2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/sched/mm.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
105 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <linux/pci.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_ingress.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
151 #include "net-sysfs.h"
153 #define MAX_GRO_SKBS 8
155 /* This should be increased if a protocol with a bigger head is added. */
156 #define GRO_MAX_HEAD (MAX_HEADER + 128)
158 static DEFINE_SPINLOCK(ptype_lock);
159 static DEFINE_SPINLOCK(offload_lock);
160 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
161 struct list_head ptype_all __read_mostly; /* Taps */
162 static struct list_head offload_base __read_mostly;
164 static int netif_rx_internal(struct sk_buff *skb);
165 static int call_netdevice_notifiers_info(unsigned long val,
166 struct netdev_notifier_info *info);
167 static int call_netdevice_notifiers_extack(unsigned long val,
168 struct net_device *dev,
169 struct netlink_ext_ack *extack);
170 static struct napi_struct *napi_by_id(unsigned int napi_id);
173 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
176 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
178 * Writers must hold the rtnl semaphore while they loop through the
179 * dev_base_head list, and hold dev_base_lock for writing when they do the
180 * actual updates. This allows pure readers to access the list even
181 * while a writer is preparing to update it.
183 * To put it another way, dev_base_lock is held for writing only to
184 * protect against pure readers; the rtnl semaphore provides the
185 * protection against other writers.
187 * See, for example usages, register_netdevice() and
188 * unregister_netdevice(), which must be called with the rtnl
191 DEFINE_RWLOCK(dev_base_lock);
192 EXPORT_SYMBOL(dev_base_lock);
194 static DEFINE_MUTEX(ifalias_mutex);
196 /* protects napi_hash addition/deletion and napi_gen_id */
197 static DEFINE_SPINLOCK(napi_hash_lock);
199 static unsigned int napi_gen_id = NR_CPUS;
200 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
202 static seqcount_t devnet_rename_seq;
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0)
210 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
212 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static void list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
253 /* Device list removal
254 * caller must respect a RCU grace period before freeing/reusing dev
256 static void unlist_netdevice(struct net_device *dev)
260 /* Unlink dev from the device chain */
261 write_lock_bh(&dev_base_lock);
262 list_del_rcu(&dev->dev_list);
263 hlist_del_rcu(&dev->name_hlist);
264 hlist_del_rcu(&dev->index_hlist);
265 write_unlock_bh(&dev_base_lock);
267 dev_base_seq_inc(dev_net(dev));
274 static RAW_NOTIFIER_HEAD(netdev_chain);
277 * Device drivers call our routines to queue packets here. We empty the
278 * queue in the local softnet handler.
281 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
282 EXPORT_PER_CPU_SYMBOL(softnet_data);
284 #ifdef CONFIG_LOCKDEP
286 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
287 * according to dev->type
289 static const unsigned short netdev_lock_type[] = {
290 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
291 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
292 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
293 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
294 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
295 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
296 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
297 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
298 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
299 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
300 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
301 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
302 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
303 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
304 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
306 static const char *const netdev_lock_name[] = {
307 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
308 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
309 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
310 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
311 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
312 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
313 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
314 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
315 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
316 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
317 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
318 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
319 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
320 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
321 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
323 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
324 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
326 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
330 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
331 if (netdev_lock_type[i] == dev_type)
333 /* the last key is used by default */
334 return ARRAY_SIZE(netdev_lock_type) - 1;
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
338 unsigned short dev_type)
342 i = netdev_lock_pos(dev_type);
343 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
351 i = netdev_lock_pos(dev->type);
352 lockdep_set_class_and_name(&dev->addr_list_lock,
353 &netdev_addr_lock_key[i],
354 netdev_lock_name[i]);
357 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
358 unsigned short dev_type)
361 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
366 /*******************************************************************************
368 * Protocol management and registration routines
370 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
392 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
394 return pt->dev ? &pt->dev->ptype_specific :
395 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
399 * dev_add_pack - add packet handler
400 * @pt: packet type declaration
402 * Add a protocol handler to the networking stack. The passed &packet_type
403 * is linked into kernel lists and may not be freed until it has been
404 * removed from the kernel lists.
406 * This call does not sleep therefore it can not
407 * guarantee all CPU's that are in middle of receiving packets
408 * will see the new packet type (until the next received packet).
411 void dev_add_pack(struct packet_type *pt)
413 struct list_head *head = ptype_head(pt);
415 spin_lock(&ptype_lock);
416 list_add_rcu(&pt->list, head);
417 spin_unlock(&ptype_lock);
419 EXPORT_SYMBOL(dev_add_pack);
422 * __dev_remove_pack - remove packet handler
423 * @pt: packet type declaration
425 * Remove a protocol handler that was previously added to the kernel
426 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
427 * from the kernel lists and can be freed or reused once this function
430 * The packet type might still be in use by receivers
431 * and must not be freed until after all the CPU's have gone
432 * through a quiescent state.
434 void __dev_remove_pack(struct packet_type *pt)
436 struct list_head *head = ptype_head(pt);
437 struct packet_type *pt1;
439 spin_lock(&ptype_lock);
441 list_for_each_entry(pt1, head, list) {
443 list_del_rcu(&pt->list);
448 pr_warn("dev_remove_pack: %p not found\n", pt);
450 spin_unlock(&ptype_lock);
452 EXPORT_SYMBOL(__dev_remove_pack);
455 * dev_remove_pack - remove packet handler
456 * @pt: packet type declaration
458 * Remove a protocol handler that was previously added to the kernel
459 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
460 * from the kernel lists and can be freed or reused once this function
463 * This call sleeps to guarantee that no CPU is looking at the packet
466 void dev_remove_pack(struct packet_type *pt)
468 __dev_remove_pack(pt);
472 EXPORT_SYMBOL(dev_remove_pack);
476 * dev_add_offload - register offload handlers
477 * @po: protocol offload declaration
479 * Add protocol offload handlers to the networking stack. The passed
480 * &proto_offload is linked into kernel lists and may not be freed until
481 * it has been removed from the kernel lists.
483 * This call does not sleep therefore it can not
484 * guarantee all CPU's that are in middle of receiving packets
485 * will see the new offload handlers (until the next received packet).
487 void dev_add_offload(struct packet_offload *po)
489 struct packet_offload *elem;
491 spin_lock(&offload_lock);
492 list_for_each_entry(elem, &offload_base, list) {
493 if (po->priority < elem->priority)
496 list_add_rcu(&po->list, elem->list.prev);
497 spin_unlock(&offload_lock);
499 EXPORT_SYMBOL(dev_add_offload);
502 * __dev_remove_offload - remove offload handler
503 * @po: packet offload declaration
505 * Remove a protocol offload handler that was previously added to the
506 * kernel offload handlers by dev_add_offload(). The passed &offload_type
507 * is removed from the kernel lists and can be freed or reused once this
510 * The packet type might still be in use by receivers
511 * and must not be freed until after all the CPU's have gone
512 * through a quiescent state.
514 static void __dev_remove_offload(struct packet_offload *po)
516 struct list_head *head = &offload_base;
517 struct packet_offload *po1;
519 spin_lock(&offload_lock);
521 list_for_each_entry(po1, head, list) {
523 list_del_rcu(&po->list);
528 pr_warn("dev_remove_offload: %p not found\n", po);
530 spin_unlock(&offload_lock);
534 * dev_remove_offload - remove packet offload handler
535 * @po: packet offload declaration
537 * Remove a packet offload handler that was previously added to the kernel
538 * offload handlers by dev_add_offload(). The passed &offload_type is
539 * removed from the kernel lists and can be freed or reused once this
542 * This call sleeps to guarantee that no CPU is looking at the packet
545 void dev_remove_offload(struct packet_offload *po)
547 __dev_remove_offload(po);
551 EXPORT_SYMBOL(dev_remove_offload);
553 /******************************************************************************
555 * Device Boot-time Settings Routines
557 ******************************************************************************/
559 /* Boot time configuration table */
560 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
563 * netdev_boot_setup_add - add new setup entry
564 * @name: name of the device
565 * @map: configured settings for the device
567 * Adds new setup entry to the dev_boot_setup list. The function
568 * returns 0 on error and 1 on success. This is a generic routine to
571 static int netdev_boot_setup_add(char *name, struct ifmap *map)
573 struct netdev_boot_setup *s;
577 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
578 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
579 memset(s[i].name, 0, sizeof(s[i].name));
580 strlcpy(s[i].name, name, IFNAMSIZ);
581 memcpy(&s[i].map, map, sizeof(s[i].map));
586 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
590 * netdev_boot_setup_check - check boot time settings
591 * @dev: the netdevice
593 * Check boot time settings for the device.
594 * The found settings are set for the device to be used
595 * later in the device probing.
596 * Returns 0 if no settings found, 1 if they are.
598 int netdev_boot_setup_check(struct net_device *dev)
600 struct netdev_boot_setup *s = dev_boot_setup;
603 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
604 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
605 !strcmp(dev->name, s[i].name)) {
606 dev->irq = s[i].map.irq;
607 dev->base_addr = s[i].map.base_addr;
608 dev->mem_start = s[i].map.mem_start;
609 dev->mem_end = s[i].map.mem_end;
615 EXPORT_SYMBOL(netdev_boot_setup_check);
619 * netdev_boot_base - get address from boot time settings
620 * @prefix: prefix for network device
621 * @unit: id for network device
623 * Check boot time settings for the base address of device.
624 * The found settings are set for the device to be used
625 * later in the device probing.
626 * Returns 0 if no settings found.
628 unsigned long netdev_boot_base(const char *prefix, int unit)
630 const struct netdev_boot_setup *s = dev_boot_setup;
634 sprintf(name, "%s%d", prefix, unit);
637 * If device already registered then return base of 1
638 * to indicate not to probe for this interface
640 if (__dev_get_by_name(&init_net, name))
643 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
644 if (!strcmp(name, s[i].name))
645 return s[i].map.base_addr;
650 * Saves at boot time configured settings for any netdevice.
652 int __init netdev_boot_setup(char *str)
657 str = get_options(str, ARRAY_SIZE(ints), ints);
662 memset(&map, 0, sizeof(map));
666 map.base_addr = ints[2];
668 map.mem_start = ints[3];
670 map.mem_end = ints[4];
672 /* Add new entry to the list */
673 return netdev_boot_setup_add(str, &map);
676 __setup("netdev=", netdev_boot_setup);
678 /*******************************************************************************
680 * Device Interface Subroutines
682 *******************************************************************************/
685 * dev_get_iflink - get 'iflink' value of a interface
686 * @dev: targeted interface
688 * Indicates the ifindex the interface is linked to.
689 * Physical interfaces have the same 'ifindex' and 'iflink' values.
692 int dev_get_iflink(const struct net_device *dev)
694 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
695 return dev->netdev_ops->ndo_get_iflink(dev);
699 EXPORT_SYMBOL(dev_get_iflink);
702 * dev_fill_metadata_dst - Retrieve tunnel egress information.
703 * @dev: targeted interface
706 * For better visibility of tunnel traffic OVS needs to retrieve
707 * egress tunnel information for a packet. Following API allows
708 * user to get this info.
710 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
712 struct ip_tunnel_info *info;
714 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
717 info = skb_tunnel_info_unclone(skb);
720 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
723 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
725 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
728 * __dev_get_by_name - find a device by its name
729 * @net: the applicable net namespace
730 * @name: name to find
732 * Find an interface by name. Must be called under RTNL semaphore
733 * or @dev_base_lock. If the name is found a pointer to the device
734 * is returned. If the name is not found then %NULL is returned. The
735 * reference counters are not incremented so the caller must be
736 * careful with locks.
739 struct net_device *__dev_get_by_name(struct net *net, const char *name)
741 struct net_device *dev;
742 struct hlist_head *head = dev_name_hash(net, name);
744 hlist_for_each_entry(dev, head, name_hlist)
745 if (!strncmp(dev->name, name, IFNAMSIZ))
750 EXPORT_SYMBOL(__dev_get_by_name);
753 * dev_get_by_name_rcu - find a device by its name
754 * @net: the applicable net namespace
755 * @name: name to find
757 * Find an interface by name.
758 * If the name is found a pointer to the device is returned.
759 * If the name is not found then %NULL is returned.
760 * The reference counters are not incremented so the caller must be
761 * careful with locks. The caller must hold RCU lock.
764 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
766 struct net_device *dev;
767 struct hlist_head *head = dev_name_hash(net, name);
769 hlist_for_each_entry_rcu(dev, head, name_hlist)
770 if (!strncmp(dev->name, name, IFNAMSIZ))
775 EXPORT_SYMBOL(dev_get_by_name_rcu);
778 * dev_get_by_name - find a device by its name
779 * @net: the applicable net namespace
780 * @name: name to find
782 * Find an interface by name. This can be called from any
783 * context and does its own locking. The returned handle has
784 * the usage count incremented and the caller must use dev_put() to
785 * release it when it is no longer needed. %NULL is returned if no
786 * matching device is found.
789 struct net_device *dev_get_by_name(struct net *net, const char *name)
791 struct net_device *dev;
794 dev = dev_get_by_name_rcu(net, name);
800 EXPORT_SYMBOL(dev_get_by_name);
803 * __dev_get_by_index - find a device by its ifindex
804 * @net: the applicable net namespace
805 * @ifindex: index of device
807 * Search for an interface by index. Returns %NULL if the device
808 * is not found or a pointer to the device. The device has not
809 * had its reference counter increased so the caller must be careful
810 * about locking. The caller must hold either the RTNL semaphore
814 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
816 struct net_device *dev;
817 struct hlist_head *head = dev_index_hash(net, ifindex);
819 hlist_for_each_entry(dev, head, index_hlist)
820 if (dev->ifindex == ifindex)
825 EXPORT_SYMBOL(__dev_get_by_index);
828 * dev_get_by_index_rcu - find a device by its ifindex
829 * @net: the applicable net namespace
830 * @ifindex: index of device
832 * Search for an interface by index. Returns %NULL if the device
833 * is not found or a pointer to the device. The device has not
834 * had its reference counter increased so the caller must be careful
835 * about locking. The caller must hold RCU lock.
838 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
840 struct net_device *dev;
841 struct hlist_head *head = dev_index_hash(net, ifindex);
843 hlist_for_each_entry_rcu(dev, head, index_hlist)
844 if (dev->ifindex == ifindex)
849 EXPORT_SYMBOL(dev_get_by_index_rcu);
853 * dev_get_by_index - find a device by its ifindex
854 * @net: the applicable net namespace
855 * @ifindex: index of device
857 * Search for an interface by index. Returns NULL if the device
858 * is not found or a pointer to the device. The device returned has
859 * had a reference added and the pointer is safe until the user calls
860 * dev_put to indicate they have finished with it.
863 struct net_device *dev_get_by_index(struct net *net, int ifindex)
865 struct net_device *dev;
868 dev = dev_get_by_index_rcu(net, ifindex);
874 EXPORT_SYMBOL(dev_get_by_index);
877 * dev_get_by_napi_id - find a device by napi_id
878 * @napi_id: ID of the NAPI struct
880 * Search for an interface by NAPI ID. Returns %NULL if the device
881 * is not found or a pointer to the device. The device has not had
882 * its reference counter increased so the caller must be careful
883 * about locking. The caller must hold RCU lock.
886 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
888 struct napi_struct *napi;
890 WARN_ON_ONCE(!rcu_read_lock_held());
892 if (napi_id < MIN_NAPI_ID)
895 napi = napi_by_id(napi_id);
897 return napi ? napi->dev : NULL;
899 EXPORT_SYMBOL(dev_get_by_napi_id);
902 * netdev_get_name - get a netdevice name, knowing its ifindex.
903 * @net: network namespace
904 * @name: a pointer to the buffer where the name will be stored.
905 * @ifindex: the ifindex of the interface to get the name from.
907 * The use of raw_seqcount_begin() and cond_resched() before
908 * retrying is required as we want to give the writers a chance
909 * to complete when CONFIG_PREEMPT is not set.
911 int netdev_get_name(struct net *net, char *name, int ifindex)
913 struct net_device *dev;
917 seq = raw_seqcount_begin(&devnet_rename_seq);
919 dev = dev_get_by_index_rcu(net, ifindex);
925 strcpy(name, dev->name);
927 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
936 * dev_getbyhwaddr_rcu - find a device by its hardware address
937 * @net: the applicable net namespace
938 * @type: media type of device
939 * @ha: hardware address
941 * Search for an interface by MAC address. Returns NULL if the device
942 * is not found or a pointer to the device.
943 * The caller must hold RCU or RTNL.
944 * The returned device has not had its ref count increased
945 * and the caller must therefore be careful about locking
949 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
952 struct net_device *dev;
954 for_each_netdev_rcu(net, dev)
955 if (dev->type == type &&
956 !memcmp(dev->dev_addr, ha, dev->addr_len))
961 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
963 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
965 struct net_device *dev;
968 for_each_netdev(net, dev)
969 if (dev->type == type)
974 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
976 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
978 struct net_device *dev, *ret = NULL;
981 for_each_netdev_rcu(net, dev)
982 if (dev->type == type) {
990 EXPORT_SYMBOL(dev_getfirstbyhwtype);
993 * __dev_get_by_flags - find any device with given flags
994 * @net: the applicable net namespace
995 * @if_flags: IFF_* values
996 * @mask: bitmask of bits in if_flags to check
998 * Search for any interface with the given flags. Returns NULL if a device
999 * is not found or a pointer to the device. Must be called inside
1000 * rtnl_lock(), and result refcount is unchanged.
1003 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1004 unsigned short mask)
1006 struct net_device *dev, *ret;
1011 for_each_netdev(net, dev) {
1012 if (((dev->flags ^ if_flags) & mask) == 0) {
1019 EXPORT_SYMBOL(__dev_get_by_flags);
1022 * dev_valid_name - check if name is okay for network device
1023 * @name: name string
1025 * Network device names need to be valid file names to
1026 * to allow sysfs to work. We also disallow any kind of
1029 bool dev_valid_name(const char *name)
1033 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1035 if (!strcmp(name, ".") || !strcmp(name, ".."))
1039 if (*name == '/' || *name == ':' || isspace(*name))
1045 EXPORT_SYMBOL(dev_valid_name);
1048 * __dev_alloc_name - allocate a name for a device
1049 * @net: network namespace to allocate the device name in
1050 * @name: name format string
1051 * @buf: scratch buffer and result name string
1053 * Passed a format string - eg "lt%d" it will try and find a suitable
1054 * id. It scans list of devices to build up a free map, then chooses
1055 * the first empty slot. The caller must hold the dev_base or rtnl lock
1056 * while allocating the name and adding the device in order to avoid
1058 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1059 * Returns the number of the unit assigned or a negative errno code.
1062 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1066 const int max_netdevices = 8*PAGE_SIZE;
1067 unsigned long *inuse;
1068 struct net_device *d;
1070 if (!dev_valid_name(name))
1073 p = strchr(name, '%');
1076 * Verify the string as this thing may have come from
1077 * the user. There must be either one "%d" and no other "%"
1080 if (p[1] != 'd' || strchr(p + 2, '%'))
1083 /* Use one page as a bit array of possible slots */
1084 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1088 for_each_netdev(net, d) {
1089 if (!sscanf(d->name, name, &i))
1091 if (i < 0 || i >= max_netdevices)
1094 /* avoid cases where sscanf is not exact inverse of printf */
1095 snprintf(buf, IFNAMSIZ, name, i);
1096 if (!strncmp(buf, d->name, IFNAMSIZ))
1100 i = find_first_zero_bit(inuse, max_netdevices);
1101 free_page((unsigned long) inuse);
1104 snprintf(buf, IFNAMSIZ, name, i);
1105 if (!__dev_get_by_name(net, buf))
1108 /* It is possible to run out of possible slots
1109 * when the name is long and there isn't enough space left
1110 * for the digits, or if all bits are used.
1115 static int dev_alloc_name_ns(struct net *net,
1116 struct net_device *dev,
1123 ret = __dev_alloc_name(net, name, buf);
1125 strlcpy(dev->name, buf, IFNAMSIZ);
1130 * dev_alloc_name - allocate a name for a device
1132 * @name: name format string
1134 * Passed a format string - eg "lt%d" it will try and find a suitable
1135 * id. It scans list of devices to build up a free map, then chooses
1136 * the first empty slot. The caller must hold the dev_base or rtnl lock
1137 * while allocating the name and adding the device in order to avoid
1139 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1140 * Returns the number of the unit assigned or a negative errno code.
