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
1411 * Takes a device from down to up state. The device's private open
1412 * function is invoked and then the multicast lists are loaded. Finally
1413 * the device is moved into the up state and a %NETDEV_UP message is
1414 * sent to the netdev notifier chain.
1416 * Calling this function on an active interface is a nop. On a failure
1417 * a negative errno code is returned.
1419 int dev_open(struct net_device *dev)
1423 if (dev->flags & IFF_UP)
1426 ret = __dev_open(dev);
1430 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1431 call_netdevice_notifiers(NETDEV_UP, dev);
1435 EXPORT_SYMBOL(dev_open);
1437 static void __dev_close_many(struct list_head *head)
1439 struct net_device *dev;
1444 list_for_each_entry(dev, head, close_list) {
1445 /* Temporarily disable netpoll until the interface is down */
1446 netpoll_poll_disable(dev);
1448 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1450 clear_bit(__LINK_STATE_START, &dev->state);
1452 /* Synchronize to scheduled poll. We cannot touch poll list, it
1453 * can be even on different cpu. So just clear netif_running().
1455 * dev->stop() will invoke napi_disable() on all of it's
1456 * napi_struct instances on this device.
1458 smp_mb__after_atomic(); /* Commit netif_running(). */
1461 dev_deactivate_many(head);
1463 list_for_each_entry(dev, head, close_list) {
1464 const struct net_device_ops *ops = dev->netdev_ops;
1467 * Call the device specific close. This cannot fail.
1468 * Only if device is UP
1470 * We allow it to be called even after a DETACH hot-plug
1476 dev->flags &= ~IFF_UP;
1477 netpoll_poll_enable(dev);
1481 static void __dev_close(struct net_device *dev)
1485 list_add(&dev->close_list, &single);
1486 __dev_close_many(&single);
1490 void dev_close_many(struct list_head *head, bool unlink)
1492 struct net_device *dev, *tmp;
1494 /* Remove the devices that don't need to be closed */
1495 list_for_each_entry_safe(dev, tmp, head, close_list)
1496 if (!(dev->flags & IFF_UP))
1497 list_del_init(&dev->close_list);
1499 __dev_close_many(head);
1501 list_for_each_entry_safe(dev, tmp, head, close_list) {
1502 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1503 call_netdevice_notifiers(NETDEV_DOWN, dev);
1505 list_del_init(&dev->close_list);
1508 EXPORT_SYMBOL(dev_close_many);
1511 * dev_close - shutdown an interface.
1512 * @dev: device to shutdown
1514 * This function moves an active device into down state. A
1515 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1516 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1519 void dev_close(struct net_device *dev)
1521 if (dev->flags & IFF_UP) {
1524 list_add(&dev->close_list, &single);
1525 dev_close_many(&single, true);
1529 EXPORT_SYMBOL(dev_close);
1533 * dev_disable_lro - disable Large Receive Offload on a device
1536 * Disable Large Receive Offload (LRO) on a net device. Must be
1537 * called under RTNL. This is needed if received packets may be
1538 * forwarded to another interface.
1540 void dev_disable_lro(struct net_device *dev)
1542 struct net_device *lower_dev;
1543 struct list_head *iter;
1545 dev->wanted_features &= ~NETIF_F_LRO;
1546 netdev_update_features(dev);
1548 if (unlikely(dev->features & NETIF_F_LRO))
1549 netdev_WARN(dev, "failed to disable LRO!\n");
1551 netdev_for_each_lower_dev(dev, lower_dev, iter)
1552 dev_disable_lro(lower_dev);
1554 EXPORT_SYMBOL(dev_disable_lro);
1557 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1560 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1561 * called under RTNL. This is needed if Generic XDP is installed on
1564 static void dev_disable_gro_hw(struct net_device *dev)
1566 dev->wanted_features &= ~NETIF_F_GRO_HW;
1567 netdev_update_features(dev);
1569 if (unlikely(dev->features & NETIF_F_GRO_HW))
1570 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1573 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1576 case NETDEV_##val: \
1577 return "NETDEV_" __stringify(val);
1579 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1580 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1581 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1582 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1583 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1584 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1585 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1586 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1587 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1590 return "UNKNOWN_NETDEV_EVENT";
1592 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1594 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1595 struct net_device *dev)
1597 struct netdev_notifier_info info = {
1601 return nb->notifier_call(nb, val, &info);
1604 static int dev_boot_phase = 1;
1607 * register_netdevice_notifier - register a network notifier block
1610 * Register a notifier to be called when network device events occur.
1611 * The notifier passed is linked into the kernel structures and must
1612 * not be reused until it has been unregistered. A negative errno code
1613 * is returned on a failure.
1615 * When registered all registration and up events are replayed
1616 * to the new notifier to allow device to have a race free
1617 * view of the network device list.
1620 int register_netdevice_notifier(struct notifier_block *nb)
1622 struct net_device *dev;
1623 struct net_device *last;
1627 /* Close race with setup_net() and cleanup_net() */
1628 down_write(&pernet_ops_rwsem);
1630 err = raw_notifier_chain_register(&netdev_chain, nb);
1636 for_each_netdev(net, dev) {
1637 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1638 err = notifier_to_errno(err);
1642 if (!(dev->flags & IFF_UP))
1645 call_netdevice_notifier(nb, NETDEV_UP, dev);
1651 up_write(&pernet_ops_rwsem);
1657 for_each_netdev(net, dev) {
1661 if (dev->flags & IFF_UP) {
1662 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1664 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1666 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1671 raw_notifier_chain_unregister(&netdev_chain, nb);
1674 EXPORT_SYMBOL(register_netdevice_notifier);
1677 * unregister_netdevice_notifier - unregister a network notifier block
1680 * Unregister a notifier previously registered by
1681 * register_netdevice_notifier(). The notifier is unlinked into the
1682 * kernel structures and may then be reused. A negative errno code
1683 * is returned on a failure.
1685 * After unregistering unregister and down device events are synthesized
1686 * for all devices on the device list to the removed notifier to remove
1687 * the need for special case cleanup code.
1690 int unregister_netdevice_notifier(struct notifier_block *nb)
1692 struct net_device *dev;
1696 /* Close race with setup_net() and cleanup_net() */
1697 down_write(&pernet_ops_rwsem);
1699 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1704 for_each_netdev(net, dev) {
1705 if (dev->flags & IFF_UP) {
1706 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1708 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1710 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1715 up_write(&pernet_ops_rwsem);
1718 EXPORT_SYMBOL(unregister_netdevice_notifier);
1721 * call_netdevice_notifiers_info - call all network notifier blocks
1722 * @val: value passed unmodified to notifier function
1723 * @info: notifier information data
1725 * Call all network notifier blocks. Parameters and return value
1726 * are as for raw_notifier_call_chain().
1729 static int call_netdevice_notifiers_info(unsigned long val,
1730 struct netdev_notifier_info *info)
1733 return raw_notifier_call_chain(&netdev_chain, val, info);
1737 * call_netdevice_notifiers - call all network notifier blocks
1738 * @val: value passed unmodified to notifier function
1739 * @dev: net_device pointer passed unmodified to notifier function
1741 * Call all network notifier blocks. Parameters and return value
1742 * are as for raw_notifier_call_chain().
1745 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1747 struct netdev_notifier_info info = {
1751 return call_netdevice_notifiers_info(val, &info);
1753 EXPORT_SYMBOL(call_netdevice_notifiers);
1756 * call_netdevice_notifiers_mtu - call all network notifier blocks
1757 * @val: value passed unmodified to notifier function
1758 * @dev: net_device pointer passed unmodified to notifier function
1759 * @arg: additional u32 argument passed to the notifier function
1761 * Call all network notifier blocks. Parameters and return value
1762 * are as for raw_notifier_call_chain().
1764 static int call_netdevice_notifiers_mtu(unsigned long val,
1765 struct net_device *dev, u32 arg)
1767 struct netdev_notifier_info_ext info = {
1772 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1774 return call_netdevice_notifiers_info(val, &info.info);
1777 #ifdef CONFIG_NET_INGRESS
1778 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1780 void net_inc_ingress_queue(void)
1782 static_branch_inc(&ingress_needed_key);
1784 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1786 void net_dec_ingress_queue(void)
1788 static_branch_dec(&ingress_needed_key);
1790 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1793 #ifdef CONFIG_NET_EGRESS
1794 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1796 void net_inc_egress_queue(void)
1798 static_branch_inc(&egress_needed_key);
1800 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1802 void net_dec_egress_queue(void)
1804 static_branch_dec(&egress_needed_key);
1806 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1809 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1810 #ifdef HAVE_JUMP_LABEL
1811 static atomic_t netstamp_needed_deferred;
1812 static atomic_t netstamp_wanted;
1813 static void netstamp_clear(struct work_struct *work)
1815 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1818 wanted = atomic_add_return(deferred, &netstamp_wanted);
1820 static_branch_enable(&netstamp_needed_key);
1822 static_branch_disable(&netstamp_needed_key);
1824 static DECLARE_WORK(netstamp_work, netstamp_clear);
1827 void net_enable_timestamp(void)
1829 #ifdef HAVE_JUMP_LABEL
1833 wanted = atomic_read(&netstamp_wanted);
1836 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1839 atomic_inc(&netstamp_needed_deferred);
1840 schedule_work(&netstamp_work);
1842 static_branch_inc(&netstamp_needed_key);
1845 EXPORT_SYMBOL(net_enable_timestamp);
1847 void net_disable_timestamp(void)
1849 #ifdef HAVE_JUMP_LABEL
1853 wanted = atomic_read(&netstamp_wanted);
1856 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1859 atomic_dec(&netstamp_needed_deferred);
1860 schedule_work(&netstamp_work);
1862 static_branch_dec(&netstamp_needed_key);
1865 EXPORT_SYMBOL(net_disable_timestamp);
1867 static inline void net_timestamp_set(struct sk_buff *skb)
1870 if (static_branch_unlikely(&netstamp_needed_key))
1871 __net_timestamp(skb);
1874 #define net_timestamp_check(COND, SKB) \
1875 if (static_branch_unlikely(&netstamp_needed_key)) { \
1876 if ((COND) && !(SKB)->tstamp) \
1877 __net_timestamp(SKB); \
1880 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1884 if (!(dev->flags & IFF_UP))
1887 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1888 if (skb->len <= len)
1891 /* if TSO is enabled, we don't care about the length as the packet
1892 * could be forwarded without being segmented before
1894 if (skb_is_gso(skb))
1899 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1901 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1903 int ret = ____dev_forward_skb(dev, skb);
1906 skb->protocol = eth_type_trans(skb, dev);
1907 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1912 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1915 * dev_forward_skb - loopback an skb to another netif
1917 * @dev: destination network device
1918 * @skb: buffer to forward
1921 * NET_RX_SUCCESS (no congestion)
1922 * NET_RX_DROP (packet was dropped, but freed)
1924 * dev_forward_skb can be used for injecting an skb from the
1925 * start_xmit function of one device into the receive queue
1926 * of another device.
1928 * The receiving device may be in another namespace, so
1929 * we have to clear all information in the skb that could
1930 * impact namespace isolation.
1932 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1934 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1936 EXPORT_SYMBOL_GPL(dev_forward_skb);
1938 static inline int deliver_skb(struct sk_buff *skb,
1939 struct packet_type *pt_prev,
1940 struct net_device *orig_dev)
1942 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1944 refcount_inc(&skb->users);
1945 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1948 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1949 struct packet_type **pt,
1950 struct net_device *orig_dev,
1952 struct list_head *ptype_list)
1954 struct packet_type *ptype, *pt_prev = *pt;
1956 list_for_each_entry_rcu(ptype, ptype_list, list) {
1957 if (ptype->type != type)
1960 deliver_skb(skb, pt_prev, orig_dev);
1966 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1968 if (!ptype->af_packet_priv || !skb->sk)
1971 if (ptype->id_match)
1972 return ptype->id_match(ptype, skb->sk);
1973 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1980 * Support routine. Sends outgoing frames to any network
1981 * taps currently in use.
1984 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1986 struct packet_type *ptype;
1987 struct sk_buff *skb2 = NULL;
1988 struct packet_type *pt_prev = NULL;
1989 struct list_head *ptype_list = &ptype_all;
1993 list_for_each_entry_rcu(ptype, ptype_list, list) {
1994 /* Never send packets back to the socket
1995 * they originated from - MvS (miquels@drinkel.ow.org)
1997 if (skb_loop_sk(ptype, skb))
2001 deliver_skb(skb2, pt_prev, skb->dev);
2006 /* need to clone skb, done only once */
2007 skb2 = skb_clone(skb, GFP_ATOMIC);
2011 net_timestamp_set(skb2);
2013 /* skb->nh should be correctly
2014 * set by sender, so that the second statement is
2015 * just protection against buggy protocols.
2017 skb_reset_mac_header(skb2);
2019 if (skb_network_header(skb2) < skb2->data ||
2020 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2021 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2022 ntohs(skb2->protocol),
2024 skb_reset_network_header(skb2);
2027 skb2->transport_header = skb2->network_header;
2028 skb2->pkt_type = PACKET_OUTGOING;
2032 if (ptype_list == &ptype_all) {
2033 ptype_list = &dev->ptype_all;
2038 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2039 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2045 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2048 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2049 * @dev: Network device
2050 * @txq: number of queues available
2052 * If real_num_tx_queues is changed the tc mappings may no longer be
2053 * valid. To resolve this verify the tc mapping remains valid and if
2054 * not NULL the mapping. With no priorities mapping to this
2055 * offset/count pair it will no longer be used. In the worst case TC0
2056 * is invalid nothing can be done so disable priority mappings. If is
2057 * expected that drivers will fix this mapping if they can before
2058 * calling netif_set_real_num_tx_queues.
2060 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2063 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2065 /* If TC0 is invalidated disable TC mapping */
2066 if (tc->offset + tc->count > txq) {
2067 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2072 /* Invalidated prio to tc mappings set to TC0 */
2073 for (i = 1; i < TC_BITMASK + 1; i++) {
2074 int q = netdev_get_prio_tc_map(dev, i);
2076 tc = &dev->tc_to_txq[q];
2077 if (tc->offset + tc->count > txq) {
2078 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2080 netdev_set_prio_tc_map(dev, i, 0);
2085 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2088 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2091 /* walk through the TCs and see if it falls into any of them */
2092 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2093 if ((txq - tc->offset) < tc->count)
2097 /* didn't find it, just return -1 to indicate no match */
2103 EXPORT_SYMBOL(netdev_txq_to_tc);
2106 struct static_key xps_needed __read_mostly;
2107 EXPORT_SYMBOL(xps_needed);
2108 struct static_key xps_rxqs_needed __read_mostly;
2109 EXPORT_SYMBOL(xps_rxqs_needed);
2110 static DEFINE_MUTEX(xps_map_mutex);
2111 #define xmap_dereference(P) \
2112 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2114 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2117 struct xps_map *map = NULL;
2121 map = xmap_dereference(dev_maps->attr_map[tci]);
2125 for (pos = map->len; pos--;) {
2126 if (map->queues[pos] != index)
2130 map->queues[pos] = map->queues[--map->len];
2134 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2135 kfree_rcu(map, rcu);
2142 static bool remove_xps_queue_cpu(struct net_device *dev,
2143 struct xps_dev_maps *dev_maps,
2144 int cpu, u16 offset, u16 count)
2146 int num_tc = dev->num_tc ? : 1;
2147 bool active = false;
2150 for (tci = cpu * num_tc; num_tc--; tci++) {
2153 for (i = count, j = offset; i--; j++) {
2154 if (!remove_xps_queue(dev_maps, tci, j))
2164 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2165 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2166 u16 offset, u16 count, bool is_rxqs_map)
2168 bool active = false;
2171 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2173 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2177 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2179 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2181 for (i = offset + (count - 1); count--; i--)
2182 netdev_queue_numa_node_write(
2183 netdev_get_tx_queue(dev, i),
2186 kfree_rcu(dev_maps, rcu);
2190 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2193 const unsigned long *possible_mask = NULL;
2194 struct xps_dev_maps *dev_maps;
2195 unsigned int nr_ids;
2197 if (!static_key_false(&xps_needed))
2201 mutex_lock(&xps_map_mutex);
2203 if (static_key_false(&xps_rxqs_needed)) {
2204 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2206 nr_ids = dev->num_rx_queues;
2207 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2208 offset, count, true);
2212 dev_maps = xmap_dereference(dev->xps_cpus_map);
2216 if (num_possible_cpus() > 1)
2217 possible_mask = cpumask_bits(cpu_possible_mask);
2218 nr_ids = nr_cpu_ids;
2219 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2223 if (static_key_enabled(&xps_rxqs_needed))
2224 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2226 static_key_slow_dec_cpuslocked(&xps_needed);
2227 mutex_unlock(&xps_map_mutex);
2231 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2233 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2236 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2237 u16 index, bool is_rxqs_map)
2239 struct xps_map *new_map;
2240 int alloc_len = XPS_MIN_MAP_ALLOC;
2243 for (pos = 0; map && pos < map->len; pos++) {
2244 if (map->queues[pos] != index)
2249 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2251 if (pos < map->alloc_len)
2254 alloc_len = map->alloc_len * 2;
2257 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2261 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2263 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2264 cpu_to_node(attr_index));
2268 for (i = 0; i < pos; i++)
2269 new_map->queues[i] = map->queues[i];
2270 new_map->alloc_len = alloc_len;
2276 /* Must be called under cpus_read_lock */
2277 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2278 u16 index, bool is_rxqs_map)
2280 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2281 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2282 int i, j, tci, numa_node_id = -2;
2283 int maps_sz, num_tc = 1, tc = 0;
2284 struct xps_map *map, *new_map;
2285 bool active = false;
2286 unsigned int nr_ids;
2289 /* Do not allow XPS on subordinate device directly */
2290 num_tc = dev->num_tc;
2294 /* If queue belongs to subordinate dev use its map */
2295 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2297 tc = netdev_txq_to_tc(dev, index);
2302 mutex_lock(&xps_map_mutex);
2304 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2305 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2306 nr_ids = dev->num_rx_queues;
2308 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2309 if (num_possible_cpus() > 1) {
2310 online_mask = cpumask_bits(cpu_online_mask);
2311 possible_mask = cpumask_bits(cpu_possible_mask);
2313 dev_maps = xmap_dereference(dev->xps_cpus_map);
2314 nr_ids = nr_cpu_ids;
2317 if (maps_sz < L1_CACHE_BYTES)
2318 maps_sz = L1_CACHE_BYTES;
2320 /* allocate memory for queue storage */
2321 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2324 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2325 if (!new_dev_maps) {
2326 mutex_unlock(&xps_map_mutex);
2330 tci = j * num_tc + tc;
2331 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2334 map = expand_xps_map(map, j, index, is_rxqs_map);
2338 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2342 goto out_no_new_maps;
2344 static_key_slow_inc_cpuslocked(&xps_needed);
2346 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2348 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2350 /* copy maps belonging to foreign traffic classes */
2351 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2352 /* fill in the new device map from the old device map */
2353 map = xmap_dereference(dev_maps->attr_map[tci]);
2354 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2357 /* We need to explicitly update tci as prevous loop
2358 * could break out early if dev_maps is NULL.
2360 tci = j * num_tc + tc;
2362 if (netif_attr_test_mask(j, mask, nr_ids) &&
2363 netif_attr_test_online(j, online_mask, nr_ids)) {
2364 /* add tx-queue to CPU/rx-queue maps */
2367 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2368 while ((pos < map->len) && (map->queues[pos] != index))
2371 if (pos == map->len)
2372 map->queues[map->len++] = index;
2375 if (numa_node_id == -2)
2376 numa_node_id = cpu_to_node(j);
2377 else if (numa_node_id != cpu_to_node(j))
2381 } else if (dev_maps) {
2382 /* fill in the new device map from the old device map */
2383 map = xmap_dereference(dev_maps->attr_map[tci]);
2384 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2387 /* copy maps belonging to foreign traffic classes */
2388 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2389 /* fill in the new device map from the old device map */
2390 map = xmap_dereference(dev_maps->attr_map[tci]);
2391 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2396 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2398 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2400 /* Cleanup old maps */
2402 goto out_no_old_maps;
2404 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2406 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2407 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2408 map = xmap_dereference(dev_maps->attr_map[tci]);
2409 if (map && map != new_map)
2410 kfree_rcu(map, rcu);
2414 kfree_rcu(dev_maps, rcu);
2417 dev_maps = new_dev_maps;
2422 /* update Tx queue numa node */
2423 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2424 (numa_node_id >= 0) ?
