1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/string.h>
84 #include <linux/socket.h>
85 #include <linux/sockios.h>
86 #include <linux/errno.h>
87 #include <linux/interrupt.h>
88 #include <linux/if_ether.h>
89 #include <linux/netdevice.h>
90 #include <linux/etherdevice.h>
91 #include <linux/ethtool.h>
92 #include <linux/skbuff.h>
93 #include <linux/bpf.h>
94 #include <linux/bpf_trace.h>
95 #include <net/net_namespace.h>
97 #include <net/busy_poll.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/stat.h>
101 #include <net/dst_metadata.h>
102 #include <net/pkt_sched.h>
103 #include <net/pkt_cls.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
139 #include <linux/crash_dump.h>
140 #include <linux/sctp.h>
141 #include <net/udp_tunnel.h>
142 #include <linux/net_namespace.h>
143 #include <linux/indirect_call_wrapper.h>
144 #include <net/devlink.h>
146 #include "net-sysfs.h"
148 #define MAX_GRO_SKBS 8
150 /* This should be increased if a protocol with a bigger head is added. */
151 #define GRO_MAX_HEAD (MAX_HEADER + 128)
153 static DEFINE_SPINLOCK(ptype_lock);
154 static DEFINE_SPINLOCK(offload_lock);
155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
156 struct list_head ptype_all __read_mostly; /* Taps */
157 static struct list_head offload_base __read_mostly;
159 static int netif_rx_internal(struct sk_buff *skb);
160 static int call_netdevice_notifiers_info(unsigned long val,
161 struct netdev_notifier_info *info);
162 static int call_netdevice_notifiers_extack(unsigned long val,
163 struct net_device *dev,
164 struct netlink_ext_ack *extack);
165 static struct napi_struct *napi_by_id(unsigned int napi_id);
168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base_head list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 DEFINE_RWLOCK(dev_base_lock);
187 EXPORT_SYMBOL(dev_base_lock);
189 static DEFINE_MUTEX(ifalias_mutex);
191 /* protects napi_hash addition/deletion and napi_gen_id */
192 static DEFINE_SPINLOCK(napi_hash_lock);
194 static unsigned int napi_gen_id = NR_CPUS;
195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
197 static seqcount_t devnet_rename_seq;
199 static inline void dev_base_seq_inc(struct net *net)
201 while (++net->dev_base_seq == 0)
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217 static inline void rps_lock(struct softnet_data *sd)
220 spin_lock(&sd->input_pkt_queue.lock);
224 static inline void rps_unlock(struct softnet_data *sd)
227 spin_unlock(&sd->input_pkt_queue.lock);
231 /* 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));
1184 /* Some auto-enslaved devices e.g. failover slaves are
1185 * special, as userspace might rename the device after
1186 * the interface had been brought up and running since
1187 * the point kernel initiated auto-enslavement. Allow
1188 * live name change even when these slave devices are
1191 * Typically, users of these auto-enslaving devices
1192 * don't actually care about slave name change, as
1193 * they are supposed to operate on master interface
1196 if (dev->flags & IFF_UP &&
1197 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1200 write_seqcount_begin(&devnet_rename_seq);
1202 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1203 write_seqcount_end(&devnet_rename_seq);
1207 memcpy(oldname, dev->name, IFNAMSIZ);
1209 err = dev_get_valid_name(net, dev, newname);
1211 write_seqcount_end(&devnet_rename_seq);
1215 if (oldname[0] && !strchr(oldname, '%'))
1216 netdev_info(dev, "renamed from %s\n", oldname);
1218 old_assign_type = dev->name_assign_type;
1219 dev->name_assign_type = NET_NAME_RENAMED;
1222 ret = device_rename(&dev->dev, dev->name);
1224 memcpy(dev->name, oldname, IFNAMSIZ);
1225 dev->name_assign_type = old_assign_type;
1226 write_seqcount_end(&devnet_rename_seq);
1230 write_seqcount_end(&devnet_rename_seq);
1232 netdev_adjacent_rename_links(dev, oldname);
1234 write_lock_bh(&dev_base_lock);
1235 hlist_del_rcu(&dev->name_hlist);
1236 write_unlock_bh(&dev_base_lock);
1240 write_lock_bh(&dev_base_lock);
1241 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1242 write_unlock_bh(&dev_base_lock);
1244 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1245 ret = notifier_to_errno(ret);
1248 /* err >= 0 after dev_alloc_name() or stores the first errno */
1251 write_seqcount_begin(&devnet_rename_seq);
1252 memcpy(dev->name, oldname, IFNAMSIZ);
1253 memcpy(oldname, newname, IFNAMSIZ);
1254 dev->name_assign_type = old_assign_type;
1255 old_assign_type = NET_NAME_RENAMED;
1258 pr_err("%s: name change rollback failed: %d\n",
1267 * dev_set_alias - change ifalias of a device
1269 * @alias: name up to IFALIASZ
1270 * @len: limit of bytes to copy from info
1272 * Set ifalias for a device,
1274 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1276 struct dev_ifalias *new_alias = NULL;
1278 if (len >= IFALIASZ)
1282 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1286 memcpy(new_alias->ifalias, alias, len);
1287 new_alias->ifalias[len] = 0;
1290 mutex_lock(&ifalias_mutex);
1291 rcu_swap_protected(dev->ifalias, new_alias,
1292 mutex_is_locked(&ifalias_mutex));
1293 mutex_unlock(&ifalias_mutex);
1296 kfree_rcu(new_alias, rcuhead);
1300 EXPORT_SYMBOL(dev_set_alias);
1303 * dev_get_alias - get ifalias of a device
1305 * @name: buffer to store name of ifalias
1306 * @len: size of buffer
1308 * get ifalias for a device. Caller must make sure dev cannot go
1309 * away, e.g. rcu read lock or own a reference count to device.
1311 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1313 const struct dev_ifalias *alias;
1317 alias = rcu_dereference(dev->ifalias);
1319 ret = snprintf(name, len, "%s", alias->ifalias);
1326 * netdev_features_change - device changes features
1327 * @dev: device to cause notification
1329 * Called to indicate a device has changed features.
1331 void netdev_features_change(struct net_device *dev)
1333 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1335 EXPORT_SYMBOL(netdev_features_change);
1338 * netdev_state_change - device changes state
1339 * @dev: device to cause notification
1341 * Called to indicate a device has changed state. This function calls
1342 * the notifier chains for netdev_chain and sends a NEWLINK message
1343 * to the routing socket.
1345 void netdev_state_change(struct net_device *dev)
1347 if (dev->flags & IFF_UP) {
1348 struct netdev_notifier_change_info change_info = {
1352 call_netdevice_notifiers_info(NETDEV_CHANGE,
1354 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1357 EXPORT_SYMBOL(netdev_state_change);
1360 * netdev_notify_peers - notify network peers about existence of @dev
1361 * @dev: network device
1363 * Generate traffic such that interested network peers are aware of
1364 * @dev, such as by generating a gratuitous ARP. This may be used when
1365 * a device wants to inform the rest of the network about some sort of
1366 * reconfiguration such as a failover event or virtual machine
1369 void netdev_notify_peers(struct net_device *dev)
1372 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1373 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1376 EXPORT_SYMBOL(netdev_notify_peers);
1378 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1380 const struct net_device_ops *ops = dev->netdev_ops;
1385 if (!netif_device_present(dev))
1388 /* Block netpoll from trying to do any rx path servicing.
1389 * If we don't do this there is a chance ndo_poll_controller
1390 * or ndo_poll may be running while we open the device
1392 netpoll_poll_disable(dev);
1394 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1395 ret = notifier_to_errno(ret);
1399 set_bit(__LINK_STATE_START, &dev->state);
1401 if (ops->ndo_validate_addr)
1402 ret = ops->ndo_validate_addr(dev);
1404 if (!ret && ops->ndo_open)
1405 ret = ops->ndo_open(dev);
1407 netpoll_poll_enable(dev);
1410 clear_bit(__LINK_STATE_START, &dev->state);
1412 dev->flags |= IFF_UP;
1413 dev_set_rx_mode(dev);
1415 add_device_randomness(dev->dev_addr, dev->addr_len);
1422 * dev_open - prepare an interface for use.
1423 * @dev: device to open
1424 * @extack: netlink extended ack
1426 * Takes a device from down to up state. The device's private open
1427 * function is invoked and then the multicast lists are loaded. Finally
1428 * the device is moved into the up state and a %NETDEV_UP message is
1429 * sent to the netdev notifier chain.
1431 * Calling this function on an active interface is a nop. On a failure
1432 * a negative errno code is returned.
1434 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1438 if (dev->flags & IFF_UP)
1441 ret = __dev_open(dev, extack);
1445 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1446 call_netdevice_notifiers(NETDEV_UP, dev);
1450 EXPORT_SYMBOL(dev_open);
1452 static void __dev_close_many(struct list_head *head)
1454 struct net_device *dev;
1459 list_for_each_entry(dev, head, close_list) {
1460 /* Temporarily disable netpoll until the interface is down */
1461 netpoll_poll_disable(dev);
1463 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1465 clear_bit(__LINK_STATE_START, &dev->state);
1467 /* Synchronize to scheduled poll. We cannot touch poll list, it
1468 * can be even on different cpu. So just clear netif_running().
1470 * dev->stop() will invoke napi_disable() on all of it's
1471 * napi_struct instances on this device.
1473 smp_mb__after_atomic(); /* Commit netif_running(). */
1476 dev_deactivate_many(head);
1478 list_for_each_entry(dev, head, close_list) {
1479 const struct net_device_ops *ops = dev->netdev_ops;
1482 * Call the device specific close. This cannot fail.
1483 * Only if device is UP
1485 * We allow it to be called even after a DETACH hot-plug
1491 dev->flags &= ~IFF_UP;
1492 netpoll_poll_enable(dev);
1496 static void __dev_close(struct net_device *dev)
1500 list_add(&dev->close_list, &single);
1501 __dev_close_many(&single);
1505 void dev_close_many(struct list_head *head, bool unlink)
1507 struct net_device *dev, *tmp;
1509 /* Remove the devices that don't need to be closed */
1510 list_for_each_entry_safe(dev, tmp, head, close_list)
1511 if (!(dev->flags & IFF_UP))
1512 list_del_init(&dev->close_list);
1514 __dev_close_many(head);
1516 list_for_each_entry_safe(dev, tmp, head, close_list) {
1517 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1518 call_netdevice_notifiers(NETDEV_DOWN, dev);
1520 list_del_init(&dev->close_list);
1523 EXPORT_SYMBOL(dev_close_many);
1526 * dev_close - shutdown an interface.
1527 * @dev: device to shutdown
1529 * This function moves an active device into down state. A
1530 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1531 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1534 void dev_close(struct net_device *dev)
1536 if (dev->flags & IFF_UP) {
1539 list_add(&dev->close_list, &single);
1540 dev_close_many(&single, true);
1544 EXPORT_SYMBOL(dev_close);
1548 * dev_disable_lro - disable Large Receive Offload on a device
1551 * Disable Large Receive Offload (LRO) on a net device. Must be
1552 * called under RTNL. This is needed if received packets may be
1553 * forwarded to another interface.
1555 void dev_disable_lro(struct net_device *dev)
1557 struct net_device *lower_dev;
1558 struct list_head *iter;
1560 dev->wanted_features &= ~NETIF_F_LRO;
1561 netdev_update_features(dev);
1563 if (unlikely(dev->features & NETIF_F_LRO))
1564 netdev_WARN(dev, "failed to disable LRO!\n");
1566 netdev_for_each_lower_dev(dev, lower_dev, iter)
1567 dev_disable_lro(lower_dev);
1569 EXPORT_SYMBOL(dev_disable_lro);
1572 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1575 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1576 * called under RTNL. This is needed if Generic XDP is installed on
1579 static void dev_disable_gro_hw(struct net_device *dev)
1581 dev->wanted_features &= ~NETIF_F_GRO_HW;
1582 netdev_update_features(dev);
1584 if (unlikely(dev->features & NETIF_F_GRO_HW))
1585 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1588 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1591 case NETDEV_##val: \
1592 return "NETDEV_" __stringify(val);
1594 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1595 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1596 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1597 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1598 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1599 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1600 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1601 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1602 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1606 return "UNKNOWN_NETDEV_EVENT";
1608 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1610 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1611 struct net_device *dev)
1613 struct netdev_notifier_info info = {
1617 return nb->notifier_call(nb, val, &info);
1620 static int dev_boot_phase = 1;
1623 * register_netdevice_notifier - register a network notifier block
1626 * Register a notifier to be called when network device events occur.
1627 * The notifier passed is linked into the kernel structures and must
1628 * not be reused until it has been unregistered. A negative errno code
1629 * is returned on a failure.
1631 * When registered all registration and up events are replayed
1632 * to the new notifier to allow device to have a race free
1633 * view of the network device list.
1636 int register_netdevice_notifier(struct notifier_block *nb)
1638 struct net_device *dev;
1639 struct net_device *last;
1643 /* Close race with setup_net() and cleanup_net() */
1644 down_write(&pernet_ops_rwsem);
1646 err = raw_notifier_chain_register(&netdev_chain, nb);
1652 for_each_netdev(net, dev) {
1653 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1654 err = notifier_to_errno(err);
1658 if (!(dev->flags & IFF_UP))
1661 call_netdevice_notifier(nb, NETDEV_UP, dev);
1667 up_write(&pernet_ops_rwsem);
1673 for_each_netdev(net, dev) {
1677 if (dev->flags & IFF_UP) {
1678 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1680 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1682 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1687 raw_notifier_chain_unregister(&netdev_chain, nb);
1690 EXPORT_SYMBOL(register_netdevice_notifier);
1693 * unregister_netdevice_notifier - unregister a network notifier block
1696 * Unregister a notifier previously registered by
1697 * register_netdevice_notifier(). The notifier is unlinked into the
1698 * kernel structures and may then be reused. A negative errno code
1699 * is returned on a failure.
1701 * After unregistering unregister and down device events are synthesized
1702 * for all devices on the device list to the removed notifier to remove
1703 * the need for special case cleanup code.
1706 int unregister_netdevice_notifier(struct notifier_block *nb)
1708 struct net_device *dev;
1712 /* Close race with setup_net() and cleanup_net() */
1713 down_write(&pernet_ops_rwsem);
1715 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1720 for_each_netdev(net, dev) {
1721 if (dev->flags & IFF_UP) {
1722 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1724 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1726 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1731 up_write(&pernet_ops_rwsem);
1734 EXPORT_SYMBOL(unregister_netdevice_notifier);
1737 * call_netdevice_notifiers_info - call all network notifier blocks
1738 * @val: value passed unmodified to notifier function
1739 * @info: notifier information data
1741 * Call all network notifier blocks. Parameters and return value
1742 * are as for raw_notifier_call_chain().
1745 static int call_netdevice_notifiers_info(unsigned long val,
1746 struct netdev_notifier_info *info)
1749 return raw_notifier_call_chain(&netdev_chain, val, info);
1752 static int call_netdevice_notifiers_extack(unsigned long val,
1753 struct net_device *dev,
1754 struct netlink_ext_ack *extack)
1756 struct netdev_notifier_info info = {
1761 return call_netdevice_notifiers_info(val, &info);
1765 * call_netdevice_notifiers - call all network notifier blocks
1766 * @val: value passed unmodified to notifier function
1767 * @dev: net_device pointer passed unmodified to notifier function
1769 * Call all network notifier blocks. Parameters and return value
1770 * are as for raw_notifier_call_chain().
1773 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1775 return call_netdevice_notifiers_extack(val, dev, NULL);
1777 EXPORT_SYMBOL(call_netdevice_notifiers);
1780 * call_netdevice_notifiers_mtu - call all network notifier blocks
1781 * @val: value passed unmodified to notifier function
1782 * @dev: net_device pointer passed unmodified to notifier function
1783 * @arg: additional u32 argument passed to the notifier function
1785 * Call all network notifier blocks. Parameters and return value
1786 * are as for raw_notifier_call_chain().
1788 static int call_netdevice_notifiers_mtu(unsigned long val,
1789 struct net_device *dev, u32 arg)
1791 struct netdev_notifier_info_ext info = {
1796 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1798 return call_netdevice_notifiers_info(val, &info.info);
1801 #ifdef CONFIG_NET_INGRESS
1802 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1804 void net_inc_ingress_queue(void)
1806 static_branch_inc(&ingress_needed_key);
1808 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1810 void net_dec_ingress_queue(void)
1812 static_branch_dec(&ingress_needed_key);
1814 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1817 #ifdef CONFIG_NET_EGRESS
1818 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1820 void net_inc_egress_queue(void)
1822 static_branch_inc(&egress_needed_key);
1824 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1826 void net_dec_egress_queue(void)
1828 static_branch_dec(&egress_needed_key);
1830 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1833 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1834 #ifdef CONFIG_JUMP_LABEL
1835 static atomic_t netstamp_needed_deferred;
1836 static atomic_t netstamp_wanted;
1837 static void netstamp_clear(struct work_struct *work)
1839 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1842 wanted = atomic_add_return(deferred, &netstamp_wanted);
1844 static_branch_enable(&netstamp_needed_key);
1846 static_branch_disable(&netstamp_needed_key);
1848 static DECLARE_WORK(netstamp_work, netstamp_clear);
1851 void net_enable_timestamp(void)
1853 #ifdef CONFIG_JUMP_LABEL
1857 wanted = atomic_read(&netstamp_wanted);
1860 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1863 atomic_inc(&netstamp_needed_deferred);
1864 schedule_work(&netstamp_work);
1866 static_branch_inc(&netstamp_needed_key);
1869 EXPORT_SYMBOL(net_enable_timestamp);
1871 void net_disable_timestamp(void)
1873 #ifdef CONFIG_JUMP_LABEL
1877 wanted = atomic_read(&netstamp_wanted);
1880 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1883 atomic_dec(&netstamp_needed_deferred);
1884 schedule_work(&netstamp_work);
1886 static_branch_dec(&netstamp_needed_key);
1889 EXPORT_SYMBOL(net_disable_timestamp);
1891 static inline void net_timestamp_set(struct sk_buff *skb)
1894 if (static_branch_unlikely(&netstamp_needed_key))
1895 __net_timestamp(skb);
1898 #define net_timestamp_check(COND, SKB) \
1899 if (static_branch_unlikely(&netstamp_needed_key)) { \
1900 if ((COND) && !(SKB)->tstamp) \
1901 __net_timestamp(SKB); \
1904 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1908 if (!(dev->flags & IFF_UP))
1911 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1912 if (skb->len <= len)
1915 /* if TSO is enabled, we don't care about the length as the packet
1916 * could be forwarded without being segmented before
1918 if (skb_is_gso(skb))
1923 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1925 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1927 int ret = ____dev_forward_skb(dev, skb);
1930 skb->protocol = eth_type_trans(skb, dev);
1931 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1936 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1939 * dev_forward_skb - loopback an skb to another netif
1941 * @dev: destination network device
1942 * @skb: buffer to forward
1945 * NET_RX_SUCCESS (no congestion)
1946 * NET_RX_DROP (packet was dropped, but freed)
1948 * dev_forward_skb can be used for injecting an skb from the
1949 * start_xmit function of one device into the receive queue
1950 * of another device.
1952 * The receiving device may be in another namespace, so
1953 * we have to clear all information in the skb that could
1954 * impact namespace isolation.
1956 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1958 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1960 EXPORT_SYMBOL_GPL(dev_forward_skb);
1962 static inline int deliver_skb(struct sk_buff *skb,
1963 struct packet_type *pt_prev,
1964 struct net_device *orig_dev)
1966 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1968 refcount_inc(&skb->users);
1969 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1972 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1973 struct packet_type **pt,
1974 struct net_device *orig_dev,
1976 struct list_head *ptype_list)
1978 struct packet_type *ptype, *pt_prev = *pt;
1980 list_for_each_entry_rcu(ptype, ptype_list, list) {
1981 if (ptype->type != type)
1984 deliver_skb(skb, pt_prev, orig_dev);
1990 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1992 if (!ptype->af_packet_priv || !skb->sk)
1995 if (ptype->id_match)
1996 return ptype->id_match(ptype, skb->sk);
1997 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2004 * dev_nit_active - return true if any network interface taps are in use
2006 * @dev: network device to check for the presence of taps
2008 bool dev_nit_active(struct net_device *dev)
2010 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2012 EXPORT_SYMBOL_GPL(dev_nit_active);
2015 * Support routine. Sends outgoing frames to any network
2016 * taps currently in use.
2019 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2021 struct packet_type *ptype;
2022 struct sk_buff *skb2 = NULL;
2023 struct packet_type *pt_prev = NULL;
2024 struct list_head *ptype_list = &ptype_all;
2028 list_for_each_entry_rcu(ptype, ptype_list, list) {
2029 if (ptype->ignore_outgoing)
2032 /* Never send packets back to the socket
2033 * they originated from - MvS (miquels@drinkel.ow.org)
2035 if (skb_loop_sk(ptype, skb))
2039 deliver_skb(skb2, pt_prev, skb->dev);
2044 /* need to clone skb, done only once */
2045 skb2 = skb_clone(skb, GFP_ATOMIC);
2049 net_timestamp_set(skb2);
2051 /* skb->nh should be correctly
2052 * set by sender, so that the second statement is
2053 * just protection against buggy protocols.
2055 skb_reset_mac_header(skb2);
2057 if (skb_network_header(skb2) < skb2->data ||
2058 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2059 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2060 ntohs(skb2->protocol),
2062 skb_reset_network_header(skb2);
2065 skb2->transport_header = skb2->network_header;
2066 skb2->pkt_type = PACKET_OUTGOING;
2070 if (ptype_list == &ptype_all) {
2071 ptype_list = &dev->ptype_all;
2076 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2077 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2083 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2086 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2087 * @dev: Network device
2088 * @txq: number of queues available
2090 * If real_num_tx_queues is changed the tc mappings may no longer be
2091 * valid. To resolve this verify the tc mapping remains valid and if
2092 * not NULL the mapping. With no priorities mapping to this
2093 * offset/count pair it will no longer be used. In the worst case TC0
2094 * is invalid nothing can be done so disable priority mappings. If is
2095 * expected that drivers will fix this mapping if they can before
2096 * calling netif_set_real_num_tx_queues.