1143 int dev_alloc_name(struct net_device *dev, const char *name)
1145 return dev_alloc_name_ns(dev_net(dev), dev, name);
1147 EXPORT_SYMBOL(dev_alloc_name);
1149 int dev_get_valid_name(struct net *net, struct net_device *dev,
1154 if (!dev_valid_name(name))
1157 if (strchr(name, '%'))
1158 return dev_alloc_name_ns(net, dev, name);
1159 else if (__dev_get_by_name(net, name))
1161 else if (dev->name != name)
1162 strlcpy(dev->name, name, IFNAMSIZ);
1166 EXPORT_SYMBOL(dev_get_valid_name);
1169 * dev_change_name - change name of a device
1171 * @newname: name (or format string) must be at least IFNAMSIZ
1173 * Change name of a device, can pass format strings "eth%d".
1176 int dev_change_name(struct net_device *dev, const char *newname)
1178 unsigned char old_assign_type;
1179 char oldname[IFNAMSIZ];
1185 BUG_ON(!dev_net(dev));
1188 if (dev->flags & IFF_UP)
1191 write_seqcount_begin(&devnet_rename_seq);
1193 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1194 write_seqcount_end(&devnet_rename_seq);
1198 memcpy(oldname, dev->name, IFNAMSIZ);
1200 err = dev_get_valid_name(net, dev, newname);
1202 write_seqcount_end(&devnet_rename_seq);
1206 if (oldname[0] && !strchr(oldname, '%'))
1207 netdev_info(dev, "renamed from %s\n", oldname);
1209 old_assign_type = dev->name_assign_type;
1210 dev->name_assign_type = NET_NAME_RENAMED;
1213 ret = device_rename(&dev->dev, dev->name);
1215 memcpy(dev->name, oldname, IFNAMSIZ);
1216 dev->name_assign_type = old_assign_type;
1217 write_seqcount_end(&devnet_rename_seq);
1221 write_seqcount_end(&devnet_rename_seq);
1223 netdev_adjacent_rename_links(dev, oldname);
1225 write_lock_bh(&dev_base_lock);
1226 hlist_del_rcu(&dev->name_hlist);
1227 write_unlock_bh(&dev_base_lock);
1231 write_lock_bh(&dev_base_lock);
1232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1233 write_unlock_bh(&dev_base_lock);
1235 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1236 ret = notifier_to_errno(ret);
1239 /* err >= 0 after dev_alloc_name() or stores the first errno */
1242 write_seqcount_begin(&devnet_rename_seq);
1243 memcpy(dev->name, oldname, IFNAMSIZ);
1244 memcpy(oldname, newname, IFNAMSIZ);
1245 dev->name_assign_type = old_assign_type;
1246 old_assign_type = NET_NAME_RENAMED;
1249 pr_err("%s: name change rollback failed: %d\n",
1258 * dev_set_alias - change ifalias of a device
1260 * @alias: name up to IFALIASZ
1261 * @len: limit of bytes to copy from info
1263 * Set ifalias for a device,
1265 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1267 struct dev_ifalias *new_alias = NULL;
1269 if (len >= IFALIASZ)
1273 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1277 memcpy(new_alias->ifalias, alias, len);
1278 new_alias->ifalias[len] = 0;
1281 mutex_lock(&ifalias_mutex);
1282 rcu_swap_protected(dev->ifalias, new_alias,
1283 mutex_is_locked(&ifalias_mutex));
1284 mutex_unlock(&ifalias_mutex);
1287 kfree_rcu(new_alias, rcuhead);
1291 EXPORT_SYMBOL(dev_set_alias);
1294 * dev_get_alias - get ifalias of a device
1296 * @name: buffer to store name of ifalias
1297 * @len: size of buffer
1299 * get ifalias for a device. Caller must make sure dev cannot go
1300 * away, e.g. rcu read lock or own a reference count to device.
1302 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1304 const struct dev_ifalias *alias;
1308 alias = rcu_dereference(dev->ifalias);
1310 ret = snprintf(name, len, "%s", alias->ifalias);
1317 * netdev_features_change - device changes features
1318 * @dev: device to cause notification
1320 * Called to indicate a device has changed features.
1322 void netdev_features_change(struct net_device *dev)
1324 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1326 EXPORT_SYMBOL(netdev_features_change);
1329 * netdev_state_change - device changes state
1330 * @dev: device to cause notification
1332 * Called to indicate a device has changed state. This function calls
1333 * the notifier chains for netdev_chain and sends a NEWLINK message
1334 * to the routing socket.
1336 void netdev_state_change(struct net_device *dev)
1338 if (dev->flags & IFF_UP) {
1339 struct netdev_notifier_change_info change_info = {
1343 call_netdevice_notifiers_info(NETDEV_CHANGE,
1345 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1348 EXPORT_SYMBOL(netdev_state_change);
1351 * netdev_notify_peers - notify network peers about existence of @dev
1352 * @dev: network device
1354 * Generate traffic such that interested network peers are aware of
1355 * @dev, such as by generating a gratuitous ARP. This may be used when
1356 * a device wants to inform the rest of the network about some sort of
1357 * reconfiguration such as a failover event or virtual machine
1360 void netdev_notify_peers(struct net_device *dev)
1363 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1364 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1367 EXPORT_SYMBOL(netdev_notify_peers);
1369 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1371 const struct net_device_ops *ops = dev->netdev_ops;
1376 if (!netif_device_present(dev))
1379 /* Block netpoll from trying to do any rx path servicing.
1380 * If we don't do this there is a chance ndo_poll_controller
1381 * or ndo_poll may be running while we open the device
1383 netpoll_poll_disable(dev);
1385 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1386 ret = notifier_to_errno(ret);
1390 set_bit(__LINK_STATE_START, &dev->state);
1392 if (ops->ndo_validate_addr)
1393 ret = ops->ndo_validate_addr(dev);
1395 if (!ret && ops->ndo_open)
1396 ret = ops->ndo_open(dev);
1398 netpoll_poll_enable(dev);
1401 clear_bit(__LINK_STATE_START, &dev->state);
1403 dev->flags |= IFF_UP;
1404 dev_set_rx_mode(dev);
1406 add_device_randomness(dev->dev_addr, dev->addr_len);
1413 * dev_open - prepare an interface for use.
1414 * @dev: device to open
1415 * @extack: netlink extended ack
1417 * Takes a device from down to up state. The device's private open
1418 * function is invoked and then the multicast lists are loaded. Finally
1419 * the device is moved into the up state and a %NETDEV_UP message is
1420 * sent to the netdev notifier chain.
1422 * Calling this function on an active interface is a nop. On a failure
1423 * a negative errno code is returned.
1425 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1429 if (dev->flags & IFF_UP)
1432 ret = __dev_open(dev, extack);
1436 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1437 call_netdevice_notifiers(NETDEV_UP, dev);
1441 EXPORT_SYMBOL(dev_open);
1443 static void __dev_close_many(struct list_head *head)
1445 struct net_device *dev;
1450 list_for_each_entry(dev, head, close_list) {
1451 /* Temporarily disable netpoll until the interface is down */
1452 netpoll_poll_disable(dev);
1454 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1456 clear_bit(__LINK_STATE_START, &dev->state);
1458 /* Synchronize to scheduled poll. We cannot touch poll list, it
1459 * can be even on different cpu. So just clear netif_running().
1461 * dev->stop() will invoke napi_disable() on all of it's
1462 * napi_struct instances on this device.
1464 smp_mb__after_atomic(); /* Commit netif_running(). */
1467 dev_deactivate_many(head);
1469 list_for_each_entry(dev, head, close_list) {
1470 const struct net_device_ops *ops = dev->netdev_ops;
1473 * Call the device specific close. This cannot fail.
1474 * Only if device is UP
1476 * We allow it to be called even after a DETACH hot-plug
1482 dev->flags &= ~IFF_UP;
1483 netpoll_poll_enable(dev);
1487 static void __dev_close(struct net_device *dev)
1491 list_add(&dev->close_list, &single);
1492 __dev_close_many(&single);
1496 void dev_close_many(struct list_head *head, bool unlink)
1498 struct net_device *dev, *tmp;
1500 /* Remove the devices that don't need to be closed */
1501 list_for_each_entry_safe(dev, tmp, head, close_list)
1502 if (!(dev->flags & IFF_UP))
1503 list_del_init(&dev->close_list);
1505 __dev_close_many(head);
1507 list_for_each_entry_safe(dev, tmp, head, close_list) {
1508 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1509 call_netdevice_notifiers(NETDEV_DOWN, dev);
1511 list_del_init(&dev->close_list);
1514 EXPORT_SYMBOL(dev_close_many);
1517 * dev_close - shutdown an interface.
1518 * @dev: device to shutdown
1520 * This function moves an active device into down state. A
1521 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1522 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1525 void dev_close(struct net_device *dev)
1527 if (dev->flags & IFF_UP) {
1530 list_add(&dev->close_list, &single);
1531 dev_close_many(&single, true);
1535 EXPORT_SYMBOL(dev_close);
1539 * dev_disable_lro - disable Large Receive Offload on a device
1542 * Disable Large Receive Offload (LRO) on a net device. Must be
1543 * called under RTNL. This is needed if received packets may be
1544 * forwarded to another interface.
1546 void dev_disable_lro(struct net_device *dev)
1548 struct net_device *lower_dev;
1549 struct list_head *iter;
1551 dev->wanted_features &= ~NETIF_F_LRO;
1552 netdev_update_features(dev);
1554 if (unlikely(dev->features & NETIF_F_LRO))
1555 netdev_WARN(dev, "failed to disable LRO!\n");
1557 netdev_for_each_lower_dev(dev, lower_dev, iter)
1558 dev_disable_lro(lower_dev);
1560 EXPORT_SYMBOL(dev_disable_lro);
1563 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1566 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1567 * called under RTNL. This is needed if Generic XDP is installed on
1570 static void dev_disable_gro_hw(struct net_device *dev)
1572 dev->wanted_features &= ~NETIF_F_GRO_HW;
1573 netdev_update_features(dev);
1575 if (unlikely(dev->features & NETIF_F_GRO_HW))
1576 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1579 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1582 case NETDEV_##val: \
1583 return "NETDEV_" __stringify(val);
1585 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1586 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1587 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1588 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1589 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1590 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1591 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1592 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1593 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1597 return "UNKNOWN_NETDEV_EVENT";
1599 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1601 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1602 struct net_device *dev)
1604 struct netdev_notifier_info info = {
1608 return nb->notifier_call(nb, val, &info);
1611 static int dev_boot_phase = 1;
1614 * register_netdevice_notifier - register a network notifier block
1617 * Register a notifier to be called when network device events occur.
1618 * The notifier passed is linked into the kernel structures and must
1619 * not be reused until it has been unregistered. A negative errno code
1620 * is returned on a failure.
1622 * When registered all registration and up events are replayed
1623 * to the new notifier to allow device to have a race free
1624 * view of the network device list.
1627 int register_netdevice_notifier(struct notifier_block *nb)
1629 struct net_device *dev;
1630 struct net_device *last;
1634 /* Close race with setup_net() and cleanup_net() */
1635 down_write(&pernet_ops_rwsem);
1637 err = raw_notifier_chain_register(&netdev_chain, nb);
1643 for_each_netdev(net, dev) {
1644 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1645 err = notifier_to_errno(err);
1649 if (!(dev->flags & IFF_UP))
1652 call_netdevice_notifier(nb, NETDEV_UP, dev);
1658 up_write(&pernet_ops_rwsem);
1664 for_each_netdev(net, dev) {
1668 if (dev->flags & IFF_UP) {
1669 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1671 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1673 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1678 raw_notifier_chain_unregister(&netdev_chain, nb);
1681 EXPORT_SYMBOL(register_netdevice_notifier);
1684 * unregister_netdevice_notifier - unregister a network notifier block
1687 * Unregister a notifier previously registered by
1688 * register_netdevice_notifier(). The notifier is unlinked into the
1689 * kernel structures and may then be reused. A negative errno code
1690 * is returned on a failure.
1692 * After unregistering unregister and down device events are synthesized
1693 * for all devices on the device list to the removed notifier to remove
1694 * the need for special case cleanup code.
1697 int unregister_netdevice_notifier(struct notifier_block *nb)
1699 struct net_device *dev;
1703 /* Close race with setup_net() and cleanup_net() */
1704 down_write(&pernet_ops_rwsem);
1706 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1711 for_each_netdev(net, dev) {
1712 if (dev->flags & IFF_UP) {
1713 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1715 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1717 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1722 up_write(&pernet_ops_rwsem);
1725 EXPORT_SYMBOL(unregister_netdevice_notifier);
1728 * call_netdevice_notifiers_info - call all network notifier blocks
1729 * @val: value passed unmodified to notifier function
1730 * @info: notifier information data
1732 * Call all network notifier blocks. Parameters and return value
1733 * are as for raw_notifier_call_chain().
1736 static int call_netdevice_notifiers_info(unsigned long val,
1737 struct netdev_notifier_info *info)
1740 return raw_notifier_call_chain(&netdev_chain, val, info);
1743 static int call_netdevice_notifiers_extack(unsigned long val,
1744 struct net_device *dev,
1745 struct netlink_ext_ack *extack)
1747 struct netdev_notifier_info info = {
1752 return call_netdevice_notifiers_info(val, &info);
1756 * call_netdevice_notifiers - call all network notifier blocks
1757 * @val: value passed unmodified to notifier function
1758 * @dev: net_device pointer passed unmodified to notifier function
1760 * Call all network notifier blocks. Parameters and return value
1761 * are as for raw_notifier_call_chain().
1764 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1766 return call_netdevice_notifiers_extack(val, dev, NULL);
1768 EXPORT_SYMBOL(call_netdevice_notifiers);
1771 * call_netdevice_notifiers_mtu - call all network notifier blocks
1772 * @val: value passed unmodified to notifier function
1773 * @dev: net_device pointer passed unmodified to notifier function
1774 * @arg: additional u32 argument passed to the notifier function
1776 * Call all network notifier blocks. Parameters and return value
1777 * are as for raw_notifier_call_chain().
1779 static int call_netdevice_notifiers_mtu(unsigned long val,
1780 struct net_device *dev, u32 arg)
1782 struct netdev_notifier_info_ext info = {
1787 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1789 return call_netdevice_notifiers_info(val, &info.info);
1792 #ifdef CONFIG_NET_INGRESS
1793 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1795 void net_inc_ingress_queue(void)
1797 static_branch_inc(&ingress_needed_key);
1799 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1801 void net_dec_ingress_queue(void)
1803 static_branch_dec(&ingress_needed_key);
1805 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1808 #ifdef CONFIG_NET_EGRESS
1809 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1811 void net_inc_egress_queue(void)
1813 static_branch_inc(&egress_needed_key);
1815 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1817 void net_dec_egress_queue(void)
1819 static_branch_dec(&egress_needed_key);
1821 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1824 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1825 #ifdef CONFIG_JUMP_LABEL
1826 static atomic_t netstamp_needed_deferred;
1827 static atomic_t netstamp_wanted;
1828 static void netstamp_clear(struct work_struct *work)
1830 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1833 wanted = atomic_add_return(deferred, &netstamp_wanted);
1835 static_branch_enable(&netstamp_needed_key);
1837 static_branch_disable(&netstamp_needed_key);
1839 static DECLARE_WORK(netstamp_work, netstamp_clear);
1842 void net_enable_timestamp(void)
1844 #ifdef CONFIG_JUMP_LABEL
1848 wanted = atomic_read(&netstamp_wanted);
1851 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1854 atomic_inc(&netstamp_needed_deferred);
1855 schedule_work(&netstamp_work);
1857 static_branch_inc(&netstamp_needed_key);
1860 EXPORT_SYMBOL(net_enable_timestamp);
1862 void net_disable_timestamp(void)
1864 #ifdef CONFIG_JUMP_LABEL
1868 wanted = atomic_read(&netstamp_wanted);
1871 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1874 atomic_dec(&netstamp_needed_deferred);
1875 schedule_work(&netstamp_work);
1877 static_branch_dec(&netstamp_needed_key);
1880 EXPORT_SYMBOL(net_disable_timestamp);
1882 static inline void net_timestamp_set(struct sk_buff *skb)
1885 if (static_branch_unlikely(&netstamp_needed_key))
1886 __net_timestamp(skb);
1889 #define net_timestamp_check(COND, SKB) \
1890 if (static_branch_unlikely(&netstamp_needed_key)) { \
1891 if ((COND) && !(SKB)->tstamp) \
1892 __net_timestamp(SKB); \
1895 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1899 if (!(dev->flags & IFF_UP))
1902 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1903 if (skb->len <= len)
1906 /* if TSO is enabled, we don't care about the length as the packet
1907 * could be forwarded without being segmented before
1909 if (skb_is_gso(skb))
1914 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1916 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1918 int ret = ____dev_forward_skb(dev, skb);
1921 skb->protocol = eth_type_trans(skb, dev);
1922 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1927 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1930 * dev_forward_skb - loopback an skb to another netif
1932 * @dev: destination network device
1933 * @skb: buffer to forward
1936 * NET_RX_SUCCESS (no congestion)
1937 * NET_RX_DROP (packet was dropped, but freed)
1939 * dev_forward_skb can be used for injecting an skb from the
1940 * start_xmit function of one device into the receive queue
1941 * of another device.
1943 * The receiving device may be in another namespace, so
1944 * we have to clear all information in the skb that could
1945 * impact namespace isolation.
1947 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1949 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1951 EXPORT_SYMBOL_GPL(dev_forward_skb);
1953 static inline int deliver_skb(struct sk_buff *skb,
1954 struct packet_type *pt_prev,
1955 struct net_device *orig_dev)
1957 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1959 refcount_inc(&skb->users);
1960 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1963 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1964 struct packet_type **pt,
1965 struct net_device *orig_dev,
1967 struct list_head *ptype_list)
1969 struct packet_type *ptype, *pt_prev = *pt;
1971 list_for_each_entry_rcu(ptype, ptype_list, list) {
1972 if (ptype->type != type)
1975 deliver_skb(skb, pt_prev, orig_dev);
1981 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1983 if (!ptype->af_packet_priv || !skb->sk)
1986 if (ptype->id_match)
1987 return ptype->id_match(ptype, skb->sk);
1988 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1995 * dev_nit_active - return true if any network interface taps are in use
1997 * @dev: network device to check for the presence of taps
1999 bool dev_nit_active(struct net_device *dev)
2001 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2003 EXPORT_SYMBOL_GPL(dev_nit_active);
2006 * Support routine. Sends outgoing frames to any network
2007 * taps currently in use.
2010 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2012 struct packet_type *ptype;
2013 struct sk_buff *skb2 = NULL;
2014 struct packet_type *pt_prev = NULL;
2015 struct list_head *ptype_list = &ptype_all;
2019 list_for_each_entry_rcu(ptype, ptype_list, list) {
2020 if (ptype->ignore_outgoing)
2023 /* Never send packets back to the socket
2024 * they originated from - MvS (miquels@drinkel.ow.org)
2026 if (skb_loop_sk(ptype, skb))
2030 deliver_skb(skb2, pt_prev, skb->dev);
2035 /* need to clone skb, done only once */
2036 skb2 = skb_clone(skb, GFP_ATOMIC);
2040 net_timestamp_set(skb2);
2042 /* skb->nh should be correctly
2043 * set by sender, so that the second statement is
2044 * just protection against buggy protocols.
2046 skb_reset_mac_header(skb2);
2048 if (skb_network_header(skb2) < skb2->data ||
2049 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2050 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2051 ntohs(skb2->protocol),
2053 skb_reset_network_header(skb2);
2056 skb2->transport_header = skb2->network_header;
2057 skb2->pkt_type = PACKET_OUTGOING;
2061 if (ptype_list == &ptype_all) {
2062 ptype_list = &dev->ptype_all;
2067 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2068 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2074 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2077 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2078 * @dev: Network device
2079 * @txq: number of queues available
2081 * If real_num_tx_queues is changed the tc mappings may no longer be
2082 * valid. To resolve this verify the tc mapping remains valid and if
2083 * not NULL the mapping. With no priorities mapping to this
2084 * offset/count pair it will no longer be used. In the worst case TC0
2085 * is invalid nothing can be done so disable priority mappings. If is
2086 * expected that drivers will fix this mapping if they can before
2087 * calling netif_set_real_num_tx_queues.