2425 numa_node_id : NUMA_NO_NODE);
2431 /* removes tx-queue from unused CPUs/rx-queues */
2432 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2434 for (i = tc, tci = j * num_tc; i--; tci++)
2435 active |= remove_xps_queue(dev_maps, tci, index);
2436 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2437 !netif_attr_test_online(j, online_mask, nr_ids))
2438 active |= remove_xps_queue(dev_maps, tci, index);
2439 for (i = num_tc - tc, tci++; --i; tci++)
2440 active |= remove_xps_queue(dev_maps, tci, index);
2443 /* free map if not active */
2446 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2448 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2449 kfree_rcu(dev_maps, rcu);
2453 mutex_unlock(&xps_map_mutex);
2457 /* remove any maps that we added */
2458 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2460 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2461 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2463 xmap_dereference(dev_maps->attr_map[tci]) :
2465 if (new_map && new_map != map)
2470 mutex_unlock(&xps_map_mutex);
2472 kfree(new_dev_maps);
2475 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2477 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2483 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2488 EXPORT_SYMBOL(netif_set_xps_queue);
2491 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2493 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2495 /* Unbind any subordinate channels */
2496 while (txq-- != &dev->_tx[0]) {
2498 netdev_unbind_sb_channel(dev, txq->sb_dev);
2502 void netdev_reset_tc(struct net_device *dev)
2505 netif_reset_xps_queues_gt(dev, 0);
2507 netdev_unbind_all_sb_channels(dev);
2509 /* Reset TC configuration of device */
2511 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2512 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2514 EXPORT_SYMBOL(netdev_reset_tc);
2516 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2518 if (tc >= dev->num_tc)
2522 netif_reset_xps_queues(dev, offset, count);
2524 dev->tc_to_txq[tc].count = count;
2525 dev->tc_to_txq[tc].offset = offset;
2528 EXPORT_SYMBOL(netdev_set_tc_queue);
2530 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2532 if (num_tc > TC_MAX_QUEUE)
2536 netif_reset_xps_queues_gt(dev, 0);
2538 netdev_unbind_all_sb_channels(dev);
2540 dev->num_tc = num_tc;
2543 EXPORT_SYMBOL(netdev_set_num_tc);
2545 void netdev_unbind_sb_channel(struct net_device *dev,
2546 struct net_device *sb_dev)
2548 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2551 netif_reset_xps_queues_gt(sb_dev, 0);
2553 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2554 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2556 while (txq-- != &dev->_tx[0]) {
2557 if (txq->sb_dev == sb_dev)
2561 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2563 int netdev_bind_sb_channel_queue(struct net_device *dev,
2564 struct net_device *sb_dev,
2565 u8 tc, u16 count, u16 offset)
2567 /* Make certain the sb_dev and dev are already configured */
2568 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2571 /* We cannot hand out queues we don't have */
2572 if ((offset + count) > dev->real_num_tx_queues)
2575 /* Record the mapping */
2576 sb_dev->tc_to_txq[tc].count = count;
2577 sb_dev->tc_to_txq[tc].offset = offset;
2579 /* Provide a way for Tx queue to find the tc_to_txq map or
2580 * XPS map for itself.
2583 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2587 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2589 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2591 /* Do not use a multiqueue device to represent a subordinate channel */
2592 if (netif_is_multiqueue(dev))
2595 /* We allow channels 1 - 32767 to be used for subordinate channels.
2596 * Channel 0 is meant to be "native" mode and used only to represent
2597 * the main root device. We allow writing 0 to reset the device back
2598 * to normal mode after being used as a subordinate channel.
2600 if (channel > S16_MAX)
2603 dev->num_tc = -channel;
2607 EXPORT_SYMBOL(netdev_set_sb_channel);
2610 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2611 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2613 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2618 disabling = txq < dev->real_num_tx_queues;
2620 if (txq < 1 || txq > dev->num_tx_queues)
2623 if (dev->reg_state == NETREG_REGISTERED ||
2624 dev->reg_state == NETREG_UNREGISTERING) {
2627 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2633 netif_setup_tc(dev, txq);
2635 dev->real_num_tx_queues = txq;
2639 qdisc_reset_all_tx_gt(dev, txq);
2641 netif_reset_xps_queues_gt(dev, txq);
2645 dev->real_num_tx_queues = txq;
2650 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2654 * netif_set_real_num_rx_queues - set actual number of RX queues used
2655 * @dev: Network device
2656 * @rxq: Actual number of RX queues
2658 * This must be called either with the rtnl_lock held or before
2659 * registration of the net device. Returns 0 on success, or a
2660 * negative error code. If called before registration, it always
2663 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2667 if (rxq < 1 || rxq > dev->num_rx_queues)
2670 if (dev->reg_state == NETREG_REGISTERED) {
2673 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2679 dev->real_num_rx_queues = rxq;
2682 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2686 * netif_get_num_default_rss_queues - default number of RSS queues
2688 * This routine should set an upper limit on the number of RSS queues
2689 * used by default by multiqueue devices.
2691 int netif_get_num_default_rss_queues(void)
2693 return is_kdump_kernel() ?
2694 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2696 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2698 static void __netif_reschedule(struct Qdisc *q)
2700 struct softnet_data *sd;
2701 unsigned long flags;
2703 local_irq_save(flags);
2704 sd = this_cpu_ptr(&softnet_data);
2705 q->next_sched = NULL;
2706 *sd->output_queue_tailp = q;
2707 sd->output_queue_tailp = &q->next_sched;
2708 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2709 local_irq_restore(flags);
2712 void __netif_schedule(struct Qdisc *q)
2714 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2715 __netif_reschedule(q);
2717 EXPORT_SYMBOL(__netif_schedule);
2719 struct dev_kfree_skb_cb {
2720 enum skb_free_reason reason;
2723 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2725 return (struct dev_kfree_skb_cb *)skb->cb;
2728 void netif_schedule_queue(struct netdev_queue *txq)
2731 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2732 struct Qdisc *q = rcu_dereference(txq->qdisc);
2734 __netif_schedule(q);
2738 EXPORT_SYMBOL(netif_schedule_queue);
2740 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2742 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2746 q = rcu_dereference(dev_queue->qdisc);
2747 __netif_schedule(q);
2751 EXPORT_SYMBOL(netif_tx_wake_queue);
2753 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2755 unsigned long flags;
2760 if (likely(refcount_read(&skb->users) == 1)) {
2762 refcount_set(&skb->users, 0);
2763 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2766 get_kfree_skb_cb(skb)->reason = reason;
2767 local_irq_save(flags);
2768 skb->next = __this_cpu_read(softnet_data.completion_queue);
2769 __this_cpu_write(softnet_data.completion_queue, skb);
2770 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2771 local_irq_restore(flags);
2773 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2775 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2777 if (in_irq() || irqs_disabled())
2778 __dev_kfree_skb_irq(skb, reason);
2782 EXPORT_SYMBOL(__dev_kfree_skb_any);
2786 * netif_device_detach - mark device as removed
2787 * @dev: network device
2789 * Mark device as removed from system and therefore no longer available.
2791 void netif_device_detach(struct net_device *dev)
2793 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2794 netif_running(dev)) {
2795 netif_tx_stop_all_queues(dev);
2798 EXPORT_SYMBOL(netif_device_detach);
2801 * netif_device_attach - mark device as attached
2802 * @dev: network device
2804 * Mark device as attached from system and restart if needed.
2806 void netif_device_attach(struct net_device *dev)
2808 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2809 netif_running(dev)) {
2810 netif_tx_wake_all_queues(dev);
2811 __netdev_watchdog_up(dev);
2814 EXPORT_SYMBOL(netif_device_attach);
2817 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2818 * to be used as a distribution range.
2820 static u16 skb_tx_hash(const struct net_device *dev,
2821 const struct net_device *sb_dev,
2822 struct sk_buff *skb)
2826 u16 qcount = dev->real_num_tx_queues;
2829 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2831 qoffset = sb_dev->tc_to_txq[tc].offset;
2832 qcount = sb_dev->tc_to_txq[tc].count;
2835 if (skb_rx_queue_recorded(skb)) {
2836 hash = skb_get_rx_queue(skb);
2837 while (unlikely(hash >= qcount))
2839 return hash + qoffset;
2842 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2845 static void skb_warn_bad_offload(const struct sk_buff *skb)
2847 static const netdev_features_t null_features;
2848 struct net_device *dev = skb->dev;
2849 const char *name = "";
2851 if (!net_ratelimit())
2855 if (dev->dev.parent)
2856 name = dev_driver_string(dev->dev.parent);
2858 name = netdev_name(dev);
2860 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2861 "gso_type=%d ip_summed=%d\n",
2862 name, dev ? &dev->features : &null_features,
2863 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2864 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2865 skb_shinfo(skb)->gso_type, skb->ip_summed);
2869 * Invalidate hardware checksum when packet is to be mangled, and
2870 * complete checksum manually on outgoing path.
2872 int skb_checksum_help(struct sk_buff *skb)
2875 int ret = 0, offset;
2877 if (skb->ip_summed == CHECKSUM_COMPLETE)
2878 goto out_set_summed;
2880 if (unlikely(skb_shinfo(skb)->gso_size)) {
2881 skb_warn_bad_offload(skb);
2885 /* Before computing a checksum, we should make sure no frag could
2886 * be modified by an external entity : checksum could be wrong.
2888 if (skb_has_shared_frag(skb)) {
2889 ret = __skb_linearize(skb);
2894 offset = skb_checksum_start_offset(skb);
2895 BUG_ON(offset >= skb_headlen(skb));
2896 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2898 offset += skb->csum_offset;
2899 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2901 if (skb_cloned(skb) &&
2902 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2903 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2908 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2910 skb->ip_summed = CHECKSUM_NONE;
2914 EXPORT_SYMBOL(skb_checksum_help);
2916 int skb_crc32c_csum_help(struct sk_buff *skb)
2919 int ret = 0, offset, start;
2921 if (skb->ip_summed != CHECKSUM_PARTIAL)
2924 if (unlikely(skb_is_gso(skb)))
2927 /* Before computing a checksum, we should make sure no frag could
2928 * be modified by an external entity : checksum could be wrong.
2930 if (unlikely(skb_has_shared_frag(skb))) {
2931 ret = __skb_linearize(skb);
2935 start = skb_checksum_start_offset(skb);
2936 offset = start + offsetof(struct sctphdr, checksum);
2937 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2941 if (skb_cloned(skb) &&
2942 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2943 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2947 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2948 skb->len - start, ~(__u32)0,
2950 *(__le32 *)(skb->data + offset) = crc32c_csum;
2951 skb->ip_summed = CHECKSUM_NONE;
2952 skb->csum_not_inet = 0;
2957 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2959 __be16 type = skb->protocol;
2961 /* Tunnel gso handlers can set protocol to ethernet. */
2962 if (type == htons(ETH_P_TEB)) {
2965 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2968 eth = (struct ethhdr *)skb->data;
2969 type = eth->h_proto;
2972 return __vlan_get_protocol(skb, type, depth);
2976 * skb_mac_gso_segment - mac layer segmentation handler.
2977 * @skb: buffer to segment
2978 * @features: features for the output path (see dev->features)
2980 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2981 netdev_features_t features)
2983 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2984 struct packet_offload *ptype;
2985 int vlan_depth = skb->mac_len;
2986 __be16 type = skb_network_protocol(skb, &vlan_depth);
2988 if (unlikely(!type))
2989 return ERR_PTR(-EINVAL);
2991 __skb_pull(skb, vlan_depth);
2994 list_for_each_entry_rcu(ptype, &offload_base, list) {
2995 if (ptype->type == type && ptype->callbacks.gso_segment) {
2996 segs = ptype->callbacks.gso_segment(skb, features);
3002 __skb_push(skb, skb->data - skb_mac_header(skb));
3006 EXPORT_SYMBOL(skb_mac_gso_segment);
3009 /* openvswitch calls this on rx path, so we need a different check.
3011 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3014 return skb->ip_summed != CHECKSUM_PARTIAL &&
3015 skb->ip_summed != CHECKSUM_UNNECESSARY;
3017 return skb->ip_summed == CHECKSUM_NONE;
3021 * __skb_gso_segment - Perform segmentation on skb.
3022 * @skb: buffer to segment
3023 * @features: features for the output path (see dev->features)
3024 * @tx_path: whether it is called in TX path
3026 * This function segments the given skb and returns a list of segments.
3028 * It may return NULL if the skb requires no segmentation. This is
3029 * only possible when GSO is used for verifying header integrity.
3031 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3033 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3034 netdev_features_t features, bool tx_path)
3036 struct sk_buff *segs;
3038 if (unlikely(skb_needs_check(skb, tx_path))) {
3041 /* We're going to init ->check field in TCP or UDP header */
3042 err = skb_cow_head(skb, 0);
3044 return ERR_PTR(err);
3047 /* Only report GSO partial support if it will enable us to
3048 * support segmentation on this frame without needing additional
3051 if (features & NETIF_F_GSO_PARTIAL) {
3052 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3053 struct net_device *dev = skb->dev;
3055 partial_features |= dev->features & dev->gso_partial_features;
3056 if (!skb_gso_ok(skb, features | partial_features))
3057 features &= ~NETIF_F_GSO_PARTIAL;
3060 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3061 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3063 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3064 SKB_GSO_CB(skb)->encap_level = 0;
3066 skb_reset_mac_header(skb);
3067 skb_reset_mac_len(skb);
3069 segs = skb_mac_gso_segment(skb, features);
3071 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3072 skb_warn_bad_offload(skb);
3076 EXPORT_SYMBOL(__skb_gso_segment);
3078 /* Take action when hardware reception checksum errors are detected. */
3080 void netdev_rx_csum_fault(struct net_device *dev)
3082 if (net_ratelimit()) {
3083 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3087 EXPORT_SYMBOL(netdev_rx_csum_fault);
3090 /* XXX: check that highmem exists at all on the given machine. */
3091 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3093 #ifdef CONFIG_HIGHMEM
3096 if (!(dev->features & NETIF_F_HIGHDMA)) {
3097 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3098 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3100 if (PageHighMem(skb_frag_page(frag)))
3108 /* If MPLS offload request, verify we are testing hardware MPLS features
3109 * instead of standard features for the netdev.
3111 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3112 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3113 netdev_features_t features,
3116 if (eth_p_mpls(type))
3117 features &= skb->dev->mpls_features;
3122 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3123 netdev_features_t features,
3130 static netdev_features_t harmonize_features(struct sk_buff *skb,
3131 netdev_features_t features)
3136 type = skb_network_protocol(skb, &tmp);
3137 features = net_mpls_features(skb, features, type);
3139 if (skb->ip_summed != CHECKSUM_NONE &&
3140 !can_checksum_protocol(features, type)) {
3141 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3143 if (illegal_highdma(skb->dev, skb))
3144 features &= ~NETIF_F_SG;
3149 netdev_features_t passthru_features_check(struct sk_buff *skb,
3150 struct net_device *dev,
3151 netdev_features_t features)
3155 EXPORT_SYMBOL(passthru_features_check);
3157 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3158 struct net_device *dev,
3159 netdev_features_t features)
3161 return vlan_features_check(skb, features);
3164 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3165 struct net_device *dev,
3166 netdev_features_t features)
3168 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3170 if (gso_segs > dev->gso_max_segs)
3171 return features & ~NETIF_F_GSO_MASK;
3173 /* Support for GSO partial features requires software
3174 * intervention before we can actually process the packets
3175 * so we need to strip support for any partial features now
3176 * and we can pull them back in after we have partially
3177 * segmented the frame.
3179 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3180 features &= ~dev->gso_partial_features;
3182 /* Make sure to clear the IPv4 ID mangling feature if the
3183 * IPv4 header has the potential to be fragmented.
3185 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3186 struct iphdr *iph = skb->encapsulation ?
3187 inner_ip_hdr(skb) : ip_hdr(skb);
3189 if (!(iph->frag_off & htons(IP_DF)))
3190 features &= ~NETIF_F_TSO_MANGLEID;
3196 netdev_features_t netif_skb_features(struct sk_buff *skb)
3198 struct net_device *dev = skb->dev;
3199 netdev_features_t features = dev->features;
3201 if (skb_is_gso(skb))
3202 features = gso_features_check(skb, dev, features);
3204 /* If encapsulation offload request, verify we are testing
3205 * hardware encapsulation features instead of standard
3206 * features for the netdev
3208 if (skb->encapsulation)
3209 features &= dev->hw_enc_features;
3211 if (skb_vlan_tagged(skb))
3212 features = netdev_intersect_features(features,
3213 dev->vlan_features |
3214 NETIF_F_HW_VLAN_CTAG_TX |
3215 NETIF_F_HW_VLAN_STAG_TX);
3217 if (dev->netdev_ops->ndo_features_check)
3218 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3221 features &= dflt_features_check(skb, dev, features);
3223 return harmonize_features(skb, features);
3225 EXPORT_SYMBOL(netif_skb_features);
3227 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3228 struct netdev_queue *txq, bool more)
3233 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3234 dev_queue_xmit_nit(skb, dev);
3237 trace_net_dev_start_xmit(skb, dev);
3238 rc = netdev_start_xmit(skb, dev, txq, more);
3239 trace_net_dev_xmit(skb, rc, dev, len);
3244 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3245 struct netdev_queue *txq, int *ret)
3247 struct sk_buff *skb = first;
3248 int rc = NETDEV_TX_OK;
3251 struct sk_buff *next = skb->next;
3254 rc = xmit_one(skb, dev, txq, next != NULL);
3255 if (unlikely(!dev_xmit_complete(rc))) {
3261 if (netif_xmit_stopped(txq) && skb) {
3262 rc = NETDEV_TX_BUSY;
3272 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3273 netdev_features_t features)
3275 if (skb_vlan_tag_present(skb) &&
3276 !vlan_hw_offload_capable(features, skb->vlan_proto))
3277 skb = __vlan_hwaccel_push_inside(skb);
3281 int skb_csum_hwoffload_help(struct sk_buff *skb,
3282 const netdev_features_t features)
3284 if (unlikely(skb->csum_not_inet))
3285 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3286 skb_crc32c_csum_help(skb);
3288 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3290 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3292 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3294 netdev_features_t features;
3296 features = netif_skb_features(skb);
3297 skb = validate_xmit_vlan(skb, features);
3301 skb = sk_validate_xmit_skb(skb, dev);
3305 if (netif_needs_gso(skb, features)) {
3306 struct sk_buff *segs;
3308 segs = skb_gso_segment(skb, features);
3316 if (skb_needs_linearize(skb, features) &&
3317 __skb_linearize(skb))
3320 /* If packet is not checksummed and device does not
3321 * support checksumming for this protocol, complete
3322 * checksumming here.
3324 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3325 if (skb->encapsulation)
3326 skb_set_inner_transport_header(skb,
3327 skb_checksum_start_offset(skb));
3329 skb_set_transport_header(skb,
3330 skb_checksum_start_offset(skb));
3331 if (skb_csum_hwoffload_help(skb, features))
3336 skb = validate_xmit_xfrm(skb, features, again);
3343 atomic_long_inc(&dev->tx_dropped);
3347 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3349 struct sk_buff *next, *head = NULL, *tail;
3351 for (; skb != NULL; skb = next) {
3355 /* in case skb wont be segmented, point to itself */
3358 skb = validate_xmit_skb(skb, dev, again);
3366 /* If skb was segmented, skb->prev points to
3367 * the last segment. If not, it still contains skb.
3373 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3375 static void qdisc_pkt_len_init(struct sk_buff *skb)
3377 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3379 qdisc_skb_cb(skb)->pkt_len = skb->len;
3381 /* To get more precise estimation of bytes sent on wire,
3382 * we add to pkt_len the headers size of all segments
3384 if (shinfo->gso_size) {
3385 unsigned int hdr_len;
3386 u16 gso_segs = shinfo->gso_segs;
3388 /* mac layer + network layer */
3389 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3391 /* + transport layer */
3392 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3393 const struct tcphdr *th;
3394 struct tcphdr _tcphdr;
3396 th = skb_header_pointer(skb, skb_transport_offset(skb),
3397 sizeof(_tcphdr), &_tcphdr);
3399 hdr_len += __tcp_hdrlen(th);
3401 struct udphdr _udphdr;
3403 if (skb_header_pointer(skb, skb_transport_offset(skb),
3404 sizeof(_udphdr), &_udphdr))
3405 hdr_len += sizeof(struct udphdr);
3408 if (shinfo->gso_type & SKB_GSO_DODGY)
3409 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3412 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3416 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3417 struct net_device *dev,
3418 struct netdev_queue *txq)
3420 spinlock_t *root_lock = qdisc_lock(q);
3421 struct sk_buff *to_free = NULL;
3425 qdisc_calculate_pkt_len(skb, q);
3427 if (q->flags & TCQ_F_NOLOCK) {
3428 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3429 __qdisc_drop(skb, &to_free);
3432 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3436 if (unlikely(to_free))
3437 kfree_skb_list(to_free);
3442 * Heuristic to force contended enqueues to serialize on a
3443 * separate lock before trying to get qdisc main lock.
3444 * This permits qdisc->running owner to get the lock more
3445 * often and dequeue packets faster.
3447 contended = qdisc_is_running(q);
3448 if (unlikely(contended))
3449 spin_lock(&q->busylock);
3451 spin_lock(root_lock);
3452 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3453 __qdisc_drop(skb, &to_free);
3455 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3456 qdisc_run_begin(q)) {
3458 * This is a work-conserving queue; there are no old skbs
3459 * waiting to be sent out; and the qdisc is not running -
3460 * xmit the skb directly.