2098 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2101 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2103 /* If TC0 is invalidated disable TC mapping */
2104 if (tc->offset + tc->count > txq) {
2105 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2110 /* Invalidated prio to tc mappings set to TC0 */
2111 for (i = 1; i < TC_BITMASK + 1; i++) {
2112 int q = netdev_get_prio_tc_map(dev, i);
2114 tc = &dev->tc_to_txq[q];
2115 if (tc->offset + tc->count > txq) {
2116 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2118 netdev_set_prio_tc_map(dev, i, 0);
2123 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2126 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2129 /* walk through the TCs and see if it falls into any of them */
2130 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2131 if ((txq - tc->offset) < tc->count)
2135 /* didn't find it, just return -1 to indicate no match */
2141 EXPORT_SYMBOL(netdev_txq_to_tc);
2144 struct static_key xps_needed __read_mostly;
2145 EXPORT_SYMBOL(xps_needed);
2146 struct static_key xps_rxqs_needed __read_mostly;
2147 EXPORT_SYMBOL(xps_rxqs_needed);
2148 static DEFINE_MUTEX(xps_map_mutex);
2149 #define xmap_dereference(P) \
2150 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2152 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2155 struct xps_map *map = NULL;
2159 map = xmap_dereference(dev_maps->attr_map[tci]);
2163 for (pos = map->len; pos--;) {
2164 if (map->queues[pos] != index)
2168 map->queues[pos] = map->queues[--map->len];
2172 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2173 kfree_rcu(map, rcu);
2180 static bool remove_xps_queue_cpu(struct net_device *dev,
2181 struct xps_dev_maps *dev_maps,
2182 int cpu, u16 offset, u16 count)
2184 int num_tc = dev->num_tc ? : 1;
2185 bool active = false;
2188 for (tci = cpu * num_tc; num_tc--; tci++) {
2191 for (i = count, j = offset; i--; j++) {
2192 if (!remove_xps_queue(dev_maps, tci, j))
2202 static void reset_xps_maps(struct net_device *dev,
2203 struct xps_dev_maps *dev_maps,
2207 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2208 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2210 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2212 static_key_slow_dec_cpuslocked(&xps_needed);
2213 kfree_rcu(dev_maps, rcu);
2216 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2217 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2218 u16 offset, u16 count, bool is_rxqs_map)
2220 bool active = false;
2223 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2225 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2228 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2231 for (i = offset + (count - 1); count--; i--) {
2232 netdev_queue_numa_node_write(
2233 netdev_get_tx_queue(dev, i),
2239 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2242 const unsigned long *possible_mask = NULL;
2243 struct xps_dev_maps *dev_maps;
2244 unsigned int nr_ids;
2246 if (!static_key_false(&xps_needed))
2250 mutex_lock(&xps_map_mutex);
2252 if (static_key_false(&xps_rxqs_needed)) {
2253 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2255 nr_ids = dev->num_rx_queues;
2256 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2257 offset, count, true);
2261 dev_maps = xmap_dereference(dev->xps_cpus_map);
2265 if (num_possible_cpus() > 1)
2266 possible_mask = cpumask_bits(cpu_possible_mask);
2267 nr_ids = nr_cpu_ids;
2268 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2272 mutex_unlock(&xps_map_mutex);
2276 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2278 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2281 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2282 u16 index, bool is_rxqs_map)
2284 struct xps_map *new_map;
2285 int alloc_len = XPS_MIN_MAP_ALLOC;
2288 for (pos = 0; map && pos < map->len; pos++) {
2289 if (map->queues[pos] != index)
2294 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2296 if (pos < map->alloc_len)
2299 alloc_len = map->alloc_len * 2;
2302 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2306 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2308 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2309 cpu_to_node(attr_index));
2313 for (i = 0; i < pos; i++)
2314 new_map->queues[i] = map->queues[i];
2315 new_map->alloc_len = alloc_len;
2321 /* Must be called under cpus_read_lock */
2322 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2323 u16 index, bool is_rxqs_map)
2325 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2326 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2327 int i, j, tci, numa_node_id = -2;
2328 int maps_sz, num_tc = 1, tc = 0;
2329 struct xps_map *map, *new_map;
2330 bool active = false;
2331 unsigned int nr_ids;
2334 /* Do not allow XPS on subordinate device directly */
2335 num_tc = dev->num_tc;
2339 /* If queue belongs to subordinate dev use its map */
2340 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2342 tc = netdev_txq_to_tc(dev, index);
2347 mutex_lock(&xps_map_mutex);
2349 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2350 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2351 nr_ids = dev->num_rx_queues;
2353 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2354 if (num_possible_cpus() > 1) {
2355 online_mask = cpumask_bits(cpu_online_mask);
2356 possible_mask = cpumask_bits(cpu_possible_mask);
2358 dev_maps = xmap_dereference(dev->xps_cpus_map);
2359 nr_ids = nr_cpu_ids;
2362 if (maps_sz < L1_CACHE_BYTES)
2363 maps_sz = L1_CACHE_BYTES;
2365 /* allocate memory for queue storage */
2366 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2369 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2370 if (!new_dev_maps) {
2371 mutex_unlock(&xps_map_mutex);
2375 tci = j * num_tc + tc;
2376 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2379 map = expand_xps_map(map, j, index, is_rxqs_map);
2383 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2387 goto out_no_new_maps;
2390 /* Increment static keys at most once per type */
2391 static_key_slow_inc_cpuslocked(&xps_needed);
2393 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2396 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2398 /* copy maps belonging to foreign traffic classes */
2399 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2400 /* fill in the new device map from the old device map */
2401 map = xmap_dereference(dev_maps->attr_map[tci]);
2402 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2405 /* We need to explicitly update tci as prevous loop
2406 * could break out early if dev_maps is NULL.
2408 tci = j * num_tc + tc;
2410 if (netif_attr_test_mask(j, mask, nr_ids) &&
2411 netif_attr_test_online(j, online_mask, nr_ids)) {
2412 /* add tx-queue to CPU/rx-queue maps */
2415 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2416 while ((pos < map->len) && (map->queues[pos] != index))
2419 if (pos == map->len)
2420 map->queues[map->len++] = index;
2423 if (numa_node_id == -2)
2424 numa_node_id = cpu_to_node(j);
2425 else if (numa_node_id != cpu_to_node(j))
2429 } else if (dev_maps) {
2430 /* fill in the new device map from the old device map */
2431 map = xmap_dereference(dev_maps->attr_map[tci]);
2432 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2435 /* copy maps belonging to foreign traffic classes */
2436 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2437 /* fill in the new device map from the old device map */
2438 map = xmap_dereference(dev_maps->attr_map[tci]);
2439 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2444 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2446 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2448 /* Cleanup old maps */
2450 goto out_no_old_maps;
2452 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2454 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2455 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2456 map = xmap_dereference(dev_maps->attr_map[tci]);
2457 if (map && map != new_map)
2458 kfree_rcu(map, rcu);
2462 kfree_rcu(dev_maps, rcu);
2465 dev_maps = new_dev_maps;
2470 /* update Tx queue numa node */
2471 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2472 (numa_node_id >= 0) ?
2473 numa_node_id : NUMA_NO_NODE);
2479 /* removes tx-queue from unused CPUs/rx-queues */
2480 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2482 for (i = tc, tci = j * num_tc; i--; tci++)
2483 active |= remove_xps_queue(dev_maps, tci, index);
2484 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2485 !netif_attr_test_online(j, online_mask, nr_ids))
2486 active |= remove_xps_queue(dev_maps, tci, index);
2487 for (i = num_tc - tc, tci++; --i; tci++)
2488 active |= remove_xps_queue(dev_maps, tci, index);
2491 /* free map if not active */
2493 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2496 mutex_unlock(&xps_map_mutex);
2500 /* remove any maps that we added */
2501 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2503 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2504 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2506 xmap_dereference(dev_maps->attr_map[tci]) :
2508 if (new_map && new_map != map)
2513 mutex_unlock(&xps_map_mutex);
2515 kfree(new_dev_maps);
2518 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2520 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2526 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2531 EXPORT_SYMBOL(netif_set_xps_queue);
2534 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2536 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2538 /* Unbind any subordinate channels */
2539 while (txq-- != &dev->_tx[0]) {
2541 netdev_unbind_sb_channel(dev, txq->sb_dev);
2545 void netdev_reset_tc(struct net_device *dev)
2548 netif_reset_xps_queues_gt(dev, 0);
2550 netdev_unbind_all_sb_channels(dev);
2552 /* Reset TC configuration of device */
2554 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2555 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2557 EXPORT_SYMBOL(netdev_reset_tc);
2559 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2561 if (tc >= dev->num_tc)
2565 netif_reset_xps_queues(dev, offset, count);
2567 dev->tc_to_txq[tc].count = count;
2568 dev->tc_to_txq[tc].offset = offset;
2571 EXPORT_SYMBOL(netdev_set_tc_queue);
2573 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2575 if (num_tc > TC_MAX_QUEUE)
2579 netif_reset_xps_queues_gt(dev, 0);
2581 netdev_unbind_all_sb_channels(dev);
2583 dev->num_tc = num_tc;
2586 EXPORT_SYMBOL(netdev_set_num_tc);
2588 void netdev_unbind_sb_channel(struct net_device *dev,
2589 struct net_device *sb_dev)
2591 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2594 netif_reset_xps_queues_gt(sb_dev, 0);
2596 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2597 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2599 while (txq-- != &dev->_tx[0]) {
2600 if (txq->sb_dev == sb_dev)
2604 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2606 int netdev_bind_sb_channel_queue(struct net_device *dev,
2607 struct net_device *sb_dev,
2608 u8 tc, u16 count, u16 offset)
2610 /* Make certain the sb_dev and dev are already configured */
2611 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2614 /* We cannot hand out queues we don't have */
2615 if ((offset + count) > dev->real_num_tx_queues)
2618 /* Record the mapping */
2619 sb_dev->tc_to_txq[tc].count = count;
2620 sb_dev->tc_to_txq[tc].offset = offset;
2622 /* Provide a way for Tx queue to find the tc_to_txq map or
2623 * XPS map for itself.
2626 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2630 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2632 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2634 /* Do not use a multiqueue device to represent a subordinate channel */
2635 if (netif_is_multiqueue(dev))
2638 /* We allow channels 1 - 32767 to be used for subordinate channels.
2639 * Channel 0 is meant to be "native" mode and used only to represent
2640 * the main root device. We allow writing 0 to reset the device back
2641 * to normal mode after being used as a subordinate channel.
2643 if (channel > S16_MAX)
2646 dev->num_tc = -channel;
2650 EXPORT_SYMBOL(netdev_set_sb_channel);
2653 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2654 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2656 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2661 disabling = txq < dev->real_num_tx_queues;
2663 if (txq < 1 || txq > dev->num_tx_queues)
2666 if (dev->reg_state == NETREG_REGISTERED ||
2667 dev->reg_state == NETREG_UNREGISTERING) {
2670 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2676 netif_setup_tc(dev, txq);
2678 dev->real_num_tx_queues = txq;
2682 qdisc_reset_all_tx_gt(dev, txq);
2684 netif_reset_xps_queues_gt(dev, txq);
2688 dev->real_num_tx_queues = txq;
2693 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2697 * netif_set_real_num_rx_queues - set actual number of RX queues used
2698 * @dev: Network device
2699 * @rxq: Actual number of RX queues
2701 * This must be called either with the rtnl_lock held or before
2702 * registration of the net device. Returns 0 on success, or a
2703 * negative error code. If called before registration, it always
2706 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2710 if (rxq < 1 || rxq > dev->num_rx_queues)
2713 if (dev->reg_state == NETREG_REGISTERED) {
2716 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2722 dev->real_num_rx_queues = rxq;
2725 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2729 * netif_get_num_default_rss_queues - default number of RSS queues
2731 * This routine should set an upper limit on the number of RSS queues
2732 * used by default by multiqueue devices.
2734 int netif_get_num_default_rss_queues(void)
2736 return is_kdump_kernel() ?
2737 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2739 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2741 static void __netif_reschedule(struct Qdisc *q)
2743 struct softnet_data *sd;
2744 unsigned long flags;
2746 local_irq_save(flags);
2747 sd = this_cpu_ptr(&softnet_data);
2748 q->next_sched = NULL;
2749 *sd->output_queue_tailp = q;
2750 sd->output_queue_tailp = &q->next_sched;
2751 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2752 local_irq_restore(flags);
2755 void __netif_schedule(struct Qdisc *q)
2757 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2758 __netif_reschedule(q);
2760 EXPORT_SYMBOL(__netif_schedule);
2762 struct dev_kfree_skb_cb {
2763 enum skb_free_reason reason;
2766 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2768 return (struct dev_kfree_skb_cb *)skb->cb;
2771 void netif_schedule_queue(struct netdev_queue *txq)
2774 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2775 struct Qdisc *q = rcu_dereference(txq->qdisc);
2777 __netif_schedule(q);
2781 EXPORT_SYMBOL(netif_schedule_queue);
2783 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2785 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2789 q = rcu_dereference(dev_queue->qdisc);
2790 __netif_schedule(q);
2794 EXPORT_SYMBOL(netif_tx_wake_queue);
2796 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2798 unsigned long flags;
2803 if (likely(refcount_read(&skb->users) == 1)) {
2805 refcount_set(&skb->users, 0);
2806 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2809 get_kfree_skb_cb(skb)->reason = reason;
2810 local_irq_save(flags);
2811 skb->next = __this_cpu_read(softnet_data.completion_queue);
2812 __this_cpu_write(softnet_data.completion_queue, skb);
2813 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2814 local_irq_restore(flags);
2816 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2818 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2820 if (in_irq() || irqs_disabled())
2821 __dev_kfree_skb_irq(skb, reason);
2825 EXPORT_SYMBOL(__dev_kfree_skb_any);
2829 * netif_device_detach - mark device as removed
2830 * @dev: network device
2832 * Mark device as removed from system and therefore no longer available.
2834 void netif_device_detach(struct net_device *dev)
2836 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2837 netif_running(dev)) {
2838 netif_tx_stop_all_queues(dev);
2841 EXPORT_SYMBOL(netif_device_detach);
2844 * netif_device_attach - mark device as attached
2845 * @dev: network device
2847 * Mark device as attached from system and restart if needed.
2849 void netif_device_attach(struct net_device *dev)
2851 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2852 netif_running(dev)) {
2853 netif_tx_wake_all_queues(dev);
2854 __netdev_watchdog_up(dev);
2857 EXPORT_SYMBOL(netif_device_attach);
2860 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2861 * to be used as a distribution range.
2863 static u16 skb_tx_hash(const struct net_device *dev,
2864 const struct net_device *sb_dev,
2865 struct sk_buff *skb)
2869 u16 qcount = dev->real_num_tx_queues;
2872 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2874 qoffset = sb_dev->tc_to_txq[tc].offset;
2875 qcount = sb_dev->tc_to_txq[tc].count;
2878 if (skb_rx_queue_recorded(skb)) {
2879 hash = skb_get_rx_queue(skb);
2880 while (unlikely(hash >= qcount))
2882 return hash + qoffset;
2885 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2888 static void skb_warn_bad_offload(const struct sk_buff *skb)
2890 static const netdev_features_t null_features;
2891 struct net_device *dev = skb->dev;
2892 const char *name = "";
2894 if (!net_ratelimit())
2898 if (dev->dev.parent)
2899 name = dev_driver_string(dev->dev.parent);
2901 name = netdev_name(dev);
2903 skb_dump(KERN_WARNING, skb, false);
2904 WARN(1, "%s: caps=(%pNF, %pNF)\n",
2905 name, dev ? &dev->features : &null_features,
2906 skb->sk ? &skb->sk->sk_route_caps : &null_features);
2910 * Invalidate hardware checksum when packet is to be mangled, and
2911 * complete checksum manually on outgoing path.
2913 int skb_checksum_help(struct sk_buff *skb)
2916 int ret = 0, offset;
2918 if (skb->ip_summed == CHECKSUM_COMPLETE)
2919 goto out_set_summed;
2921 if (unlikely(skb_shinfo(skb)->gso_size)) {
2922 skb_warn_bad_offload(skb);
2926 /* Before computing a checksum, we should make sure no frag could
2927 * be modified by an external entity : checksum could be wrong.
2929 if (skb_has_shared_frag(skb)) {
2930 ret = __skb_linearize(skb);
2935 offset = skb_checksum_start_offset(skb);
2936 BUG_ON(offset >= skb_headlen(skb));
2937 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2939 offset += skb->csum_offset;
2940 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2942 if (skb_cloned(skb) &&
2943 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2944 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2949 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2951 skb->ip_summed = CHECKSUM_NONE;
2955 EXPORT_SYMBOL(skb_checksum_help);
2957 int skb_crc32c_csum_help(struct sk_buff *skb)
2960 int ret = 0, offset, start;
2962 if (skb->ip_summed != CHECKSUM_PARTIAL)
2965 if (unlikely(skb_is_gso(skb)))
2968 /* Before computing a checksum, we should make sure no frag could
2969 * be modified by an external entity : checksum could be wrong.
2971 if (unlikely(skb_has_shared_frag(skb))) {
2972 ret = __skb_linearize(skb);
2976 start = skb_checksum_start_offset(skb);
2977 offset = start + offsetof(struct sctphdr, checksum);
2978 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2982 if (skb_cloned(skb) &&
2983 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2984 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2988 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2989 skb->len - start, ~(__u32)0,
2991 *(__le32 *)(skb->data + offset) = crc32c_csum;
2992 skb->ip_summed = CHECKSUM_NONE;
2993 skb->csum_not_inet = 0;
2998 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3000 __be16 type = skb->protocol;
3002 /* Tunnel gso handlers can set protocol to ethernet. */
3003 if (type == htons(ETH_P_TEB)) {
3006 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3009 eth = (struct ethhdr *)skb->data;
3010 type = eth->h_proto;
3013 return __vlan_get_protocol(skb, type, depth);
3017 * skb_mac_gso_segment - mac layer segmentation handler.
3018 * @skb: buffer to segment
3019 * @features: features for the output path (see dev->features)
3021 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3022 netdev_features_t features)
3024 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3025 struct packet_offload *ptype;
3026 int vlan_depth = skb->mac_len;
3027 __be16 type = skb_network_protocol(skb, &vlan_depth);
3029 if (unlikely(!type))
3030 return ERR_PTR(-EINVAL);
3032 __skb_pull(skb, vlan_depth);
3035 list_for_each_entry_rcu(ptype, &offload_base, list) {
3036 if (ptype->type == type && ptype->callbacks.gso_segment) {
3037 segs = ptype->callbacks.gso_segment(skb, features);
3043 __skb_push(skb, skb->data - skb_mac_header(skb));
3047 EXPORT_SYMBOL(skb_mac_gso_segment);
3050 /* openvswitch calls this on rx path, so we need a different check.
3052 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3055 return skb->ip_summed != CHECKSUM_PARTIAL &&
3056 skb->ip_summed != CHECKSUM_UNNECESSARY;
3058 return skb->ip_summed == CHECKSUM_NONE;
3062 * __skb_gso_segment - Perform segmentation on skb.
3063 * @skb: buffer to segment
3064 * @features: features for the output path (see dev->features)
3065 * @tx_path: whether it is called in TX path
3067 * This function segments the given skb and returns a list of segments.
3069 * It may return NULL if the skb requires no segmentation. This is
3070 * only possible when GSO is used for verifying header integrity.
3072 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3074 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3075 netdev_features_t features, bool tx_path)
3077 struct sk_buff *segs;
3079 if (unlikely(skb_needs_check(skb, tx_path))) {
3082 /* We're going to init ->check field in TCP or UDP header */
3083 err = skb_cow_head(skb, 0);
3085 return ERR_PTR(err);
3088 /* Only report GSO partial support if it will enable us to
3089 * support segmentation on this frame without needing additional
3092 if (features & NETIF_F_GSO_PARTIAL) {
3093 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3094 struct net_device *dev = skb->dev;
3096 partial_features |= dev->features & dev->gso_partial_features;
3097 if (!skb_gso_ok(skb, features | partial_features))
3098 features &= ~NETIF_F_GSO_PARTIAL;
3101 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3102 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3104 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3105 SKB_GSO_CB(skb)->encap_level = 0;
3107 skb_reset_mac_header(skb);
3108 skb_reset_mac_len(skb);
3110 segs = skb_mac_gso_segment(skb, features);
3112 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3113 skb_warn_bad_offload(skb);
3117 EXPORT_SYMBOL(__skb_gso_segment);
3119 /* Take action when hardware reception checksum errors are detected. */
3121 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3123 if (net_ratelimit()) {
3124 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3125 skb_dump(KERN_ERR, skb, true);
3129 EXPORT_SYMBOL(netdev_rx_csum_fault);
3132 /* XXX: check that highmem exists at all on the given machine. */
3133 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3135 #ifdef CONFIG_HIGHMEM
3138 if (!(dev->features & NETIF_F_HIGHDMA)) {
3139 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3140 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3142 if (PageHighMem(skb_frag_page(frag)))
3150 /* If MPLS offload request, verify we are testing hardware MPLS features
3151 * instead of standard features for the netdev.
3153 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3154 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3155 netdev_features_t features,
3158 if (eth_p_mpls(type))
3159 features &= skb->dev->mpls_features;
3164 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3165 netdev_features_t features,
3172 static netdev_features_t harmonize_features(struct sk_buff *skb,
3173 netdev_features_t features)
3178 type = skb_network_protocol(skb, &tmp);
3179 features = net_mpls_features(skb, features, type);
3181 if (skb->ip_summed != CHECKSUM_NONE &&
3182 !can_checksum_protocol(features, type)) {
3183 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3185 if (illegal_highdma(skb->dev, skb))
3186 features &= ~NETIF_F_SG;
3191 netdev_features_t passthru_features_check(struct sk_buff *skb,
3192 struct net_device *dev,
3193 netdev_features_t features)
3197 EXPORT_SYMBOL(passthru_features_check);
3199 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3200 struct net_device *dev,
3201 netdev_features_t features)
3203 return vlan_features_check(skb, features);
3206 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3207 struct net_device *dev,
3208 netdev_features_t features)
3210 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3212 if (gso_segs > dev->gso_max_segs)
3213 return features & ~NETIF_F_GSO_MASK;
3215 /* Support for GSO partial features requires software
3216 * intervention before we can actually process the packets
3217 * so we need to strip support for any partial features now
3218 * and we can pull them back in after we have partially
3219 * segmented the frame.
3221 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3222 features &= ~dev->gso_partial_features;
3224 /* Make sure to clear the IPv4 ID mangling feature if the
3225 * IPv4 header has the potential to be fragmented.
3227 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3228 struct iphdr *iph = skb->encapsulation ?
3229 inner_ip_hdr(skb) : ip_hdr(skb);
3231 if (!(iph->frag_off & htons(IP_DF)))
3232 features &= ~NETIF_F_TSO_MANGLEID;
3238 netdev_features_t netif_skb_features(struct sk_buff *skb)
3240 struct net_device *dev = skb->dev;
3241 netdev_features_t features = dev->features;
3243 if (skb_is_gso(skb))
3244 features = gso_features_check(skb, dev, features);
3246 /* If encapsulation offload request, verify we are testing
3247 * hardware encapsulation features instead of standard
3248 * features for the netdev
3250 if (skb->encapsulation)
3251 features &= dev->hw_enc_features;
3253 if (skb_vlan_tagged(skb))
3254 features = netdev_intersect_features(features,
3255 dev->vlan_features |
3256 NETIF_F_HW_VLAN_CTAG_TX |
3257 NETIF_F_HW_VLAN_STAG_TX);
3259 if (dev->netdev_ops->ndo_features_check)
3260 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3263 features &= dflt_features_check(skb, dev, features);
3265 return harmonize_features(skb, features);
3267 EXPORT_SYMBOL(netif_skb_features);
3269 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3270 struct netdev_queue *txq, bool more)
3275 if (dev_nit_active(dev))
3276 dev_queue_xmit_nit(skb, dev);
3279 trace_net_dev_start_xmit(skb, dev);
3280 rc = netdev_start_xmit(skb, dev, txq, more);
3281 trace_net_dev_xmit(skb, rc, dev, len);
3286 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3287 struct netdev_queue *txq, int *ret)
3289 struct sk_buff *skb = first;
3290 int rc = NETDEV_TX_OK;
3293 struct sk_buff *next = skb->next;
3295 skb_mark_not_on_list(skb);
3296 rc = xmit_one(skb, dev, txq, next != NULL);
3297 if (unlikely(!dev_xmit_complete(rc))) {
3303 if (netif_tx_queue_stopped(txq) && skb) {
3304 rc = NETDEV_TX_BUSY;
3314 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3315 netdev_features_t features)
3317 if (skb_vlan_tag_present(skb) &&
3318 !vlan_hw_offload_capable(features, skb->vlan_proto))
3319 skb = __vlan_hwaccel_push_inside(skb);
3323 int skb_csum_hwoffload_help(struct sk_buff *skb,
3324 const netdev_features_t features)
3326 if (unlikely(skb->csum_not_inet))
3327 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3328 skb_crc32c_csum_help(skb);
3330 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3332 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3334 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3336 netdev_features_t features;
3338 features = netif_skb_features(skb);
3339 skb = validate_xmit_vlan(skb, features);
3343 skb = sk_validate_xmit_skb(skb, dev);
3347 if (netif_needs_gso(skb, features)) {
3348 struct sk_buff *segs;
3350 segs = skb_gso_segment(skb, features);
3358 if (skb_needs_linearize(skb, features) &&
3359 __skb_linearize(skb))
3362 /* If packet is not checksummed and device does not
3363 * support checksumming for this protocol, complete
3364 * checksumming here.