2089 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2092 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2094 /* If TC0 is invalidated disable TC mapping */
2095 if (tc->offset + tc->count > txq) {
2096 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2101 /* Invalidated prio to tc mappings set to TC0 */
2102 for (i = 1; i < TC_BITMASK + 1; i++) {
2103 int q = netdev_get_prio_tc_map(dev, i);
2105 tc = &dev->tc_to_txq[q];
2106 if (tc->offset + tc->count > txq) {
2107 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2109 netdev_set_prio_tc_map(dev, i, 0);
2114 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2117 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2120 /* walk through the TCs and see if it falls into any of them */
2121 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2122 if ((txq - tc->offset) < tc->count)
2126 /* didn't find it, just return -1 to indicate no match */
2132 EXPORT_SYMBOL(netdev_txq_to_tc);
2135 struct static_key xps_needed __read_mostly;
2136 EXPORT_SYMBOL(xps_needed);
2137 struct static_key xps_rxqs_needed __read_mostly;
2138 EXPORT_SYMBOL(xps_rxqs_needed);
2139 static DEFINE_MUTEX(xps_map_mutex);
2140 #define xmap_dereference(P) \
2141 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2143 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2146 struct xps_map *map = NULL;
2150 map = xmap_dereference(dev_maps->attr_map[tci]);
2154 for (pos = map->len; pos--;) {
2155 if (map->queues[pos] != index)
2159 map->queues[pos] = map->queues[--map->len];
2163 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2164 kfree_rcu(map, rcu);
2171 static bool remove_xps_queue_cpu(struct net_device *dev,
2172 struct xps_dev_maps *dev_maps,
2173 int cpu, u16 offset, u16 count)
2175 int num_tc = dev->num_tc ? : 1;
2176 bool active = false;
2179 for (tci = cpu * num_tc; num_tc--; tci++) {
2182 for (i = count, j = offset; i--; j++) {
2183 if (!remove_xps_queue(dev_maps, tci, j))
2193 static void reset_xps_maps(struct net_device *dev,
2194 struct xps_dev_maps *dev_maps,
2198 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2199 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2201 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2203 static_key_slow_dec_cpuslocked(&xps_needed);
2204 kfree_rcu(dev_maps, rcu);
2207 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2208 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2209 u16 offset, u16 count, bool is_rxqs_map)
2211 bool active = false;
2214 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2216 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2219 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2222 for (i = offset + (count - 1); count--; i--) {
2223 netdev_queue_numa_node_write(
2224 netdev_get_tx_queue(dev, i),
2230 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2233 const unsigned long *possible_mask = NULL;
2234 struct xps_dev_maps *dev_maps;
2235 unsigned int nr_ids;
2237 if (!static_key_false(&xps_needed))
2241 mutex_lock(&xps_map_mutex);
2243 if (static_key_false(&xps_rxqs_needed)) {
2244 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2246 nr_ids = dev->num_rx_queues;
2247 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2248 offset, count, true);
2252 dev_maps = xmap_dereference(dev->xps_cpus_map);
2256 if (num_possible_cpus() > 1)
2257 possible_mask = cpumask_bits(cpu_possible_mask);
2258 nr_ids = nr_cpu_ids;
2259 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2263 mutex_unlock(&xps_map_mutex);
2267 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2269 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2272 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2273 u16 index, bool is_rxqs_map)
2275 struct xps_map *new_map;
2276 int alloc_len = XPS_MIN_MAP_ALLOC;
2279 for (pos = 0; map && pos < map->len; pos++) {
2280 if (map->queues[pos] != index)
2285 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2287 if (pos < map->alloc_len)
2290 alloc_len = map->alloc_len * 2;
2293 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2297 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2299 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2300 cpu_to_node(attr_index));
2304 for (i = 0; i < pos; i++)
2305 new_map->queues[i] = map->queues[i];
2306 new_map->alloc_len = alloc_len;
2312 /* Must be called under cpus_read_lock */
2313 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2314 u16 index, bool is_rxqs_map)
2316 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2317 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2318 int i, j, tci, numa_node_id = -2;
2319 int maps_sz, num_tc = 1, tc = 0;
2320 struct xps_map *map, *new_map;
2321 bool active = false;
2322 unsigned int nr_ids;
2325 /* Do not allow XPS on subordinate device directly */
2326 num_tc = dev->num_tc;
2330 /* If queue belongs to subordinate dev use its map */
2331 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2333 tc = netdev_txq_to_tc(dev, index);
2338 mutex_lock(&xps_map_mutex);
2340 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2341 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2342 nr_ids = dev->num_rx_queues;
2344 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2345 if (num_possible_cpus() > 1) {
2346 online_mask = cpumask_bits(cpu_online_mask);
2347 possible_mask = cpumask_bits(cpu_possible_mask);
2349 dev_maps = xmap_dereference(dev->xps_cpus_map);
2350 nr_ids = nr_cpu_ids;
2353 if (maps_sz < L1_CACHE_BYTES)
2354 maps_sz = L1_CACHE_BYTES;
2356 /* allocate memory for queue storage */
2357 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2360 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2361 if (!new_dev_maps) {
2362 mutex_unlock(&xps_map_mutex);
2366 tci = j * num_tc + tc;
2367 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2370 map = expand_xps_map(map, j, index, is_rxqs_map);
2374 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2378 goto out_no_new_maps;
2381 /* Increment static keys at most once per type */
2382 static_key_slow_inc_cpuslocked(&xps_needed);
2384 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2387 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2389 /* copy maps belonging to foreign traffic classes */
2390 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2391 /* fill in the new device map from the old device map */
2392 map = xmap_dereference(dev_maps->attr_map[tci]);
2393 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2396 /* We need to explicitly update tci as prevous loop
2397 * could break out early if dev_maps is NULL.
2399 tci = j * num_tc + tc;
2401 if (netif_attr_test_mask(j, mask, nr_ids) &&
2402 netif_attr_test_online(j, online_mask, nr_ids)) {
2403 /* add tx-queue to CPU/rx-queue maps */
2406 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2407 while ((pos < map->len) && (map->queues[pos] != index))
2410 if (pos == map->len)
2411 map->queues[map->len++] = index;
2414 if (numa_node_id == -2)
2415 numa_node_id = cpu_to_node(j);
2416 else if (numa_node_id != cpu_to_node(j))
2420 } else if (dev_maps) {
2421 /* fill in the new device map from the old device map */
2422 map = xmap_dereference(dev_maps->attr_map[tci]);
2423 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2426 /* copy maps belonging to foreign traffic classes */
2427 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2428 /* fill in the new device map from the old device map */
2429 map = xmap_dereference(dev_maps->attr_map[tci]);
2430 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2435 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2437 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2439 /* Cleanup old maps */
2441 goto out_no_old_maps;
2443 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2445 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2446 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2447 map = xmap_dereference(dev_maps->attr_map[tci]);
2448 if (map && map != new_map)
2449 kfree_rcu(map, rcu);
2453 kfree_rcu(dev_maps, rcu);
2456 dev_maps = new_dev_maps;
2461 /* update Tx queue numa node */
2462 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2463 (numa_node_id >= 0) ?
2464 numa_node_id : NUMA_NO_NODE);
2470 /* removes tx-queue from unused CPUs/rx-queues */
2471 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2473 for (i = tc, tci = j * num_tc; i--; tci++)
2474 active |= remove_xps_queue(dev_maps, tci, index);
2475 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2476 !netif_attr_test_online(j, online_mask, nr_ids))
2477 active |= remove_xps_queue(dev_maps, tci, index);
2478 for (i = num_tc - tc, tci++; --i; tci++)
2479 active |= remove_xps_queue(dev_maps, tci, index);
2482 /* free map if not active */
2484 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2487 mutex_unlock(&xps_map_mutex);
2491 /* remove any maps that we added */
2492 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2494 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2495 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2497 xmap_dereference(dev_maps->attr_map[tci]) :
2499 if (new_map && new_map != map)
2504 mutex_unlock(&xps_map_mutex);
2506 kfree(new_dev_maps);
2509 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2511 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2517 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2522 EXPORT_SYMBOL(netif_set_xps_queue);
2525 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2527 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2529 /* Unbind any subordinate channels */
2530 while (txq-- != &dev->_tx[0]) {
2532 netdev_unbind_sb_channel(dev, txq->sb_dev);
2536 void netdev_reset_tc(struct net_device *dev)
2539 netif_reset_xps_queues_gt(dev, 0);
2541 netdev_unbind_all_sb_channels(dev);
2543 /* Reset TC configuration of device */
2545 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2546 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2548 EXPORT_SYMBOL(netdev_reset_tc);
2550 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2552 if (tc >= dev->num_tc)
2556 netif_reset_xps_queues(dev, offset, count);
2558 dev->tc_to_txq[tc].count = count;
2559 dev->tc_to_txq[tc].offset = offset;
2562 EXPORT_SYMBOL(netdev_set_tc_queue);
2564 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2566 if (num_tc > TC_MAX_QUEUE)
2570 netif_reset_xps_queues_gt(dev, 0);
2572 netdev_unbind_all_sb_channels(dev);
2574 dev->num_tc = num_tc;
2577 EXPORT_SYMBOL(netdev_set_num_tc);
2579 void netdev_unbind_sb_channel(struct net_device *dev,
2580 struct net_device *sb_dev)
2582 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2585 netif_reset_xps_queues_gt(sb_dev, 0);
2587 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2588 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2590 while (txq-- != &dev->_tx[0]) {
2591 if (txq->sb_dev == sb_dev)
2595 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2597 int netdev_bind_sb_channel_queue(struct net_device *dev,
2598 struct net_device *sb_dev,
2599 u8 tc, u16 count, u16 offset)
2601 /* Make certain the sb_dev and dev are already configured */
2602 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2605 /* We cannot hand out queues we don't have */
2606 if ((offset + count) > dev->real_num_tx_queues)
2609 /* Record the mapping */
2610 sb_dev->tc_to_txq[tc].count = count;
2611 sb_dev->tc_to_txq[tc].offset = offset;
2613 /* Provide a way for Tx queue to find the tc_to_txq map or
2614 * XPS map for itself.
2617 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2621 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2623 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2625 /* Do not use a multiqueue device to represent a subordinate channel */
2626 if (netif_is_multiqueue(dev))
2629 /* We allow channels 1 - 32767 to be used for subordinate channels.
2630 * Channel 0 is meant to be "native" mode and used only to represent
2631 * the main root device. We allow writing 0 to reset the device back
2632 * to normal mode after being used as a subordinate channel.
2634 if (channel > S16_MAX)
2637 dev->num_tc = -channel;
2641 EXPORT_SYMBOL(netdev_set_sb_channel);
2644 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2645 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2647 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2652 disabling = txq < dev->real_num_tx_queues;
2654 if (txq < 1 || txq > dev->num_tx_queues)
2657 if (dev->reg_state == NETREG_REGISTERED ||
2658 dev->reg_state == NETREG_UNREGISTERING) {
2661 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2667 netif_setup_tc(dev, txq);
2669 dev->real_num_tx_queues = txq;
2673 qdisc_reset_all_tx_gt(dev, txq);
2675 netif_reset_xps_queues_gt(dev, txq);
2679 dev->real_num_tx_queues = txq;
2684 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2688 * netif_set_real_num_rx_queues - set actual number of RX queues used
2689 * @dev: Network device
2690 * @rxq: Actual number of RX queues
2692 * This must be called either with the rtnl_lock held or before
2693 * registration of the net device. Returns 0 on success, or a
2694 * negative error code. If called before registration, it always
2697 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2701 if (rxq < 1 || rxq > dev->num_rx_queues)
2704 if (dev->reg_state == NETREG_REGISTERED) {
2707 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2713 dev->real_num_rx_queues = rxq;
2716 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2720 * netif_get_num_default_rss_queues - default number of RSS queues
2722 * This routine should set an upper limit on the number of RSS queues
2723 * used by default by multiqueue devices.
2725 int netif_get_num_default_rss_queues(void)
2727 return is_kdump_kernel() ?
2728 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2730 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2732 static void __netif_reschedule(struct Qdisc *q)
2734 struct softnet_data *sd;
2735 unsigned long flags;
2737 local_irq_save(flags);
2738 sd = this_cpu_ptr(&softnet_data);
2739 q->next_sched = NULL;
2740 *sd->output_queue_tailp = q;
2741 sd->output_queue_tailp = &q->next_sched;
2742 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2743 local_irq_restore(flags);
2746 void __netif_schedule(struct Qdisc *q)
2748 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2749 __netif_reschedule(q);
2751 EXPORT_SYMBOL(__netif_schedule);
2753 struct dev_kfree_skb_cb {
2754 enum skb_free_reason reason;
2757 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2759 return (struct dev_kfree_skb_cb *)skb->cb;
2762 void netif_schedule_queue(struct netdev_queue *txq)
2765 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2766 struct Qdisc *q = rcu_dereference(txq->qdisc);
2768 __netif_schedule(q);
2772 EXPORT_SYMBOL(netif_schedule_queue);
2774 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2776 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2780 q = rcu_dereference(dev_queue->qdisc);
2781 __netif_schedule(q);
2785 EXPORT_SYMBOL(netif_tx_wake_queue);
2787 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2789 unsigned long flags;
2794 if (likely(refcount_read(&skb->users) == 1)) {
2796 refcount_set(&skb->users, 0);
2797 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2800 get_kfree_skb_cb(skb)->reason = reason;
2801 local_irq_save(flags);
2802 skb->next = __this_cpu_read(softnet_data.completion_queue);
2803 __this_cpu_write(softnet_data.completion_queue, skb);
2804 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2805 local_irq_restore(flags);
2807 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2809 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2811 if (in_irq() || irqs_disabled())
2812 __dev_kfree_skb_irq(skb, reason);
2816 EXPORT_SYMBOL(__dev_kfree_skb_any);
2820 * netif_device_detach - mark device as removed
2821 * @dev: network device
2823 * Mark device as removed from system and therefore no longer available.
2825 void netif_device_detach(struct net_device *dev)
2827 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2828 netif_running(dev)) {
2829 netif_tx_stop_all_queues(dev);
2832 EXPORT_SYMBOL(netif_device_detach);
2835 * netif_device_attach - mark device as attached
2836 * @dev: network device
2838 * Mark device as attached from system and restart if needed.
2840 void netif_device_attach(struct net_device *dev)
2842 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2843 netif_running(dev)) {
2844 netif_tx_wake_all_queues(dev);
2845 __netdev_watchdog_up(dev);
2848 EXPORT_SYMBOL(netif_device_attach);
2851 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2852 * to be used as a distribution range.
2854 static u16 skb_tx_hash(const struct net_device *dev,
2855 const struct net_device *sb_dev,
2856 struct sk_buff *skb)
2860 u16 qcount = dev->real_num_tx_queues;
2863 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2865 qoffset = sb_dev->tc_to_txq[tc].offset;
2866 qcount = sb_dev->tc_to_txq[tc].count;
2869 if (skb_rx_queue_recorded(skb)) {
2870 hash = skb_get_rx_queue(skb);
2871 while (unlikely(hash >= qcount))
2873 return hash + qoffset;
2876 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2879 static void skb_warn_bad_offload(const struct sk_buff *skb)
2881 static const netdev_features_t null_features;
2882 struct net_device *dev = skb->dev;
2883 const char *name = "";
2885 if (!net_ratelimit())
2889 if (dev->dev.parent)
2890 name = dev_driver_string(dev->dev.parent);
2892 name = netdev_name(dev);
2894 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2895 "gso_type=%d ip_summed=%d\n",
2896 name, dev ? &dev->features : &null_features,
2897 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2898 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2899 skb_shinfo(skb)->gso_type, skb->ip_summed);
2903 * Invalidate hardware checksum when packet is to be mangled, and
2904 * complete checksum manually on outgoing path.
2906 int skb_checksum_help(struct sk_buff *skb)
2909 int ret = 0, offset;
2911 if (skb->ip_summed == CHECKSUM_COMPLETE)
2912 goto out_set_summed;
2914 if (unlikely(skb_shinfo(skb)->gso_size)) {
2915 skb_warn_bad_offload(skb);
2919 /* Before computing a checksum, we should make sure no frag could
2920 * be modified by an external entity : checksum could be wrong.
2922 if (skb_has_shared_frag(skb)) {
2923 ret = __skb_linearize(skb);
2928 offset = skb_checksum_start_offset(skb);
2929 BUG_ON(offset >= skb_headlen(skb));
2930 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2932 offset += skb->csum_offset;
2933 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2935 if (skb_cloned(skb) &&
2936 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2937 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2942 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2944 skb->ip_summed = CHECKSUM_NONE;
2948 EXPORT_SYMBOL(skb_checksum_help);
2950 int skb_crc32c_csum_help(struct sk_buff *skb)
2953 int ret = 0, offset, start;
2955 if (skb->ip_summed != CHECKSUM_PARTIAL)
2958 if (unlikely(skb_is_gso(skb)))
2961 /* Before computing a checksum, we should make sure no frag could
2962 * be modified by an external entity : checksum could be wrong.
2964 if (unlikely(skb_has_shared_frag(skb))) {
2965 ret = __skb_linearize(skb);
2969 start = skb_checksum_start_offset(skb);
2970 offset = start + offsetof(struct sctphdr, checksum);
2971 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2975 if (skb_cloned(skb) &&
2976 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2977 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2981 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2982 skb->len - start, ~(__u32)0,
2984 *(__le32 *)(skb->data + offset) = crc32c_csum;
2985 skb->ip_summed = CHECKSUM_NONE;
2986 skb->csum_not_inet = 0;
2991 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2993 __be16 type = skb->protocol;
2995 /* Tunnel gso handlers can set protocol to ethernet. */
2996 if (type == htons(ETH_P_TEB)) {
2999 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3002 eth = (struct ethhdr *)skb->data;
3003 type = eth->h_proto;
3006 return __vlan_get_protocol(skb, type, depth);
3010 * skb_mac_gso_segment - mac layer segmentation handler.
3011 * @skb: buffer to segment
3012 * @features: features for the output path (see dev->features)
3014 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3015 netdev_features_t features)
3017 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3018 struct packet_offload *ptype;
3019 int vlan_depth = skb->mac_len;
3020 __be16 type = skb_network_protocol(skb, &vlan_depth);
3022 if (unlikely(!type))
3023 return ERR_PTR(-EINVAL);
3025 __skb_pull(skb, vlan_depth);
3028 list_for_each_entry_rcu(ptype, &offload_base, list) {
3029 if (ptype->type == type && ptype->callbacks.gso_segment) {
3030 segs = ptype->callbacks.gso_segment(skb, features);
3036 __skb_push(skb, skb->data - skb_mac_header(skb));
3040 EXPORT_SYMBOL(skb_mac_gso_segment);
3043 /* openvswitch calls this on rx path, so we need a different check.
3045 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3048 return skb->ip_summed != CHECKSUM_PARTIAL &&
3049 skb->ip_summed != CHECKSUM_UNNECESSARY;
3051 return skb->ip_summed == CHECKSUM_NONE;
3055 * __skb_gso_segment - Perform segmentation on skb.
3056 * @skb: buffer to segment
3057 * @features: features for the output path (see dev->features)
3058 * @tx_path: whether it is called in TX path
3060 * This function segments the given skb and returns a list of segments.
3062 * It may return NULL if the skb requires no segmentation. This is
3063 * only possible when GSO is used for verifying header integrity.