3463 qdisc_bstats_update(q, skb);
3465 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3466 if (unlikely(contended)) {
3467 spin_unlock(&q->busylock);
3474 rc = NET_XMIT_SUCCESS;
3476 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3477 if (qdisc_run_begin(q)) {
3478 if (unlikely(contended)) {
3479 spin_unlock(&q->busylock);
3486 spin_unlock(root_lock);
3487 if (unlikely(to_free))
3488 kfree_skb_list(to_free);
3489 if (unlikely(contended))
3490 spin_unlock(&q->busylock);
3494 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3495 static void skb_update_prio(struct sk_buff *skb)
3497 const struct netprio_map *map;
3498 const struct sock *sk;
3499 unsigned int prioidx;
3503 map = rcu_dereference_bh(skb->dev->priomap);
3506 sk = skb_to_full_sk(skb);
3510 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3512 if (prioidx < map->priomap_len)
3513 skb->priority = map->priomap[prioidx];
3516 #define skb_update_prio(skb)
3519 DEFINE_PER_CPU(int, xmit_recursion);
3520 EXPORT_SYMBOL(xmit_recursion);
3523 * dev_loopback_xmit - loop back @skb
3524 * @net: network namespace this loopback is happening in
3525 * @sk: sk needed to be a netfilter okfn
3526 * @skb: buffer to transmit
3528 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3530 skb_reset_mac_header(skb);
3531 __skb_pull(skb, skb_network_offset(skb));
3532 skb->pkt_type = PACKET_LOOPBACK;
3533 skb->ip_summed = CHECKSUM_UNNECESSARY;
3534 WARN_ON(!skb_dst(skb));
3539 EXPORT_SYMBOL(dev_loopback_xmit);
3541 #ifdef CONFIG_NET_EGRESS
3542 static struct sk_buff *
3543 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3545 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3546 struct tcf_result cl_res;
3551 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3552 mini_qdisc_bstats_cpu_update(miniq, skb);
3554 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3556 case TC_ACT_RECLASSIFY:
3557 skb->tc_index = TC_H_MIN(cl_res.classid);
3560 mini_qdisc_qstats_cpu_drop(miniq);
3561 *ret = NET_XMIT_DROP;
3567 *ret = NET_XMIT_SUCCESS;
3570 case TC_ACT_REDIRECT:
3571 /* No need to push/pop skb's mac_header here on egress! */
3572 skb_do_redirect(skb);
3573 *ret = NET_XMIT_SUCCESS;
3581 #endif /* CONFIG_NET_EGRESS */
3584 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3585 struct xps_dev_maps *dev_maps, unsigned int tci)
3587 struct xps_map *map;
3588 int queue_index = -1;
3592 tci += netdev_get_prio_tc_map(dev, skb->priority);
3595 map = rcu_dereference(dev_maps->attr_map[tci]);
3598 queue_index = map->queues[0];
3600 queue_index = map->queues[reciprocal_scale(
3601 skb_get_hash(skb), map->len)];
3602 if (unlikely(queue_index >= dev->real_num_tx_queues))
3609 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3610 struct sk_buff *skb)
3613 struct xps_dev_maps *dev_maps;
3614 struct sock *sk = skb->sk;
3615 int queue_index = -1;
3617 if (!static_key_false(&xps_needed))
3621 if (!static_key_false(&xps_rxqs_needed))
3624 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3626 int tci = sk_rx_queue_get(sk);
3628 if (tci >= 0 && tci < dev->num_rx_queues)
3629 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3634 if (queue_index < 0) {
3635 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3637 unsigned int tci = skb->sender_cpu - 1;
3639 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3651 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3652 struct net_device *sb_dev,
3653 select_queue_fallback_t fallback)
3657 EXPORT_SYMBOL(dev_pick_tx_zero);
3659 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3660 struct net_device *sb_dev,
3661 select_queue_fallback_t fallback)
3663 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3665 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3667 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3668 struct net_device *sb_dev)
3670 struct sock *sk = skb->sk;
3671 int queue_index = sk_tx_queue_get(sk);
3673 sb_dev = sb_dev ? : dev;
3675 if (queue_index < 0 || skb->ooo_okay ||
3676 queue_index >= dev->real_num_tx_queues) {
3677 int new_index = get_xps_queue(dev, sb_dev, skb);
3680 new_index = skb_tx_hash(dev, sb_dev, skb);
3682 if (queue_index != new_index && sk &&
3684 rcu_access_pointer(sk->sk_dst_cache))
3685 sk_tx_queue_set(sk, new_index);
3687 queue_index = new_index;
3693 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3694 struct sk_buff *skb,
3695 struct net_device *sb_dev)
3697 int queue_index = 0;
3700 u32 sender_cpu = skb->sender_cpu - 1;
3702 if (sender_cpu >= (u32)NR_CPUS)
3703 skb->sender_cpu = raw_smp_processor_id() + 1;
3706 if (dev->real_num_tx_queues != 1) {
3707 const struct net_device_ops *ops = dev->netdev_ops;
3709 if (ops->ndo_select_queue)
3710 queue_index = ops->ndo_select_queue(dev, skb, sb_dev,
3713 queue_index = __netdev_pick_tx(dev, skb, sb_dev);
3715 queue_index = netdev_cap_txqueue(dev, queue_index);
3718 skb_set_queue_mapping(skb, queue_index);
3719 return netdev_get_tx_queue(dev, queue_index);
3723 * __dev_queue_xmit - transmit a buffer
3724 * @skb: buffer to transmit
3725 * @sb_dev: suboordinate device used for L2 forwarding offload
3727 * Queue a buffer for transmission to a network device. The caller must
3728 * have set the device and priority and built the buffer before calling
3729 * this function. The function can be called from an interrupt.
3731 * A negative errno code is returned on a failure. A success does not
3732 * guarantee the frame will be transmitted as it may be dropped due
3733 * to congestion or traffic shaping.
3735 * -----------------------------------------------------------------------------------
3736 * I notice this method can also return errors from the queue disciplines,
3737 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3740 * Regardless of the return value, the skb is consumed, so it is currently
3741 * difficult to retry a send to this method. (You can bump the ref count
3742 * before sending to hold a reference for retry if you are careful.)
3744 * When calling this method, interrupts MUST be enabled. This is because
3745 * the BH enable code must have IRQs enabled so that it will not deadlock.
3748 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3750 struct net_device *dev = skb->dev;
3751 struct netdev_queue *txq;
3756 skb_reset_mac_header(skb);
3758 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3759 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3761 /* Disable soft irqs for various locks below. Also
3762 * stops preemption for RCU.
3766 skb_update_prio(skb);
3768 qdisc_pkt_len_init(skb);
3769 #ifdef CONFIG_NET_CLS_ACT
3770 skb->tc_at_ingress = 0;
3771 # ifdef CONFIG_NET_EGRESS
3772 if (static_branch_unlikely(&egress_needed_key)) {
3773 skb = sch_handle_egress(skb, &rc, dev);
3779 /* If device/qdisc don't need skb->dst, release it right now while
3780 * its hot in this cpu cache.
3782 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3787 txq = netdev_pick_tx(dev, skb, sb_dev);
3788 q = rcu_dereference_bh(txq->qdisc);
3790 trace_net_dev_queue(skb);
3792 rc = __dev_xmit_skb(skb, q, dev, txq);
3796 /* The device has no queue. Common case for software devices:
3797 * loopback, all the sorts of tunnels...
3799 * Really, it is unlikely that netif_tx_lock protection is necessary
3800 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3802 * However, it is possible, that they rely on protection
3805 * Check this and shot the lock. It is not prone from deadlocks.
3806 *Either shot noqueue qdisc, it is even simpler 8)
3808 if (dev->flags & IFF_UP) {
3809 int cpu = smp_processor_id(); /* ok because BHs are off */
3811 if (txq->xmit_lock_owner != cpu) {
3812 if (unlikely(__this_cpu_read(xmit_recursion) >
3813 XMIT_RECURSION_LIMIT))
3814 goto recursion_alert;
3816 skb = validate_xmit_skb(skb, dev, &again);
3820 HARD_TX_LOCK(dev, txq, cpu);
3822 if (!netif_xmit_stopped(txq)) {
3823 __this_cpu_inc(xmit_recursion);
3824 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3825 __this_cpu_dec(xmit_recursion);
3826 if (dev_xmit_complete(rc)) {
3827 HARD_TX_UNLOCK(dev, txq);
3831 HARD_TX_UNLOCK(dev, txq);
3832 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3835 /* Recursion is detected! It is possible,
3839 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3845 rcu_read_unlock_bh();
3847 atomic_long_inc(&dev->tx_dropped);
3848 kfree_skb_list(skb);
3851 rcu_read_unlock_bh();
3855 int dev_queue_xmit(struct sk_buff *skb)
3857 return __dev_queue_xmit(skb, NULL);
3859 EXPORT_SYMBOL(dev_queue_xmit);
3861 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3863 return __dev_queue_xmit(skb, sb_dev);
3865 EXPORT_SYMBOL(dev_queue_xmit_accel);
3867 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3869 struct net_device *dev = skb->dev;
3870 struct sk_buff *orig_skb = skb;
3871 struct netdev_queue *txq;
3872 int ret = NETDEV_TX_BUSY;
3875 if (unlikely(!netif_running(dev) ||
3876 !netif_carrier_ok(dev)))
3879 skb = validate_xmit_skb_list(skb, dev, &again);
3880 if (skb != orig_skb)
3883 skb_set_queue_mapping(skb, queue_id);
3884 txq = skb_get_tx_queue(dev, skb);
3888 HARD_TX_LOCK(dev, txq, smp_processor_id());
3889 if (!netif_xmit_frozen_or_drv_stopped(txq))
3890 ret = netdev_start_xmit(skb, dev, txq, false);
3891 HARD_TX_UNLOCK(dev, txq);
3895 if (!dev_xmit_complete(ret))
3900 atomic_long_inc(&dev->tx_dropped);
3901 kfree_skb_list(skb);
3902 return NET_XMIT_DROP;
3904 EXPORT_SYMBOL(dev_direct_xmit);
3906 /*************************************************************************
3908 *************************************************************************/
3910 int netdev_max_backlog __read_mostly = 1000;
3911 EXPORT_SYMBOL(netdev_max_backlog);
3913 int netdev_tstamp_prequeue __read_mostly = 1;
3914 int netdev_budget __read_mostly = 300;
3915 unsigned int __read_mostly netdev_budget_usecs = 2000;
3916 int weight_p __read_mostly = 64; /* old backlog weight */
3917 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3918 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3919 int dev_rx_weight __read_mostly = 64;
3920 int dev_tx_weight __read_mostly = 64;
3922 /* Called with irq disabled */
3923 static inline void ____napi_schedule(struct softnet_data *sd,
3924 struct napi_struct *napi)
3926 list_add_tail(&napi->poll_list, &sd->poll_list);
3927 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3932 /* One global table that all flow-based protocols share. */
3933 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3934 EXPORT_SYMBOL(rps_sock_flow_table);
3935 u32 rps_cpu_mask __read_mostly;
3936 EXPORT_SYMBOL(rps_cpu_mask);
3938 struct static_key rps_needed __read_mostly;
3939 EXPORT_SYMBOL(rps_needed);
3940 struct static_key rfs_needed __read_mostly;
3941 EXPORT_SYMBOL(rfs_needed);
3943 static struct rps_dev_flow *
3944 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3945 struct rps_dev_flow *rflow, u16 next_cpu)
3947 if (next_cpu < nr_cpu_ids) {
3948 #ifdef CONFIG_RFS_ACCEL
3949 struct netdev_rx_queue *rxqueue;
3950 struct rps_dev_flow_table *flow_table;
3951 struct rps_dev_flow *old_rflow;
3956 /* Should we steer this flow to a different hardware queue? */
3957 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3958 !(dev->features & NETIF_F_NTUPLE))
3960 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3961 if (rxq_index == skb_get_rx_queue(skb))
3964 rxqueue = dev->_rx + rxq_index;
3965 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3968 flow_id = skb_get_hash(skb) & flow_table->mask;
3969 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3970 rxq_index, flow_id);
3974 rflow = &flow_table->flows[flow_id];
3976 if (old_rflow->filter == rflow->filter)
3977 old_rflow->filter = RPS_NO_FILTER;
3981 per_cpu(softnet_data, next_cpu).input_queue_head;
3984 rflow->cpu = next_cpu;
3989 * get_rps_cpu is called from netif_receive_skb and returns the target
3990 * CPU from the RPS map of the receiving queue for a given skb.
3991 * rcu_read_lock must be held on entry.
3993 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3994 struct rps_dev_flow **rflowp)
3996 const struct rps_sock_flow_table *sock_flow_table;
3997 struct netdev_rx_queue *rxqueue = dev->_rx;
3998 struct rps_dev_flow_table *flow_table;
3999 struct rps_map *map;
4004 if (skb_rx_queue_recorded(skb)) {
4005 u16 index = skb_get_rx_queue(skb);
4007 if (unlikely(index >= dev->real_num_rx_queues)) {
4008 WARN_ONCE(dev->real_num_rx_queues > 1,
4009 "%s received packet on queue %u, but number "
4010 "of RX queues is %u\n",
4011 dev->name, index, dev->real_num_rx_queues);
4017 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4019 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4020 map = rcu_dereference(rxqueue->rps_map);
4021 if (!flow_table && !map)
4024 skb_reset_network_header(skb);
4025 hash = skb_get_hash(skb);
4029 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4030 if (flow_table && sock_flow_table) {
4031 struct rps_dev_flow *rflow;
4035 /* First check into global flow table if there is a match */
4036 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4037 if ((ident ^ hash) & ~rps_cpu_mask)
4040 next_cpu = ident & rps_cpu_mask;
4042 /* OK, now we know there is a match,
4043 * we can look at the local (per receive queue) flow table
4045 rflow = &flow_table->flows[hash & flow_table->mask];
4049 * If the desired CPU (where last recvmsg was done) is
4050 * different from current CPU (one in the rx-queue flow
4051 * table entry), switch if one of the following holds:
4052 * - Current CPU is unset (>= nr_cpu_ids).
4053 * - Current CPU is offline.
4054 * - The current CPU's queue tail has advanced beyond the
4055 * last packet that was enqueued using this table entry.
4056 * This guarantees that all previous packets for the flow
4057 * have been dequeued, thus preserving in order delivery.
4059 if (unlikely(tcpu != next_cpu) &&
4060 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4061 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4062 rflow->last_qtail)) >= 0)) {
4064 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4067 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4077 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4078 if (cpu_online(tcpu)) {
4088 #ifdef CONFIG_RFS_ACCEL
4091 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4092 * @dev: Device on which the filter was set
4093 * @rxq_index: RX queue index
4094 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4095 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4097 * Drivers that implement ndo_rx_flow_steer() should periodically call
4098 * this function for each installed filter and remove the filters for
4099 * which it returns %true.
4101 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4102 u32 flow_id, u16 filter_id)
4104 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4105 struct rps_dev_flow_table *flow_table;
4106 struct rps_dev_flow *rflow;
4111 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4112 if (flow_table && flow_id <= flow_table->mask) {
4113 rflow = &flow_table->flows[flow_id];
4114 cpu = READ_ONCE(rflow->cpu);
4115 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4116 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4117 rflow->last_qtail) <
4118 (int)(10 * flow_table->mask)))
4124 EXPORT_SYMBOL(rps_may_expire_flow);
4126 #endif /* CONFIG_RFS_ACCEL */
4128 /* Called from hardirq (IPI) context */
4129 static void rps_trigger_softirq(void *data)
4131 struct softnet_data *sd = data;
4133 ____napi_schedule(sd, &sd->backlog);
4137 #endif /* CONFIG_RPS */
4140 * Check if this softnet_data structure is another cpu one
4141 * If yes, queue it to our IPI list and return 1
4144 static int rps_ipi_queued(struct softnet_data *sd)
4147 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4150 sd->rps_ipi_next = mysd->rps_ipi_list;
4151 mysd->rps_ipi_list = sd;
4153 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4156 #endif /* CONFIG_RPS */
4160 #ifdef CONFIG_NET_FLOW_LIMIT
4161 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4164 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4166 #ifdef CONFIG_NET_FLOW_LIMIT
4167 struct sd_flow_limit *fl;
4168 struct softnet_data *sd;
4169 unsigned int old_flow, new_flow;
4171 if (qlen < (netdev_max_backlog >> 1))
4174 sd = this_cpu_ptr(&softnet_data);
4177 fl = rcu_dereference(sd->flow_limit);
4179 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4180 old_flow = fl->history[fl->history_head];
4181 fl->history[fl->history_head] = new_flow;
4184 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4186 if (likely(fl->buckets[old_flow]))
4187 fl->buckets[old_flow]--;
4189 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4201 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4202 * queue (may be a remote CPU queue).
4204 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4205 unsigned int *qtail)
4207 struct softnet_data *sd;
4208 unsigned long flags;
4211 sd = &per_cpu(softnet_data, cpu);
4213 local_irq_save(flags);
4216 if (!netif_running(skb->dev))
4218 qlen = skb_queue_len(&sd->input_pkt_queue);
4219 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4222 __skb_queue_tail(&sd->input_pkt_queue, skb);
4223 input_queue_tail_incr_save(sd, qtail);
4225 local_irq_restore(flags);
4226 return NET_RX_SUCCESS;
4229 /* Schedule NAPI for backlog device
4230 * We can use non atomic operation since we own the queue lock
4232 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4233 if (!rps_ipi_queued(sd))
4234 ____napi_schedule(sd, &sd->backlog);
4243 local_irq_restore(flags);
4245 atomic_long_inc(&skb->dev->rx_dropped);
4250 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4252 struct net_device *dev = skb->dev;
4253 struct netdev_rx_queue *rxqueue;
4257 if (skb_rx_queue_recorded(skb)) {
4258 u16 index = skb_get_rx_queue(skb);
4260 if (unlikely(index >= dev->real_num_rx_queues)) {
4261 WARN_ONCE(dev->real_num_rx_queues > 1,
4262 "%s received packet on queue %u, but number "
4263 "of RX queues is %u\n",
4264 dev->name, index, dev->real_num_rx_queues);
4266 return rxqueue; /* Return first rxqueue */
4273 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4274 struct xdp_buff *xdp,
4275 struct bpf_prog *xdp_prog)
4277 struct netdev_rx_queue *rxqueue;
4278 void *orig_data, *orig_data_end;
4279 u32 metalen, act = XDP_DROP;
4283 /* Reinjected packets coming from act_mirred or similar should
4284 * not get XDP generic processing.
4286 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4289 /* XDP packets must be linear and must have sufficient headroom
4290 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4291 * native XDP provides, thus we need to do it here as well.
4293 if (skb_is_nonlinear(skb) ||
4294 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4295 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4296 int troom = skb->tail + skb->data_len - skb->end;
4298 /* In case we have to go down the path and also linearize,
4299 * then lets do the pskb_expand_head() work just once here.
4301 if (pskb_expand_head(skb,
4302 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4303 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4305 if (skb_linearize(skb))
4309 /* The XDP program wants to see the packet starting at the MAC
4312 mac_len = skb->data - skb_mac_header(skb);
4313 hlen = skb_headlen(skb) + mac_len;
4314 xdp->data = skb->data - mac_len;
4315 xdp->data_meta = xdp->data;
4316 xdp->data_end = xdp->data + hlen;
4317 xdp->data_hard_start = skb->data - skb_headroom(skb);
4318 orig_data_end = xdp->data_end;
4319 orig_data = xdp->data;
4321 rxqueue = netif_get_rxqueue(skb);
4322 xdp->rxq = &rxqueue->xdp_rxq;
4324 act = bpf_prog_run_xdp(xdp_prog, xdp);
4326 off = xdp->data - orig_data;
4328 __skb_pull(skb, off);
4330 __skb_push(skb, -off);
4331 skb->mac_header += off;
4333 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4336 off = orig_data_end - xdp->data_end;
4338 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4346 __skb_push(skb, mac_len);
4349 metalen = xdp->data - xdp->data_meta;
4351 skb_metadata_set(skb, metalen);
4354 bpf_warn_invalid_xdp_action(act);
4357 trace_xdp_exception(skb->dev, xdp_prog, act);
4368 /* When doing generic XDP we have to bypass the qdisc layer and the
4369 * network taps in order to match in-driver-XDP behavior.
4371 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4373 struct net_device *dev = skb->dev;
4374 struct netdev_queue *txq;
4375 bool free_skb = true;
4378 txq = netdev_pick_tx(dev, skb, NULL);
4379 cpu = smp_processor_id();
4380 HARD_TX_LOCK(dev, txq, cpu);
4381 if (!netif_xmit_stopped(txq)) {
4382 rc = netdev_start_xmit(skb, dev, txq, 0);
4383 if (dev_xmit_complete(rc))
4386 HARD_TX_UNLOCK(dev, txq);
4388 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4392 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4394 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4396 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4399 struct xdp_buff xdp;
4403 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4404 if (act != XDP_PASS) {
4407 err = xdp_do_generic_redirect(skb->dev, skb,
4413 generic_xdp_tx(skb, xdp_prog);
4424 EXPORT_SYMBOL_GPL(do_xdp_generic);
4426 static int netif_rx_internal(struct sk_buff *skb)
4430 net_timestamp_check(netdev_tstamp_prequeue, skb);
4432 trace_netif_rx(skb);
4434 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4439 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4443 /* Consider XDP consuming the packet a success from
4444 * the netdev point of view we do not want to count
4447 if (ret != XDP_PASS)
4448 return NET_RX_SUCCESS;
4452 if (static_key_false(&rps_needed)) {
4453 struct rps_dev_flow voidflow, *rflow = &voidflow;
4459 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4461 cpu = smp_processor_id();
4463 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4472 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4479 * netif_rx - post buffer to the network code
4480 * @skb: buffer to post
4482 * This function receives a packet from a device driver and queues it for
4483 * the upper (protocol) levels to process. It always succeeds. The buffer
4484 * may be dropped during processing for congestion control or by the
4488 * NET_RX_SUCCESS (no congestion)
4489 * NET_RX_DROP (packet was dropped)
4493 int netif_rx(struct sk_buff *skb)
4495 trace_netif_rx_entry(skb);
4497 return netif_rx_internal(skb);
4499 EXPORT_SYMBOL(netif_rx);
4501 int netif_rx_ni(struct sk_buff *skb)
4505 trace_netif_rx_ni_entry(skb);
4508 err = netif_rx_internal(skb);
4509 if (local_softirq_pending())
4515 EXPORT_SYMBOL(netif_rx_ni);
4517 static __latent_entropy void net_tx_action(struct softirq_action *h)
4519 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4521 if (sd->completion_queue) {
4522 struct sk_buff *clist;
4524 local_irq_disable();
4525 clist = sd->completion_queue;
4526 sd->completion_queue = NULL;
4530 struct sk_buff *skb = clist;
4532 clist = clist->next;
4534 WARN_ON(refcount_read(&skb->users));
4535 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4536 trace_consume_skb(skb);
4538 trace_kfree_skb(skb, net_tx_action);
4540 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4543 __kfree_skb_defer(skb);
4546 __kfree_skb_flush();
4549 if (sd->output_queue) {
4552 local_irq_disable();
4553 head = sd->output_queue;
4554 sd->output_queue = NULL;
4555 sd->output_queue_tailp = &sd->output_queue;
4559 struct Qdisc *q = head;
4560 spinlock_t *root_lock = NULL;
4562 head = head->next_sched;
4564 if (!(q->flags & TCQ_F_NOLOCK)) {
4565 root_lock = qdisc_lock(q);
4566 spin_lock(root_lock);
4568 /* We need to make sure head->next_sched is read
4569 * before clearing __QDISC_STATE_SCHED
4571 smp_mb__before_atomic();
4572 clear_bit(__QDISC_STATE_SCHED, &q->state);
4575 spin_unlock(root_lock);
4579 xfrm_dev_backlog(sd);
4582 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4583 /* This hook is defined here for ATM LANE */
4584 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4585 unsigned char *addr) __read_mostly;
4586 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4589 static inline struct sk_buff *
4590 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4591 struct net_device *orig_dev)
4593 #ifdef CONFIG_NET_CLS_ACT
4594 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4595 struct tcf_result cl_res;
4597 /* If there's at least one ingress present somewhere (so
4598 * we get here via enabled static key), remaining devices
4599 * that are not configured with an ingress qdisc will bail
4606 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4610 qdisc_skb_cb(skb)->pkt_len = skb->len;
4611 skb->tc_at_ingress = 1;
4612 mini_qdisc_bstats_cpu_update(miniq, skb);
4614 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4616 case TC_ACT_RECLASSIFY:
4617 skb->tc_index = TC_H_MIN(cl_res.classid);
4620 mini_qdisc_qstats_cpu_drop(miniq);
4628 case TC_ACT_REDIRECT:
4629 /* skb_mac_header check was done by cls/act_bpf, so
4630 * we can safely push the L2 header back before
4631 * redirecting to another netdev
4633 __skb_push(skb, skb->mac_len);
4634 skb_do_redirect(skb);
4636 case TC_ACT_REINSERT:
4637 /* this does not scrub the packet, and updates stats on error */
4638 skb_tc_reinsert(skb, &cl_res);
4643 #endif /* CONFIG_NET_CLS_ACT */
4648 * netdev_is_rx_handler_busy - check if receive handler is registered
4649 * @dev: device to check
4651 * Check if a receive handler is already registered for a given device.