3366 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3367 if (skb->encapsulation)
3368 skb_set_inner_transport_header(skb,
3369 skb_checksum_start_offset(skb));
3371 skb_set_transport_header(skb,
3372 skb_checksum_start_offset(skb));
3373 if (skb_csum_hwoffload_help(skb, features))
3378 skb = validate_xmit_xfrm(skb, features, again);
3385 atomic_long_inc(&dev->tx_dropped);
3389 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3391 struct sk_buff *next, *head = NULL, *tail;
3393 for (; skb != NULL; skb = next) {
3395 skb_mark_not_on_list(skb);
3397 /* in case skb wont be segmented, point to itself */
3400 skb = validate_xmit_skb(skb, dev, again);
3408 /* If skb was segmented, skb->prev points to
3409 * the last segment. If not, it still contains skb.
3415 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3417 static void qdisc_pkt_len_init(struct sk_buff *skb)
3419 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3421 qdisc_skb_cb(skb)->pkt_len = skb->len;
3423 /* To get more precise estimation of bytes sent on wire,
3424 * we add to pkt_len the headers size of all segments
3426 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3427 unsigned int hdr_len;
3428 u16 gso_segs = shinfo->gso_segs;
3430 /* mac layer + network layer */
3431 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3433 /* + transport layer */
3434 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3435 const struct tcphdr *th;
3436 struct tcphdr _tcphdr;
3438 th = skb_header_pointer(skb, skb_transport_offset(skb),
3439 sizeof(_tcphdr), &_tcphdr);
3441 hdr_len += __tcp_hdrlen(th);
3443 struct udphdr _udphdr;
3445 if (skb_header_pointer(skb, skb_transport_offset(skb),
3446 sizeof(_udphdr), &_udphdr))
3447 hdr_len += sizeof(struct udphdr);
3450 if (shinfo->gso_type & SKB_GSO_DODGY)
3451 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3454 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3458 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3459 struct net_device *dev,
3460 struct netdev_queue *txq)
3462 spinlock_t *root_lock = qdisc_lock(q);
3463 struct sk_buff *to_free = NULL;
3467 qdisc_calculate_pkt_len(skb, q);
3469 if (q->flags & TCQ_F_NOLOCK) {
3470 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3471 __qdisc_drop(skb, &to_free);
3473 } else if ((q->flags & TCQ_F_CAN_BYPASS) && q->empty &&
3474 qdisc_run_begin(q)) {
3475 qdisc_bstats_cpu_update(q, skb);
3477 if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
3481 rc = NET_XMIT_SUCCESS;
3483 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3487 if (unlikely(to_free))
3488 kfree_skb_list(to_free);
3493 * Heuristic to force contended enqueues to serialize on a
3494 * separate lock before trying to get qdisc main lock.
3495 * This permits qdisc->running owner to get the lock more
3496 * often and dequeue packets faster.
3498 contended = qdisc_is_running(q);
3499 if (unlikely(contended))
3500 spin_lock(&q->busylock);
3502 spin_lock(root_lock);
3503 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3504 __qdisc_drop(skb, &to_free);
3506 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3507 qdisc_run_begin(q)) {
3509 * This is a work-conserving queue; there are no old skbs
3510 * waiting to be sent out; and the qdisc is not running -
3511 * xmit the skb directly.
3514 qdisc_bstats_update(q, skb);
3516 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3517 if (unlikely(contended)) {
3518 spin_unlock(&q->busylock);
3525 rc = NET_XMIT_SUCCESS;
3527 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3528 if (qdisc_run_begin(q)) {
3529 if (unlikely(contended)) {
3530 spin_unlock(&q->busylock);
3537 spin_unlock(root_lock);
3538 if (unlikely(to_free))
3539 kfree_skb_list(to_free);
3540 if (unlikely(contended))
3541 spin_unlock(&q->busylock);
3545 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3546 static void skb_update_prio(struct sk_buff *skb)
3548 const struct netprio_map *map;
3549 const struct sock *sk;
3550 unsigned int prioidx;
3554 map = rcu_dereference_bh(skb->dev->priomap);
3557 sk = skb_to_full_sk(skb);
3561 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3563 if (prioidx < map->priomap_len)
3564 skb->priority = map->priomap[prioidx];
3567 #define skb_update_prio(skb)
3571 * dev_loopback_xmit - loop back @skb
3572 * @net: network namespace this loopback is happening in
3573 * @sk: sk needed to be a netfilter okfn
3574 * @skb: buffer to transmit
3576 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3578 skb_reset_mac_header(skb);
3579 __skb_pull(skb, skb_network_offset(skb));
3580 skb->pkt_type = PACKET_LOOPBACK;
3581 skb->ip_summed = CHECKSUM_UNNECESSARY;
3582 WARN_ON(!skb_dst(skb));
3587 EXPORT_SYMBOL(dev_loopback_xmit);
3589 #ifdef CONFIG_NET_EGRESS
3590 static struct sk_buff *
3591 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3593 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3594 struct tcf_result cl_res;
3599 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3600 mini_qdisc_bstats_cpu_update(miniq, skb);
3602 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3604 case TC_ACT_RECLASSIFY:
3605 skb->tc_index = TC_H_MIN(cl_res.classid);
3608 mini_qdisc_qstats_cpu_drop(miniq);
3609 *ret = NET_XMIT_DROP;
3615 *ret = NET_XMIT_SUCCESS;
3618 case TC_ACT_REDIRECT:
3619 /* No need to push/pop skb's mac_header here on egress! */
3620 skb_do_redirect(skb);
3621 *ret = NET_XMIT_SUCCESS;
3629 #endif /* CONFIG_NET_EGRESS */
3632 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3633 struct xps_dev_maps *dev_maps, unsigned int tci)
3635 struct xps_map *map;
3636 int queue_index = -1;
3640 tci += netdev_get_prio_tc_map(dev, skb->priority);
3643 map = rcu_dereference(dev_maps->attr_map[tci]);
3646 queue_index = map->queues[0];
3648 queue_index = map->queues[reciprocal_scale(
3649 skb_get_hash(skb), map->len)];
3650 if (unlikely(queue_index >= dev->real_num_tx_queues))
3657 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3658 struct sk_buff *skb)
3661 struct xps_dev_maps *dev_maps;
3662 struct sock *sk = skb->sk;
3663 int queue_index = -1;
3665 if (!static_key_false(&xps_needed))
3669 if (!static_key_false(&xps_rxqs_needed))
3672 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3674 int tci = sk_rx_queue_get(sk);
3676 if (tci >= 0 && tci < dev->num_rx_queues)
3677 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3682 if (queue_index < 0) {
3683 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3685 unsigned int tci = skb->sender_cpu - 1;
3687 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3699 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3700 struct net_device *sb_dev)
3704 EXPORT_SYMBOL(dev_pick_tx_zero);
3706 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3707 struct net_device *sb_dev)
3709 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3711 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3713 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3714 struct net_device *sb_dev)
3716 struct sock *sk = skb->sk;
3717 int queue_index = sk_tx_queue_get(sk);
3719 sb_dev = sb_dev ? : dev;
3721 if (queue_index < 0 || skb->ooo_okay ||
3722 queue_index >= dev->real_num_tx_queues) {
3723 int new_index = get_xps_queue(dev, sb_dev, skb);
3726 new_index = skb_tx_hash(dev, sb_dev, skb);
3728 if (queue_index != new_index && sk &&
3730 rcu_access_pointer(sk->sk_dst_cache))
3731 sk_tx_queue_set(sk, new_index);
3733 queue_index = new_index;
3738 EXPORT_SYMBOL(netdev_pick_tx);
3740 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3741 struct sk_buff *skb,
3742 struct net_device *sb_dev)
3744 int queue_index = 0;
3747 u32 sender_cpu = skb->sender_cpu - 1;
3749 if (sender_cpu >= (u32)NR_CPUS)
3750 skb->sender_cpu = raw_smp_processor_id() + 1;
3753 if (dev->real_num_tx_queues != 1) {
3754 const struct net_device_ops *ops = dev->netdev_ops;
3756 if (ops->ndo_select_queue)
3757 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3759 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3761 queue_index = netdev_cap_txqueue(dev, queue_index);
3764 skb_set_queue_mapping(skb, queue_index);
3765 return netdev_get_tx_queue(dev, queue_index);
3769 * __dev_queue_xmit - transmit a buffer
3770 * @skb: buffer to transmit
3771 * @sb_dev: suboordinate device used for L2 forwarding offload
3773 * Queue a buffer for transmission to a network device. The caller must
3774 * have set the device and priority and built the buffer before calling
3775 * this function. The function can be called from an interrupt.
3777 * A negative errno code is returned on a failure. A success does not
3778 * guarantee the frame will be transmitted as it may be dropped due
3779 * to congestion or traffic shaping.
3781 * -----------------------------------------------------------------------------------
3782 * I notice this method can also return errors from the queue disciplines,
3783 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3786 * Regardless of the return value, the skb is consumed, so it is currently
3787 * difficult to retry a send to this method. (You can bump the ref count
3788 * before sending to hold a reference for retry if you are careful.)
3790 * When calling this method, interrupts MUST be enabled. This is because
3791 * the BH enable code must have IRQs enabled so that it will not deadlock.
3794 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3796 struct net_device *dev = skb->dev;
3797 struct netdev_queue *txq;
3802 skb_reset_mac_header(skb);
3804 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3805 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3807 /* Disable soft irqs for various locks below. Also
3808 * stops preemption for RCU.
3812 skb_update_prio(skb);
3814 qdisc_pkt_len_init(skb);
3815 #ifdef CONFIG_NET_CLS_ACT
3816 skb->tc_at_ingress = 0;
3817 # ifdef CONFIG_NET_EGRESS
3818 if (static_branch_unlikely(&egress_needed_key)) {
3819 skb = sch_handle_egress(skb, &rc, dev);
3825 /* If device/qdisc don't need skb->dst, release it right now while
3826 * its hot in this cpu cache.
3828 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3833 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3834 q = rcu_dereference_bh(txq->qdisc);
3836 trace_net_dev_queue(skb);
3838 rc = __dev_xmit_skb(skb, q, dev, txq);
3842 /* The device has no queue. Common case for software devices:
3843 * loopback, all the sorts of tunnels...
3845 * Really, it is unlikely that netif_tx_lock protection is necessary
3846 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3848 * However, it is possible, that they rely on protection
3851 * Check this and shot the lock. It is not prone from deadlocks.
3852 *Either shot noqueue qdisc, it is even simpler 8)
3854 if (dev->flags & IFF_UP) {
3855 int cpu = smp_processor_id(); /* ok because BHs are off */
3857 if (txq->xmit_lock_owner != cpu) {
3858 if (dev_xmit_recursion())
3859 goto recursion_alert;
3861 skb = validate_xmit_skb(skb, dev, &again);
3865 HARD_TX_LOCK(dev, txq, cpu);
3867 if (!netif_xmit_stopped(txq)) {
3868 dev_xmit_recursion_inc();
3869 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3870 dev_xmit_recursion_dec();
3871 if (dev_xmit_complete(rc)) {
3872 HARD_TX_UNLOCK(dev, txq);
3876 HARD_TX_UNLOCK(dev, txq);
3877 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3880 /* Recursion is detected! It is possible,
3884 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3890 rcu_read_unlock_bh();
3892 atomic_long_inc(&dev->tx_dropped);
3893 kfree_skb_list(skb);
3896 rcu_read_unlock_bh();
3900 int dev_queue_xmit(struct sk_buff *skb)
3902 return __dev_queue_xmit(skb, NULL);
3904 EXPORT_SYMBOL(dev_queue_xmit);
3906 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3908 return __dev_queue_xmit(skb, sb_dev);
3910 EXPORT_SYMBOL(dev_queue_xmit_accel);
3912 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3914 struct net_device *dev = skb->dev;
3915 struct sk_buff *orig_skb = skb;
3916 struct netdev_queue *txq;
3917 int ret = NETDEV_TX_BUSY;
3920 if (unlikely(!netif_running(dev) ||
3921 !netif_carrier_ok(dev)))
3924 skb = validate_xmit_skb_list(skb, dev, &again);
3925 if (skb != orig_skb)
3928 skb_set_queue_mapping(skb, queue_id);
3929 txq = skb_get_tx_queue(dev, skb);
3933 HARD_TX_LOCK(dev, txq, smp_processor_id());
3934 if (!netif_xmit_frozen_or_drv_stopped(txq))
3935 ret = netdev_start_xmit(skb, dev, txq, false);
3936 HARD_TX_UNLOCK(dev, txq);
3940 if (!dev_xmit_complete(ret))
3945 atomic_long_inc(&dev->tx_dropped);
3946 kfree_skb_list(skb);
3947 return NET_XMIT_DROP;
3949 EXPORT_SYMBOL(dev_direct_xmit);
3951 /*************************************************************************
3953 *************************************************************************/
3955 int netdev_max_backlog __read_mostly = 1000;
3956 EXPORT_SYMBOL(netdev_max_backlog);
3958 int netdev_tstamp_prequeue __read_mostly = 1;
3959 int netdev_budget __read_mostly = 300;
3960 unsigned int __read_mostly netdev_budget_usecs = 2000;
3961 int weight_p __read_mostly = 64; /* old backlog weight */
3962 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3963 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3964 int dev_rx_weight __read_mostly = 64;
3965 int dev_tx_weight __read_mostly = 64;
3967 /* Called with irq disabled */
3968 static inline void ____napi_schedule(struct softnet_data *sd,
3969 struct napi_struct *napi)
3971 list_add_tail(&napi->poll_list, &sd->poll_list);
3972 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3977 /* One global table that all flow-based protocols share. */
3978 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3979 EXPORT_SYMBOL(rps_sock_flow_table);
3980 u32 rps_cpu_mask __read_mostly;
3981 EXPORT_SYMBOL(rps_cpu_mask);
3983 struct static_key_false rps_needed __read_mostly;
3984 EXPORT_SYMBOL(rps_needed);
3985 struct static_key_false rfs_needed __read_mostly;
3986 EXPORT_SYMBOL(rfs_needed);
3988 static struct rps_dev_flow *
3989 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3990 struct rps_dev_flow *rflow, u16 next_cpu)
3992 if (next_cpu < nr_cpu_ids) {
3993 #ifdef CONFIG_RFS_ACCEL
3994 struct netdev_rx_queue *rxqueue;
3995 struct rps_dev_flow_table *flow_table;
3996 struct rps_dev_flow *old_rflow;
4001 /* Should we steer this flow to a different hardware queue? */
4002 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4003 !(dev->features & NETIF_F_NTUPLE))
4005 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4006 if (rxq_index == skb_get_rx_queue(skb))
4009 rxqueue = dev->_rx + rxq_index;
4010 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4013 flow_id = skb_get_hash(skb) & flow_table->mask;
4014 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4015 rxq_index, flow_id);
4019 rflow = &flow_table->flows[flow_id];
4021 if (old_rflow->filter == rflow->filter)
4022 old_rflow->filter = RPS_NO_FILTER;
4026 per_cpu(softnet_data, next_cpu).input_queue_head;
4029 rflow->cpu = next_cpu;
4034 * get_rps_cpu is called from netif_receive_skb and returns the target
4035 * CPU from the RPS map of the receiving queue for a given skb.
4036 * rcu_read_lock must be held on entry.
4038 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4039 struct rps_dev_flow **rflowp)
4041 const struct rps_sock_flow_table *sock_flow_table;
4042 struct netdev_rx_queue *rxqueue = dev->_rx;
4043 struct rps_dev_flow_table *flow_table;
4044 struct rps_map *map;
4049 if (skb_rx_queue_recorded(skb)) {
4050 u16 index = skb_get_rx_queue(skb);
4052 if (unlikely(index >= dev->real_num_rx_queues)) {
4053 WARN_ONCE(dev->real_num_rx_queues > 1,
4054 "%s received packet on queue %u, but number "
4055 "of RX queues is %u\n",
4056 dev->name, index, dev->real_num_rx_queues);
4062 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4064 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4065 map = rcu_dereference(rxqueue->rps_map);
4066 if (!flow_table && !map)
4069 skb_reset_network_header(skb);
4070 hash = skb_get_hash(skb);
4074 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4075 if (flow_table && sock_flow_table) {
4076 struct rps_dev_flow *rflow;
4080 /* First check into global flow table if there is a match */
4081 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4082 if ((ident ^ hash) & ~rps_cpu_mask)
4085 next_cpu = ident & rps_cpu_mask;
4087 /* OK, now we know there is a match,
4088 * we can look at the local (per receive queue) flow table
4090 rflow = &flow_table->flows[hash & flow_table->mask];
4094 * If the desired CPU (where last recvmsg was done) is
4095 * different from current CPU (one in the rx-queue flow
4096 * table entry), switch if one of the following holds:
4097 * - Current CPU is unset (>= nr_cpu_ids).
4098 * - Current CPU is offline.
4099 * - The current CPU's queue tail has advanced beyond the
4100 * last packet that was enqueued using this table entry.
4101 * This guarantees that all previous packets for the flow
4102 * have been dequeued, thus preserving in order delivery.
4104 if (unlikely(tcpu != next_cpu) &&
4105 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4106 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4107 rflow->last_qtail)) >= 0)) {
4109 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4112 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4122 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4123 if (cpu_online(tcpu)) {
4133 #ifdef CONFIG_RFS_ACCEL
4136 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4137 * @dev: Device on which the filter was set
4138 * @rxq_index: RX queue index
4139 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4140 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4142 * Drivers that implement ndo_rx_flow_steer() should periodically call
4143 * this function for each installed filter and remove the filters for
4144 * which it returns %true.
4146 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4147 u32 flow_id, u16 filter_id)
4149 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4150 struct rps_dev_flow_table *flow_table;
4151 struct rps_dev_flow *rflow;
4156 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4157 if (flow_table && flow_id <= flow_table->mask) {
4158 rflow = &flow_table->flows[flow_id];
4159 cpu = READ_ONCE(rflow->cpu);
4160 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4161 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4162 rflow->last_qtail) <
4163 (int)(10 * flow_table->mask)))
4169 EXPORT_SYMBOL(rps_may_expire_flow);
4171 #endif /* CONFIG_RFS_ACCEL */
4173 /* Called from hardirq (IPI) context */
4174 static void rps_trigger_softirq(void *data)
4176 struct softnet_data *sd = data;
4178 ____napi_schedule(sd, &sd->backlog);
4182 #endif /* CONFIG_RPS */
4185 * Check if this softnet_data structure is another cpu one
4186 * If yes, queue it to our IPI list and return 1
4189 static int rps_ipi_queued(struct softnet_data *sd)
4192 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4195 sd->rps_ipi_next = mysd->rps_ipi_list;
4196 mysd->rps_ipi_list = sd;
4198 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4201 #endif /* CONFIG_RPS */
4205 #ifdef CONFIG_NET_FLOW_LIMIT
4206 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4209 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4211 #ifdef CONFIG_NET_FLOW_LIMIT
4212 struct sd_flow_limit *fl;
4213 struct softnet_data *sd;
4214 unsigned int old_flow, new_flow;
4216 if (qlen < (netdev_max_backlog >> 1))
4219 sd = this_cpu_ptr(&softnet_data);
4222 fl = rcu_dereference(sd->flow_limit);
4224 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4225 old_flow = fl->history[fl->history_head];
4226 fl->history[fl->history_head] = new_flow;
4229 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4231 if (likely(fl->buckets[old_flow]))
4232 fl->buckets[old_flow]--;
4234 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4246 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4247 * queue (may be a remote CPU queue).
4249 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4250 unsigned int *qtail)
4252 struct softnet_data *sd;
4253 unsigned long flags;
4256 sd = &per_cpu(softnet_data, cpu);
4258 local_irq_save(flags);
4261 if (!netif_running(skb->dev))
4263 qlen = skb_queue_len(&sd->input_pkt_queue);
4264 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4267 __skb_queue_tail(&sd->input_pkt_queue, skb);
4268 input_queue_tail_incr_save(sd, qtail);
4270 local_irq_restore(flags);
4271 return NET_RX_SUCCESS;
4274 /* Schedule NAPI for backlog device
4275 * We can use non atomic operation since we own the queue lock
4277 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4278 if (!rps_ipi_queued(sd))
4279 ____napi_schedule(sd, &sd->backlog);
4288 local_irq_restore(flags);
4290 atomic_long_inc(&skb->dev->rx_dropped);
4295 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4297 struct net_device *dev = skb->dev;
4298 struct netdev_rx_queue *rxqueue;
4302 if (skb_rx_queue_recorded(skb)) {
4303 u16 index = skb_get_rx_queue(skb);
4305 if (unlikely(index >= dev->real_num_rx_queues)) {
4306 WARN_ONCE(dev->real_num_rx_queues > 1,
4307 "%s received packet on queue %u, but number "
4308 "of RX queues is %u\n",
4309 dev->name, index, dev->real_num_rx_queues);
4311 return rxqueue; /* Return first rxqueue */
4318 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4319 struct xdp_buff *xdp,
4320 struct bpf_prog *xdp_prog)
4322 struct netdev_rx_queue *rxqueue;
4323 void *orig_data, *orig_data_end;
4324 u32 metalen, act = XDP_DROP;
4325 __be16 orig_eth_type;
4331 /* Reinjected packets coming from act_mirred or similar should
4332 * not get XDP generic processing.
4334 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4337 /* XDP packets must be linear and must have sufficient headroom
4338 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4339 * native XDP provides, thus we need to do it here as well.
4341 if (skb_is_nonlinear(skb) ||
4342 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4343 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4344 int troom = skb->tail + skb->data_len - skb->end;
4346 /* In case we have to go down the path and also linearize,
4347 * then lets do the pskb_expand_head() work just once here.
4349 if (pskb_expand_head(skb,
4350 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4351 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4353 if (skb_linearize(skb))
4357 /* The XDP program wants to see the packet starting at the MAC
4360 mac_len = skb->data - skb_mac_header(skb);
4361 hlen = skb_headlen(skb) + mac_len;
4362 xdp->data = skb->data - mac_len;
4363 xdp->data_meta = xdp->data;
4364 xdp->data_end = xdp->data + hlen;
4365 xdp->data_hard_start = skb->data - skb_headroom(skb);
4366 orig_data_end = xdp->data_end;
4367 orig_data = xdp->data;
4368 eth = (struct ethhdr *)xdp->data;
4369 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4370 orig_eth_type = eth->h_proto;
4372 rxqueue = netif_get_rxqueue(skb);
4373 xdp->rxq = &rxqueue->xdp_rxq;
4375 act = bpf_prog_run_xdp(xdp_prog, xdp);
4377 off = xdp->data - orig_data;
4379 __skb_pull(skb, off);
4381 __skb_push(skb, -off);
4382 skb->mac_header += off;
4384 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4387 off = orig_data_end - xdp->data_end;
4389 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4394 /* check if XDP changed eth hdr such SKB needs update */
4395 eth = (struct ethhdr *)xdp->data;
4396 if ((orig_eth_type != eth->h_proto) ||
4397 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4398 __skb_push(skb, ETH_HLEN);
4399 skb->protocol = eth_type_trans(skb, skb->dev);
4405 __skb_push(skb, mac_len);
4408 metalen = xdp->data - xdp->data_meta;
4410 skb_metadata_set(skb, metalen);
4413 bpf_warn_invalid_xdp_action(act);
4416 trace_xdp_exception(skb->dev, xdp_prog, act);
4427 /* When doing generic XDP we have to bypass the qdisc layer and the
4428 * network taps in order to match in-driver-XDP behavior.
