dev_add_offload - register offload handlers
472 * @po: protocol offload declaration
474 * Add protocol offload handlers to the networking stack. The passed
475 * &proto_offload is linked into kernel lists and may not be freed until
476 * it has been removed from the kernel lists.
478 * This call does not sleep therefore it can not
479 * guarantee all CPU's that are in middle of receiving packets
480 * will see the new offload handlers (until the next received packet).
482 void dev_add_offload(struct packet_offload *po)
484 struct packet_offload *elem;
486 spin_lock(&offload_lock);
487 list_for_each_entry(elem, &offload_base, list) {
488 if (po->priority < elem->priority)
491 list_add_rcu(&po->list, elem->list.prev);
492 spin_unlock(&offload_lock);
494 EXPORT_SYMBOL(dev_add_offload);
497 * __dev_remove_offload - remove offload handler
498 * @po: packet offload declaration
500 * Remove a protocol offload handler that was previously added to the
501 * kernel offload handlers by dev_add_offload(). The passed &offload_type
502 * is removed from the kernel lists and can be freed or reused once this
505 * The packet type might still be in use by receivers
506 * and must not be freed until after all the CPU's have gone
507 * through a quiescent state.
509 static void __dev_remove_offload(struct packet_offload *po)
511 struct list_head *head = &offload_base;
512 struct packet_offload *po1;
514 spin_lock(&offload_lock);
516 list_for_each_entry(po1, head, list) {
518 list_del_rcu(&po->list);
523 pr_warn("dev_remove_offload: %p not found\n", po);
525 spin_unlock(&offload_lock);
529 * dev_remove_offload - remove packet offload handler
530 * @po: packet offload declaration
532 * Remove a packet offload handler that was previously added to the kernel
533 * offload handlers by dev_add_offload(). The passed &offload_type is
534 * removed from the kernel lists and can be freed or reused once this
537 * This call sleeps to guarantee that no CPU is looking at the packet
540 void dev_remove_offload(struct packet_offload *po)
542 __dev_remove_offload(po);
546 EXPORT_SYMBOL(dev_remove_offload);
548 /******************************************************************************
550 * Device Boot-time Settings Routines
552 ******************************************************************************/
554 /* Boot time configuration table */
555 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
558 * netdev_boot_setup_add - add new setup entry
559 * @name: name of the device
560 * @map: configured settings for the device
562 * Adds new setup entry to the dev_boot_setup list. The function
563 * returns 0 on error and 1 on success. This is a generic routine to
566 static int netdev_boot_setup_add(char *name, struct ifmap *map)
568 struct netdev_boot_setup *s;
572 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
573 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
574 memset(s[i].name, 0, sizeof(s[i].name));
575 strlcpy(s[i].name, name, IFNAMSIZ);
576 memcpy(&s[i].map, map, sizeof(s[i].map));
581 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
585 * netdev_boot_setup_check - check boot time settings
586 * @dev: the netdevice
588 * Check boot time settings for the device.
589 * The found settings are set for the device to be used
590 * later in the device probing.
591 * Returns 0 if no settings found, 1 if they are.
593 int netdev_boot_setup_check(struct net_device *dev)
595 struct netdev_boot_setup *s = dev_boot_setup;
598 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
599 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
600 !strcmp(dev->name, s[i].name)) {
601 dev->irq = s[i].map.irq;
602 dev->base_addr = s[i].map.base_addr;
603 dev->mem_start = s[i].map.mem_start;
604 dev->mem_end = s[i].map.mem_end;
610 EXPORT_SYMBOL(netdev_boot_setup_check);
614 * netdev_boot_base - get address from boot time settings
615 * @prefix: prefix for network device
616 * @unit: id for network device
618 * Check boot time settings for the base address of device.
619 * The found settings are set for the device to be used
620 * later in the device probing.
621 * Returns 0 if no settings found.
623 unsigned long netdev_boot_base(const char *prefix, int unit)
625 const struct netdev_boot_setup *s = dev_boot_setup;
629 sprintf(name, "%s%d", prefix, unit);
632 * If device already registered then return base of 1
633 * to indicate not to probe for this interface
635 if (__dev_get_by_name(&init_net, name))
638 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
639 if (!strcmp(name, s[i].name))
640 return s[i].map.base_addr;
645 * Saves at boot time configured settings for any netdevice.
647 int __init netdev_boot_setup(char *str)
652 str = get_options(str, ARRAY_SIZE(ints), ints);
657 memset(&map, 0, sizeof(map));
661 map.base_addr = ints[2];
663 map.mem_start = ints[3];
665 map.mem_end = ints[4];
667 /* Add new entry to the list */
668 return netdev_boot_setup_add(str, &map);
671 __setup("netdev=", netdev_boot_setup);
673 /*******************************************************************************
675 * Device Interface Subroutines
677 *******************************************************************************/
680 * dev_get_iflink - get 'iflink' value of a interface
681 * @dev: targeted interface
683 * Indicates the ifindex the interface is linked to.
684 * Physical interfaces have the same 'ifindex' and 'iflink' values.
687 int dev_get_iflink(const struct net_device *dev)
689 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
690 return dev->netdev_ops->ndo_get_iflink(dev);
694 EXPORT_SYMBOL(dev_get_iflink);
697 * dev_fill_metadata_dst - Retrieve tunnel egress information.
698 * @dev: targeted interface
701 * For better visibility of tunnel traffic OVS needs to retrieve
702 * egress tunnel information for a packet. Following API allows
703 * user to get this info.
705 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
707 struct ip_tunnel_info *info;
709 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
712 info = skb_tunnel_info_unclone(skb);
715 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
718 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
720 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
723 * __dev_get_by_name - find a device by its name
724 * @net: the applicable net namespace
725 * @name: name to find
727 * Find an interface by name. Must be called under RTNL semaphore
728 * or @dev_base_lock. If the name is found a pointer to the device
729 * is returned. If the name is not found then %NULL is returned. The
730 * reference counters are not incremented so the caller must be
731 * careful with locks.
734 struct net_device *__dev_get_by_name(struct net *net, const char *name)
736 struct net_device *dev;
737 struct hlist_head *head = dev_name_hash(net, name);
739 hlist_for_each_entry(dev, head, name_hlist)
740 if (!strncmp(dev->name, name, IFNAMSIZ))
745 EXPORT_SYMBOL(__dev_get_by_name);
748 * dev_get_by_name_rcu - find a device by its name
749 * @net: the applicable net namespace
750 * @name: name to find
752 * Find an interface by name.
753 * If the name is found a pointer to the device is returned.
754 * If the name is not found then %NULL is returned.
755 * The reference counters are not incremented so the caller must be
756 * careful with locks. The caller must hold RCU lock.
759 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
761 struct net_device *dev;
762 struct hlist_head *head = dev_name_hash(net, name);
764 hlist_for_each_entry_rcu(dev, head, name_hlist)
765 if (!strncmp(dev->name, name, IFNAMSIZ))
770 EXPORT_SYMBOL(dev_get_by_name_rcu);
773 * dev_get_by_name - find a device by its name
774 * @net: the applicable net namespace
775 * @name: name to find
777 * Find an interface by name. This can be called from any
778 * context and does its own locking. The returned handle has
779 * the usage count incremented and the caller must use dev_put() to
780 * release it when it is no longer needed. %NULL is returned if no
781 * matching device is found.
784 struct net_device *dev_get_by_name(struct net *net, const char *name)
786 struct net_device *dev;
789 dev = dev_get_by_name_rcu(net, name);
795 EXPORT_SYMBOL(dev_get_by_name);
798 * __dev_get_by_index - find a device by its ifindex
799 * @net: the applicable net namespace
800 * @ifindex: index of device
802 * Search for an interface by index. Returns %NULL if the device
803 * is not found or a pointer to the device. The device has not
804 * had its reference counter increased so the caller must be careful
805 * about locking. The caller must hold either the RTNL semaphore
809 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
811 struct net_device *dev;
812 struct hlist_head *head = dev_index_hash(net, ifindex);
814 hlist_for_each_entry(dev, head, index_hlist)
815 if (dev->ifindex == ifindex)
820 EXPORT_SYMBOL(__dev_get_by_index);
823 * dev_get_by_index_rcu - find a device by its ifindex
824 * @net: the applicable net namespace
825 * @ifindex: index of device
827 * Search for an interface by index. Returns %NULL if the device
828 * is not found or a pointer to the device. The device has not
829 * had its reference counter increased so the caller must be careful
830 * about locking. The caller must hold RCU lock.
833 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
835 struct net_device *dev;
836 struct hlist_head *head = dev_index_hash(net, ifindex);
838 hlist_for_each_entry_rcu(dev, head, index_hlist)
839 if (dev->ifindex == ifindex)
844 EXPORT_SYMBOL(dev_get_by_index_rcu);
848 * dev_get_by_index - find a device by its ifindex
849 * @net: the applicable net namespace
850 * @ifindex: index of device
852 * Search for an interface by index. Returns NULL if the device
853 * is not found or a pointer to the device. The device returned has
854 * had a reference added and the pointer is safe until the user calls
855 * dev_put to indicate they have finished with it.
858 struct net_device *dev_get_by_index(struct net *net, int ifindex)
860 struct net_device *dev;
863 dev = dev_get_by_index_rcu(net, ifindex);
869 EXPORT_SYMBOL(dev_get_by_index);
872 * dev_get_by_napi_id - find a device by napi_id
873 * @napi_id: ID of the NAPI struct
875 * Search for an interface by NAPI ID. Returns %NULL if the device
876 * is not found or a pointer to the device. The device has not had
877 * its reference counter increased so the caller must be careful
878 * about locking. The caller must hold RCU lock.
881 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
883 struct napi_struct *napi;
885 WARN_ON_ONCE(!rcu_read_lock_held());
887 if (napi_id < MIN_NAPI_ID)
890 napi = napi_by_id(napi_id);
892 return napi ? napi->dev : NULL;
894 EXPORT_SYMBOL(dev_get_by_napi_id);
897 * netdev_get_name - get a netdevice name, knowing its ifindex.
898 * @net: network namespace
899 * @name: a pointer to the buffer where the name will be stored.
900 * @ifindex: the ifindex of the interface to get the name from.
902 * The use of raw_seqcount_begin() and cond_resched() before
903 * retrying is required as we want to give the writers a chance
904 * to complete when CONFIG_PREEMPT is not set.
906 int netdev_get_name(struct net *net, char *name, int ifindex)
908 struct net_device *dev;
912 seq = raw_seqcount_begin(&devnet_rename_seq);
914 dev = dev_get_by_index_rcu(net, ifindex);
920 strcpy(name, dev->name);
922 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
931 * dev_getbyhwaddr_rcu - find a device by its hardware address
932 * @net: the applicable net namespace
933 * @type: media type of device
934 * @ha: hardware address
936 * Search for an interface by MAC address. Returns NULL if the device
937 * is not found or a pointer to the device.
938 * The caller must hold RCU or RTNL.
939 * The returned device has not had its ref count increased
940 * and the caller must therefore be careful about locking
944 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
947 struct net_device *dev;
949 for_each_netdev_rcu(net, dev)
950 if (dev->type == type &&
951 !memcmp(dev->dev_addr, ha, dev->addr_len))
956 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
958 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
960 struct net_device *dev;
963 for_each_netdev(net, dev)
964 if (dev->type == type)
969 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
971 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
973 struct net_device *dev, *ret = NULL;
976 for_each_netdev_rcu(net, dev)
977 if (dev->type == type) {
985 EXPORT_SYMBOL(dev_getfirstbyhwtype);
988 * __dev_get_by_flags - find any device with given flags
989 * @net: the applicable net namespace
990 * @if_flags: IFF_* values
991 * @mask: bitmask of bits in if_flags to check
993 * Search for any interface with the given flags. Returns NULL if a device
994 * is not found or a pointer to the device. Must be called inside
995 * rtnl_lock(), and result refcount is unchanged.
998 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1001 struct net_device *dev, *ret;
1006 for_each_netdev(net, dev) {
1007 if (((dev->flags ^ if_flags) & mask) == 0) {
1014 EXPORT_SYMBOL(__dev_get_by_flags);
1017 * dev_valid_name - check if name is okay for network device
1018 * @name: name string
1020 * Network device names need to be valid file names to
1021 * to allow sysfs to work. We also disallow any kind of
1024 bool dev_valid_name(const char *name)
1028 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1030 if (!strcmp(name, ".") || !strcmp(name, ".."))
1034 if (*name == '/' || *name == ':' || isspace(*name))
1040 EXPORT_SYMBOL(dev_valid_name);
1043 * __dev_alloc_name - allocate a name for a device
1044 * @net: network namespace to allocate the device name in
1045 * @name: name format string
1046 * @buf: scratch buffer and result name string
1048 * Passed a format string - eg "lt%d" it will try and find a suitable
1049 * id. It scans list of devices to build up a free map, then chooses
1050 * the first empty slot. The caller must hold the dev_base or rtnl lock
1051 * while allocating the name and adding the device in order to avoid
1053 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1054 * Returns the number of the unit assigned or a negative errno code.
1057 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1061 const int max_netdevices = 8*PAGE_SIZE;
1062 unsigned long *inuse;
1063 struct net_device *d;
1065 if (!dev_valid_name(name))
1068 p = strchr(name, '%');
1071 * Verify the string as this thing may have come from
1072 * the user. There must be either one "%d" and no other "%"
1075 if (p[1] != 'd' || strchr(p + 2, '%'))
1078 /* Use one page as a bit array of possible slots */
1079 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1083 for_each_netdev(net, d) {
1084 if (!sscanf(d->name, name, &i))
1086 if (i < 0 || i >= max_netdevices)
1089 /* avoid cases where sscanf is not exact inverse of printf */
1090 snprintf(buf, IFNAMSIZ, name, i);
1091 if (!strncmp(buf, d->name, IFNAMSIZ))
1095 i = find_first_zero_bit(inuse, max_netdevices);
1096 free_page((unsigned long) inuse);
1099 snprintf(buf, IFNAMSIZ, name, i);
1100 if (!__dev_get_by_name(net, buf))
1103 /* It is possible to run out of possible slots
1104 * when the name is long and there isn't enough space left
1105 * for the digits, or if all bits are used.
1110 static int dev_alloc_name_ns(struct net *net,
1111 struct net_device *dev,
1118 ret = __dev_alloc_name(net, name, buf);
1120 strlcpy(dev->name, buf, IFNAMSIZ);
1125 * dev_alloc_name - allocate a name for a device
1127 * @name: name format string
1129 * Passed a format string - eg "lt%d" it will try and find a suitable
1130 * id. It scans list of devices to build up a free map, then chooses
1131 * the first empty slot. The caller must hold the dev_base or rtnl lock
1132 * while allocating the name and adding the device in order to avoid
1134 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1135 * Returns the number of the unit assigned or a negative errno code.
1138 int dev_alloc_name(struct net_device *dev, const char *name)
1140 return dev_alloc_name_ns(dev_net(dev), dev, name);
1142 EXPORT_SYMBOL(dev_alloc_name);
1144 int dev_get_valid_name(struct net *net, struct net_device *dev,
1149 if (!dev_valid_name(name))
1152 if (strchr(name, '%'))
1153 return dev_alloc_name_ns(net, dev, name);
1154 else if (__dev_get_by_name(net, name))
1156 else if (dev->name != name)
1157 strlcpy(dev->name, name, IFNAMSIZ);
1161 EXPORT_SYMBOL(dev_get_valid_name);
1164 * dev_change_name - change name of a device
1166 * @newname: name (or format string) must be at least IFNAMSIZ
1168 * Change name of a device, can pass format strings "eth%d".
1171 int dev_change_name(struct net_device *dev, const char *newname)
1173 unsigned char old_assign_type;
1174 char oldname[IFNAMSIZ];
1180 BUG_ON(!dev_net(dev));
1183 if (dev->flags & IFF_UP)
1186 write_seqcount_begin(&devnet_rename_seq);
1188 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1189 write_seqcount_end(&devnet_rename_seq);
1193 memcpy(oldname, dev->name, IFNAMSIZ);
1195 err = dev_get_valid_name(net, dev, newname);
1197 write_seqcount_end(&devnet_rename_seq);
1201 if (oldname[0] && !strchr(oldname, '%'))
1202 netdev_info(dev, "renamed from %s\n", oldname);
1204 old_assign_type = dev->name_assign_type;
1205 dev->name_assign_type = NET_NAME_RENAMED;
1208 ret = device_rename(&dev->dev, dev->name);
1210 memcpy(dev->name, oldname, IFNAMSIZ);
1211 dev->name_assign_type = old_assign_type;
1212 write_seqcount_end(&devnet_rename_seq);
1216 write_seqcount_end(&devnet_rename_seq);
1218 netdev_adjacent_rename_links(dev, oldname);
1220 write_lock_bh(&dev_base_lock);
1221 hlist_del_rcu(&dev->name_hlist);
1222 write_unlock_bh(&dev_base_lock);
1226 write_lock_bh(&dev_base_lock);
1227 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1228 write_unlock_bh(&dev_base_lock);
1230 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1231 ret = notifier_to_errno(ret);
1234 /* err >= 0 after dev_alloc_name() or stores the first errno */
1237 write_seqcount_begin(&devnet_rename_seq);
1238 memcpy(dev->name, oldname, IFNAMSIZ);
1239 memcpy(oldname, newname, IFNAMSIZ);
1240 dev->name_assign_type = old_assign_type;
1241 old_assign_type = NET_NAME_RENAMED;
1244 pr_err("%s: name change rollback failed: %d\n",
1253 * dev_set_alias - change ifalias of a device
1255 * @alias: name up to IFALIASZ
1256 * @len: limit of bytes to copy from info
1258 * Set ifalias for a device,
1260 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1262 struct dev_ifalias *new_alias = NULL;
1264 if (len >= IFALIASZ)
1268 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1272 memcpy(new_alias->ifalias, alias, len);
1273 new_alias->ifalias[len] = 0;
1276 mutex_lock(&ifalias_mutex);
1277 rcu_swap_protected(dev->ifalias, new_alias,
1278 mutex_is_locked(&ifalias_mutex));
1279 mutex_unlock(&ifalias_mutex);
1282 kfree_rcu(new_alias, rcuhead);
1286 EXPORT_SYMBOL(dev_set_alias);
1289 * dev_get_alias - get ifalias of a device
1291 * @name: buffer to store name of ifalias
1292 * @len: size of buffer
1294 * get ifalias for a device. Caller must make sure dev cannot go
1295 * away, e.g. rcu read lock or own a reference count to device.
1297 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1299 const struct dev_ifalias *alias;
1303 alias = rcu_dereference(dev->ifalias);
1305 ret = snprintf(name, len, "%s", alias->ifalias);
1312 * netdev_features_change - device changes features
1313 * @dev: device to cause notification
1315 * Called to indicate a device has changed features.
1317 void netdev_features_change(struct net_device *dev)
1319 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1321 EXPORT_SYMBOL(netdev_features_change);
1324 * netdev_state_change - device changes state
1325 * @dev: device to cause notification
1327 * Called to indicate a device has changed state. This function calls
1328 * the notifier chains for netdev_chain and sends a NEWLINK message
1329 * to the routing socket.
1331 void netdev_state_change(struct net_device *dev)
1333 if (dev->flags & IFF_UP) {
1334 struct netdev_notifier_change_info change_info = {
1338 call_netdevice_notifiers_info(NETDEV_CHANGE,
1340 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1343 EXPORT_SYMBOL(netdev_state_change);
1346 * netdev_notify_peers - notify network peers about existence of @dev
1347 * @dev: network device
1349 * Generate traffic such that interested network peers are aware of
1350 * @dev, such as by generating a gratuitous ARP. This may be used when
1351 * a device wants to inform the rest of the network about some sort of
1352 * reconfiguration such as a failover event or virtual machine
1355 void netdev_notify_peers(struct net_device *dev)
1358 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1359 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1362 EXPORT_SYMBOL(netdev_notify_peers);
1364 static int __dev_open(struct net_device *dev)
1366 const struct net_device_ops *ops = dev->netdev_ops;
1371 if (!netif_device_present(dev))
1374 /* Block netpoll from trying to do any rx path servicing.
1375 * If we don't do this there is a chance ndo_poll_controller
1376 * or ndo_poll may be running while we open the device
1378 netpoll_poll_disable(dev);
1380 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1381 ret = notifier_to_errno(ret);
1385 set_bit(__LINK_STATE_START, &dev->state);
1387 if (ops->ndo_validate_addr)
1388 ret = ops->ndo_validate_addr(dev);
1390 if (!ret && ops->ndo_open)
1391 ret = ops->ndo_open(dev);
1393 netpoll_poll_enable(dev);
1396 clear_bit(__LINK_STATE_START, &dev->state);
1398 dev->flags |= IFF_UP;
1399 dev_set_rx_mode(dev);
1401 add_device_randomness(dev->dev_addr, dev->addr_len);
1408 * dev_open - prepare an interface for use.
1409 * @dev: device to open
1410 * @extack: netlink extended ack
1412 * Takes a device from down to up state. The device's private open
1413 * function is invoked and then the multicast lists are loaded. Finally
1414 * the device is moved into the up state and a %NETDEV_UP message is
1415 * sent to the netdev notifier chain.
1417 * Calling this function on an active interface is a nop. On a failure
1418 * a negative errno code is returned.
1420 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1424 if (dev->flags & IFF_UP)
1427 ret = __dev_open(dev);
1431 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1432 call_netdevice_notifiers(NETDEV_UP, dev);
1436 EXPORT_SYMBOL(dev_open);
1438 static void __dev_close_many(struct list_head *head)
1440 struct net_device *dev;
1445 list_for_each_entry(dev, head, close_list) {
1446 /* Temporarily disable netpoll until the interface is down */
1447 netpoll_poll_disable(dev);
1449 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1451 clear_bit(__LINK_STATE_START, &dev->state);
1453 /* Synchronize to scheduled poll. We cannot touch poll list, it
1454 * can be even on different cpu. So just clear netif_running().
1456 * dev->stop() will invoke napi_disable() on all of it's
1457 * napi_struct instances on this device.
1459 smp_mb__after_atomic(); /* Commit netif_running(). */
1462 dev_deactivate_many(head);
1464 list_for_each_entry(dev, head, close_list) {
1465 const struct net_device_ops *ops = dev->netdev_ops;
1468 * Call the device specific close. This cannot fail.
1469 * Only if device is UP
1471 * We allow it to be called even after a DETACH hot-plug
1477 dev->flags &= ~IFF_UP;
1478 netpoll_poll_enable(dev);
1482 static void __dev_close(struct net_device *dev)
1486 list_add(&dev->close_list, &single);
1487 __dev_close_many(&single);
1491 void dev_close_many(struct list_head *head, bool unlink)
1493 struct net_device *dev, *tmp;
1495 /* Remove the devices that don't need to be closed */
1496 list_for_each_entry_safe(dev, tmp, head, close_list)
1497 if (!(dev->flags & IFF_UP))
1498 list_del_init(&dev->close_list);
1500 __dev_close_many(head);
1502 list_for_each_entry_safe(dev, tmp, head, close_list) {
1503 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1504 call_netdevice_notifiers(NETDEV_DOWN, dev);
1506 list_del_init(&dev->close_list);
1509 EXPORT_SYMBOL(dev_close_many);
1512 * dev_close - shutdown an interface.
1513 * @dev: device to shutdown
1515 * This function moves an active device into down state. A
1516 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1517 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1520 void dev_close(struct net_device *dev)
1522 if (dev->flags & IFF_UP) {
1525 list_add(&dev->close_list, &single);
1526 dev_close_many(&single, true);
1530 EXPORT_SYMBOL(dev_close);
1534 * dev_disable_lro - disable Large Receive Offload on a device
1537 * Disable Large Receive Offload (LRO) on a net device. Must be
1538 * called under RTNL. This is needed if received packets may be
1539 * forwarded to another interface.
1541 void dev_disable_lro(struct net_device *dev)
1543 struct net_device *lower_dev;
1544 struct list_head *iter;
1546 dev->wanted_features &= ~NETIF_F_LRO;
1547 netdev_update_features(dev);
1549 if (unlikely(dev->features & NETIF_F_LRO))
1550 netdev_WARN(dev, "failed to disable LRO!\n");
1552 netdev_for_each_lower_dev(dev, lower_dev, iter)
1553 dev_disable_lro(lower_dev);
1555 EXPORT_SYMBOL(dev_disable_lro);
1558 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1561 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1562 * called under RTNL. This is needed if Generic XDP is installed on
1565 static void dev_disable_gro_hw(struct net_device *dev)
1567 dev->wanted_features &= ~NETIF_F_GRO_HW;
1568 netdev_update_features(dev);
1570 if (unlikely(dev->features & NETIF_F_GRO_HW))
1571 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1574 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1577 case NETDEV_##val: \
1578 return "NETDEV_" __stringify(val);
1580 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1581 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1582 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1583 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1584 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1585 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1586 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1587 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1588 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1591 return "UNKNOWN_NETDEV_EVENT";
1593 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1595 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1596 struct net_device *dev)
1598 struct netdev_notifier_info info = {
1602 return nb->notifier_call(nb, val, &info);
1605 static int dev_boot_phase = 1;
1608 * register_netdevice_notifier - register a network notifier block
1611 * Register a notifier to be called when network device events occur.