4652 * Return true if there one.
4654 * The caller must hold the rtnl_mutex.
4656 bool netdev_is_rx_handler_busy(struct net_device *dev)
4659 return dev && rtnl_dereference(dev->rx_handler);
4661 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4664 * netdev_rx_handler_register - register receive handler
4665 * @dev: device to register a handler for
4666 * @rx_handler: receive handler to register
4667 * @rx_handler_data: data pointer that is used by rx handler
4669 * Register a receive handler for a device. This handler will then be
4670 * called from __netif_receive_skb. A negative errno code is returned
4673 * The caller must hold the rtnl_mutex.
4675 * For a general description of rx_handler, see enum rx_handler_result.
4677 int netdev_rx_handler_register(struct net_device *dev,
4678 rx_handler_func_t *rx_handler,
4679 void *rx_handler_data)
4681 if (netdev_is_rx_handler_busy(dev))
4684 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4687 /* Note: rx_handler_data must be set before rx_handler */
4688 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4689 rcu_assign_pointer(dev->rx_handler, rx_handler);
4693 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4696 * netdev_rx_handler_unregister - unregister receive handler
4697 * @dev: device to unregister a handler from
4699 * Unregister a receive handler from a device.
4701 * The caller must hold the rtnl_mutex.
4703 void netdev_rx_handler_unregister(struct net_device *dev)
4707 RCU_INIT_POINTER(dev->rx_handler, NULL);
4708 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4709 * section has a guarantee to see a non NULL rx_handler_data
4713 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4715 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4718 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4719 * the special handling of PFMEMALLOC skbs.
4721 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4723 switch (skb->protocol) {
4724 case htons(ETH_P_ARP):
4725 case htons(ETH_P_IP):
4726 case htons(ETH_P_IPV6):
4727 case htons(ETH_P_8021Q):
4728 case htons(ETH_P_8021AD):
4735 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4736 int *ret, struct net_device *orig_dev)
4738 #ifdef CONFIG_NETFILTER_INGRESS
4739 if (nf_hook_ingress_active(skb)) {
4743 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4748 ingress_retval = nf_hook_ingress(skb);
4750 return ingress_retval;
4752 #endif /* CONFIG_NETFILTER_INGRESS */
4756 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4757 struct packet_type **ppt_prev)
4759 struct packet_type *ptype, *pt_prev;
4760 rx_handler_func_t *rx_handler;
4761 struct net_device *orig_dev;
4762 bool deliver_exact = false;
4763 int ret = NET_RX_DROP;
4766 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4768 trace_netif_receive_skb(skb);
4770 orig_dev = skb->dev;
4772 skb_reset_network_header(skb);
4773 if (!skb_transport_header_was_set(skb))
4774 skb_reset_transport_header(skb);
4775 skb_reset_mac_len(skb);
4780 skb->skb_iif = skb->dev->ifindex;
4782 __this_cpu_inc(softnet_data.processed);
4784 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4785 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4786 skb = skb_vlan_untag(skb);
4791 if (skb_skip_tc_classify(skb))
4797 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4799 ret = deliver_skb(skb, pt_prev, orig_dev);
4803 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4805 ret = deliver_skb(skb, pt_prev, orig_dev);
4810 #ifdef CONFIG_NET_INGRESS
4811 if (static_branch_unlikely(&ingress_needed_key)) {
4812 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4816 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4822 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4825 if (skb_vlan_tag_present(skb)) {
4827 ret = deliver_skb(skb, pt_prev, orig_dev);
4830 if (vlan_do_receive(&skb))
4832 else if (unlikely(!skb))
4836 rx_handler = rcu_dereference(skb->dev->rx_handler);
4839 ret = deliver_skb(skb, pt_prev, orig_dev);
4842 switch (rx_handler(&skb)) {
4843 case RX_HANDLER_CONSUMED:
4844 ret = NET_RX_SUCCESS;
4846 case RX_HANDLER_ANOTHER:
4848 case RX_HANDLER_EXACT:
4849 deliver_exact = true;
4850 case RX_HANDLER_PASS:
4857 if (unlikely(skb_vlan_tag_present(skb))) {
4858 if (skb_vlan_tag_get_id(skb))
4859 skb->pkt_type = PACKET_OTHERHOST;
4860 /* Note: we might in the future use prio bits
4861 * and set skb->priority like in vlan_do_receive()
4862 * For the time being, just ignore Priority Code Point
4867 type = skb->protocol;
4869 /* deliver only exact match when indicated */
4870 if (likely(!deliver_exact)) {
4871 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4872 &ptype_base[ntohs(type) &
4876 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4877 &orig_dev->ptype_specific);
4879 if (unlikely(skb->dev != orig_dev)) {
4880 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4881 &skb->dev->ptype_specific);
4885 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4887 *ppt_prev = pt_prev;
4891 atomic_long_inc(&skb->dev->rx_dropped);
4893 atomic_long_inc(&skb->dev->rx_nohandler);
4895 /* Jamal, now you will not able to escape explaining
4896 * me how you were going to use this. :-)
4905 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4907 struct net_device *orig_dev = skb->dev;
4908 struct packet_type *pt_prev = NULL;
4911 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4913 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4918 * netif_receive_skb_core - special purpose version of netif_receive_skb
4919 * @skb: buffer to process
4921 * More direct receive version of netif_receive_skb(). It should
4922 * only be used by callers that have a need to skip RPS and Generic XDP.
4923 * Caller must also take care of handling if (page_is_)pfmemalloc.
4925 * This function may only be called from softirq context and interrupts
4926 * should be enabled.
4928 * Return values (usually ignored):
4929 * NET_RX_SUCCESS: no congestion
4930 * NET_RX_DROP: packet was dropped
4932 int netif_receive_skb_core(struct sk_buff *skb)
4937 ret = __netif_receive_skb_one_core(skb, false);
4942 EXPORT_SYMBOL(netif_receive_skb_core);
4944 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
4945 struct packet_type *pt_prev,
4946 struct net_device *orig_dev)
4948 struct sk_buff *skb, *next;
4952 if (list_empty(head))
4954 if (pt_prev->list_func != NULL)
4955 pt_prev->list_func(head, pt_prev, orig_dev);
4957 list_for_each_entry_safe(skb, next, head, list)
4958 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4961 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
4963 /* Fast-path assumptions:
4964 * - There is no RX handler.
4965 * - Only one packet_type matches.
4966 * If either of these fails, we will end up doing some per-packet
4967 * processing in-line, then handling the 'last ptype' for the whole
4968 * sublist. This can't cause out-of-order delivery to any single ptype,
4969 * because the 'last ptype' must be constant across the sublist, and all
4970 * other ptypes are handled per-packet.
4972 /* Current (common) ptype of sublist */
4973 struct packet_type *pt_curr = NULL;
4974 /* Current (common) orig_dev of sublist */
4975 struct net_device *od_curr = NULL;
4976 struct list_head sublist;
4977 struct sk_buff *skb, *next;
4979 INIT_LIST_HEAD(&sublist);
4980 list_for_each_entry_safe(skb, next, head, list) {
4981 struct net_device *orig_dev = skb->dev;
4982 struct packet_type *pt_prev = NULL;
4984 list_del(&skb->list);
4985 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4988 if (pt_curr != pt_prev || od_curr != orig_dev) {
4989 /* dispatch old sublist */
4990 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
4991 /* start new sublist */
4992 INIT_LIST_HEAD(&sublist);
4996 list_add_tail(&skb->list, &sublist);
4999 /* dispatch final sublist */
5000 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5003 static int __netif_receive_skb(struct sk_buff *skb)
5007 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5008 unsigned int noreclaim_flag;
5011 * PFMEMALLOC skbs are special, they should
5012 * - be delivered to SOCK_MEMALLOC sockets only
5013 * - stay away from userspace
5014 * - have bounded memory usage
5016 * Use PF_MEMALLOC as this saves us from propagating the allocation
5017 * context down to all allocation sites.
5019 noreclaim_flag = memalloc_noreclaim_save();
5020 ret = __netif_receive_skb_one_core(skb, true);
5021 memalloc_noreclaim_restore(noreclaim_flag);
5023 ret = __netif_receive_skb_one_core(skb, false);
5028 static void __netif_receive_skb_list(struct list_head *head)
5030 unsigned long noreclaim_flag = 0;
5031 struct sk_buff *skb, *next;
5032 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5034 list_for_each_entry_safe(skb, next, head, list) {
5035 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5036 struct list_head sublist;
5038 /* Handle the previous sublist */
5039 list_cut_before(&sublist, head, &skb->list);
5040 if (!list_empty(&sublist))
5041 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5042 pfmemalloc = !pfmemalloc;
5043 /* See comments in __netif_receive_skb */
5045 noreclaim_flag = memalloc_noreclaim_save();
5047 memalloc_noreclaim_restore(noreclaim_flag);
5050 /* Handle the remaining sublist */
5051 if (!list_empty(head))
5052 __netif_receive_skb_list_core(head, pfmemalloc);
5053 /* Restore pflags */
5055 memalloc_noreclaim_restore(noreclaim_flag);
5058 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5060 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5061 struct bpf_prog *new = xdp->prog;
5064 switch (xdp->command) {
5065 case XDP_SETUP_PROG:
5066 rcu_assign_pointer(dev->xdp_prog, new);
5071 static_branch_dec(&generic_xdp_needed_key);
5072 } else if (new && !old) {
5073 static_branch_inc(&generic_xdp_needed_key);
5074 dev_disable_lro(dev);
5075 dev_disable_gro_hw(dev);
5079 case XDP_QUERY_PROG:
5080 xdp->prog_id = old ? old->aux->id : 0;
5091 static int netif_receive_skb_internal(struct sk_buff *skb)
5095 net_timestamp_check(netdev_tstamp_prequeue, skb);
5097 if (skb_defer_rx_timestamp(skb))
5098 return NET_RX_SUCCESS;
5100 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5105 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5109 if (ret != XDP_PASS)
5115 if (static_key_false(&rps_needed)) {
5116 struct rps_dev_flow voidflow, *rflow = &voidflow;
5117 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5120 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5126 ret = __netif_receive_skb(skb);
5131 static void netif_receive_skb_list_internal(struct list_head *head)
5133 struct bpf_prog *xdp_prog = NULL;
5134 struct sk_buff *skb, *next;
5135 struct list_head sublist;
5137 INIT_LIST_HEAD(&sublist);
5138 list_for_each_entry_safe(skb, next, head, list) {
5139 net_timestamp_check(netdev_tstamp_prequeue, skb);
5140 list_del(&skb->list);
5141 if (!skb_defer_rx_timestamp(skb))
5142 list_add_tail(&skb->list, &sublist);
5144 list_splice_init(&sublist, head);
5146 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5149 list_for_each_entry_safe(skb, next, head, list) {
5150 xdp_prog = rcu_dereference(skb->dev->xdp_prog);
5151 list_del(&skb->list);
5152 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS)
5153 list_add_tail(&skb->list, &sublist);
5157 /* Put passed packets back on main list */
5158 list_splice_init(&sublist, head);
5163 if (static_key_false(&rps_needed)) {
5164 list_for_each_entry_safe(skb, next, head, list) {
5165 struct rps_dev_flow voidflow, *rflow = &voidflow;
5166 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5169 /* Will be handled, remove from list */
5170 list_del(&skb->list);
5171 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5176 __netif_receive_skb_list(head);
5181 * netif_receive_skb - process receive buffer from network
5182 * @skb: buffer to process
5184 * netif_receive_skb() is the main receive data processing function.
5185 * It always succeeds. The buffer may be dropped during processing
5186 * for congestion control or by the protocol layers.
5188 * This function may only be called from softirq context and interrupts
5189 * should be enabled.
5191 * Return values (usually ignored):
5192 * NET_RX_SUCCESS: no congestion
5193 * NET_RX_DROP: packet was dropped
5195 int netif_receive_skb(struct sk_buff *skb)
5197 trace_netif_receive_skb_entry(skb);
5199 return netif_receive_skb_internal(skb);
5201 EXPORT_SYMBOL(netif_receive_skb);
5204 * netif_receive_skb_list - process many receive buffers from network
5205 * @head: list of skbs to process.
5207 * Since return value of netif_receive_skb() is normally ignored, and
5208 * wouldn't be meaningful for a list, this function returns void.
5210 * This function may only be called from softirq context and interrupts
5211 * should be enabled.
5213 void netif_receive_skb_list(struct list_head *head)
5215 struct sk_buff *skb;
5217 if (list_empty(head))
5219 list_for_each_entry(skb, head, list)
5220 trace_netif_receive_skb_list_entry(skb);
5221 netif_receive_skb_list_internal(head);
5223 EXPORT_SYMBOL(netif_receive_skb_list);
5225 DEFINE_PER_CPU(struct work_struct, flush_works);
5227 /* Network device is going away, flush any packets still pending */
5228 static void flush_backlog(struct work_struct *work)
5230 struct sk_buff *skb, *tmp;
5231 struct softnet_data *sd;
5234 sd = this_cpu_ptr(&softnet_data);
5236 local_irq_disable();
5238 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5239 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5240 __skb_unlink(skb, &sd->input_pkt_queue);
5242 input_queue_head_incr(sd);
5248 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5249 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5250 __skb_unlink(skb, &sd->process_queue);
5252 input_queue_head_incr(sd);
5258 static void flush_all_backlogs(void)
5264 for_each_online_cpu(cpu)
5265 queue_work_on(cpu, system_highpri_wq,
5266 per_cpu_ptr(&flush_works, cpu));
5268 for_each_online_cpu(cpu)
5269 flush_work(per_cpu_ptr(&flush_works, cpu));
5274 static int napi_gro_complete(struct sk_buff *skb)
5276 struct packet_offload *ptype;
5277 __be16 type = skb->protocol;
5278 struct list_head *head = &offload_base;
5281 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5283 if (NAPI_GRO_CB(skb)->count == 1) {
5284 skb_shinfo(skb)->gso_size = 0;
5289 list_for_each_entry_rcu(ptype, head, list) {
5290 if (ptype->type != type || !ptype->callbacks.gro_complete)
5293 err = ptype->callbacks.gro_complete(skb, 0);
5299 WARN_ON(&ptype->list == head);
5301 return NET_RX_SUCCESS;
5305 return netif_receive_skb_internal(skb);
5308 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5311 struct list_head *head = &napi->gro_hash[index].list;
5312 struct sk_buff *skb, *p;
5314 list_for_each_entry_safe_reverse(skb, p, head, list) {
5315 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5317 list_del(&skb->list);
5319 napi_gro_complete(skb);
5320 napi->gro_hash[index].count--;
5323 if (!napi->gro_hash[index].count)
5324 __clear_bit(index, &napi->gro_bitmask);
5327 /* napi->gro_hash[].list contains packets ordered by age.
5328 * youngest packets at the head of it.
5329 * Complete skbs in reverse order to reduce latencies.
5331 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5335 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
5336 if (test_bit(i, &napi->gro_bitmask))
5337 __napi_gro_flush_chain(napi, i, flush_old);
5340 EXPORT_SYMBOL(napi_gro_flush);
5342 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5343 struct sk_buff *skb)
5345 unsigned int maclen = skb->dev->hard_header_len;
5346 u32 hash = skb_get_hash_raw(skb);
5347 struct list_head *head;
5350 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5351 list_for_each_entry(p, head, list) {
5352 unsigned long diffs;
5354 NAPI_GRO_CB(p)->flush = 0;
5356 if (hash != skb_get_hash_raw(p)) {
5357 NAPI_GRO_CB(p)->same_flow = 0;
5361 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5362 diffs |= p->vlan_tci ^ skb->vlan_tci;
5363 diffs |= skb_metadata_dst_cmp(p, skb);
5364 diffs |= skb_metadata_differs(p, skb);
5365 if (maclen == ETH_HLEN)
5366 diffs |= compare_ether_header(skb_mac_header(p),
5367 skb_mac_header(skb));
5369 diffs = memcmp(skb_mac_header(p),
5370 skb_mac_header(skb),
5372 NAPI_GRO_CB(p)->same_flow = !diffs;
5378 static void skb_gro_reset_offset(struct sk_buff *skb)
5380 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5381 const skb_frag_t *frag0 = &pinfo->frags[0];
5383 NAPI_GRO_CB(skb)->data_offset = 0;
5384 NAPI_GRO_CB(skb)->frag0 = NULL;
5385 NAPI_GRO_CB(skb)->frag0_len = 0;
5387 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5389 !PageHighMem(skb_frag_page(frag0))) {
5390 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5391 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5392 skb_frag_size(frag0),
5393 skb->end - skb->tail);
5397 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5399 struct skb_shared_info *pinfo = skb_shinfo(skb);
5401 BUG_ON(skb->end - skb->tail < grow);
5403 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5405 skb->data_len -= grow;
5408 pinfo->frags[0].page_offset += grow;
5409 skb_frag_size_sub(&pinfo->frags[0], grow);
5411 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5412 skb_frag_unref(skb, 0);
5413 memmove(pinfo->frags, pinfo->frags + 1,
5414 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5418 static void gro_flush_oldest(struct list_head *head)
5420 struct sk_buff *oldest;
5422 oldest = list_last_entry(head, struct sk_buff, list);
5424 /* We are called with head length >= MAX_GRO_SKBS, so this is
5427 if (WARN_ON_ONCE(!oldest))
5430 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5433 list_del(&oldest->list);
5434 napi_gro_complete(oldest);
5437 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5439 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5440 struct list_head *head = &offload_base;
5441 struct packet_offload *ptype;
5442 __be16 type = skb->protocol;
5443 struct list_head *gro_head;
5444 struct sk_buff *pp = NULL;
5445 enum gro_result ret;
5449 if (netif_elide_gro(skb->dev))
5452 gro_head = gro_list_prepare(napi, skb);
5455 list_for_each_entry_rcu(ptype, head, list) {
5456 if (ptype->type != type || !ptype->callbacks.gro_receive)
5459 skb_set_network_header(skb, skb_gro_offset(skb));
5460 skb_reset_mac_len(skb);
5461 NAPI_GRO_CB(skb)->same_flow = 0;
5462 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5463 NAPI_GRO_CB(skb)->free = 0;
5464 NAPI_GRO_CB(skb)->encap_mark = 0;
5465 NAPI_GRO_CB(skb)->recursion_counter = 0;
5466 NAPI_GRO_CB(skb)->is_fou = 0;
5467 NAPI_GRO_CB(skb)->is_atomic = 1;
5468 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5470 /* Setup for GRO checksum validation */
5471 switch (skb->ip_summed) {
5472 case CHECKSUM_COMPLETE:
5473 NAPI_GRO_CB(skb)->csum = skb->csum;
5474 NAPI_GRO_CB(skb)->csum_valid = 1;
5475 NAPI_GRO_CB(skb)->csum_cnt = 0;
5477 case CHECKSUM_UNNECESSARY:
5478 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5479 NAPI_GRO_CB(skb)->csum_valid = 0;
5482 NAPI_GRO_CB(skb)->csum_cnt = 0;
5483 NAPI_GRO_CB(skb)->csum_valid = 0;
5486 pp = ptype->callbacks.gro_receive(gro_head, skb);
5491 if (&ptype->list == head)
5494 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5499 same_flow = NAPI_GRO_CB(skb)->same_flow;
5500 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5503 list_del(&pp->list);
5505 napi_gro_complete(pp);
5506 napi->gro_hash[hash].count--;
5512 if (NAPI_GRO_CB(skb)->flush)
5515 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5516 gro_flush_oldest(gro_head);
5518 napi->gro_hash[hash].count++;
5520 NAPI_GRO_CB(skb)->count = 1;
5521 NAPI_GRO_CB(skb)->age = jiffies;
5522 NAPI_GRO_CB(skb)->last = skb;
5523 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5524 list_add(&skb->list, gro_head);
5528 grow = skb_gro_offset(skb) - skb_headlen(skb);
5530 gro_pull_from_frag0(skb, grow);
5532 if (napi->gro_hash[hash].count) {
5533 if (!test_bit(hash, &napi->gro_bitmask))
5534 __set_bit(hash, &napi->gro_bitmask);
5535 } else if (test_bit(hash, &napi->gro_bitmask)) {
5536 __clear_bit(hash, &napi->gro_bitmask);
5546 struct packet_offload *gro_find_receive_by_type(__be16 type)
5548 struct list_head *offload_head = &offload_base;
5549 struct packet_offload *ptype;
5551 list_for_each_entry_rcu(ptype, offload_head, list) {
5552 if (ptype->type != type || !ptype->callbacks.gro_receive)
5558 EXPORT_SYMBOL(gro_find_receive_by_type);
5560 struct packet_offload *gro_find_complete_by_type(__be16 type)
5562 struct list_head *offload_head = &offload_base;
5563 struct packet_offload *ptype;
5565 list_for_each_entry_rcu(ptype, offload_head, list) {
5566 if (ptype->type != type || !ptype->callbacks.gro_complete)
5572 EXPORT_SYMBOL(gro_find_complete_by_type);
5574 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5578 kmem_cache_free(skbuff_head_cache, skb);
5581 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5585 if (netif_receive_skb_internal(skb))
5593 case GRO_MERGED_FREE:
5594 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5595 napi_skb_free_stolen_head(skb);
5609 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5611 skb_mark_napi_id(skb, napi);
5612 trace_napi_gro_receive_entry(skb);
5614 skb_gro_reset_offset(skb);
5616 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
5618 EXPORT_SYMBOL(napi_gro_receive);
5620 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5622 if (unlikely(skb->pfmemalloc)) {
5626 __skb_pull(skb, skb_headlen(skb));
5627 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5628 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5630 skb->dev = napi->dev;
5632 skb->encapsulation = 0;
5633 skb_shinfo(skb)->gso_type = 0;
5634 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5640 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5642 struct sk_buff *skb = napi->skb;
5645 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5648 skb_mark_napi_id(skb, napi);
5653 EXPORT_SYMBOL(napi_get_frags);
5655 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5656 struct sk_buff *skb,
5662 __skb_push(skb, ETH_HLEN);
5663 skb->protocol = eth_type_trans(skb, skb->dev);
5664 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5669 napi_reuse_skb(napi, skb);
5672 case GRO_MERGED_FREE:
5673 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5674 napi_skb_free_stolen_head(skb);
5676 napi_reuse_skb(napi, skb);
5687 /* Upper GRO stack assumes network header starts at gro_offset=0
5688 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5689 * We copy ethernet header into skb->data to have a common layout.