4430 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4432 struct net_device *dev = skb->dev;
4433 struct netdev_queue *txq;
4434 bool free_skb = true;
4437 txq = netdev_core_pick_tx(dev, skb, NULL);
4438 cpu = smp_processor_id();
4439 HARD_TX_LOCK(dev, txq, cpu);
4440 if (!netif_xmit_stopped(txq)) {
4441 rc = netdev_start_xmit(skb, dev, txq, 0);
4442 if (dev_xmit_complete(rc))
4445 HARD_TX_UNLOCK(dev, txq);
4447 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4451 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4453 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4455 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4458 struct xdp_buff xdp;
4462 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4463 if (act != XDP_PASS) {
4466 err = xdp_do_generic_redirect(skb->dev, skb,
4472 generic_xdp_tx(skb, xdp_prog);
4483 EXPORT_SYMBOL_GPL(do_xdp_generic);
4485 static int netif_rx_internal(struct sk_buff *skb)
4489 net_timestamp_check(netdev_tstamp_prequeue, skb);
4491 trace_netif_rx(skb);
4494 if (static_branch_unlikely(&rps_needed)) {
4495 struct rps_dev_flow voidflow, *rflow = &voidflow;
4501 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4503 cpu = smp_processor_id();
4505 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4514 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4521 * netif_rx - post buffer to the network code
4522 * @skb: buffer to post
4524 * This function receives a packet from a device driver and queues it for
4525 * the upper (protocol) levels to process. It always succeeds. The buffer
4526 * may be dropped during processing for congestion control or by the
4530 * NET_RX_SUCCESS (no congestion)
4531 * NET_RX_DROP (packet was dropped)
4535 int netif_rx(struct sk_buff *skb)
4539 trace_netif_rx_entry(skb);
4541 ret = netif_rx_internal(skb);
4542 trace_netif_rx_exit(ret);
4546 EXPORT_SYMBOL(netif_rx);
4548 int netif_rx_ni(struct sk_buff *skb)
4552 trace_netif_rx_ni_entry(skb);
4555 err = netif_rx_internal(skb);
4556 if (local_softirq_pending())
4559 trace_netif_rx_ni_exit(err);
4563 EXPORT_SYMBOL(netif_rx_ni);
4565 static __latent_entropy void net_tx_action(struct softirq_action *h)
4567 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4569 if (sd->completion_queue) {
4570 struct sk_buff *clist;
4572 local_irq_disable();
4573 clist = sd->completion_queue;
4574 sd->completion_queue = NULL;
4578 struct sk_buff *skb = clist;
4580 clist = clist->next;
4582 WARN_ON(refcount_read(&skb->users));
4583 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4584 trace_consume_skb(skb);
4586 trace_kfree_skb(skb, net_tx_action);
4588 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4591 __kfree_skb_defer(skb);
4594 __kfree_skb_flush();
4597 if (sd->output_queue) {
4600 local_irq_disable();
4601 head = sd->output_queue;
4602 sd->output_queue = NULL;
4603 sd->output_queue_tailp = &sd->output_queue;
4607 struct Qdisc *q = head;
4608 spinlock_t *root_lock = NULL;
4610 head = head->next_sched;
4612 if (!(q->flags & TCQ_F_NOLOCK)) {
4613 root_lock = qdisc_lock(q);
4614 spin_lock(root_lock);
4616 /* We need to make sure head->next_sched is read
4617 * before clearing __QDISC_STATE_SCHED
4619 smp_mb__before_atomic();
4620 clear_bit(__QDISC_STATE_SCHED, &q->state);
4623 spin_unlock(root_lock);
4627 xfrm_dev_backlog(sd);
4630 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4631 /* This hook is defined here for ATM LANE */
4632 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4633 unsigned char *addr) __read_mostly;
4634 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4637 static inline struct sk_buff *
4638 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4639 struct net_device *orig_dev)
4641 #ifdef CONFIG_NET_CLS_ACT
4642 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4643 struct tcf_result cl_res;
4645 /* If there's at least one ingress present somewhere (so
4646 * we get here via enabled static key), remaining devices
4647 * that are not configured with an ingress qdisc will bail
4654 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4658 qdisc_skb_cb(skb)->pkt_len = skb->len;
4659 skb->tc_at_ingress = 1;
4660 mini_qdisc_bstats_cpu_update(miniq, skb);
4662 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4664 case TC_ACT_RECLASSIFY:
4665 skb->tc_index = TC_H_MIN(cl_res.classid);
4668 mini_qdisc_qstats_cpu_drop(miniq);
4676 case TC_ACT_REDIRECT:
4677 /* skb_mac_header check was done by cls/act_bpf, so
4678 * we can safely push the L2 header back before
4679 * redirecting to another netdev
4681 __skb_push(skb, skb->mac_len);
4682 skb_do_redirect(skb);
4684 case TC_ACT_CONSUMED:
4689 #endif /* CONFIG_NET_CLS_ACT */
4694 * netdev_is_rx_handler_busy - check if receive handler is registered
4695 * @dev: device to check
4697 * Check if a receive handler is already registered for a given device.
4698 * Return true if there one.
4700 * The caller must hold the rtnl_mutex.
4702 bool netdev_is_rx_handler_busy(struct net_device *dev)
4705 return dev && rtnl_dereference(dev->rx_handler);
4707 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4710 * netdev_rx_handler_register - register receive handler
4711 * @dev: device to register a handler for
4712 * @rx_handler: receive handler to register
4713 * @rx_handler_data: data pointer that is used by rx handler
4715 * Register a receive handler for a device. This handler will then be
4716 * called from __netif_receive_skb. A negative errno code is returned
4719 * The caller must hold the rtnl_mutex.
4721 * For a general description of rx_handler, see enum rx_handler_result.
4723 int netdev_rx_handler_register(struct net_device *dev,
4724 rx_handler_func_t *rx_handler,
4725 void *rx_handler_data)
4727 if (netdev_is_rx_handler_busy(dev))
4730 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4733 /* Note: rx_handler_data must be set before rx_handler */
4734 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4735 rcu_assign_pointer(dev->rx_handler, rx_handler);
4739 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4742 * netdev_rx_handler_unregister - unregister receive handler
4743 * @dev: device to unregister a handler from
4745 * Unregister a receive handler from a device.
4747 * The caller must hold the rtnl_mutex.
4749 void netdev_rx_handler_unregister(struct net_device *dev)
4753 RCU_INIT_POINTER(dev->rx_handler, NULL);
4754 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4755 * section has a guarantee to see a non NULL rx_handler_data
4759 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4761 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4764 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4765 * the special handling of PFMEMALLOC skbs.
4767 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4769 switch (skb->protocol) {
4770 case htons(ETH_P_ARP):
4771 case htons(ETH_P_IP):
4772 case htons(ETH_P_IPV6):
4773 case htons(ETH_P_8021Q):
4774 case htons(ETH_P_8021AD):
4781 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4782 int *ret, struct net_device *orig_dev)
4784 #ifdef CONFIG_NETFILTER_INGRESS
4785 if (nf_hook_ingress_active(skb)) {
4789 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4794 ingress_retval = nf_hook_ingress(skb);
4796 return ingress_retval;
4798 #endif /* CONFIG_NETFILTER_INGRESS */
4802 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4803 struct packet_type **ppt_prev)
4805 struct packet_type *ptype, *pt_prev;
4806 rx_handler_func_t *rx_handler;
4807 struct net_device *orig_dev;
4808 bool deliver_exact = false;
4809 int ret = NET_RX_DROP;
4812 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4814 trace_netif_receive_skb(skb);
4816 orig_dev = skb->dev;
4818 skb_reset_network_header(skb);
4819 if (!skb_transport_header_was_set(skb))
4820 skb_reset_transport_header(skb);
4821 skb_reset_mac_len(skb);
4826 skb->skb_iif = skb->dev->ifindex;
4828 __this_cpu_inc(softnet_data.processed);
4830 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4834 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4837 if (ret2 != XDP_PASS)
4839 skb_reset_mac_len(skb);
4842 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4843 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4844 skb = skb_vlan_untag(skb);
4849 if (skb_skip_tc_classify(skb))
4855 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4857 ret = deliver_skb(skb, pt_prev, orig_dev);
4861 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4863 ret = deliver_skb(skb, pt_prev, orig_dev);
4868 #ifdef CONFIG_NET_INGRESS
4869 if (static_branch_unlikely(&ingress_needed_key)) {
4870 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4874 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4880 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4883 if (skb_vlan_tag_present(skb)) {
4885 ret = deliver_skb(skb, pt_prev, orig_dev);
4888 if (vlan_do_receive(&skb))
4890 else if (unlikely(!skb))
4894 rx_handler = rcu_dereference(skb->dev->rx_handler);
4897 ret = deliver_skb(skb, pt_prev, orig_dev);
4900 switch (rx_handler(&skb)) {
4901 case RX_HANDLER_CONSUMED:
4902 ret = NET_RX_SUCCESS;
4904 case RX_HANDLER_ANOTHER:
4906 case RX_HANDLER_EXACT:
4907 deliver_exact = true;
4908 case RX_HANDLER_PASS:
4915 if (unlikely(skb_vlan_tag_present(skb))) {
4917 if (skb_vlan_tag_get_id(skb)) {
4918 /* Vlan id is non 0 and vlan_do_receive() above couldn't
4921 skb->pkt_type = PACKET_OTHERHOST;
4922 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4923 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4924 /* Outer header is 802.1P with vlan 0, inner header is
4925 * 802.1Q or 802.1AD and vlan_do_receive() above could
4926 * not find vlan dev for vlan id 0.
4928 __vlan_hwaccel_clear_tag(skb);
4929 skb = skb_vlan_untag(skb);
4932 if (vlan_do_receive(&skb))
4933 /* After stripping off 802.1P header with vlan 0
4934 * vlan dev is found for inner header.
4937 else if (unlikely(!skb))
4940 /* We have stripped outer 802.1P vlan 0 header.
4941 * But could not find vlan dev.
4942 * check again for vlan id to set OTHERHOST.
4946 /* Note: we might in the future use prio bits
4947 * and set skb->priority like in vlan_do_receive()
4948 * For the time being, just ignore Priority Code Point
4950 __vlan_hwaccel_clear_tag(skb);
4953 type = skb->protocol;
4955 /* deliver only exact match when indicated */
4956 if (likely(!deliver_exact)) {
4957 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4958 &ptype_base[ntohs(type) &
4962 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4963 &orig_dev->ptype_specific);
4965 if (unlikely(skb->dev != orig_dev)) {
4966 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4967 &skb->dev->ptype_specific);
4971 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4973 *ppt_prev = pt_prev;
4977 atomic_long_inc(&skb->dev->rx_dropped);
4979 atomic_long_inc(&skb->dev->rx_nohandler);
4981 /* Jamal, now you will not able to escape explaining
4982 * me how you were going to use this. :-)
4991 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4993 struct net_device *orig_dev = skb->dev;
4994 struct packet_type *pt_prev = NULL;
4997 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4999 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5000 skb->dev, pt_prev, orig_dev);
5005 * netif_receive_skb_core - special purpose version of netif_receive_skb
5006 * @skb: buffer to process
5008 * More direct receive version of netif_receive_skb(). It should
5009 * only be used by callers that have a need to skip RPS and Generic XDP.
5010 * Caller must also take care of handling if (page_is_)pfmemalloc.
5012 * This function may only be called from softirq context and interrupts
5013 * should be enabled.
5015 * Return values (usually ignored):
5016 * NET_RX_SUCCESS: no congestion
5017 * NET_RX_DROP: packet was dropped
5019 int netif_receive_skb_core(struct sk_buff *skb)
5024 ret = __netif_receive_skb_one_core(skb, false);
5029 EXPORT_SYMBOL(netif_receive_skb_core);
5031 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5032 struct packet_type *pt_prev,
5033 struct net_device *orig_dev)
5035 struct sk_buff *skb, *next;
5039 if (list_empty(head))
5041 if (pt_prev->list_func != NULL)
5042 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5043 ip_list_rcv, head, pt_prev, orig_dev);
5045 list_for_each_entry_safe(skb, next, head, list) {
5046 skb_list_del_init(skb);
5047 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5051 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5053 /* Fast-path assumptions:
5054 * - There is no RX handler.
5055 * - Only one packet_type matches.
5056 * If either of these fails, we will end up doing some per-packet
5057 * processing in-line, then handling the 'last ptype' for the whole
5058 * sublist. This can't cause out-of-order delivery to any single ptype,
5059 * because the 'last ptype' must be constant across the sublist, and all
5060 * other ptypes are handled per-packet.
5062 /* Current (common) ptype of sublist */
5063 struct packet_type *pt_curr = NULL;
5064 /* Current (common) orig_dev of sublist */
5065 struct net_device *od_curr = NULL;
5066 struct list_head sublist;
5067 struct sk_buff *skb, *next;
5069 INIT_LIST_HEAD(&sublist);
5070 list_for_each_entry_safe(skb, next, head, list) {
5071 struct net_device *orig_dev = skb->dev;
5072 struct packet_type *pt_prev = NULL;
5074 skb_list_del_init(skb);
5075 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5078 if (pt_curr != pt_prev || od_curr != orig_dev) {
5079 /* dispatch old sublist */
5080 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5081 /* start new sublist */
5082 INIT_LIST_HEAD(&sublist);
5086 list_add_tail(&skb->list, &sublist);
5089 /* dispatch final sublist */
5090 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5093 static int __netif_receive_skb(struct sk_buff *skb)
5097 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5098 unsigned int noreclaim_flag;
5101 * PFMEMALLOC skbs are special, they should
5102 * - be delivered to SOCK_MEMALLOC sockets only
5103 * - stay away from userspace
5104 * - have bounded memory usage
5106 * Use PF_MEMALLOC as this saves us from propagating the allocation
5107 * context down to all allocation sites.
5109 noreclaim_flag = memalloc_noreclaim_save();
5110 ret = __netif_receive_skb_one_core(skb, true);
5111 memalloc_noreclaim_restore(noreclaim_flag);
5113 ret = __netif_receive_skb_one_core(skb, false);
5118 static void __netif_receive_skb_list(struct list_head *head)
5120 unsigned long noreclaim_flag = 0;
5121 struct sk_buff *skb, *next;
5122 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5124 list_for_each_entry_safe(skb, next, head, list) {
5125 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5126 struct list_head sublist;
5128 /* Handle the previous sublist */
5129 list_cut_before(&sublist, head, &skb->list);
5130 if (!list_empty(&sublist))
5131 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5132 pfmemalloc = !pfmemalloc;
5133 /* See comments in __netif_receive_skb */
5135 noreclaim_flag = memalloc_noreclaim_save();
5137 memalloc_noreclaim_restore(noreclaim_flag);
5140 /* Handle the remaining sublist */
5141 if (!list_empty(head))
5142 __netif_receive_skb_list_core(head, pfmemalloc);
5143 /* Restore pflags */
5145 memalloc_noreclaim_restore(noreclaim_flag);
5148 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5150 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5151 struct bpf_prog *new = xdp->prog;
5154 switch (xdp->command) {
5155 case XDP_SETUP_PROG:
5156 rcu_assign_pointer(dev->xdp_prog, new);
5161 static_branch_dec(&generic_xdp_needed_key);
5162 } else if (new && !old) {
5163 static_branch_inc(&generic_xdp_needed_key);
5164 dev_disable_lro(dev);
5165 dev_disable_gro_hw(dev);
5169 case XDP_QUERY_PROG:
5170 xdp->prog_id = old ? old->aux->id : 0;
5181 static int netif_receive_skb_internal(struct sk_buff *skb)
5185 net_timestamp_check(netdev_tstamp_prequeue, skb);
5187 if (skb_defer_rx_timestamp(skb))
5188 return NET_RX_SUCCESS;
5192 if (static_branch_unlikely(&rps_needed)) {
5193 struct rps_dev_flow voidflow, *rflow = &voidflow;
5194 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5197 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5203 ret = __netif_receive_skb(skb);
5208 static void netif_receive_skb_list_internal(struct list_head *head)
5210 struct sk_buff *skb, *next;
5211 struct list_head sublist;
5213 INIT_LIST_HEAD(&sublist);
5214 list_for_each_entry_safe(skb, next, head, list) {
5215 net_timestamp_check(netdev_tstamp_prequeue, skb);
5216 skb_list_del_init(skb);
5217 if (!skb_defer_rx_timestamp(skb))
5218 list_add_tail(&skb->list, &sublist);
5220 list_splice_init(&sublist, head);
5224 if (static_branch_unlikely(&rps_needed)) {
5225 list_for_each_entry_safe(skb, next, head, list) {
5226 struct rps_dev_flow voidflow, *rflow = &voidflow;
5227 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5230 /* Will be handled, remove from list */
5231 skb_list_del_init(skb);
5232 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5237 __netif_receive_skb_list(head);
5242 * netif_receive_skb - process receive buffer from network
5243 * @skb: buffer to process
5245 * netif_receive_skb() is the main receive data processing function.
5246 * It always succeeds. The buffer may be dropped during processing
5247 * for congestion control or by the protocol layers.
5249 * This function may only be called from softirq context and interrupts
5250 * should be enabled.
5252 * Return values (usually ignored):
5253 * NET_RX_SUCCESS: no congestion
5254 * NET_RX_DROP: packet was dropped
5256 int netif_receive_skb(struct sk_buff *skb)
5260 trace_netif_receive_skb_entry(skb);
5262 ret = netif_receive_skb_internal(skb);
5263 trace_netif_receive_skb_exit(ret);
5267 EXPORT_SYMBOL(netif_receive_skb);
5270 * netif_receive_skb_list - process many receive buffers from network
5271 * @head: list of skbs to process.
5273 * Since return value of netif_receive_skb() is normally ignored, and
5274 * wouldn't be meaningful for a list, this function returns void.
5276 * This function may only be called from softirq context and interrupts
5277 * should be enabled.
5279 void netif_receive_skb_list(struct list_head *head)
5281 struct sk_buff *skb;
5283 if (list_empty(head))
5285 if (trace_netif_receive_skb_list_entry_enabled()) {
5286 list_for_each_entry(skb, head, list)
5287 trace_netif_receive_skb_list_entry(skb);
5289 netif_receive_skb_list_internal(head);
5290 trace_netif_receive_skb_list_exit(0);
5292 EXPORT_SYMBOL(netif_receive_skb_list);
5294 DEFINE_PER_CPU(struct work_struct, flush_works);
5296 /* Network device is going away, flush any packets still pending */
5297 static void flush_backlog(struct work_struct *work)
5299 struct sk_buff *skb, *tmp;
5300 struct softnet_data *sd;
5303 sd = this_cpu_ptr(&softnet_data);
5305 local_irq_disable();
5307 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5308 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5309 __skb_unlink(skb, &sd->input_pkt_queue);
5311 input_queue_head_incr(sd);
5317 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5318 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5319 __skb_unlink(skb, &sd->process_queue);
5321 input_queue_head_incr(sd);
5327 static void flush_all_backlogs(void)
5333 for_each_online_cpu(cpu)
5334 queue_work_on(cpu, system_highpri_wq,
5335 per_cpu_ptr(&flush_works, cpu));
5337 for_each_online_cpu(cpu)
5338 flush_work(per_cpu_ptr(&flush_works, cpu));
5343 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5344 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5345 static int napi_gro_complete(struct sk_buff *skb)
5347 struct packet_offload *ptype;
5348 __be16 type = skb->protocol;
5349 struct list_head *head = &offload_base;
5352 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5354 if (NAPI_GRO_CB(skb)->count == 1) {
5355 skb_shinfo(skb)->gso_size = 0;
5360 list_for_each_entry_rcu(ptype, head, list) {
5361 if (ptype->type != type || !ptype->callbacks.gro_complete)
5364 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5365 ipv6_gro_complete, inet_gro_complete,
5372 WARN_ON(&ptype->list == head);
5374 return NET_RX_SUCCESS;
5378 return netif_receive_skb_internal(skb);
5381 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5384 struct list_head *head = &napi->gro_hash[index].list;
5385 struct sk_buff *skb, *p;
5387 list_for_each_entry_safe_reverse(skb, p, head, list) {
5388 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5390 skb_list_del_init(skb);
5391 napi_gro_complete(skb);
5392 napi->gro_hash[index].count--;
5395 if (!napi->gro_hash[index].count)
5396 __clear_bit(index, &napi->gro_bitmask);
5399 /* napi->gro_hash[].list contains packets ordered by age.
5400 * youngest packets at the head of it.
5401 * Complete skbs in reverse order to reduce latencies.