3065 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3067 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3068 netdev_features_t features, bool tx_path)
3070 struct sk_buff *segs;
3072 if (unlikely(skb_needs_check(skb, tx_path))) {
3075 /* We're going to init ->check field in TCP or UDP header */
3076 err = skb_cow_head(skb, 0);
3078 return ERR_PTR(err);
3081 /* Only report GSO partial support if it will enable us to
3082 * support segmentation on this frame without needing additional
3085 if (features & NETIF_F_GSO_PARTIAL) {
3086 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3087 struct net_device *dev = skb->dev;
3089 partial_features |= dev->features & dev->gso_partial_features;
3090 if (!skb_gso_ok(skb, features | partial_features))
3091 features &= ~NETIF_F_GSO_PARTIAL;
3094 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3095 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3097 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3098 SKB_GSO_CB(skb)->encap_level = 0;
3100 skb_reset_mac_header(skb);
3101 skb_reset_mac_len(skb);
3103 segs = skb_mac_gso_segment(skb, features);
3105 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3106 skb_warn_bad_offload(skb);
3110 EXPORT_SYMBOL(__skb_gso_segment);
3112 /* Take action when hardware reception checksum errors are detected. */
3114 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3116 if (net_ratelimit()) {
3117 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3119 pr_err("dev features: %pNF\n", &dev->features);
3120 pr_err("skb len=%u data_len=%u pkt_type=%u gso_size=%u gso_type=%u nr_frags=%u ip_summed=%u csum=%x csum_complete_sw=%d csum_valid=%d csum_level=%u\n",
3121 skb->len, skb->data_len, skb->pkt_type,
3122 skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type,
3123 skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum,
3124 skb->csum_complete_sw, skb->csum_valid, skb->csum_level);
3128 EXPORT_SYMBOL(netdev_rx_csum_fault);
3131 /* XXX: check that highmem exists at all on the given machine. */
3132 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3134 #ifdef CONFIG_HIGHMEM
3137 if (!(dev->features & NETIF_F_HIGHDMA)) {
3138 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3139 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3141 if (PageHighMem(skb_frag_page(frag)))
3149 /* If MPLS offload request, verify we are testing hardware MPLS features
3150 * instead of standard features for the netdev.
3152 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3153 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3154 netdev_features_t features,
3157 if (eth_p_mpls(type))
3158 features &= skb->dev->mpls_features;
3163 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3164 netdev_features_t features,
3171 static netdev_features_t harmonize_features(struct sk_buff *skb,
3172 netdev_features_t features)
3177 type = skb_network_protocol(skb, &tmp);
3178 features = net_mpls_features(skb, features, type);
3180 if (skb->ip_summed != CHECKSUM_NONE &&
3181 !can_checksum_protocol(features, type)) {
3182 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3184 if (illegal_highdma(skb->dev, skb))
3185 features &= ~NETIF_F_SG;
3190 netdev_features_t passthru_features_check(struct sk_buff *skb,
3191 struct net_device *dev,
3192 netdev_features_t features)
3196 EXPORT_SYMBOL(passthru_features_check);
3198 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3199 struct net_device *dev,
3200 netdev_features_t features)
3202 return vlan_features_check(skb, features);
3205 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3206 struct net_device *dev,
3207 netdev_features_t features)
3209 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3211 if (gso_segs > dev->gso_max_segs)
3212 return features & ~NETIF_F_GSO_MASK;
3214 /* Support for GSO partial features requires software
3215 * intervention before we can actually process the packets
3216 * so we need to strip support for any partial features now
3217 * and we can pull them back in after we have partially
3218 * segmented the frame.
3220 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3221 features &= ~dev->gso_partial_features;
3223 /* Make sure to clear the IPv4 ID mangling feature if the
3224 * IPv4 header has the potential to be fragmented.
3226 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3227 struct iphdr *iph = skb->encapsulation ?
3228 inner_ip_hdr(skb) : ip_hdr(skb);
3230 if (!(iph->frag_off & htons(IP_DF)))
3231 features &= ~NETIF_F_TSO_MANGLEID;
3237 netdev_features_t netif_skb_features(struct sk_buff *skb)
3239 struct net_device *dev = skb->dev;
3240 netdev_features_t features = dev->features;
3242 if (skb_is_gso(skb))
3243 features = gso_features_check(skb, dev, features);
3245 /* If encapsulation offload request, verify we are testing
3246 * hardware encapsulation features instead of standard
3247 * features for the netdev
3249 if (skb->encapsulation)
3250 features &= dev->hw_enc_features;
3252 if (skb_vlan_tagged(skb))
3253 features = netdev_intersect_features(features,
3254 dev->vlan_features |
3255 NETIF_F_HW_VLAN_CTAG_TX |
3256 NETIF_F_HW_VLAN_STAG_TX);
3258 if (dev->netdev_ops->ndo_features_check)
3259 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3262 features &= dflt_features_check(skb, dev, features);
3264 return harmonize_features(skb, features);
3266 EXPORT_SYMBOL(netif_skb_features);
3268 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3269 struct netdev_queue *txq, bool more)
3274 if (dev_nit_active(dev))
3275 dev_queue_xmit_nit(skb, dev);
3278 trace_net_dev_start_xmit(skb, dev);
3279 rc = netdev_start_xmit(skb, dev, txq, more);
3280 trace_net_dev_xmit(skb, rc, dev, len);
3285 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3286 struct netdev_queue *txq, int *ret)
3288 struct sk_buff *skb = first;
3289 int rc = NETDEV_TX_OK;
3292 struct sk_buff *next = skb->next;
3294 skb_mark_not_on_list(skb);
3295 rc = xmit_one(skb, dev, txq, next != NULL);
3296 if (unlikely(!dev_xmit_complete(rc))) {
3302 if (netif_tx_queue_stopped(txq) && skb) {
3303 rc = NETDEV_TX_BUSY;
3313 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3314 netdev_features_t features)
3316 if (skb_vlan_tag_present(skb) &&
3317 !vlan_hw_offload_capable(features, skb->vlan_proto))
3318 skb = __vlan_hwaccel_push_inside(skb);
3322 int skb_csum_hwoffload_help(struct sk_buff *skb,
3323 const netdev_features_t features)
3325 if (unlikely(skb->csum_not_inet))
3326 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3327 skb_crc32c_csum_help(skb);
3329 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3331 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3333 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3335 netdev_features_t features;
3337 features = netif_skb_features(skb);
3338 skb = validate_xmit_vlan(skb, features);
3342 skb = sk_validate_xmit_skb(skb, dev);
3346 if (netif_needs_gso(skb, features)) {
3347 struct sk_buff *segs;
3349 segs = skb_gso_segment(skb, features);
3357 if (skb_needs_linearize(skb, features) &&
3358 __skb_linearize(skb))
3361 /* If packet is not checksummed and device does not
3362 * support checksumming for this protocol, complete
3363 * checksumming here.
3365 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3366 if (skb->encapsulation)
3367 skb_set_inner_transport_header(skb,
3368 skb_checksum_start_offset(skb));
3370 skb_set_transport_header(skb,
3371 skb_checksum_start_offset(skb));
3372 if (skb_csum_hwoffload_help(skb, features))
3377 skb = validate_xmit_xfrm(skb, features, again);
3384 atomic_long_inc(&dev->tx_dropped);
3388 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3390 struct sk_buff *next, *head = NULL, *tail;
3392 for (; skb != NULL; skb = next) {
3394 skb_mark_not_on_list(skb);
3396 /* in case skb wont be segmented, point to itself */
3399 skb = validate_xmit_skb(skb, dev, again);
3407 /* If skb was segmented, skb->prev points to
3408 * the last segment. If not, it still contains skb.
3414 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3416 static void qdisc_pkt_len_init(struct sk_buff *skb)
3418 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3420 qdisc_skb_cb(skb)->pkt_len = skb->len;
3422 /* To get more precise estimation of bytes sent on wire,
3423 * we add to pkt_len the headers size of all segments
3425 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3426 unsigned int hdr_len;
3427 u16 gso_segs = shinfo->gso_segs;
3429 /* mac layer + network layer */
3430 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3432 /* + transport layer */
3433 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3434 const struct tcphdr *th;
3435 struct tcphdr _tcphdr;
3437 th = skb_header_pointer(skb, skb_transport_offset(skb),
3438 sizeof(_tcphdr), &_tcphdr);
3440 hdr_len += __tcp_hdrlen(th);
3442 struct udphdr _udphdr;
3444 if (skb_header_pointer(skb, skb_transport_offset(skb),
3445 sizeof(_udphdr), &_udphdr))
3446 hdr_len += sizeof(struct udphdr);
3449 if (shinfo->gso_type & SKB_GSO_DODGY)
3450 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3453 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3457 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3458 struct net_device *dev,
3459 struct netdev_queue *txq)
3461 spinlock_t *root_lock = qdisc_lock(q);
3462 struct sk_buff *to_free = NULL;
3466 qdisc_calculate_pkt_len(skb, q);
3468 if (q->flags & TCQ_F_NOLOCK) {
3469 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3470 __qdisc_drop(skb, &to_free);
3472 } else if ((q->flags & TCQ_F_CAN_BYPASS) && q->empty &&
3473 qdisc_run_begin(q)) {
3474 qdisc_bstats_cpu_update(q, skb);
3476 if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
3480 rc = NET_XMIT_SUCCESS;
3482 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3486 if (unlikely(to_free))
3487 kfree_skb_list(to_free);
3492 * Heuristic to force contended enqueues to serialize on a
3493 * separate lock before trying to get qdisc main lock.
3494 * This permits qdisc->running owner to get the lock more
3495 * often and dequeue packets faster.
3497 contended = qdisc_is_running(q);
3498 if (unlikely(contended))
3499 spin_lock(&q->busylock);
3501 spin_lock(root_lock);
3502 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3503 __qdisc_drop(skb, &to_free);
3505 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3506 qdisc_run_begin(q)) {
3508 * This is a work-conserving queue; there are no old skbs
3509 * waiting to be sent out; and the qdisc is not running -
3510 * xmit the skb directly.
3513 qdisc_bstats_update(q, skb);
3515 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3516 if (unlikely(contended)) {
3517 spin_unlock(&q->busylock);
3524 rc = NET_XMIT_SUCCESS;
3526 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3527 if (qdisc_run_begin(q)) {
3528 if (unlikely(contended)) {
3529 spin_unlock(&q->busylock);
3536 spin_unlock(root_lock);
3537 if (unlikely(to_free))
3538 kfree_skb_list(to_free);
3539 if (unlikely(contended))
3540 spin_unlock(&q->busylock);
3544 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3545 static void skb_update_prio(struct sk_buff *skb)
3547 const struct netprio_map *map;
3548 const struct sock *sk;
3549 unsigned int prioidx;
3553 map = rcu_dereference_bh(skb->dev->priomap);
3556 sk = skb_to_full_sk(skb);
3560 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3562 if (prioidx < map->priomap_len)
3563 skb->priority = map->priomap[prioidx];
3566 #define skb_update_prio(skb)
3569 DEFINE_PER_CPU(int, xmit_recursion);
3570 EXPORT_SYMBOL(xmit_recursion);
3573 * dev_loopback_xmit - loop back @skb
3574 * @net: network namespace this loopback is happening in
3575 * @sk: sk needed to be a netfilter okfn
3576 * @skb: buffer to transmit
3578 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3580 skb_reset_mac_header(skb);
3581 __skb_pull(skb, skb_network_offset(skb));
3582 skb->pkt_type = PACKET_LOOPBACK;
3583 skb->ip_summed = CHECKSUM_UNNECESSARY;
3584 WARN_ON(!skb_dst(skb));
3589 EXPORT_SYMBOL(dev_loopback_xmit);
3591 #ifdef CONFIG_NET_EGRESS
3592 static struct sk_buff *
3593 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3595 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3596 struct tcf_result cl_res;
3601 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3602 mini_qdisc_bstats_cpu_update(miniq, skb);
3604 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3606 case TC_ACT_RECLASSIFY:
3607 skb->tc_index = TC_H_MIN(cl_res.classid);
3610 mini_qdisc_qstats_cpu_drop(miniq);
3611 *ret = NET_XMIT_DROP;
3617 *ret = NET_XMIT_SUCCESS;
3620 case TC_ACT_REDIRECT:
3621 /* No need to push/pop skb's mac_header here on egress! */
3622 skb_do_redirect(skb);
3623 *ret = NET_XMIT_SUCCESS;
3631 #endif /* CONFIG_NET_EGRESS */
3634 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3635 struct xps_dev_maps *dev_maps, unsigned int tci)
3637 struct xps_map *map;
3638 int queue_index = -1;
3642 tci += netdev_get_prio_tc_map(dev, skb->priority);
3645 map = rcu_dereference(dev_maps->attr_map[tci]);
3648 queue_index = map->queues[0];
3650 queue_index = map->queues[reciprocal_scale(
3651 skb_get_hash(skb), map->len)];
3652 if (unlikely(queue_index >= dev->real_num_tx_queues))
3659 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3660 struct sk_buff *skb)
3663 struct xps_dev_maps *dev_maps;
3664 struct sock *sk = skb->sk;
3665 int queue_index = -1;
3667 if (!static_key_false(&xps_needed))
3671 if (!static_key_false(&xps_rxqs_needed))
3674 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3676 int tci = sk_rx_queue_get(sk);
3678 if (tci >= 0 && tci < dev->num_rx_queues)
3679 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3684 if (queue_index < 0) {
3685 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3687 unsigned int tci = skb->sender_cpu - 1;
3689 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3701 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3702 struct net_device *sb_dev)
3706 EXPORT_SYMBOL(dev_pick_tx_zero);
3708 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3709 struct net_device *sb_dev)
3711 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3713 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3715 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3716 struct net_device *sb_dev)
3718 struct sock *sk = skb->sk;
3719 int queue_index = sk_tx_queue_get(sk);
3721 sb_dev = sb_dev ? : dev;
3723 if (queue_index < 0 || skb->ooo_okay ||
3724 queue_index >= dev->real_num_tx_queues) {
3725 int new_index = get_xps_queue(dev, sb_dev, skb);
3728 new_index = skb_tx_hash(dev, sb_dev, skb);
3730 if (queue_index != new_index && sk &&
3732 rcu_access_pointer(sk->sk_dst_cache))
3733 sk_tx_queue_set(sk, new_index);
3735 queue_index = new_index;
3740 EXPORT_SYMBOL(netdev_pick_tx);
3742 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3743 struct sk_buff *skb,
3744 struct net_device *sb_dev)
3746 int queue_index = 0;
3749 u32 sender_cpu = skb->sender_cpu - 1;
3751 if (sender_cpu >= (u32)NR_CPUS)
3752 skb->sender_cpu = raw_smp_processor_id() + 1;
3755 if (dev->real_num_tx_queues != 1) {
3756 const struct net_device_ops *ops = dev->netdev_ops;
3758 if (ops->ndo_select_queue)
3759 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3761 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3763 queue_index = netdev_cap_txqueue(dev, queue_index);
3766 skb_set_queue_mapping(skb, queue_index);
3767 return netdev_get_tx_queue(dev, queue_index);
3771 * __dev_queue_xmit - transmit a buffer
3772 * @skb: buffer to transmit
3773 * @sb_dev: suboordinate device used for L2 forwarding offload
3775 * Queue a buffer for transmission to a network device. The caller must
3776 * have set the device and priority and built the buffer before calling
3777 * this function. The function can be called from an interrupt.
3779 * A negative errno code is returned on a failure. A success does not
3780 * guarantee the frame will be transmitted as it may be dropped due
3781 * to congestion or traffic shaping.
3783 * -----------------------------------------------------------------------------------
3784 * I notice this method can also return errors from the queue disciplines,
3785 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3788 * Regardless of the return value, the skb is consumed, so it is currently
3789 * difficult to retry a send to this method. (You can bump the ref count
3790 * before sending to hold a reference for retry if you are careful.)
3792 * When calling this method, interrupts MUST be enabled. This is because
3793 * the BH enable code must have IRQs enabled so that it will not deadlock.
3796 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3798 struct net_device *dev = skb->dev;
3799 struct netdev_queue *txq;
3804 skb_reset_mac_header(skb);
3806 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3807 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3809 /* Disable soft irqs for various locks below. Also
3810 * stops preemption for RCU.
3814 skb_update_prio(skb);
3816 qdisc_pkt_len_init(skb);
3817 #ifdef CONFIG_NET_CLS_ACT
3818 skb->tc_at_ingress = 0;
3819 # ifdef CONFIG_NET_EGRESS
3820 if (static_branch_unlikely(&egress_needed_key)) {
3821 skb = sch_handle_egress(skb, &rc, dev);
3827 /* If device/qdisc don't need skb->dst, release it right now while
3828 * its hot in this cpu cache.
3830 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3835 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3836 q = rcu_dereference_bh(txq->qdisc);
3838 trace_net_dev_queue(skb);
3840 rc = __dev_xmit_skb(skb, q, dev, txq);
3844 /* The device has no queue. Common case for software devices:
3845 * loopback, all the sorts of tunnels...
3847 * Really, it is unlikely that netif_tx_lock protection is necessary
3848 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3850 * However, it is possible, that they rely on protection
3853 * Check this and shot the lock. It is not prone from deadlocks.
3854 *Either shot noqueue qdisc, it is even simpler 8)
3856 if (dev->flags & IFF_UP) {
3857 int cpu = smp_processor_id(); /* ok because BHs are off */
3859 if (txq->xmit_lock_owner != cpu) {
3860 if (unlikely(__this_cpu_read(xmit_recursion) >
3861 XMIT_RECURSION_LIMIT))
3862 goto recursion_alert;
3864 skb = validate_xmit_skb(skb, dev, &again);
3868 HARD_TX_LOCK(dev, txq, cpu);
3870 if (!netif_xmit_stopped(txq)) {
3871 __this_cpu_inc(xmit_recursion);
3872 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3873 __this_cpu_dec(xmit_recursion);
3874 if (dev_xmit_complete(rc)) {
3875 HARD_TX_UNLOCK(dev, txq);
3879 HARD_TX_UNLOCK(dev, txq);
3880 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3883 /* Recursion is detected! It is possible,
3887 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3893 rcu_read_unlock_bh();
3895 atomic_long_inc(&dev->tx_dropped);
3896 kfree_skb_list(skb);
3899 rcu_read_unlock_bh();
3903 int dev_queue_xmit(struct sk_buff *skb)
3905 return __dev_queue_xmit(skb, NULL);
3907 EXPORT_SYMBOL(dev_queue_xmit);
3909 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3911 return __dev_queue_xmit(skb, sb_dev);
3913 EXPORT_SYMBOL(dev_queue_xmit_accel);
3915 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3917 struct net_device *dev = skb->dev;
3918 struct sk_buff *orig_skb = skb;
3919 struct netdev_queue *txq;
3920 int ret = NETDEV_TX_BUSY;
3923 if (unlikely(!netif_running(dev) ||
3924 !netif_carrier_ok(dev)))
3927 skb = validate_xmit_skb_list(skb, dev, &again);
3928 if (skb != orig_skb)
3931 skb_set_queue_mapping(skb, queue_id);
3932 txq = skb_get_tx_queue(dev, skb);
3936 HARD_TX_LOCK(dev, txq, smp_processor_id());
3937 if (!netif_xmit_frozen_or_drv_stopped(txq))
3938 ret = netdev_start_xmit(skb, dev, txq, false);
3939 HARD_TX_UNLOCK(dev, txq);
3943 if (!dev_xmit_complete(ret))
3948 atomic_long_inc(&dev->tx_dropped);
3949 kfree_skb_list(skb);
3950 return NET_XMIT_DROP;
3952 EXPORT_SYMBOL(dev_direct_xmit);
3954 /*************************************************************************
3956 *************************************************************************/
3958 int netdev_max_backlog __read_mostly = 1000;
3959 EXPORT_SYMBOL(netdev_max_backlog);
3961 int netdev_tstamp_prequeue __read_mostly = 1;
3962 int netdev_budget __read_mostly = 300;
3963 unsigned int __read_mostly netdev_budget_usecs = 2000;
3964 int weight_p __read_mostly = 64; /* old backlog weight */
3965 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3966 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3967 int dev_rx_weight __read_mostly = 64;
3968 int dev_tx_weight __read_mostly = 64;
3970 /* Called with irq disabled */
3971 static inline void ____napi_schedule(struct softnet_data *sd,
3972 struct napi_struct *napi)
3974 list_add_tail(&napi->poll_list, &sd->poll_list);
3975 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3980 /* One global table that all flow-based protocols share. */
3981 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3982 EXPORT_SYMBOL(rps_sock_flow_table);
3983 u32 rps_cpu_mask __read_mostly;
3984 EXPORT_SYMBOL(rps_cpu_mask);
3986 struct static_key_false rps_needed __read_mostly;
3987 EXPORT_SYMBOL(rps_needed);
3988 struct static_key_false rfs_needed __read_mostly;
3989 EXPORT_SYMBOL(rfs_needed);
3991 static struct rps_dev_flow *
3992 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3993 struct rps_dev_flow *rflow, u16 next_cpu)
3995 if (next_cpu < nr_cpu_ids) {
3996 #ifdef CONFIG_RFS_ACCEL
3997 struct netdev_rx_queue *rxqueue;
3998 struct rps_dev_flow_table *flow_table;
3999 struct rps_dev_flow *old_rflow;
4004 /* Should we steer this flow to a different hardware queue? */
4005 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4006 !(dev->features & NETIF_F_NTUPLE))
4008 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4009 if (rxq_index == skb_get_rx_queue(skb))
4012 rxqueue = dev->_rx + rxq_index;
4013 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4016 flow_id = skb_get_hash(skb) & flow_table->mask;
4017 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4018 rxq_index, flow_id);
4022 rflow = &flow_table->flows[flow_id];
4024 if (old_rflow->filter == rflow->filter)
4025 old_rflow->filter = RPS_NO_FILTER;
4029 per_cpu(softnet_data, next_cpu).input_queue_head;
4032 rflow->cpu = next_cpu;
4037 * get_rps_cpu is called from netif_receive_skb and returns the target
4038 * CPU from the RPS map of the receiving queue for a given skb.