5691 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5693 struct sk_buff *skb = napi->skb;
5694 const struct ethhdr *eth;
5695 unsigned int hlen = sizeof(*eth);
5699 skb_reset_mac_header(skb);
5700 skb_gro_reset_offset(skb);
5702 eth = skb_gro_header_fast(skb, 0);
5703 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5704 eth = skb_gro_header_slow(skb, hlen, 0);
5705 if (unlikely(!eth)) {
5706 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5707 __func__, napi->dev->name);
5708 napi_reuse_skb(napi, skb);
5712 gro_pull_from_frag0(skb, hlen);
5713 NAPI_GRO_CB(skb)->frag0 += hlen;
5714 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5716 __skb_pull(skb, hlen);
5719 * This works because the only protocols we care about don't require
5721 * We'll fix it up properly in napi_frags_finish()
5723 skb->protocol = eth->h_proto;
5728 gro_result_t napi_gro_frags(struct napi_struct *napi)
5730 struct sk_buff *skb = napi_frags_skb(napi);
5735 trace_napi_gro_frags_entry(skb);
5737 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5739 EXPORT_SYMBOL(napi_gro_frags);
5741 /* Compute the checksum from gro_offset and return the folded value
5742 * after adding in any pseudo checksum.
5744 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5749 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5751 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5752 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5754 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5755 !skb->csum_complete_sw)
5756 netdev_rx_csum_fault(skb->dev);
5759 NAPI_GRO_CB(skb)->csum = wsum;
5760 NAPI_GRO_CB(skb)->csum_valid = 1;
5764 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5766 static void net_rps_send_ipi(struct softnet_data *remsd)
5770 struct softnet_data *next = remsd->rps_ipi_next;
5772 if (cpu_online(remsd->cpu))
5773 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5780 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5781 * Note: called with local irq disabled, but exits with local irq enabled.
5783 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5786 struct softnet_data *remsd = sd->rps_ipi_list;
5789 sd->rps_ipi_list = NULL;
5793 /* Send pending IPI's to kick RPS processing on remote cpus. */
5794 net_rps_send_ipi(remsd);
5800 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5803 return sd->rps_ipi_list != NULL;
5809 static int process_backlog(struct napi_struct *napi, int quota)
5811 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5815 /* Check if we have pending ipi, its better to send them now,
5816 * not waiting net_rx_action() end.
5818 if (sd_has_rps_ipi_waiting(sd)) {
5819 local_irq_disable();
5820 net_rps_action_and_irq_enable(sd);
5823 napi->weight = dev_rx_weight;
5825 struct sk_buff *skb;
5827 while ((skb = __skb_dequeue(&sd->process_queue))) {
5829 __netif_receive_skb(skb);
5831 input_queue_head_incr(sd);
5832 if (++work >= quota)
5837 local_irq_disable();
5839 if (skb_queue_empty(&sd->input_pkt_queue)) {
5841 * Inline a custom version of __napi_complete().
5842 * only current cpu owns and manipulates this napi,
5843 * and NAPI_STATE_SCHED is the only possible flag set
5845 * We can use a plain write instead of clear_bit(),
5846 * and we dont need an smp_mb() memory barrier.
5851 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5852 &sd->process_queue);
5862 * __napi_schedule - schedule for receive
5863 * @n: entry to schedule
5865 * The entry's receive function will be scheduled to run.
5866 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5868 void __napi_schedule(struct napi_struct *n)
5870 unsigned long flags;
5872 local_irq_save(flags);
5873 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5874 local_irq_restore(flags);
5876 EXPORT_SYMBOL(__napi_schedule);
5879 * napi_schedule_prep - check if napi can be scheduled
5882 * Test if NAPI routine is already running, and if not mark
5883 * it as running. This is used as a condition variable
5884 * insure only one NAPI poll instance runs. We also make
5885 * sure there is no pending NAPI disable.
5887 bool napi_schedule_prep(struct napi_struct *n)
5889 unsigned long val, new;
5892 val = READ_ONCE(n->state);
5893 if (unlikely(val & NAPIF_STATE_DISABLE))
5895 new = val | NAPIF_STATE_SCHED;
5897 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5898 * This was suggested by Alexander Duyck, as compiler
5899 * emits better code than :
5900 * if (val & NAPIF_STATE_SCHED)
5901 * new |= NAPIF_STATE_MISSED;
5903 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5905 } while (cmpxchg(&n->state, val, new) != val);
5907 return !(val & NAPIF_STATE_SCHED);
5909 EXPORT_SYMBOL(napi_schedule_prep);
5912 * __napi_schedule_irqoff - schedule for receive
5913 * @n: entry to schedule
5915 * Variant of __napi_schedule() assuming hard irqs are masked
5917 void __napi_schedule_irqoff(struct napi_struct *n)
5919 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5921 EXPORT_SYMBOL(__napi_schedule_irqoff);
5923 bool napi_complete_done(struct napi_struct *n, int work_done)
5925 unsigned long flags, val, new;
5928 * 1) Don't let napi dequeue from the cpu poll list
5929 * just in case its running on a different cpu.
5930 * 2) If we are busy polling, do nothing here, we have
5931 * the guarantee we will be called later.
5933 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5934 NAPIF_STATE_IN_BUSY_POLL)))
5937 if (n->gro_bitmask) {
5938 unsigned long timeout = 0;
5941 timeout = n->dev->gro_flush_timeout;
5944 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5945 HRTIMER_MODE_REL_PINNED);
5947 napi_gro_flush(n, false);
5949 if (unlikely(!list_empty(&n->poll_list))) {
5950 /* If n->poll_list is not empty, we need to mask irqs */
5951 local_irq_save(flags);
5952 list_del_init(&n->poll_list);
5953 local_irq_restore(flags);
5957 val = READ_ONCE(n->state);
5959 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5961 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
5963 /* If STATE_MISSED was set, leave STATE_SCHED set,
5964 * because we will call napi->poll() one more time.
5965 * This C code was suggested by Alexander Duyck to help gcc.
5967 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5969 } while (cmpxchg(&n->state, val, new) != val);
5971 if (unlikely(val & NAPIF_STATE_MISSED)) {
5978 EXPORT_SYMBOL(napi_complete_done);
5980 /* must be called under rcu_read_lock(), as we dont take a reference */
5981 static struct napi_struct *napi_by_id(unsigned int napi_id)
5983 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5984 struct napi_struct *napi;
5986 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5987 if (napi->napi_id == napi_id)
5993 #if defined(CONFIG_NET_RX_BUSY_POLL)
5995 #define BUSY_POLL_BUDGET 8
5997 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6001 /* Busy polling means there is a high chance device driver hard irq
6002 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6003 * set in napi_schedule_prep().
6004 * Since we are about to call napi->poll() once more, we can safely
6005 * clear NAPI_STATE_MISSED.
6007 * Note: x86 could use a single "lock and ..." instruction
6008 * to perform these two clear_bit()
6010 clear_bit(NAPI_STATE_MISSED, &napi->state);
6011 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6015 /* All we really want here is to re-enable device interrupts.
6016 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6018 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6019 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6020 netpoll_poll_unlock(have_poll_lock);
6021 if (rc == BUSY_POLL_BUDGET)
6022 __napi_schedule(napi);
6026 void napi_busy_loop(unsigned int napi_id,
6027 bool (*loop_end)(void *, unsigned long),
6030 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6031 int (*napi_poll)(struct napi_struct *napi, int budget);
6032 void *have_poll_lock = NULL;
6033 struct napi_struct *napi;
6040 napi = napi_by_id(napi_id);
6050 unsigned long val = READ_ONCE(napi->state);
6052 /* If multiple threads are competing for this napi,
6053 * we avoid dirtying napi->state as much as we can.
6055 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6056 NAPIF_STATE_IN_BUSY_POLL))
6058 if (cmpxchg(&napi->state, val,
6059 val | NAPIF_STATE_IN_BUSY_POLL |
6060 NAPIF_STATE_SCHED) != val)
6062 have_poll_lock = netpoll_poll_lock(napi);
6063 napi_poll = napi->poll;
6065 work = napi_poll(napi, BUSY_POLL_BUDGET);
6066 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6069 __NET_ADD_STATS(dev_net(napi->dev),
6070 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6073 if (!loop_end || loop_end(loop_end_arg, start_time))
6076 if (unlikely(need_resched())) {
6078 busy_poll_stop(napi, have_poll_lock);
6082 if (loop_end(loop_end_arg, start_time))
6089 busy_poll_stop(napi, have_poll_lock);
6094 EXPORT_SYMBOL(napi_busy_loop);
6096 #endif /* CONFIG_NET_RX_BUSY_POLL */
6098 static void napi_hash_add(struct napi_struct *napi)
6100 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6101 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6104 spin_lock(&napi_hash_lock);
6106 /* 0..NR_CPUS range is reserved for sender_cpu use */
6108 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6109 napi_gen_id = MIN_NAPI_ID;
6110 } while (napi_by_id(napi_gen_id));
6111 napi->napi_id = napi_gen_id;
6113 hlist_add_head_rcu(&napi->napi_hash_node,
6114 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6116 spin_unlock(&napi_hash_lock);
6119 /* Warning : caller is responsible to make sure rcu grace period
6120 * is respected before freeing memory containing @napi
6122 bool napi_hash_del(struct napi_struct *napi)
6124 bool rcu_sync_needed = false;
6126 spin_lock(&napi_hash_lock);
6128 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6129 rcu_sync_needed = true;
6130 hlist_del_rcu(&napi->napi_hash_node);
6132 spin_unlock(&napi_hash_lock);
6133 return rcu_sync_needed;
6135 EXPORT_SYMBOL_GPL(napi_hash_del);
6137 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6139 struct napi_struct *napi;
6141 napi = container_of(timer, struct napi_struct, timer);
6143 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6144 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6146 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6147 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6148 __napi_schedule_irqoff(napi);
6150 return HRTIMER_NORESTART;
6153 static void init_gro_hash(struct napi_struct *napi)
6157 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6158 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6159 napi->gro_hash[i].count = 0;
6161 napi->gro_bitmask = 0;
6164 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6165 int (*poll)(struct napi_struct *, int), int weight)
6167 INIT_LIST_HEAD(&napi->poll_list);
6168 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6169 napi->timer.function = napi_watchdog;
6170 init_gro_hash(napi);
6173 if (weight > NAPI_POLL_WEIGHT)
6174 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
6176 napi->weight = weight;
6177 list_add(&napi->dev_list, &dev->napi_list);
6179 #ifdef CONFIG_NETPOLL
6180 napi->poll_owner = -1;
6182 set_bit(NAPI_STATE_SCHED, &napi->state);
6183 napi_hash_add(napi);
6185 EXPORT_SYMBOL(netif_napi_add);
6187 void napi_disable(struct napi_struct *n)
6190 set_bit(NAPI_STATE_DISABLE, &n->state);
6192 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6194 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6197 hrtimer_cancel(&n->timer);
6199 clear_bit(NAPI_STATE_DISABLE, &n->state);
6201 EXPORT_SYMBOL(napi_disable);
6203 static void flush_gro_hash(struct napi_struct *napi)
6207 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6208 struct sk_buff *skb, *n;
6210 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6212 napi->gro_hash[i].count = 0;
6216 /* Must be called in process context */
6217 void netif_napi_del(struct napi_struct *napi)
6220 if (napi_hash_del(napi))
6222 list_del_init(&napi->dev_list);
6223 napi_free_frags(napi);
6225 flush_gro_hash(napi);
6226 napi->gro_bitmask = 0;
6228 EXPORT_SYMBOL(netif_napi_del);
6230 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6235 list_del_init(&n->poll_list);
6237 have = netpoll_poll_lock(n);
6241 /* This NAPI_STATE_SCHED test is for avoiding a race
6242 * with netpoll's poll_napi(). Only the entity which
6243 * obtains the lock and sees NAPI_STATE_SCHED set will
6244 * actually make the ->poll() call. Therefore we avoid
6245 * accidentally calling ->poll() when NAPI is not scheduled.
6248 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6249 work = n->poll(n, weight);
6250 trace_napi_poll(n, work, weight);
6253 WARN_ON_ONCE(work > weight);
6255 if (likely(work < weight))
6258 /* Drivers must not modify the NAPI state if they
6259 * consume the entire weight. In such cases this code
6260 * still "owns" the NAPI instance and therefore can
6261 * move the instance around on the list at-will.
6263 if (unlikely(napi_disable_pending(n))) {
6268 if (n->gro_bitmask) {
6269 /* flush too old packets
6270 * If HZ < 1000, flush all packets.
6272 napi_gro_flush(n, HZ >= 1000);
6275 /* Some drivers may have called napi_schedule
6276 * prior to exhausting their budget.
6278 if (unlikely(!list_empty(&n->poll_list))) {
6279 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6280 n->dev ? n->dev->name : "backlog");
6284 list_add_tail(&n->poll_list, repoll);
6287 netpoll_poll_unlock(have);
6292 static __latent_entropy void net_rx_action(struct softirq_action *h)
6294 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6295 unsigned long time_limit = jiffies +
6296 usecs_to_jiffies(netdev_budget_usecs);
6297 int budget = netdev_budget;
6301 local_irq_disable();
6302 list_splice_init(&sd->poll_list, &list);
6306 struct napi_struct *n;
6308 if (list_empty(&list)) {
6309 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6314 n = list_first_entry(&list, struct napi_struct, poll_list);
6315 budget -= napi_poll(n, &repoll);
6317 /* If softirq window is exhausted then punt.
6318 * Allow this to run for 2 jiffies since which will allow
6319 * an average latency of 1.5/HZ.
6321 if (unlikely(budget <= 0 ||
6322 time_after_eq(jiffies, time_limit))) {
6328 local_irq_disable();
6330 list_splice_tail_init(&sd->poll_list, &list);
6331 list_splice_tail(&repoll, &list);
6332 list_splice(&list, &sd->poll_list);
6333 if (!list_empty(&sd->poll_list))
6334 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6336 net_rps_action_and_irq_enable(sd);
6338 __kfree_skb_flush();
6341 struct netdev_adjacent {
6342 struct net_device *dev;
6344 /* upper master flag, there can only be one master device per list */
6347 /* counter for the number of times this device was added to us */
6350 /* private field for the users */
6353 struct list_head list;
6354 struct rcu_head rcu;
6357 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6358 struct list_head *adj_list)
6360 struct netdev_adjacent *adj;
6362 list_for_each_entry(adj, adj_list, list) {
6363 if (adj->dev == adj_dev)
6369 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6371 struct net_device *dev = data;
6373 return upper_dev == dev;
6377 * netdev_has_upper_dev - Check if device is linked to an upper device
6379 * @upper_dev: upper device to check
6381 * Find out if a device is linked to specified upper device and return true
6382 * in case it is. Note that this checks only immediate upper device,
6383 * not through a complete stack of devices. The caller must hold the RTNL lock.
6385 bool netdev_has_upper_dev(struct net_device *dev,
6386 struct net_device *upper_dev)
6390 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6393 EXPORT_SYMBOL(netdev_has_upper_dev);
6396 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6398 * @upper_dev: upper device to check
6400 * Find out if a device is linked to specified upper device and return true
6401 * in case it is. Note that this checks the entire upper device chain.
6402 * The caller must hold rcu lock.
6405 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6406 struct net_device *upper_dev)
6408 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6411 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6414 * netdev_has_any_upper_dev - Check if device is linked to some device
6417 * Find out if a device is linked to an upper device and return true in case
6418 * it is. The caller must hold the RTNL lock.
6420 bool netdev_has_any_upper_dev(struct net_device *dev)
6424 return !list_empty(&dev->adj_list.upper);
6426 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6429 * netdev_master_upper_dev_get - Get master upper device
6432 * Find a master upper device and return pointer to it or NULL in case
6433 * it's not there. The caller must hold the RTNL lock.
6435 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6437 struct netdev_adjacent *upper;
6441 if (list_empty(&dev->adj_list.upper))
6444 upper = list_first_entry(&dev->adj_list.upper,
6445 struct netdev_adjacent, list);
6446 if (likely(upper->master))
6450 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6453 * netdev_has_any_lower_dev - Check if device is linked to some device
6456 * Find out if a device is linked to a lower device and return true in case
6457 * it is. The caller must hold the RTNL lock.
6459 static bool netdev_has_any_lower_dev(struct net_device *dev)
6463 return !list_empty(&dev->adj_list.lower);
6466 void *netdev_adjacent_get_private(struct list_head *adj_list)
6468 struct netdev_adjacent *adj;
6470 adj = list_entry(adj_list, struct netdev_adjacent, list);
6472 return adj->private;
6474 EXPORT_SYMBOL(netdev_adjacent_get_private);
6477 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6479 * @iter: list_head ** of the current position
6481 * Gets the next device from the dev's upper list, starting from iter
6482 * position. The caller must hold RCU read lock.
6484 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6485 struct list_head **iter)
6487 struct netdev_adjacent *upper;
6489 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6491 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6493 if (&upper->list == &dev->adj_list.upper)
6496 *iter = &upper->list;
6500 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6502 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6503 struct list_head **iter)
6505 struct netdev_adjacent *upper;
6507 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6509 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6511 if (&upper->list == &dev->adj_list.upper)
6514 *iter = &upper->list;
6519 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6520 int (*fn)(struct net_device *dev,
6524 struct net_device *udev;
6525 struct list_head *iter;
6528 for (iter = &dev->adj_list.upper,
6529 udev = netdev_next_upper_dev_rcu(dev, &iter);
6531 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6532 /* first is the upper device itself */
6533 ret = fn(udev, data);
6537 /* then look at all of its upper devices */
6538 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6545 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6548 * netdev_lower_get_next_private - Get the next ->private from the
6549 * lower neighbour list
6551 * @iter: list_head ** of the current position
6553 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6554 * list, starting from iter position. The caller must hold either hold the
6555 * RTNL lock or its own locking that guarantees that the neighbour lower
6556 * list will remain unchanged.
6558 void *netdev_lower_get_next_private(struct net_device *dev,
6559 struct list_head **iter)
6561 struct netdev_adjacent *lower;
6563 lower = list_entry(*iter, struct netdev_adjacent, list);
6565 if (&lower->list == &dev->adj_list.lower)
6568 *iter = lower->list.next;
6570 return lower->private;
6572 EXPORT_SYMBOL(netdev_lower_get_next_private);
6575 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6576 * lower neighbour list, RCU
6579 * @iter: list_head ** of the current position
6581 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6582 * list, starting from iter position. The caller must hold RCU read lock.
6584 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6585 struct list_head **iter)
6587 struct netdev_adjacent *lower;
6589 WARN_ON_ONCE(!rcu_read_lock_held());
6591 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6593 if (&lower->list == &dev->adj_list.lower)
6596 *iter = &lower->list;
6598 return lower->private;
6600 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6603 * netdev_lower_get_next - Get the next device from the lower neighbour
6606 * @iter: list_head ** of the current position
6608 * Gets the next netdev_adjacent from the dev's lower neighbour
6609 * list, starting from iter position. The caller must hold RTNL lock or
6610 * its own locking that guarantees that the neighbour lower
6611 * list will remain unchanged.
6613 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6615 struct netdev_adjacent *lower;
6617 lower = list_entry(*iter, struct netdev_adjacent, list);
6619 if (&lower->list == &dev->adj_list.lower)
6622 *iter = lower->list.next;
6626 EXPORT_SYMBOL(netdev_lower_get_next);
6628 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6629 struct list_head **iter)
6631 struct netdev_adjacent *lower;
6633 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6635 if (&lower->list == &dev->adj_list.lower)
6638 *iter = &lower->list;
6643 int netdev_walk_all_lower_dev(struct net_device *dev,
6644 int (*fn)(struct net_device *dev,
6648 struct net_device *ldev;
6649 struct list_head *iter;
6652 for (iter = &dev->adj_list.lower,
6653 ldev = netdev_next_lower_dev(dev, &iter);
6655 ldev = netdev_next_lower_dev(dev, &iter)) {
6656 /* first is the lower device itself */
6657 ret = fn(ldev, data);
6661 /* then look at all of its lower devices */
6662 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6669 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6671 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6672 struct list_head **iter)
6674 struct netdev_adjacent *lower;
6676 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6677 if (&lower->list == &dev->adj_list.lower)
6680 *iter = &lower->list;
6685 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6686 int (*fn)(struct net_device *dev,
6690 struct net_device *ldev;
6691 struct list_head *iter;
6694 for (iter = &dev->adj_list.lower,
6695 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6697 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6698 /* first is the lower device itself */
6699 ret = fn(ldev, data);
6703 /* then look at all of its lower devices */
6704 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6711 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6714 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6715 * lower neighbour list, RCU
6719 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6720 * list. The caller must hold RCU read lock.