5403 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5405 unsigned long bitmask = napi->gro_bitmask;
5406 unsigned int i, base = ~0U;
5408 while ((i = ffs(bitmask)) != 0) {
5411 __napi_gro_flush_chain(napi, base, flush_old);
5414 EXPORT_SYMBOL(napi_gro_flush);
5416 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5417 struct sk_buff *skb)
5419 unsigned int maclen = skb->dev->hard_header_len;
5420 u32 hash = skb_get_hash_raw(skb);
5421 struct list_head *head;
5424 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5425 list_for_each_entry(p, head, list) {
5426 unsigned long diffs;
5428 NAPI_GRO_CB(p)->flush = 0;
5430 if (hash != skb_get_hash_raw(p)) {
5431 NAPI_GRO_CB(p)->same_flow = 0;
5435 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5436 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5437 if (skb_vlan_tag_present(p))
5438 diffs |= p->vlan_tci ^ skb->vlan_tci;
5439 diffs |= skb_metadata_dst_cmp(p, skb);
5440 diffs |= skb_metadata_differs(p, skb);
5441 if (maclen == ETH_HLEN)
5442 diffs |= compare_ether_header(skb_mac_header(p),
5443 skb_mac_header(skb));
5445 diffs = memcmp(skb_mac_header(p),
5446 skb_mac_header(skb),
5448 NAPI_GRO_CB(p)->same_flow = !diffs;
5454 static void skb_gro_reset_offset(struct sk_buff *skb)
5456 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5457 const skb_frag_t *frag0 = &pinfo->frags[0];
5459 NAPI_GRO_CB(skb)->data_offset = 0;
5460 NAPI_GRO_CB(skb)->frag0 = NULL;
5461 NAPI_GRO_CB(skb)->frag0_len = 0;
5463 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5465 !PageHighMem(skb_frag_page(frag0))) {
5466 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5467 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5468 skb_frag_size(frag0),
5469 skb->end - skb->tail);
5473 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5475 struct skb_shared_info *pinfo = skb_shinfo(skb);
5477 BUG_ON(skb->end - skb->tail < grow);
5479 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5481 skb->data_len -= grow;
5484 pinfo->frags[0].page_offset += grow;
5485 skb_frag_size_sub(&pinfo->frags[0], grow);
5487 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5488 skb_frag_unref(skb, 0);
5489 memmove(pinfo->frags, pinfo->frags + 1,
5490 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5494 static void gro_flush_oldest(struct list_head *head)
5496 struct sk_buff *oldest;
5498 oldest = list_last_entry(head, struct sk_buff, list);
5500 /* We are called with head length >= MAX_GRO_SKBS, so this is
5503 if (WARN_ON_ONCE(!oldest))
5506 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5509 skb_list_del_init(oldest);
5510 napi_gro_complete(oldest);
5513 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5515 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5517 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5519 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5520 struct list_head *head = &offload_base;
5521 struct packet_offload *ptype;
5522 __be16 type = skb->protocol;
5523 struct list_head *gro_head;
5524 struct sk_buff *pp = NULL;
5525 enum gro_result ret;
5529 if (netif_elide_gro(skb->dev))
5532 gro_head = gro_list_prepare(napi, skb);
5535 list_for_each_entry_rcu(ptype, head, list) {
5536 if (ptype->type != type || !ptype->callbacks.gro_receive)
5539 skb_set_network_header(skb, skb_gro_offset(skb));
5540 skb_reset_mac_len(skb);
5541 NAPI_GRO_CB(skb)->same_flow = 0;
5542 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5543 NAPI_GRO_CB(skb)->free = 0;
5544 NAPI_GRO_CB(skb)->encap_mark = 0;
5545 NAPI_GRO_CB(skb)->recursion_counter = 0;
5546 NAPI_GRO_CB(skb)->is_fou = 0;
5547 NAPI_GRO_CB(skb)->is_atomic = 1;
5548 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5550 /* Setup for GRO checksum validation */
5551 switch (skb->ip_summed) {
5552 case CHECKSUM_COMPLETE:
5553 NAPI_GRO_CB(skb)->csum = skb->csum;
5554 NAPI_GRO_CB(skb)->csum_valid = 1;
5555 NAPI_GRO_CB(skb)->csum_cnt = 0;
5557 case CHECKSUM_UNNECESSARY:
5558 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5559 NAPI_GRO_CB(skb)->csum_valid = 0;
5562 NAPI_GRO_CB(skb)->csum_cnt = 0;
5563 NAPI_GRO_CB(skb)->csum_valid = 0;
5566 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5567 ipv6_gro_receive, inet_gro_receive,
5573 if (&ptype->list == head)
5576 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5581 same_flow = NAPI_GRO_CB(skb)->same_flow;
5582 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5585 skb_list_del_init(pp);
5586 napi_gro_complete(pp);
5587 napi->gro_hash[hash].count--;
5593 if (NAPI_GRO_CB(skb)->flush)
5596 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5597 gro_flush_oldest(gro_head);
5599 napi->gro_hash[hash].count++;
5601 NAPI_GRO_CB(skb)->count = 1;
5602 NAPI_GRO_CB(skb)->age = jiffies;
5603 NAPI_GRO_CB(skb)->last = skb;
5604 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5605 list_add(&skb->list, gro_head);
5609 grow = skb_gro_offset(skb) - skb_headlen(skb);
5611 gro_pull_from_frag0(skb, grow);
5613 if (napi->gro_hash[hash].count) {
5614 if (!test_bit(hash, &napi->gro_bitmask))
5615 __set_bit(hash, &napi->gro_bitmask);
5616 } else if (test_bit(hash, &napi->gro_bitmask)) {
5617 __clear_bit(hash, &napi->gro_bitmask);
5627 struct packet_offload *gro_find_receive_by_type(__be16 type)
5629 struct list_head *offload_head = &offload_base;
5630 struct packet_offload *ptype;
5632 list_for_each_entry_rcu(ptype, offload_head, list) {
5633 if (ptype->type != type || !ptype->callbacks.gro_receive)
5639 EXPORT_SYMBOL(gro_find_receive_by_type);
5641 struct packet_offload *gro_find_complete_by_type(__be16 type)
5643 struct list_head *offload_head = &offload_base;
5644 struct packet_offload *ptype;
5646 list_for_each_entry_rcu(ptype, offload_head, list) {
5647 if (ptype->type != type || !ptype->callbacks.gro_complete)
5653 EXPORT_SYMBOL(gro_find_complete_by_type);
5655 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5659 kmem_cache_free(skbuff_head_cache, skb);
5662 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5666 if (netif_receive_skb_internal(skb))
5674 case GRO_MERGED_FREE:
5675 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5676 napi_skb_free_stolen_head(skb);
5690 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5694 skb_mark_napi_id(skb, napi);
5695 trace_napi_gro_receive_entry(skb);
5697 skb_gro_reset_offset(skb);
5699 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb);
5700 trace_napi_gro_receive_exit(ret);
5704 EXPORT_SYMBOL(napi_gro_receive);
5706 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5708 if (unlikely(skb->pfmemalloc)) {
5712 __skb_pull(skb, skb_headlen(skb));
5713 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5714 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5715 __vlan_hwaccel_clear_tag(skb);
5716 skb->dev = napi->dev;
5719 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5720 skb->pkt_type = PACKET_HOST;
5722 skb->encapsulation = 0;
5723 skb_shinfo(skb)->gso_type = 0;
5724 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5730 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5732 struct sk_buff *skb = napi->skb;
5735 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5738 skb_mark_napi_id(skb, napi);
5743 EXPORT_SYMBOL(napi_get_frags);
5745 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5746 struct sk_buff *skb,
5752 __skb_push(skb, ETH_HLEN);
5753 skb->protocol = eth_type_trans(skb, skb->dev);
5754 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5759 napi_reuse_skb(napi, skb);
5762 case GRO_MERGED_FREE:
5763 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5764 napi_skb_free_stolen_head(skb);
5766 napi_reuse_skb(napi, skb);
5777 /* Upper GRO stack assumes network header starts at gro_offset=0
5778 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5779 * We copy ethernet header into skb->data to have a common layout.
5781 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5783 struct sk_buff *skb = napi->skb;
5784 const struct ethhdr *eth;
5785 unsigned int hlen = sizeof(*eth);
5789 skb_reset_mac_header(skb);
5790 skb_gro_reset_offset(skb);
5792 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5793 eth = skb_gro_header_slow(skb, hlen, 0);
5794 if (unlikely(!eth)) {
5795 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5796 __func__, napi->dev->name);
5797 napi_reuse_skb(napi, skb);
5801 eth = (const struct ethhdr *)skb->data;
5802 gro_pull_from_frag0(skb, hlen);
5803 NAPI_GRO_CB(skb)->frag0 += hlen;
5804 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5806 __skb_pull(skb, hlen);
5809 * This works because the only protocols we care about don't require
5811 * We'll fix it up properly in napi_frags_finish()
5813 skb->protocol = eth->h_proto;
5818 gro_result_t napi_gro_frags(struct napi_struct *napi)
5821 struct sk_buff *skb = napi_frags_skb(napi);
5826 trace_napi_gro_frags_entry(skb);
5828 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5829 trace_napi_gro_frags_exit(ret);
5833 EXPORT_SYMBOL(napi_gro_frags);
5835 /* Compute the checksum from gro_offset and return the folded value
5836 * after adding in any pseudo checksum.
5838 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5843 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5845 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5846 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5847 /* See comments in __skb_checksum_complete(). */
5849 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5850 !skb->csum_complete_sw)
5851 netdev_rx_csum_fault(skb->dev, skb);
5854 NAPI_GRO_CB(skb)->csum = wsum;
5855 NAPI_GRO_CB(skb)->csum_valid = 1;
5859 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5861 static void net_rps_send_ipi(struct softnet_data *remsd)
5865 struct softnet_data *next = remsd->rps_ipi_next;
5867 if (cpu_online(remsd->cpu))
5868 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5875 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5876 * Note: called with local irq disabled, but exits with local irq enabled.
5878 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5881 struct softnet_data *remsd = sd->rps_ipi_list;
5884 sd->rps_ipi_list = NULL;
5888 /* Send pending IPI's to kick RPS processing on remote cpus. */
5889 net_rps_send_ipi(remsd);
5895 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5898 return sd->rps_ipi_list != NULL;
5904 static int process_backlog(struct napi_struct *napi, int quota)
5906 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5910 /* Check if we have pending ipi, its better to send them now,
5911 * not waiting net_rx_action() end.
5913 if (sd_has_rps_ipi_waiting(sd)) {
5914 local_irq_disable();
5915 net_rps_action_and_irq_enable(sd);
5918 napi->weight = dev_rx_weight;
5920 struct sk_buff *skb;
5922 while ((skb = __skb_dequeue(&sd->process_queue))) {
5924 __netif_receive_skb(skb);
5926 input_queue_head_incr(sd);
5927 if (++work >= quota)
5932 local_irq_disable();
5934 if (skb_queue_empty(&sd->input_pkt_queue)) {
5936 * Inline a custom version of __napi_complete().
5937 * only current cpu owns and manipulates this napi,
5938 * and NAPI_STATE_SCHED is the only possible flag set
5940 * We can use a plain write instead of clear_bit(),
5941 * and we dont need an smp_mb() memory barrier.
5946 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5947 &sd->process_queue);
5957 * __napi_schedule - schedule for receive
5958 * @n: entry to schedule
5960 * The entry's receive function will be scheduled to run.
5961 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5963 void __napi_schedule(struct napi_struct *n)
5965 unsigned long flags;
5967 local_irq_save(flags);
5968 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5969 local_irq_restore(flags);
5971 EXPORT_SYMBOL(__napi_schedule);
5974 * napi_schedule_prep - check if napi can be scheduled
5977 * Test if NAPI routine is already running, and if not mark
5978 * it as running. This is used as a condition variable
5979 * insure only one NAPI poll instance runs. We also make
5980 * sure there is no pending NAPI disable.
5982 bool napi_schedule_prep(struct napi_struct *n)
5984 unsigned long val, new;
5987 val = READ_ONCE(n->state);
5988 if (unlikely(val & NAPIF_STATE_DISABLE))
5990 new = val | NAPIF_STATE_SCHED;
5992 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5993 * This was suggested by Alexander Duyck, as compiler
5994 * emits better code than :
5995 * if (val & NAPIF_STATE_SCHED)
5996 * new |= NAPIF_STATE_MISSED;
5998 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6000 } while (cmpxchg(&n->state, val, new) != val);
6002 return !(val & NAPIF_STATE_SCHED);
6004 EXPORT_SYMBOL(napi_schedule_prep);
6007 * __napi_schedule_irqoff - schedule for receive
6008 * @n: entry to schedule
6010 * Variant of __napi_schedule() assuming hard irqs are masked
6012 void __napi_schedule_irqoff(struct napi_struct *n)
6014 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6016 EXPORT_SYMBOL(__napi_schedule_irqoff);
6018 bool napi_complete_done(struct napi_struct *n, int work_done)
6020 unsigned long flags, val, new;
6023 * 1) Don't let napi dequeue from the cpu poll list
6024 * just in case its running on a different cpu.
6025 * 2) If we are busy polling, do nothing here, we have
6026 * the guarantee we will be called later.
6028 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6029 NAPIF_STATE_IN_BUSY_POLL)))
6032 if (n->gro_bitmask) {
6033 unsigned long timeout = 0;
6036 timeout = n->dev->gro_flush_timeout;
6038 /* When the NAPI instance uses a timeout and keeps postponing
6039 * it, we need to bound somehow the time packets are kept in
6042 napi_gro_flush(n, !!timeout);
6044 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6045 HRTIMER_MODE_REL_PINNED);
6047 if (unlikely(!list_empty(&n->poll_list))) {
6048 /* If n->poll_list is not empty, we need to mask irqs */
6049 local_irq_save(flags);
6050 list_del_init(&n->poll_list);
6051 local_irq_restore(flags);
6055 val = READ_ONCE(n->state);
6057 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6059 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6061 /* If STATE_MISSED was set, leave STATE_SCHED set,
6062 * because we will call napi->poll() one more time.
6063 * This C code was suggested by Alexander Duyck to help gcc.
6065 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6067 } while (cmpxchg(&n->state, val, new) != val);
6069 if (unlikely(val & NAPIF_STATE_MISSED)) {
6076 EXPORT_SYMBOL(napi_complete_done);
6078 /* must be called under rcu_read_lock(), as we dont take a reference */
6079 static struct napi_struct *napi_by_id(unsigned int napi_id)
6081 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6082 struct napi_struct *napi;
6084 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6085 if (napi->napi_id == napi_id)
6091 #if defined(CONFIG_NET_RX_BUSY_POLL)
6093 #define BUSY_POLL_BUDGET 8
6095 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6099 /* Busy polling means there is a high chance device driver hard irq
6100 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6101 * set in napi_schedule_prep().
6102 * Since we are about to call napi->poll() once more, we can safely
6103 * clear NAPI_STATE_MISSED.
6105 * Note: x86 could use a single "lock and ..." instruction
6106 * to perform these two clear_bit()
6108 clear_bit(NAPI_STATE_MISSED, &napi->state);
6109 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6113 /* All we really want here is to re-enable device interrupts.
6114 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6116 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6117 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6118 netpoll_poll_unlock(have_poll_lock);
6119 if (rc == BUSY_POLL_BUDGET)
6120 __napi_schedule(napi);
6124 void napi_busy_loop(unsigned int napi_id,
6125 bool (*loop_end)(void *, unsigned long),
6128 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6129 int (*napi_poll)(struct napi_struct *napi, int budget);
6130 void *have_poll_lock = NULL;
6131 struct napi_struct *napi;
6138 napi = napi_by_id(napi_id);
6148 unsigned long val = READ_ONCE(napi->state);
6150 /* If multiple threads are competing for this napi,
6151 * we avoid dirtying napi->state as much as we can.
6153 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6154 NAPIF_STATE_IN_BUSY_POLL))
6156 if (cmpxchg(&napi->state, val,
6157 val | NAPIF_STATE_IN_BUSY_POLL |
6158 NAPIF_STATE_SCHED) != val)
6160 have_poll_lock = netpoll_poll_lock(napi);
6161 napi_poll = napi->poll;
6163 work = napi_poll(napi, BUSY_POLL_BUDGET);
6164 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6167 __NET_ADD_STATS(dev_net(napi->dev),
6168 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6171 if (!loop_end || loop_end(loop_end_arg, start_time))
6174 if (unlikely(need_resched())) {
6176 busy_poll_stop(napi, have_poll_lock);
6180 if (loop_end(loop_end_arg, start_time))
6187 busy_poll_stop(napi, have_poll_lock);
6192 EXPORT_SYMBOL(napi_busy_loop);
6194 #endif /* CONFIG_NET_RX_BUSY_POLL */
6196 static void napi_hash_add(struct napi_struct *napi)
6198 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6199 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6202 spin_lock(&napi_hash_lock);
6204 /* 0..NR_CPUS range is reserved for sender_cpu use */
6206 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6207 napi_gen_id = MIN_NAPI_ID;
6208 } while (napi_by_id(napi_gen_id));
6209 napi->napi_id = napi_gen_id;
6211 hlist_add_head_rcu(&napi->napi_hash_node,
6212 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6214 spin_unlock(&napi_hash_lock);
6217 /* Warning : caller is responsible to make sure rcu grace period
6218 * is respected before freeing memory containing @napi
6220 bool napi_hash_del(struct napi_struct *napi)
6222 bool rcu_sync_needed = false;
6224 spin_lock(&napi_hash_lock);
6226 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6227 rcu_sync_needed = true;
6228 hlist_del_rcu(&napi->napi_hash_node);
6230 spin_unlock(&napi_hash_lock);
6231 return rcu_sync_needed;
6233 EXPORT_SYMBOL_GPL(napi_hash_del);
6235 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6237 struct napi_struct *napi;
6239 napi = container_of(timer, struct napi_struct, timer);
6241 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6242 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6244 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6245 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6246 __napi_schedule_irqoff(napi);
6248 return HRTIMER_NORESTART;
6251 static void init_gro_hash(struct napi_struct *napi)
6255 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6256 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6257 napi->gro_hash[i].count = 0;
6259 napi->gro_bitmask = 0;
6262 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6263 int (*poll)(struct napi_struct *, int), int weight)
6265 INIT_LIST_HEAD(&napi->poll_list);
6266 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6267 napi->timer.function = napi_watchdog;
6268 init_gro_hash(napi);
6271 if (weight > NAPI_POLL_WEIGHT)
6272 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6274 napi->weight = weight;
6275 list_add(&napi->dev_list, &dev->napi_list);
6277 #ifdef CONFIG_NETPOLL
6278 napi->poll_owner = -1;
6280 set_bit(NAPI_STATE_SCHED, &napi->state);
6281 napi_hash_add(napi);
6283 EXPORT_SYMBOL(netif_napi_add);
6285 void napi_disable(struct napi_struct *n)
6288 set_bit(NAPI_STATE_DISABLE, &n->state);
6290 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6292 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6295 hrtimer_cancel(&n->timer);
6297 clear_bit(NAPI_STATE_DISABLE, &n->state);
6299 EXPORT_SYMBOL(napi_disable);
6301 static void flush_gro_hash(struct napi_struct *napi)
6305 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6306 struct sk_buff *skb, *n;
6308 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6310 napi->gro_hash[i].count = 0;
6314 /* Must be called in process context */
6315 void netif_napi_del(struct napi_struct *napi)
6318 if (napi_hash_del(napi))
6320 list_del_init(&napi->dev_list);
6321 napi_free_frags(napi);
6323 flush_gro_hash(napi);
6324 napi->gro_bitmask = 0;
6326 EXPORT_SYMBOL(netif_napi_del);
6328 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6333 list_del_init(&n->poll_list);
6335 have = netpoll_poll_lock(n);
6339 /* This NAPI_STATE_SCHED test is for avoiding a race
6340 * with netpoll's poll_napi(). Only the entity which
6341 * obtains the lock and sees NAPI_STATE_SCHED set will
6342 * actually make the ->poll() call. Therefore we avoid
6343 * accidentally calling ->poll() when NAPI is not scheduled.
6346 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6347 work = n->poll(n, weight);
6348 trace_napi_poll(n, work, weight);
6351 WARN_ON_ONCE(work > weight);
6353 if (likely(work < weight))
6356 /* Drivers must not modify the NAPI state if they
6357 * consume the entire weight. In such cases this code
6358 * still "owns" the NAPI instance and therefore can
6359 * move the instance around on the list at-will.
6361 if (unlikely(napi_disable_pending(n))) {
6366 if (n->gro_bitmask) {
6367 /* flush too old packets
6368 * If HZ < 1000, flush all packets.
6370 napi_gro_flush(n, HZ >= 1000);
6373 /* Some drivers may have called napi_schedule
6374 * prior to exhausting their budget.
6376 if (unlikely(!list_empty(&n->poll_list))) {
6377 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6378 n->dev ? n->dev->name : "backlog");
6382 list_add_tail(&n->poll_list, repoll);
6385 netpoll_poll_unlock(have);
6390 static __latent_entropy void net_rx_action(struct softirq_action *h)
6392 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6393 unsigned long time_limit = jiffies +
6394 usecs_to_jiffies(netdev_budget_usecs);
6395 int budget = netdev_budget;
6399 local_irq_disable();
6400 list_splice_init(&sd->poll_list, &list);
6404 struct napi_struct *n;
6406 if (list_empty(&list)) {
6407 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6412 n = list_first_entry(&list, struct napi_struct, poll_list);
6413 budget -= napi_poll(n, &repoll);
6415 /* If softirq window is exhausted then punt.
6416 * Allow this to run for 2 jiffies since which will allow
6417 * an average latency of 1.5/HZ.
6419 if (unlikely(budget <= 0 ||
6420 time_after_eq(jiffies, time_limit))) {
6426 local_irq_disable();
6428 list_splice_tail_init(&sd->poll_list, &list);
6429 list_splice_tail(&repoll, &list);
6430 list_splice(&list, &sd->poll_list);
6431 if (!list_empty(&sd->poll_list))
6432 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6434 net_rps_action_and_irq_enable(sd);
6436 __kfree_skb_flush();
6439 struct netdev_adjacent {
6440 struct net_device *dev;
6442 /* upper master flag, there can only be one master device per list */
6445 /* counter for the number of times this device was added to us */
6448 /* private field for the users */
6451 struct list_head list;
6452 struct rcu_head rcu;
6455 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6456 struct list_head *adj_list)
6458 struct netdev_adjacent *adj;
6460 list_for_each_entry(adj, adj_list, list) {
6461 if (adj->dev == adj_dev)
6467 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6469 struct net_device *dev = data;
6471 return upper_dev == dev;
6475 * netdev_has_upper_dev - Check if device is linked to an upper device
6477 * @upper_dev: upper device to check
6479 * Find out if a device is linked to specified upper device and return true
6480 * in case it is. Note that this checks only immediate upper device,
6481 * not through a complete stack of devices. The caller must hold the RTNL lock.
6483 bool netdev_has_upper_dev(struct net_device *dev,
6484 struct net_device *upper_dev)
6488 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6491 EXPORT_SYMBOL(netdev_has_upper_dev);
6494 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6496 * @upper_dev: upper device to check
6498 * Find out if a device is linked to specified upper device and return true
6499 * in case it is. Note that this checks the entire upper device chain.
6500 * The caller must hold rcu lock.
6503 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6504 struct net_device *upper_dev)
6506 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6509 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6512 * netdev_has_any_upper_dev - Check if device is linked to some device
6515 * Find out if a device is linked to an upper device and return true in case
6516 * it is. The caller must hold the RTNL lock.
6518 bool netdev_has_any_upper_dev(struct net_device *dev)
6522 return !list_empty(&dev->adj_list.upper);
6524 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6527 * netdev_master_upper_dev_get - Get master upper device
6530 * Find a master upper device and return pointer to it or NULL in case
6531 * it's not there. The caller must hold the RTNL lock.
6533 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6535 struct netdev_adjacent *upper;
6539 if (list_empty(&dev->adj_list.upper))
6542 upper = list_first_entry(&dev->adj_list.upper,
6543 struct netdev_adjacent, list);
6544 if (likely(upper->master))
6548 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6551 * netdev_has_any_lower_dev - Check if device is linked to some device
6554 * Find out if a device is linked to a lower device and return true in case
6555 * it is. The caller must hold the RTNL lock.
6557 static bool netdev_has_any_lower_dev(struct net_device *dev)
6561 return !list_empty(&dev->adj_list.lower);
6564 void *netdev_adjacent_get_private(struct list_head *adj_list)
6566 struct netdev_adjacent *adj;
6568 adj = list_entry(adj_list, struct netdev_adjacent, list);
6570 return adj->private;
6572 EXPORT_SYMBOL(netdev_adjacent_get_private);
6575 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6577 * @iter: list_head ** of the current position
6579 * Gets the next device from the dev's upper list, starting from iter
6580 * position. The caller must hold RCU read lock.
6582 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6583 struct list_head **iter)
6585 struct netdev_adjacent *upper;
6587 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6589 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6591 if (&upper->list == &dev->adj_list.upper)
6594 *iter = &upper->list;
6598 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6600 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6601 struct list_head **iter)
6603 struct netdev_adjacent *upper;
6605 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6607 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6609 if (&upper->list == &dev->adj_list.upper)
6612 *iter = &upper->list;
6617 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6618 int (*fn)(struct net_device *dev,
6622 struct net_device *udev;
6623 struct list_head *iter;
6626 for (iter = &dev->adj_list.upper,
6627 udev = netdev_next_upper_dev_rcu(dev, &iter);
6629 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6630 /* first is the upper device itself */
6631 ret = fn(udev, data);
6635 /* then look at all of its upper devices */
6636 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6643 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6646 * netdev_lower_get_next_private - Get the next ->private from the
6647 * lower neighbour list
6649 * @iter: list_head ** of the current position
6651 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6652 * list, starting from iter position. The caller must hold either hold the
6653 * RTNL lock or its own locking that guarantees that the neighbour lower
6654 * list will remain unchanged.