1612 * The notifier passed is linked into the kernel structures and must
1613 * not be reused until it has been unregistered. A negative errno code
1614 * is returned on a failure.
1616 * When registered all registration and up events are replayed
1617 * to the new notifier to allow device to have a race free
1618 * view of the network device list.
1621 int register_netdevice_notifier(struct notifier_block *nb)
1623 struct net_device *dev;
1624 struct net_device *last;
1628 /* Close race with setup_net() and cleanup_net() */
1629 down_write(&pernet_ops_rwsem);
1631 err = raw_notifier_chain_register(&netdev_chain, nb);
1637 for_each_netdev(net, dev) {
1638 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1639 err = notifier_to_errno(err);
1643 if (!(dev->flags & IFF_UP))
1646 call_netdevice_notifier(nb, NETDEV_UP, dev);
1652 up_write(&pernet_ops_rwsem);
1658 for_each_netdev(net, dev) {
1662 if (dev->flags & IFF_UP) {
1663 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1665 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1667 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1672 raw_notifier_chain_unregister(&netdev_chain, nb);
1675 EXPORT_SYMBOL(register_netdevice_notifier);
1678 * unregister_netdevice_notifier - unregister a network notifier block
1681 * Unregister a notifier previously registered by
1682 * register_netdevice_notifier(). The notifier is unlinked into the
1683 * kernel structures and may then be reused. A negative errno code
1684 * is returned on a failure.
1686 * After unregistering unregister and down device events are synthesized
1687 * for all devices on the device list to the removed notifier to remove
1688 * the need for special case cleanup code.
1691 int unregister_netdevice_notifier(struct notifier_block *nb)
1693 struct net_device *dev;
1697 /* Close race with setup_net() and cleanup_net() */
1698 down_write(&pernet_ops_rwsem);
1700 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1705 for_each_netdev(net, dev) {
1706 if (dev->flags & IFF_UP) {
1707 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1709 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1711 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1716 up_write(&pernet_ops_rwsem);
1719 EXPORT_SYMBOL(unregister_netdevice_notifier);
1722 * call_netdevice_notifiers_info - call all network notifier blocks
1723 * @val: value passed unmodified to notifier function
1724 * @info: notifier information data
1726 * Call all network notifier blocks. Parameters and return value
1727 * are as for raw_notifier_call_chain().
1730 static int call_netdevice_notifiers_info(unsigned long val,
1731 struct netdev_notifier_info *info)
1734 return raw_notifier_call_chain(&netdev_chain, val, info);
1738 * call_netdevice_notifiers - call all network notifier blocks
1739 * @val: value passed unmodified to notifier function
1740 * @dev: net_device pointer passed unmodified to notifier function
1742 * Call all network notifier blocks. Parameters and return value
1743 * are as for raw_notifier_call_chain().
1746 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1748 struct netdev_notifier_info info = {
1752 return call_netdevice_notifiers_info(val, &info);
1754 EXPORT_SYMBOL(call_netdevice_notifiers);
1757 * call_netdevice_notifiers_mtu - call all network notifier blocks
1758 * @val: value passed unmodified to notifier function
1759 * @dev: net_device pointer passed unmodified to notifier function
1760 * @arg: additional u32 argument passed to the notifier function
1762 * Call all network notifier blocks. Parameters and return value
1763 * are as for raw_notifier_call_chain().
1765 static int call_netdevice_notifiers_mtu(unsigned long val,
1766 struct net_device *dev, u32 arg)
1768 struct netdev_notifier_info_ext info = {
1773 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1775 return call_netdevice_notifiers_info(val, &info.info);
1778 #ifdef CONFIG_NET_INGRESS
1779 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1781 void net_inc_ingress_queue(void)
1783 static_branch_inc(&ingress_needed_key);
1785 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1787 void net_dec_ingress_queue(void)
1789 static_branch_dec(&ingress_needed_key);
1791 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1794 #ifdef CONFIG_NET_EGRESS
1795 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1797 void net_inc_egress_queue(void)
1799 static_branch_inc(&egress_needed_key);
1801 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1803 void net_dec_egress_queue(void)
1805 static_branch_dec(&egress_needed_key);
1807 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1810 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1811 #ifdef HAVE_JUMP_LABEL
1812 static atomic_t netstamp_needed_deferred;
1813 static atomic_t netstamp_wanted;
1814 static void netstamp_clear(struct work_struct *work)
1816 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1819 wanted = atomic_add_return(deferred, &netstamp_wanted);
1821 static_branch_enable(&netstamp_needed_key);
1823 static_branch_disable(&netstamp_needed_key);
1825 static DECLARE_WORK(netstamp_work, netstamp_clear);
1828 void net_enable_timestamp(void)
1830 #ifdef HAVE_JUMP_LABEL
1834 wanted = atomic_read(&netstamp_wanted);
1837 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1840 atomic_inc(&netstamp_needed_deferred);
1841 schedule_work(&netstamp_work);
1843 static_branch_inc(&netstamp_needed_key);
1846 EXPORT_SYMBOL(net_enable_timestamp);
1848 void net_disable_timestamp(void)
1850 #ifdef HAVE_JUMP_LABEL
1854 wanted = atomic_read(&netstamp_wanted);
1857 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1860 atomic_dec(&netstamp_needed_deferred);
1861 schedule_work(&netstamp_work);
1863 static_branch_dec(&netstamp_needed_key);
1866 EXPORT_SYMBOL(net_disable_timestamp);
1868 static inline void net_timestamp_set(struct sk_buff *skb)
1871 if (static_branch_unlikely(&netstamp_needed_key))
1872 __net_timestamp(skb);
1875 #define net_timestamp_check(COND, SKB) \
1876 if (static_branch_unlikely(&netstamp_needed_key)) { \
1877 if ((COND) && !(SKB)->tstamp) \
1878 __net_timestamp(SKB); \
1881 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1885 if (!(dev->flags & IFF_UP))
1888 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1889 if (skb->len <= len)
1892 /* if TSO is enabled, we don't care about the length as the packet
1893 * could be forwarded without being segmented before
1895 if (skb_is_gso(skb))
1900 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1902 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1904 int ret = ____dev_forward_skb(dev, skb);
1907 skb->protocol = eth_type_trans(skb, dev);
1908 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1913 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1916 * dev_forward_skb - loopback an skb to another netif
1918 * @dev: destination network device
1919 * @skb: buffer to forward
1922 * NET_RX_SUCCESS (no congestion)
1923 * NET_RX_DROP (packet was dropped, but freed)
1925 * dev_forward_skb can be used for injecting an skb from the
1926 * start_xmit function of one device into the receive queue
1927 * of another device.
1929 * The receiving device may be in another namespace, so
1930 * we have to clear all information in the skb that could
1931 * impact namespace isolation.
1933 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1935 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1937 EXPORT_SYMBOL_GPL(dev_forward_skb);
1939 static inline int deliver_skb(struct sk_buff *skb,
1940 struct packet_type *pt_prev,
1941 struct net_device *orig_dev)
1943 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1945 refcount_inc(&skb->users);
1946 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1949 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1950 struct packet_type **pt,
1951 struct net_device *orig_dev,
1953 struct list_head *ptype_list)
1955 struct packet_type *ptype, *pt_prev = *pt;
1957 list_for_each_entry_rcu(ptype, ptype_list, list) {
1958 if (ptype->type != type)
1961 deliver_skb(skb, pt_prev, orig_dev);
1967 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1969 if (!ptype->af_packet_priv || !skb->sk)
1972 if (ptype->id_match)
1973 return ptype->id_match(ptype, skb->sk);
1974 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1981 * dev_nit_active - return true if any network interface taps are in use
1983 * @dev: network device to check for the presence of taps
1985 bool dev_nit_active(struct net_device *dev)
1987 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
1989 EXPORT_SYMBOL_GPL(dev_nit_active);
1992 * Support routine. Sends outgoing frames to any network
1993 * taps currently in use.
1996 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1998 struct packet_type *ptype;
1999 struct sk_buff *skb2 = NULL;
2000 struct packet_type *pt_prev = NULL;
2001 struct list_head *ptype_list = &ptype_all;
2005 list_for_each_entry_rcu(ptype, ptype_list, list) {
2006 if (ptype->ignore_outgoing)
2009 /* Never send packets back to the socket
2010 * they originated from - MvS (miquels@drinkel.ow.org)
2012 if (skb_loop_sk(ptype, skb))
2016 deliver_skb(skb2, pt_prev, skb->dev);
2021 /* need to clone skb, done only once */
2022 skb2 = skb_clone(skb, GFP_ATOMIC);
2026 net_timestamp_set(skb2);
2028 /* skb->nh should be correctly
2029 * set by sender, so that the second statement is
2030 * just protection against buggy protocols.
2032 skb_reset_mac_header(skb2);
2034 if (skb_network_header(skb2) < skb2->data ||
2035 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2036 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2037 ntohs(skb2->protocol),
2039 skb_reset_network_header(skb2);
2042 skb2->transport_header = skb2->network_header;
2043 skb2->pkt_type = PACKET_OUTGOING;
2047 if (ptype_list == &ptype_all) {
2048 ptype_list = &dev->ptype_all;
2053 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2054 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2060 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2063 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2064 * @dev: Network device
2065 * @txq: number of queues available
2067 * If real_num_tx_queues is changed the tc mappings may no longer be
2068 * valid. To resolve this verify the tc mapping remains valid and if
2069 * not NULL the mapping. With no priorities mapping to this
2070 * offset/count pair it will no longer be used. In the worst case TC0
2071 * is invalid nothing can be done so disable priority mappings. If is
2072 * expected that drivers will fix this mapping if they can before
2073 * calling netif_set_real_num_tx_queues.
2075 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2078 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2080 /* If TC0 is invalidated disable TC mapping */
2081 if (tc->offset + tc->count > txq) {
2082 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2087 /* Invalidated prio to tc mappings set to TC0 */
2088 for (i = 1; i < TC_BITMASK + 1; i++) {
2089 int q = netdev_get_prio_tc_map(dev, i);
2091 tc = &dev->tc_to_txq[q];
2092 if (tc->offset + tc->count > txq) {
2093 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2095 netdev_set_prio_tc_map(dev, i, 0);
2100 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2103 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2106 /* walk through the TCs and see if it falls into any of them */
2107 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2108 if ((txq - tc->offset) < tc->count)
2112 /* didn't find it, just return -1 to indicate no match */
2118 EXPORT_SYMBOL(netdev_txq_to_tc);
2121 struct static_key xps_needed __read_mostly;
2122 EXPORT_SYMBOL(xps_needed);
2123 struct static_key xps_rxqs_needed __read_mostly;
2124 EXPORT_SYMBOL(xps_rxqs_needed);
2125 static DEFINE_MUTEX(xps_map_mutex);
2126 #define xmap_dereference(P) \
2127 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2129 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2132 struct xps_map *map = NULL;
2136 map = xmap_dereference(dev_maps->attr_map[tci]);
2140 for (pos = map->len; pos--;) {
2141 if (map->queues[pos] != index)
2145 map->queues[pos] = map->queues[--map->len];
2149 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2150 kfree_rcu(map, rcu);
2157 static bool remove_xps_queue_cpu(struct net_device *dev,
2158 struct xps_dev_maps *dev_maps,
2159 int cpu, u16 offset, u16 count)
2161 int num_tc = dev->num_tc ? : 1;
2162 bool active = false;
2165 for (tci = cpu * num_tc; num_tc--; tci++) {
2168 for (i = count, j = offset; i--; j++) {
2169 if (!remove_xps_queue(dev_maps, tci, j))
2179 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2180 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2181 u16 offset, u16 count, bool is_rxqs_map)
2183 bool active = false;
2186 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2188 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2192 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2194 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2196 for (i = offset + (count - 1); count--; i--)
2197 netdev_queue_numa_node_write(
2198 netdev_get_tx_queue(dev, i),
2201 kfree_rcu(dev_maps, rcu);
2205 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2208 const unsigned long *possible_mask = NULL;
2209 struct xps_dev_maps *dev_maps;
2210 unsigned int nr_ids;
2212 if (!static_key_false(&xps_needed))
2216 mutex_lock(&xps_map_mutex);
2218 if (static_key_false(&xps_rxqs_needed)) {
2219 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2221 nr_ids = dev->num_rx_queues;
2222 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2223 offset, count, true);
2227 dev_maps = xmap_dereference(dev->xps_cpus_map);
2231 if (num_possible_cpus() > 1)
2232 possible_mask = cpumask_bits(cpu_possible_mask);
2233 nr_ids = nr_cpu_ids;
2234 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2238 if (static_key_enabled(&xps_rxqs_needed))
2239 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2241 static_key_slow_dec_cpuslocked(&xps_needed);
2242 mutex_unlock(&xps_map_mutex);
2246 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2248 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2251 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2252 u16 index, bool is_rxqs_map)
2254 struct xps_map *new_map;
2255 int alloc_len = XPS_MIN_MAP_ALLOC;
2258 for (pos = 0; map && pos < map->len; pos++) {
2259 if (map->queues[pos] != index)
2264 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2266 if (pos < map->alloc_len)
2269 alloc_len = map->alloc_len * 2;
2272 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2276 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2278 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2279 cpu_to_node(attr_index));
2283 for (i = 0; i < pos; i++)
2284 new_map->queues[i] = map->queues[i];
2285 new_map->alloc_len = alloc_len;
2291 /* Must be called under cpus_read_lock */
2292 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2293 u16 index, bool is_rxqs_map)
2295 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2296 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2297 int i, j, tci, numa_node_id = -2;
2298 int maps_sz, num_tc = 1, tc = 0;
2299 struct xps_map *map, *new_map;
2300 bool active = false;
2301 unsigned int nr_ids;
2304 /* Do not allow XPS on subordinate device directly */
2305 num_tc = dev->num_tc;
2309 /* If queue belongs to subordinate dev use its map */
2310 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2312 tc = netdev_txq_to_tc(dev, index);
2317 mutex_lock(&xps_map_mutex);
2319 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2320 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2321 nr_ids = dev->num_rx_queues;
2323 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2324 if (num_possible_cpus() > 1) {
2325 online_mask = cpumask_bits(cpu_online_mask);
2326 possible_mask = cpumask_bits(cpu_possible_mask);
2328 dev_maps = xmap_dereference(dev->xps_cpus_map);
2329 nr_ids = nr_cpu_ids;
2332 if (maps_sz < L1_CACHE_BYTES)
2333 maps_sz = L1_CACHE_BYTES;
2335 /* allocate memory for queue storage */
2336 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2339 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2340 if (!new_dev_maps) {
2341 mutex_unlock(&xps_map_mutex);
2345 tci = j * num_tc + tc;
2346 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2349 map = expand_xps_map(map, j, index, is_rxqs_map);
2353 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2357 goto out_no_new_maps;
2359 static_key_slow_inc_cpuslocked(&xps_needed);
2361 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2363 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2365 /* copy maps belonging to foreign traffic classes */
2366 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2367 /* fill in the new device map from the old device map */
2368 map = xmap_dereference(dev_maps->attr_map[tci]);
2369 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2372 /* We need to explicitly update tci as prevous loop
2373 * could break out early if dev_maps is NULL.
2375 tci = j * num_tc + tc;
2377 if (netif_attr_test_mask(j, mask, nr_ids) &&
2378 netif_attr_test_online(j, online_mask, nr_ids)) {
2379 /* add tx-queue to CPU/rx-queue maps */
2382 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2383 while ((pos < map->len) && (map->queues[pos] != index))
2386 if (pos == map->len)
2387 map->queues[map->len++] = index;
2390 if (numa_node_id == -2)
2391 numa_node_id = cpu_to_node(j);
2392 else if (numa_node_id != cpu_to_node(j))
2396 } else if (dev_maps) {
2397 /* fill in the new device map from the old device map */
2398 map = xmap_dereference(dev_maps->attr_map[tci]);
2399 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2402 /* copy maps belonging to foreign traffic classes */
2403 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2404 /* fill in the new device map from the old device map */
2405 map = xmap_dereference(dev_maps->attr_map[tci]);
2406 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2411 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2413 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2415 /* Cleanup old maps */
2417 goto out_no_old_maps;
2419 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2421 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2422 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2423 map = xmap_dereference(dev_maps->attr_map[tci]);
2424 if (map && map != new_map)
2425 kfree_rcu(map, rcu);
2429 kfree_rcu(dev_maps, rcu);
2432 dev_maps = new_dev_maps;
2437 /* update Tx queue numa node */
2438 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2439 (numa_node_id >= 0) ?
2440 numa_node_id : NUMA_NO_NODE);
2446 /* removes tx-queue from unused CPUs/rx-queues */
2447 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2449 for (i = tc, tci = j * num_tc; i--; tci++)
2450 active |= remove_xps_queue(dev_maps, tci, index);
2451 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2452 !netif_attr_test_online(j, online_mask, nr_ids))
2453 active |= remove_xps_queue(dev_maps, tci, index);
2454 for (i = num_tc - tc, tci++; --i; tci++)
2455 active |= remove_xps_queue(dev_maps, tci, index);
2458 /* free map if not active */
2461 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2463 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2464 kfree_rcu(dev_maps, rcu);
2468 mutex_unlock(&xps_map_mutex);
2472 /* remove any maps that we added */
2473 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2475 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2476 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2478 xmap_dereference(dev_maps->attr_map[tci]) :
2480 if (new_map && new_map != map)
2485 mutex_unlock(&xps_map_mutex);
2487 kfree(new_dev_maps);
2490 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2492 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2498 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2503 EXPORT_SYMBOL(netif_set_xps_queue);
2506 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2508 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2510 /* Unbind any subordinate channels */
2511 while (txq-- != &dev->_tx[0]) {
2513 netdev_unbind_sb_channel(dev, txq->sb_dev);
2517 void netdev_reset_tc(struct net_device *dev)
2520 netif_reset_xps_queues_gt(dev, 0);
2522 netdev_unbind_all_sb_channels(dev);
2524 /* Reset TC configuration of device */
2526 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2527 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2529 EXPORT_SYMBOL(netdev_reset_tc);
2531 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2533 if (tc >= dev->num_tc)
2537 netif_reset_xps_queues(dev, offset, count);
2539 dev->tc_to_txq[tc].count = count;
2540 dev->tc_to_txq[tc].offset = offset;
2543 EXPORT_SYMBOL(netdev_set_tc_queue);
2545 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2547 if (num_tc > TC_MAX_QUEUE)
2551 netif_reset_xps_queues_gt(dev, 0);
2553 netdev_unbind_all_sb_channels(dev);
2555 dev->num_tc = num_tc;
2558 EXPORT_SYMBOL(netdev_set_num_tc);
2560 void netdev_unbind_sb_channel(struct net_device *dev,
2561 struct net_device *sb_dev)
2563 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2566 netif_reset_xps_queues_gt(sb_dev, 0);
2568 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2569 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2571 while (txq-- != &dev->_tx[0]) {
2572 if (txq->sb_dev == sb_dev)
2576 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2578 int netdev_bind_sb_channel_queue(struct net_device *dev,
2579 struct net_device *sb_dev,
2580 u8 tc, u16 count, u16 offset)
2582 /* Make certain the sb_dev and dev are already configured */
2583 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2586 /* We cannot hand out queues we don't have */
2587 if ((offset + count) > dev->real_num_tx_queues)
2590 /* Record the mapping */
2591 sb_dev->tc_to_txq[tc].count = count;
2592 sb_dev->tc_to_txq[tc].offset = offset;
2594 /* Provide a way for Tx queue to find the tc_to_txq map or
2595 * XPS map for itself.
2598 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2602 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2604 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2606 /* Do not use a multiqueue device to represent a subordinate channel */
2607 if (netif_is_multiqueue(dev))
2610 /* We allow channels 1 - 32767 to be used for subordinate channels.
2611 * Channel 0 is meant to be "native" mode and used only to represent
2612 * the main root device. We allow writing 0 to reset the device back
2613 * to normal mode after being used as a subordinate channel.
2615 if (channel > S16_MAX)
2618 dev->num_tc = -channel;
2622 EXPORT_SYMBOL(netdev_set_sb_channel);
2625 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2626 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2628 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2633 disabling = txq < dev->real_num_tx_queues;
2635 if (txq < 1 || txq > dev->num_tx_queues)
2638 if (dev->reg_state == NETREG_REGISTERED ||
2639 dev->reg_state == NETREG_UNREGISTERING) {
2642 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2648 netif_setup_tc(dev, txq);
2650 dev->real_num_tx_queues = txq;
2654 qdisc_reset_all_tx_gt(dev, txq);
2656 netif_reset_xps_queues_gt(dev, txq);
2660 dev->real_num_tx_queues = txq;
2665 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2669 * netif_set_real_num_rx_queues - set actual number of RX queues used
2670 * @dev: Network device
2671 * @rxq: Actual number of RX queues
2673 * This must be called either with the rtnl_lock held or before
2674 * registration of the net device. Returns 0 on success, or a
2675 * negative error code. If called before registration, it always
2678 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2682 if (rxq < 1 || rxq > dev->num_rx_queues)
2685 if (dev->reg_state == NETREG_REGISTERED) {
2688 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2694 dev->real_num_rx_queues = rxq;
2697 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2701 * netif_get_num_default_rss_queues - default number of RSS queues
2703 * This routine should set an upper limit on the number of RSS queues
2704 * used by default by multiqueue devices.
2706 int netif_get_num_default_rss_queues(void)
2708 return is_kdump_kernel() ?
2709 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2711 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2713 static void __netif_reschedule(struct Qdisc *q)
2715 struct softnet_data *sd;
2716 unsigned long flags;
2718 local_irq_save(flags);
2719 sd = this_cpu_ptr(&softnet_data);
2720 q->next_sched = NULL;
2721 *sd->output_queue_tailp = q;
2722 sd->output_queue_tailp = &q->next_sched;
2723 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2724 local_irq_restore(flags);
2727 void __netif_schedule(struct Qdisc *q)
2729 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2730 __netif_reschedule(q);
2732 EXPORT_SYMBOL(__netif_schedule);
2734 struct dev_kfree_skb_cb {
2735 enum skb_free_reason reason;
2738 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2740 return (struct dev_kfree_skb_cb *)skb->cb;
2743 void netif_schedule_queue(struct netdev_queue *txq)
2746 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2747 struct Qdisc *q = rcu_dereference(txq->qdisc);
2749 __netif_schedule(q);
2753 EXPORT_SYMBOL(netif_schedule_queue);
2755 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2757 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2761 q = rcu_dereference(dev_queue->qdisc);
2762 __netif_schedule(q);
2766 EXPORT_SYMBOL(netif_tx_wake_queue);
2768 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2770 unsigned long flags;
2775 if (likely(refcount_read(&skb->users) == 1)) {
2777 refcount_set(&skb->users, 0);
2778 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2781 get_kfree_skb_cb(skb)->reason = reason;
2782 local_irq_save(flags);
2783 skb->next = __this_cpu_read(softnet_data.completion_queue);
2784 __this_cpu_write(softnet_data.completion_queue, skb);
2785 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2786 local_irq_restore(flags);
2788 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2790 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2792 if (in_irq() || irqs_disabled())
2793 __dev_kfree_skb_irq(skb, reason);
2797 EXPORT_SYMBOL(__dev_kfree_skb_any);
2801 * netif_device_detach - mark device as removed
2802 * @dev: network device
2804 * Mark device as removed from system and therefore no longer available.
2806 void netif_device_detach(struct net_device *dev)
2808 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2809 netif_running(dev)) {
2810 netif_tx_stop_all_queues(dev);
2813 EXPORT_SYMBOL(netif_device_detach);
2816 * netif_device_attach - mark device as attached
2817 * @dev: network device
2819 * Mark device as attached from system and restart if needed.
2821 void netif_device_attach(struct net_device *dev)
2823 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2824 netif_running(dev)) {
2825 netif_tx_wake_all_queues(dev);
2826 __netdev_watchdog_up(dev);
2829 EXPORT_SYMBOL(netif_device_attach);
2832 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2833 * to be used as a distribution range.