6722 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6724 struct netdev_adjacent *lower;
6726 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6727 struct netdev_adjacent, list);
6729 return lower->private;
6732 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6735 * netdev_master_upper_dev_get_rcu - Get master upper device
6738 * Find a master upper device and return pointer to it or NULL in case
6739 * it's not there. The caller must hold the RCU read lock.
6741 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6743 struct netdev_adjacent *upper;
6745 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6746 struct netdev_adjacent, list);
6747 if (upper && likely(upper->master))
6751 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6753 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6754 struct net_device *adj_dev,
6755 struct list_head *dev_list)
6757 char linkname[IFNAMSIZ+7];
6759 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6760 "upper_%s" : "lower_%s", adj_dev->name);
6761 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6764 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6766 struct list_head *dev_list)
6768 char linkname[IFNAMSIZ+7];
6770 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6771 "upper_%s" : "lower_%s", name);
6772 sysfs_remove_link(&(dev->dev.kobj), linkname);
6775 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6776 struct net_device *adj_dev,
6777 struct list_head *dev_list)
6779 return (dev_list == &dev->adj_list.upper ||
6780 dev_list == &dev->adj_list.lower) &&
6781 net_eq(dev_net(dev), dev_net(adj_dev));
6784 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6785 struct net_device *adj_dev,
6786 struct list_head *dev_list,
6787 void *private, bool master)
6789 struct netdev_adjacent *adj;
6792 adj = __netdev_find_adj(adj_dev, dev_list);
6796 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6797 dev->name, adj_dev->name, adj->ref_nr);
6802 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6807 adj->master = master;
6809 adj->private = private;
6812 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6813 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6815 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6816 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6821 /* Ensure that master link is always the first item in list. */
6823 ret = sysfs_create_link(&(dev->dev.kobj),
6824 &(adj_dev->dev.kobj), "master");
6826 goto remove_symlinks;
6828 list_add_rcu(&adj->list, dev_list);
6830 list_add_tail_rcu(&adj->list, dev_list);
6836 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6837 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6845 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6846 struct net_device *adj_dev,
6848 struct list_head *dev_list)
6850 struct netdev_adjacent *adj;
6852 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6853 dev->name, adj_dev->name, ref_nr);
6855 adj = __netdev_find_adj(adj_dev, dev_list);
6858 pr_err("Adjacency does not exist for device %s from %s\n",
6859 dev->name, adj_dev->name);
6864 if (adj->ref_nr > ref_nr) {
6865 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6866 dev->name, adj_dev->name, ref_nr,
6867 adj->ref_nr - ref_nr);
6868 adj->ref_nr -= ref_nr;
6873 sysfs_remove_link(&(dev->dev.kobj), "master");
6875 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6876 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6878 list_del_rcu(&adj->list);
6879 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6880 adj_dev->name, dev->name, adj_dev->name);
6882 kfree_rcu(adj, rcu);
6885 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6886 struct net_device *upper_dev,
6887 struct list_head *up_list,
6888 struct list_head *down_list,
6889 void *private, bool master)
6893 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6898 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6901 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6908 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6909 struct net_device *upper_dev,
6911 struct list_head *up_list,
6912 struct list_head *down_list)
6914 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6915 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6918 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6919 struct net_device *upper_dev,
6920 void *private, bool master)
6922 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6923 &dev->adj_list.upper,
6924 &upper_dev->adj_list.lower,
6928 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6929 struct net_device *upper_dev)
6931 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
6932 &dev->adj_list.upper,
6933 &upper_dev->adj_list.lower);
6936 static int __netdev_upper_dev_link(struct net_device *dev,
6937 struct net_device *upper_dev, bool master,
6938 void *upper_priv, void *upper_info,
6939 struct netlink_ext_ack *extack)
6941 struct netdev_notifier_changeupper_info changeupper_info = {
6946 .upper_dev = upper_dev,
6949 .upper_info = upper_info,
6951 struct net_device *master_dev;
6956 if (dev == upper_dev)
6959 /* To prevent loops, check if dev is not upper device to upper_dev. */
6960 if (netdev_has_upper_dev(upper_dev, dev))
6964 if (netdev_has_upper_dev(dev, upper_dev))
6967 master_dev = netdev_master_upper_dev_get(dev);
6969 return master_dev == upper_dev ? -EEXIST : -EBUSY;
6972 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
6973 &changeupper_info.info);
6974 ret = notifier_to_errno(ret);
6978 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
6983 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
6984 &changeupper_info.info);
6985 ret = notifier_to_errno(ret);
6992 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6998 * netdev_upper_dev_link - Add a link to the upper device
7000 * @upper_dev: new upper device
7001 * @extack: netlink extended ack
7003 * Adds a link to device which is upper to this one. The caller must hold
7004 * the RTNL lock. On a failure a negative errno code is returned.
7005 * On success the reference counts are adjusted and the function
7008 int netdev_upper_dev_link(struct net_device *dev,
7009 struct net_device *upper_dev,
7010 struct netlink_ext_ack *extack)
7012 return __netdev_upper_dev_link(dev, upper_dev, false,
7013 NULL, NULL, extack);
7015 EXPORT_SYMBOL(netdev_upper_dev_link);
7018 * netdev_master_upper_dev_link - Add a master link to the upper device
7020 * @upper_dev: new upper device
7021 * @upper_priv: upper device private
7022 * @upper_info: upper info to be passed down via notifier
7023 * @extack: netlink extended ack
7025 * Adds a link to device which is upper to this one. In this case, only
7026 * one master upper device can be linked, although other non-master devices
7027 * might be linked as well. The caller must hold the RTNL lock.
7028 * On a failure a negative errno code is returned. On success the reference
7029 * counts are adjusted and the function returns zero.
7031 int netdev_master_upper_dev_link(struct net_device *dev,
7032 struct net_device *upper_dev,
7033 void *upper_priv, void *upper_info,
7034 struct netlink_ext_ack *extack)
7036 return __netdev_upper_dev_link(dev, upper_dev, true,
7037 upper_priv, upper_info, extack);
7039 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7042 * netdev_upper_dev_unlink - Removes a link to upper device
7044 * @upper_dev: new upper device
7046 * Removes a link to device which is upper to this one. The caller must hold
7049 void netdev_upper_dev_unlink(struct net_device *dev,
7050 struct net_device *upper_dev)
7052 struct netdev_notifier_changeupper_info changeupper_info = {
7056 .upper_dev = upper_dev,
7062 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7064 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7065 &changeupper_info.info);
7067 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7069 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7070 &changeupper_info.info);
7072 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7075 * netdev_bonding_info_change - Dispatch event about slave change
7077 * @bonding_info: info to dispatch
7079 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7080 * The caller must hold the RTNL lock.
7082 void netdev_bonding_info_change(struct net_device *dev,
7083 struct netdev_bonding_info *bonding_info)
7085 struct netdev_notifier_bonding_info info = {
7089 memcpy(&info.bonding_info, bonding_info,
7090 sizeof(struct netdev_bonding_info));
7091 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7094 EXPORT_SYMBOL(netdev_bonding_info_change);
7096 static void netdev_adjacent_add_links(struct net_device *dev)
7098 struct netdev_adjacent *iter;
7100 struct net *net = dev_net(dev);
7102 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7103 if (!net_eq(net, dev_net(iter->dev)))
7105 netdev_adjacent_sysfs_add(iter->dev, dev,
7106 &iter->dev->adj_list.lower);
7107 netdev_adjacent_sysfs_add(dev, iter->dev,
7108 &dev->adj_list.upper);
7111 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7112 if (!net_eq(net, dev_net(iter->dev)))
7114 netdev_adjacent_sysfs_add(iter->dev, dev,
7115 &iter->dev->adj_list.upper);
7116 netdev_adjacent_sysfs_add(dev, iter->dev,
7117 &dev->adj_list.lower);
7121 static void netdev_adjacent_del_links(struct net_device *dev)
7123 struct netdev_adjacent *iter;
7125 struct net *net = dev_net(dev);
7127 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7128 if (!net_eq(net, dev_net(iter->dev)))
7130 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7131 &iter->dev->adj_list.lower);
7132 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7133 &dev->adj_list.upper);
7136 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7137 if (!net_eq(net, dev_net(iter->dev)))
7139 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7140 &iter->dev->adj_list.upper);
7141 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7142 &dev->adj_list.lower);
7146 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7148 struct netdev_adjacent *iter;
7150 struct net *net = dev_net(dev);
7152 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7153 if (!net_eq(net, dev_net(iter->dev)))
7155 netdev_adjacent_sysfs_del(iter->dev, oldname,
7156 &iter->dev->adj_list.lower);
7157 netdev_adjacent_sysfs_add(iter->dev, dev,
7158 &iter->dev->adj_list.lower);
7161 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7162 if (!net_eq(net, dev_net(iter->dev)))
7164 netdev_adjacent_sysfs_del(iter->dev, oldname,
7165 &iter->dev->adj_list.upper);
7166 netdev_adjacent_sysfs_add(iter->dev, dev,
7167 &iter->dev->adj_list.upper);
7171 void *netdev_lower_dev_get_private(struct net_device *dev,
7172 struct net_device *lower_dev)
7174 struct netdev_adjacent *lower;
7178 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7182 return lower->private;
7184 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7187 int dev_get_nest_level(struct net_device *dev)
7189 struct net_device *lower = NULL;
7190 struct list_head *iter;
7196 netdev_for_each_lower_dev(dev, lower, iter) {
7197 nest = dev_get_nest_level(lower);
7198 if (max_nest < nest)
7202 return max_nest + 1;
7204 EXPORT_SYMBOL(dev_get_nest_level);
7207 * netdev_lower_change - Dispatch event about lower device state change
7208 * @lower_dev: device
7209 * @lower_state_info: state to dispatch
7211 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7212 * The caller must hold the RTNL lock.
7214 void netdev_lower_state_changed(struct net_device *lower_dev,
7215 void *lower_state_info)
7217 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7218 .info.dev = lower_dev,
7222 changelowerstate_info.lower_state_info = lower_state_info;
7223 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7224 &changelowerstate_info.info);
7226 EXPORT_SYMBOL(netdev_lower_state_changed);
7228 static void dev_change_rx_flags(struct net_device *dev, int flags)
7230 const struct net_device_ops *ops = dev->netdev_ops;
7232 if (ops->ndo_change_rx_flags)
7233 ops->ndo_change_rx_flags(dev, flags);
7236 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7238 unsigned int old_flags = dev->flags;
7244 dev->flags |= IFF_PROMISC;
7245 dev->promiscuity += inc;
7246 if (dev->promiscuity == 0) {
7249 * If inc causes overflow, untouch promisc and return error.
7252 dev->flags &= ~IFF_PROMISC;
7254 dev->promiscuity -= inc;
7255 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7260 if (dev->flags != old_flags) {
7261 pr_info("device %s %s promiscuous mode\n",
7263 dev->flags & IFF_PROMISC ? "entered" : "left");
7264 if (audit_enabled) {
7265 current_uid_gid(&uid, &gid);
7266 audit_log(audit_context(), GFP_ATOMIC,
7267 AUDIT_ANOM_PROMISCUOUS,
7268 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7269 dev->name, (dev->flags & IFF_PROMISC),
7270 (old_flags & IFF_PROMISC),
7271 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7272 from_kuid(&init_user_ns, uid),
7273 from_kgid(&init_user_ns, gid),
7274 audit_get_sessionid(current));
7277 dev_change_rx_flags(dev, IFF_PROMISC);
7280 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7285 * dev_set_promiscuity - update promiscuity count on a device
7289 * Add or remove promiscuity from a device. While the count in the device
7290 * remains above zero the interface remains promiscuous. Once it hits zero
7291 * the device reverts back to normal filtering operation. A negative inc
7292 * value is used to drop promiscuity on the device.
7293 * Return 0 if successful or a negative errno code on error.
7295 int dev_set_promiscuity(struct net_device *dev, int inc)
7297 unsigned int old_flags = dev->flags;
7300 err = __dev_set_promiscuity(dev, inc, true);
7303 if (dev->flags != old_flags)
7304 dev_set_rx_mode(dev);
7307 EXPORT_SYMBOL(dev_set_promiscuity);
7309 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7311 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7315 dev->flags |= IFF_ALLMULTI;
7316 dev->allmulti += inc;
7317 if (dev->allmulti == 0) {
7320 * If inc causes overflow, untouch allmulti and return error.
7323 dev->flags &= ~IFF_ALLMULTI;
7325 dev->allmulti -= inc;
7326 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7331 if (dev->flags ^ old_flags) {
7332 dev_change_rx_flags(dev, IFF_ALLMULTI);
7333 dev_set_rx_mode(dev);
7335 __dev_notify_flags(dev, old_flags,
7336 dev->gflags ^ old_gflags);
7342 * dev_set_allmulti - update allmulti count on a device
7346 * Add or remove reception of all multicast frames to a device. While the
7347 * count in the device remains above zero the interface remains listening
7348 * to all interfaces. Once it hits zero the device reverts back to normal
7349 * filtering operation. A negative @inc value is used to drop the counter
7350 * when releasing a resource needing all multicasts.
7351 * Return 0 if successful or a negative errno code on error.
7354 int dev_set_allmulti(struct net_device *dev, int inc)
7356 return __dev_set_allmulti(dev, inc, true);
7358 EXPORT_SYMBOL(dev_set_allmulti);
7361 * Upload unicast and multicast address lists to device and
7362 * configure RX filtering. When the device doesn't support unicast
7363 * filtering it is put in promiscuous mode while unicast addresses
7366 void __dev_set_rx_mode(struct net_device *dev)
7368 const struct net_device_ops *ops = dev->netdev_ops;
7370 /* dev_open will call this function so the list will stay sane. */
7371 if (!(dev->flags&IFF_UP))
7374 if (!netif_device_present(dev))
7377 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7378 /* Unicast addresses changes may only happen under the rtnl,
7379 * therefore calling __dev_set_promiscuity here is safe.
7381 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7382 __dev_set_promiscuity(dev, 1, false);
7383 dev->uc_promisc = true;
7384 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7385 __dev_set_promiscuity(dev, -1, false);
7386 dev->uc_promisc = false;
7390 if (ops->ndo_set_rx_mode)
7391 ops->ndo_set_rx_mode(dev);
7394 void dev_set_rx_mode(struct net_device *dev)
7396 netif_addr_lock_bh(dev);
7397 __dev_set_rx_mode(dev);
7398 netif_addr_unlock_bh(dev);
7402 * dev_get_flags - get flags reported to userspace
7405 * Get the combination of flag bits exported through APIs to userspace.
7407 unsigned int dev_get_flags(const struct net_device *dev)
7411 flags = (dev->flags & ~(IFF_PROMISC |
7416 (dev->gflags & (IFF_PROMISC |
7419 if (netif_running(dev)) {
7420 if (netif_oper_up(dev))
7421 flags |= IFF_RUNNING;
7422 if (netif_carrier_ok(dev))
7423 flags |= IFF_LOWER_UP;
7424 if (netif_dormant(dev))
7425 flags |= IFF_DORMANT;
7430 EXPORT_SYMBOL(dev_get_flags);
7432 int __dev_change_flags(struct net_device *dev, unsigned int flags)
7434 unsigned int old_flags = dev->flags;
7440 * Set the flags on our device.
7443 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7444 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7446 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7450 * Load in the correct multicast list now the flags have changed.
7453 if ((old_flags ^ flags) & IFF_MULTICAST)
7454 dev_change_rx_flags(dev, IFF_MULTICAST);
7456 dev_set_rx_mode(dev);
7459 * Have we downed the interface. We handle IFF_UP ourselves
7460 * according to user attempts to set it, rather than blindly
7465 if ((old_flags ^ flags) & IFF_UP) {
7466 if (old_flags & IFF_UP)
7469 ret = __dev_open(dev);
7472 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7473 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7474 unsigned int old_flags = dev->flags;
7476 dev->gflags ^= IFF_PROMISC;
7478 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7479 if (dev->flags != old_flags)
7480 dev_set_rx_mode(dev);
7483 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7484 * is important. Some (broken) drivers set IFF_PROMISC, when
7485 * IFF_ALLMULTI is requested not asking us and not reporting.
7487 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7488 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7490 dev->gflags ^= IFF_ALLMULTI;
7491 __dev_set_allmulti(dev, inc, false);
7497 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7498 unsigned int gchanges)
7500 unsigned int changes = dev->flags ^ old_flags;
7503 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7505 if (changes & IFF_UP) {
7506 if (dev->flags & IFF_UP)
7507 call_netdevice_notifiers(NETDEV_UP, dev);
7509 call_netdevice_notifiers(NETDEV_DOWN, dev);
7512 if (dev->flags & IFF_UP &&
7513 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7514 struct netdev_notifier_change_info change_info = {
7518 .flags_changed = changes,
7521 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7526 * dev_change_flags - change device settings
7528 * @flags: device state flags
7530 * Change settings on device based state flags. The flags are
7531 * in the userspace exported format.
7533 int dev_change_flags(struct net_device *dev, unsigned int flags)
7536 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7538 ret = __dev_change_flags(dev, flags);
7542 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7543 __dev_notify_flags(dev, old_flags, changes);
7546 EXPORT_SYMBOL(dev_change_flags);
7548 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7550 const struct net_device_ops *ops = dev->netdev_ops;
7552 if (ops->ndo_change_mtu)
7553 return ops->ndo_change_mtu(dev, new_mtu);
7558 EXPORT_SYMBOL(__dev_set_mtu);
7561 * dev_set_mtu_ext - Change maximum transfer unit
7563 * @new_mtu: new transfer unit
7564 * @extack: netlink extended ack
7566 * Change the maximum transfer size of the network device.
7568 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7569 struct netlink_ext_ack *extack)
7573 if (new_mtu == dev->mtu)
7576 /* MTU must be positive, and in range */
7577 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7578 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7582 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7583 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7587 if (!netif_device_present(dev))
7590 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7591 err = notifier_to_errno(err);
7595 orig_mtu = dev->mtu;
7596 err = __dev_set_mtu(dev, new_mtu);
7599 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7601 err = notifier_to_errno(err);
7603 /* setting mtu back and notifying everyone again,
7604 * so that they have a chance to revert changes.
7606 __dev_set_mtu(dev, orig_mtu);
7607 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7614 int dev_set_mtu(struct net_device *dev, int new_mtu)
7616 struct netlink_ext_ack extack;
7619 memset(&extack, 0, sizeof(extack));
7620 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7621 if (err && extack._msg)
7622 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7625 EXPORT_SYMBOL(dev_set_mtu);
7628 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7630 * @new_len: new tx queue length
7632 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7634 unsigned int orig_len = dev->tx_queue_len;
7637 if (new_len != (unsigned int)new_len)
7640 if (new_len != orig_len) {
7641 dev->tx_queue_len = new_len;
7642 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7643 res = notifier_to_errno(res);
7646 res = dev_qdisc_change_tx_queue_len(dev);
7654 netdev_err(dev, "refused to change device tx_queue_len\n");
7655 dev->tx_queue_len = orig_len;
7660 * dev_set_group - Change group this device belongs to
7662 * @new_group: group this device should belong to
7664 void dev_set_group(struct net_device *dev, int new_group)
7666 dev->group = new_group;
7668 EXPORT_SYMBOL(dev_set_group);
7671 * dev_set_mac_address - Change Media Access Control Address
7675 * Change the hardware (MAC) address of the device
7677 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
7679 const struct net_device_ops *ops = dev->netdev_ops;
7682 if (!ops->ndo_set_mac_address)
7684 if (sa->sa_family != dev->type)
7686 if (!netif_device_present(dev))
7688 err = ops->ndo_set_mac_address(dev, sa);
7691 dev->addr_assign_type = NET_ADDR_SET;
7692 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7693 add_device_randomness(dev->dev_addr, dev->addr_len);
7696 EXPORT_SYMBOL(dev_set_mac_address);
7699 * dev_change_carrier - Change device carrier
7701 * @new_carrier: new value
7703 * Change device carrier
7705 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7707 const struct net_device_ops *ops = dev->netdev_ops;
7709 if (!ops->ndo_change_carrier)
7711 if (!netif_device_present(dev))
7713 return ops->ndo_change_carrier(dev, new_carrier);
7715 EXPORT_SYMBOL(dev_change_carrier);
7718 * dev_get_phys_port_id - Get device physical port ID
7722 * Get device physical port ID
7724 int dev_get_phys_port_id(struct net_device *dev,
7725 struct netdev_phys_item_id *ppid)
7727 const struct net_device_ops *ops = dev->netdev_ops;
7729 if (!ops->ndo_get_phys_port_id)
7731 return ops->ndo_get_phys_port_id(dev, ppid);
7733 EXPORT_SYMBOL(dev_get_phys_port_id);
7736 * dev_get_phys_port_name - Get device physical port name
7739 * @len: limit of bytes to copy to name
7741 * Get device physical port name
7743 int dev_get_phys_port_name(struct net_device *dev,
7744 char *name, size_t len)
7746 const struct net_device_ops *ops = dev->netdev_ops;
7748 if (!ops->ndo_get_phys_port_name)
7750 return ops->ndo_get_phys_port_name(dev, name, len);
7752 EXPORT_SYMBOL(dev_get_phys_port_name);
7755 * dev_change_proto_down - update protocol port state information
7757 * @proto_down: new value
7759 * This info can be used by switch drivers to set the phys state of the
7762 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7764 const struct net_device_ops *ops = dev->netdev_ops;
7766 if (!ops->ndo_change_proto_down)
7768 if (!netif_device_present(dev))
7770 return ops->ndo_change_proto_down(dev, proto_down);
7772 EXPORT_SYMBOL(dev_change_proto_down);
7774 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
7775 enum bpf_netdev_command cmd)
7777 struct netdev_bpf xdp;
7782 memset(&xdp, 0, sizeof(xdp));
7785 /* Query must always succeed. */
7786 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
7791 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
7792 struct netlink_ext_ack *extack, u32 flags,
7793 struct bpf_prog *prog)
7795 struct netdev_bpf xdp;
7797 memset(&xdp, 0, sizeof(xdp));
7798 if (flags & XDP_FLAGS_HW_MODE)
7799 xdp.command = XDP_SETUP_PROG_HW;
7801 xdp.command = XDP_SETUP_PROG;
7802 xdp.extack = extack;
7806 return bpf_op(dev, &xdp);
7809 static void dev_xdp_uninstall(struct net_device *dev)
7811 struct netdev_bpf xdp;
7814 /* Remove generic XDP */
7815 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
7817 /* Remove from the driver */
7818 ndo_bpf = dev->netdev_ops->ndo_bpf;
7822 memset(&xdp, 0, sizeof(xdp));
7823 xdp.command = XDP_QUERY_PROG;
7824 WARN_ON(ndo_bpf(dev, &xdp));
7826 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
7829 /* Remove HW offload */
7830 memset(&xdp, 0, sizeof(xdp));
7831 xdp.command = XDP_QUERY_PROG_HW;
7832 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
7833 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
7838 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
7840 * @extack: netlink extended ack
7841 * @fd: new program fd or negative value to clear
7842 * @flags: xdp-related flags
7844 * Set or clear a bpf program for a device
7846 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
7849 const struct net_device_ops *ops = dev->netdev_ops;
7850 enum bpf_netdev_command query;
7851 struct bpf_prog *prog = NULL;
7852 bpf_op_t bpf_op, bpf_chk;
7857 query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
7859 bpf_op = bpf_chk = ops->ndo_bpf;
7860 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7862 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
7863 bpf_op = generic_xdp_install;
7864 if (bpf_op == bpf_chk)
7865 bpf_chk = generic_xdp_install;
7868 if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) ||
7869 __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW))
7871 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7872 __dev_xdp_query(dev, bpf_op, query))
7875 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
7876 bpf_op == ops->ndo_bpf);
7878 return PTR_ERR(prog);
7880 if (!(flags & XDP_FLAGS_HW_MODE) &&
7881 bpf_prog_is_dev_bound(prog->aux)) {
7882 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
7888 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
7889 if (err < 0 && prog)
7896 * dev_new_index - allocate an ifindex
7897 * @net: the applicable net namespace
7899 * Returns a suitable unique value for a new device interface
7900 * number. The caller must hold the rtnl semaphore or the
7901 * dev_base_lock to be sure it remains unique.