6656 void *netdev_lower_get_next_private(struct net_device *dev,
6657 struct list_head **iter)
6659 struct netdev_adjacent *lower;
6661 lower = list_entry(*iter, struct netdev_adjacent, list);
6663 if (&lower->list == &dev->adj_list.lower)
6666 *iter = lower->list.next;
6668 return lower->private;
6670 EXPORT_SYMBOL(netdev_lower_get_next_private);
6673 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6674 * lower neighbour list, RCU
6677 * @iter: list_head ** of the current position
6679 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6680 * list, starting from iter position. The caller must hold RCU read lock.
6682 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6683 struct list_head **iter)
6685 struct netdev_adjacent *lower;
6687 WARN_ON_ONCE(!rcu_read_lock_held());
6689 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6691 if (&lower->list == &dev->adj_list.lower)
6694 *iter = &lower->list;
6696 return lower->private;
6698 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6701 * netdev_lower_get_next - Get the next device from the lower neighbour
6704 * @iter: list_head ** of the current position
6706 * Gets the next netdev_adjacent from the dev's lower neighbour
6707 * list, starting from iter position. The caller must hold RTNL lock or
6708 * its own locking that guarantees that the neighbour lower
6709 * list will remain unchanged.
6711 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6713 struct netdev_adjacent *lower;
6715 lower = list_entry(*iter, struct netdev_adjacent, list);
6717 if (&lower->list == &dev->adj_list.lower)
6720 *iter = lower->list.next;
6724 EXPORT_SYMBOL(netdev_lower_get_next);
6726 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6727 struct list_head **iter)
6729 struct netdev_adjacent *lower;
6731 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6733 if (&lower->list == &dev->adj_list.lower)
6736 *iter = &lower->list;
6741 int netdev_walk_all_lower_dev(struct net_device *dev,
6742 int (*fn)(struct net_device *dev,
6746 struct net_device *ldev;
6747 struct list_head *iter;
6750 for (iter = &dev->adj_list.lower,
6751 ldev = netdev_next_lower_dev(dev, &iter);
6753 ldev = netdev_next_lower_dev(dev, &iter)) {
6754 /* first is the lower device itself */
6755 ret = fn(ldev, data);
6759 /* then look at all of its lower devices */
6760 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6767 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6769 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6770 struct list_head **iter)
6772 struct netdev_adjacent *lower;
6774 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6775 if (&lower->list == &dev->adj_list.lower)
6778 *iter = &lower->list;
6783 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6784 int (*fn)(struct net_device *dev,
6788 struct net_device *ldev;
6789 struct list_head *iter;
6792 for (iter = &dev->adj_list.lower,
6793 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6795 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6796 /* first is the lower device itself */
6797 ret = fn(ldev, data);
6801 /* then look at all of its lower devices */
6802 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6809 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6812 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6813 * lower neighbour list, RCU
6817 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6818 * list. The caller must hold RCU read lock.
6820 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6822 struct netdev_adjacent *lower;
6824 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6825 struct netdev_adjacent, list);
6827 return lower->private;
6830 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6833 * netdev_master_upper_dev_get_rcu - Get master upper device
6836 * Find a master upper device and return pointer to it or NULL in case
6837 * it's not there. The caller must hold the RCU read lock.
6839 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6841 struct netdev_adjacent *upper;
6843 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6844 struct netdev_adjacent, list);
6845 if (upper && likely(upper->master))
6849 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6851 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6852 struct net_device *adj_dev,
6853 struct list_head *dev_list)
6855 char linkname[IFNAMSIZ+7];
6857 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6858 "upper_%s" : "lower_%s", adj_dev->name);
6859 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6862 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6864 struct list_head *dev_list)
6866 char linkname[IFNAMSIZ+7];
6868 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6869 "upper_%s" : "lower_%s", name);
6870 sysfs_remove_link(&(dev->dev.kobj), linkname);
6873 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6874 struct net_device *adj_dev,
6875 struct list_head *dev_list)
6877 return (dev_list == &dev->adj_list.upper ||
6878 dev_list == &dev->adj_list.lower) &&
6879 net_eq(dev_net(dev), dev_net(adj_dev));
6882 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6883 struct net_device *adj_dev,
6884 struct list_head *dev_list,
6885 void *private, bool master)
6887 struct netdev_adjacent *adj;
6890 adj = __netdev_find_adj(adj_dev, dev_list);
6894 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6895 dev->name, adj_dev->name, adj->ref_nr);
6900 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6905 adj->master = master;
6907 adj->private = private;
6910 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6911 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6913 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6914 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6919 /* Ensure that master link is always the first item in list. */
6921 ret = sysfs_create_link(&(dev->dev.kobj),
6922 &(adj_dev->dev.kobj), "master");
6924 goto remove_symlinks;
6926 list_add_rcu(&adj->list, dev_list);
6928 list_add_tail_rcu(&adj->list, dev_list);
6934 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6935 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6943 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6944 struct net_device *adj_dev,
6946 struct list_head *dev_list)
6948 struct netdev_adjacent *adj;
6950 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6951 dev->name, adj_dev->name, ref_nr);
6953 adj = __netdev_find_adj(adj_dev, dev_list);
6956 pr_err("Adjacency does not exist for device %s from %s\n",
6957 dev->name, adj_dev->name);
6962 if (adj->ref_nr > ref_nr) {
6963 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6964 dev->name, adj_dev->name, ref_nr,
6965 adj->ref_nr - ref_nr);
6966 adj->ref_nr -= ref_nr;
6971 sysfs_remove_link(&(dev->dev.kobj), "master");
6973 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6974 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6976 list_del_rcu(&adj->list);
6977 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6978 adj_dev->name, dev->name, adj_dev->name);
6980 kfree_rcu(adj, rcu);
6983 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6984 struct net_device *upper_dev,
6985 struct list_head *up_list,
6986 struct list_head *down_list,
6987 void *private, bool master)
6991 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6996 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6999 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7006 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7007 struct net_device *upper_dev,
7009 struct list_head *up_list,
7010 struct list_head *down_list)
7012 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7013 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7016 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7017 struct net_device *upper_dev,
7018 void *private, bool master)
7020 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7021 &dev->adj_list.upper,
7022 &upper_dev->adj_list.lower,
7026 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7027 struct net_device *upper_dev)
7029 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7030 &dev->adj_list.upper,
7031 &upper_dev->adj_list.lower);
7034 static int __netdev_upper_dev_link(struct net_device *dev,
7035 struct net_device *upper_dev, bool master,
7036 void *upper_priv, void *upper_info,
7037 struct netlink_ext_ack *extack)
7039 struct netdev_notifier_changeupper_info changeupper_info = {
7044 .upper_dev = upper_dev,
7047 .upper_info = upper_info,
7049 struct net_device *master_dev;
7054 if (dev == upper_dev)
7057 /* To prevent loops, check if dev is not upper device to upper_dev. */
7058 if (netdev_has_upper_dev(upper_dev, dev))
7062 if (netdev_has_upper_dev(dev, upper_dev))
7065 master_dev = netdev_master_upper_dev_get(dev);
7067 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7070 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7071 &changeupper_info.info);
7072 ret = notifier_to_errno(ret);
7076 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7081 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7082 &changeupper_info.info);
7083 ret = notifier_to_errno(ret);
7090 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7096 * netdev_upper_dev_link - Add a link to the upper device
7098 * @upper_dev: new upper device
7099 * @extack: netlink extended ack
7101 * Adds a link to device which is upper to this one. The caller must hold
7102 * the RTNL lock. On a failure a negative errno code is returned.
7103 * On success the reference counts are adjusted and the function
7106 int netdev_upper_dev_link(struct net_device *dev,
7107 struct net_device *upper_dev,
7108 struct netlink_ext_ack *extack)
7110 return __netdev_upper_dev_link(dev, upper_dev, false,
7111 NULL, NULL, extack);
7113 EXPORT_SYMBOL(netdev_upper_dev_link);
7116 * netdev_master_upper_dev_link - Add a master link to the upper device
7118 * @upper_dev: new upper device
7119 * @upper_priv: upper device private
7120 * @upper_info: upper info to be passed down via notifier
7121 * @extack: netlink extended ack
7123 * Adds a link to device which is upper to this one. In this case, only
7124 * one master upper device can be linked, although other non-master devices
7125 * might be linked as well. The caller must hold the RTNL lock.
7126 * On a failure a negative errno code is returned. On success the reference
7127 * counts are adjusted and the function returns zero.
7129 int netdev_master_upper_dev_link(struct net_device *dev,
7130 struct net_device *upper_dev,
7131 void *upper_priv, void *upper_info,
7132 struct netlink_ext_ack *extack)
7134 return __netdev_upper_dev_link(dev, upper_dev, true,
7135 upper_priv, upper_info, extack);
7137 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7140 * netdev_upper_dev_unlink - Removes a link to upper device
7142 * @upper_dev: new upper device
7144 * Removes a link to device which is upper to this one. The caller must hold
7147 void netdev_upper_dev_unlink(struct net_device *dev,
7148 struct net_device *upper_dev)
7150 struct netdev_notifier_changeupper_info changeupper_info = {
7154 .upper_dev = upper_dev,
7160 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7162 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7163 &changeupper_info.info);
7165 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7167 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7168 &changeupper_info.info);
7170 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7173 * netdev_bonding_info_change - Dispatch event about slave change
7175 * @bonding_info: info to dispatch
7177 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7178 * The caller must hold the RTNL lock.
7180 void netdev_bonding_info_change(struct net_device *dev,
7181 struct netdev_bonding_info *bonding_info)
7183 struct netdev_notifier_bonding_info info = {
7187 memcpy(&info.bonding_info, bonding_info,
7188 sizeof(struct netdev_bonding_info));
7189 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7192 EXPORT_SYMBOL(netdev_bonding_info_change);
7194 static void netdev_adjacent_add_links(struct net_device *dev)
7196 struct netdev_adjacent *iter;
7198 struct net *net = dev_net(dev);
7200 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7201 if (!net_eq(net, dev_net(iter->dev)))
7203 netdev_adjacent_sysfs_add(iter->dev, dev,
7204 &iter->dev->adj_list.lower);
7205 netdev_adjacent_sysfs_add(dev, iter->dev,
7206 &dev->adj_list.upper);
7209 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7210 if (!net_eq(net, dev_net(iter->dev)))
7212 netdev_adjacent_sysfs_add(iter->dev, dev,
7213 &iter->dev->adj_list.upper);
7214 netdev_adjacent_sysfs_add(dev, iter->dev,
7215 &dev->adj_list.lower);
7219 static void netdev_adjacent_del_links(struct net_device *dev)
7221 struct netdev_adjacent *iter;
7223 struct net *net = dev_net(dev);
7225 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7226 if (!net_eq(net, dev_net(iter->dev)))
7228 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7229 &iter->dev->adj_list.lower);
7230 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7231 &dev->adj_list.upper);
7234 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7235 if (!net_eq(net, dev_net(iter->dev)))
7237 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7238 &iter->dev->adj_list.upper);
7239 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7240 &dev->adj_list.lower);
7244 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7246 struct netdev_adjacent *iter;
7248 struct net *net = dev_net(dev);
7250 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7251 if (!net_eq(net, dev_net(iter->dev)))
7253 netdev_adjacent_sysfs_del(iter->dev, oldname,
7254 &iter->dev->adj_list.lower);
7255 netdev_adjacent_sysfs_add(iter->dev, dev,
7256 &iter->dev->adj_list.lower);
7259 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7260 if (!net_eq(net, dev_net(iter->dev)))
7262 netdev_adjacent_sysfs_del(iter->dev, oldname,
7263 &iter->dev->adj_list.upper);
7264 netdev_adjacent_sysfs_add(iter->dev, dev,
7265 &iter->dev->adj_list.upper);
7269 void *netdev_lower_dev_get_private(struct net_device *dev,
7270 struct net_device *lower_dev)
7272 struct netdev_adjacent *lower;
7276 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7280 return lower->private;
7282 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7285 int dev_get_nest_level(struct net_device *dev)
7287 struct net_device *lower = NULL;
7288 struct list_head *iter;
7294 netdev_for_each_lower_dev(dev, lower, iter) {
7295 nest = dev_get_nest_level(lower);
7296 if (max_nest < nest)
7300 return max_nest + 1;
7302 EXPORT_SYMBOL(dev_get_nest_level);
7305 * netdev_lower_change - Dispatch event about lower device state change
7306 * @lower_dev: device
7307 * @lower_state_info: state to dispatch
7309 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7310 * The caller must hold the RTNL lock.
7312 void netdev_lower_state_changed(struct net_device *lower_dev,
7313 void *lower_state_info)
7315 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7316 .info.dev = lower_dev,
7320 changelowerstate_info.lower_state_info = lower_state_info;
7321 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7322 &changelowerstate_info.info);
7324 EXPORT_SYMBOL(netdev_lower_state_changed);
7326 static void dev_change_rx_flags(struct net_device *dev, int flags)
7328 const struct net_device_ops *ops = dev->netdev_ops;
7330 if (ops->ndo_change_rx_flags)
7331 ops->ndo_change_rx_flags(dev, flags);
7334 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7336 unsigned int old_flags = dev->flags;
7342 dev->flags |= IFF_PROMISC;
7343 dev->promiscuity += inc;
7344 if (dev->promiscuity == 0) {
7347 * If inc causes overflow, untouch promisc and return error.
7350 dev->flags &= ~IFF_PROMISC;
7352 dev->promiscuity -= inc;
7353 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7358 if (dev->flags != old_flags) {
7359 pr_info("device %s %s promiscuous mode\n",
7361 dev->flags & IFF_PROMISC ? "entered" : "left");
7362 if (audit_enabled) {
7363 current_uid_gid(&uid, &gid);
7364 audit_log(audit_context(), GFP_ATOMIC,
7365 AUDIT_ANOM_PROMISCUOUS,
7366 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7367 dev->name, (dev->flags & IFF_PROMISC),
7368 (old_flags & IFF_PROMISC),
7369 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7370 from_kuid(&init_user_ns, uid),
7371 from_kgid(&init_user_ns, gid),
7372 audit_get_sessionid(current));
7375 dev_change_rx_flags(dev, IFF_PROMISC);
7378 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7383 * dev_set_promiscuity - update promiscuity count on a device
7387 * Add or remove promiscuity from a device. While the count in the device
7388 * remains above zero the interface remains promiscuous. Once it hits zero
7389 * the device reverts back to normal filtering operation. A negative inc
7390 * value is used to drop promiscuity on the device.
7391 * Return 0 if successful or a negative errno code on error.
7393 int dev_set_promiscuity(struct net_device *dev, int inc)
7395 unsigned int old_flags = dev->flags;
7398 err = __dev_set_promiscuity(dev, inc, true);
7401 if (dev->flags != old_flags)
7402 dev_set_rx_mode(dev);
7405 EXPORT_SYMBOL(dev_set_promiscuity);
7407 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7409 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7413 dev->flags |= IFF_ALLMULTI;
7414 dev->allmulti += inc;
7415 if (dev->allmulti == 0) {
7418 * If inc causes overflow, untouch allmulti and return error.
7421 dev->flags &= ~IFF_ALLMULTI;
7423 dev->allmulti -= inc;
7424 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7429 if (dev->flags ^ old_flags) {
7430 dev_change_rx_flags(dev, IFF_ALLMULTI);
7431 dev_set_rx_mode(dev);
7433 __dev_notify_flags(dev, old_flags,
7434 dev->gflags ^ old_gflags);
7440 * dev_set_allmulti - update allmulti count on a device
7444 * Add or remove reception of all multicast frames to a device. While the
7445 * count in the device remains above zero the interface remains listening
7446 * to all interfaces. Once it hits zero the device reverts back to normal
7447 * filtering operation. A negative @inc value is used to drop the counter
7448 * when releasing a resource needing all multicasts.
7449 * Return 0 if successful or a negative errno code on error.
7452 int dev_set_allmulti(struct net_device *dev, int inc)
7454 return __dev_set_allmulti(dev, inc, true);
7456 EXPORT_SYMBOL(dev_set_allmulti);
7459 * Upload unicast and multicast address lists to device and
7460 * configure RX filtering. When the device doesn't support unicast
7461 * filtering it is put in promiscuous mode while unicast addresses
7464 void __dev_set_rx_mode(struct net_device *dev)
7466 const struct net_device_ops *ops = dev->netdev_ops;
7468 /* dev_open will call this function so the list will stay sane. */
7469 if (!(dev->flags&IFF_UP))
7472 if (!netif_device_present(dev))
7475 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7476 /* Unicast addresses changes may only happen under the rtnl,
7477 * therefore calling __dev_set_promiscuity here is safe.
7479 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7480 __dev_set_promiscuity(dev, 1, false);
7481 dev->uc_promisc = true;
7482 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7483 __dev_set_promiscuity(dev, -1, false);
7484 dev->uc_promisc = false;
7488 if (ops->ndo_set_rx_mode)
7489 ops->ndo_set_rx_mode(dev);
7492 void dev_set_rx_mode(struct net_device *dev)
7494 netif_addr_lock_bh(dev);
7495 __dev_set_rx_mode(dev);
7496 netif_addr_unlock_bh(dev);
7500 * dev_get_flags - get flags reported to userspace
7503 * Get the combination of flag bits exported through APIs to userspace.
7505 unsigned int dev_get_flags(const struct net_device *dev)
7509 flags = (dev->flags & ~(IFF_PROMISC |
7514 (dev->gflags & (IFF_PROMISC |
7517 if (netif_running(dev)) {
7518 if (netif_oper_up(dev))
7519 flags |= IFF_RUNNING;
7520 if (netif_carrier_ok(dev))
7521 flags |= IFF_LOWER_UP;
7522 if (netif_dormant(dev))
7523 flags |= IFF_DORMANT;
7528 EXPORT_SYMBOL(dev_get_flags);
7530 int __dev_change_flags(struct net_device *dev, unsigned int flags,
7531 struct netlink_ext_ack *extack)
7533 unsigned int old_flags = dev->flags;
7539 * Set the flags on our device.
7542 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7543 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7545 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7549 * Load in the correct multicast list now the flags have changed.
7552 if ((old_flags ^ flags) & IFF_MULTICAST)
7553 dev_change_rx_flags(dev, IFF_MULTICAST);
7555 dev_set_rx_mode(dev);
7558 * Have we downed the interface. We handle IFF_UP ourselves
7559 * according to user attempts to set it, rather than blindly
7564 if ((old_flags ^ flags) & IFF_UP) {
7565 if (old_flags & IFF_UP)
7568 ret = __dev_open(dev, extack);
7571 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7572 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7573 unsigned int old_flags = dev->flags;
7575 dev->gflags ^= IFF_PROMISC;
7577 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7578 if (dev->flags != old_flags)
7579 dev_set_rx_mode(dev);
7582 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7583 * is important. Some (broken) drivers set IFF_PROMISC, when
7584 * IFF_ALLMULTI is requested not asking us and not reporting.
7586 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7587 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7589 dev->gflags ^= IFF_ALLMULTI;
7590 __dev_set_allmulti(dev, inc, false);
7596 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7597 unsigned int gchanges)
7599 unsigned int changes = dev->flags ^ old_flags;
7602 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7604 if (changes & IFF_UP) {
7605 if (dev->flags & IFF_UP)
7606 call_netdevice_notifiers(NETDEV_UP, dev);
7608 call_netdevice_notifiers(NETDEV_DOWN, dev);
7611 if (dev->flags & IFF_UP &&
7612 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7613 struct netdev_notifier_change_info change_info = {
7617 .flags_changed = changes,
7620 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7625 * dev_change_flags - change device settings
7627 * @flags: device state flags
7628 * @extack: netlink extended ack
7630 * Change settings on device based state flags. The flags are
7631 * in the userspace exported format.
7633 int dev_change_flags(struct net_device *dev, unsigned int flags,
7634 struct netlink_ext_ack *extack)
7637 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7639 ret = __dev_change_flags(dev, flags, extack);
7643 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7644 __dev_notify_flags(dev, old_flags, changes);
7647 EXPORT_SYMBOL(dev_change_flags);
7649 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7651 const struct net_device_ops *ops = dev->netdev_ops;
7653 if (ops->ndo_change_mtu)
7654 return ops->ndo_change_mtu(dev, new_mtu);
7659 EXPORT_SYMBOL(__dev_set_mtu);
7662 * dev_set_mtu_ext - Change maximum transfer unit
7664 * @new_mtu: new transfer unit
7665 * @extack: netlink extended ack
7667 * Change the maximum transfer size of the network device.
7669 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7670 struct netlink_ext_ack *extack)
7674 if (new_mtu == dev->mtu)
7677 /* MTU must be positive, and in range */
7678 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7679 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7683 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7684 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7688 if (!netif_device_present(dev))
7691 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7692 err = notifier_to_errno(err);
7696 orig_mtu = dev->mtu;
7697 err = __dev_set_mtu(dev, new_mtu);
7700 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7702 err = notifier_to_errno(err);
7704 /* setting mtu back and notifying everyone again,
7705 * so that they have a chance to revert changes.