4039 * rcu_read_lock must be held on entry.
4041 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4042 struct rps_dev_flow **rflowp)
4044 const struct rps_sock_flow_table *sock_flow_table;
4045 struct netdev_rx_queue *rxqueue = dev->_rx;
4046 struct rps_dev_flow_table *flow_table;
4047 struct rps_map *map;
4052 if (skb_rx_queue_recorded(skb)) {
4053 u16 index = skb_get_rx_queue(skb);
4055 if (unlikely(index >= dev->real_num_rx_queues)) {
4056 WARN_ONCE(dev->real_num_rx_queues > 1,
4057 "%s received packet on queue %u, but number "
4058 "of RX queues is %u\n",
4059 dev->name, index, dev->real_num_rx_queues);
4065 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4067 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4068 map = rcu_dereference(rxqueue->rps_map);
4069 if (!flow_table && !map)
4072 skb_reset_network_header(skb);
4073 hash = skb_get_hash(skb);
4077 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4078 if (flow_table && sock_flow_table) {
4079 struct rps_dev_flow *rflow;
4083 /* First check into global flow table if there is a match */
4084 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4085 if ((ident ^ hash) & ~rps_cpu_mask)
4088 next_cpu = ident & rps_cpu_mask;
4090 /* OK, now we know there is a match,
4091 * we can look at the local (per receive queue) flow table
4093 rflow = &flow_table->flows[hash & flow_table->mask];
4097 * If the desired CPU (where last recvmsg was done) is
4098 * different from current CPU (one in the rx-queue flow
4099 * table entry), switch if one of the following holds:
4100 * - Current CPU is unset (>= nr_cpu_ids).
4101 * - Current CPU is offline.
4102 * - The current CPU's queue tail has advanced beyond the
4103 * last packet that was enqueued using this table entry.
4104 * This guarantees that all previous packets for the flow
4105 * have been dequeued, thus preserving in order delivery.
4107 if (unlikely(tcpu != next_cpu) &&
4108 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4109 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4110 rflow->last_qtail)) >= 0)) {
4112 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4115 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4125 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4126 if (cpu_online(tcpu)) {
4136 #ifdef CONFIG_RFS_ACCEL
4139 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4140 * @dev: Device on which the filter was set
4141 * @rxq_index: RX queue index
4142 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4143 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4145 * Drivers that implement ndo_rx_flow_steer() should periodically call
4146 * this function for each installed filter and remove the filters for
4147 * which it returns %true.
4149 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4150 u32 flow_id, u16 filter_id)
4152 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4153 struct rps_dev_flow_table *flow_table;
4154 struct rps_dev_flow *rflow;
4159 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4160 if (flow_table && flow_id <= flow_table->mask) {
4161 rflow = &flow_table->flows[flow_id];
4162 cpu = READ_ONCE(rflow->cpu);
4163 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4164 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4165 rflow->last_qtail) <
4166 (int)(10 * flow_table->mask)))
4172 EXPORT_SYMBOL(rps_may_expire_flow);
4174 #endif /* CONFIG_RFS_ACCEL */
4176 /* Called from hardirq (IPI) context */
4177 static void rps_trigger_softirq(void *data)
4179 struct softnet_data *sd = data;
4181 ____napi_schedule(sd, &sd->backlog);
4185 #endif /* CONFIG_RPS */
4188 * Check if this softnet_data structure is another cpu one
4189 * If yes, queue it to our IPI list and return 1
4192 static int rps_ipi_queued(struct softnet_data *sd)
4195 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4198 sd->rps_ipi_next = mysd->rps_ipi_list;
4199 mysd->rps_ipi_list = sd;
4201 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4204 #endif /* CONFIG_RPS */
4208 #ifdef CONFIG_NET_FLOW_LIMIT
4209 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4212 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4214 #ifdef CONFIG_NET_FLOW_LIMIT
4215 struct sd_flow_limit *fl;
4216 struct softnet_data *sd;
4217 unsigned int old_flow, new_flow;
4219 if (qlen < (netdev_max_backlog >> 1))
4222 sd = this_cpu_ptr(&softnet_data);
4225 fl = rcu_dereference(sd->flow_limit);
4227 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4228 old_flow = fl->history[fl->history_head];
4229 fl->history[fl->history_head] = new_flow;
4232 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4234 if (likely(fl->buckets[old_flow]))
4235 fl->buckets[old_flow]--;
4237 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4249 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4250 * queue (may be a remote CPU queue).
4252 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4253 unsigned int *qtail)
4255 struct softnet_data *sd;
4256 unsigned long flags;
4259 sd = &per_cpu(softnet_data, cpu);
4261 local_irq_save(flags);
4264 if (!netif_running(skb->dev))
4266 qlen = skb_queue_len(&sd->input_pkt_queue);
4267 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4270 __skb_queue_tail(&sd->input_pkt_queue, skb);
4271 input_queue_tail_incr_save(sd, qtail);
4273 local_irq_restore(flags);
4274 return NET_RX_SUCCESS;
4277 /* Schedule NAPI for backlog device
4278 * We can use non atomic operation since we own the queue lock
4280 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4281 if (!rps_ipi_queued(sd))
4282 ____napi_schedule(sd, &sd->backlog);
4291 local_irq_restore(flags);
4293 atomic_long_inc(&skb->dev->rx_dropped);
4298 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4300 struct net_device *dev = skb->dev;
4301 struct netdev_rx_queue *rxqueue;
4305 if (skb_rx_queue_recorded(skb)) {
4306 u16 index = skb_get_rx_queue(skb);
4308 if (unlikely(index >= dev->real_num_rx_queues)) {
4309 WARN_ONCE(dev->real_num_rx_queues > 1,
4310 "%s received packet on queue %u, but number "
4311 "of RX queues is %u\n",
4312 dev->name, index, dev->real_num_rx_queues);
4314 return rxqueue; /* Return first rxqueue */
4321 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4322 struct xdp_buff *xdp,
4323 struct bpf_prog *xdp_prog)
4325 struct netdev_rx_queue *rxqueue;
4326 void *orig_data, *orig_data_end;
4327 u32 metalen, act = XDP_DROP;
4328 __be16 orig_eth_type;
4334 /* Reinjected packets coming from act_mirred or similar should
4335 * not get XDP generic processing.
4337 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4340 /* XDP packets must be linear and must have sufficient headroom
4341 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4342 * native XDP provides, thus we need to do it here as well.
4344 if (skb_is_nonlinear(skb) ||
4345 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4346 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4347 int troom = skb->tail + skb->data_len - skb->end;
4349 /* In case we have to go down the path and also linearize,
4350 * then lets do the pskb_expand_head() work just once here.
4352 if (pskb_expand_head(skb,
4353 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4354 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4356 if (skb_linearize(skb))
4360 /* The XDP program wants to see the packet starting at the MAC
4363 mac_len = skb->data - skb_mac_header(skb);
4364 hlen = skb_headlen(skb) + mac_len;
4365 xdp->data = skb->data - mac_len;
4366 xdp->data_meta = xdp->data;
4367 xdp->data_end = xdp->data + hlen;
4368 xdp->data_hard_start = skb->data - skb_headroom(skb);
4369 orig_data_end = xdp->data_end;
4370 orig_data = xdp->data;
4371 eth = (struct ethhdr *)xdp->data;
4372 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4373 orig_eth_type = eth->h_proto;
4375 rxqueue = netif_get_rxqueue(skb);
4376 xdp->rxq = &rxqueue->xdp_rxq;
4378 act = bpf_prog_run_xdp(xdp_prog, xdp);
4380 off = xdp->data - orig_data;
4382 __skb_pull(skb, off);
4384 __skb_push(skb, -off);
4385 skb->mac_header += off;
4387 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4390 off = orig_data_end - xdp->data_end;
4392 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4397 /* check if XDP changed eth hdr such SKB needs update */
4398 eth = (struct ethhdr *)xdp->data;
4399 if ((orig_eth_type != eth->h_proto) ||
4400 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4401 __skb_push(skb, ETH_HLEN);
4402 skb->protocol = eth_type_trans(skb, skb->dev);
4408 __skb_push(skb, mac_len);
4411 metalen = xdp->data - xdp->data_meta;
4413 skb_metadata_set(skb, metalen);
4416 bpf_warn_invalid_xdp_action(act);
4419 trace_xdp_exception(skb->dev, xdp_prog, act);
4430 /* When doing generic XDP we have to bypass the qdisc layer and the
4431 * network taps in order to match in-driver-XDP behavior.
4433 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4435 struct net_device *dev = skb->dev;
4436 struct netdev_queue *txq;
4437 bool free_skb = true;
4440 txq = netdev_core_pick_tx(dev, skb, NULL);
4441 cpu = smp_processor_id();
4442 HARD_TX_LOCK(dev, txq, cpu);
4443 if (!netif_xmit_stopped(txq)) {
4444 rc = netdev_start_xmit(skb, dev, txq, 0);
4445 if (dev_xmit_complete(rc))
4448 HARD_TX_UNLOCK(dev, txq);
4450 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4454 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4456 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4458 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4461 struct xdp_buff xdp;
4465 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4466 if (act != XDP_PASS) {
4469 err = xdp_do_generic_redirect(skb->dev, skb,
4475 generic_xdp_tx(skb, xdp_prog);
4486 EXPORT_SYMBOL_GPL(do_xdp_generic);
4488 static int netif_rx_internal(struct sk_buff *skb)
4492 net_timestamp_check(netdev_tstamp_prequeue, skb);
4494 trace_netif_rx(skb);
4496 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4501 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4505 /* Consider XDP consuming the packet a success from
4506 * the netdev point of view we do not want to count
4509 if (ret != XDP_PASS)
4510 return NET_RX_SUCCESS;
4514 if (static_branch_unlikely(&rps_needed)) {
4515 struct rps_dev_flow voidflow, *rflow = &voidflow;
4521 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4523 cpu = smp_processor_id();
4525 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4534 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4541 * netif_rx - post buffer to the network code
4542 * @skb: buffer to post
4544 * This function receives a packet from a device driver and queues it for
4545 * the upper (protocol) levels to process. It always succeeds. The buffer
4546 * may be dropped during processing for congestion control or by the
4550 * NET_RX_SUCCESS (no congestion)
4551 * NET_RX_DROP (packet was dropped)
4555 int netif_rx(struct sk_buff *skb)
4559 trace_netif_rx_entry(skb);
4561 ret = netif_rx_internal(skb);
4562 trace_netif_rx_exit(ret);
4566 EXPORT_SYMBOL(netif_rx);
4568 int netif_rx_ni(struct sk_buff *skb)
4572 trace_netif_rx_ni_entry(skb);
4575 err = netif_rx_internal(skb);
4576 if (local_softirq_pending())
4579 trace_netif_rx_ni_exit(err);
4583 EXPORT_SYMBOL(netif_rx_ni);
4585 static __latent_entropy void net_tx_action(struct softirq_action *h)
4587 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4589 if (sd->completion_queue) {
4590 struct sk_buff *clist;
4592 local_irq_disable();
4593 clist = sd->completion_queue;
4594 sd->completion_queue = NULL;
4598 struct sk_buff *skb = clist;
4600 clist = clist->next;
4602 WARN_ON(refcount_read(&skb->users));
4603 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4604 trace_consume_skb(skb);
4606 trace_kfree_skb(skb, net_tx_action);
4608 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4611 __kfree_skb_defer(skb);
4614 __kfree_skb_flush();
4617 if (sd->output_queue) {
4620 local_irq_disable();
4621 head = sd->output_queue;
4622 sd->output_queue = NULL;
4623 sd->output_queue_tailp = &sd->output_queue;
4627 struct Qdisc *q = head;
4628 spinlock_t *root_lock = NULL;
4630 head = head->next_sched;
4632 if (!(q->flags & TCQ_F_NOLOCK)) {
4633 root_lock = qdisc_lock(q);
4634 spin_lock(root_lock);
4636 /* We need to make sure head->next_sched is read
4637 * before clearing __QDISC_STATE_SCHED
4639 smp_mb__before_atomic();
4640 clear_bit(__QDISC_STATE_SCHED, &q->state);
4643 spin_unlock(root_lock);
4647 xfrm_dev_backlog(sd);
4650 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4651 /* This hook is defined here for ATM LANE */
4652 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4653 unsigned char *addr) __read_mostly;
4654 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4657 static inline struct sk_buff *
4658 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4659 struct net_device *orig_dev)
4661 #ifdef CONFIG_NET_CLS_ACT
4662 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4663 struct tcf_result cl_res;
4665 /* If there's at least one ingress present somewhere (so
4666 * we get here via enabled static key), remaining devices
4667 * that are not configured with an ingress qdisc will bail
4674 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4678 qdisc_skb_cb(skb)->pkt_len = skb->len;
4679 skb->tc_at_ingress = 1;
4680 mini_qdisc_bstats_cpu_update(miniq, skb);
4682 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4684 case TC_ACT_RECLASSIFY:
4685 skb->tc_index = TC_H_MIN(cl_res.classid);
4688 mini_qdisc_qstats_cpu_drop(miniq);
4696 case TC_ACT_REDIRECT:
4697 /* skb_mac_header check was done by cls/act_bpf, so
4698 * we can safely push the L2 header back before
4699 * redirecting to another netdev
4701 __skb_push(skb, skb->mac_len);
4702 skb_do_redirect(skb);
4704 case TC_ACT_REINSERT:
4705 /* this does not scrub the packet, and updates stats on error */
4706 skb_tc_reinsert(skb, &cl_res);
4711 #endif /* CONFIG_NET_CLS_ACT */
4716 * netdev_is_rx_handler_busy - check if receive handler is registered
4717 * @dev: device to check
4719 * Check if a receive handler is already registered for a given device.
4720 * Return true if there one.
4722 * The caller must hold the rtnl_mutex.
4724 bool netdev_is_rx_handler_busy(struct net_device *dev)
4727 return dev && rtnl_dereference(dev->rx_handler);
4729 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4732 * netdev_rx_handler_register - register receive handler
4733 * @dev: device to register a handler for
4734 * @rx_handler: receive handler to register
4735 * @rx_handler_data: data pointer that is used by rx handler
4737 * Register a receive handler for a device. This handler will then be
4738 * called from __netif_receive_skb. A negative errno code is returned
4741 * The caller must hold the rtnl_mutex.
4743 * For a general description of rx_handler, see enum rx_handler_result.
4745 int netdev_rx_handler_register(struct net_device *dev,
4746 rx_handler_func_t *rx_handler,
4747 void *rx_handler_data)
4749 if (netdev_is_rx_handler_busy(dev))
4752 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4755 /* Note: rx_handler_data must be set before rx_handler */
4756 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4757 rcu_assign_pointer(dev->rx_handler, rx_handler);
4761 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4764 * netdev_rx_handler_unregister - unregister receive handler
4765 * @dev: device to unregister a handler from
4767 * Unregister a receive handler from a device.
4769 * The caller must hold the rtnl_mutex.
4771 void netdev_rx_handler_unregister(struct net_device *dev)
4775 RCU_INIT_POINTER(dev->rx_handler, NULL);
4776 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4777 * section has a guarantee to see a non NULL rx_handler_data
4781 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4783 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4786 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4787 * the special handling of PFMEMALLOC skbs.
4789 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4791 switch (skb->protocol) {
4792 case htons(ETH_P_ARP):
4793 case htons(ETH_P_IP):
4794 case htons(ETH_P_IPV6):
4795 case htons(ETH_P_8021Q):
4796 case htons(ETH_P_8021AD):
4803 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4804 int *ret, struct net_device *orig_dev)
4806 #ifdef CONFIG_NETFILTER_INGRESS
4807 if (nf_hook_ingress_active(skb)) {
4811 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4816 ingress_retval = nf_hook_ingress(skb);
4818 return ingress_retval;
4820 #endif /* CONFIG_NETFILTER_INGRESS */
4824 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4825 struct packet_type **ppt_prev)
4827 struct packet_type *ptype, *pt_prev;
4828 rx_handler_func_t *rx_handler;
4829 struct net_device *orig_dev;
4830 bool deliver_exact = false;
4831 int ret = NET_RX_DROP;
4834 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4836 trace_netif_receive_skb(skb);
4838 orig_dev = skb->dev;
4840 skb_reset_network_header(skb);
4841 if (!skb_transport_header_was_set(skb))
4842 skb_reset_transport_header(skb);
4843 skb_reset_mac_len(skb);
4848 skb->skb_iif = skb->dev->ifindex;
4850 __this_cpu_inc(softnet_data.processed);
4852 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4853 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4854 skb = skb_vlan_untag(skb);
4859 if (skb_skip_tc_classify(skb))
4865 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4867 ret = deliver_skb(skb, pt_prev, orig_dev);
4871 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4873 ret = deliver_skb(skb, pt_prev, orig_dev);
4878 #ifdef CONFIG_NET_INGRESS
4879 if (static_branch_unlikely(&ingress_needed_key)) {
4880 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4884 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4890 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4893 if (skb_vlan_tag_present(skb)) {
4895 ret = deliver_skb(skb, pt_prev, orig_dev);
4898 if (vlan_do_receive(&skb))
4900 else if (unlikely(!skb))
4904 rx_handler = rcu_dereference(skb->dev->rx_handler);
4907 ret = deliver_skb(skb, pt_prev, orig_dev);
4910 switch (rx_handler(&skb)) {
4911 case RX_HANDLER_CONSUMED:
4912 ret = NET_RX_SUCCESS;
4914 case RX_HANDLER_ANOTHER:
4916 case RX_HANDLER_EXACT:
4917 deliver_exact = true;
4918 case RX_HANDLER_PASS:
4925 if (unlikely(skb_vlan_tag_present(skb))) {
4926 if (skb_vlan_tag_get_id(skb))
4927 skb->pkt_type = PACKET_OTHERHOST;
4928 /* Note: we might in the future use prio bits
4929 * and set skb->priority like in vlan_do_receive()
4930 * For the time being, just ignore Priority Code Point
4932 __vlan_hwaccel_clear_tag(skb);
4935 type = skb->protocol;
4937 /* deliver only exact match when indicated */
4938 if (likely(!deliver_exact)) {
4939 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4940 &ptype_base[ntohs(type) &
4944 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4945 &orig_dev->ptype_specific);
4947 if (unlikely(skb->dev != orig_dev)) {
4948 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4949 &skb->dev->ptype_specific);
4953 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4955 *ppt_prev = pt_prev;
4959 atomic_long_inc(&skb->dev->rx_dropped);
4961 atomic_long_inc(&skb->dev->rx_nohandler);
4963 /* Jamal, now you will not able to escape explaining
4964 * me how you were going to use this. :-)
4973 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4975 struct net_device *orig_dev = skb->dev;
4976 struct packet_type *pt_prev = NULL;
4979 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
4981 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4986 * netif_receive_skb_core - special purpose version of netif_receive_skb
4987 * @skb: buffer to process
4989 * More direct receive version of netif_receive_skb(). It should
4990 * only be used by callers that have a need to skip RPS and Generic XDP.
4991 * Caller must also take care of handling if (page_is_)pfmemalloc.
4993 * This function may only be called from softirq context and interrupts
4994 * should be enabled.
4996 * Return values (usually ignored):
4997 * NET_RX_SUCCESS: no congestion
4998 * NET_RX_DROP: packet was dropped
5000 int netif_receive_skb_core(struct sk_buff *skb)
5005 ret = __netif_receive_skb_one_core(skb, false);
5010 EXPORT_SYMBOL(netif_receive_skb_core);
5012 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5013 struct packet_type *pt_prev,
5014 struct net_device *orig_dev)
5016 struct sk_buff *skb, *next;
5020 if (list_empty(head))
5022 if (pt_prev->list_func != NULL)
5023 pt_prev->list_func(head, pt_prev, orig_dev);
5025 list_for_each_entry_safe(skb, next, head, list)
5026 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5029 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5031 /* Fast-path assumptions:
5032 * - There is no RX handler.