7903 static int dev_new_index(struct net *net)
7905 int ifindex = net->ifindex;
7910 if (!__dev_get_by_index(net, ifindex))
7911 return net->ifindex = ifindex;
7915 /* Delayed registration/unregisteration */
7916 static LIST_HEAD(net_todo_list);
7917 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7919 static void net_set_todo(struct net_device *dev)
7921 list_add_tail(&dev->todo_list, &net_todo_list);
7922 dev_net(dev)->dev_unreg_count++;
7925 static void rollback_registered_many(struct list_head *head)
7927 struct net_device *dev, *tmp;
7928 LIST_HEAD(close_head);
7930 BUG_ON(dev_boot_phase);
7933 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
7934 /* Some devices call without registering
7935 * for initialization unwind. Remove those
7936 * devices and proceed with the remaining.
7938 if (dev->reg_state == NETREG_UNINITIALIZED) {
7939 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7943 list_del(&dev->unreg_list);
7946 dev->dismantle = true;
7947 BUG_ON(dev->reg_state != NETREG_REGISTERED);
7950 /* If device is running, close it first. */
7951 list_for_each_entry(dev, head, unreg_list)
7952 list_add_tail(&dev->close_list, &close_head);
7953 dev_close_many(&close_head, true);
7955 list_for_each_entry(dev, head, unreg_list) {
7956 /* And unlink it from device chain. */
7957 unlist_netdevice(dev);
7959 dev->reg_state = NETREG_UNREGISTERING;
7961 flush_all_backlogs();
7965 list_for_each_entry(dev, head, unreg_list) {
7966 struct sk_buff *skb = NULL;
7968 /* Shutdown queueing discipline. */
7971 dev_xdp_uninstall(dev);
7973 /* Notify protocols, that we are about to destroy
7974 * this device. They should clean all the things.
7976 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7978 if (!dev->rtnl_link_ops ||
7979 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7980 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
7981 GFP_KERNEL, NULL, 0);
7984 * Flush the unicast and multicast chains
7989 if (dev->netdev_ops->ndo_uninit)
7990 dev->netdev_ops->ndo_uninit(dev);
7993 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
7995 /* Notifier chain MUST detach us all upper devices. */
7996 WARN_ON(netdev_has_any_upper_dev(dev));
7997 WARN_ON(netdev_has_any_lower_dev(dev));
7999 /* Remove entries from kobject tree */
8000 netdev_unregister_kobject(dev);
8002 /* Remove XPS queueing entries */
8003 netif_reset_xps_queues_gt(dev, 0);
8009 list_for_each_entry(dev, head, unreg_list)
8013 static void rollback_registered(struct net_device *dev)
8017 list_add(&dev->unreg_list, &single);
8018 rollback_registered_many(&single);
8022 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8023 struct net_device *upper, netdev_features_t features)
8025 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8026 netdev_features_t feature;
8029 for_each_netdev_feature(&upper_disables, feature_bit) {
8030 feature = __NETIF_F_BIT(feature_bit);
8031 if (!(upper->wanted_features & feature)
8032 && (features & feature)) {
8033 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8034 &feature, upper->name);
8035 features &= ~feature;
8042 static void netdev_sync_lower_features(struct net_device *upper,
8043 struct net_device *lower, netdev_features_t features)
8045 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8046 netdev_features_t feature;
8049 for_each_netdev_feature(&upper_disables, feature_bit) {
8050 feature = __NETIF_F_BIT(feature_bit);
8051 if (!(features & feature) && (lower->features & feature)) {
8052 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8053 &feature, lower->name);
8054 lower->wanted_features &= ~feature;
8055 netdev_update_features(lower);
8057 if (unlikely(lower->features & feature))
8058 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8059 &feature, lower->name);
8064 static netdev_features_t netdev_fix_features(struct net_device *dev,
8065 netdev_features_t features)
8067 /* Fix illegal checksum combinations */
8068 if ((features & NETIF_F_HW_CSUM) &&
8069 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8070 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8071 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8074 /* TSO requires that SG is present as well. */
8075 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8076 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8077 features &= ~NETIF_F_ALL_TSO;
8080 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8081 !(features & NETIF_F_IP_CSUM)) {
8082 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8083 features &= ~NETIF_F_TSO;
8084 features &= ~NETIF_F_TSO_ECN;
8087 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8088 !(features & NETIF_F_IPV6_CSUM)) {
8089 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8090 features &= ~NETIF_F_TSO6;
8093 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8094 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8095 features &= ~NETIF_F_TSO_MANGLEID;
8097 /* TSO ECN requires that TSO is present as well. */
8098 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8099 features &= ~NETIF_F_TSO_ECN;
8101 /* Software GSO depends on SG. */
8102 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8103 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8104 features &= ~NETIF_F_GSO;
8107 /* GSO partial features require GSO partial be set */
8108 if ((features & dev->gso_partial_features) &&
8109 !(features & NETIF_F_GSO_PARTIAL)) {
8111 "Dropping partially supported GSO features since no GSO partial.\n");
8112 features &= ~dev->gso_partial_features;
8115 if (!(features & NETIF_F_RXCSUM)) {
8116 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8117 * successfully merged by hardware must also have the
8118 * checksum verified by hardware. If the user does not
8119 * want to enable RXCSUM, logically, we should disable GRO_HW.
8121 if (features & NETIF_F_GRO_HW) {
8122 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8123 features &= ~NETIF_F_GRO_HW;
8127 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8128 if (features & NETIF_F_RXFCS) {
8129 if (features & NETIF_F_LRO) {
8130 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8131 features &= ~NETIF_F_LRO;
8134 if (features & NETIF_F_GRO_HW) {
8135 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8136 features &= ~NETIF_F_GRO_HW;
8143 int __netdev_update_features(struct net_device *dev)
8145 struct net_device *upper, *lower;
8146 netdev_features_t features;
8147 struct list_head *iter;
8152 features = netdev_get_wanted_features(dev);
8154 if (dev->netdev_ops->ndo_fix_features)
8155 features = dev->netdev_ops->ndo_fix_features(dev, features);
8157 /* driver might be less strict about feature dependencies */
8158 features = netdev_fix_features(dev, features);
8160 /* some features can't be enabled if they're off an an upper device */
8161 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8162 features = netdev_sync_upper_features(dev, upper, features);
8164 if (dev->features == features)
8167 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8168 &dev->features, &features);
8170 if (dev->netdev_ops->ndo_set_features)
8171 err = dev->netdev_ops->ndo_set_features(dev, features);
8175 if (unlikely(err < 0)) {
8177 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8178 err, &features, &dev->features);
8179 /* return non-0 since some features might have changed and
8180 * it's better to fire a spurious notification than miss it
8186 /* some features must be disabled on lower devices when disabled
8187 * on an upper device (think: bonding master or bridge)
8189 netdev_for_each_lower_dev(dev, lower, iter)
8190 netdev_sync_lower_features(dev, lower, features);
8193 netdev_features_t diff = features ^ dev->features;
8195 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8196 /* udp_tunnel_{get,drop}_rx_info both need
8197 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8198 * device, or they won't do anything.
8199 * Thus we need to update dev->features
8200 * *before* calling udp_tunnel_get_rx_info,
8201 * but *after* calling udp_tunnel_drop_rx_info.
8203 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8204 dev->features = features;
8205 udp_tunnel_get_rx_info(dev);
8207 udp_tunnel_drop_rx_info(dev);
8211 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8212 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8213 dev->features = features;
8214 err |= vlan_get_rx_ctag_filter_info(dev);
8216 vlan_drop_rx_ctag_filter_info(dev);
8220 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8221 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8222 dev->features = features;
8223 err |= vlan_get_rx_stag_filter_info(dev);
8225 vlan_drop_rx_stag_filter_info(dev);
8229 dev->features = features;
8232 return err < 0 ? 0 : 1;
8236 * netdev_update_features - recalculate device features
8237 * @dev: the device to check
8239 * Recalculate dev->features set and send notifications if it
8240 * has changed. Should be called after driver or hardware dependent
8241 * conditions might have changed that influence the features.
8243 void netdev_update_features(struct net_device *dev)
8245 if (__netdev_update_features(dev))
8246 netdev_features_change(dev);
8248 EXPORT_SYMBOL(netdev_update_features);
8251 * netdev_change_features - recalculate device features
8252 * @dev: the device to check
8254 * Recalculate dev->features set and send notifications even
8255 * if they have not changed. Should be called instead of
8256 * netdev_update_features() if also dev->vlan_features might
8257 * have changed to allow the changes to be propagated to stacked
8260 void netdev_change_features(struct net_device *dev)
8262 __netdev_update_features(dev);
8263 netdev_features_change(dev);
8265 EXPORT_SYMBOL(netdev_change_features);
8268 * netif_stacked_transfer_operstate - transfer operstate
8269 * @rootdev: the root or lower level device to transfer state from
8270 * @dev: the device to transfer operstate to
8272 * Transfer operational state from root to device. This is normally
8273 * called when a stacking relationship exists between the root
8274 * device and the device(a leaf device).
8276 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8277 struct net_device *dev)
8279 if (rootdev->operstate == IF_OPER_DORMANT)
8280 netif_dormant_on(dev);
8282 netif_dormant_off(dev);
8284 if (netif_carrier_ok(rootdev))
8285 netif_carrier_on(dev);
8287 netif_carrier_off(dev);
8289 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8291 static int netif_alloc_rx_queues(struct net_device *dev)
8293 unsigned int i, count = dev->num_rx_queues;
8294 struct netdev_rx_queue *rx;
8295 size_t sz = count * sizeof(*rx);
8300 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8306 for (i = 0; i < count; i++) {
8309 /* XDP RX-queue setup */
8310 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8317 /* Rollback successful reg's and free other resources */
8319 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8325 static void netif_free_rx_queues(struct net_device *dev)
8327 unsigned int i, count = dev->num_rx_queues;
8329 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8333 for (i = 0; i < count; i++)
8334 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8339 static void netdev_init_one_queue(struct net_device *dev,
8340 struct netdev_queue *queue, void *_unused)
8342 /* Initialize queue lock */
8343 spin_lock_init(&queue->_xmit_lock);
8344 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8345 queue->xmit_lock_owner = -1;
8346 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8349 dql_init(&queue->dql, HZ);
8353 static void netif_free_tx_queues(struct net_device *dev)
8358 static int netif_alloc_netdev_queues(struct net_device *dev)
8360 unsigned int count = dev->num_tx_queues;
8361 struct netdev_queue *tx;
8362 size_t sz = count * sizeof(*tx);
8364 if (count < 1 || count > 0xffff)
8367 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8373 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
8374 spin_lock_init(&dev->tx_global_lock);
8379 void netif_tx_stop_all_queues(struct net_device *dev)
8383 for (i = 0; i < dev->num_tx_queues; i++) {
8384 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
8386 netif_tx_stop_queue(txq);
8389 EXPORT_SYMBOL(netif_tx_stop_all_queues);
8392 * register_netdevice - register a network device
8393 * @dev: device to register
8395 * Take a completed network device structure and add it to the kernel
8396 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8397 * chain. 0 is returned on success. A negative errno code is returned
8398 * on a failure to set up the device, or if the name is a duplicate.
8400 * Callers must hold the rtnl semaphore. You may want
8401 * register_netdev() instead of this.
8404 * The locking appears insufficient to guarantee two parallel registers
8405 * will not get the same name.
8408 int register_netdevice(struct net_device *dev)
8411 struct net *net = dev_net(dev);
8413 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
8414 NETDEV_FEATURE_COUNT);
8415 BUG_ON(dev_boot_phase);
8420 /* When net_device's are persistent, this will be fatal. */
8421 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8424 spin_lock_init(&dev->addr_list_lock);
8425 netdev_set_addr_lockdep_class(dev);
8427 ret = dev_get_valid_name(net, dev, dev->name);
8431 /* Init, if this function is available */
8432 if (dev->netdev_ops->ndo_init) {
8433 ret = dev->netdev_ops->ndo_init(dev);
8441 if (((dev->hw_features | dev->features) &
8442 NETIF_F_HW_VLAN_CTAG_FILTER) &&
8443 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
8444 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
8445 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
8452 dev->ifindex = dev_new_index(net);
8453 else if (__dev_get_by_index(net, dev->ifindex))
8456 /* Transfer changeable features to wanted_features and enable
8457 * software offloads (GSO and GRO).
8459 dev->hw_features |= NETIF_F_SOFT_FEATURES;
8460 dev->features |= NETIF_F_SOFT_FEATURES;
8462 if (dev->netdev_ops->ndo_udp_tunnel_add) {
8463 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8464 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8467 dev->wanted_features = dev->features & dev->hw_features;
8469 if (!(dev->flags & IFF_LOOPBACK))
8470 dev->hw_features |= NETIF_F_NOCACHE_COPY;
8472 /* If IPv4 TCP segmentation offload is supported we should also
8473 * allow the device to enable segmenting the frame with the option
8474 * of ignoring a static IP ID value. This doesn't enable the
8475 * feature itself but allows the user to enable it later.
8477 if (dev->hw_features & NETIF_F_TSO)
8478 dev->hw_features |= NETIF_F_TSO_MANGLEID;
8479 if (dev->vlan_features & NETIF_F_TSO)
8480 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
8481 if (dev->mpls_features & NETIF_F_TSO)
8482 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
8483 if (dev->hw_enc_features & NETIF_F_TSO)
8484 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
8486 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8488 dev->vlan_features |= NETIF_F_HIGHDMA;
8490 /* Make NETIF_F_SG inheritable to tunnel devices.
8492 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
8494 /* Make NETIF_F_SG inheritable to MPLS.
8496 dev->mpls_features |= NETIF_F_SG;
8498 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
8499 ret = notifier_to_errno(ret);
8503 ret = netdev_register_kobject(dev);
8506 dev->reg_state = NETREG_REGISTERED;
8508 __netdev_update_features(dev);
8511 * Default initial state at registry is that the
8512 * device is present.
8515 set_bit(__LINK_STATE_PRESENT, &dev->state);
8517 linkwatch_init_dev(dev);
8519 dev_init_scheduler(dev);
8521 list_netdevice(dev);
8522 add_device_randomness(dev->dev_addr, dev->addr_len);
8524 /* If the device has permanent device address, driver should
8525 * set dev_addr and also addr_assign_type should be set to
8526 * NET_ADDR_PERM (default value).
8528 if (dev->addr_assign_type == NET_ADDR_PERM)
8529 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8531 /* Notify protocols, that a new device appeared. */
8532 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8533 ret = notifier_to_errno(ret);
8535 rollback_registered(dev);
8536 dev->reg_state = NETREG_UNREGISTERED;
8539 * Prevent userspace races by waiting until the network
8540 * device is fully setup before sending notifications.
8542 if (!dev->rtnl_link_ops ||
8543 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8544 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8550 if (dev->netdev_ops->ndo_uninit)
8551 dev->netdev_ops->ndo_uninit(dev);
8552 if (dev->priv_destructor)
8553 dev->priv_destructor(dev);
8556 EXPORT_SYMBOL(register_netdevice);
8559 * init_dummy_netdev - init a dummy network device for NAPI
8560 * @dev: device to init
8562 * This takes a network device structure and initialize the minimum
8563 * amount of fields so it can be used to schedule NAPI polls without
8564 * registering a full blown interface. This is to be used by drivers
8565 * that need to tie several hardware interfaces to a single NAPI
8566 * poll scheduler due to HW limitations.
8568 int init_dummy_netdev(struct net_device *dev)
8570 /* Clear everything. Note we don't initialize spinlocks
8571 * are they aren't supposed to be taken by any of the
8572 * NAPI code and this dummy netdev is supposed to be
8573 * only ever used for NAPI polls
8575 memset(dev, 0, sizeof(struct net_device));
8577 /* make sure we BUG if trying to hit standard
8578 * register/unregister code path
8580 dev->reg_state = NETREG_DUMMY;
8582 /* NAPI wants this */
8583 INIT_LIST_HEAD(&dev->napi_list);
8585 /* a dummy interface is started by default */
8586 set_bit(__LINK_STATE_PRESENT, &dev->state);
8587 set_bit(__LINK_STATE_START, &dev->state);
8589 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8590 * because users of this 'device' dont need to change
8596 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8600 * register_netdev - register a network device
8601 * @dev: device to register
8603 * Take a completed network device structure and add it to the kernel
8604 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8605 * chain. 0 is returned on success. A negative errno code is returned
8606 * on a failure to set up the device, or if the name is a duplicate.
8608 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8609 * and expands the device name if you passed a format string to
8612 int register_netdev(struct net_device *dev)
8616 if (rtnl_lock_killable())
8618 err = register_netdevice(dev);
8622 EXPORT_SYMBOL(register_netdev);
8624 int netdev_refcnt_read(const struct net_device *dev)
8628 for_each_possible_cpu(i)
8629 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8632 EXPORT_SYMBOL(netdev_refcnt_read);
8635 * netdev_wait_allrefs - wait until all references are gone.
8636 * @dev: target net_device
8638 * This is called when unregistering network devices.
8640 * Any protocol or device that holds a reference should register
8641 * for netdevice notification, and cleanup and put back the
8642 * reference if they receive an UNREGISTER event.
8643 * We can get stuck here if buggy protocols don't correctly
8646 static void netdev_wait_allrefs(struct net_device *dev)
8648 unsigned long rebroadcast_time, warning_time;
8651 linkwatch_forget_dev(dev);
8653 rebroadcast_time = warning_time = jiffies;
8654 refcnt = netdev_refcnt_read(dev);
8656 while (refcnt != 0) {
8657 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8660 /* Rebroadcast unregister notification */
8661 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8667 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8669 /* We must not have linkwatch events
8670 * pending on unregister. If this
8671 * happens, we simply run the queue
8672 * unscheduled, resulting in a noop
8675 linkwatch_run_queue();
8680 rebroadcast_time = jiffies;
8685 refcnt = netdev_refcnt_read(dev);
8687 if (time_after(jiffies, warning_time + 10 * HZ)) {
8688 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8690 warning_time = jiffies;
8699 * register_netdevice(x1);
8700 * register_netdevice(x2);
8702 * unregister_netdevice(y1);
8703 * unregister_netdevice(y2);
8709 * We are invoked by rtnl_unlock().
8710 * This allows us to deal with problems:
8711 * 1) We can delete sysfs objects which invoke hotplug
8712 * without deadlocking with linkwatch via keventd.
8713 * 2) Since we run with the RTNL semaphore not held, we can sleep
8714 * safely in order to wait for the netdev refcnt to drop to zero.
8716 * We must not return until all unregister events added during
8717 * the interval the lock was held have been completed.