7707 __dev_set_mtu(dev, orig_mtu);
7708 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7715 int dev_set_mtu(struct net_device *dev, int new_mtu)
7717 struct netlink_ext_ack extack;
7720 memset(&extack, 0, sizeof(extack));
7721 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7722 if (err && extack._msg)
7723 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7726 EXPORT_SYMBOL(dev_set_mtu);
7729 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7731 * @new_len: new tx queue length
7733 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7735 unsigned int orig_len = dev->tx_queue_len;
7738 if (new_len != (unsigned int)new_len)
7741 if (new_len != orig_len) {
7742 dev->tx_queue_len = new_len;
7743 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7744 res = notifier_to_errno(res);
7747 res = dev_qdisc_change_tx_queue_len(dev);
7755 netdev_err(dev, "refused to change device tx_queue_len\n");
7756 dev->tx_queue_len = orig_len;
7761 * dev_set_group - Change group this device belongs to
7763 * @new_group: group this device should belong to
7765 void dev_set_group(struct net_device *dev, int new_group)
7767 dev->group = new_group;
7769 EXPORT_SYMBOL(dev_set_group);
7772 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
7774 * @addr: new address
7775 * @extack: netlink extended ack
7777 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
7778 struct netlink_ext_ack *extack)
7780 struct netdev_notifier_pre_changeaddr_info info = {
7782 .info.extack = extack,
7787 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
7788 return notifier_to_errno(rc);
7790 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
7793 * dev_set_mac_address - Change Media Access Control Address
7796 * @extack: netlink extended ack
7798 * Change the hardware (MAC) address of the device
7800 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
7801 struct netlink_ext_ack *extack)
7803 const struct net_device_ops *ops = dev->netdev_ops;
7806 if (!ops->ndo_set_mac_address)
7808 if (sa->sa_family != dev->type)
7810 if (!netif_device_present(dev))
7812 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
7815 err = ops->ndo_set_mac_address(dev, sa);
7818 dev->addr_assign_type = NET_ADDR_SET;
7819 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7820 add_device_randomness(dev->dev_addr, dev->addr_len);
7823 EXPORT_SYMBOL(dev_set_mac_address);
7826 * dev_change_carrier - Change device carrier
7828 * @new_carrier: new value
7830 * Change device carrier
7832 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7834 const struct net_device_ops *ops = dev->netdev_ops;
7836 if (!ops->ndo_change_carrier)
7838 if (!netif_device_present(dev))
7840 return ops->ndo_change_carrier(dev, new_carrier);
7842 EXPORT_SYMBOL(dev_change_carrier);
7845 * dev_get_phys_port_id - Get device physical port ID
7849 * Get device physical port ID
7851 int dev_get_phys_port_id(struct net_device *dev,
7852 struct netdev_phys_item_id *ppid)
7854 const struct net_device_ops *ops = dev->netdev_ops;
7856 if (!ops->ndo_get_phys_port_id)
7858 return ops->ndo_get_phys_port_id(dev, ppid);
7860 EXPORT_SYMBOL(dev_get_phys_port_id);
7863 * dev_get_phys_port_name - Get device physical port name
7866 * @len: limit of bytes to copy to name
7868 * Get device physical port name
7870 int dev_get_phys_port_name(struct net_device *dev,
7871 char *name, size_t len)
7873 const struct net_device_ops *ops = dev->netdev_ops;
7876 if (ops->ndo_get_phys_port_name) {
7877 err = ops->ndo_get_phys_port_name(dev, name, len);
7878 if (err != -EOPNOTSUPP)
7881 return devlink_compat_phys_port_name_get(dev, name, len);
7883 EXPORT_SYMBOL(dev_get_phys_port_name);
7886 * dev_get_port_parent_id - Get the device's port parent identifier
7887 * @dev: network device
7888 * @ppid: pointer to a storage for the port's parent identifier
7889 * @recurse: allow/disallow recursion to lower devices
7891 * Get the devices's port parent identifier
7893 int dev_get_port_parent_id(struct net_device *dev,
7894 struct netdev_phys_item_id *ppid,
7897 const struct net_device_ops *ops = dev->netdev_ops;
7898 struct netdev_phys_item_id first = { };
7899 struct net_device *lower_dev;
7900 struct list_head *iter;
7903 if (ops->ndo_get_port_parent_id) {
7904 err = ops->ndo_get_port_parent_id(dev, ppid);
7905 if (err != -EOPNOTSUPP)
7909 err = devlink_compat_switch_id_get(dev, ppid);
7910 if (!err || err != -EOPNOTSUPP)
7916 netdev_for_each_lower_dev(dev, lower_dev, iter) {
7917 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
7922 else if (memcmp(&first, ppid, sizeof(*ppid)))
7928 EXPORT_SYMBOL(dev_get_port_parent_id);
7931 * netdev_port_same_parent_id - Indicate if two network devices have
7932 * the same port parent identifier
7933 * @a: first network device
7934 * @b: second network device
7936 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
7938 struct netdev_phys_item_id a_id = { };
7939 struct netdev_phys_item_id b_id = { };
7941 if (dev_get_port_parent_id(a, &a_id, true) ||
7942 dev_get_port_parent_id(b, &b_id, true))
7945 return netdev_phys_item_id_same(&a_id, &b_id);
7947 EXPORT_SYMBOL(netdev_port_same_parent_id);
7950 * dev_change_proto_down - update protocol port state information
7952 * @proto_down: new value
7954 * This info can be used by switch drivers to set the phys state of the
7957 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7959 const struct net_device_ops *ops = dev->netdev_ops;
7961 if (!ops->ndo_change_proto_down)
7963 if (!netif_device_present(dev))
7965 return ops->ndo_change_proto_down(dev, proto_down);
7967 EXPORT_SYMBOL(dev_change_proto_down);
7970 * dev_change_proto_down_generic - generic implementation for
7971 * ndo_change_proto_down that sets carrier according to
7975 * @proto_down: new value
7977 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
7980 netif_carrier_off(dev);
7982 netif_carrier_on(dev);
7983 dev->proto_down = proto_down;
7986 EXPORT_SYMBOL(dev_change_proto_down_generic);
7988 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
7989 enum bpf_netdev_command cmd)
7991 struct netdev_bpf xdp;
7996 memset(&xdp, 0, sizeof(xdp));
7999 /* Query must always succeed. */
8000 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8005 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8006 struct netlink_ext_ack *extack, u32 flags,
8007 struct bpf_prog *prog)
8009 struct netdev_bpf xdp;
8011 memset(&xdp, 0, sizeof(xdp));
8012 if (flags & XDP_FLAGS_HW_MODE)
8013 xdp.command = XDP_SETUP_PROG_HW;
8015 xdp.command = XDP_SETUP_PROG;
8016 xdp.extack = extack;
8020 return bpf_op(dev, &xdp);
8023 static void dev_xdp_uninstall(struct net_device *dev)
8025 struct netdev_bpf xdp;
8028 /* Remove generic XDP */
8029 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8031 /* Remove from the driver */
8032 ndo_bpf = dev->netdev_ops->ndo_bpf;
8036 memset(&xdp, 0, sizeof(xdp));
8037 xdp.command = XDP_QUERY_PROG;
8038 WARN_ON(ndo_bpf(dev, &xdp));
8040 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8043 /* Remove HW offload */
8044 memset(&xdp, 0, sizeof(xdp));
8045 xdp.command = XDP_QUERY_PROG_HW;
8046 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8047 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8052 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8054 * @extack: netlink extended ack
8055 * @fd: new program fd or negative value to clear
8056 * @flags: xdp-related flags
8058 * Set or clear a bpf program for a device
8060 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8063 const struct net_device_ops *ops = dev->netdev_ops;
8064 enum bpf_netdev_command query;
8065 struct bpf_prog *prog = NULL;
8066 bpf_op_t bpf_op, bpf_chk;
8072 offload = flags & XDP_FLAGS_HW_MODE;
8073 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8075 bpf_op = bpf_chk = ops->ndo_bpf;
8076 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8077 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8080 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8081 bpf_op = generic_xdp_install;
8082 if (bpf_op == bpf_chk)
8083 bpf_chk = generic_xdp_install;
8086 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8087 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8090 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
8091 __dev_xdp_query(dev, bpf_op, query)) {
8092 NL_SET_ERR_MSG(extack, "XDP program already attached");
8096 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8097 bpf_op == ops->ndo_bpf);
8099 return PTR_ERR(prog);
8101 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8102 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8108 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8109 if (err < 0 && prog)
8116 * dev_new_index - allocate an ifindex
8117 * @net: the applicable net namespace
8119 * Returns a suitable unique value for a new device interface
8120 * number. The caller must hold the rtnl semaphore or the
8121 * dev_base_lock to be sure it remains unique.
8123 static int dev_new_index(struct net *net)
8125 int ifindex = net->ifindex;
8130 if (!__dev_get_by_index(net, ifindex))
8131 return net->ifindex = ifindex;
8135 /* Delayed registration/unregisteration */
8136 static LIST_HEAD(net_todo_list);
8137 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8139 static void net_set_todo(struct net_device *dev)
8141 list_add_tail(&dev->todo_list, &net_todo_list);
8142 dev_net(dev)->dev_unreg_count++;
8145 static void rollback_registered_many(struct list_head *head)
8147 struct net_device *dev, *tmp;
8148 LIST_HEAD(close_head);
8150 BUG_ON(dev_boot_phase);
8153 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8154 /* Some devices call without registering
8155 * for initialization unwind. Remove those
8156 * devices and proceed with the remaining.
8158 if (dev->reg_state == NETREG_UNINITIALIZED) {
8159 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8163 list_del(&dev->unreg_list);
8166 dev->dismantle = true;
8167 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8170 /* If device is running, close it first. */
8171 list_for_each_entry(dev, head, unreg_list)
8172 list_add_tail(&dev->close_list, &close_head);
8173 dev_close_many(&close_head, true);
8175 list_for_each_entry(dev, head, unreg_list) {
8176 /* And unlink it from device chain. */
8177 unlist_netdevice(dev);
8179 dev->reg_state = NETREG_UNREGISTERING;
8181 flush_all_backlogs();
8185 list_for_each_entry(dev, head, unreg_list) {
8186 struct sk_buff *skb = NULL;
8188 /* Shutdown queueing discipline. */
8191 dev_xdp_uninstall(dev);
8193 /* Notify protocols, that we are about to destroy
8194 * this device. They should clean all the things.
8196 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8198 if (!dev->rtnl_link_ops ||
8199 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8200 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8201 GFP_KERNEL, NULL, 0);
8204 * Flush the unicast and multicast chains
8209 if (dev->netdev_ops->ndo_uninit)
8210 dev->netdev_ops->ndo_uninit(dev);
8213 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8215 /* Notifier chain MUST detach us all upper devices. */
8216 WARN_ON(netdev_has_any_upper_dev(dev));
8217 WARN_ON(netdev_has_any_lower_dev(dev));
8219 /* Remove entries from kobject tree */
8220 netdev_unregister_kobject(dev);
8222 /* Remove XPS queueing entries */
8223 netif_reset_xps_queues_gt(dev, 0);
8229 list_for_each_entry(dev, head, unreg_list)
8233 static void rollback_registered(struct net_device *dev)
8237 list_add(&dev->unreg_list, &single);
8238 rollback_registered_many(&single);
8242 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8243 struct net_device *upper, netdev_features_t features)
8245 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8246 netdev_features_t feature;
8249 for_each_netdev_feature(upper_disables, feature_bit) {
8250 feature = __NETIF_F_BIT(feature_bit);
8251 if (!(upper->wanted_features & feature)
8252 && (features & feature)) {
8253 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8254 &feature, upper->name);
8255 features &= ~feature;
8262 static void netdev_sync_lower_features(struct net_device *upper,
8263 struct net_device *lower, netdev_features_t features)
8265 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8266 netdev_features_t feature;
8269 for_each_netdev_feature(upper_disables, feature_bit) {
8270 feature = __NETIF_F_BIT(feature_bit);
8271 if (!(features & feature) && (lower->features & feature)) {
8272 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8273 &feature, lower->name);
8274 lower->wanted_features &= ~feature;
8275 netdev_update_features(lower);
8277 if (unlikely(lower->features & feature))
8278 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8279 &feature, lower->name);
8284 static netdev_features_t netdev_fix_features(struct net_device *dev,
8285 netdev_features_t features)
8287 /* Fix illegal checksum combinations */
8288 if ((features & NETIF_F_HW_CSUM) &&
8289 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8290 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8291 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8294 /* TSO requires that SG is present as well. */
8295 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8296 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8297 features &= ~NETIF_F_ALL_TSO;
8300 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8301 !(features & NETIF_F_IP_CSUM)) {
8302 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8303 features &= ~NETIF_F_TSO;
8304 features &= ~NETIF_F_TSO_ECN;
8307 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8308 !(features & NETIF_F_IPV6_CSUM)) {
8309 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8310 features &= ~NETIF_F_TSO6;
8313 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8314 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8315 features &= ~NETIF_F_TSO_MANGLEID;
8317 /* TSO ECN requires that TSO is present as well. */
8318 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8319 features &= ~NETIF_F_TSO_ECN;
8321 /* Software GSO depends on SG. */
8322 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8323 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8324 features &= ~NETIF_F_GSO;
8327 /* GSO partial features require GSO partial be set */
8328 if ((features & dev->gso_partial_features) &&
8329 !(features & NETIF_F_GSO_PARTIAL)) {
8331 "Dropping partially supported GSO features since no GSO partial.\n");
8332 features &= ~dev->gso_partial_features;
8335 if (!(features & NETIF_F_RXCSUM)) {
8336 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8337 * successfully merged by hardware must also have the
8338 * checksum verified by hardware. If the user does not
8339 * want to enable RXCSUM, logically, we should disable GRO_HW.
8341 if (features & NETIF_F_GRO_HW) {
8342 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8343 features &= ~NETIF_F_GRO_HW;
8347 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8348 if (features & NETIF_F_RXFCS) {
8349 if (features & NETIF_F_LRO) {
8350 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8351 features &= ~NETIF_F_LRO;
8354 if (features & NETIF_F_GRO_HW) {
8355 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8356 features &= ~NETIF_F_GRO_HW;
8363 int __netdev_update_features(struct net_device *dev)
8365 struct net_device *upper, *lower;
8366 netdev_features_t features;
8367 struct list_head *iter;
8372 features = netdev_get_wanted_features(dev);
8374 if (dev->netdev_ops->ndo_fix_features)
8375 features = dev->netdev_ops->ndo_fix_features(dev, features);
8377 /* driver might be less strict about feature dependencies */
8378 features = netdev_fix_features(dev, features);
8380 /* some features can't be enabled if they're off an an upper device */
8381 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8382 features = netdev_sync_upper_features(dev, upper, features);
8384 if (dev->features == features)
8387 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8388 &dev->features, &features);
8390 if (dev->netdev_ops->ndo_set_features)
8391 err = dev->netdev_ops->ndo_set_features(dev, features);
8395 if (unlikely(err < 0)) {
8397 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8398 err, &features, &dev->features);
8399 /* return non-0 since some features might have changed and
8400 * it's better to fire a spurious notification than miss it
8406 /* some features must be disabled on lower devices when disabled
8407 * on an upper device (think: bonding master or bridge)
8409 netdev_for_each_lower_dev(dev, lower, iter)
8410 netdev_sync_lower_features(dev, lower, features);
8413 netdev_features_t diff = features ^ dev->features;
8415 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8416 /* udp_tunnel_{get,drop}_rx_info both need
8417 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8418 * device, or they won't do anything.
8419 * Thus we need to update dev->features
8420 * *before* calling udp_tunnel_get_rx_info,
8421 * but *after* calling udp_tunnel_drop_rx_info.
8423 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8424 dev->features = features;
8425 udp_tunnel_get_rx_info(dev);
8427 udp_tunnel_drop_rx_info(dev);
8431 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8432 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8433 dev->features = features;
8434 err |= vlan_get_rx_ctag_filter_info(dev);
8436 vlan_drop_rx_ctag_filter_info(dev);
8440 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8441 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8442 dev->features = features;
8443 err |= vlan_get_rx_stag_filter_info(dev);
8445 vlan_drop_rx_stag_filter_info(dev);
8449 dev->features = features;
8452 return err < 0 ? 0 : 1;
8456 * netdev_update_features - recalculate device features
8457 * @dev: the device to check
8459 * Recalculate dev->features set and send notifications if it
8460 * has changed. Should be called after driver or hardware dependent
8461 * conditions might have changed that influence the features.
8463 void netdev_update_features(struct net_device *dev)
8465 if (__netdev_update_features(dev))
8466 netdev_features_change(dev);
8468 EXPORT_SYMBOL(netdev_update_features);
8471 * netdev_change_features - recalculate device features
8472 * @dev: the device to check
8474 * Recalculate dev->features set and send notifications even
8475 * if they have not changed. Should be called instead of
8476 * netdev_update_features() if also dev->vlan_features might
8477 * have changed to allow the changes to be propagated to stacked
8480 void netdev_change_features(struct net_device *dev)
8482 __netdev_update_features(dev);
8483 netdev_features_change(dev);
8485 EXPORT_SYMBOL(netdev_change_features);
8488 * netif_stacked_transfer_operstate - transfer operstate
8489 * @rootdev: the root or lower level device to transfer state from
8490 * @dev: the device to transfer operstate to
8492 * Transfer operational state from root to device. This is normally
8493 * called when a stacking relationship exists between the root
8494 * device and the device(a leaf device).
8496 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8497 struct net_device *dev)
8499 if (rootdev->operstate == IF_OPER_DORMANT)
8500 netif_dormant_on(dev);
8502 netif_dormant_off(dev);
8504 if (netif_carrier_ok(rootdev))
8505 netif_carrier_on(dev);
8507 netif_carrier_off(dev);
8509 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8511 static int netif_alloc_rx_queues(struct net_device *dev)
8513 unsigned int i, count = dev->num_rx_queues;
8514 struct netdev_rx_queue *rx;
8515 size_t sz = count * sizeof(*rx);
8520 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8526 for (i = 0; i < count; i++) {
8529 /* XDP RX-queue setup */
8530 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8537 /* Rollback successful reg's and free other resources */
8539 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8545 static void netif_free_rx_queues(struct net_device *dev)
8547 unsigned int i, count = dev->num_rx_queues;
8549 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8553 for (i = 0; i < count; i++)
8554 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8559 static void netdev_init_one_queue(struct net_device *dev,
8560 struct netdev_queue *queue, void *_unused)
8562 /* Initialize queue lock */
8563 spin_lock_init(&queue->_xmit_lock);
8564 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8565 queue->xmit_lock_owner = -1;
8566 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8569 dql_init(&queue->dql, HZ);
8573 static void netif_free_tx_queues(struct net_device *dev)
8578 static int netif_alloc_netdev_queues(struct net_device *dev)
8580 unsigned int count = dev->num_tx_queues;
8581 struct netdev_queue *tx;
8582 size_t sz = count * sizeof(*tx);
8584 if (count < 1 || count > 0xffff)
8587 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8593 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
8594 spin_lock_init(&dev->tx_global_lock);
8599 void netif_tx_stop_all_queues(struct net_device *dev)
8603 for (i = 0; i < dev->num_tx_queues; i++) {
8604 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
8606 netif_tx_stop_queue(txq);
8609 EXPORT_SYMBOL(netif_tx_stop_all_queues);
8612 * register_netdevice - register a network device
8613 * @dev: device to register
8615 * Take a completed network device structure and add it to the kernel
8616 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8617 * chain. 0 is returned on success. A negative errno code is returned
8618 * on a failure to set up the device, or if the name is a duplicate.
8620 * Callers must hold the rtnl semaphore. You may want
8621 * register_netdev() instead of this.
8624 * The locking appears insufficient to guarantee two parallel registers
8625 * will not get the same name.
8628 int register_netdevice(struct net_device *dev)
8631 struct net *net = dev_net(dev);
8633 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
8634 NETDEV_FEATURE_COUNT);
8635 BUG_ON(dev_boot_phase);
8640 /* When net_device's are persistent, this will be fatal. */
8641 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8644 spin_lock_init(&dev->addr_list_lock);
8645 netdev_set_addr_lockdep_class(dev);
8647 ret = dev_get_valid_name(net, dev, dev->name);
8651 /* Init, if this function is available */
8652 if (dev->netdev_ops->ndo_init) {
8653 ret = dev->netdev_ops->ndo_init(dev);
8661 if (((dev->hw_features | dev->features) &
8662 NETIF_F_HW_VLAN_CTAG_FILTER) &&
8663 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
8664 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
8665 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
8672 dev->ifindex = dev_new_index(net);
8673 else if (__dev_get_by_index(net, dev->ifindex))
8676 /* Transfer changeable features to wanted_features and enable
8677 * software offloads (GSO and GRO).
8679 dev->hw_features |= NETIF_F_SOFT_FEATURES;
8680 dev->features |= NETIF_F_SOFT_FEATURES;
8682 if (dev->netdev_ops->ndo_udp_tunnel_add) {
8683 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8684 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
8687 dev->wanted_features = dev->features & dev->hw_features;
8689 if (!(dev->flags & IFF_LOOPBACK))
8690 dev->hw_features |= NETIF_F_NOCACHE_COPY;
8692 /* If IPv4 TCP segmentation offload is supported we should also
8693 * allow the device to enable segmenting the frame with the option
8694 * of ignoring a static IP ID value. This doesn't enable the
8695 * feature itself but allows the user to enable it later.
8697 if (dev->hw_features & NETIF_F_TSO)
8698 dev->hw_features |= NETIF_F_TSO_MANGLEID;
8699 if (dev->vlan_features & NETIF_F_TSO)
8700 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
8701 if (dev->mpls_features & NETIF_F_TSO)
8702 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
8703 if (dev->hw_enc_features & NETIF_F_TSO)
8704 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
8706 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
8708 dev->vlan_features |= NETIF_F_HIGHDMA;
8710 /* Make NETIF_F_SG inheritable to tunnel devices.
8712 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
8714 /* Make NETIF_F_SG inheritable to MPLS.
8716 dev->mpls_features |= NETIF_F_SG;
8718 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
8719 ret = notifier_to_errno(ret);
8723 ret = netdev_register_kobject(dev);
8726 dev->reg_state = NETREG_REGISTERED;
8728 __netdev_update_features(dev);
8731 * Default initial state at registry is that the
8732 * device is present.
8735 set_bit(__LINK_STATE_PRESENT, &dev->state);
8737 linkwatch_init_dev(dev);
8739 dev_init_scheduler(dev);
8741 list_netdevice(dev);
8742 add_device_randomness(dev->dev_addr, dev->addr_len);
8744 /* If the device has permanent device address, driver should
8745 * set dev_addr and also addr_assign_type should be set to
8746 * NET_ADDR_PERM (default value).
8748 if (dev->addr_assign_type == NET_ADDR_PERM)
8749 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8751 /* Notify protocols, that a new device appeared. */
8752 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8753 ret = notifier_to_errno(ret);
8755 rollback_registered(dev);
8756 dev->reg_state = NETREG_UNREGISTERED;
8759 * Prevent userspace races by waiting until the network
8760 * device is fully setup before sending notifications.
8762 if (!dev->rtnl_link_ops ||
8763 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8764 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8770 if (dev->netdev_ops->ndo_uninit)
8771 dev->netdev_ops->ndo_uninit(dev);
8772 if (dev->priv_destructor)
8773 dev->priv_destructor(dev);
8776 EXPORT_SYMBOL(register_netdevice);
8779 * init_dummy_netdev - init a dummy network device for NAPI
8780 * @dev: device to init
8782 * This takes a network device structure and initialize the minimum
8783 * amount of fields so it can be used to schedule NAPI polls without
8784 * registering a full blown interface. This is to be used by drivers
8785 * that need to tie several hardware interfaces to a single NAPI
8786 * poll scheduler due to HW limitations.
8788 int init_dummy_netdev(struct net_device *dev)
8790 /* Clear everything. Note we don't initialize spinlocks
8791 * are they aren't supposed to be taken by any of the
8792 * NAPI code and this dummy netdev is supposed to be
8793 * only ever used for NAPI polls
8795 memset(dev, 0, sizeof(struct net_device));
8797 /* make sure we BUG if trying to hit standard
8798 * register/unregister code path
8800 dev->reg_state = NETREG_DUMMY;
8802 /* NAPI wants this */
8803 INIT_LIST_HEAD(&dev->napi_list);
8805 /* a dummy interface is started by default */
8806 set_bit(__LINK_STATE_PRESENT, &dev->state);
8807 set_bit(__LINK_STATE_START, &dev->state);
8809 /* napi_busy_loop stats accounting wants this */
8810 dev_net_set(dev, &init_net);
8812 /* Note : We dont allocate pcpu_refcnt for dummy devices,
8813 * because users of this 'device' dont need to change
8819 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8823 * register_netdev - register a network device
8824 * @dev: device to register
8826 * Take a completed network device structure and add it to the kernel
8827 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8828 * chain. 0 is returned on success. A negative errno code is returned
8829 * on a failure to set up the device, or if the name is a duplicate.
8831 * This is a wrapper around register_netdevice that takes the rtnl semaphore
8832 * and expands the device name if you passed a format string to
8835 int register_netdev(struct net_device *dev)
8839 if (rtnl_lock_killable())
8841 err = register_netdevice(dev);
8845 EXPORT_SYMBOL(register_netdev);
8847 int netdev_refcnt_read(const struct net_device *dev)
8851 for_each_possible_cpu(i)
8852 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8855 EXPORT_SYMBOL(netdev_refcnt_read);
8858 * netdev_wait_allrefs - wait until all references are gone.
8859 * @dev: target net_device
8861 * This is called when unregistering network devices.
8863 * Any protocol or device that holds a reference should register
8864 * for netdevice notification, and cleanup and put back the
8865 * reference if they receive an UNREGISTER event.