2835 static u16 skb_tx_hash(const struct net_device *dev,
2836 const struct net_device *sb_dev,
2837 struct sk_buff *skb)
2841 u16 qcount = dev->real_num_tx_queues;
2844 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2846 qoffset = sb_dev->tc_to_txq[tc].offset;
2847 qcount = sb_dev->tc_to_txq[tc].count;
2850 if (skb_rx_queue_recorded(skb)) {
2851 hash = skb_get_rx_queue(skb);
2852 while (unlikely(hash >= qcount))
2854 return hash + qoffset;
2857 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2860 static void skb_warn_bad_offload(const struct sk_buff *skb)
2862 static const netdev_features_t null_features;
2863 struct net_device *dev = skb->dev;
2864 const char *name = "";
2866 if (!net_ratelimit())
2870 if (dev->dev.parent)
2871 name = dev_driver_string(dev->dev.parent);
2873 name = netdev_name(dev);
2875 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2876 "gso_type=%d ip_summed=%d\n",
2877 name, dev ? &dev->features : &null_features,
2878 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2879 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2880 skb_shinfo(skb)->gso_type, skb->ip_summed);
2884 * Invalidate hardware checksum when packet is to be mangled, and
2885 * complete checksum manually on outgoing path.
2887 int skb_checksum_help(struct sk_buff *skb)
2890 int ret = 0, offset;
2892 if (skb->ip_summed == CHECKSUM_COMPLETE)
2893 goto out_set_summed;
2895 if (unlikely(skb_shinfo(skb)->gso_size)) {
2896 skb_warn_bad_offload(skb);
2900 /* Before computing a checksum, we should make sure no frag could
2901 * be modified by an external entity : checksum could be wrong.
2903 if (skb_has_shared_frag(skb)) {
2904 ret = __skb_linearize(skb);
2909 offset = skb_checksum_start_offset(skb);
2910 BUG_ON(offset >= skb_headlen(skb));
2911 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2913 offset += skb->csum_offset;
2914 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2916 if (skb_cloned(skb) &&
2917 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2918 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2923 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2925 skb->ip_summed = CHECKSUM_NONE;
2929 EXPORT_SYMBOL(skb_checksum_help);
2931 int skb_crc32c_csum_help(struct sk_buff *skb)
2934 int ret = 0, offset, start;
2936 if (skb->ip_summed != CHECKSUM_PARTIAL)
2939 if (unlikely(skb_is_gso(skb)))
2942 /* Before computing a checksum, we should make sure no frag could
2943 * be modified by an external entity : checksum could be wrong.
2945 if (unlikely(skb_has_shared_frag(skb))) {
2946 ret = __skb_linearize(skb);
2950 start = skb_checksum_start_offset(skb);
2951 offset = start + offsetof(struct sctphdr, checksum);
2952 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2956 if (skb_cloned(skb) &&
2957 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2958 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2962 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2963 skb->len - start, ~(__u32)0,
2965 *(__le32 *)(skb->data + offset) = crc32c_csum;
2966 skb->ip_summed = CHECKSUM_NONE;
2967 skb->csum_not_inet = 0;
2972 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2974 __be16 type = skb->protocol;
2976 /* Tunnel gso handlers can set protocol to ethernet. */
2977 if (type == htons(ETH_P_TEB)) {
2980 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2983 eth = (struct ethhdr *)skb->data;
2984 type = eth->h_proto;
2987 return __vlan_get_protocol(skb, type, depth);
2991 * skb_mac_gso_segment - mac layer segmentation handler.
2992 * @skb: buffer to segment
2993 * @features: features for the output path (see dev->features)
2995 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2996 netdev_features_t features)
2998 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2999 struct packet_offload *ptype;
3000 int vlan_depth = skb->mac_len;
3001 __be16 type = skb_network_protocol(skb, &vlan_depth);
3003 if (unlikely(!type))
3004 return ERR_PTR(-EINVAL);
3006 __skb_pull(skb, vlan_depth);
3009 list_for_each_entry_rcu(ptype, &offload_base, list) {
3010 if (ptype->type == type && ptype->callbacks.gso_segment) {
3011 segs = ptype->callbacks.gso_segment(skb, features);
3017 __skb_push(skb, skb->data - skb_mac_header(skb));
3021 EXPORT_SYMBOL(skb_mac_gso_segment);
3024 /* openvswitch calls this on rx path, so we need a different check.
3026 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3029 return skb->ip_summed != CHECKSUM_PARTIAL &&
3030 skb->ip_summed != CHECKSUM_UNNECESSARY;
3032 return skb->ip_summed == CHECKSUM_NONE;
3036 * __skb_gso_segment - Perform segmentation on skb.
3037 * @skb: buffer to segment
3038 * @features: features for the output path (see dev->features)
3039 * @tx_path: whether it is called in TX path
3041 * This function segments the given skb and returns a list of segments.
3043 * It may return NULL if the skb requires no segmentation. This is
3044 * only possible when GSO is used for verifying header integrity.
3046 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3048 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3049 netdev_features_t features, bool tx_path)
3051 struct sk_buff *segs;
3053 if (unlikely(skb_needs_check(skb, tx_path))) {
3056 /* We're going to init ->check field in TCP or UDP header */
3057 err = skb_cow_head(skb, 0);
3059 return ERR_PTR(err);
3062 /* Only report GSO partial support if it will enable us to
3063 * support segmentation on this frame without needing additional
3066 if (features & NETIF_F_GSO_PARTIAL) {
3067 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3068 struct net_device *dev = skb->dev;
3070 partial_features |= dev->features & dev->gso_partial_features;
3071 if (!skb_gso_ok(skb, features | partial_features))
3072 features &= ~NETIF_F_GSO_PARTIAL;
3075 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3076 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3078 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3079 SKB_GSO_CB(skb)->encap_level = 0;
3081 skb_reset_mac_header(skb);
3082 skb_reset_mac_len(skb);
3084 segs = skb_mac_gso_segment(skb, features);
3086 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3087 skb_warn_bad_offload(skb);
3091 EXPORT_SYMBOL(__skb_gso_segment);
3093 /* Take action when hardware reception checksum errors are detected. */
3095 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3097 if (net_ratelimit()) {
3098 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3100 pr_err("dev features: %pNF\n", &dev->features);
3101 pr_err("skb len=%u data_len=%u pkt_type=%u gso_size=%u gso_type=%u nr_frags=%u ip_summed=%u csum=%x csum_complete_sw=%d csum_valid=%d csum_level=%u\n",
3102 skb->len, skb->data_len, skb->pkt_type,
3103 skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type,
3104 skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum,
3105 skb->csum_complete_sw, skb->csum_valid, skb->csum_level);
3109 EXPORT_SYMBOL(netdev_rx_csum_fault);
3112 /* XXX: check that highmem exists at all on the given machine. */
3113 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3115 #ifdef CONFIG_HIGHMEM
3118 if (!(dev->features & NETIF_F_HIGHDMA)) {
3119 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3120 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3122 if (PageHighMem(skb_frag_page(frag)))
3130 /* If MPLS offload request, verify we are testing hardware MPLS features
3131 * instead of standard features for the netdev.
3133 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3134 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3135 netdev_features_t features,
3138 if (eth_p_mpls(type))
3139 features &= skb->dev->mpls_features;
3144 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3145 netdev_features_t features,
3152 static netdev_features_t harmonize_features(struct sk_buff *skb,
3153 netdev_features_t features)
3158 type = skb_network_protocol(skb, &tmp);
3159 features = net_mpls_features(skb, features, type);
3161 if (skb->ip_summed != CHECKSUM_NONE &&
3162 !can_checksum_protocol(features, type)) {
3163 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3165 if (illegal_highdma(skb->dev, skb))
3166 features &= ~NETIF_F_SG;
3171 netdev_features_t passthru_features_check(struct sk_buff *skb,
3172 struct net_device *dev,
3173 netdev_features_t features)
3177 EXPORT_SYMBOL(passthru_features_check);
3179 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3180 struct net_device *dev,
3181 netdev_features_t features)
3183 return vlan_features_check(skb, features);
3186 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3187 struct net_device *dev,
3188 netdev_features_t features)
3190 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3192 if (gso_segs > dev->gso_max_segs)
3193 return features & ~NETIF_F_GSO_MASK;
3195 /* Support for GSO partial features requires software
3196 * intervention before we can actually process the packets
3197 * so we need to strip support for any partial features now
3198 * and we can pull them back in after we have partially
3199 * segmented the frame.
3201 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3202 features &= ~dev->gso_partial_features;
3204 /* Make sure to clear the IPv4 ID mangling feature if the
3205 * IPv4 header has the potential to be fragmented.
3207 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3208 struct iphdr *iph = skb->encapsulation ?
3209 inner_ip_hdr(skb) : ip_hdr(skb);
3211 if (!(iph->frag_off & htons(IP_DF)))
3212 features &= ~NETIF_F_TSO_MANGLEID;
3218 netdev_features_t netif_skb_features(struct sk_buff *skb)
3220 struct net_device *dev = skb->dev;
3221 netdev_features_t features = dev->features;
3223 if (skb_is_gso(skb))
3224 features = gso_features_check(skb, dev, features);
3226 /* If encapsulation offload request, verify we are testing
3227 * hardware encapsulation features instead of standard
3228 * features for the netdev
3230 if (skb->encapsulation)
3231 features &= dev->hw_enc_features;
3233 if (skb_vlan_tagged(skb))
3234 features = netdev_intersect_features(features,
3235 dev->vlan_features |
3236 NETIF_F_HW_VLAN_CTAG_TX |
3237 NETIF_F_HW_VLAN_STAG_TX);
3239 if (dev->netdev_ops->ndo_features_check)
3240 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3243 features &= dflt_features_check(skb, dev, features);
3245 return harmonize_features(skb, features);
3247 EXPORT_SYMBOL(netif_skb_features);
3249 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3250 struct netdev_queue *txq, bool more)
3255 if (dev_nit_active(dev))
3256 dev_queue_xmit_nit(skb, dev);
3259 trace_net_dev_start_xmit(skb, dev);
3260 rc = netdev_start_xmit(skb, dev, txq, more);
3261 trace_net_dev_xmit(skb, rc, dev, len);
3266 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3267 struct netdev_queue *txq, int *ret)
3269 struct sk_buff *skb = first;
3270 int rc = NETDEV_TX_OK;
3273 struct sk_buff *next = skb->next;
3275 skb_mark_not_on_list(skb);
3276 rc = xmit_one(skb, dev, txq, next != NULL);
3277 if (unlikely(!dev_xmit_complete(rc))) {
3283 if (netif_tx_queue_stopped(txq) && skb) {
3284 rc = NETDEV_TX_BUSY;
3294 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3295 netdev_features_t features)
3297 if (skb_vlan_tag_present(skb) &&
3298 !vlan_hw_offload_capable(features, skb->vlan_proto))
3299 skb = __vlan_hwaccel_push_inside(skb);
3303 int skb_csum_hwoffload_help(struct sk_buff *skb,
3304 const netdev_features_t features)
3306 if (unlikely(skb->csum_not_inet))
3307 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3308 skb_crc32c_csum_help(skb);
3310 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3312 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3314 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3316 netdev_features_t features;
3318 features = netif_skb_features(skb);
3319 skb = validate_xmit_vlan(skb, features);
3323 skb = sk_validate_xmit_skb(skb, dev);
3327 if (netif_needs_gso(skb, features)) {
3328 struct sk_buff *segs;
3330 segs = skb_gso_segment(skb, features);
3338 if (skb_needs_linearize(skb, features) &&
3339 __skb_linearize(skb))
3342 /* If packet is not checksummed and device does not
3343 * support checksumming for this protocol, complete
3344 * checksumming here.
3346 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3347 if (skb->encapsulation)
3348 skb_set_inner_transport_header(skb,
3349 skb_checksum_start_offset(skb));
3351 skb_set_transport_header(skb,
3352 skb_checksum_start_offset(skb));
3353 if (skb_csum_hwoffload_help(skb, features))
3358 skb = validate_xmit_xfrm(skb, features, again);
3365 atomic_long_inc(&dev->tx_dropped);
3369 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3371 struct sk_buff *next, *head = NULL, *tail;
3373 for (; skb != NULL; skb = next) {
3375 skb_mark_not_on_list(skb);
3377 /* in case skb wont be segmented, point to itself */
3380 skb = validate_xmit_skb(skb, dev, again);
3388 /* If skb was segmented, skb->prev points to
3389 * the last segment. If not, it still contains skb.
3395 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3397 static void qdisc_pkt_len_init(struct sk_buff *skb)
3399 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3401 qdisc_skb_cb(skb)->pkt_len = skb->len;
3403 /* To get more precise estimation of bytes sent on wire,
3404 * we add to pkt_len the headers size of all segments
3406 if (shinfo->gso_size) {
3407 unsigned int hdr_len;
3408 u16 gso_segs = shinfo->gso_segs;
3410 /* mac layer + network layer */
3411 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3413 /* + transport layer */
3414 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3415 const struct tcphdr *th;
3416 struct tcphdr _tcphdr;
3418 th = skb_header_pointer(skb, skb_transport_offset(skb),
3419 sizeof(_tcphdr), &_tcphdr);
3421 hdr_len += __tcp_hdrlen(th);
3423 struct udphdr _udphdr;
3425 if (skb_header_pointer(skb, skb_transport_offset(skb),
3426 sizeof(_udphdr), &_udphdr))
3427 hdr_len += sizeof(struct udphdr);
3430 if (shinfo->gso_type & SKB_GSO_DODGY)
3431 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3434 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3438 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3439 struct net_device *dev,
3440 struct netdev_queue *txq)
3442 spinlock_t *root_lock = qdisc_lock(q);
3443 struct sk_buff *to_free = NULL;
3447 qdisc_calculate_pkt_len(skb, q);
3449 if (q->flags & TCQ_F_NOLOCK) {
3450 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3451 __qdisc_drop(skb, &to_free);
3454 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3458 if (unlikely(to_free))
3459 kfree_skb_list(to_free);
3464 * Heuristic to force contended enqueues to serialize on a
3465 * separate lock before trying to get qdisc main lock.
3466 * This permits qdisc->running owner to get the lock more
3467 * often and dequeue packets faster.
3469 contended = qdisc_is_running(q);
3470 if (unlikely(contended))
3471 spin_lock(&q->busylock);
3473 spin_lock(root_lock);
3474 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3475 __qdisc_drop(skb, &to_free);
3477 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3478 qdisc_run_begin(q)) {
3480 * This is a work-conserving queue; there are no old skbs
3481 * waiting to be sent out; and the qdisc is not running -
3482 * xmit the skb directly.
3485 qdisc_bstats_update(q, skb);
3487 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3488 if (unlikely(contended)) {
3489 spin_unlock(&q->busylock);
3496 rc = NET_XMIT_SUCCESS;
3498 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3499 if (qdisc_run_begin(q)) {
3500 if (unlikely(contended)) {
3501 spin_unlock(&q->busylock);
3508 spin_unlock(root_lock);
3509 if (unlikely(to_free))
3510 kfree_skb_list(to_free);
3511 if (unlikely(contended))
3512 spin_unlock(&q->busylock);
3516 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3517 static void skb_update_prio(struct sk_buff *skb)
3519 const struct netprio_map *map;
3520 const struct sock *sk;
3521 unsigned int prioidx;
3525 map = rcu_dereference_bh(skb->dev->priomap);
3528 sk = skb_to_full_sk(skb);
3532 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3534 if (prioidx < map->priomap_len)
3535 skb->priority = map->priomap[prioidx];
3538 #define skb_update_prio(skb)
3541 DEFINE_PER_CPU(int, xmit_recursion);
3542 EXPORT_SYMBOL(xmit_recursion);
3545 * dev_loopback_xmit - loop back @skb
3546 * @net: network namespace this loopback is happening in
3547 * @sk: sk needed to be a netfilter okfn
3548 * @skb: buffer to transmit
3550 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3552 skb_reset_mac_header(skb);
3553 __skb_pull(skb, skb_network_offset(skb));
3554 skb->pkt_type = PACKET_LOOPBACK;
3555 skb->ip_summed = CHECKSUM_UNNECESSARY;
3556 WARN_ON(!skb_dst(skb));
3561 EXPORT_SYMBOL(dev_loopback_xmit);
3563 #ifdef CONFIG_NET_EGRESS
3564 static struct sk_buff *
3565 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3567 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3568 struct tcf_result cl_res;
3573 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3574 mini_qdisc_bstats_cpu_update(miniq, skb);
3576 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3578 case TC_ACT_RECLASSIFY:
3579 skb->tc_index = TC_H_MIN(cl_res.classid);
3582 mini_qdisc_qstats_cpu_drop(miniq);
3583 *ret = NET_XMIT_DROP;
3589 *ret = NET_XMIT_SUCCESS;
3592 case TC_ACT_REDIRECT:
3593 /* No need to push/pop skb's mac_header here on egress! */
3594 skb_do_redirect(skb);
3595 *ret = NET_XMIT_SUCCESS;
3603 #endif /* CONFIG_NET_EGRESS */
3606 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3607 struct xps_dev_maps *dev_maps, unsigned int tci)
3609 struct xps_map *map;
3610 int queue_index = -1;
3614 tci += netdev_get_prio_tc_map(dev, skb->priority);
3617 map = rcu_dereference(dev_maps->attr_map[tci]);
3620 queue_index = map->queues[0];
3622 queue_index = map->queues[reciprocal_scale(
3623 skb_get_hash(skb), map->len)];
3624 if (unlikely(queue_index >= dev->real_num_tx_queues))
3631 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3632 struct sk_buff *skb)
3635 struct xps_dev_maps *dev_maps;
3636 struct sock *sk = skb->sk;
3637 int queue_index = -1;
3639 if (!static_key_false(&xps_needed))
3643 if (!static_key_false(&xps_rxqs_needed))
3646 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3648 int tci = sk_rx_queue_get(sk);
3650 if (tci >= 0 && tci < dev->num_rx_queues)
3651 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3656 if (queue_index < 0) {
3657 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3659 unsigned int tci = skb->sender_cpu - 1;
3661 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3673 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3674 struct net_device *sb_dev,
3675 select_queue_fallback_t fallback)
3679 EXPORT_SYMBOL(dev_pick_tx_zero);
3681 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3682 struct net_device *sb_dev,
3683 select_queue_fallback_t fallback)
3685 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3687 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3689 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3690 struct net_device *sb_dev)
3692 struct sock *sk = skb->sk;
3693 int queue_index = sk_tx_queue_get(sk);
3695 sb_dev = sb_dev ? : dev;
3697 if (queue_index < 0 || skb->ooo_okay ||
3698 queue_index >= dev->real_num_tx_queues) {
3699 int new_index = get_xps_queue(dev, sb_dev, skb);
3702 new_index = skb_tx_hash(dev, sb_dev, skb);
3704 if (queue_index != new_index && sk &&
3706 rcu_access_pointer(sk->sk_dst_cache))
3707 sk_tx_queue_set(sk, new_index);
3709 queue_index = new_index;
3715 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3716 struct sk_buff *skb,
3717 struct net_device *sb_dev)
3719 int queue_index = 0;
3722 u32 sender_cpu = skb->sender_cpu - 1;
3724 if (sender_cpu >= (u32)NR_CPUS)
3725 skb->sender_cpu = raw_smp_processor_id() + 1;
3728 if (dev->real_num_tx_queues != 1) {
3729 const struct net_device_ops *ops = dev->netdev_ops;
3731 if (ops->ndo_select_queue)
3732 queue_index = ops->ndo_select_queue(dev, skb, sb_dev,
3735 queue_index = __netdev_pick_tx(dev, skb, sb_dev);
3737 queue_index = netdev_cap_txqueue(dev, queue_index);
3740 skb_set_queue_mapping(skb, queue_index);
3741 return netdev_get_tx_queue(dev, queue_index);
3745 * __dev_queue_xmit - transmit a buffer
3746 * @skb: buffer to transmit
3747 * @sb_dev: suboordinate device used for L2 forwarding offload
3749 * Queue a buffer for transmission to a network device. The caller must
3750 * have set the device and priority and built the buffer before calling
3751 * this function. The function can be called from an interrupt.
3753 * A negative errno code is returned on a failure. A success does not
3754 * guarantee the frame will be transmitted as it may be dropped due
3755 * to congestion or traffic shaping.
3757 * -----------------------------------------------------------------------------------
3758 * I notice this method can also return errors from the queue disciplines,
3759 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3762 * Regardless of the return value, the skb is consumed, so it is currently
3763 * difficult to retry a send to this method. (You can bump the ref count
3764 * before sending to hold a reference for retry if you are careful.)
3766 * When calling this method, interrupts MUST be enabled. This is because
3767 * the BH enable code must have IRQs enabled so that it will not deadlock.
3770 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3772 struct net_device *dev = skb->dev;
3773 struct netdev_queue *txq;
3778 skb_reset_mac_header(skb);
3780 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3781 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3783 /* Disable soft irqs for various locks below. Also
3784 * stops preemption for RCU.
3788 skb_update_prio(skb);
3790 qdisc_pkt_len_init(skb);
3791 #ifdef CONFIG_NET_CLS_ACT
3792 skb->tc_at_ingress = 0;
3793 # ifdef CONFIG_NET_EGRESS
3794 if (static_branch_unlikely(&egress_needed_key)) {
3795 skb = sch_handle_egress(skb, &rc, dev);
3801 /* If device/qdisc don't need skb->dst, release it right now while
3802 * its hot in this cpu cache.
3804 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3809 txq = netdev_pick_tx(dev, skb, sb_dev);
3810 q = rcu_dereference_bh(txq->qdisc);
3812 trace_net_dev_queue(skb);
3814 rc = __dev_xmit_skb(skb, q, dev, txq);
3818 /* The device has no queue. Common case for software devices:
3819 * loopback, all the sorts of tunnels...
3821 * Really, it is unlikely that netif_tx_lock protection is necessary
3822 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3824 * However, it is possible, that they rely on protection
3827 * Check this and shot the lock. It is not prone from deadlocks.
3828 *Either shot noqueue qdisc, it is even simpler 8)
3830 if (dev->flags & IFF_UP) {
3831 int cpu = smp_processor_id(); /* ok because BHs are off */
3833 if (txq->xmit_lock_owner != cpu) {
3834 if (unlikely(__this_cpu_read(xmit_recursion) >
3835 XMIT_RECURSION_LIMIT))
3836 goto recursion_alert;
3838 skb = validate_xmit_skb(skb, dev, &again);
3842 HARD_TX_LOCK(dev, txq, cpu);
3844 if (!netif_xmit_stopped(txq)) {
3845 __this_cpu_inc(xmit_recursion);
3846 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3847 __this_cpu_dec(xmit_recursion);
3848 if (dev_xmit_complete(rc)) {
3849 HARD_TX_UNLOCK(dev, txq);
3853 HARD_TX_UNLOCK(dev, txq);
3854 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3857 /* Recursion is detected! It is possible,
3861 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3867 rcu_read_unlock_bh();
3869 atomic_long_inc(&dev->tx_dropped);
3870 kfree_skb_list(skb);
3873 rcu_read_unlock_bh();
3877 int dev_queue_xmit(struct sk_buff *skb)
3879 return __dev_queue_xmit(skb, NULL);
3881 EXPORT_SYMBOL(dev_queue_xmit);
3883 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3885 return __dev_queue_xmit(skb, sb_dev);
3887 EXPORT_SYMBOL(dev_queue_xmit_accel);
3889 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3891 struct net_device *dev = skb->dev;
3892 struct sk_buff *orig_skb = skb;
3893 struct netdev_queue *txq;
3894 int ret = NETDEV_TX_BUSY;
3897 if (unlikely(!netif_running(dev) ||
3898 !netif_carrier_ok(dev)))
3901 skb = validate_xmit_skb_list(skb, dev, &again);
3902 if (skb != orig_skb)
3905 skb_set_queue_mapping(skb, queue_id);
3906 txq = skb_get_tx_queue(dev, skb);
3910 HARD_TX_LOCK(dev, txq, smp_processor_id());
3911 if (!netif_xmit_frozen_or_drv_stopped(txq))
3912 ret = netdev_start_xmit(skb, dev, txq, false);
3913 HARD_TX_UNLOCK(dev, txq);
3917 if (!dev_xmit_complete(ret))
3922 atomic_long_inc(&dev->tx_dropped);
3923 kfree_skb_list(skb);
3924 return NET_XMIT_DROP;
3926 EXPORT_SYMBOL(dev_direct_xmit);
3928 /*************************************************************************
3930 *************************************************************************/
3932 int netdev_max_backlog __read_mostly = 1000;
3933 EXPORT_SYMBOL(netdev_max_backlog);
3935 int netdev_tstamp_prequeue __read_mostly = 1;
3936 int netdev_budget __read_mostly = 300;
3937 unsigned int __read_mostly netdev_budget_usecs = 2000;
3938 int weight_p __read_mostly = 64; /* old backlog weight */
3939 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3940 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3941 int dev_rx_weight __read_mostly = 64;
3942 int dev_tx_weight __read_mostly = 64;
3944 /* Called with irq disabled */
3945 static inline void ____napi_schedule(struct softnet_data *sd,
3946 struct napi_struct *napi)
3948 list_add_tail(&napi->poll_list, &sd->poll_list);
3949 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3954 /* One global table that all flow-based protocols share. */
3955 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3956 EXPORT_SYMBOL(rps_sock_flow_table);
3957 u32 rps_cpu_mask __read_mostly;
3958 EXPORT_SYMBOL(rps_cpu_mask);
3960 struct static_key rps_needed __read_mostly;
3961 EXPORT_SYMBOL(rps_needed);
3962 struct static_key rfs_needed __read_mostly;
3963 EXPORT_SYMBOL(rfs_needed);
3965 static struct rps_dev_flow *
3966 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3967 struct rps_dev_flow *rflow, u16 next_cpu)
3969 if (next_cpu < nr_cpu_ids) {
3970 #ifdef CONFIG_RFS_ACCEL
3971 struct netdev_rx_queue *rxqueue;
3972 struct rps_dev_flow_table *flow_table;
3973 struct rps_dev_flow *old_rflow;
3978 /* Should we steer this flow to a different hardware queue? */
3979 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3980 !(dev->features & NETIF_F_NTUPLE))
3982 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3983 if (rxq_index == skb_get_rx_queue(skb))
3986 rxqueue = dev->_rx + rxq_index;
3987 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3990 flow_id = skb_get_hash(skb) & flow_table->mask;
3991 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3992 rxq_index, flow_id);
3996 rflow = &flow_table->flows[flow_id];
3998 if (old_rflow->filter == rflow->filter)
3999 old_rflow->filter = RPS_NO_FILTER;
4003 per_cpu(softnet_data, next_cpu).input_queue_head;
4006 rflow->cpu = next_cpu;
4011 * get_rps_cpu is called from netif_receive_skb and returns the target
4012 * CPU from the RPS map of the receiving queue for a given skb.