8719 void netdev_run_todo(void)
8721 struct list_head list;
8723 /* Snapshot list, allow later requests */
8724 list_replace_init(&net_todo_list, &list);
8729 /* Wait for rcu callbacks to finish before next phase */
8730 if (!list_empty(&list))
8733 while (!list_empty(&list)) {
8734 struct net_device *dev
8735 = list_first_entry(&list, struct net_device, todo_list);
8736 list_del(&dev->todo_list);
8738 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8739 pr_err("network todo '%s' but state %d\n",
8740 dev->name, dev->reg_state);
8745 dev->reg_state = NETREG_UNREGISTERED;
8747 netdev_wait_allrefs(dev);
8750 BUG_ON(netdev_refcnt_read(dev));
8751 BUG_ON(!list_empty(&dev->ptype_all));
8752 BUG_ON(!list_empty(&dev->ptype_specific));
8753 WARN_ON(rcu_access_pointer(dev->ip_ptr));
8754 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
8755 #if IS_ENABLED(CONFIG_DECNET)
8756 WARN_ON(dev->dn_ptr);
8758 if (dev->priv_destructor)
8759 dev->priv_destructor(dev);
8760 if (dev->needs_free_netdev)
8763 /* Report a network device has been unregistered */
8765 dev_net(dev)->dev_unreg_count--;
8767 wake_up(&netdev_unregistering_wq);
8769 /* Free network device */
8770 kobject_put(&dev->dev.kobj);
8774 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8775 * all the same fields in the same order as net_device_stats, with only
8776 * the type differing, but rtnl_link_stats64 may have additional fields
8777 * at the end for newer counters.
8779 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
8780 const struct net_device_stats *netdev_stats)
8782 #if BITS_PER_LONG == 64
8783 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
8784 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
8785 /* zero out counters that only exist in rtnl_link_stats64 */
8786 memset((char *)stats64 + sizeof(*netdev_stats), 0,
8787 sizeof(*stats64) - sizeof(*netdev_stats));
8789 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
8790 const unsigned long *src = (const unsigned long *)netdev_stats;
8791 u64 *dst = (u64 *)stats64;
8793 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
8794 for (i = 0; i < n; i++)
8796 /* zero out counters that only exist in rtnl_link_stats64 */
8797 memset((char *)stats64 + n * sizeof(u64), 0,
8798 sizeof(*stats64) - n * sizeof(u64));
8801 EXPORT_SYMBOL(netdev_stats_to_stats64);
8804 * dev_get_stats - get network device statistics
8805 * @dev: device to get statistics from
8806 * @storage: place to store stats
8808 * Get network statistics from device. Return @storage.
8809 * The device driver may provide its own method by setting
8810 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8811 * otherwise the internal statistics structure is used.
8813 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
8814 struct rtnl_link_stats64 *storage)
8816 const struct net_device_ops *ops = dev->netdev_ops;
8818 if (ops->ndo_get_stats64) {
8819 memset(storage, 0, sizeof(*storage));
8820 ops->ndo_get_stats64(dev, storage);
8821 } else if (ops->ndo_get_stats) {
8822 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
8824 netdev_stats_to_stats64(storage, &dev->stats);
8826 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
8827 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
8828 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
8831 EXPORT_SYMBOL(dev_get_stats);
8833 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
8835 struct netdev_queue *queue = dev_ingress_queue(dev);
8837 #ifdef CONFIG_NET_CLS_ACT
8840 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
8843 netdev_init_one_queue(dev, queue, NULL);
8844 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
8845 queue->qdisc_sleeping = &noop_qdisc;
8846 rcu_assign_pointer(dev->ingress_queue, queue);
8851 static const struct ethtool_ops default_ethtool_ops;
8853 void netdev_set_default_ethtool_ops(struct net_device *dev,
8854 const struct ethtool_ops *ops)
8856 if (dev->ethtool_ops == &default_ethtool_ops)
8857 dev->ethtool_ops = ops;
8859 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
8861 void netdev_freemem(struct net_device *dev)
8863 char *addr = (char *)dev - dev->padded;
8869 * alloc_netdev_mqs - allocate network device
8870 * @sizeof_priv: size of private data to allocate space for
8871 * @name: device name format string
8872 * @name_assign_type: origin of device name
8873 * @setup: callback to initialize device
8874 * @txqs: the number of TX subqueues to allocate
8875 * @rxqs: the number of RX subqueues to allocate
8877 * Allocates a struct net_device with private data area for driver use
8878 * and performs basic initialization. Also allocates subqueue structs
8879 * for each queue on the device.
8881 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
8882 unsigned char name_assign_type,
8883 void (*setup)(struct net_device *),
8884 unsigned int txqs, unsigned int rxqs)
8886 struct net_device *dev;
8887 unsigned int alloc_size;
8888 struct net_device *p;
8890 BUG_ON(strlen(name) >= sizeof(dev->name));
8893 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8898 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
8902 alloc_size = sizeof(struct net_device);
8904 /* ensure 32-byte alignment of private area */
8905 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
8906 alloc_size += sizeof_priv;
8908 /* ensure 32-byte alignment of whole construct */
8909 alloc_size += NETDEV_ALIGN - 1;
8911 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8915 dev = PTR_ALIGN(p, NETDEV_ALIGN);
8916 dev->padded = (char *)dev - (char *)p;
8918 dev->pcpu_refcnt = alloc_percpu(int);
8919 if (!dev->pcpu_refcnt)
8922 if (dev_addr_init(dev))
8928 dev_net_set(dev, &init_net);
8930 dev->gso_max_size = GSO_MAX_SIZE;
8931 dev->gso_max_segs = GSO_MAX_SEGS;
8933 INIT_LIST_HEAD(&dev->napi_list);
8934 INIT_LIST_HEAD(&dev->unreg_list);
8935 INIT_LIST_HEAD(&dev->close_list);
8936 INIT_LIST_HEAD(&dev->link_watch_list);
8937 INIT_LIST_HEAD(&dev->adj_list.upper);
8938 INIT_LIST_HEAD(&dev->adj_list.lower);
8939 INIT_LIST_HEAD(&dev->ptype_all);
8940 INIT_LIST_HEAD(&dev->ptype_specific);
8941 #ifdef CONFIG_NET_SCHED
8942 hash_init(dev->qdisc_hash);
8944 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
8947 if (!dev->tx_queue_len) {
8948 dev->priv_flags |= IFF_NO_QUEUE;
8949 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
8952 dev->num_tx_queues = txqs;
8953 dev->real_num_tx_queues = txqs;
8954 if (netif_alloc_netdev_queues(dev))
8957 dev->num_rx_queues = rxqs;
8958 dev->real_num_rx_queues = rxqs;
8959 if (netif_alloc_rx_queues(dev))
8962 strcpy(dev->name, name);
8963 dev->name_assign_type = name_assign_type;
8964 dev->group = INIT_NETDEV_GROUP;
8965 if (!dev->ethtool_ops)
8966 dev->ethtool_ops = &default_ethtool_ops;
8968 nf_hook_ingress_init(dev);
8977 free_percpu(dev->pcpu_refcnt);
8979 netdev_freemem(dev);
8982 EXPORT_SYMBOL(alloc_netdev_mqs);
8985 * free_netdev - free network device
8988 * This function does the last stage of destroying an allocated device
8989 * interface. The reference to the device object is released. If this
8990 * is the last reference then it will be freed.Must be called in process
8993 void free_netdev(struct net_device *dev)
8995 struct napi_struct *p, *n;
8998 netif_free_tx_queues(dev);
8999 netif_free_rx_queues(dev);
9001 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9003 /* Flush device addresses */
9004 dev_addr_flush(dev);
9006 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9009 free_percpu(dev->pcpu_refcnt);
9010 dev->pcpu_refcnt = NULL;
9012 /* Compatibility with error handling in drivers */
9013 if (dev->reg_state == NETREG_UNINITIALIZED) {
9014 netdev_freemem(dev);
9018 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9019 dev->reg_state = NETREG_RELEASED;
9021 /* will free via device release */
9022 put_device(&dev->dev);
9024 EXPORT_SYMBOL(free_netdev);
9027 * synchronize_net - Synchronize with packet receive processing
9029 * Wait for packets currently being received to be done.
9030 * Does not block later packets from starting.
9032 void synchronize_net(void)
9035 if (rtnl_is_locked())
9036 synchronize_rcu_expedited();
9040 EXPORT_SYMBOL(synchronize_net);
9043 * unregister_netdevice_queue - remove device from the kernel
9047 * This function shuts down a device interface and removes it
9048 * from the kernel tables.
9049 * If head not NULL, device is queued to be unregistered later.
9051 * Callers must hold the rtnl semaphore. You may want
9052 * unregister_netdev() instead of this.
9055 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9060 list_move_tail(&dev->unreg_list, head);
9062 rollback_registered(dev);
9063 /* Finish processing unregister after unlock */
9067 EXPORT_SYMBOL(unregister_netdevice_queue);
9070 * unregister_netdevice_many - unregister many devices
9071 * @head: list of devices
9073 * Note: As most callers use a stack allocated list_head,
9074 * we force a list_del() to make sure stack wont be corrupted later.
9076 void unregister_netdevice_many(struct list_head *head)
9078 struct net_device *dev;
9080 if (!list_empty(head)) {
9081 rollback_registered_many(head);
9082 list_for_each_entry(dev, head, unreg_list)
9087 EXPORT_SYMBOL(unregister_netdevice_many);
9090 * unregister_netdev - remove device from the kernel
9093 * This function shuts down a device interface and removes it
9094 * from the kernel tables.
9096 * This is just a wrapper for unregister_netdevice that takes
9097 * the rtnl semaphore. In general you want to use this and not
9098 * unregister_netdevice.
9100 void unregister_netdev(struct net_device *dev)
9103 unregister_netdevice(dev);
9106 EXPORT_SYMBOL(unregister_netdev);
9109 * dev_change_net_namespace - move device to different nethost namespace
9111 * @net: network namespace
9112 * @pat: If not NULL name pattern to try if the current device name
9113 * is already taken in the destination network namespace.
9115 * This function shuts down a device interface and moves it
9116 * to a new network namespace. On success 0 is returned, on
9117 * a failure a netagive errno code is returned.
9119 * Callers must hold the rtnl semaphore.
9122 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9124 int err, new_nsid, new_ifindex;
9128 /* Don't allow namespace local devices to be moved. */
9130 if (dev->features & NETIF_F_NETNS_LOCAL)
9133 /* Ensure the device has been registrered */
9134 if (dev->reg_state != NETREG_REGISTERED)
9137 /* Get out if there is nothing todo */
9139 if (net_eq(dev_net(dev), net))
9142 /* Pick the destination device name, and ensure
9143 * we can use it in the destination network namespace.
9146 if (__dev_get_by_name(net, dev->name)) {
9147 /* We get here if we can't use the current device name */
9150 err = dev_get_valid_name(net, dev, pat);
9156 * And now a mini version of register_netdevice unregister_netdevice.
9159 /* If device is running close it first. */
9162 /* And unlink it from device chain */
9163 unlist_netdevice(dev);
9167 /* Shutdown queueing discipline. */
9170 /* Notify protocols, that we are about to destroy
9171 * this device. They should clean all the things.
9173 * Note that dev->reg_state stays at NETREG_REGISTERED.
9174 * This is wanted because this way 8021q and macvlan know
9175 * the device is just moving and can keep their slaves up.
9177 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9180 new_nsid = peernet2id_alloc(dev_net(dev), net);
9181 /* If there is an ifindex conflict assign a new one */
9182 if (__dev_get_by_index(net, dev->ifindex))
9183 new_ifindex = dev_new_index(net);
9185 new_ifindex = dev->ifindex;
9187 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9191 * Flush the unicast and multicast chains
9196 /* Send a netdev-removed uevent to the old namespace */
9197 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9198 netdev_adjacent_del_links(dev);
9200 /* Actually switch the network namespace */
9201 dev_net_set(dev, net);
9202 dev->ifindex = new_ifindex;
9204 /* Send a netdev-add uevent to the new namespace */
9205 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9206 netdev_adjacent_add_links(dev);
9208 /* Fixup kobjects */
9209 err = device_rename(&dev->dev, dev->name);
9212 /* Add the device back in the hashes */
9213 list_netdevice(dev);
9215 /* Notify protocols, that a new device appeared. */
9216 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9219 * Prevent userspace races by waiting until the network
9220 * device is fully setup before sending notifications.
9222 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9229 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9231 static int dev_cpu_dead(unsigned int oldcpu)
9233 struct sk_buff **list_skb;
9234 struct sk_buff *skb;
9236 struct softnet_data *sd, *oldsd, *remsd = NULL;
9238 local_irq_disable();
9239 cpu = smp_processor_id();
9240 sd = &per_cpu(softnet_data, cpu);
9241 oldsd = &per_cpu(softnet_data, oldcpu);
9243 /* Find end of our completion_queue. */
9244 list_skb = &sd->completion_queue;
9246 list_skb = &(*list_skb)->next;
9247 /* Append completion queue from offline CPU. */
9248 *list_skb = oldsd->completion_queue;
9249 oldsd->completion_queue = NULL;
9251 /* Append output queue from offline CPU. */
9252 if (oldsd->output_queue) {
9253 *sd->output_queue_tailp = oldsd->output_queue;
9254 sd->output_queue_tailp = oldsd->output_queue_tailp;
9255 oldsd->output_queue = NULL;
9256 oldsd->output_queue_tailp = &oldsd->output_queue;
9258 /* Append NAPI poll list from offline CPU, with one exception :
9259 * process_backlog() must be called by cpu owning percpu backlog.
9260 * We properly handle process_queue & input_pkt_queue later.
9262 while (!list_empty(&oldsd->poll_list)) {
9263 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9267 list_del_init(&napi->poll_list);
9268 if (napi->poll == process_backlog)
9271 ____napi_schedule(sd, napi);
9274 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9278 remsd = oldsd->rps_ipi_list;
9279 oldsd->rps_ipi_list = NULL;
9281 /* send out pending IPI's on offline CPU */
9282 net_rps_send_ipi(remsd);
9284 /* Process offline CPU's input_pkt_queue */
9285 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9287 input_queue_head_incr(oldsd);
9289 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9291 input_queue_head_incr(oldsd);
9298 * netdev_increment_features - increment feature set by one
9299 * @all: current feature set
9300 * @one: new feature set
9301 * @mask: mask feature set
9303 * Computes a new feature set after adding a device with feature set
9304 * @one to the master device with current feature set @all. Will not
9305 * enable anything that is off in @mask. Returns the new feature set.
9307 netdev_features_t netdev_increment_features(netdev_features_t all,
9308 netdev_features_t one, netdev_features_t mask)
9310 if (mask & NETIF_F_HW_CSUM)
9311 mask |= NETIF_F_CSUM_MASK;
9312 mask |= NETIF_F_VLAN_CHALLENGED;
9314 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9315 all &= one | ~NETIF_F_ALL_FOR_ALL;
9317 /* If one device supports hw checksumming, set for all. */
9318 if (all & NETIF_F_HW_CSUM)
9319 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
9323 EXPORT_SYMBOL(netdev_increment_features);
9325 static struct hlist_head * __net_init netdev_create_hash(void)
9328 struct hlist_head *hash;
9330 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
9332 for (i = 0; i < NETDEV_HASHENTRIES; i++)
9333 INIT_HLIST_HEAD(&hash[i]);
9338 /* Initialize per network namespace state */
9339 static int __net_init netdev_init(struct net *net)
9341 BUILD_BUG_ON(GRO_HASH_BUCKETS >
9342 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
9344 if (net != &init_net)
9345 INIT_LIST_HEAD(&net->dev_base_head);
9347 net->dev_name_head = netdev_create_hash();
9348 if (net->dev_name_head == NULL)
9351 net->dev_index_head = netdev_create_hash();
9352 if (net->dev_index_head == NULL)
9358 kfree(net->dev_name_head);
9364 * netdev_drivername - network driver for the device
9365 * @dev: network device
9367 * Determine network driver for device.
9369 const char *netdev_drivername(const struct net_device *dev)
9371 const struct device_driver *driver;
9372 const struct device *parent;
9373 const char *empty = "";
9375 parent = dev->dev.parent;
9379 driver = parent->driver;
9380 if (driver && driver->name)
9381 return driver->name;
9385 static void __netdev_printk(const char *level, const struct net_device *dev,
9386 struct va_format *vaf)
9388 if (dev && dev->dev.parent) {
9389 dev_printk_emit(level[1] - '0',
9392 dev_driver_string(dev->dev.parent),
9393 dev_name(dev->dev.parent),
9394 netdev_name(dev), netdev_reg_state(dev),
9397 printk("%s%s%s: %pV",
9398 level, netdev_name(dev), netdev_reg_state(dev), vaf);
9400 printk("%s(NULL net_device): %pV", level, vaf);
9404 void netdev_printk(const char *level, const struct net_device *dev,
9405 const char *format, ...)
9407 struct va_format vaf;
9410 va_start(args, format);
9415 __netdev_printk(level, dev, &vaf);
9419 EXPORT_SYMBOL(netdev_printk);
9421 #define define_netdev_printk_level(func, level) \
9422 void func(const struct net_device *dev, const char *fmt, ...) \
9424 struct va_format vaf; \
9427 va_start(args, fmt); \
9432 __netdev_printk(level, dev, &vaf); \
9436 EXPORT_SYMBOL(func);
9438 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
9439 define_netdev_printk_level(netdev_alert, KERN_ALERT);
9440 define_netdev_printk_level(netdev_crit, KERN_CRIT);
9441 define_netdev_printk_level(netdev_err, KERN_ERR);
9442 define_netdev_printk_level(netdev_warn, KERN_WARNING);
9443 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
9444 define_netdev_printk_level(netdev_info, KERN_INFO);
9446 static void __net_exit netdev_exit(struct net *net)
9448 kfree(net->dev_name_head);
9449 kfree(net->dev_index_head);
9450 if (net != &init_net)
9451 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
9454 static struct pernet_operations __net_initdata netdev_net_ops = {
9455 .init = netdev_init,
9456 .exit = netdev_exit,
9459 static void __net_exit default_device_exit(struct net *net)
9461 struct net_device *dev, *aux;
9463 * Push all migratable network devices back to the
9464 * initial network namespace
9467 for_each_netdev_safe(net, dev, aux) {
9469 char fb_name[IFNAMSIZ];
9471 /* Ignore unmoveable devices (i.e. loopback) */
9472 if (dev->features & NETIF_F_NETNS_LOCAL)
9475 /* Leave virtual devices for the generic cleanup */
9476 if (dev->rtnl_link_ops)
9479 /* Push remaining network devices to init_net */
9480 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
9481 err = dev_change_net_namespace(dev, &init_net, fb_name);
9483 pr_emerg("%s: failed to move %s to init_net: %d\n",
9484 __func__, dev->name, err);
9491 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
9493 /* Return with the rtnl_lock held when there are no network
9494 * devices unregistering in any network namespace in net_list.
9498 DEFINE_WAIT_FUNC(wait, woken_wake_function);
9500 add_wait_queue(&netdev_unregistering_wq, &wait);
9502 unregistering = false;
9504 list_for_each_entry(net, net_list, exit_list) {
9505 if (net->dev_unreg_count > 0) {
9506 unregistering = true;
9514 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
9516 remove_wait_queue(&netdev_unregistering_wq, &wait);
9519 static void __net_exit default_device_exit_batch(struct list_head *net_list)
9521 /* At exit all network devices most be removed from a network
9522 * namespace. Do this in the reverse order of registration.
9523 * Do this across as many network namespaces as possible to
9524 * improve batching efficiency.
9526 struct net_device *dev;
9528 LIST_HEAD(dev_kill_list);
9530 /* To prevent network device cleanup code from dereferencing
9531 * loopback devices or network devices that have been freed
9532 * wait here for all pending unregistrations to complete,
9533 * before unregistring the loopback device and allowing the
9534 * network namespace be freed.
9536 * The netdev todo list containing all network devices
9537 * unregistrations that happen in default_device_exit_batch
9538 * will run in the rtnl_unlock() at the end of
9539 * default_device_exit_batch.
9541 rtnl_lock_unregistering(net_list);
9542 list_for_each_entry(net, net_list, exit_list) {
9543 for_each_netdev_reverse(net, dev) {
9544 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
9545 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
9547 unregister_netdevice_queue(dev, &dev_kill_list);
9550 unregister_netdevice_many(&dev_kill_list);
9554 static struct pernet_operations __net_initdata default_device_ops = {
9555 .exit = default_device_exit,
9556 .exit_batch = default_device_exit_batch,
9560 * Initialize the DEV module. At boot time this walks the device list and
9561 * unhooks any devices that fail to initialise (normally hardware not
9562 * present) and leaves us with a valid list of present and active devices.
9567 * This is called single threaded during boot, so no need
9568 * to take the rtnl semaphore.
9570 static int __init net_dev_init(void)
9572 int i, rc = -ENOMEM;
9574 BUG_ON(!dev_boot_phase);
9576 if (dev_proc_init())
9579 if (netdev_kobject_init())
9582 INIT_LIST_HEAD(&ptype_all);
9583 for (i = 0; i < PTYPE_HASH_SIZE; i++)
9584 INIT_LIST_HEAD(&ptype_base[i]);
9586 INIT_LIST_HEAD(&offload_base);
9588 if (register_pernet_subsys(&netdev_net_ops))
9592 * Initialise the packet receive queues.
9595 for_each_possible_cpu(i) {
9596 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
9597 struct softnet_data *sd = &per_cpu(softnet_data, i);
9599 INIT_WORK(flush, flush_backlog);
9601 skb_queue_head_init(&sd->input_pkt_queue);
9602 skb_queue_head_init(&sd->process_queue);
9603 #ifdef CONFIG_XFRM_OFFLOAD
9604 skb_queue_head_init(&sd->xfrm_backlog);
9606 INIT_LIST_HEAD(&sd->poll_list);
9607 sd->output_queue_tailp = &sd->output_queue;
9609 sd->csd.func = rps_trigger_softirq;
9614 init_gro_hash(&sd->backlog);
9615 sd->backlog.poll = process_backlog;
9616 sd->backlog.weight = weight_p;
9621 /* The loopback device is special if any other network devices
9622 * is present in a network namespace the loopback device must
9623 * be present. Since we now dynamically allocate and free the
9624 * loopback device ensure this invariant is maintained by
9625 * keeping the loopback device as the first device on the
9626 * list of network devices. Ensuring the loopback devices
9627 * is the first device that appears and the last network device
9630 if (register_pernet_device(&loopback_net_ops))
9633 if (register_pernet_device(&default_device_ops))
9636 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
9637 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
9639 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
9640 NULL, dev_cpu_dead);
9647 subsys_initcall(net_dev_init);