5033 * - Only one packet_type matches.
5034 * If either of these fails, we will end up doing some per-packet
5035 * processing in-line, then handling the 'last ptype' for the whole
5036 * sublist. This can't cause out-of-order delivery to any single ptype,
5037 * because the 'last ptype' must be constant across the sublist, and all
5038 * other ptypes are handled per-packet.
5040 /* Current (common) ptype of sublist */
5041 struct packet_type *pt_curr = NULL;
5042 /* Current (common) orig_dev of sublist */
5043 struct net_device *od_curr = NULL;
5044 struct list_head sublist;
5045 struct sk_buff *skb, *next;
5047 INIT_LIST_HEAD(&sublist);
5048 list_for_each_entry_safe(skb, next, head, list) {
5049 struct net_device *orig_dev = skb->dev;
5050 struct packet_type *pt_prev = NULL;
5052 skb_list_del_init(skb);
5053 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5056 if (pt_curr != pt_prev || od_curr != orig_dev) {
5057 /* dispatch old sublist */
5058 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5059 /* start new sublist */
5060 INIT_LIST_HEAD(&sublist);
5064 list_add_tail(&skb->list, &sublist);
5067 /* dispatch final sublist */
5068 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5071 static int __netif_receive_skb(struct sk_buff *skb)
5075 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5076 unsigned int noreclaim_flag;
5079 * PFMEMALLOC skbs are special, they should
5080 * - be delivered to SOCK_MEMALLOC sockets only
5081 * - stay away from userspace
5082 * - have bounded memory usage
5084 * Use PF_MEMALLOC as this saves us from propagating the allocation
5085 * context down to all allocation sites.
5087 noreclaim_flag = memalloc_noreclaim_save();
5088 ret = __netif_receive_skb_one_core(skb, true);
5089 memalloc_noreclaim_restore(noreclaim_flag);
5091 ret = __netif_receive_skb_one_core(skb, false);
5096 static void __netif_receive_skb_list(struct list_head *head)
5098 unsigned long noreclaim_flag = 0;
5099 struct sk_buff *skb, *next;
5100 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5102 list_for_each_entry_safe(skb, next, head, list) {
5103 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5104 struct list_head sublist;
5106 /* Handle the previous sublist */
5107 list_cut_before(&sublist, head, &skb->list);
5108 if (!list_empty(&sublist))
5109 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5110 pfmemalloc = !pfmemalloc;
5111 /* See comments in __netif_receive_skb */
5113 noreclaim_flag = memalloc_noreclaim_save();
5115 memalloc_noreclaim_restore(noreclaim_flag);
5118 /* Handle the remaining sublist */
5119 if (!list_empty(head))
5120 __netif_receive_skb_list_core(head, pfmemalloc);
5121 /* Restore pflags */
5123 memalloc_noreclaim_restore(noreclaim_flag);
5126 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5128 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5129 struct bpf_prog *new = xdp->prog;
5132 switch (xdp->command) {
5133 case XDP_SETUP_PROG:
5134 rcu_assign_pointer(dev->xdp_prog, new);
5139 static_branch_dec(&generic_xdp_needed_key);
5140 } else if (new && !old) {
5141 static_branch_inc(&generic_xdp_needed_key);
5142 dev_disable_lro(dev);
5143 dev_disable_gro_hw(dev);
5147 case XDP_QUERY_PROG:
5148 xdp->prog_id = old ? old->aux->id : 0;
5159 static int netif_receive_skb_internal(struct sk_buff *skb)
5163 net_timestamp_check(netdev_tstamp_prequeue, skb);
5165 if (skb_defer_rx_timestamp(skb))
5166 return NET_RX_SUCCESS;
5168 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5173 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5177 if (ret != XDP_PASS)
5183 if (static_branch_unlikely(&rps_needed)) {
5184 struct rps_dev_flow voidflow, *rflow = &voidflow;
5185 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5188 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5194 ret = __netif_receive_skb(skb);
5199 static void netif_receive_skb_list_internal(struct list_head *head)
5201 struct bpf_prog *xdp_prog = NULL;
5202 struct sk_buff *skb, *next;
5203 struct list_head sublist;
5205 INIT_LIST_HEAD(&sublist);
5206 list_for_each_entry_safe(skb, next, head, list) {
5207 net_timestamp_check(netdev_tstamp_prequeue, skb);
5208 skb_list_del_init(skb);
5209 if (!skb_defer_rx_timestamp(skb))
5210 list_add_tail(&skb->list, &sublist);
5212 list_splice_init(&sublist, head);
5214 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5217 list_for_each_entry_safe(skb, next, head, list) {
5218 xdp_prog = rcu_dereference(skb->dev->xdp_prog);
5219 skb_list_del_init(skb);
5220 if (do_xdp_generic(xdp_prog, skb) == XDP_PASS)
5221 list_add_tail(&skb->list, &sublist);
5225 /* Put passed packets back on main list */
5226 list_splice_init(&sublist, head);
5231 if (static_branch_unlikely(&rps_needed)) {
5232 list_for_each_entry_safe(skb, next, head, list) {
5233 struct rps_dev_flow voidflow, *rflow = &voidflow;
5234 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5237 /* Will be handled, remove from list */
5238 skb_list_del_init(skb);
5239 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5244 __netif_receive_skb_list(head);
5249 * netif_receive_skb - process receive buffer from network
5250 * @skb: buffer to process
5252 * netif_receive_skb() is the main receive data processing function.
5253 * It always succeeds. The buffer may be dropped during processing
5254 * for congestion control or by the protocol layers.
5256 * This function may only be called from softirq context and interrupts
5257 * should be enabled.
5259 * Return values (usually ignored):
5260 * NET_RX_SUCCESS: no congestion
5261 * NET_RX_DROP: packet was dropped
5263 int netif_receive_skb(struct sk_buff *skb)
5267 trace_netif_receive_skb_entry(skb);
5269 ret = netif_receive_skb_internal(skb);
5270 trace_netif_receive_skb_exit(ret);
5274 EXPORT_SYMBOL(netif_receive_skb);
5277 * netif_receive_skb_list - process many receive buffers from network
5278 * @head: list of skbs to process.
5280 * Since return value of netif_receive_skb() is normally ignored, and
5281 * wouldn't be meaningful for a list, this function returns void.
5283 * This function may only be called from softirq context and interrupts
5284 * should be enabled.
5286 void netif_receive_skb_list(struct list_head *head)
5288 struct sk_buff *skb;
5290 if (list_empty(head))
5292 if (trace_netif_receive_skb_list_entry_enabled()) {
5293 list_for_each_entry(skb, head, list)
5294 trace_netif_receive_skb_list_entry(skb);
5296 netif_receive_skb_list_internal(head);
5297 trace_netif_receive_skb_list_exit(0);
5299 EXPORT_SYMBOL(netif_receive_skb_list);
5301 DEFINE_PER_CPU(struct work_struct, flush_works);
5303 /* Network device is going away, flush any packets still pending */
5304 static void flush_backlog(struct work_struct *work)
5306 struct sk_buff *skb, *tmp;
5307 struct softnet_data *sd;
5310 sd = this_cpu_ptr(&softnet_data);
5312 local_irq_disable();
5314 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5315 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5316 __skb_unlink(skb, &sd->input_pkt_queue);
5318 input_queue_head_incr(sd);
5324 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5325 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5326 __skb_unlink(skb, &sd->process_queue);
5328 input_queue_head_incr(sd);
5334 static void flush_all_backlogs(void)
5340 for_each_online_cpu(cpu)
5341 queue_work_on(cpu, system_highpri_wq,
5342 per_cpu_ptr(&flush_works, cpu));
5344 for_each_online_cpu(cpu)
5345 flush_work(per_cpu_ptr(&flush_works, cpu));
5350 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5351 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5352 static int napi_gro_complete(struct sk_buff *skb)
5354 struct packet_offload *ptype;
5355 __be16 type = skb->protocol;
5356 struct list_head *head = &offload_base;
5359 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5361 if (NAPI_GRO_CB(skb)->count == 1) {
5362 skb_shinfo(skb)->gso_size = 0;
5367 list_for_each_entry_rcu(ptype, head, list) {
5368 if (ptype->type != type || !ptype->callbacks.gro_complete)
5371 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5372 ipv6_gro_complete, inet_gro_complete,
5379 WARN_ON(&ptype->list == head);
5381 return NET_RX_SUCCESS;
5385 return netif_receive_skb_internal(skb);
5388 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5391 struct list_head *head = &napi->gro_hash[index].list;
5392 struct sk_buff *skb, *p;
5394 list_for_each_entry_safe_reverse(skb, p, head, list) {
5395 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5397 skb_list_del_init(skb);
5398 napi_gro_complete(skb);
5399 napi->gro_hash[index].count--;
5402 if (!napi->gro_hash[index].count)
5403 __clear_bit(index, &napi->gro_bitmask);
5406 /* napi->gro_hash[].list contains packets ordered by age.
5407 * youngest packets at the head of it.
5408 * Complete skbs in reverse order to reduce latencies.
5410 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5412 unsigned long bitmask = napi->gro_bitmask;
5413 unsigned int i, base = ~0U;
5415 while ((i = ffs(bitmask)) != 0) {
5418 __napi_gro_flush_chain(napi, base, flush_old);
5421 EXPORT_SYMBOL(napi_gro_flush);
5423 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5424 struct sk_buff *skb)
5426 unsigned int maclen = skb->dev->hard_header_len;
5427 u32 hash = skb_get_hash_raw(skb);
5428 struct list_head *head;
5431 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5432 list_for_each_entry(p, head, list) {
5433 unsigned long diffs;
5435 NAPI_GRO_CB(p)->flush = 0;
5437 if (hash != skb_get_hash_raw(p)) {
5438 NAPI_GRO_CB(p)->same_flow = 0;
5442 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5443 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5444 if (skb_vlan_tag_present(p))
5445 diffs |= p->vlan_tci ^ skb->vlan_tci;
5446 diffs |= skb_metadata_dst_cmp(p, skb);
5447 diffs |= skb_metadata_differs(p, skb);
5448 if (maclen == ETH_HLEN)
5449 diffs |= compare_ether_header(skb_mac_header(p),
5450 skb_mac_header(skb));
5452 diffs = memcmp(skb_mac_header(p),
5453 skb_mac_header(skb),
5455 NAPI_GRO_CB(p)->same_flow = !diffs;
5461 static void skb_gro_reset_offset(struct sk_buff *skb)
5463 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5464 const skb_frag_t *frag0 = &pinfo->frags[0];
5466 NAPI_GRO_CB(skb)->data_offset = 0;
5467 NAPI_GRO_CB(skb)->frag0 = NULL;
5468 NAPI_GRO_CB(skb)->frag0_len = 0;
5470 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5472 !PageHighMem(skb_frag_page(frag0))) {
5473 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5474 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5475 skb_frag_size(frag0),
5476 skb->end - skb->tail);
5480 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5482 struct skb_shared_info *pinfo = skb_shinfo(skb);
5484 BUG_ON(skb->end - skb->tail < grow);
5486 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5488 skb->data_len -= grow;
5491 pinfo->frags[0].page_offset += grow;
5492 skb_frag_size_sub(&pinfo->frags[0], grow);
5494 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5495 skb_frag_unref(skb, 0);
5496 memmove(pinfo->frags, pinfo->frags + 1,
5497 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5501 static void gro_flush_oldest(struct list_head *head)
5503 struct sk_buff *oldest;
5505 oldest = list_last_entry(head, struct sk_buff, list);
5507 /* We are called with head length >= MAX_GRO_SKBS, so this is
5510 if (WARN_ON_ONCE(!oldest))
5513 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5516 skb_list_del_init(oldest);
5517 napi_gro_complete(oldest);
5520 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5522 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5524 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5526 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5527 struct list_head *head = &offload_base;
5528 struct packet_offload *ptype;
5529 __be16 type = skb->protocol;
5530 struct list_head *gro_head;
5531 struct sk_buff *pp = NULL;
5532 enum gro_result ret;
5536 if (netif_elide_gro(skb->dev))
5539 gro_head = gro_list_prepare(napi, skb);
5542 list_for_each_entry_rcu(ptype, head, list) {
5543 if (ptype->type != type || !ptype->callbacks.gro_receive)
5546 skb_set_network_header(skb, skb_gro_offset(skb));
5547 skb_reset_mac_len(skb);
5548 NAPI_GRO_CB(skb)->same_flow = 0;
5549 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5550 NAPI_GRO_CB(skb)->free = 0;
5551 NAPI_GRO_CB(skb)->encap_mark = 0;
5552 NAPI_GRO_CB(skb)->recursion_counter = 0;
5553 NAPI_GRO_CB(skb)->is_fou = 0;
5554 NAPI_GRO_CB(skb)->is_atomic = 1;
5555 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5557 /* Setup for GRO checksum validation */
5558 switch (skb->ip_summed) {
5559 case CHECKSUM_COMPLETE:
5560 NAPI_GRO_CB(skb)->csum = skb->csum;
5561 NAPI_GRO_CB(skb)->csum_valid = 1;
5562 NAPI_GRO_CB(skb)->csum_cnt = 0;
5564 case CHECKSUM_UNNECESSARY:
5565 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5566 NAPI_GRO_CB(skb)->csum_valid = 0;
5569 NAPI_GRO_CB(skb)->csum_cnt = 0;
5570 NAPI_GRO_CB(skb)->csum_valid = 0;
5573 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5574 ipv6_gro_receive, inet_gro_receive,
5580 if (&ptype->list == head)
5583 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5588 same_flow = NAPI_GRO_CB(skb)->same_flow;
5589 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5592 skb_list_del_init(pp);
5593 napi_gro_complete(pp);
5594 napi->gro_hash[hash].count--;
5600 if (NAPI_GRO_CB(skb)->flush)
5603 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5604 gro_flush_oldest(gro_head);
5606 napi->gro_hash[hash].count++;
5608 NAPI_GRO_CB(skb)->count = 1;
5609 NAPI_GRO_CB(skb)->age = jiffies;
5610 NAPI_GRO_CB(skb)->last = skb;
5611 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5612 list_add(&skb->list, gro_head);
5616 grow = skb_gro_offset(skb) - skb_headlen(skb);
5618 gro_pull_from_frag0(skb, grow);
5620 if (napi->gro_hash[hash].count) {
5621 if (!test_bit(hash, &napi->gro_bitmask))
5622 __set_bit(hash, &napi->gro_bitmask);
5623 } else if (test_bit(hash, &napi->gro_bitmask)) {
5624 __clear_bit(hash, &napi->gro_bitmask);
5634 struct packet_offload *gro_find_receive_by_type(__be16 type)
5636 struct list_head *offload_head = &offload_base;
5637 struct packet_offload *ptype;
5639 list_for_each_entry_rcu(ptype, offload_head, list) {
5640 if (ptype->type != type || !ptype->callbacks.gro_receive)
5646 EXPORT_SYMBOL(gro_find_receive_by_type);
5648 struct packet_offload *gro_find_complete_by_type(__be16 type)
5650 struct list_head *offload_head = &offload_base;
5651 struct packet_offload *ptype;
5653 list_for_each_entry_rcu(ptype, offload_head, list) {
5654 if (ptype->type != type || !ptype->callbacks.gro_complete)
5660 EXPORT_SYMBOL(gro_find_complete_by_type);
5662 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5666 kmem_cache_free(skbuff_head_cache, skb);
5669 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5673 if (netif_receive_skb_internal(skb))
5681 case GRO_MERGED_FREE:
5682 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5683 napi_skb_free_stolen_head(skb);
5697 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5701 skb_mark_napi_id(skb, napi);
5702 trace_napi_gro_receive_entry(skb);
5704 skb_gro_reset_offset(skb);
5706 ret = napi_skb_finish(dev_gro_receive(napi, skb), skb);
5707 trace_napi_gro_receive_exit(ret);
5711 EXPORT_SYMBOL(napi_gro_receive);
5713 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5715 if (unlikely(skb->pfmemalloc)) {
5719 __skb_pull(skb, skb_headlen(skb));
5720 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5721 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5722 __vlan_hwaccel_clear_tag(skb);
5723 skb->dev = napi->dev;
5726 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5727 skb->pkt_type = PACKET_HOST;
5729 skb->encapsulation = 0;
5730 skb_shinfo(skb)->gso_type = 0;
5731 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5737 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5739 struct sk_buff *skb = napi->skb;
5742 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5745 skb_mark_napi_id(skb, napi);
5750 EXPORT_SYMBOL(napi_get_frags);
5752 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5753 struct sk_buff *skb,
5759 __skb_push(skb, ETH_HLEN);
5760 skb->protocol = eth_type_trans(skb, skb->dev);
5761 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5766 napi_reuse_skb(napi, skb);
5769 case GRO_MERGED_FREE:
5770 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5771 napi_skb_free_stolen_head(skb);
5773 napi_reuse_skb(napi, skb);
5784 /* Upper GRO stack assumes network header starts at gro_offset=0
5785 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5786 * We copy ethernet header into skb->data to have a common layout.
5788 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5790 struct sk_buff *skb = napi->skb;
5791 const struct ethhdr *eth;
5792 unsigned int hlen = sizeof(*eth);
5796 skb_reset_mac_header(skb);
5797 skb_gro_reset_offset(skb);
5799 eth = skb_gro_header_fast(skb, 0);
5800 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5801 eth = skb_gro_header_slow(skb, hlen, 0);
5802 if (unlikely(!eth)) {
5803 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5804 __func__, napi->dev->name);
5805 napi_reuse_skb(napi, skb);
5809 gro_pull_from_frag0(skb, hlen);
5810 NAPI_GRO_CB(skb)->frag0 += hlen;
5811 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5813 __skb_pull(skb, hlen);
5816 * This works because the only protocols we care about don't require
5818 * We'll fix it up properly in napi_frags_finish()
5820 skb->protocol = eth->h_proto;
5825 gro_result_t napi_gro_frags(struct napi_struct *napi)
5828 struct sk_buff *skb = napi_frags_skb(napi);
5833 trace_napi_gro_frags_entry(skb);
5835 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5836 trace_napi_gro_frags_exit(ret);
5840 EXPORT_SYMBOL(napi_gro_frags);
5842 /* Compute the checksum from gro_offset and return the folded value
5843 * after adding in any pseudo checksum.
5845 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5850 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5852 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5853 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5854 /* See comments in __skb_checksum_complete(). */
5856 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5857 !skb->csum_complete_sw)
5858 netdev_rx_csum_fault(skb->dev, skb);
5861 NAPI_GRO_CB(skb)->csum = wsum;
5862 NAPI_GRO_CB(skb)->csum_valid = 1;
5866 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5868 static void net_rps_send_ipi(struct softnet_data *remsd)
5872 struct softnet_data *next = remsd->rps_ipi_next;
5874 if (cpu_online(remsd->cpu))
5875 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5882 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5883 * Note: called with local irq disabled, but exits with local irq enabled.
5885 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5888 struct softnet_data *remsd = sd->rps_ipi_list;
5891 sd->rps_ipi_list = NULL;
5895 /* Send pending IPI's to kick RPS processing on remote cpus. */
5896 net_rps_send_ipi(remsd);
5902 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5905 return sd->rps_ipi_list != NULL;
5911 static int process_backlog(struct napi_struct *napi, int quota)
5913 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5917 /* Check if we have pending ipi, its better to send them now,
5918 * not waiting net_rx_action() end.
5920 if (sd_has_rps_ipi_waiting(sd)) {
5921 local_irq_disable();
5922 net_rps_action_and_irq_enable(sd);
5925 napi->weight = dev_rx_weight;
5927 struct sk_buff *skb;
5929 while ((skb = __skb_dequeue(&sd->process_queue))) {
5931 __netif_receive_skb(skb);
5933 input_queue_head_incr(sd);
5934 if (++work >= quota)
5939 local_irq_disable();
5941 if (skb_queue_empty(&sd->input_pkt_queue)) {
5943 * Inline a custom version of __napi_complete().
5944 * only current cpu owns and manipulates this napi,
5945 * and NAPI_STATE_SCHED is the only possible flag set
5947 * We can use a plain write instead of clear_bit(),
5948 * and we dont need an smp_mb() memory barrier.