4013 * rcu_read_lock must be held on entry.
4015 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4016 struct rps_dev_flow **rflowp)
4018 const struct rps_sock_flow_table *sock_flow_table;
4019 struct netdev_rx_queue *rxqueue = dev->_rx;
4020 struct rps_dev_flow_table *flow_table;
4021 struct rps_map *map;
4026 if (skb_rx_queue_recorded(skb)) {
4027 u16 index = skb_get_rx_queue(skb);
4029 if (unlikely(index >= dev->real_num_rx_queues)) {
4030 WARN_ONCE(dev->real_num_rx_queues > 1,
4031 "%s received packet on queue %u, but number "
4032 "of RX queues is %u\n",
4033 dev->name, index, dev->real_num_rx_queues);
4039 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4041 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4042 map = rcu_dereference(rxqueue->rps_map);
4043 if (!flow_table && !map)
4046 skb_reset_network_header(skb);
4047 hash = skb_get_hash(skb);
4051 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4052 if (flow_table && sock_flow_table) {
4053 struct rps_dev_flow *rflow;
4057 /* First check into global flow table if there is a match */
4058 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4059 if ((ident ^ hash) & ~rps_cpu_mask)
4062 next_cpu = ident & rps_cpu_mask;
4064 /* OK, now we know there is a match,
4065 * we can look at the local (per receive queue) flow table
4067 rflow = &flow_table->flows[hash & flow_table->mask];
4071 * If the desired CPU (where last recvmsg was done) is
4072 * different from current CPU (one in the rx-queue flow
4073 * table entry), switch if one of the following holds:
4074 * - Current CPU is unset (>= nr_cpu_ids).
4075 * - Current CPU is offline.
4076 * - The current CPU's queue tail has advanced beyond the
4077 * last packet that was enqueued using this table entry.
4078 * This guarantees that all previous packets for the flow
4079 * have been dequeued, thus preserving in order delivery.
4081 if (unlikely(tcpu != next_cpu) &&
4082 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4083 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4084 rflow->last_qtail)) >= 0)) {
4086 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4089 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4099 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4100 if (cpu_online(tcpu)) {
4110 #ifdef CONFIG_RFS_ACCEL
4113 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4114 * @dev: Device on which the filter was set
4115 * @rxq_index: RX queue index
4116 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4117 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4119 * Drivers that implement ndo_rx_flow_steer() should periodically call
4120 * this function for each installed filter and remove the filters for
4121 * which it returns %true.
4123 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4124 u32 flow_id, u16 filter_id)
4126 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4127 struct rps_dev_flow_table *flow_table;
4128 struct rps_dev_flow *rflow;
4133 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4134 if (flow_table && flow_id <= flow_table->mask) {
4135 rflow = &flow_table->flows[flow_id];
4136 cpu = READ_ONCE(rflow->cpu);
4137 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4138 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4139 rflow->last_qtail) <
4140 (int)(10 * flow_table->mask)))
4146 EXPORT_SYMBOL(rps_may_expire_flow);
4148 #endif /* CONFIG_RFS_ACCEL */
4150 /* Called from hardirq (IPI) context */
4151 static void rps_trigger_softirq(void *data)
4153 struct softnet_data *sd = data;
4155 ____napi_schedule(sd, &sd->backlog);
4159 #endif /* CONFIG_RPS */
4162 * Check if this softnet_data structure is another cpu one
4163 * If yes, queue it to our IPI list and return 1
4166 static int rps_ipi_queued(struct softnet_data *sd)
4169 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4172 sd->rps_ipi_next = mysd->rps_ipi_list;
4173 mysd->rps_ipi_list = sd;
4175 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4178 #endif /* CONFIG_RPS */
4182 #ifdef CONFIG_NET_FLOW_LIMIT
4183 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4186 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4188 #ifdef CONFIG_NET_FLOW_LIMIT
4189 struct sd_flow_limit *fl;
4190 struct softnet_data *sd;
4191 unsigned int old_flow, new_flow;
4193 if (qlen < (netdev_max_backlog >> 1))
4196 sd = this_cpu_ptr(&softnet_data);
4199 fl = rcu_dereference(sd->flow_limit);
4201 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4202 old_flow = fl->history[fl->history_head];
4203 fl->history[fl->history_head] = new_flow;
4206 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4208 if (likely(fl->buckets[old_flow]))
4209 fl->buckets[old_flow]--;
4211 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4223 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4224 * queue (may be a remote CPU queue).
4226 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4227 unsigned int *qtail)
4229 struct softnet_data *sd;
4230 unsigned long flags;
4233 sd = &per_cpu(softnet_data, cpu);
4235 local_irq_save(flags);
4238 if (!netif_running(skb->dev))
4240 qlen = skb_queue_len(&sd->input_pkt_queue);
4241 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4244 __skb_queue_tail(&sd->input_pkt_queue, skb);
4245 input_queue_tail_incr_save(sd, qtail);
4247 local_irq_restore(flags);
4248 return NET_RX_SUCCESS;
4251 /* Schedule NAPI for backlog device
4252 * We can use non atomic operation since we own the queue lock
4254 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4255 if (!rps_ipi_queued(sd))
4256 ____napi_schedule(sd, &sd->backlog);
4265 local_irq_restore(flags);
4267 atomic_long_inc(&skb->dev->rx_dropped);
4272 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4274 struct net_device *dev = skb->dev;
4275 struct netdev_rx_queue *rxqueue;
4279 if (skb_rx_queue_recorded(skb)) {
4280 u16 index = skb_get_rx_queue(skb);
4282 if (unlikely(index >= dev->real_num_rx_queues)) {
4283 WARN_ONCE(dev->real_num_rx_queues > 1,
4284 "%s received packet on queue %u, but number "
4285 "of RX queues is %u\n",
4286 dev->name, index, dev->real_num_rx_queues);
4288 return rxqueue; /* Return first rxqueue */
4295 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4296 struct xdp_buff *xdp,
4297 struct bpf_prog *xdp_prog)
4299 struct netdev_rx_queue *rxqueue;
4300 void *orig_data, *orig_data_end;
4301 u32 metalen, act = XDP_DROP;
4302 __be16 orig_eth_type;
4308 /* Reinjected packets coming from act_mirred or similar should
4309 * not get XDP generic processing.
4311 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4314 /* XDP packets must be linear and must have sufficient headroom
4315 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4316 * native XDP provides, thus we need to do it here as well.
4318 if (skb_is_nonlinear(skb) ||
4319 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4320 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4321 int troom = skb->tail + skb->data_len - skb->end;
4323 /* In case we have to go down the path and also linearize,
4324 * then lets do the pskb_expand_head() work just once here.
4326 if (pskb_expand_head(skb,
4327 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4328 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4330 if (skb_linearize(skb))
4334 /* The XDP program wants to see the packet starting at the MAC
4337 mac_len = skb->data - skb_mac_header(skb);
4338 hlen = skb_headlen(skb) + mac_len;
4339 xdp->data = skb->data - mac_len;
4340 xdp->data_meta = xdp->data;
4341 xdp->data_end = xdp->data + hlen;
4342 xdp->data_hard_start = skb->data - skb_headroom(skb);
4343 orig_data_end = xdp->data_end;
4344 orig_data = xdp->data;
4345 eth = (struct ethhdr *)xdp->data;
4346 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4347 orig_eth_type = eth->h_proto;
4349 rxqueue = netif_get_rxqueue(skb);
4350 xdp->rxq = &rxqueue->xdp_rxq;
4352 act = bpf_prog_run_xdp(xdp_prog, xdp);
4354 off = xdp->data - orig_data;
4356 __skb_pull(skb, off);
4358 __skb_push(skb, -off);
4359 skb->mac_header += off;
4361 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4364 off = orig_data_end - xdp->data_end;
4366 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4371 /* check if XDP changed eth hdr such SKB needs update */
4372 eth = (struct ethhdr *)xdp->data;
4373 if ((orig_eth_type != eth->h_proto) ||
4374 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4375 __skb_push(skb, ETH_HLEN);
4376 skb->protocol = eth_type_trans(skb, skb->dev);
4382 __skb_push(skb, mac_len);
4385 metalen = xdp->data - xdp->data_meta;
4387 skb_metadata_set(skb, metalen);
4390 bpf_warn_invalid_xdp_action(act);
4393 trace_xdp_exception(skb->dev, xdp_prog, act);
4404 /* When doing generic XDP we have to bypass the qdisc layer and the
4405 * network taps in order to match in-driver-XDP behavior.
4407 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4409 struct net_device *dev = skb->dev;
4410 struct netdev_queue *txq;
4411 bool free_skb = true;
4414 txq = netdev_pick_tx(dev, skb, NULL);
4415 cpu = smp_processor_id();
4416 HARD_TX_LOCK(dev, txq, cpu);
4417 if (!netif_xmit_stopped(txq)) {
4418 rc = netdev_start_xmit(skb, dev, txq, 0);
4419 if (dev_xmit_complete(rc))
4422 HARD_TX_UNLOCK(dev, txq);
4424 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4428 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4430 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4432 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4435 struct xdp_buff xdp;
4439 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4440 if (act != XDP_PASS) {
4443 err = xdp_do_generic_redirect(skb->dev, skb,
4449 generic_xdp_tx(skb, xdp_prog);
4460 EXPORT_SYMBOL_GPL(do_xdp_generic);
4462 static int netif_rx_internal(struct sk_buff *skb)
4466 net_timestamp_check(netdev_tstamp_prequeue, skb);
4468 trace_netif_rx(skb);
4470 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4475 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4479 /* Consider XDP consuming the packet a success from
4480 * the netdev point of view we do not want to count
4483 if (ret != XDP_PASS)
4484 return NET_RX_SUCCESS;
4488 if (static_key_false(&rps_needed)) {
4489 struct rps_dev_flow voidflow, *rflow = &voidflow;
4495 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4497 cpu = smp_processor_id();
4499 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4508 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4515 * netif_rx - post buffer to the network code
4516 * @skb: buffer to post
4518 * This function receives a packet from a device driver and queues it for
4519 * the upper (protocol) levels to process. It always succeeds. The buffer
4520 * may be dropped during processing for congestion control or by the
4524 * NET_RX_SUCCESS (no congestion)
4525 * NET_RX_DROP (packet was dropped)
4529 int netif_rx(struct sk_buff *skb)
4533 trace_netif_rx_entry(skb);
4535 ret = netif_rx_internal(skb);
4536 trace_netif_rx_exit(ret);
4540 EXPORT_SYMBOL(netif_rx);
4542 int netif_rx_ni(struct sk_buff *skb)
4546 trace_netif_rx_ni_entry(skb);
4549 err = netif_rx_internal(skb);
4550 if (local_softirq_pending())
4553 trace_netif_rx_ni_exit(err);
4557 EXPORT_SYMBOL(netif_rx_ni);
4559 static __latent_entropy void net_tx_action(struct softirq_action *h)
4561 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4563 if (sd->completion_queue) {
4564 struct sk_buff *clist;
4566 local_irq_disable();
4567 clist = sd->completion_queue;
4568 sd->completion_queue = NULL;
4572 struct sk_buff *skb = clist;
4574 clist = clist->next;
4576 WARN_ON(refcount_read(&skb->users));
4577 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4578 trace_consume_skb(skb);
4580 trace_kfree_skb(skb, net_tx_action);
4582 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4585 __kfree_skb_defer(skb);
4588 __kfree_skb_flush();
4591 if (sd->output_queue) {
4594 local_irq_disable();
4595 head = sd->output_queue;
4596 sd->output_queue = NULL;
4597 sd->output_queue_tailp = &sd->output_queue;
4601 struct Qdisc *q = head;
4602 spinlock_t *root_lock = NULL;
4604 head = head->next_sched;
4606 if (!(q->flags & TCQ_F_NOLOCK)) {
4607 root_lock = qdisc_lock(q);
4608 spin_lock(root_lock);
4610 /* We need to make sure head->next_sched is read
4611 * before clearing __QDISC_STATE_SCHED
4613 smp_mb__before_atomic();
4614 clear_bit(__QDISC_STATE_SCHED, &q->state);
4617 spin_unlock(root_lock);
4621 xfrm_dev_backlog(sd);
4624 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4625 /* This hook is defined here for ATM LANE */
4626 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4627 unsigned char *addr) __read_mostly;
4628 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4631 static inline struct sk_buff *
4632 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4633 struct net_device *orig_dev)
4635 #ifdef CONFIG_NET_CLS_ACT
4636 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4637 struct tcf_result cl_res;
4639 /* If there's at least one ingress present somewhere (so
4640 * we get here via enabled static key), remaining devices
4641 * that are not configured with an ingress qdisc will bail
4648 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4652 qdisc_skb_cb(skb)->pkt_len = skb->len;
4653 skb->tc_at_ingress = 1;
4654 mini_qdisc_bstats_cpu_update(miniq, skb);
4656 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4658 case TC_ACT_RECLASSIFY:
4659 skb->tc_index = TC_H_MIN(cl_res.classid);
4662 mini_qdisc_qstats_cpu_drop(miniq);
4670 case TC_ACT_REDIRECT:
4671 /* skb_mac_header check was done by cls/act_bpf, so
4672 * we can safely push the L2 header back before
4673 * redirecting to another netdev
4675 __skb_push(skb, skb->mac_len);
4676 skb_do_redirect(skb);
4678 case TC_ACT_REINSERT:
4679 /* this does not scrub the packet, and updates stats on error */
4680 skb_tc_reinsert(skb, &cl_res);
4685 #endif /* CONFIG_NET_CLS_ACT */
4690 * netdev_is_rx_handler_busy - check if receive handler is registered
4691 * @dev: device to check
4693 * Check if a receive handler is already registered for a given device.
4694 * Return true if there one.
4696 * The caller must hold the rtnl_mutex.
4698 bool netdev_is_rx_handler_busy(struct net_device *dev)
4701 return dev && rtnl_dereference(dev->rx_handler);
4703 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4706 * netdev_rx_handler_register - register receive handler
4707 * @dev: device to register a handler for
4708 * @rx_handler: receive handler to register
4709 * @rx_handler_data: data pointer that is used by rx handler
4711 * Register a receive handler for a device. This handler will then be
4712 * called from __netif_receive_skb. A negative errno code is returned
4715 * The caller must hold the rtnl_mutex.
4717 * For a general description of rx_handler, see enum rx_handler_result.
4719 int netdev_rx_handler_register(struct net_device *dev,
4720 rx_handler_func_t *rx_handler,
4721 void *rx_handler_data)
4723 if (netdev_is_rx_handler_busy(dev))
4726 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4729 /* Note: rx_handler_data must be set before rx_handler */
4730 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4731 rcu_assign_pointer(dev->rx_handler, rx_handler);
4735 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4738 * netdev_rx_handler_unregister - unregister receive handler
4739 * @dev: device to unregister a handler from
4741 * Unregister a receive handler from a device.
4743 * The caller must hold the rtnl_mutex.
4745 void netdev_rx_handler_unregister(struct net_device *dev)
4749 RCU_INIT_POINTER(dev->rx_handler, NULL);
4750 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4751 * section has a guarantee to see a non NULL rx_handler_data
4755 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4757 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4760 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4761 * the special handling of PFMEMALLOC skbs.
4763 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4765 switch (skb->protocol) {
4766 case htons(ETH_P_ARP):
4767 case htons(ETH_P_IP):
4768 case htons(ETH_P_IPV6):
4769 case htons(ETH_P_8021Q):
4770 case htons(ETH_P_8021AD):
4777 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4778 int *ret, struct net_device *orig_dev)
4780 #ifdef CONFIG_NETFILTER_INGRESS
4781 if (nf_hook_ingress_active(skb)) {
4785 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4790 ingress_retval = nf_hook_ingress(skb);
4792 return ingress_retval;
4794 #endif /* CONFIG_NETFILTER_INGRESS */
4798 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4799 struct packet_type **ppt_prev)
4801 struct packet_type *ptype, *pt_prev;
4802 rx_handler_func_t *rx_handler;
4803 struct net_device *orig_dev;
4804 bool deliver_exact = false;
4805 int ret = NET_RX_DROP;
4808 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4810 trace_netif_receive_skb(skb);
4812 orig_dev = skb->dev;
4814 skb_reset_network_header(skb);
4815 if (!skb_transport_header_was_set(skb))
4816 skb_reset_transport_header(skb);
4817 skb_reset_mac_len(skb);
4822 skb->skb_iif = skb->dev->ifindex;
4824 __this_cpu_inc(softnet_data.processed);
4826 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4827 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4828 skb = skb_vlan_untag(skb);
4833 if (skb_skip_tc_classify(skb))
4839 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4841 ret = deliver_skb(skb, pt_prev, orig_dev);
4845 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4847 ret = deliver_skb(skb, pt_prev, orig_dev);
4852 #ifdef CONFIG_NET_INGRESS
4853 if (static_branch_unlikely(&ingress_needed_key)) {
4854 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4858 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4864 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4867 if (skb_vlan_tag_present(skb)) {
4869 ret = deliver_skb(skb, pt_prev, orig_dev);
4872 if (vlan_do_receive(&skb))
4874 else if (unlikely(!skb))
4878 rx_handler = rcu_dereference(skb->dev->rx_handler);
4881 ret = deliver_skb(skb, pt_prev, orig_dev);
4884 switch (rx_handler(&skb)) {
4885 case RX_HANDLER_CONSUMED:
4886 ret = NET_RX_SUCCESS;
4888 case RX_HANDLER_ANOTHER:
4890 case RX_HANDLER_EXACT:
4891 deliver_exact = true;
4892 case RX_HANDLER_PASS:
4899 if (unlikely(skb_vlan_tag_present(skb))) {
4900 if (skb_vlan_tag_get_id(skb))
4901 skb->pkt_type = PACKET_OTHERHOST;
4902 /* Note: we might in the future use prio bits
4903 * and set skb->priority like in vlan_do_receive()
4904 * For the time being, just ignore Priority Code Point
4906 __vlan_hwaccel_clear_tag(skb);
4909 type = skb->protocol;
4911 /* deliver only exact match when indicated */
4912 if (likely(!deliver_exact)) {
4913 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4914 &ptype_base[ntohs(type) &
4918 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4919 &orig_dev->ptype_specific);
4921 if (unlikely(skb->dev != orig_dev)) {
4922 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4923 &skb->dev->ptype_specific);
4927 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4929 *ppt_prev = pt_prev;
4933 atomic_long_inc(&skb->dev->rx_dropped);
4935 atomic_long_inc(&skb->dev->rx_nohandler);
4937 /* Jamal, now you will not able to escape explaining
4938 * me how you were going to use this. :-)
4947 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4949 struct net_device *orig_dev = skb->dev;
4950 struct packet_type *pt_prev = NULL;
4953 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4955 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4960 * netif_receive_skb_core - special purpose version of netif_receive_skb
4961 * @skb: buffer to process
4963 * More direct receive version of netif_receive_skb(). It should
4964 * only be used by callers that have a need to skip RPS and Generic XDP.
4965 * Caller must also take care of handling if (page_is_)pfmemalloc.
4967 * This function may only be called from softirq context and interrupts
4968 * should be enabled.
4970 * Return values (usually ignored):
4971 * NET_RX_SUCCESS: no congestion
4972 * NET_RX_DROP: packet was dropped
4974 int netif_receive_skb_core(struct sk_buff *skb)
4979 ret = __netif_receive_skb_one_core(skb, false);
4984 EXPORT_SYMBOL(netif_receive_skb_core);
4986 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
4987 struct packet_type *pt_prev,
4988 struct net_device *orig_dev)
4990 struct sk_buff *skb, *next;
4994 if (list_empty(head))
4996 if (pt_prev->list_func != NULL)
4997 pt_prev->list_func(head, pt_prev, orig_dev);
4999 list_for_each_entry_safe(skb, next, head, list)
5000 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5003 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5005 /* Fast-path assumptions:
5006 * - There is no RX handler.
5007 * - Only one packet_type matches.
5008 * If either of these fails, we will end up doing some per-packet
5009 * processing in-line, then handling the 'last ptype' for the whole
5010 * sublist. This can't cause out-of-order delivery to any single ptype,
5011 * because the 'last ptype' must be constant across the sublist, and all
5012 * other ptypes are handled per-packet.
5014 /* Current (common) ptype of sublist */
5015 struct packet_type *pt_curr = NULL;
5016 /* Current (common) orig_dev of sublist */
5017 struct net_device *od_curr = NULL;
5018 struct list_head sublist;
5019 struct sk_buff *skb, *next;
5021 INIT_LIST_HEAD(&sublist);
5022 list_for_each_entry_safe(skb, next, head, list) {
5023 struct net_device *orig_dev = skb->dev;
5024 struct packet_type *pt_prev = NULL;
5026 list_del(&skb->list);
5027 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5030 if (pt_curr != pt_prev || od_curr != orig_dev) {
5031 /* dispatch old sublist */
5032 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5033 /* start new sublist */
5034 INIT_LIST_HEAD(&sublist);
5038 list_add_tail(&skb->list, &sublist);
5041 /* dispatch final sublist */
5042 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5045 static int __netif_receive_skb(struct sk_buff *skb)
5049 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5050 unsigned int noreclaim_flag;
5053 * PFMEMALLOC skbs are special, they should
5054 * - be delivered to SOCK_MEMALLOC sockets only
5055 * - stay away from userspace
5056 * - have bounded memory usage
5058 * Use PF_MEMALLOC as this saves us from propagating the allocation
5059 * context down to all allocation sites.
5061 noreclaim_flag = memalloc_noreclaim_save();
5062 ret = __netif_receive_skb_one_core(skb, true);
5063 memalloc_noreclaim_restore(noreclaim_flag);
5065 ret = __netif_receive_skb_one_core(skb, false);
5070 static void __netif_receive_skb_list(struct list_head *head)
5072 unsigned long noreclaim_flag = 0;
5073 struct sk_buff *skb, *next;
5074 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5076 list_for_each_entry_safe(skb, next, head, list) {
5077 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5078 struct list_head sublist;
5080 /* Handle the previous sublist */
5081 list_cut_before(&sublist, head, &skb->list);
5082 if (!list_empty(&sublist))
5083 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5084 pfmemalloc = !pfmemalloc;
5085 /* See comments in __netif_receive_skb */
5087 noreclaim_flag = memalloc_noreclaim_save();
5089 memalloc_noreclaim_restore(noreclaim_flag);
5092 /* Handle the remaining sublist */
5093 if (!list_empty(head))
5094 __netif_receive_skb_list_core(head, pfmemalloc);
5095 /* Restore pflags */
5097 memalloc_noreclaim_restore(noreclaim_flag);
5100 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5102 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5103 struct bpf_prog *new = xdp->prog;
5106 switch (xdp->command) {
5107 case XDP_SETUP_PROG:
5108 rcu_assign_pointer(dev->xdp_prog, new);
5113 static_branch_dec(&generic_xdp_needed_key);
5114 } else if (new && !old) {
5115 static_branch_inc(&generic_xdp_needed_key);
5116 dev_disable_lro(dev);
5117 dev_disable_gro_hw(dev);
5121 case XDP_QUERY_PROG:
5122 xdp->prog_id = old ? old->aux->id : 0;
5133 static int netif_receive_skb_internal(struct sk_buff *skb)
5137 net_timestamp_check(netdev_tstamp_prequeue, skb);
5139 if (skb_defer_rx_timestamp(skb))
5140 return NET_RX_SUCCESS;
5142 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5147 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5151 if (ret != XDP_PASS)
5157 if (static_key_false(&rps_needed)) {
5158 struct rps_dev_flow voidflow, *rflow = &voidflow;
5159 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5162 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5168 ret = __netif_receive_skb(skb);
5173 static void netif_receive_skb_list_internal(struct list_head *head)
5175 struct bpf_prog *xdp_prog = NULL;
5176 struct sk_buff *skb, *next;
5177 struct list_head sublist;
5179 INIT_LIST_HEAD(&sublist);
5180 list_for_each_entry_safe(skb, next, head, list) {
5181 net_timestamp_check(netdev_tstamp_prequeue, skb);
5182 list_del(&skb->list);
5183 if (!skb_defer_rx_timestamp(skb))
5184 list_add_tail(&skb->list, &sublist);
5186 list_splice_init(&sublist, head);
5188 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5191 list_for_each_entry_safe(skb, next, head, list) {
5192 xdp_prog = rcu_dereference(skb->dev->xdp_prog);
5193 list_del(&skb->list);
5194 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS)
5195 list_add_tail(&skb->list, &sublist);
5199 /* Put passed packets back on main list */
5200 list_splice_init(&sublist, head);
5205 if (static_key_false(&rps_needed)) {
5206 list_for_each_entry_safe(skb, next, head, list) {
5207 struct rps_dev_flow voidflow, *rflow = &voidflow;
5208 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5211 /* Will be handled, remove from list */
5212 list_del(&skb->list);
5213 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5218 __netif_receive_skb_list(head);
5223 * netif_receive_skb - process receive buffer from network
5224 * @skb: buffer to process
5226 * netif_receive_skb() is the main receive data processing function.