8866 * We can get stuck here if buggy protocols don't correctly
8869 static void netdev_wait_allrefs(struct net_device *dev)
8871 unsigned long rebroadcast_time, warning_time;
8874 linkwatch_forget_dev(dev);
8876 rebroadcast_time = warning_time = jiffies;
8877 refcnt = netdev_refcnt_read(dev);
8879 while (refcnt != 0) {
8880 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8883 /* Rebroadcast unregister notification */
8884 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8890 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8892 /* We must not have linkwatch events
8893 * pending on unregister. If this
8894 * happens, we simply run the queue
8895 * unscheduled, resulting in a noop
8898 linkwatch_run_queue();
8903 rebroadcast_time = jiffies;
8908 refcnt = netdev_refcnt_read(dev);
8910 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
8911 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8913 warning_time = jiffies;
8922 * register_netdevice(x1);
8923 * register_netdevice(x2);
8925 * unregister_netdevice(y1);
8926 * unregister_netdevice(y2);
8932 * We are invoked by rtnl_unlock().
8933 * This allows us to deal with problems:
8934 * 1) We can delete sysfs objects which invoke hotplug
8935 * without deadlocking with linkwatch via keventd.
8936 * 2) Since we run with the RTNL semaphore not held, we can sleep
8937 * safely in order to wait for the netdev refcnt to drop to zero.
8939 * We must not return until all unregister events added during
8940 * the interval the lock was held have been completed.
8942 void netdev_run_todo(void)
8944 struct list_head list;
8946 /* Snapshot list, allow later requests */
8947 list_replace_init(&net_todo_list, &list);
8952 /* Wait for rcu callbacks to finish before next phase */
8953 if (!list_empty(&list))
8956 while (!list_empty(&list)) {
8957 struct net_device *dev
8958 = list_first_entry(&list, struct net_device, todo_list);
8959 list_del(&dev->todo_list);
8961 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8962 pr_err("network todo '%s' but state %d\n",
8963 dev->name, dev->reg_state);
8968 dev->reg_state = NETREG_UNREGISTERED;
8970 netdev_wait_allrefs(dev);
8973 BUG_ON(netdev_refcnt_read(dev));
8974 BUG_ON(!list_empty(&dev->ptype_all));
8975 BUG_ON(!list_empty(&dev->ptype_specific));
8976 WARN_ON(rcu_access_pointer(dev->ip_ptr));
8977 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
8978 #if IS_ENABLED(CONFIG_DECNET)
8979 WARN_ON(dev->dn_ptr);
8981 if (dev->priv_destructor)
8982 dev->priv_destructor(dev);
8983 if (dev->needs_free_netdev)
8986 /* Report a network device has been unregistered */
8988 dev_net(dev)->dev_unreg_count--;
8990 wake_up(&netdev_unregistering_wq);
8992 /* Free network device */
8993 kobject_put(&dev->dev.kobj);
8997 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8998 * all the same fields in the same order as net_device_stats, with only
8999 * the type differing, but rtnl_link_stats64 may have additional fields
9000 * at the end for newer counters.
9002 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9003 const struct net_device_stats *netdev_stats)
9005 #if BITS_PER_LONG == 64
9006 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9007 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9008 /* zero out counters that only exist in rtnl_link_stats64 */
9009 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9010 sizeof(*stats64) - sizeof(*netdev_stats));
9012 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9013 const unsigned long *src = (const unsigned long *)netdev_stats;
9014 u64 *dst = (u64 *)stats64;
9016 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9017 for (i = 0; i < n; i++)
9019 /* zero out counters that only exist in rtnl_link_stats64 */
9020 memset((char *)stats64 + n * sizeof(u64), 0,
9021 sizeof(*stats64) - n * sizeof(u64));
9024 EXPORT_SYMBOL(netdev_stats_to_stats64);
9027 * dev_get_stats - get network device statistics
9028 * @dev: device to get statistics from
9029 * @storage: place to store stats
9031 * Get network statistics from device. Return @storage.
9032 * The device driver may provide its own method by setting
9033 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9034 * otherwise the internal statistics structure is used.
9036 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9037 struct rtnl_link_stats64 *storage)
9039 const struct net_device_ops *ops = dev->netdev_ops;
9041 if (ops->ndo_get_stats64) {
9042 memset(storage, 0, sizeof(*storage));
9043 ops->ndo_get_stats64(dev, storage);
9044 } else if (ops->ndo_get_stats) {
9045 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9047 netdev_stats_to_stats64(storage, &dev->stats);
9049 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9050 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9051 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9054 EXPORT_SYMBOL(dev_get_stats);
9056 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9058 struct netdev_queue *queue = dev_ingress_queue(dev);
9060 #ifdef CONFIG_NET_CLS_ACT
9063 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9066 netdev_init_one_queue(dev, queue, NULL);
9067 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9068 queue->qdisc_sleeping = &noop_qdisc;
9069 rcu_assign_pointer(dev->ingress_queue, queue);
9074 static const struct ethtool_ops default_ethtool_ops;
9076 void netdev_set_default_ethtool_ops(struct net_device *dev,
9077 const struct ethtool_ops *ops)
9079 if (dev->ethtool_ops == &default_ethtool_ops)
9080 dev->ethtool_ops = ops;
9082 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9084 void netdev_freemem(struct net_device *dev)
9086 char *addr = (char *)dev - dev->padded;
9092 * alloc_netdev_mqs - allocate network device
9093 * @sizeof_priv: size of private data to allocate space for
9094 * @name: device name format string
9095 * @name_assign_type: origin of device name
9096 * @setup: callback to initialize device
9097 * @txqs: the number of TX subqueues to allocate
9098 * @rxqs: the number of RX subqueues to allocate
9100 * Allocates a struct net_device with private data area for driver use
9101 * and performs basic initialization. Also allocates subqueue structs
9102 * for each queue on the device.
9104 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9105 unsigned char name_assign_type,
9106 void (*setup)(struct net_device *),
9107 unsigned int txqs, unsigned int rxqs)
9109 struct net_device *dev;
9110 unsigned int alloc_size;
9111 struct net_device *p;
9113 BUG_ON(strlen(name) >= sizeof(dev->name));
9116 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9121 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9125 alloc_size = sizeof(struct net_device);
9127 /* ensure 32-byte alignment of private area */
9128 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9129 alloc_size += sizeof_priv;
9131 /* ensure 32-byte alignment of whole construct */
9132 alloc_size += NETDEV_ALIGN - 1;
9134 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9138 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9139 dev->padded = (char *)dev - (char *)p;
9141 dev->pcpu_refcnt = alloc_percpu(int);
9142 if (!dev->pcpu_refcnt)
9145 if (dev_addr_init(dev))
9151 dev_net_set(dev, &init_net);
9153 dev->gso_max_size = GSO_MAX_SIZE;
9154 dev->gso_max_segs = GSO_MAX_SEGS;
9156 INIT_LIST_HEAD(&dev->napi_list);
9157 INIT_LIST_HEAD(&dev->unreg_list);
9158 INIT_LIST_HEAD(&dev->close_list);
9159 INIT_LIST_HEAD(&dev->link_watch_list);
9160 INIT_LIST_HEAD(&dev->adj_list.upper);
9161 INIT_LIST_HEAD(&dev->adj_list.lower);
9162 INIT_LIST_HEAD(&dev->ptype_all);
9163 INIT_LIST_HEAD(&dev->ptype_specific);
9164 #ifdef CONFIG_NET_SCHED
9165 hash_init(dev->qdisc_hash);
9167 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9170 if (!dev->tx_queue_len) {
9171 dev->priv_flags |= IFF_NO_QUEUE;
9172 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9175 dev->num_tx_queues = txqs;
9176 dev->real_num_tx_queues = txqs;
9177 if (netif_alloc_netdev_queues(dev))
9180 dev->num_rx_queues = rxqs;
9181 dev->real_num_rx_queues = rxqs;
9182 if (netif_alloc_rx_queues(dev))
9185 strcpy(dev->name, name);
9186 dev->name_assign_type = name_assign_type;
9187 dev->group = INIT_NETDEV_GROUP;
9188 if (!dev->ethtool_ops)
9189 dev->ethtool_ops = &default_ethtool_ops;
9191 nf_hook_ingress_init(dev);
9200 free_percpu(dev->pcpu_refcnt);
9202 netdev_freemem(dev);
9205 EXPORT_SYMBOL(alloc_netdev_mqs);
9208 * free_netdev - free network device
9211 * This function does the last stage of destroying an allocated device
9212 * interface. The reference to the device object is released. If this
9213 * is the last reference then it will be freed.Must be called in process
9216 void free_netdev(struct net_device *dev)
9218 struct napi_struct *p, *n;
9221 netif_free_tx_queues(dev);
9222 netif_free_rx_queues(dev);
9224 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9226 /* Flush device addresses */
9227 dev_addr_flush(dev);
9229 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9232 free_percpu(dev->pcpu_refcnt);
9233 dev->pcpu_refcnt = NULL;
9235 /* Compatibility with error handling in drivers */
9236 if (dev->reg_state == NETREG_UNINITIALIZED) {
9237 netdev_freemem(dev);
9241 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9242 dev->reg_state = NETREG_RELEASED;
9244 /* will free via device release */
9245 put_device(&dev->dev);
9247 EXPORT_SYMBOL(free_netdev);
9250 * synchronize_net - Synchronize with packet receive processing
9252 * Wait for packets currently being received to be done.
9253 * Does not block later packets from starting.
9255 void synchronize_net(void)
9258 if (rtnl_is_locked())
9259 synchronize_rcu_expedited();
9263 EXPORT_SYMBOL(synchronize_net);
9266 * unregister_netdevice_queue - remove device from the kernel
9270 * This function shuts down a device interface and removes it
9271 * from the kernel tables.
9272 * If head not NULL, device is queued to be unregistered later.
9274 * Callers must hold the rtnl semaphore. You may want
9275 * unregister_netdev() instead of this.
9278 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9283 list_move_tail(&dev->unreg_list, head);
9285 rollback_registered(dev);
9286 /* Finish processing unregister after unlock */
9290 EXPORT_SYMBOL(unregister_netdevice_queue);
9293 * unregister_netdevice_many - unregister many devices
9294 * @head: list of devices
9296 * Note: As most callers use a stack allocated list_head,
9297 * we force a list_del() to make sure stack wont be corrupted later.
9299 void unregister_netdevice_many(struct list_head *head)
9301 struct net_device *dev;
9303 if (!list_empty(head)) {
9304 rollback_registered_many(head);
9305 list_for_each_entry(dev, head, unreg_list)
9310 EXPORT_SYMBOL(unregister_netdevice_many);
9313 * unregister_netdev - remove device from the kernel
9316 * This function shuts down a device interface and removes it
9317 * from the kernel tables.
9319 * This is just a wrapper for unregister_netdevice that takes
9320 * the rtnl semaphore. In general you want to use this and not
9321 * unregister_netdevice.
9323 void unregister_netdev(struct net_device *dev)
9326 unregister_netdevice(dev);
9329 EXPORT_SYMBOL(unregister_netdev);
9332 * dev_change_net_namespace - move device to different nethost namespace
9334 * @net: network namespace
9335 * @pat: If not NULL name pattern to try if the current device name
9336 * is already taken in the destination network namespace.
9338 * This function shuts down a device interface and moves it
9339 * to a new network namespace. On success 0 is returned, on
9340 * a failure a netagive errno code is returned.
9342 * Callers must hold the rtnl semaphore.
9345 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9347 int err, new_nsid, new_ifindex;
9351 /* Don't allow namespace local devices to be moved. */
9353 if (dev->features & NETIF_F_NETNS_LOCAL)
9356 /* Ensure the device has been registrered */
9357 if (dev->reg_state != NETREG_REGISTERED)
9360 /* Get out if there is nothing todo */
9362 if (net_eq(dev_net(dev), net))
9365 /* Pick the destination device name, and ensure
9366 * we can use it in the destination network namespace.
9369 if (__dev_get_by_name(net, dev->name)) {
9370 /* We get here if we can't use the current device name */
9373 err = dev_get_valid_name(net, dev, pat);
9379 * And now a mini version of register_netdevice unregister_netdevice.
9382 /* If device is running close it first. */
9385 /* And unlink it from device chain */
9386 unlist_netdevice(dev);
9390 /* Shutdown queueing discipline. */
9393 /* Notify protocols, that we are about to destroy
9394 * this device. They should clean all the things.
9396 * Note that dev->reg_state stays at NETREG_REGISTERED.
9397 * This is wanted because this way 8021q and macvlan know
9398 * the device is just moving and can keep their slaves up.
9400 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9403 new_nsid = peernet2id_alloc(dev_net(dev), net);
9404 /* If there is an ifindex conflict assign a new one */
9405 if (__dev_get_by_index(net, dev->ifindex))
9406 new_ifindex = dev_new_index(net);
9408 new_ifindex = dev->ifindex;
9410 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9414 * Flush the unicast and multicast chains
9419 /* Send a netdev-removed uevent to the old namespace */
9420 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9421 netdev_adjacent_del_links(dev);
9423 /* Actually switch the network namespace */
9424 dev_net_set(dev, net);
9425 dev->ifindex = new_ifindex;
9427 /* Send a netdev-add uevent to the new namespace */
9428 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9429 netdev_adjacent_add_links(dev);
9431 /* Fixup kobjects */
9432 err = device_rename(&dev->dev, dev->name);
9435 /* Add the device back in the hashes */
9436 list_netdevice(dev);
9438 /* Notify protocols, that a new device appeared. */
9439 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9442 * Prevent userspace races by waiting until the network
9443 * device is fully setup before sending notifications.
9445 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9452 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9454 static int dev_cpu_dead(unsigned int oldcpu)
9456 struct sk_buff **list_skb;
9457 struct sk_buff *skb;
9459 struct softnet_data *sd, *oldsd, *remsd = NULL;
9461 local_irq_disable();
9462 cpu = smp_processor_id();
9463 sd = &per_cpu(softnet_data, cpu);
9464 oldsd = &per_cpu(softnet_data, oldcpu);
9466 /* Find end of our completion_queue. */
9467 list_skb = &sd->completion_queue;
9469 list_skb = &(*list_skb)->next;
9470 /* Append completion queue from offline CPU. */
9471 *list_skb = oldsd->completion_queue;
9472 oldsd->completion_queue = NULL;
9474 /* Append output queue from offline CPU. */
9475 if (oldsd->output_queue) {
9476 *sd->output_queue_tailp = oldsd->output_queue;
9477 sd->output_queue_tailp = oldsd->output_queue_tailp;
9478 oldsd->output_queue = NULL;
9479 oldsd->output_queue_tailp = &oldsd->output_queue;
9481 /* Append NAPI poll list from offline CPU, with one exception :
9482 * process_backlog() must be called by cpu owning percpu backlog.
9483 * We properly handle process_queue & input_pkt_queue later.
9485 while (!list_empty(&oldsd->poll_list)) {
9486 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9490 list_del_init(&napi->poll_list);
9491 if (napi->poll == process_backlog)
9494 ____napi_schedule(sd, napi);
9497 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9501 remsd = oldsd->rps_ipi_list;
9502 oldsd->rps_ipi_list = NULL;
9504 /* send out pending IPI's on offline CPU */
9505 net_rps_send_ipi(remsd);
9507 /* Process offline CPU's input_pkt_queue */
9508 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9510 input_queue_head_incr(oldsd);
9512 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9514 input_queue_head_incr(oldsd);
9521 * netdev_increment_features - increment feature set by one
9522 * @all: current feature set
9523 * @one: new feature set
9524 * @mask: mask feature set
9526 * Computes a new feature set after adding a device with feature set
9527 * @one to the master device with current feature set @all. Will not
9528 * enable anything that is off in @mask. Returns the new feature set.
9530 netdev_features_t netdev_increment_features(netdev_features_t all,
9531 netdev_features_t one, netdev_features_t mask)
9533 if (mask & NETIF_F_HW_CSUM)
9534 mask |= NETIF_F_CSUM_MASK;
9535 mask |= NETIF_F_VLAN_CHALLENGED;
9537 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9538 all &= one | ~NETIF_F_ALL_FOR_ALL;
9540 /* If one device supports hw checksumming, set for all. */
9541 if (all & NETIF_F_HW_CSUM)
9542 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
9546 EXPORT_SYMBOL(netdev_increment_features);
9548 static struct hlist_head * __net_init netdev_create_hash(void)
9551 struct hlist_head *hash;
9553 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
9555 for (i = 0; i < NETDEV_HASHENTRIES; i++)
9556 INIT_HLIST_HEAD(&hash[i]);
9561 /* Initialize per network namespace state */
9562 static int __net_init netdev_init(struct net *net)
9564 BUILD_BUG_ON(GRO_HASH_BUCKETS >
9565 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
9567 if (net != &init_net)
9568 INIT_LIST_HEAD(&net->dev_base_head);
9570 net->dev_name_head = netdev_create_hash();
9571 if (net->dev_name_head == NULL)
9574 net->dev_index_head = netdev_create_hash();
9575 if (net->dev_index_head == NULL)
9581 kfree(net->dev_name_head);
9587 * netdev_drivername - network driver for the device
9588 * @dev: network device
9590 * Determine network driver for device.
9592 const char *netdev_drivername(const struct net_device *dev)
9594 const struct device_driver *driver;
9595 const struct device *parent;
9596 const char *empty = "";
9598 parent = dev->dev.parent;
9602 driver = parent->driver;
9603 if (driver && driver->name)
9604 return driver->name;
9608 static void __netdev_printk(const char *level, const struct net_device *dev,
9609 struct va_format *vaf)
9611 if (dev && dev->dev.parent) {
9612 dev_printk_emit(level[1] - '0',
9615 dev_driver_string(dev->dev.parent),
9616 dev_name(dev->dev.parent),
9617 netdev_name(dev), netdev_reg_state(dev),
9620 printk("%s%s%s: %pV",
9621 level, netdev_name(dev), netdev_reg_state(dev), vaf);
9623 printk("%s(NULL net_device): %pV", level, vaf);
9627 void netdev_printk(const char *level, const struct net_device *dev,
9628 const char *format, ...)
9630 struct va_format vaf;
9633 va_start(args, format);
9638 __netdev_printk(level, dev, &vaf);
9642 EXPORT_SYMBOL(netdev_printk);
9644 #define define_netdev_printk_level(func, level) \
9645 void func(const struct net_device *dev, const char *fmt, ...) \
9647 struct va_format vaf; \
9650 va_start(args, fmt); \
9655 __netdev_printk(level, dev, &vaf); \
9659 EXPORT_SYMBOL(func);
9661 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
9662 define_netdev_printk_level(netdev_alert, KERN_ALERT);
9663 define_netdev_printk_level(netdev_crit, KERN_CRIT);
9664 define_netdev_printk_level(netdev_err, KERN_ERR);
9665 define_netdev_printk_level(netdev_warn, KERN_WARNING);
9666 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
9667 define_netdev_printk_level(netdev_info, KERN_INFO);
9669 static void __net_exit netdev_exit(struct net *net)
9671 kfree(net->dev_name_head);
9672 kfree(net->dev_index_head);
9673 if (net != &init_net)
9674 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
9677 static struct pernet_operations __net_initdata netdev_net_ops = {
9678 .init = netdev_init,
9679 .exit = netdev_exit,
9682 static void __net_exit default_device_exit(struct net *net)
9684 struct net_device *dev, *aux;
9686 * Push all migratable network devices back to the
9687 * initial network namespace
9690 for_each_netdev_safe(net, dev, aux) {
9692 char fb_name[IFNAMSIZ];
9694 /* Ignore unmoveable devices (i.e. loopback) */
9695 if (dev->features & NETIF_F_NETNS_LOCAL)
9698 /* Leave virtual devices for the generic cleanup */
9699 if (dev->rtnl_link_ops)
9702 /* Push remaining network devices to init_net */
9703 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
9704 err = dev_change_net_namespace(dev, &init_net, fb_name);
9706 pr_emerg("%s: failed to move %s to init_net: %d\n",
9707 __func__, dev->name, err);
9714 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
9716 /* Return with the rtnl_lock held when there are no network
9717 * devices unregistering in any network namespace in net_list.
9721 DEFINE_WAIT_FUNC(wait, woken_wake_function);
9723 add_wait_queue(&netdev_unregistering_wq, &wait);
9725 unregistering = false;
9727 list_for_each_entry(net, net_list, exit_list) {
9728 if (net->dev_unreg_count > 0) {
9729 unregistering = true;
9737 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
9739 remove_wait_queue(&netdev_unregistering_wq, &wait);
9742 static void __net_exit default_device_exit_batch(struct list_head *net_list)
9744 /* At exit all network devices most be removed from a network
9745 * namespace. Do this in the reverse order of registration.
9746 * Do this across as many network namespaces as possible to
9747 * improve batching efficiency.
9749 struct net_device *dev;
9751 LIST_HEAD(dev_kill_list);
9753 /* To prevent network device cleanup code from dereferencing
9754 * loopback devices or network devices that have been freed
9755 * wait here for all pending unregistrations to complete,
9756 * before unregistring the loopback device and allowing the
9757 * network namespace be freed.
9759 * The netdev todo list containing all network devices
9760 * unregistrations that happen in default_device_exit_batch
9761 * will run in the rtnl_unlock() at the end of
9762 * default_device_exit_batch.
9764 rtnl_lock_unregistering(net_list);
9765 list_for_each_entry(net, net_list, exit_list) {
9766 for_each_netdev_reverse(net, dev) {
9767 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
9768 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
9770 unregister_netdevice_queue(dev, &dev_kill_list);
9773 unregister_netdevice_many(&dev_kill_list);
9777 static struct pernet_operations __net_initdata default_device_ops = {
9778 .exit = default_device_exit,
9779 .exit_batch = default_device_exit_batch,
9783 * Initialize the DEV module. At boot time this walks the device list and
9784 * unhooks any devices that fail to initialise (normally hardware not
9785 * present) and leaves us with a valid list of present and active devices.
9790 * This is called single threaded during boot, so no need
9791 * to take the rtnl semaphore.
9793 static int __init net_dev_init(void)
9795 int i, rc = -ENOMEM;
9797 BUG_ON(!dev_boot_phase);
9799 if (dev_proc_init())
9802 if (netdev_kobject_init())
9805 INIT_LIST_HEAD(&ptype_all);
9806 for (i = 0; i < PTYPE_HASH_SIZE; i++)
9807 INIT_LIST_HEAD(&ptype_base[i]);
9809 INIT_LIST_HEAD(&offload_base);
9811 if (register_pernet_subsys(&netdev_net_ops))
9815 * Initialise the packet receive queues.
9818 for_each_possible_cpu(i) {
9819 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
9820 struct softnet_data *sd = &per_cpu(softnet_data, i);
9822 INIT_WORK(flush, flush_backlog);
9824 skb_queue_head_init(&sd->input_pkt_queue);
9825 skb_queue_head_init(&sd->process_queue);
9826 #ifdef CONFIG_XFRM_OFFLOAD
9827 skb_queue_head_init(&sd->xfrm_backlog);
9829 INIT_LIST_HEAD(&sd->poll_list);
9830 sd->output_queue_tailp = &sd->output_queue;
9832 sd->csd.func = rps_trigger_softirq;
9837 init_gro_hash(&sd->backlog);
9838 sd->backlog.poll = process_backlog;
9839 sd->backlog.weight = weight_p;
9844 /* The loopback device is special if any other network devices
9845 * is present in a network namespace the loopback device must
9846 * be present. Since we now dynamically allocate and free the
9847 * loopback device ensure this invariant is maintained by
9848 * keeping the loopback device as the first device on the
9849 * list of network devices. Ensuring the loopback devices
9850 * is the first device that appears and the last network device
9853 if (register_pernet_device(&loopback_net_ops))
9856 if (register_pernet_device(&default_device_ops))
9859 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
9860 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
9862 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
9863 NULL, dev_cpu_dead);
9870 subsys_initcall(net_dev_init);