5953 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5954 &sd->process_queue);
5964 * __napi_schedule - schedule for receive
5965 * @n: entry to schedule
5967 * The entry's receive function will be scheduled to run.
5968 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5970 void __napi_schedule(struct napi_struct *n)
5972 unsigned long flags;
5974 local_irq_save(flags);
5975 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5976 local_irq_restore(flags);
5978 EXPORT_SYMBOL(__napi_schedule);
5981 * napi_schedule_prep - check if napi can be scheduled
5984 * Test if NAPI routine is already running, and if not mark
5985 * it as running. This is used as a condition variable
5986 * insure only one NAPI poll instance runs. We also make
5987 * sure there is no pending NAPI disable.
5989 bool napi_schedule_prep(struct napi_struct *n)
5991 unsigned long val, new;
5994 val = READ_ONCE(n->state);
5995 if (unlikely(val & NAPIF_STATE_DISABLE))
5997 new = val | NAPIF_STATE_SCHED;
5999 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6000 * This was suggested by Alexander Duyck, as compiler
6001 * emits better code than :
6002 * if (val & NAPIF_STATE_SCHED)
6003 * new |= NAPIF_STATE_MISSED;
6005 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6007 } while (cmpxchg(&n->state, val, new) != val);
6009 return !(val & NAPIF_STATE_SCHED);
6011 EXPORT_SYMBOL(napi_schedule_prep);
6014 * __napi_schedule_irqoff - schedule for receive
6015 * @n: entry to schedule
6017 * Variant of __napi_schedule() assuming hard irqs are masked
6019 void __napi_schedule_irqoff(struct napi_struct *n)
6021 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6023 EXPORT_SYMBOL(__napi_schedule_irqoff);
6025 bool napi_complete_done(struct napi_struct *n, int work_done)
6027 unsigned long flags, val, new;
6030 * 1) Don't let napi dequeue from the cpu poll list
6031 * just in case its running on a different cpu.
6032 * 2) If we are busy polling, do nothing here, we have
6033 * the guarantee we will be called later.
6035 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6036 NAPIF_STATE_IN_BUSY_POLL)))
6039 if (n->gro_bitmask) {
6040 unsigned long timeout = 0;
6043 timeout = n->dev->gro_flush_timeout;
6045 /* When the NAPI instance uses a timeout and keeps postponing
6046 * it, we need to bound somehow the time packets are kept in
6049 napi_gro_flush(n, !!timeout);
6051 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6052 HRTIMER_MODE_REL_PINNED);
6054 if (unlikely(!list_empty(&n->poll_list))) {
6055 /* If n->poll_list is not empty, we need to mask irqs */
6056 local_irq_save(flags);
6057 list_del_init(&n->poll_list);
6058 local_irq_restore(flags);
6062 val = READ_ONCE(n->state);
6064 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6066 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6068 /* If STATE_MISSED was set, leave STATE_SCHED set,
6069 * because we will call napi->poll() one more time.
6070 * This C code was suggested by Alexander Duyck to help gcc.
6072 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6074 } while (cmpxchg(&n->state, val, new) != val);
6076 if (unlikely(val & NAPIF_STATE_MISSED)) {
6083 EXPORT_SYMBOL(napi_complete_done);
6085 /* must be called under rcu_read_lock(), as we dont take a reference */
6086 static struct napi_struct *napi_by_id(unsigned int napi_id)
6088 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6089 struct napi_struct *napi;
6091 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6092 if (napi->napi_id == napi_id)
6098 #if defined(CONFIG_NET_RX_BUSY_POLL)
6100 #define BUSY_POLL_BUDGET 8
6102 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6106 /* Busy polling means there is a high chance device driver hard irq
6107 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6108 * set in napi_schedule_prep().
6109 * Since we are about to call napi->poll() once more, we can safely
6110 * clear NAPI_STATE_MISSED.
6112 * Note: x86 could use a single "lock and ..." instruction
6113 * to perform these two clear_bit()
6115 clear_bit(NAPI_STATE_MISSED, &napi->state);
6116 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6120 /* All we really want here is to re-enable device interrupts.
6121 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6123 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6124 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6125 netpoll_poll_unlock(have_poll_lock);
6126 if (rc == BUSY_POLL_BUDGET)
6127 __napi_schedule(napi);
6131 void napi_busy_loop(unsigned int napi_id,
6132 bool (*loop_end)(void *, unsigned long),
6135 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6136 int (*napi_poll)(struct napi_struct *napi, int budget);
6137 void *have_poll_lock = NULL;
6138 struct napi_struct *napi;
6145 napi = napi_by_id(napi_id);
6155 unsigned long val = READ_ONCE(napi->state);
6157 /* If multiple threads are competing for this napi,
6158 * we avoid dirtying napi->state as much as we can.
6160 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6161 NAPIF_STATE_IN_BUSY_POLL))
6163 if (cmpxchg(&napi->state, val,
6164 val | NAPIF_STATE_IN_BUSY_POLL |
6165 NAPIF_STATE_SCHED) != val)
6167 have_poll_lock = netpoll_poll_lock(napi);
6168 napi_poll = napi->poll;
6170 work = napi_poll(napi, BUSY_POLL_BUDGET);
6171 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6174 __NET_ADD_STATS(dev_net(napi->dev),
6175 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6178 if (!loop_end || loop_end(loop_end_arg, start_time))
6181 if (unlikely(need_resched())) {
6183 busy_poll_stop(napi, have_poll_lock);
6187 if (loop_end(loop_end_arg, start_time))
6194 busy_poll_stop(napi, have_poll_lock);
6199 EXPORT_SYMBOL(napi_busy_loop);
6201 #endif /* CONFIG_NET_RX_BUSY_POLL */
6203 static void napi_hash_add(struct napi_struct *napi)
6205 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6206 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6209 spin_lock(&napi_hash_lock);
6211 /* 0..NR_CPUS range is reserved for sender_cpu use */
6213 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6214 napi_gen_id = MIN_NAPI_ID;
6215 } while (napi_by_id(napi_gen_id));
6216 napi->napi_id = napi_gen_id;
6218 hlist_add_head_rcu(&napi->napi_hash_node,
6219 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6221 spin_unlock(&napi_hash_lock);
6224 /* Warning : caller is responsible to make sure rcu grace period
6225 * is respected before freeing memory containing @napi
6227 bool napi_hash_del(struct napi_struct *napi)
6229 bool rcu_sync_needed = false;
6231 spin_lock(&napi_hash_lock);
6233 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6234 rcu_sync_needed = true;
6235 hlist_del_rcu(&napi->napi_hash_node);
6237 spin_unlock(&napi_hash_lock);
6238 return rcu_sync_needed;
6240 EXPORT_SYMBOL_GPL(napi_hash_del);
6242 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6244 struct napi_struct *napi;
6246 napi = container_of(timer, struct napi_struct, timer);
6248 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6249 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6251 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6252 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6253 __napi_schedule_irqoff(napi);
6255 return HRTIMER_NORESTART;
6258 static void init_gro_hash(struct napi_struct *napi)
6262 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6263 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6264 napi->gro_hash[i].count = 0;
6266 napi->gro_bitmask = 0;
6269 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6270 int (*poll)(struct napi_struct *, int), int weight)
6272 INIT_LIST_HEAD(&napi->poll_list);
6273 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6274 napi->timer.function = napi_watchdog;
6275 init_gro_hash(napi);
6278 if (weight > NAPI_POLL_WEIGHT)
6279 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6281 napi->weight = weight;
6282 list_add(&napi->dev_list, &dev->napi_list);
6284 #ifdef CONFIG_NETPOLL
6285 napi->poll_owner = -1;
6287 set_bit(NAPI_STATE_SCHED, &napi->state);
6288 napi_hash_add(napi);
6290 EXPORT_SYMBOL(netif_napi_add);
6292 void napi_disable(struct napi_struct *n)
6295 set_bit(NAPI_STATE_DISABLE, &n->state);
6297 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6299 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6302 hrtimer_cancel(&n->timer);
6304 clear_bit(NAPI_STATE_DISABLE, &n->state);
6306 EXPORT_SYMBOL(napi_disable);
6308 static void flush_gro_hash(struct napi_struct *napi)
6312 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6313 struct sk_buff *skb, *n;
6315 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6317 napi->gro_hash[i].count = 0;
6321 /* Must be called in process context */
6322 void netif_napi_del(struct napi_struct *napi)
6325 if (napi_hash_del(napi))
6327 list_del_init(&napi->dev_list);
6328 napi_free_frags(napi);
6330 flush_gro_hash(napi);
6331 napi->gro_bitmask = 0;
6333 EXPORT_SYMBOL(netif_napi_del);
6335 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6340 list_del_init(&n->poll_list);
6342 have = netpoll_poll_lock(n);
6346 /* This NAPI_STATE_SCHED test is for avoiding a race
6347 * with netpoll's poll_napi(). Only the entity which
6348 * obtains the lock and sees NAPI_STATE_SCHED set will
6349 * actually make the ->poll() call. Therefore we avoid
6350 * accidentally calling ->poll() when NAPI is not scheduled.
6353 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6354 work = n->poll(n, weight);
6355 trace_napi_poll(n, work, weight);
6358 WARN_ON_ONCE(work > weight);
6360 if (likely(work < weight))
6363 /* Drivers must not modify the NAPI state if they
6364 * consume the entire weight. In such cases this code
6365 * still "owns" the NAPI instance and therefore can
6366 * move the instance around on the list at-will.
6368 if (unlikely(napi_disable_pending(n))) {
6373 if (n->gro_bitmask) {
6374 /* flush too old packets
6375 * If HZ < 1000, flush all packets.
6377 napi_gro_flush(n, HZ >= 1000);
6380 /* Some drivers may have called napi_schedule
6381 * prior to exhausting their budget.
6383 if (unlikely(!list_empty(&n->poll_list))) {
6384 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6385 n->dev ? n->dev->name : "backlog");
6389 list_add_tail(&n->poll_list, repoll);
6392 netpoll_poll_unlock(have);
6397 static __latent_entropy void net_rx_action(struct softirq_action *h)
6399 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6400 unsigned long time_limit = jiffies +
6401 usecs_to_jiffies(netdev_budget_usecs);
6402 int budget = netdev_budget;
6406 local_irq_disable();
6407 list_splice_init(&sd->poll_list, &list);
6411 struct napi_struct *n;
6413 if (list_empty(&list)) {
6414 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6419 n = list_first_entry(&list, struct napi_struct, poll_list);
6420 budget -= napi_poll(n, &repoll);
6422 /* If softirq window is exhausted then punt.
6423 * Allow this to run for 2 jiffies since which will allow
6424 * an average latency of 1.5/HZ.
6426 if (unlikely(budget <= 0 ||
6427 time_after_eq(jiffies, time_limit))) {
6433 local_irq_disable();
6435 list_splice_tail_init(&sd->poll_list, &list);
6436 list_splice_tail(&repoll, &list);
6437 list_splice(&list, &sd->poll_list);
6438 if (!list_empty(&sd->poll_list))
6439 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6441 net_rps_action_and_irq_enable(sd);
6443 __kfree_skb_flush();
6446 struct netdev_adjacent {
6447 struct net_device *dev;
6449 /* upper master flag, there can only be one master device per list */
6452 /* counter for the number of times this device was added to us */
6455 /* private field for the users */
6458 struct list_head list;
6459 struct rcu_head rcu;
6462 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6463 struct list_head *adj_list)
6465 struct netdev_adjacent *adj;
6467 list_for_each_entry(adj, adj_list, list) {
6468 if (adj->dev == adj_dev)
6474 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6476 struct net_device *dev = data;
6478 return upper_dev == dev;
6482 * netdev_has_upper_dev - Check if device is linked to an upper device
6484 * @upper_dev: upper device to check
6486 * Find out if a device is linked to specified upper device and return true
6487 * in case it is. Note that this checks only immediate upper device,
6488 * not through a complete stack of devices. The caller must hold the RTNL lock.
6490 bool netdev_has_upper_dev(struct net_device *dev,
6491 struct net_device *upper_dev)
6495 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6498 EXPORT_SYMBOL(netdev_has_upper_dev);
6501 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6503 * @upper_dev: upper device to check
6505 * Find out if a device is linked to specified upper device and return true
6506 * in case it is. Note that this checks the entire upper device chain.
6507 * The caller must hold rcu lock.
6510 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6511 struct net_device *upper_dev)
6513 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6516 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6519 * netdev_has_any_upper_dev - Check if device is linked to some device
6522 * Find out if a device is linked to an upper device and return true in case
6523 * it is. The caller must hold the RTNL lock.
6525 bool netdev_has_any_upper_dev(struct net_device *dev)
6529 return !list_empty(&dev->adj_list.upper);
6531 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6534 * netdev_master_upper_dev_get - Get master upper device
6537 * Find a master upper device and return pointer to it or NULL in case
6538 * it's not there. The caller must hold the RTNL lock.
6540 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6542 struct netdev_adjacent *upper;
6546 if (list_empty(&dev->adj_list.upper))
6549 upper = list_first_entry(&dev->adj_list.upper,
6550 struct netdev_adjacent, list);
6551 if (likely(upper->master))
6555 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6558 * netdev_has_any_lower_dev - Check if device is linked to some device
6561 * Find out if a device is linked to a lower device and return true in case
6562 * it is. The caller must hold the RTNL lock.
6564 static bool netdev_has_any_lower_dev(struct net_device *dev)
6568 return !list_empty(&dev->adj_list.lower);
6571 void *netdev_adjacent_get_private(struct list_head *adj_list)
6573 struct netdev_adjacent *adj;
6575 adj = list_entry(adj_list, struct netdev_adjacent, list);
6577 return adj->private;
6579 EXPORT_SYMBOL(netdev_adjacent_get_private);
6582 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6584 * @iter: list_head ** of the current position
6586 * Gets the next device from the dev's upper list, starting from iter
6587 * position. The caller must hold RCU read lock.
6589 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6590 struct list_head **iter)
6592 struct netdev_adjacent *upper;
6594 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6596 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6598 if (&upper->list == &dev->adj_list.upper)
6601 *iter = &upper->list;
6605 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6607 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6608 struct list_head **iter)
6610 struct netdev_adjacent *upper;
6612 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6614 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6616 if (&upper->list == &dev->adj_list.upper)
6619 *iter = &upper->list;
6624 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6625 int (*fn)(struct net_device *dev,
6629 struct net_device *udev;
6630 struct list_head *iter;
6633 for (iter = &dev->adj_list.upper,
6634 udev = netdev_next_upper_dev_rcu(dev, &iter);
6636 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6637 /* first is the upper device itself */
6638 ret = fn(udev, data);
6642 /* then look at all of its upper devices */
6643 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6650 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6653 * netdev_lower_get_next_private - Get the next ->private from the
6654 * lower neighbour list
6656 * @iter: list_head ** of the current position
6658 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6659 * list, starting from iter position. The caller must hold either hold the
6660 * RTNL lock or its own locking that guarantees that the neighbour lower
6661 * list will remain unchanged.
6663 void *netdev_lower_get_next_private(struct net_device *dev,
6664 struct list_head **iter)
6666 struct netdev_adjacent *lower;
6668 lower = list_entry(*iter, struct netdev_adjacent, list);
6670 if (&lower->list == &dev->adj_list.lower)
6673 *iter = lower->list.next;
6675 return lower->private;
6677 EXPORT_SYMBOL(netdev_lower_get_next_private);
6680 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6681 * lower neighbour list, RCU
6684 * @iter: list_head ** of the current position
6686 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6687 * list, starting from iter position. The caller must hold RCU read lock.
6689 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6690 struct list_head **iter)
6692 struct netdev_adjacent *lower;
6694 WARN_ON_ONCE(!rcu_read_lock_held());
6696 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6698 if (&lower->list == &dev->adj_list.lower)
6701 *iter = &lower->list;
6703 return lower->private;
6705 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6708 * netdev_lower_get_next - Get the next device from the lower neighbour
6711 * @iter: list_head ** of the current position
6713 * Gets the next netdev_adjacent from the dev's lower neighbour
6714 * list, starting from iter position. The caller must hold RTNL lock or
6715 * its own locking that guarantees that the neighbour lower
6716 * list will remain unchanged.
6718 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6720 struct netdev_adjacent *lower;
6722 lower = list_entry(*iter, struct netdev_adjacent, list);
6724 if (&lower->list == &dev->adj_list.lower)
6727 *iter = lower->list.next;
6731 EXPORT_SYMBOL(netdev_lower_get_next);
6733 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6734 struct list_head **iter)
6736 struct netdev_adjacent *lower;
6738 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6740 if (&lower->list == &dev->adj_list.lower)
6743 *iter = &lower->list;
6748 int netdev_walk_all_lower_dev(struct net_device *dev,
6749 int (*fn)(struct net_device *dev,
6753 struct net_device *ldev;
6754 struct list_head *iter;
6757 for (iter = &dev->adj_list.lower,
6758 ldev = netdev_next_lower_dev(dev, &iter);
6760 ldev = netdev_next_lower_dev(dev, &iter)) {
6761 /* first is the lower device itself */
6762 ret = fn(ldev, data);
6766 /* then look at all of its lower devices */
6767 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6774 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6776 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6777 struct list_head **iter)
6779 struct netdev_adjacent *lower;
6781 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6782 if (&lower->list == &dev->adj_list.lower)
6785 *iter = &lower->list;
6790 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6791 int (*fn)(struct net_device *dev,
6795 struct net_device *ldev;
6796 struct list_head *iter;
6799 for (iter = &dev->adj_list.lower,
6800 ldev = netdev_next_lower_dev_rcu(dev, &iter);
6802 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6803 /* first is the lower device itself */
6804 ret = fn(ldev, data);
6808 /* then look at all of its lower devices */
6809 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6816 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6819 * netdev_lower_get_first_private_rcu - Get the first ->private from the
6820 * lower neighbour list, RCU
6824 * Gets the first netdev_adjacent->private from the dev's lower neighbour
6825 * list. The caller must hold RCU read lock.
6827 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6829 struct netdev_adjacent *lower;
6831 lower = list_first_or_null_rcu(&dev->adj_list.lower,
6832 struct netdev_adjacent, list);
6834 return lower->private;
6837 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6840 * netdev_master_upper_dev_get_rcu - Get master upper device
6843 * Find a master upper device and return pointer to it or NULL in case
6844 * it's not there. The caller must hold the RCU read lock.