5227 * It always succeeds. The buffer may be dropped during processing
5228 * for congestion control or by the protocol layers.
5230 * This function may only be called from softirq context and interrupts
5231 * should be enabled.
5233 * Return values (usually ignored):
5234 * NET_RX_SUCCESS: no congestion
5235 * NET_RX_DROP: packet was dropped
5237 int netif_receive_skb(struct sk_buff *skb)
5241 trace_netif_receive_skb_entry(skb);
5243 ret = netif_receive_skb_internal(skb);
5244 trace_netif_receive_skb_exit(ret);
5248 EXPORT_SYMBOL(netif_receive_skb);
5251 * netif_receive_skb_list - process many receive buffers from network
5252 * @head: list of skbs to process.
5254 * Since return value of netif_receive_skb() is normally ignored, and
5255 * wouldn't be meaningful for a list, this function returns void.
5257 * This function may only be called from softirq context and interrupts
5258 * should be enabled.
5260 void netif_receive_skb_list(struct list_head *head)
5262 struct sk_buff *skb;
5264 if (list_empty(head))
5266 if (trace_netif_receive_skb_list_entry_enabled()) {
5267 list_for_each_entry(skb, head, list)
5268 trace_netif_receive_skb_list_entry(skb);
5270 netif_receive_skb_list_internal(head);
5271 trace_netif_receive_skb_list_exit(0);
5273 EXPORT_SYMBOL(netif_receive_skb_list);
5275 DEFINE_PER_CPU(struct work_struct, flush_works);
5277 /* Network device is going away, flush any packets still pending */
5278 static void flush_backlog(struct work_struct *work)
5280 struct sk_buff *skb, *tmp;
5281 struct softnet_data *sd;
5284 sd = this_cpu_ptr(&softnet_data);
5286 local_irq_disable();
5288 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5289 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5290 __skb_unlink(skb, &sd->input_pkt_queue);
5292 input_queue_head_incr(sd);
5298 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5299 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5300 __skb_unlink(skb, &sd->process_queue);
5302 input_queue_head_incr(sd);
5308 static void flush_all_backlogs(void)
5314 for_each_online_cpu(cpu)
5315 queue_work_on(cpu, system_highpri_wq,
5316 per_cpu_ptr(&flush_works, cpu));
5318 for_each_online_cpu(cpu)
5319 flush_work(per_cpu_ptr(&flush_works, cpu));
5324 static int napi_gro_complete(struct sk_buff *skb)
5326 struct packet_offload *ptype;
5327 __be16 type = skb->protocol;
5328 struct list_head *head = &offload_base;
5331 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5333 if (NAPI_GRO_CB(skb)->count == 1) {
5334 skb_shinfo(skb)->gso_size = 0;
5339 list_for_each_entry_rcu(ptype, head, list) {
5340 if (ptype->type != type || !ptype->callbacks.gro_complete)
5343 err = ptype->callbacks.gro_complete(skb, 0);
5349 WARN_ON(&ptype->list == head);
5351 return NET_RX_SUCCESS;
5355 return netif_receive_skb_internal(skb);
5358 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5361 struct list_head *head = &napi->gro_hash[index].list;
5362 struct sk_buff *skb, *p;
5364 list_for_each_entry_safe_reverse(skb, p, head, list) {
5365 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5367 skb_list_del_init(skb);
5368 napi_gro_complete(skb);
5369 napi->gro_hash[index].count--;
5372 if (!napi->gro_hash[index].count)
5373 __clear_bit(index, &napi->gro_bitmask);
5376 /* napi->gro_hash[].list contains packets ordered by age.
5377 * youngest packets at the head of it.
5378 * Complete skbs in reverse order to reduce latencies.
5380 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5382 unsigned long bitmask = napi->gro_bitmask;
5383 unsigned int i, base = ~0U;
5385 while ((i = ffs(bitmask)) != 0) {
5388 __napi_gro_flush_chain(napi, base, flush_old);
5391 EXPORT_SYMBOL(napi_gro_flush);
5393 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5394 struct sk_buff *skb)
5396 unsigned int maclen = skb->dev->hard_header_len;
5397 u32 hash = skb_get_hash_raw(skb);
5398 struct list_head *head;
5401 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5402 list_for_each_entry(p, head, list) {
5403 unsigned long diffs;
5405 NAPI_GRO_CB(p)->flush = 0;
5407 if (hash != skb_get_hash_raw(p)) {
5408 NAPI_GRO_CB(p)->same_flow = 0;
5412 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5413 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5414 if (skb_vlan_tag_present(p))
5415 diffs |= p->vlan_tci ^ skb->vlan_tci;
5416 diffs |= skb_metadata_dst_cmp(p, skb);
5417 diffs |= skb_metadata_differs(p, skb);
5418 if (maclen == ETH_HLEN)
5419 diffs |= compare_ether_header(skb_mac_header(p),
5420 skb_mac_header(skb));
5422 diffs = memcmp(skb_mac_header(p),
5423 skb_mac_header(skb),
5425 NAPI_GRO_CB(p)->same_flow = !diffs;
5431 static void skb_gro_reset_offset(struct sk_buff *skb)
5433 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5434 const skb_frag_t *frag0 = &pinfo->frags[0];
5436 NAPI_GRO_CB(skb)->data_offset = 0;
5437 NAPI_GRO_CB(skb)->frag0 = NULL;
5438 NAPI_GRO_CB(skb)->frag0_len = 0;
5440 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5442 !PageHighMem(skb_frag_page(frag0))) {
5443 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5444 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5445 skb_frag_size(frag0),
5446 skb->end - skb->tail);
5450 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5452 struct skb_shared_info *pinfo = skb_shinfo(skb);
5454 BUG_ON(skb->end - skb->tail < grow);
5456 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5458 skb->data_len -= grow;
5461 pinfo->frags[0].page_offset += grow;
5462 skb_frag_size_sub(&pinfo->frags[0], grow);
5464 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5465 skb_frag_unref(skb, 0);
5466 memmove(pinfo->frags, pinfo->frags + 1,
5467 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5471 static void gro_flush_oldest(struct list_head *head)
5473 struct sk_buff *oldest;
5475 oldest = list_last_entry(head, struct sk_buff, list);
5477 /* We are called with head length >= MAX_GRO_SKBS, so this is
5480 if (WARN_ON_ONCE(!oldest))
5483 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5486 skb_list_del_init(oldest);
5487 napi_gro_complete(oldest);
5490 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5492 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5493 struct list_head *head = &offload_base;
5494 struct packet_offload *ptype;
5495 __be16 type = skb->protocol;
5496 struct list_head *gro_head;
5497 struct sk_buff *pp = NULL;
5498 enum gro_result ret;
5502 if (netif_elide_gro(skb->dev))
5505 gro_head = gro_list_prepare(napi, skb);
5508 list_for_each_entry_rcu(ptype, head, list) {
5509 if (ptype->type != type || !ptype->callbacks.gro_receive)
5512 skb_set_network_header(skb, skb_gro_offset(skb));
5513 skb_reset_mac_len(skb);
5514 NAPI_GRO_CB(skb)->same_flow = 0;
5515 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5516 NAPI_GRO_CB(skb)->free = 0;
5517 NAPI_GRO_CB(skb)->encap_mark = 0;
5518 NAPI_GRO_CB(skb)->recursion_counter = 0;
5519 NAPI_GRO_CB(skb)->is_fou = 0;
5520 NAPI_GRO_CB(skb)->is_atomic = 1;
5521 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5523 /* Setup for GRO checksum validation */
5524 switch (skb->ip_summed) {
5525 case CHECKSUM_COMPLETE:
5526 NAPI_GRO_CB(skb)->csum = skb->csum;
5527 NAPI_GRO_CB(skb)->csum_valid = 1;
5528 NAPI_GRO_CB(skb)->csum_cnt = 0;
5530 case CHECKSUM_UNNECESSARY:
5531 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5532 NAPI_GRO_CB(skb)->csum_valid = 0;
5535 NAPI_GRO_CB(skb)->csum_cnt = 0;
5536 NAPI_GRO_CB(skb)->csum_valid = 0;
5539 pp = ptype->callbacks.gro_receive(gro_head, skb);
5544 if (&ptype->list == head)
5547 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5552 same_flow = NAPI_GRO_CB(skb)->same_flow;
5553 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5556 skb_list_del_init(pp);
5557 napi_gro_complete(pp);
5558 napi->gro_hash[hash].count--;
5564 if (NAPI_GRO_CB(skb)->flush)
5567 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5568 gro_flush_oldest(gro_head);
5570 napi->gro_hash[hash].count++;
5572 NAPI_GRO_CB(skb)->count = 1;
5573 NAPI_GRO_CB(skb)->age = jiffies;
5574 NAPI_GRO_CB(skb)->last = skb;
5575 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5576 list_add(&skb->list, gro_head);
5580 grow = skb_gro_offset(skb) - skb_headlen(skb);
5582 gro_pull_from_frag0(skb, grow);
5584 if (napi->gro_hash[hash].count) {
5585 if (!test_bit(hash, &napi->gro_bitmask))
5586 __set_bit(hash, &napi->gro_bitmask);
5587 } else if (test_bit(hash, &napi->gro_bitmask)) {
5588 __clear_bit(hash, &napi->gro_bitmask);
5598 struct packet_offload *gro_find_receive_by_type(__be16 type)
5600 struct list_head *offload_head = &offload_base;
5601 struct packet_offload *ptype;
5603 list_for_each_entry_rcu(ptype, offload_head, list) {
5604 if (ptype->type != type || !ptype->callbacks.gro_receive)
5610 EXPORT_SYMBOL(gro_find_receive_by_type);
5612 struct packet_offload *gro_find_complete_by_type(__be16 type)
5614 struct list_head *offload_head = &offload_base;
5615 struct packet_offload *ptype;
5617 list_for_each_entry_rcu(ptype, offload_head, list) {
5618 if (ptype->type != type || !ptype->callbacks.gro_complete)
5624 EXPORT_SYMBOL(gro_find_complete_by_type);
5626 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5630 kmem_cache_free(skbuff_head_cache, skb);
5633 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5637 if (netif_receive_skb_internal(skb))
5645 case GRO_MERGED_FREE:
5646 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5647 napi_skb_free_stolen_head(skb);
5661 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5665 skb_mark_napi_id(skb, napi);
5666 trace_napi_gro_receive_entry(skb);
5668 skb_gro_reset_offset(skb);
5670 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb);
5671 trace_napi_gro_receive_exit(ret);
5675 EXPORT_SYMBOL(napi_gro_receive);
5677 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5679 if (unlikely(skb->pfmemalloc)) {
5683 __skb_pull(skb, skb_headlen(skb));
5684 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5685 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5686 __vlan_hwaccel_clear_tag(skb);
5687 skb->dev = napi->dev;
5690 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5691 skb->pkt_type = PACKET_HOST;
5693 skb->encapsulation = 0;
5694 skb_shinfo(skb)->gso_type = 0;
5695 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5701 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5703 struct sk_buff *skb = napi->skb;
5706 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5709 skb_mark_napi_id(skb, napi);
5714 EXPORT_SYMBOL(napi_get_frags);
5716 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5717 struct sk_buff *skb,
5723 __skb_push(skb, ETH_HLEN);
5724 skb->protocol = eth_type_trans(skb, skb->dev);
5725 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5730 napi_reuse_skb(napi, skb);
5733 case GRO_MERGED_FREE:
5734 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5735 napi_skb_free_stolen_head(skb);
5737 napi_reuse_skb(napi, skb);
5748 /* Upper GRO stack assumes network header starts at gro_offset=0
5749 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5750 * We copy ethernet header into skb->data to have a common layout.
5752 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5754 struct sk_buff *skb = napi->skb;
5755 const struct ethhdr *eth;
5756 unsigned int hlen = sizeof(*eth);
5760 skb_reset_mac_header(skb);
5761 skb_gro_reset_offset(skb);
5763 eth = skb_gro_header_fast(skb, 0);
5764 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5765 eth = skb_gro_header_slow(skb, hlen, 0);
5766 if (unlikely(!eth)) {
5767 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5768 __func__, napi->dev->name);
5769 napi_reuse_skb(napi, skb);
5773 gro_pull_from_frag0(skb, hlen);
5774 NAPI_GRO_CB(skb)->frag0 += hlen;
5775 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5777 __skb_pull(skb, hlen);
5780 * This works because the only protocols we care about don't require
5782 * We'll fix it up properly in napi_frags_finish()
5784 skb->protocol = eth->h_proto;
5789 gro_result_t napi_gro_frags(struct napi_struct *napi)
5792 struct sk_buff *skb = napi_frags_skb(napi);
5797 trace_napi_gro_frags_entry(skb);
5799 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5800 trace_napi_gro_frags_exit(ret);
5804 EXPORT_SYMBOL(napi_gro_frags);
5806 /* Compute the checksum from gro_offset and return the folded value
5807 * after adding in any pseudo checksum.
5809 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5814 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5816 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5817 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5818 /* See comments in __skb_checksum_complete(). */
5820 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5821 !skb->csum_complete_sw)
5822 netdev_rx_csum_fault(skb->dev, skb);
5825 NAPI_GRO_CB(skb)->csum = wsum;
5826 NAPI_GRO_CB(skb)->csum_valid = 1;
5830 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5832 static void net_rps_send_ipi(struct softnet_data *remsd)
5836 struct softnet_data *next = remsd->rps_ipi_next;
5838 if (cpu_online(remsd->cpu))
5839 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5846 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5847 * Note: called with local irq disabled, but exits with local irq enabled.
5849 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5852 struct softnet_data *remsd = sd->rps_ipi_list;
5855 sd->rps_ipi_list = NULL;
5859 /* Send pending IPI's to kick RPS processing on remote cpus. */
5860 net_rps_send_ipi(remsd);
5866 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5869 return sd->rps_ipi_list != NULL;
5875 static int process_backlog(struct napi_struct *napi, int quota)
5877 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5881 /* Check if we have pending ipi, its better to send them now,
5882 * not waiting net_rx_action() end.
5884 if (sd_has_rps_ipi_waiting(sd)) {
5885 local_irq_disable();
5886 net_rps_action_and_irq_enable(sd);
5889 napi->weight = dev_rx_weight;
5891 struct sk_buff *skb;
5893 while ((skb = __skb_dequeue(&sd->process_queue))) {
5895 __netif_receive_skb(skb);
5897 input_queue_head_incr(sd);
5898 if (++work >= quota)
5903 local_irq_disable();
5905 if (skb_queue_empty(&sd->input_pkt_queue)) {
5907 * Inline a custom version of __napi_complete().
5908 * only current cpu owns and manipulates this napi,
5909 * and NAPI_STATE_SCHED is the only possible flag set
5911 * We can use a plain write instead of clear_bit(),
5912 * and we dont need an smp_mb() memory barrier.
5917 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5918 &sd->process_queue);
5928 * __napi_schedule - schedule for receive
5929 * @n: entry to schedule
5931 * The entry's receive function will be scheduled to run.
5932 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5934 void __napi_schedule(struct napi_struct *n)
5936 unsigned long flags;
5938 local_irq_save(flags);
5939 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5940 local_irq_restore(flags);
5942 EXPORT_SYMBOL(__napi_schedule);
5945 * napi_schedule_prep - check if napi can be scheduled
5948 * Test if NAPI routine is already running, and if not mark
5949 * it as running. This is used as a condition variable
5950 * insure only one NAPI poll instance runs. We also make
5951 * sure there is no pending NAPI disable.
5953 bool napi_schedule_prep(struct napi_struct *n)
5955 unsigned long val, new;
5958 val = READ_ONCE(n->state);
5959 if (unlikely(val & NAPIF_STATE_DISABLE))
5961 new = val | NAPIF_STATE_SCHED;
5963 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5964 * This was suggested by Alexander Duyck, as compiler
5965 * emits better code than :
5966 * if (val & NAPIF_STATE_SCHED)
5967 * new |= NAPIF_STATE_MISSED;
5969 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5971 } while (cmpxchg(&n->state, val, new) != val);
5973 return !(val & NAPIF_STATE_SCHED);
5975 EXPORT_SYMBOL(napi_schedule_prep);
5978 * __napi_schedule_irqoff - schedule for receive
5979 * @n: entry to schedule
5981 * Variant of __napi_schedule() assuming hard irqs are masked
5983 void __napi_schedule_irqoff(struct napi_struct *n)
5985 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5987 EXPORT_SYMBOL(__napi_schedule_irqoff);
5989 bool napi_complete_done(struct napi_struct *n, int work_done)
5991 unsigned long flags, val, new;
5994 * 1) Don't let napi dequeue from the cpu poll list
5995 * just in case its running on a different cpu.
5996 * 2) If we are busy polling, do nothing here, we have
5997 * the guarantee we will be called later.
5999 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6000 NAPIF_STATE_IN_BUSY_POLL)))
6003 if (n->gro_bitmask) {
6004 unsigned long timeout = 0;
6007 timeout = n->dev->gro_flush_timeout;
6009 /* When the NAPI instance uses a timeout and keeps postponing
6010 * it, we need to bound somehow the time packets are kept in
6013 napi_gro_flush(n, !!timeout);
6015 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6016 HRTIMER_MODE_REL_PINNED);
6018 if (unlikely(!list_empty(&n->poll_list))) {
6019 /* If n->poll_list is not empty, we need to mask irqs */
6020 local_irq_save(flags);
6021 list_del_init(&n->poll_list);
6022 local_irq_restore(flags);
6026 val = READ_ONCE(n->state);
6028 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6030 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6032 /* If STATE_MISSED was set, leave STATE_SCHED set,
6033 * because we will call napi->poll() one more time.
6034 * This C code was suggested by Alexander Duyck to help gcc.
6036 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6038 } while (cmpxchg(&n->state, val, new) != val);
6040 if (unlikely(val & NAPIF_STATE_MISSED)) {
6047 EXPORT_SYMBOL(napi_complete_done);
6049 /* must be called under rcu_read_lock(), as we dont take a reference */
6050 static struct napi_struct *napi_by_id(unsigned int napi_id)
6052 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6053 struct napi_struct *napi;
6055 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6056 if (napi->napi_id == napi_id)
6062 #if defined(CONFIG_NET_RX_BUSY_POLL)
6064 #define BUSY_POLL_BUDGET 8
6066 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6070 /* Busy polling means there is a high chance device driver hard irq
6071 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6072 * set in napi_schedule_prep().
6073 * Since we are about to call napi->poll() once more, we can safely
6074 * clear NAPI_STATE_MISSED.
6076 * Note: x86 could use a single "lock and ..." instruction
6077 * to perform these two clear_bit()
6079 clear_bit(NAPI_STATE_MISSED, &napi->state);
6080 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6084 /* All we really want here is to re-enable device interrupts.
6085 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6087 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6088 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6089 netpoll_poll_unlock(have_poll_lock);
6090 if (rc == BUSY_POLL_BUDGET)
6091 __napi_schedule(napi);
6095 void napi_busy_loop(unsigned int napi_id,
6096 bool (*loop_end)(void *, unsigned long),
6099 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6100 int (*napi_poll)(struct napi_struct *napi, int budget);
6101 void *have_poll_lock = NULL;
6102 struct napi_struct *napi;
6109 napi = napi_by_id(napi_id);
6119 unsigned long val = READ_ONCE(napi->state);
6121 /* If multiple threads are competing for this napi,
6122 * we avoid dirtying napi->state as much as we can.
6124 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6125 NAPIF_STATE_IN_BUSY_POLL))
6127 if (cmpxchg(&napi->state, val,
6128 val | NAPIF_STATE_IN_BUSY_POLL |
6129 NAPIF_STATE_SCHED) != val)
6131 have_poll_lock = netpoll_poll_lock(napi);
6132 napi_poll = napi->poll;
6134 work = napi_poll(napi, BUSY_POLL_BUDGET);
6135 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6138 __NET_ADD_STATS(dev_net(napi->dev),
6139 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6142 if (!loop_end || loop_end(loop_end_arg, start_time))
6145 if (unlikely(need_resched())) {
6147 busy_poll_stop(napi, have_poll_lock);
6151 if (loop_end(loop_end_arg, start_time))
6158 busy_poll_stop(napi, have_poll_lock);
6163 EXPORT_SYMBOL(napi_busy_loop);
6165 #endif /* CONFIG_NET_RX_BUSY_POLL */
6167 static void napi_hash_add(struct napi_struct *napi)
6169 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6170 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6173 spin_lock(&napi_hash_lock);
6175 /* 0..NR_CPUS range is reserved for sender_cpu use */
6177 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6178 napi_gen_id = MIN_NAPI_ID;
6179 } while (napi_by_id(napi_gen_id));
6180 napi->napi_id = napi_gen_id;
6182 hlist_add_head_rcu(&napi->napi_hash_node,
6183 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6185 spin_unlock(&napi_hash_lock);
6188 /* Warning : caller is responsible to make sure rcu grace period
6189 * is respected before freeing memory containing @napi
6191 bool napi_hash_del(struct napi_struct *napi)
6193 bool rcu_sync_needed = false;
6195 spin_lock(&napi_hash_lock);
6197 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6198 rcu_sync_needed = true;
6199 hlist_del_rcu(&napi->napi_hash_node);
6201 spin_unlock(&napi_hash_lock);
6202 return rcu_sync_needed;
6204 EXPORT_SYMBOL_GPL(napi_hash_del);
6206 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6208 struct napi_struct *napi;
6210 napi = container_of(timer, struct napi_struct, timer);
6212 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6213 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6215 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6216 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6217 __napi_schedule_irqoff(napi);
6219 return HRTIMER_NORESTART;
6222 static void init_gro_hash(struct napi_struct *napi)
6226 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6227 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6228 napi->gro_hash[i].count = 0;
6230 napi->gro_bitmask = 0;
6233 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6234 int (*poll)(struct napi_struct *, int), int weight)
6236 INIT_LIST_HEAD(&napi->poll_list);
6237 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6238 napi->timer.function = napi_watchdog;
6239 init_gro_hash(napi);
6242 if (weight > NAPI_POLL_WEIGHT)
6243 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
6245 napi->weight = weight;
6246 list_add(&napi->dev_list, &dev->napi_list);
6248 #ifdef CONFIG_NETPOLL
6249 napi->poll_owner = -1;
6251 set_bit(NAPI_STATE_SCHED, &napi->state);
6252 napi_hash_add(napi);
6254 EXPORT_SYMBOL(netif_napi_add);
6256 void napi_disable(struct napi_struct *n)
6259 set_bit(NAPI_STATE_DISABLE, &n->state);
6261 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6263 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6266 hrtimer_cancel(&n->timer);
6268 clear_bit(NAPI_STATE_DISABLE, &n->state);
6270 EXPORT_SYMBOL(napi_disable);
6272 static void flush_gro_hash(struct napi_struct *napi)
6276 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6277 struct sk_buff *skb, *n;
6279 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6281 napi->gro_hash[i].count = 0;
6285 /* Must be called in process context */
6286 void netif_napi_del(struct napi_struct *napi)
6289 if (napi_hash_del(napi))
6291 list_del_init(&napi->dev_list);
6292 napi_free_frags(napi);
6294 flush_gro_hash(napi);
6295 napi->gro_bitmask = 0;
6297 EXPORT_SYMBOL(netif_napi_del);
6299 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6304 list_del_init(&n->poll_list);
6306 have = netpoll_poll_lock(n);
6310 /* This NAPI_STATE_SCHED test is for avoiding a race
6311 * with netpoll's poll_napi(). Only the entity which
6312 * obtains the lock and sees NAPI_STATE_SCHED set will
6313 * actually make the ->poll() call. Therefore we avoid
6314 * accidentally calling ->poll() when NAPI is not scheduled.
6317 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6318 work = n->poll(n, weight);
6319 trace_napi_poll(n, work, weight);
6322 WARN_ON_ONCE(work > weight);
6324 if (likely(work < weight))
6327 /* Drivers must not modify the NAPI state if they
6328 * consume the entire weight. In such cases this code
6329 * still "owns" the NAPI instance and therefore can
6330 * move the instance around on the list at-will.
6332 if (unlikely(napi_disable_pending(n))) {
6337 if (n->gro_bitmask) {
6338 /* flush too old packets
6339 * If HZ < 1000, flush all packets.
6341 napi_gro_flush(n, HZ >= 1000);
6344 /* Some drivers may have called napi_schedule
6345 * prior to exhausting their budget.
6347 if (unlikely(!list_empty(&n->poll_list))) {
6348 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6349 n->dev ? n->dev->name : "backlog");
6353 list_add_tail(&n->poll_list, repoll);
6356 netpoll_poll_unlock(have);
6361 static __latent_entropy void net_rx_action(struct softirq_action *h)
6363 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6364 unsigned long time_limit = jiffies +
6365 usecs_to_jiffies(netdev_budget_usecs);
6366 int budget = netdev_budget;
6370 local_irq_disable();
6371 list_splice_init(&sd->poll_list, &list);
6375 struct napi_struct *n;
6377 if (list_empty(&list)) {
6378 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6383 n = list_first_entry(&list, struct napi_struct, poll_list);
6384 budget -= napi_poll(n, &repoll);
6386 /* If softirq window is exhausted then punt.
6387 * Allow this to run for 2 jiffies since which will allow
6388 * an average latency of 1.5/HZ.
6390 if (unlikely(budget <= 0 ||
6391 time_after_eq(jiffies, time_limit))) {
6397 local_irq_disable();
6399 list_splice_tail_init(&sd->poll_list, &list);
6400 list_splice_tail(&repoll, &list);
6401 list_splice(&list, &sd->poll_list);
6402 if (!list_empty(&sd->poll_list))
6403 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6405 net_rps_action_and_irq_enable(sd);
6407 __kfree_skb_flush();
6410 struct netdev_adjacent {
6411 struct net_device *dev;
6413 /* upper master flag, there can only be one master device per list */
6416 /* counter for the number of times this device was added to us */
6419 /* private field for the users */
6422 struct list_head list;
6423 struct rcu_head rcu;
6426 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6427 struct list_head *adj_list)
6429 struct netdev_adjacent *adj;
6431 list_for_each_entry(adj, adj_list, list) {
6432 if (adj->dev == adj_dev)
6438 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6440 struct net_device *dev = data;
6442 return upper_dev == dev;
6446 * netdev_has_upper_dev - Check if device is linked to an upper device
6448 * @upper_dev: upper device to check
6450 * Find out if a device is linked to specified upper device and return true
6451 * in case it is. Note that this checks only immediate upper device,
6452 * not through a complete stack of devices. The caller must hold the RTNL lock.
6454 bool netdev_has_upper_dev(struct net_device *dev,
6455 struct net_device *upper_dev)
6459 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6462 EXPORT_SYMBOL(netdev_has_upper_dev);
6465 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6467 * @upper_dev: upper device to check
6469 * Find out if a device is linked to specified upper device and return true
6470 * in case it is. Note that this checks the entire upper device chain.
6471 * The caller must hold rcu lock.
6474 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6475 struct net_device *upper_dev)
6477 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6480 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6483 * netdev_has_any_upper_dev - Check if device is linked to some device
6486 * Find out if a device is linked to an upper device and return true in case
6487 * it is. The caller must hold the RTNL lock.
6489 bool netdev_has_any_upper_dev(struct net_device *dev)
6493 return !list_empty(&dev->adj_list.upper);
6495 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6498 * netdev_master_upper_dev_get - Get master upper device
6501 * Find a master upper device and return pointer to it or NULL in case
6502 * it's not there. The caller must hold the RTNL lock.
6504 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6506 struct netdev_adjacent *upper;
6510 if (list_empty(&dev->adj_list.upper))
6513 upper = list_first_entry(&dev->adj_list.upper,
6514 struct netdev_adjacent, list);
6515 if (likely(upper->master))
6519 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6522 * netdev_has_any_lower_dev - Check if device is linked to some device
6525 * Find out if a device is linked to a lower device and return true in case
6526 * it is. The caller must hold the RTNL lock.
6528 static bool netdev_has_any_lower_dev(struct net_device *dev)
6532 return !list_empty(&dev->adj_list.lower);
6535 void *netdev_adjacent_get_private(struct list_head *adj_list)
6537 struct netdev_adjacent *adj;
6539 adj = list_entry(adj_list, struct netdev_adjacent, list);
6541 return adj->private;
6543 EXPORT_SYMBOL(netdev_adjacent_get_private);
6546 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6548 * @iter: list_head ** of the current position
6550 * Gets the next device from the dev's upper list, starting from iter
6551 * position. The caller must hold RCU read lock.
6553 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6554 struct list_head **iter)
6556 struct netdev_adjacent *upper;
6558 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6560 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6562 if (&upper->list == &dev->adj_list.upper)
6565 *iter = &upper->list;
6569 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6571 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6572 struct list_head **iter)
6574 struct netdev_adjacent *upper;
6576 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6578 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6580 if (&upper->list == &dev->adj_list.upper)
6583 *iter = &upper->list;
6588 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6589 int (*fn)(struct net_device *dev,
6593 struct net_device *udev;
6594 struct list_head *iter;
6597 for (iter = &dev->adj_list.upper,
6598 udev = netdev_next_upper_dev_rcu(dev, &iter);
6600 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6601 /* first is the upper device itself */
6602 ret = fn(udev, data);
6606 /* then look at all of its upper devices */
6607 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6614 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6617 * netdev_lower_get_next_private - Get the next ->private from the
6618 * lower neighbour list
6620 * @iter: list_head ** of the current position
6622 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6623 * list, starting from iter position. The caller must hold either hold the
6624 * RTNL lock or its own locking that guarantees that the neighbour lower
6625 * list will remain unchanged.
6627 void *netdev_lower_get_next_private(struct net_device *dev,
6628 struct list_head **iter)
6630 struct netdev_adjacent *lower;
6632 lower = list_entry(*iter, struct netdev_adjacent, list);
6634 if (&lower->list == &dev->adj_list.lower)
6637 *iter = lower->list.next;
6639 return lower->private;
6641 EXPORT_SYMBOL(netdev_lower_get_next_private);
6644 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6645 * lower neighbour list, RCU
6648 * @iter: list_head ** of the current position
6650 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6651 * list, starting from iter position. The caller must hold RCU read lock.
6653 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6654 struct list_head **iter)
6656 struct netdev_adjacent *lower;
6658 WARN_ON_ONCE(!rcu_read_lock_held());
6660 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6662 if (&lower->list == &dev->adj_list.lower)
6665 *iter = &lower->list;
6667 return lower->private;
6669 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6672 * netdev_lower_get_next - Get the next device from the lower neighbour
6675 * @iter: list_head ** of the current position
6677 * Gets the next netdev_adjacent from the dev's lower neighbour
6678 * list, starting from iter position. The caller must hold RTNL lock or
6679 * its own locking that guarantees that the neighbour lower
6680 * list will remain unchanged.
6682 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6684 struct netdev_adjacent *lower;
6686 lower = list_entry(*iter, struct netdev_adjacent, list);
6688 if (&lower->list == &dev->adj_list.lower)
6691 *iter = lower->list.next;
6695 EXPORT_SYMBOL(netdev_lower_get_next);
6697 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6698 struct list_head **iter)
6700 struct netdev_adjacent *lower;
6702 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6704 if (&lower->list == &dev->adj_list.lower)
6707 *iter = &lower->list;
6712 int netdev_walk_all_lower_dev(struct net_device *dev,
6713 int (*fn)(struct net_device *dev,
6717 struct net_device *ldev;
6718 struct list_head *iter;
6721 for (iter = &dev->adj_list.lower,
6722 ldev = netdev_next_lower_dev(dev, &iter);
6724 ldev = netdev_next_lower_dev(dev, &iter)) {
6725 /* first is the lower device itself */
6726 ret = fn(ldev, data);
6730 /* then look at all of its lower devices */
6731 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6738 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6740 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6741 struct list_head **iter)
6743 struct netdev_adjacent *lower;
6745 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6746 if (&lower->list == &dev->adj_list.lower)
6749 *iter = &lower->list;
6754 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6755 int (*fn)(struct net_device *dev,
6759 struct net_device *ldev;
6760 struct list_head *iter;
6763 for (iter = &dev->adj_list.lower,
6764 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6766 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6767 /* first is the lower device itself */
6768 ret = fn(ldev, data);
6772 /* then look at all of its lower devices */
6773 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6780 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6783 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6784 * lower neighbour list, RCU
6788 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6789 * list. The caller must hold RCU read lock.
6791 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6793 struct netdev_adjacent *lower;
6795 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6796 struct netdev_adjacent, list);
6798 return lower->private;
6801 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6804 * netdev_master_upper_dev_get_rcu - Get master upper device
6807 * Find a master upper device and return pointer to it or NULL in case
6808 * it's not there. The caller must hold the RCU read lock.
6810 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6812 struct netdev_adjacent *upper;
6814 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6815 struct netdev_adjacent, list);
6816 if (upper && likely(upper->master))
6820 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6822 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6823 struct net_device *adj_dev,
6824 struct list_head *dev_list)
6826 char linkname[IFNAMSIZ+7];
6828 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6829 "upper_%s" : "lower_%s", adj_dev->name);
6830 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6833 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6835 struct list_head *dev_list)
6837 char linkname[IFNAMSIZ+7];
6839 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6840 "upper_%s" : "lower_%s", name);
6841 sysfs_remove_link(&(dev->dev.kobj), linkname);
6844 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6845 struct net_device *adj_dev,
6846 struct list_head *dev_list)
6848 return (dev_list == &dev->adj_list.upper ||
6849 dev_list == &dev->adj_list.lower) &&
6850 net_eq(dev_net(dev), dev_net(adj_dev));
6853 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6854 struct net_device *adj_dev,
6855 struct list_head *dev_list,
6856 void *private, bool master)
6858 struct netdev_adjacent *adj;
6861 adj = __netdev_find_adj(adj_dev, dev_list);
6865 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6866 dev->name, adj_dev->name, adj->ref_nr);
6871 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6876 adj->master = master;
6878 adj->private = private;
6881 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6882 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6884 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6885 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6890 /* Ensure that master link is always the first item in list. */
6892 ret = sysfs_create_link(&(dev->dev.kobj),
6893 &(adj_dev->dev.kobj), "master");
6895 goto remove_symlinks;
6897 list_add_rcu(&adj->list, dev_list);
6899 list_add_tail_rcu(&adj->list, dev_list);
6905 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6906 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6914 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6915 struct net_device *adj_dev,
6917 struct list_head *dev_list)
6919 struct netdev_adjacent *adj;
6921 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6922 dev->name, adj_dev->name, ref_nr);
6924 adj = __netdev_find_adj(adj_dev, dev_list);
6927 pr_err("Adjacency does not exist for device %s from %s\n",
6928 dev->name, adj_dev->name);
6933 if (adj->ref_nr > ref_nr) {
6934 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6935 dev->name, adj_dev->name, ref_nr,
6936 adj->ref_nr - ref_nr);
6937 adj->ref_nr -= ref_nr;
6942 sysfs_remove_link(&(dev->dev.kobj), "master");
6944 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6945 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6947 list_del_rcu(&adj->list);
6948 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6949 adj_dev->name, dev->name, adj_dev->name);
6951 kfree_rcu(adj, rcu);
6954 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6955 struct net_device *upper_dev,
6956 struct list_head *up_list,
6957 struct list_head *down_list,
6958 void *private, bool master)
6962 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6967 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6970 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6977 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6978 struct net_device *upper_dev,
6980 struct list_head *up_list,
6981 struct list_head *down_list)
6983 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6984 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6987 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6988 struct net_device *upper_dev,
6989 void *private, bool master)
6991 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6992 &dev->adj_list.upper,
6993 &upper_dev->adj_list.lower,
6997 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6998 struct net_device *upper_dev)
7000 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7001 &dev->adj_list.upper,
7002 &upper_dev->adj_list.lower);
7005 static int __netdev_upper_dev_link(struct net_device *dev,
7006 struct net_device *upper_dev, bool master,
7007 void *upper_priv, void *upper_info,
7008 struct netlink_ext_ack *extack)
7010 struct netdev_notifier_changeupper_info changeupper_info = {
7015 .upper_dev = upper_dev,
7018 .upper_info = upper_info,
7020 struct net_device *master_dev;
7025 if (dev == upper_dev)
7028 /* To prevent loops, check if dev is not upper device to upper_dev. */
7029 if (netdev_has_upper_dev(upper_dev, dev))
7033 if (netdev_has_upper_dev(dev, upper_dev))
7036 master_dev = netdev_master_upper_dev_get(dev);
7038 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7041 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7042 &changeupper_info.info);
7043 ret = notifier_to_errno(ret);
7047 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7052 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7053 &changeupper_info.info);
7054 ret = notifier_to_errno(ret);
7061 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7067 * netdev_upper_dev_link - Add a link to the upper device
7069 * @upper_dev: new upper device
7070 * @extack: netlink extended ack
7072 * Adds a link to device which is upper to this one. The caller must hold
7073 * the RTNL lock. On a failure a negative errno code is returned.
7074 * On success the reference counts are adjusted and the function
7077 int netdev_upper_dev_link(struct net_device *dev,
7078 struct net_device *upper_dev,
7079 struct netlink_ext_ack *extack)
7081 return __netdev_upper_dev_link(dev, upper_dev, false,
7082 NULL, NULL, extack);
7084 EXPORT_SYMBOL(netdev_upper_dev_link);
7087 * netdev_master_upper_dev_link - Add a master link to the upper device
7089 * @upper_dev: new upper device
7090 * @upper_priv: upper device private
7091 * @upper_info: upper info to be passed down via notifier
7092 * @extack: netlink extended ack
7094 * Adds a link to device which is upper to this one. In this case, only
7095 * one master upper device can be linked, although other non-master devices
7096 * might be linked as well. The caller must hold the RTNL lock.
7097 * On a failure a negative errno code is returned. On success the reference
7098 * counts are adjusted and the function returns zero.
7100 int netdev_master_upper_dev_link(struct net_device *dev,
7101 struct net_device *upper_dev,
7102 void *upper_priv, void *upper_info,
7103 struct netlink_ext_ack *extack)
7105 return __netdev_upper_dev_link(dev, upper_dev, true,
7106 upper_priv, upper_info, extack);
7108 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7111 * netdev_upper_dev_unlink - Removes a link to upper device
7113 * @upper_dev: new upper device
7115 * Removes a link to device which is upper to this one. The caller must hold
7118 void netdev_upper_dev_unlink(struct net_device *dev,
7119 struct net_device *upper_dev)
7121 struct netdev_notifier_changeupper_info changeupper_info = {
7125 .upper_dev = upper_dev,
7131 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7133 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7134 &changeupper_info.info);
7136 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7138 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7139 &changeupper_info.info);
7141 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7144 * netdev_bonding_info_change - Dispatch event about slave change
7146 * @bonding_info: info to dispatch
7148 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7149 * The caller must hold the RTNL lock.
7151 void netdev_bonding_info_change(struct net_device *dev,
7152 struct netdev_bonding_info *bonding_info)
7154 struct netdev_notifier_bonding_info info = {
7158 memcpy(&info.bonding_info, bonding_info,
7159 sizeof(struct netdev_bonding_info));
7160 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7163 EXPORT_SYMBOL(netdev_bonding_info_change);
7165 static void netdev_adjacent_add_links(struct net_device *dev)
7167 struct netdev_adjacent *iter;
7169 struct net *net = dev_net(dev);
7171 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7172 if (!net_eq(net, dev_net(iter->dev)))
7174 netdev_adjacent_sysfs_add(iter->dev, dev,
7175 &iter->dev->adj_list.lower);
7176 netdev_adjacent_sysfs_add(dev, iter->dev,
7177 &dev->adj_list.upper);
7180 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7181 if (!net_eq(net, dev_net(iter->dev)))
7183 netdev_adjacent_sysfs_add(iter->dev, dev,
7184 &iter->dev->adj_list.upper);
7185 netdev_adjacent_sysfs_add(dev, iter->dev,
7186 &dev->adj_list.lower);
7190 static void netdev_adjacent_del_links(struct net_device *dev)
7192 struct netdev_adjacent *iter;
7194 struct net *net = dev_net(dev);
7196 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7197 if (!net_eq(net, dev_net(iter->dev)))
7199 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7200 &iter->dev->adj_list.lower);
7201 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7202 &dev->adj_list.upper);
7205 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7206 if (!net_eq(net, dev_net(iter->dev)))
7208 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7209 &iter->dev->adj_list.upper);
7210 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7211 &dev->adj_list.lower);
7215 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7217 struct netdev_adjacent *iter;
7219 struct net *net = dev_net(dev);
7221 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7222 if (!net_eq(net, dev_net(iter->dev)))
7224 netdev_adjacent_sysfs_del(iter->dev, oldname,
7225 &iter->dev->adj_list.lower);
7226 netdev_adjacent_sysfs_add(iter->dev, dev,
7227 &iter->dev->adj_list.lower);
7230 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7231 if (!net_eq(net, dev_net(iter->dev)))
7233 netdev_adjacent_sysfs_del(iter->dev, oldname,
7234 &iter->dev->adj_list.upper);
7235 netdev_adjacent_sysfs_add(iter->dev, dev,
7236 &iter->dev->adj_list.upper);
7240 void *netdev_lower_dev_get_private(struct net_device *dev,
7241 struct net_device *lower_dev)
7243 struct netdev_adjacent *lower;
7247 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7251 return lower->private;
7253 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7256 int dev_get_nest_level(struct net_device *dev)
7258 struct net_device *lower = NULL;
7259 struct list_head *iter;
7265 netdev_for_each_lower_dev(dev, lower, iter) {
7266 nest = dev_get_nest_level(lower);
7267 if (max_nest < nest)
7271 return max_nest + 1;
7273 EXPORT_SYMBOL(dev_get_nest_level);
7276 * netdev_lower_change - Dispatch event about lower device state change
7277 * @lower_dev: device
7278 * @lower_state_info: state to dispatch
7280 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7281 * The caller must hold the RTNL lock.
7283 void netdev_lower_state_changed(struct net_device *lower_dev,
7284 void *lower_state_info)
7286 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7287 .info.dev = lower_dev,
7291 changelowerstate_info.lower_state_info = lower_state_info;
7292 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7293 &changelowerstate_info.info);
7295 EXPORT_SYMBOL(netdev_lower_state_changed);
7297 static void dev_change_rx_flags(struct net_device *dev, int flags)
7299 const struct net_device_ops *ops = dev->netdev_ops;
7301 if (ops->ndo_change_rx_flags)
7302 ops->ndo_change_rx_flags(dev, flags);
7305 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7307 unsigned int old_flags = dev->flags;
7313 dev->flags |= IFF_PROMISC;
7314 dev->promiscuity += inc;
7315 if (dev->promiscuity == 0) {
7318 * If inc causes overflow, untouch promisc and return error.
7321 dev->flags &= ~IFF_PROMISC;
7323 dev->promiscuity -= inc;
7324 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7329 if (dev->flags != old_flags) {
7330 pr_info("device %s %s promiscuous mode\n",
7332 dev->flags & IFF_PROMISC ? "entered" : "left");
7333 if (audit_enabled) {
7334 current_uid_gid(&uid, &gid);
7335 audit_log(audit_context(), GFP_ATOMIC,
7336 AUDIT_ANOM_PROMISCUOUS,
7337 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7338 dev->name, (dev->flags & IFF_PROMISC),
7339 (old_flags & IFF_PROMISC),
7340 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7341 from_kuid(&init_user_ns, uid),
7342 from_kgid(&init_user_ns, gid),
7343 audit_get_sessionid(current));
7346 dev_change_rx_flags(dev, IFF_PROMISC);
7349 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7354 * dev_set_promiscuity - update promiscuity count on a device
7358 * Add or remove promiscuity from a device. While the count in the device
7359 * remains above zero the interface remains promiscuous. Once it hits zero
7360 * the device reverts back to normal filtering operation. A negative inc
7361 * value is used to drop promiscuity on the device.
7362 * Return 0 if successful or a negative errno code on error.
7364 int dev_set_promiscuity(struct net_device *dev, int inc)
7366 unsigned int old_flags = dev->flags;
7369 err = __dev_set_promiscuity(dev, inc, true);
7372 if (dev->flags != old_flags)
7373 dev_set_rx_mode(dev);
7376 EXPORT_SYMBOL(dev_set_promiscuity);
7378 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7380 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7384 dev->flags |= IFF_ALLMULTI;
7385 dev->allmulti += inc;
7386 if (dev->allmulti == 0) {
7389 * If inc causes overflow, untouch allmulti and return error.
7392 dev->flags &= ~IFF_ALLMULTI;
7394 dev->allmulti -= inc;
7395 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7400 if (dev->flags ^ old_flags) {
7401 dev_change_rx_flags(dev, IFF_ALLMULTI);
7402 dev_set_rx_mode(dev);
7404 __dev_notify_flags(dev, old_flags,
7405 dev->gflags ^ old_gflags);
7411 * dev_set_allmulti - update allmulti count on a device
7415 * Add or remove reception of all multicast frames to a device. While the
7416 * count in the device remains above zero the interface remains listening
7417 * to all interfaces. Once it hits zero the device reverts back to normal
7418 * filtering operation. A negative @inc value is used to drop the counter
7419 * when releasing a resource needing all multicasts.
7420 * Return 0 if successful or a negative errno code on error.
7423 int dev_set_allmulti(struct net_device *dev, int inc)
7425 return __dev_set_allmulti(dev, inc, true);
7427 EXPORT_SYMBOL(dev_set_allmulti);
7430 * Upload unicast and multicast address lists to device and
7431 * configure RX filtering. When the device doesn't support unicast
7432 * filtering it is put in promiscuous mode while unicast addresses
7435 void __dev_set_rx_mode(struct net_device *dev)
7437 const struct net_device_ops *ops = dev->netdev_ops;
7439 /* dev_open will call this function so the list will stay sane. */
7440 if (!(dev->flags&IFF_UP))
7443 if (!netif_device_present(dev))
7446 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7447 /* Unicast addresses changes may only happen under the rtnl,
7448 * therefore calling __dev_set_promiscuity here is safe.
7450 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7451 __dev_set_promiscuity(dev, 1, false);
7452 dev->uc_promisc = true;
7453 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7454 __dev_set_promiscuity(dev, -1, false);
7455 dev->uc_promisc = false;
7459 if (ops->ndo_set_rx_mode)
7460 ops->ndo_set_rx_mode(dev);
7463 void dev_set_rx_mode(struct net_device *dev)
7465 netif_addr_lock_bh(dev);
7466 __dev_set_rx_mode(dev);
7467 netif_addr_unlock_bh(dev);
7471 * dev_get_flags - get flags reported to userspace
7474 * Get the combination of flag bits exported through APIs to userspace.
7476 unsigned int dev_get_flags(const struct net_device *dev)
7480 flags = (dev->flags & ~(IFF_PROMISC |
7485 (dev->gflags & (IFF_PROMISC |
7488 if (netif_running(dev)) {
7489 if (netif_oper_up(dev))
7490 flags |= IFF_RUNNING;
7491 if (netif_carrier_ok(dev))
7492 flags |= IFF_LOWER_UP;
7493 if (netif_dormant(dev))
7494 flags |= IFF_DORMANT;
7499 EXPORT_SYMBOL(dev_get_flags);
7501 int __dev_change_flags(struct net_device *dev, unsigned int flags)
7503 unsigned int old_flags = dev->flags;
7509 * Set the flags on our device.
7512 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7513 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7515 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7519 * Load in the correct multicast list now the flags have changed.
7522 if ((old_flags ^ flags) & IFF_MULTICAST)
7523 dev_change_rx_flags(dev, IFF_MULTICAST);
7525 dev_set_rx_mode(dev);
7528 * Have we downed the interface. We handle IFF_UP ourselves
7529 * according to user attempts to set it, rather than blindly
7534 if ((old_flags ^ flags) & IFF_UP) {
7535 if (old_flags & IFF_UP)
7538 ret = __dev_open(dev);
7541 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7542 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7543 unsigned int old_flags = dev->flags;
7545 dev->gflags ^= IFF_PROMISC;
7547 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7548 if (dev->flags != old_flags)
7549 dev_set_rx_mode(dev);
7552 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7553 * is important. Some (broken) drivers set IFF_PROMISC, when
7554 * IFF_ALLMULTI is requested not asking us and not reporting.