6846 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6848 struct netdev_adjacent *upper;
6850 upper = list_first_or_null_rcu(&dev->adj_list.upper,
6851 struct netdev_adjacent, list);
6852 if (upper && likely(upper->master))
6856 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6858 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6859 struct net_device *adj_dev,
6860 struct list_head *dev_list)
6862 char linkname[IFNAMSIZ+7];
6864 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6865 "upper_%s" : "lower_%s", adj_dev->name);
6866 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6869 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6871 struct list_head *dev_list)
6873 char linkname[IFNAMSIZ+7];
6875 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6876 "upper_%s" : "lower_%s", name);
6877 sysfs_remove_link(&(dev->dev.kobj), linkname);
6880 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6881 struct net_device *adj_dev,
6882 struct list_head *dev_list)
6884 return (dev_list == &dev->adj_list.upper ||
6885 dev_list == &dev->adj_list.lower) &&
6886 net_eq(dev_net(dev), dev_net(adj_dev));
6889 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6890 struct net_device *adj_dev,
6891 struct list_head *dev_list,
6892 void *private, bool master)
6894 struct netdev_adjacent *adj;
6897 adj = __netdev_find_adj(adj_dev, dev_list);
6901 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6902 dev->name, adj_dev->name, adj->ref_nr);
6907 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6912 adj->master = master;
6914 adj->private = private;
6917 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6918 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6920 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6921 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6926 /* Ensure that master link is always the first item in list. */
6928 ret = sysfs_create_link(&(dev->dev.kobj),
6929 &(adj_dev->dev.kobj), "master");
6931 goto remove_symlinks;
6933 list_add_rcu(&adj->list, dev_list);
6935 list_add_tail_rcu(&adj->list, dev_list);
6941 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6942 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6950 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6951 struct net_device *adj_dev,
6953 struct list_head *dev_list)
6955 struct netdev_adjacent *adj;
6957 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6958 dev->name, adj_dev->name, ref_nr);
6960 adj = __netdev_find_adj(adj_dev, dev_list);
6963 pr_err("Adjacency does not exist for device %s from %s\n",
6964 dev->name, adj_dev->name);
6969 if (adj->ref_nr > ref_nr) {
6970 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6971 dev->name, adj_dev->name, ref_nr,
6972 adj->ref_nr - ref_nr);
6973 adj->ref_nr -= ref_nr;
6978 sysfs_remove_link(&(dev->dev.kobj), "master");
6980 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6981 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6983 list_del_rcu(&adj->list);
6984 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6985 adj_dev->name, dev->name, adj_dev->name);
6987 kfree_rcu(adj, rcu);
6990 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6991 struct net_device *upper_dev,
6992 struct list_head *up_list,
6993 struct list_head *down_list,
6994 void *private, bool master)
6998 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7003 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7006 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7013 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7014 struct net_device *upper_dev,
7016 struct list_head *up_list,
7017 struct list_head *down_list)
7019 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7020 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7023 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7024 struct net_device *upper_dev,
7025 void *private, bool master)
7027 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7028 &dev->adj_list.upper,
7029 &upper_dev->adj_list.lower,
7033 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7034 struct net_device *upper_dev)
7036 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7037 &dev->adj_list.upper,
7038 &upper_dev->adj_list.lower);
7041 static int __netdev_upper_dev_link(struct net_device *dev,
7042 struct net_device *upper_dev, bool master,
7043 void *upper_priv, void *upper_info,
7044 struct netlink_ext_ack *extack)
7046 struct netdev_notifier_changeupper_info changeupper_info = {
7051 .upper_dev = upper_dev,
7054 .upper_info = upper_info,
7056 struct net_device *master_dev;
7061 if (dev == upper_dev)
7064 /* To prevent loops, check if dev is not upper device to upper_dev. */
7065 if (netdev_has_upper_dev(upper_dev, dev))
7069 if (netdev_has_upper_dev(dev, upper_dev))
7072 master_dev = netdev_master_upper_dev_get(dev);
7074 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7077 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7078 &changeupper_info.info);
7079 ret = notifier_to_errno(ret);
7083 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7088 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7089 &changeupper_info.info);
7090 ret = notifier_to_errno(ret);
7097 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7103 * netdev_upper_dev_link - Add a link to the upper device
7105 * @upper_dev: new upper device
7106 * @extack: netlink extended ack
7108 * Adds a link to device which is upper to this one. The caller must hold
7109 * the RTNL lock. On a failure a negative errno code is returned.
7110 * On success the reference counts are adjusted and the function
7113 int netdev_upper_dev_link(struct net_device *dev,
7114 struct net_device *upper_dev,
7115 struct netlink_ext_ack *extack)
7117 return __netdev_upper_dev_link(dev, upper_dev, false,
7118 NULL, NULL, extack);
7120 EXPORT_SYMBOL(netdev_upper_dev_link);
7123 * netdev_master_upper_dev_link - Add a master link to the upper device
7125 * @upper_dev: new upper device
7126 * @upper_priv: upper device private
7127 * @upper_info: upper info to be passed down via notifier
7128 * @extack: netlink extended ack
7130 * Adds a link to device which is upper to this one. In this case, only
7131 * one master upper device can be linked, although other non-master devices
7132 * might be linked as well. The caller must hold the RTNL lock.
7133 * On a failure a negative errno code is returned. On success the reference
7134 * counts are adjusted and the function returns zero.
7136 int netdev_master_upper_dev_link(struct net_device *dev,
7137 struct net_device *upper_dev,
7138 void *upper_priv, void *upper_info,
7139 struct netlink_ext_ack *extack)
7141 return __netdev_upper_dev_link(dev, upper_dev, true,
7142 upper_priv, upper_info, extack);
7144 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7147 * netdev_upper_dev_unlink - Removes a link to upper device
7149 * @upper_dev: new upper device
7151 * Removes a link to device which is upper to this one. The caller must hold
7154 void netdev_upper_dev_unlink(struct net_device *dev,
7155 struct net_device *upper_dev)
7157 struct netdev_notifier_changeupper_info changeupper_info = {
7161 .upper_dev = upper_dev,
7167 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7169 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7170 &changeupper_info.info);
7172 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7174 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7175 &changeupper_info.info);
7177 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7180 * netdev_bonding_info_change - Dispatch event about slave change
7182 * @bonding_info: info to dispatch
7184 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7185 * The caller must hold the RTNL lock.
7187 void netdev_bonding_info_change(struct net_device *dev,
7188 struct netdev_bonding_info *bonding_info)
7190 struct netdev_notifier_bonding_info info = {
7194 memcpy(&info.bonding_info, bonding_info,
7195 sizeof(struct netdev_bonding_info));
7196 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7199 EXPORT_SYMBOL(netdev_bonding_info_change);
7201 static void netdev_adjacent_add_links(struct net_device *dev)
7203 struct netdev_adjacent *iter;
7205 struct net *net = dev_net(dev);
7207 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7208 if (!net_eq(net, dev_net(iter->dev)))
7210 netdev_adjacent_sysfs_add(iter->dev, dev,
7211 &iter->dev->adj_list.lower);
7212 netdev_adjacent_sysfs_add(dev, iter->dev,
7213 &dev->adj_list.upper);
7216 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7217 if (!net_eq(net, dev_net(iter->dev)))
7219 netdev_adjacent_sysfs_add(iter->dev, dev,
7220 &iter->dev->adj_list.upper);
7221 netdev_adjacent_sysfs_add(dev, iter->dev,
7222 &dev->adj_list.lower);
7226 static void netdev_adjacent_del_links(struct net_device *dev)
7228 struct netdev_adjacent *iter;
7230 struct net *net = dev_net(dev);
7232 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7233 if (!net_eq(net, dev_net(iter->dev)))
7235 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7236 &iter->dev->adj_list.lower);
7237 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7238 &dev->adj_list.upper);
7241 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7242 if (!net_eq(net, dev_net(iter->dev)))
7244 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7245 &iter->dev->adj_list.upper);
7246 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7247 &dev->adj_list.lower);
7251 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7253 struct netdev_adjacent *iter;
7255 struct net *net = dev_net(dev);
7257 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7258 if (!net_eq(net, dev_net(iter->dev)))
7260 netdev_adjacent_sysfs_del(iter->dev, oldname,
7261 &iter->dev->adj_list.lower);
7262 netdev_adjacent_sysfs_add(iter->dev, dev,
7263 &iter->dev->adj_list.lower);
7266 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7267 if (!net_eq(net, dev_net(iter->dev)))
7269 netdev_adjacent_sysfs_del(iter->dev, oldname,
7270 &iter->dev->adj_list.upper);
7271 netdev_adjacent_sysfs_add(iter->dev, dev,
7272 &iter->dev->adj_list.upper);
7276 void *netdev_lower_dev_get_private(struct net_device *dev,
7277 struct net_device *lower_dev)
7279 struct netdev_adjacent *lower;
7283 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7287 return lower->private;
7289 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7292 int dev_get_nest_level(struct net_device *dev)
7294 struct net_device *lower = NULL;
7295 struct list_head *iter;
7301 netdev_for_each_lower_dev(dev, lower, iter) {
7302 nest = dev_get_nest_level(lower);
7303 if (max_nest < nest)
7307 return max_nest + 1;
7309 EXPORT_SYMBOL(dev_get_nest_level);
7312 * netdev_lower_change - Dispatch event about lower device state change
7313 * @lower_dev: device
7314 * @lower_state_info: state to dispatch
7316 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7317 * The caller must hold the RTNL lock.
7319 void netdev_lower_state_changed(struct net_device *lower_dev,
7320 void *lower_state_info)
7322 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7323 .info.dev = lower_dev,
7327 changelowerstate_info.lower_state_info = lower_state_info;
7328 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7329 &changelowerstate_info.info);
7331 EXPORT_SYMBOL(netdev_lower_state_changed);
7333 static void dev_change_rx_flags(struct net_device *dev, int flags)
7335 const struct net_device_ops *ops = dev->netdev_ops;
7337 if (ops->ndo_change_rx_flags)
7338 ops->ndo_change_rx_flags(dev, flags);
7341 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7343 unsigned int old_flags = dev->flags;
7349 dev->flags |= IFF_PROMISC;
7350 dev->promiscuity += inc;
7351 if (dev->promiscuity == 0) {
7354 * If inc causes overflow, untouch promisc and return error.
7357 dev->flags &= ~IFF_PROMISC;
7359 dev->promiscuity -= inc;
7360 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7365 if (dev->flags != old_flags) {
7366 pr_info("device %s %s promiscuous mode\n",
7368 dev->flags & IFF_PROMISC ? "entered" : "left");
7369 if (audit_enabled) {
7370 current_uid_gid(&uid, &gid);
7371 audit_log(audit_context(), GFP_ATOMIC,
7372 AUDIT_ANOM_PROMISCUOUS,
7373 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7374 dev->name, (dev->flags & IFF_PROMISC),
7375 (old_flags & IFF_PROMISC),
7376 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7377 from_kuid(&init_user_ns, uid),
7378 from_kgid(&init_user_ns, gid),
7379 audit_get_sessionid(current));
7382 dev_change_rx_flags(dev, IFF_PROMISC);
7385 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7390 * dev_set_promiscuity - update promiscuity count on a device
7394 * Add or remove promiscuity from a device. While the count in the device
7395 * remains above zero the interface remains promiscuous. Once it hits zero
7396 * the device reverts back to normal filtering operation. A negative inc
7397 * value is used to drop promiscuity on the device.
7398 * Return 0 if successful or a negative errno code on error.
7400 int dev_set_promiscuity(struct net_device *dev, int inc)
7402 unsigned int old_flags = dev->flags;
7405 err = __dev_set_promiscuity(dev, inc, true);
7408 if (dev->flags != old_flags)
7409 dev_set_rx_mode(dev);
7412 EXPORT_SYMBOL(dev_set_promiscuity);
7414 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7416 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7420 dev->flags |= IFF_ALLMULTI;
7421 dev->allmulti += inc;
7422 if (dev->allmulti == 0) {
7425 * If inc causes overflow, untouch allmulti and return error.
7428 dev->flags &= ~IFF_ALLMULTI;
7430 dev->allmulti -= inc;
7431 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7436 if (dev->flags ^ old_flags) {
7437 dev_change_rx_flags(dev, IFF_ALLMULTI);
7438 dev_set_rx_mode(dev);
7440 __dev_notify_flags(dev, old_flags,
7441 dev->gflags ^ old_gflags);
7447 * dev_set_allmulti - update allmulti count on a device
7451 * Add or remove reception of all multicast frames to a device. While the
7452 * count in the device remains above zero the interface remains listening
7453 * to all interfaces. Once it hits zero the device reverts back to normal
7454 * filtering operation. A negative @inc value is used to drop the counter
7455 * when releasing a resource needing all multicasts.
7456 * Return 0 if successful or a negative errno code on error.
7459 int dev_set_allmulti(struct net_device *dev, int inc)
7461 return __dev_set_allmulti(dev, inc, true);
7463 EXPORT_SYMBOL(dev_set_allmulti);
7466 * Upload unicast and multicast address lists to device and
7467 * configure RX filtering. When the device doesn't support unicast
7468 * filtering it is put in promiscuous mode while unicast addresses
7471 void __dev_set_rx_mode(struct net_device *dev)
7473 const struct net_device_ops *ops = dev->netdev_ops;
7475 /* dev_open will call this function so the list will stay sane. */
7476 if (!(dev->flags&IFF_UP))
7479 if (!netif_device_present(dev))
7482 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7483 /* Unicast addresses changes may only happen under the rtnl,
7484 * therefore calling __dev_set_promiscuity here is safe.
7486 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7487 __dev_set_promiscuity(dev, 1, false);
7488 dev->uc_promisc = true;
7489 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7490 __dev_set_promiscuity(dev, -1, false);
7491 dev->uc_promisc = false;
7495 if (ops->ndo_set_rx_mode)
7496 ops->ndo_set_rx_mode(dev);
7499 void dev_set_rx_mode(struct net_device *dev)
7501 netif_addr_lock_bh(dev);
7502 __dev_set_rx_mode(dev);
7503 netif_addr_unlock_bh(dev);
7507 * dev_get_flags - get flags reported to userspace
7510 * Get the combination of flag bits exported through APIs to userspace.
7512 unsigned int dev_get_flags(const struct net_device *dev)
7516 flags = (dev->flags & ~(IFF_PROMISC |
7521 (dev->gflags & (IFF_PROMISC |
7524 if (netif_running(dev)) {
7525 if (netif_oper_up(dev))
7526 flags |= IFF_RUNNING;
7527 if (netif_carrier_ok(dev))
7528 flags |= IFF_LOWER_UP;
7529 if (netif_dormant(dev))
7530 flags |= IFF_DORMANT;
7535 EXPORT_SYMBOL(dev_get_flags);
7537 int __dev_change_flags(struct net_device *dev, unsigned int flags,
7538 struct netlink_ext_ack *extack)
7540 unsigned int old_flags = dev->flags;
7546 * Set the flags on our device.
7549 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7550 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7552 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7556 * Load in the correct multicast list now the flags have changed.
7559 if ((old_flags ^ flags) & IFF_MULTICAST)
7560 dev_change_rx_flags(dev, IFF_MULTICAST);
7562 dev_set_rx_mode(dev);
7565 * Have we downed the interface. We handle IFF_UP ourselves
7566 * according to user attempts to set it, rather than blindly
7571 if ((old_flags ^ flags) & IFF_UP) {
7572 if (old_flags & IFF_UP)
7575 ret = __dev_open(dev, extack);
7578 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7579 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7580 unsigned int old_flags = dev->flags;
7582 dev->gflags ^= IFF_PROMISC;
7584 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7585 if (dev->flags != old_flags)
7586 dev_set_rx_mode(dev);
7589 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7590 * is important. Some (broken) drivers set IFF_PROMISC, when
7591 * IFF_ALLMULTI is requested not asking us and not reporting.
7593 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7594 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7596 dev->gflags ^= IFF_ALLMULTI;
7597 __dev_set_allmulti(dev, inc, false);
7603 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7604 unsigned int gchanges)
7606 unsigned int changes = dev->flags ^ old_flags;
7609 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7611 if (changes & IFF_UP) {
7612 if (dev->flags & IFF_UP)
7613 call_netdevice_notifiers(NETDEV_UP, dev);
7615 call_netdevice_notifiers(NETDEV_DOWN, dev);
7618 if (dev->flags & IFF_UP &&
7619 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7620 struct netdev_notifier_change_info change_info = {
7624 .flags_changed = changes,
7627 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7632 * dev_change_flags - change device settings
7634 * @flags: device state flags
7635 * @extack: netlink extended ack
7637 * Change settings on device based state flags. The flags are
7638 * in the userspace exported format.
7640 int dev_change_flags(struct net_device *dev, unsigned int flags,
7641 struct netlink_ext_ack *extack)
7644 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7646 ret = __dev_change_flags(dev, flags, extack);
7650 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7651 __dev_notify_flags(dev, old_flags, changes);
7654 EXPORT_SYMBOL(dev_change_flags);
7656 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7658 const struct net_device_ops *ops = dev->netdev_ops;
7660 if (ops->ndo_change_mtu)
7661 return ops->ndo_change_mtu(dev, new_mtu);
7666 EXPORT_SYMBOL(__dev_set_mtu);
7669 * dev_set_mtu_ext - Change maximum transfer unit
7671 * @new_mtu: new transfer unit
7672 * @extack: netlink extended ack
7674 * Change the maximum transfer size of the network device.
7676 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
7677 struct netlink_ext_ack *extack)
7681 if (new_mtu == dev->mtu)
7684 /* MTU must be positive, and in range */
7685 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7686 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
7690 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7691 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
7695 if (!netif_device_present(dev))
7698 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7699 err = notifier_to_errno(err);
7703 orig_mtu = dev->mtu;
7704 err = __dev_set_mtu(dev, new_mtu);
7707 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7709 err = notifier_to_errno(err);
7711 /* setting mtu back and notifying everyone again,
7712 * so that they have a chance to revert changes.
7714 __dev_set_mtu(dev, orig_mtu);
7715 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
7722 int dev_set_mtu(struct net_device *dev, int new_mtu)
7724 struct netlink_ext_ack extack;
7727 memset(&extack, 0, sizeof(extack));
7728 err = dev_set_mtu_ext(dev, new_mtu, &extack);
7729 if (err && extack._msg)
7730 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
7733 EXPORT_SYMBOL(dev_set_mtu);
7736 * dev_change_tx_queue_len - Change TX queue length of a netdevice
7738 * @new_len: new tx queue length
7740 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7742 unsigned int orig_len = dev->tx_queue_len;
7745 if (new_len != (unsigned int)new_len)
7748 if (new_len != orig_len) {
7749 dev->tx_queue_len = new_len;
7750 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7751 res = notifier_to_errno(res);
7754 res = dev_qdisc_change_tx_queue_len(dev);
7762 netdev_err(dev, "refused to change device tx_queue_len\n");
7763 dev->tx_queue_len = orig_len;
7768 * dev_set_group - Change group this device belongs to
7770 * @new_group: group this device should belong to
7772 void dev_set_group(struct net_device *dev, int new_group)
7774 dev->group = new_group;
7776 EXPORT_SYMBOL(dev_set_group);
7779 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
7781 * @addr: new address
7782 * @extack: netlink extended ack
7784 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
7785 struct netlink_ext_ack *extack)
7787 struct netdev_notifier_pre_changeaddr_info info = {
7789 .info.extack = extack,
7794 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
7795 return notifier_to_errno(rc);
7797 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
7800 * dev_set_mac_address - Change Media Access Control Address
7803 * @extack: netlink extended ack
7805 * Change the hardware (MAC) address of the device
7807 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
7808 struct netlink_ext_ack *extack)
7810 const struct net_device_ops *ops = dev->netdev_ops;
7813 if (!ops->ndo_set_mac_address)
7815 if (sa->sa_family != dev->type)
7817 if (!netif_device_present(dev))
7819 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
7822 err = ops->ndo_set_mac_address(dev, sa);
7825 dev->addr_assign_type = NET_ADDR_SET;
7826 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7827 add_device_randomness(dev->dev_addr, dev->addr_len);
7830 EXPORT_SYMBOL(dev_set_mac_address);
7833 * dev_change_carrier - Change device carrier
7835 * @new_carrier: new value
7837 * Change device carrier
7839 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7841 const struct net_device_ops *ops = dev->netdev_ops;
7843 if (!ops->ndo_change_carrier)
7845 if (!netif_device_present(dev))
7847 return ops->ndo_change_carrier(dev, new_carrier);
7849 EXPORT_SYMBOL(dev_change_carrier);
7852 * dev_get_phys_port_id - Get device physical port ID
7856 * Get device physical port ID
7858 int dev_get_phys_port_id(struct net_device *dev,
7859 struct netdev_phys_item_id *ppid)
7861 const struct net_device_ops *ops = dev->netdev_ops;
7863 if (!ops->ndo_get_phys_port_id)
7865 return ops->ndo_get_phys_port_id(dev, ppid);
7867 EXPORT_SYMBOL(dev_get_phys_port_id);
7870 * dev_get_phys_port_name - Get device physical port name
7873 * @len: limit of bytes to copy to name
7875 * Get device physical port name
7877 int dev_get_phys_port_name(struct net_device *dev,
7878 char *name, size_t len)
7880 const struct net_device_ops *ops = dev->netdev_ops;
7883 if (ops->ndo_get_phys_port_name) {
7884 err = ops->ndo_get_phys_port_name(dev, name, len);
7885 if (err != -EOPNOTSUPP)
7888 return devlink_compat_phys_port_name_get(dev, name, len);
7890 EXPORT_SYMBOL(dev_get_phys_port_name);
7893 * dev_get_port_parent_id - Get the device's port parent identifier
7894 * @dev: network device
7895 * @ppid: pointer to a storage for the port's parent identifier
7896 * @recurse: allow/disallow recursion to lower devices
7898 * Get the devices's port parent identifier
7900 int dev_get_port_parent_id(struct net_device *dev,
7901 struct netdev_phys_item_id *ppid,
7904 const struct net_device_ops *ops = dev->netdev_ops;
7905 struct netdev_phys_item_id first = { };
7906 struct net_device *lower_dev;
7907 struct list_head *iter;
7908 int err = -EOPNOTSUPP;
7910 if (ops->ndo_get_port_parent_id)
7911 return ops->ndo_get_port_parent_id(dev, ppid);
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 (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);