7556 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7557 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7559 dev->gflags ^= IFF_ALLMULTI;
7560 __dev_set_allmulti(dev, inc, false);
7566 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7567 unsigned int gchanges)
7569 unsigned int changes = dev->flags ^ old_flags;
7572 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7574 if (changes & IFF_UP) {
7575 if (dev->flags & IFF_UP)
7576 call_netdevice_notifiers(NETDEV_UP, dev);
7578 call_netdevice_notifiers(NETDEV_DOWN, dev);
7581 if (dev->flags & IFF_UP &&
7582 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7583 struct netdev_notifier_change_info change_info = {
7587 .flags_changed = changes,
7590 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7595 * dev_change_flags - change device settings
7597 * @flags: device state flags
7598 * @extack: netlink extended ack
7600 * Change settings on device based state flags. The flags are
7601 * in the userspace exported format.
7603 int dev_change_flags(struct net_device *dev, unsigned int flags,
7604 struct netlink_ext_ack *extack)
7607 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7609 ret = __dev_change_flags(dev, flags);
7613 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7614 __dev_notify_flags(dev, old_flags, changes);
7617 EXPORT_SYMBOL(dev_change_flags);
7619 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7621 const struct net_device_ops *ops = dev->netdev_ops;
7623 if (ops->ndo_change_mtu)
7624 return ops->ndo_change_mtu(dev, new_mtu);
7629 EXPORT_SYMBOL(__dev_set_mtu);
7632 * dev_set_mtu_ext - Change maximum transfer unit
7634 * @new_mtu: new transfer unit
7635 * @extack: netlink extended ack
7637 * Change the maximum transfer size of the network device.
7639 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7640 struct netlink_ext_ack *extack)
7644 if (new_mtu == dev->mtu)
7647 /* MTU must be positive, and in range */
7648 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7649 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7653 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7654 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7658 if (!netif_device_present(dev))
7661 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7662 err = notifier_to_errno(err);
7666 orig_mtu = dev->mtu;
7667 err = __dev_set_mtu(dev, new_mtu);
7670 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7672 err = notifier_to_errno(err);
7674 /* setting mtu back and notifying everyone again,
7675 * so that they have a chance to revert changes.
7677 __dev_set_mtu(dev, orig_mtu);
7678 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7685 int dev_set_mtu(struct net_device *dev, int new_mtu)
7687 struct netlink_ext_ack extack;
7690 memset(&extack, 0, sizeof(extack));
7691 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7692 if (err && extack._msg)
7693 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7696 EXPORT_SYMBOL(dev_set_mtu);
7699 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7701 * @new_len: new tx queue length
7703 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7705 unsigned int orig_len = dev->tx_queue_len;
7708 if (new_len != (unsigned int)new_len)
7711 if (new_len != orig_len) {
7712 dev->tx_queue_len = new_len;
7713 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7714 res = notifier_to_errno(res);
7717 res = dev_qdisc_change_tx_queue_len(dev);
7725 netdev_err(dev, "refused to change device tx_queue_len\n");
7726 dev->tx_queue_len = orig_len;
7731 * dev_set_group - Change group this device belongs to
7733 * @new_group: group this device should belong to
7735 void dev_set_group(struct net_device *dev, int new_group)
7737 dev->group = new_group;
7739 EXPORT_SYMBOL(dev_set_group);
7742 * dev_set_mac_address - Change Media Access Control Address
7746 * Change the hardware (MAC) address of the device
7748 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
7750 const struct net_device_ops *ops = dev->netdev_ops;
7753 if (!ops->ndo_set_mac_address)
7755 if (sa->sa_family != dev->type)
7757 if (!netif_device_present(dev))
7759 err = ops->ndo_set_mac_address(dev, sa);
7762 dev->addr_assign_type = NET_ADDR_SET;
7763 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7764 add_device_randomness(dev->dev_addr, dev->addr_len);
7767 EXPORT_SYMBOL(dev_set_mac_address);
7770 * dev_change_carrier - Change device carrier
7772 * @new_carrier: new value
7774 * Change device carrier
7776 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7778 const struct net_device_ops *ops = dev->netdev_ops;
7780 if (!ops->ndo_change_carrier)
7782 if (!netif_device_present(dev))
7784 return ops->ndo_change_carrier(dev, new_carrier);
7786 EXPORT_SYMBOL(dev_change_carrier);
7789 * dev_get_phys_port_id - Get device physical port ID
7793 * Get device physical port ID
7795 int dev_get_phys_port_id(struct net_device *dev,
7796 struct netdev_phys_item_id *ppid)
7798 const struct net_device_ops *ops = dev->netdev_ops;
7800 if (!ops->ndo_get_phys_port_id)
7802 return ops->ndo_get_phys_port_id(dev, ppid);
7804 EXPORT_SYMBOL(dev_get_phys_port_id);
7807 * dev_get_phys_port_name - Get device physical port name
7810 * @len: limit of bytes to copy to name
7812 * Get device physical port name
7814 int dev_get_phys_port_name(struct net_device *dev,
7815 char *name, size_t len)
7817 const struct net_device_ops *ops = dev->netdev_ops;
7819 if (!ops->ndo_get_phys_port_name)
7821 return ops->ndo_get_phys_port_name(dev, name, len);
7823 EXPORT_SYMBOL(dev_get_phys_port_name);
7826 * dev_change_proto_down - update protocol port state information
7828 * @proto_down: new value
7830 * This info can be used by switch drivers to set the phys state of the
7833 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7835 const struct net_device_ops *ops = dev->netdev_ops;
7837 if (!ops->ndo_change_proto_down)
7839 if (!netif_device_present(dev))
7841 return ops->ndo_change_proto_down(dev, proto_down);
7843 EXPORT_SYMBOL(dev_change_proto_down);
7845 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
7846 enum bpf_netdev_command cmd)
7848 struct netdev_bpf xdp;
7853 memset(&xdp, 0, sizeof(xdp));
7856 /* Query must always succeed. */
7857 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
7862 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
7863 struct netlink_ext_ack *extack, u32 flags,
7864 struct bpf_prog *prog)
7866 struct netdev_bpf xdp;
7868 memset(&xdp, 0, sizeof(xdp));
7869 if (flags & XDP_FLAGS_HW_MODE)
7870 xdp.command = XDP_SETUP_PROG_HW;
7872 xdp.command = XDP_SETUP_PROG;
7873 xdp.extack = extack;
7877 return bpf_op(dev, &xdp);
7880 static void dev_xdp_uninstall(struct net_device *dev)
7882 struct netdev_bpf xdp;
7885 /* Remove generic XDP */
7886 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
7888 /* Remove from the driver */
7889 ndo_bpf = dev->netdev_ops->ndo_bpf;
7893 memset(&xdp, 0, sizeof(xdp));
7894 xdp.command = XDP_QUERY_PROG;
7895 WARN_ON(ndo_bpf(dev, &xdp));
7897 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
7900 /* Remove HW offload */
7901 memset(&xdp, 0, sizeof(xdp));
7902 xdp.command = XDP_QUERY_PROG_HW;
7903 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
7904 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
7909 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
7911 * @extack: netlink extended ack
7912 * @fd: new program fd or negative value to clear
7913 * @flags: xdp-related flags
7915 * Set or clear a bpf program for a device
7917 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
7920 const struct net_device_ops *ops = dev->netdev_ops;
7921 enum bpf_netdev_command query;
7922 struct bpf_prog *prog = NULL;
7923 bpf_op_t bpf_op, bpf_chk;
7928 query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
7930 bpf_op = bpf_chk = ops->ndo_bpf;
7931 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7933 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
7934 bpf_op = generic_xdp_install;
7935 if (bpf_op == bpf_chk)
7936 bpf_chk = generic_xdp_install;
7939 if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) ||
7940 __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW))
7942 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7943 __dev_xdp_query(dev, bpf_op, query))
7946 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
7947 bpf_op == ops->ndo_bpf);
7949 return PTR_ERR(prog);
7951 if (!(flags & XDP_FLAGS_HW_MODE) &&
7952 bpf_prog_is_dev_bound(prog->aux)) {
7953 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
7959 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
7960 if (err < 0 && prog)
7967 * dev_new_index - allocate an ifindex
7968 * @net: the applicable net namespace
7970 * Returns a suitable unique value for a new device interface
7971 * number. The caller must hold the rtnl semaphore or the
7972 * dev_base_lock to be sure it remains unique.
7974 static int dev_new_index(struct net *net)
7976 int ifindex = net->ifindex;
7981 if (!__dev_get_by_index(net, ifindex))
7982 return net->ifindex = ifindex;
7986 /* Delayed registration/unregisteration */
7987 static LIST_HEAD(net_todo_list);
7988 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7990 static void net_set_todo(struct net_device *dev)
7992 list_add_tail(&dev->todo_list, &net_todo_list);
7993 dev_net(dev)->dev_unreg_count++;
7996 static void rollback_registered_many(struct list_head *head)
7998 struct net_device *dev, *tmp;
7999 LIST_HEAD(close_head);
8001 BUG_ON(dev_boot_phase);
8004 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8005 /* Some devices call without registering
8006 * for initialization unwind. Remove those
8007 * devices and proceed with the remaining.
8009 if (dev->reg_state == NETREG_UNINITIALIZED) {
8010 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8014 list_del(&dev->unreg_list);
8017 dev->dismantle = true;
8018 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8021 /* If device is running, close it first. */
8022 list_for_each_entry(dev, head, unreg_list)
8023 list_add_tail(&dev->close_list, &close_head);
8024 dev_close_many(&close_head, true);
8026 list_for_each_entry(dev, head, unreg_list) {
8027 /* And unlink it from device chain. */
8028 unlist_netdevice(dev);
8030 dev->reg_state = NETREG_UNREGISTERING;
8032 flush_all_backlogs();
8036 list_for_each_entry(dev, head, unreg_list) {
8037 struct sk_buff *skb = NULL;
8039 /* Shutdown queueing discipline. */
8042 dev_xdp_uninstall(dev);
8044 /* Notify protocols, that we are about to destroy
8045 * this device. They should clean all the things.
8047 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8049 if (!dev->rtnl_link_ops ||
8050 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8051 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8052 GFP_KERNEL, NULL, 0);
8055 * Flush the unicast and multicast chains
8060 if (dev->netdev_ops->ndo_uninit)
8061 dev->netdev_ops->ndo_uninit(dev);
8064 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8066 /* Notifier chain MUST detach us all upper devices. */
8067 WARN_ON(netdev_has_any_upper_dev(dev));
8068 WARN_ON(netdev_has_any_lower_dev(dev));
8070 /* Remove entries from kobject tree */
8071 netdev_unregister_kobject(dev);
8073 /* Remove XPS queueing entries */
8074 netif_reset_xps_queues_gt(dev, 0);
8080 list_for_each_entry(dev, head, unreg_list)
8084 static void rollback_registered(struct net_device *dev)
8088 list_add(&dev->unreg_list, &single);
8089 rollback_registered_many(&single);
8093 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8094 struct net_device *upper, netdev_features_t features)
8096 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8097 netdev_features_t feature;
8100 for_each_netdev_feature(&upper_disables, feature_bit) {
8101 feature = __NETIF_F_BIT(feature_bit);
8102 if (!(upper->wanted_features & feature)
8103 && (features & feature)) {
8104 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8105 &feature, upper->name);
8106 features &= ~feature;
8113 static void netdev_sync_lower_features(struct net_device *upper,
8114 struct net_device *lower, netdev_features_t features)
8116 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8117 netdev_features_t feature;
8120 for_each_netdev_feature(&upper_disables, feature_bit) {
8121 feature = __NETIF_F_BIT(feature_bit);
8122 if (!(features & feature) && (lower->features & feature)) {
8123 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8124 &feature, lower->name);
8125 lower->wanted_features &= ~feature;
8126 netdev_update_features(lower);
8128 if (unlikely(lower->features & feature))
8129 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8130 &feature, lower->name);
8135 static netdev_features_t netdev_fix_features(struct net_device *dev,
8136 netdev_features_t features)
8138 /* Fix illegal checksum combinations */
8139 if ((features & NETIF_F_HW_CSUM) &&
8140 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8141 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8142 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8145 /* TSO requires that SG is present as well. */
8146 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8147 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8148 features &= ~NETIF_F_ALL_TSO;
8151 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8152 !(features & NETIF_F_IP_CSUM)) {
8153 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8154 features &= ~NETIF_F_TSO;
8155 features &= ~NETIF_F_TSO_ECN;
8158 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8159 !(features & NETIF_F_IPV6_CSUM)) {
8160 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8161 features &= ~NETIF_F_TSO6;
8164 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8165 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8166 features &= ~NETIF_F_TSO_MANGLEID;
8168 /* TSO ECN requires that TSO is present as well. */
8169 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8170 features &= ~NETIF_F_TSO_ECN;
8172 /* Software GSO depends on SG. */
8173 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8174 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8175 features &= ~NETIF_F_GSO;
8178 /* GSO partial features require GSO partial be set */
8179 if ((features & dev->gso_partial_features) &&
8180 !(features & NETIF_F_GSO_PARTIAL)) {
8182 "Dropping partially supported GSO features since no GSO partial.\n");
8183 features &= ~dev->gso_partial_features;
8186 if (!(features & NETIF_F_RXCSUM)) {
8187 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8188 * successfully merged by hardware must also have the
8189 * checksum verified by hardware. If the user does not
8190 * want to enable RXCSUM, logically, we should disable GRO_HW.
8192 if (features & NETIF_F_GRO_HW) {
8193 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8194 features &= ~NETIF_F_GRO_HW;
8198 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8199 if (features & NETIF_F_RXFCS) {
8200 if (features & NETIF_F_LRO) {
8201 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8202 features &= ~NETIF_F_LRO;
8205 if (features & NETIF_F_GRO_HW) {
8206 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8207 features &= ~NETIF_F_GRO_HW;
8214 int __netdev_update_features(struct net_device *dev)
8216 struct net_device *upper, *lower;
8217 netdev_features_t features;
8218 struct list_head *iter;
8223 features = netdev_get_wanted_features(dev);
8225 if (dev->netdev_ops->ndo_fix_features)
8226 features = dev->netdev_ops->ndo_fix_features(dev, features);
8228 /* driver might be less strict about feature dependencies */
8229 features = netdev_fix_features(dev, features);
8231 /* some features can't be enabled if they're off an an upper device */
8232 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8233 features = netdev_sync_upper_features(dev, upper, features);
8235 if (dev->features == features)
8238 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8239 &dev->features, &features);
8241 if (dev->netdev_ops->ndo_set_features)
8242 err = dev->netdev_ops->ndo_set_features(dev, features);
8246 if (unlikely(err < 0)) {
8248 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8249 err, &features, &dev->features);
8250 /* return non-0 since some features might have changed and
8251 * it's better to fire a spurious notification than miss it
8257 /* some features must be disabled on lower devices when disabled
8258 * on an upper device (think: bonding master or bridge)
8260 netdev_for_each_lower_dev(dev, lower, iter)
8261 netdev_sync_lower_features(dev, lower, features);
8264 netdev_features_t diff = features ^ dev->features;
8266 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8267 /* udp_tunnel_{get,drop}_rx_info both need
8268 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8269 * device, or they won't do anything.
8270 * Thus we need to update dev->features
8271 * *before* calling udp_tunnel_get_rx_info,
8272 * but *after* calling udp_tunnel_drop_rx_info.
8274 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8275 dev->features = features;
8276 udp_tunnel_get_rx_info(dev);
8278 udp_tunnel_drop_rx_info(dev);
8282 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8283 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8284 dev->features = features;
8285 err |= vlan_get_rx_ctag_filter_info(dev);
8287 vlan_drop_rx_ctag_filter_info(dev);
8291 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8292 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8293 dev->features = features;
8294 err |= vlan_get_rx_stag_filter_info(dev);
8296 vlan_drop_rx_stag_filter_info(dev);
8300 dev->features = features;
8303 return err < 0 ? 0 : 1;
8307 * netdev_update_features - recalculate device features
8308 * @dev: the device to check
8310 * Recalculate dev->features set and send notifications if it
8311 * has changed. Should be called after driver or hardware dependent
8312 * conditions might have changed that influence the features.
8314 void netdev_update_features(struct net_device *dev)
8316 if (__netdev_update_features(dev))
8317 netdev_features_change(dev);
8319 EXPORT_SYMBOL(netdev_update_features);
8322 * netdev_change_features - recalculate device features
8323 * @dev: the device to check
8325 * Recalculate dev->features set and send notifications even
8326 * if they have not changed. Should be called instead of
8327 * netdev_update_features() if also dev->vlan_features might
8328 * have changed to allow the changes to be propagated to stacked
8331 void netdev_change_features(struct net_device *dev)
8333 __netdev_update_features(dev);
8334 netdev_features_change(dev);
8336 EXPORT_SYMBOL(netdev_change_features);
8339 * netif_stacked_transfer_operstate - transfer operstate
8340 * @rootdev: the root or lower level device to transfer state from
8341 * @dev: the device to transfer operstate to
8343 * Transfer operational state from root to device. This is normally
8344 * called when a stacking relationship exists between the root
8345 * device and the device(a leaf device).
8347 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8348 struct net_device *dev)
8350 if (rootdev->operstate == IF_OPER_DORMANT)
8351 netif_dormant_on(dev);
8353 netif_dormant_off(dev);
8355 if (netif_carrier_ok(rootdev))
8356 netif_carrier_on(dev);
8358 netif_carrier_off(dev);
8360 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8362 static int netif_alloc_rx_queues(struct net_device *dev)
8364 unsigned int i, count = dev->num_rx_queues;
8365 struct netdev_rx_queue *rx;
8366 size_t sz = count * sizeof(*rx);
8371 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8377 for (i = 0; i < count; i++) {
8380 /* XDP RX-queue setup */
8381 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8388 /* Rollback successful reg's and free other resources */
8390 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8396 static void netif_free_rx_queues(struct net_device *dev)
8398 unsigned int i, count = dev->num_rx_queues;
8400 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8404 for (i = 0; i < count; i++)
8405 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8410 static void netdev_init_one_queue(struct net_device *dev,
8411 struct netdev_queue *queue, void *_unused)
8413 /* Initialize queue lock */
8414 spin_lock_init(&queue->_xmit_lock);
8415 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8416 queue->xmit_lock_owner = -1;
8417 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8420 dql_init(&queue->dql, HZ);
8424 static void netif_free_tx_queues(struct net_device *dev)
8429 static int netif_alloc_netdev_queues(struct net_device *dev)
8431 unsigned int count = dev->num_tx_queues;
8432 struct netdev_queue *tx;
8433 size_t sz = count * sizeof(*tx);
8435 if (count < 1 || count > 0xffff)
8438 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8444 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
8445 spin_lock_init(&dev->tx_global_lock);
8450 void netif_tx_stop_all_queues(struct net_device *dev)
8454 for (i = 0; i < dev->num_tx_queues; i++) {
8455 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
8457 netif_tx_stop_queue(txq);
8460 EXPORT_SYMBOL(netif_tx_stop_all_queues);
8463 * register_netdevice - register a network device
8464 * @dev: device to register
8466 * Take a completed network device structure and add it to the kernel
8467 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8468 * chain. 0 is returned on success. A negative errno code is returned
8469 * on a failure to set up the device, or if the name is a duplicate.
8471 * Callers must hold the rtnl semaphore. You may want
8472 * register_netdev() instead of this.
8475 * The locking appears insufficient to guarantee two parallel registers
8476 * will not get the same name.
8479 int register_netdevice(struct net_device *dev)
8482 struct net *net = dev_net(dev);
8484 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
8485 NETDEV_FEATURE_COUNT);
8486 BUG_ON(dev_boot_phase);
8491 /* When net_device's are persistent, this will be fatal. */
8492 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8495 spin_lock_init(&dev->addr_list_lock);
8496 netdev_set_addr_lockdep_class(dev);
8498 ret = dev_get_valid_name(net, dev, dev->name);
8502 /* Init, if this function is available */
8503 if (dev->netdev_ops->ndo_init) {
8504 ret = dev->netdev_ops->ndo_init(dev);
8512 if (((dev->hw_features | dev->features) &
8513 NETIF_F_HW_VLAN_CTAG_FILTER) &&
8514 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
8515 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
8516 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
8523 dev->ifindex = dev_new_index(net);
8524 else if (__dev_get_by_index(net, dev->ifindex))
8527 /* Transfer changeable features to wanted_features and enable
8528 * software offloads (GSO and GRO).
8530 dev->hw_features |= NETIF_F_SOFT_FEATURES;
8531 dev->features |= NETIF_F_SOFT_FEATURES;
8533 if (dev->netdev_ops->ndo_udp_tunnel_add) {
8534 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8535 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8538 dev->wanted_features = dev->features & dev->hw_features;
8540 if (!(dev->flags & IFF_LOOPBACK))
8541 dev->hw_features |= NETIF_F_NOCACHE_COPY;
8543 /* If IPv4 TCP segmentation offload is supported we should also
8544 * allow the device to enable segmenting the frame with the option
8545 * of ignoring a static IP ID value. This doesn't enable the
8546 * feature itself but allows the user to enable it later.
8548 if (dev->hw_features & NETIF_F_TSO)
8549 dev->hw_features |= NETIF_F_TSO_MANGLEID;
8550 if (dev->vlan_features & NETIF_F_TSO)
8551 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
8552 if (dev->mpls_features & NETIF_F_TSO)
8553 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
8554 if (dev->hw_enc_features & NETIF_F_TSO)
8555 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
8557 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8559 dev->vlan_features |= NETIF_F_HIGHDMA;
8561 /* Make NETIF_F_SG inheritable to tunnel devices.
8563 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
8565 /* Make NETIF_F_SG inheritable to MPLS.
8567 dev->mpls_features |= NETIF_F_SG;
8569 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
8570 ret = notifier_to_errno(ret);
8574 ret = netdev_register_kobject(dev);
8577 dev->reg_state = NETREG_REGISTERED;
8579 __netdev_update_features(dev);
8582 * Default initial state at registry is that the
8583 * device is present.
8586 set_bit(__LINK_STATE_PRESENT, &dev->state);
8588 linkwatch_init_dev(dev);
8590 dev_init_scheduler(dev);
8592 list_netdevice(dev);
8593 add_device_randomness(dev->dev_addr, dev->addr_len);
8595 /* If the device has permanent device address, driver should
8596 * set dev_addr and also addr_assign_type should be set to
8597 * NET_ADDR_PERM (default value).
8599 if (dev->addr_assign_type == NET_ADDR_PERM)
8600 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8602 /* Notify protocols, that a new device appeared. */
8603 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8604 ret = notifier_to_errno(ret);
8606 rollback_registered(dev);
8607 dev->reg_state = NETREG_UNREGISTERED;
8610 * Prevent userspace races by waiting until the network
8611 * device is fully setup before sending notifications.
8613 if (!dev->rtnl_link_ops ||
8614 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8615 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8621 if (dev->netdev_ops->ndo_uninit)
8622 dev->netdev_ops->ndo_uninit(dev);
8623 if (dev->priv_destructor)
8624 dev->priv_destructor(dev);
8627 EXPORT_SYMBOL(register_netdevice);
8630 * init_dummy_netdev - init a dummy network device for NAPI
8631 * @dev: device to init
8633 * This takes a network device structure and initialize the minimum
8634 * amount of fields so it can be used to schedule NAPI polls without
8635 * registering a full blown interface. This is to be used by drivers
8636 * that need to tie several hardware interfaces to a single NAPI
8637 * poll scheduler due to HW limitations.
8639 int init_dummy_netdev(struct net_device *dev)
8641 /* Clear everything. Note we don't initialize spinlocks
8642 * are they aren't supposed to be taken by any of the
8643 * NAPI code and this dummy netdev is supposed to be
8644 * only ever used for NAPI polls
8646 memset(dev, 0, sizeof(struct net_device));
8648 /* make sure we BUG if trying to hit standard
8649 * register/unregister code path
8651 dev->reg_state = NETREG_DUMMY;
8653 /* NAPI wants this */
8654 INIT_LIST_HEAD(&dev->napi_list);
8656 /* a dummy interface is started by default */
8657 set_bit(__LINK_STATE_PRESENT, &dev->state);
8658 set_bit(__LINK_STATE_START, &dev->state);
8660 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8661 * because users of this 'device' dont need to change
8667 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8671 * register_netdev - register a network device
8672 * @dev: device to register
8674 * Take a completed network device structure and add it to the kernel
8675 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8676 * chain. 0 is returned on success. A negative errno code is returned
8677 * on a failure to set up the device, or if the name is a duplicate.
8679 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8680 * and expands the device name if you passed a format string to
8683 int register_netdev(struct net_device *dev)
8687 if (rtnl_lock_killable())
8689 err = register_netdevice(dev);
8693 EXPORT_SYMBOL(register_netdev);
8695 int netdev_refcnt_read(const struct net_device *dev)
8699 for_each_possible_cpu(i)
8700 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8703 EXPORT_SYMBOL(netdev_refcnt_read);
8706 * netdev_wait_allrefs - wait until all references are gone.
8707 * @dev: target net_device
8709 * This is called when unregistering network devices.
8711 * Any protocol or device that holds a reference should register
8712 * for netdevice notification, and cleanup and put back the
8713 * reference if they receive an UNREGISTER event.
8714 * We can get stuck here if buggy protocols don't correctly
8717 static void netdev_wait_allrefs(struct net_device *dev)
8719 unsigned long rebroadcast_time, warning_time;
8722 linkwatch_forget_dev(dev);
8724 rebroadcast_time = warning_time = jiffies;
8725 refcnt = netdev_refcnt_read(dev);
8727 while (refcnt != 0) {
8728 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8731 /* Rebroadcast unregister notification */
8732 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8738 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8740 /* We must not have linkwatch events
8741 * pending on unregister. If this
8742 * happens, we simply run the queue
8743 * unscheduled, resulting in a noop
8746 linkwatch_run_queue();
8751 rebroadcast_time = jiffies;
8756 refcnt = netdev_refcnt_read(dev);
8758 if (time_after(jiffies, warning_time + 10 * HZ)) {
8759 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8761 warning_time = jiffies;
8770 * register_netdevice(x1);
8771 * register_netdevice(x2);
8773 * unregister_netdevice(y1);
8774 * unregister_netdevice(y2);
8780 * We are invoked by rtnl_unlock().
8781 * This allows us to deal with problems:
8782 * 1) We can delete sysfs objects which invoke hotplug
8783 * without deadlocking with linkwatch via keventd.
8784 * 2) Since we run with the RTNL semaphore not held, we can sleep
8785 * safely in order to wait for the netdev refcnt to drop to zero.
8787 * We must not return until all unregister events added during
8788 * the interval the lock was held have been completed.
8790 void netdev_run_todo(void)
8792 struct list_head list;
8794 /* Snapshot list, allow later requests */
8795 list_replace_init(&net_todo_list, &list);
8800 /* Wait for rcu callbacks to finish before next phase */
8801 if (!list_empty(&list))
8804 while (!list_empty(&list)) {
8805 struct net_device *dev
8806 = list_first_entry(&list, struct net_device, todo_list);
8807 list_del(&dev->todo_list);
8809 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8810 pr_err("network todo '%s' but state %d\n",
8811 dev->name, dev->reg_state);
8816 dev->reg_state = NETREG_UNREGISTERED;
8818 netdev_wait_allrefs(dev);
8821 BUG_ON(netdev_refcnt_read(dev));
8822 BUG_ON(!list_empty(&dev->ptype_all));
8823 BUG_ON(!list_empty(&dev->ptype_specific));
8824 WARN_ON(rcu_access_pointer(dev->ip_ptr));
8825 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
8826 #if IS_ENABLED(CONFIG_DECNET)
8827 WARN_ON(dev->dn_ptr);
8829 if (dev->priv_destructor)
8830 dev->priv_destructor(dev);
8831 if (dev->needs_free_netdev)
8834 /* Report a network device has been unregistered */
8836 dev_net(dev)->dev_unreg_count--;
8838 wake_up(&netdev_unregistering_wq);
8840 /* Free network device */
8841 kobject_put(&dev->dev.kobj);
8845 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8846 * all the same fields in the same order as net_device_stats, with only
8847 * the type differing, but rtnl_link_stats64 may have additional fields
8848 * at the end for newer counters.
8850 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
8851 const struct net_device_stats *netdev_stats)
8853 #if BITS_PER_LONG == 64
8854 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
8855 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
8856 /* zero out counters that only exist in rtnl_link_stats64 */
8857 memset((char *)stats64 + sizeof(*netdev_stats), 0,
8858 sizeof(*stats64) - sizeof(*netdev_stats));
8860 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
8861 const unsigned long *src = (const unsigned long *)netdev_stats;
8862 u64 *dst = (u64 *)stats64;
8864 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
8865 for (i = 0; i < n; i++)
8867 /* zero out counters that only exist in rtnl_link_stats64 */
8868 memset((char *)stats64 + n * sizeof(u64), 0,
8869 sizeof(*stats64) - n * sizeof(u64));
8872 EXPORT_SYMBOL(netdev_stats_to_stats64);
8875 * dev_get_stats - get network device statistics
8876 * @dev: device to get statistics from
8877 * @storage: place to store stats
8879 * Get network statistics from device. Return @storage.
8880 * The device driver may provide its own method by setting
8881 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8882 * otherwise the internal statistics structure is used.
8884 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
8885 struct rtnl_link_stats64 *storage)
8887 const struct net_device_ops *ops = dev->netdev_ops;
8889 if (ops->ndo_get_stats64) {
8890 memset(storage, 0, sizeof(*storage));
8891 ops->ndo_get_stats64(dev, storage);
8892 } else if (ops->ndo_get_stats) {
8893 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
8895 netdev_stats_to_stats64(storage, &dev->stats);
8897 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
8898 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
8899 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
8902 EXPORT_SYMBOL(dev_get_stats);
8904 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
8906 struct netdev_queue *queue = dev_ingress_queue(dev);
8908 #ifdef CONFIG_NET_CLS_ACT
8911 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
8914 netdev_init_one_queue(dev, queue, NULL);
8915 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
8916 queue->qdisc_sleeping = &noop_qdisc;
8917 rcu_assign_pointer(dev->ingress_queue, queue);
8922 static const struct ethtool_ops default_ethtool_ops;
8924 void netdev_set_default_ethtool_ops(struct net_device *dev,
8925 const struct ethtool_ops *ops)
8927 if (dev->ethtool_ops == &default_ethtool_ops)
8928 dev->ethtool_ops = ops;
8930 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
8932 void netdev_freemem(struct net_device *dev)
8934 char *addr = (char *)dev - dev->padded;
8940 * alloc_netdev_mqs - allocate network device
8941 * @sizeof_priv: size of private data to allocate space for
8942 * @name: device name format string
8943 * @name_assign_type: origin of device name
8944 * @setup: callback to initialize device
8945 * @txqs: the number of TX subqueues to allocate
8946 * @rxqs: the number of RX subqueues to allocate
8948 * Allocates a struct net_device with private data area for driver use
8949 * and performs basic initialization. Also allocates subqueue structs
8950 * for each queue on the device.
8952 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
8953 unsigned char name_assign_type,
8954 void (*setup)(struct net_device *),
8955 unsigned int txqs, unsigned int rxqs)
8957 struct net_device *dev;
8958 unsigned int alloc_size;
8959 struct net_device *p;
8961 BUG_ON(strlen(name) >= sizeof(dev->name));
8964 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8969 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
8973 alloc_size = sizeof(struct net_device);
8975 /* ensure 32-byte alignment of private area */
8976 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
8977 alloc_size += sizeof_priv;
8979 /* ensure 32-byte alignment of whole construct */
8980 alloc_size += NETDEV_ALIGN - 1;
8982 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8986 dev = PTR_ALIGN(p, NETDEV_ALIGN);
8987 dev->padded = (char *)dev - (char *)p;
8989 dev->pcpu_refcnt = alloc_percpu(int);
8990 if (!dev->pcpu_refcnt)
8993 if (dev_addr_init(dev))
8999 dev_net_set(dev, &init_net);
9001 dev->gso_max_size = GSO_MAX_SIZE;
9002 dev->gso_max_segs = GSO_MAX_SEGS;
9004 INIT_LIST_HEAD(&dev->napi_list);
9005 INIT_LIST_HEAD(&dev->unreg_list);
9006 INIT_LIST_HEAD(&dev->close_list);
9007 INIT_LIST_HEAD(&dev->link_watch_list);
9008 INIT_LIST_HEAD(&dev->adj_list.upper);
9009 INIT_LIST_HEAD(&dev->adj_list.lower);
9010 INIT_LIST_HEAD(&dev->ptype_all);
9011 INIT_LIST_HEAD(&dev->ptype_specific);
9012 #ifdef CONFIG_NET_SCHED
9013 hash_init(dev->qdisc_hash);
9015 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9018 if (!dev->tx_queue_len) {
9019 dev->priv_flags |= IFF_NO_QUEUE;
9020 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9023 dev->num_tx_queues = txqs;
9024 dev->real_num_tx_queues = txqs;
9025 if (netif_alloc_netdev_queues(dev))
9028 dev->num_rx_queues = rxqs;
9029 dev->real_num_rx_queues = rxqs;
9030 if (netif_alloc_rx_queues(dev))
9033 strcpy(dev->name, name);
9034 dev->name_assign_type = name_assign_type;
9035 dev->group = INIT_NETDEV_GROUP;
9036 if (!dev->ethtool_ops)
9037 dev->ethtool_ops = &default_ethtool_ops;
9039 nf_hook_ingress_init(dev);
9048 free_percpu(dev->pcpu_refcnt);
9050 netdev_freemem(dev);
9053 EXPORT_SYMBOL(alloc_netdev_mqs);
9056 * free_netdev - free network device
9059 * This function does the last stage of destroying an allocated device
9060 * interface. The reference to the device object is released. If this
9061 * is the last reference then it will be freed.Must be called in process
9064 void free_netdev(struct net_device *dev)
9066 struct napi_struct *p, *n;
9069 netif_free_tx_queues(dev);
9070 netif_free_rx_queues(dev);
9072 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9074 /* Flush device addresses */
9075 dev_addr_flush(dev);
9077 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9080 free_percpu(dev->pcpu_refcnt);
9081 dev->pcpu_refcnt = NULL;
9083 /* Compatibility with error handling in drivers */
9084 if (dev->reg_state == NETREG_UNINITIALIZED) {
9085 netdev_freemem(dev);
9089 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9090 dev->reg_state = NETREG_RELEASED;
9092 /* will free via device release */
9093 put_device(&dev->dev);
9095 EXPORT_SYMBOL(free_netdev);
9098 * synchronize_net - Synchronize with packet receive processing
9100 * Wait for packets currently being received to be done.
9101 * Does not block later packets from starting.
9103 void synchronize_net(void)
9106 if (rtnl_is_locked())
9107 synchronize_rcu_expedited();
9111 EXPORT_SYMBOL(synchronize_net);
9114 * unregister_netdevice_queue - remove device from the kernel
9118 * This function shuts down a device interface and removes it
9119 * from the kernel tables.
9120 * If head not NULL, device is queued to be unregistered later.
9122 * Callers must hold the rtnl semaphore. You may want
9123 * unregister_netdev() instead of this.
9126 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9131 list_move_tail(&dev->unreg_list, head);
9133 rollback_registered(dev);
9134 /* Finish processing unregister after unlock */
9138 EXPORT_SYMBOL(unregister_netdevice_queue);
9141 * unregister_netdevice_many - unregister many devices
9142 * @head: list of devices
9144 * Note: As most callers use a stack allocated list_head,
9145 * we force a list_del() to make sure stack wont be corrupted later.
9147 void unregister_netdevice_many(struct list_head *head)
9149 struct net_device *dev;
9151 if (!list_empty(head)) {
9152 rollback_registered_many(head);
9153 list_for_each_entry(dev, head, unreg_list)
9158 EXPORT_SYMBOL(unregister_netdevice_many);
9161 * unregister_netdev - remove device from the kernel
9164 * This function shuts down a device interface and removes it
9165 * from the kernel tables.
9167 * This is just a wrapper for unregister_netdevice that takes
9168 * the rtnl semaphore. In general you want to use this and not
9169 * unregister_netdevice.
9171 void unregister_netdev(struct net_device *dev)
9174 unregister_netdevice(dev);
9177 EXPORT_SYMBOL(unregister_netdev);
9180 * dev_change_net_namespace - move device to different nethost namespace
9182 * @net: network namespace
9183 * @pat: If not NULL name pattern to try if the current device name
9184 * is already taken in the destination network namespace.
9186 * This function shuts down a device interface and moves it
9187 * to a new network namespace. On success 0 is returned, on
9188 * a failure a netagive errno code is returned.
9190 * Callers must hold the rtnl semaphore.
9193 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9195 int err, new_nsid, new_ifindex;
9199 /* Don't allow namespace local devices to be moved. */
9201 if (dev->features & NETIF_F_NETNS_LOCAL)
9204 /* Ensure the device has been registrered */
9205 if (dev->reg_state != NETREG_REGISTERED)
9208 /* Get out if there is nothing todo */
9210 if (net_eq(dev_net(dev), net))
9213 /* Pick the destination device name, and ensure
9214 * we can use it in the destination network namespace.
9217 if (__dev_get_by_name(net, dev->name)) {
9218 /* We get here if we can't use the current device name */
9221 err = dev_get_valid_name(net, dev, pat);
9227 * And now a mini version of register_netdevice unregister_netdevice.
9230 /* If device is running close it first. */
9233 /* And unlink it from device chain */
9234 unlist_netdevice(dev);
9238 /* Shutdown queueing discipline. */
9241 /* Notify protocols, that we are about to destroy
9242 * this device. They should clean all the things.
9244 * Note that dev->reg_state stays at NETREG_REGISTERED.
9245 * This is wanted because this way 8021q and macvlan know
9246 * the device is just moving and can keep their slaves up.
9248 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9251 new_nsid = peernet2id_alloc(dev_net(dev), net);
9252 /* If there is an ifindex conflict assign a new one */
9253 if (__dev_get_by_index(net, dev->ifindex))
9254 new_ifindex = dev_new_index(net);
9256 new_ifindex = dev->ifindex;
9258 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9262 * Flush the unicast and multicast chains
9267 /* Send a netdev-removed uevent to the old namespace */
9268 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9269 netdev_adjacent_del_links(dev);
9271 /* Actually switch the network namespace */
9272 dev_net_set(dev, net);
9273 dev->ifindex = new_ifindex;
9275 /* Send a netdev-add uevent to the new namespace */
9276 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9277 netdev_adjacent_add_links(dev);
9279 /* Fixup kobjects */
9280 err = device_rename(&dev->dev, dev->name);
9283 /* Add the device back in the hashes */
9284 list_netdevice(dev);
9286 /* Notify protocols, that a new device appeared. */
9287 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9290 * Prevent userspace races by waiting until the network
9291 * device is fully setup before sending notifications.
9293 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9300 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9302 static int dev_cpu_dead(unsigned int oldcpu)
9304 struct sk_buff **list_skb;
9305 struct sk_buff *skb;
9307 struct softnet_data *sd, *oldsd, *remsd = NULL;
9309 local_irq_disable();
9310 cpu = smp_processor_id();
9311 sd = &per_cpu(softnet_data, cpu);
9312 oldsd = &per_cpu(softnet_data, oldcpu);
9314 /* Find end of our completion_queue. */
9315 list_skb = &sd->completion_queue;
9317 list_skb = &(*list_skb)->next;
9318 /* Append completion queue from offline CPU. */
9319 *list_skb = oldsd->completion_queue;
9320 oldsd->completion_queue = NULL;
9322 /* Append output queue from offline CPU. */
9323 if (oldsd->output_queue) {
9324 *sd->output_queue_tailp = oldsd->output_queue;
9325 sd->output_queue_tailp = oldsd->output_queue_tailp;
9326 oldsd->output_queue = NULL;
9327 oldsd->output_queue_tailp = &oldsd->output_queue;
9329 /* Append NAPI poll list from offline CPU, with one exception :
9330 * process_backlog() must be called by cpu owning percpu backlog.
9331 * We properly handle process_queue & input_pkt_queue later.
9333 while (!list_empty(&oldsd->poll_list)) {
9334 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9338 list_del_init(&napi->poll_list);
9339 if (napi->poll == process_backlog)
9342 ____napi_schedule(sd, napi);
9345 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9349 remsd = oldsd->rps_ipi_list;
9350 oldsd->rps_ipi_list = NULL;
9352 /* send out pending IPI's on offline CPU */
9353 net_rps_send_ipi(remsd);
9355 /* Process offline CPU's input_pkt_queue */
9356 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9358 input_queue_head_incr(oldsd);
9360 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9362 input_queue_head_incr(oldsd);
9369 * netdev_increment_features - increment feature set by one
9370 * @all: current feature set
9371 * @one: new feature set
9372 * @mask: mask feature set
9374 * Computes a new feature set after adding a device with feature set
9375 * @one to the master device with current feature set @all. Will not
9376 * enable anything that is off in @mask. Returns the new feature set.
9378 netdev_features_t netdev_increment_features(netdev_features_t all,
9379 netdev_features_t one, netdev_features_t mask)
9381 if (mask & NETIF_F_HW_CSUM)
9382 mask |= NETIF_F_CSUM_MASK;
9383 mask |= NETIF_F_VLAN_CHALLENGED;
9385 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9386 all &= one | ~NETIF_F_ALL_FOR_ALL;
9388 /* If one device supports hw checksumming, set for all. */
9389 if (all & NETIF_F_HW_CSUM)
9390 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
9394 EXPORT_SYMBOL(netdev_increment_features);
9396 static struct hlist_head * __net_init netdev_create_hash(void)
9399 struct hlist_head *hash;
9401 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
9403 for (i = 0; i < NETDEV_HASHENTRIES; i++)
9404 INIT_HLIST_HEAD(&hash[i]);
9409 /* Initialize per network namespace state */
9410 static int __net_init netdev_init(struct net *net)
9412 BUILD_BUG_ON(GRO_HASH_BUCKETS >
9413 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
9415 if (net != &init_net)
9416 INIT_LIST_HEAD(&net->dev_base_head);
9418 net->dev_name_head = netdev_create_hash();
9419 if (net->dev_name_head == NULL)
9422 net->dev_index_head = netdev_create_hash();
9423 if (net->dev_index_head == NULL)
9429 kfree(net->dev_name_head);
9435 * netdev_drivername - network driver for the device
9436 * @dev: network device
9438 * Determine network driver for device.
9440 const char *netdev_drivername(const struct net_device *dev)
9442 const struct device_driver *driver;
9443 const struct device *parent;
9444 const char *empty = "";
9446 parent = dev->dev.parent;
9450 driver = parent->driver;
9451 if (driver && driver->name)
9452 return driver->name;
9456 static void __netdev_printk(const char *level, const struct net_device *dev,
9457 struct va_format *vaf)
9459 if (dev && dev->dev.parent) {
9460 dev_printk_emit(level[1] - '0',
9463 dev_driver_string(dev->dev.parent),
9464 dev_name(dev->dev.parent),
9465 netdev_name(dev), netdev_reg_state(dev),
9468 printk("%s%s%s: %pV",
9469 level, netdev_name(dev), netdev_reg_state(dev), vaf);
9471 printk("%s(NULL net_device): %pV", level, vaf);
9475 void netdev_printk(const char *level, const struct net_device *dev,
9476 const char *format, ...)
9478 struct va_format vaf;
9481 va_start(args, format);
9486 __netdev_printk(level, dev, &vaf);
9490 EXPORT_SYMBOL(netdev_printk);
9492 #define define_netdev_printk_level(func, level) \
9493 void func(const struct net_device *dev, const char *fmt, ...) \
9495 struct va_format vaf; \
9498 va_start(args, fmt); \
9503 __netdev_printk(level, dev, &vaf); \
9507 EXPORT_SYMBOL(func);
9509 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
9510 define_netdev_printk_level(netdev_alert, KERN_ALERT);
9511 define_netdev_printk_level(netdev_crit, KERN_CRIT);
9512 define_netdev_printk_level(netdev_err, KERN_ERR);
9513 define_netdev_printk_level(netdev_warn, KERN_WARNING);
9514 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
9515 define_netdev_printk_level(netdev_info, KERN_INFO);
9517 static void __net_exit netdev_exit(struct net *net)
9519 kfree(net->dev_name_head);
9520 kfree(net->dev_index_head);
9521 if (net != &init_net)
9522 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
9525 static struct pernet_operations __net_initdata netdev_net_ops = {
9526 .init = netdev_init,
9527 .exit = netdev_exit,
9530 static void __net_exit default_device_exit(struct net *net)
9532 struct net_device *dev, *aux;
9534 * Push all migratable network devices back to the
9535 * initial network namespace
9538 for_each_netdev_safe(net, dev, aux) {
9540 char fb_name[IFNAMSIZ];
9542 /* Ignore unmoveable devices (i.e. loopback) */
9543 if (dev->features & NETIF_F_NETNS_LOCAL)
9546 /* Leave virtual devices for the generic cleanup */
9547 if (dev->rtnl_link_ops)
9550 /* Push remaining network devices to init_net */
9551 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
9552 err = dev_change_net_namespace(dev, &init_net, fb_name);
9554 pr_emerg("%s: failed to move %s to init_net: %d\n",
9555 __func__, dev->name, err);
9562 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
9564 /* Return with the rtnl_lock held when there are no network
9565 * devices unregistering in any network namespace in net_list.
9569 DEFINE_WAIT_FUNC(wait, woken_wake_function);
9571 add_wait_queue(&netdev_unregistering_wq, &wait);
9573 unregistering = false;
9575 list_for_each_entry(net, net_list, exit_list) {
9576 if (net->dev_unreg_count > 0) {
9577 unregistering = true;
9585 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
9587 remove_wait_queue(&netdev_unregistering_wq, &wait);
9590 static void __net_exit default_device_exit_batch(struct list_head *net_list)
9592 /* At exit all network devices most be removed from a network
9593 * namespace. Do this in the reverse order of registration.
9594 * Do this across as many network namespaces as possible to
9595 * improve batching efficiency.
9597 struct net_device *dev;
9599 LIST_HEAD(dev_kill_list);
9601 /* To prevent network device cleanup code from dereferencing
9602 * loopback devices or network devices that have been freed
9603 * wait here for all pending unregistrations to complete,
9604 * before unregistring the loopback device and allowing the
9605 * network namespace be freed.
9607 * The netdev todo list containing all network devices
9608 * unregistrations that happen in default_device_exit_batch
9609 * will run in the rtnl_unlock() at the end of
9610 * default_device_exit_batch.
9612 rtnl_lock_unregistering(net_list);
9613 list_for_each_entry(net, net_list, exit_list) {
9614 for_each_netdev_reverse(net, dev) {
9615 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
9616 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
9618 unregister_netdevice_queue(dev, &dev_kill_list);
9621 unregister_netdevice_many(&dev_kill_list);
9625 static struct pernet_operations __net_initdata default_device_ops = {
9626 .exit = default_device_exit,
9627 .exit_batch = default_device_exit_batch,
9631 * Initialize the DEV module. At boot time this walks the device list and
9632 * unhooks any devices that fail to initialise (normally hardware not
9633 * present) and leaves us with a valid list of present and active devices.
9638 * This is called single threaded during boot, so no need
9639 * to take the rtnl semaphore.
9641 static int __init net_dev_init(void)
9643 int i, rc = -ENOMEM;
9645 BUG_ON(!dev_boot_phase);
9647 if (dev_proc_init())
9650 if (netdev_kobject_init())
9653 INIT_LIST_HEAD(&ptype_all);
9654 for (i = 0; i < PTYPE_HASH_SIZE; i++)
9655 INIT_LIST_HEAD(&ptype_base[i]);
9657 INIT_LIST_HEAD(&offload_base);
9659 if (register_pernet_subsys(&netdev_net_ops))
9663 * Initialise the packet receive queues.
9666 for_each_possible_cpu(i) {
9667 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
9668 struct softnet_data *sd = &per_cpu(softnet_data, i);
9670 INIT_WORK(flush, flush_backlog);
9672 skb_queue_head_init(&sd->input_pkt_queue);
9673 skb_queue_head_init(&sd->process_queue);
9674 #ifdef CONFIG_XFRM_OFFLOAD
9675 skb_queue_head_init(&sd->xfrm_backlog);
9677 INIT_LIST_HEAD(&sd->poll_list);
9678 sd->output_queue_tailp = &sd->output_queue;
9680 sd->csd.func = rps_trigger_softirq;
9685 init_gro_hash(&sd->backlog);
9686 sd->backlog.poll = process_backlog;
9687 sd->backlog.weight = weight_p;
9692 /* The loopback device is special if any other network devices
9693 * is present in a network namespace the loopback device must
9694 * be present. Since we now dynamically allocate and free the
9695 * loopback device ensure this invariant is maintained by
9696 * keeping the loopback device as the first device on the
9697 * list of network devices. Ensuring the loopback devices
9698 * is the first device that appears and the last network device
9701 if (register_pernet_device(&loopback_net_ops))
9704 if (register_pernet_device(&default_device_ops))
9707 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
9708 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
9710 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
9711 NULL, dev_cpu_dead);
9718 subsys_initcall(net_dev_init);