1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #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/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/static_key.h>
137 #include <linux/hashtable.h>
138 #include <linux/vmalloc.h>
139 #include <linux/if_macvlan.h>
140 #include <linux/errqueue.h>
141 #include <linux/hrtimer.h>
142 #include <linux/netfilter_ingress.h>
143 #include <linux/crash_dump.h>
144 #include <linux/sctp.h>
145 #include <net/udp_tunnel.h>
146 #include <linux/net_namespace.h>
147 #include <linux/indirect_call_wrapper.h>
148 #include <net/devlink.h>
149 #include <linux/pm_runtime.h>
150 #include <linux/prandom.h>
152 #include "net-sysfs.h"
154 #define MAX_GRO_SKBS 8
156 /* This should be increased if a protocol with a bigger head is added. */
157 #define GRO_MAX_HEAD (MAX_HEADER + 128)
159 static DEFINE_SPINLOCK(ptype_lock);
160 static DEFINE_SPINLOCK(offload_lock);
161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
162 struct list_head ptype_all __read_mostly; /* Taps */
163 static struct list_head offload_base __read_mostly;
165 static int netif_rx_internal(struct sk_buff *skb);
166 static int call_netdevice_notifiers_info(unsigned long val,
167 struct netdev_notifier_info *info);
168 static int call_netdevice_notifiers_extack(unsigned long val,
169 struct net_device *dev,
170 struct netlink_ext_ack *extack);
171 static struct napi_struct *napi_by_id(unsigned int napi_id);
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
192 DEFINE_RWLOCK(dev_base_lock);
193 EXPORT_SYMBOL(dev_base_lock);
195 static DEFINE_MUTEX(ifalias_mutex);
197 /* protects napi_hash addition/deletion and napi_gen_id */
198 static DEFINE_SPINLOCK(napi_hash_lock);
200 static unsigned int napi_gen_id = NR_CPUS;
201 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
203 static DECLARE_RWSEM(devnet_rename_sem);
205 static inline void dev_base_seq_inc(struct net *net)
207 while (++net->dev_base_seq == 0)
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 static inline void rps_lock(struct softnet_data *sd)
226 spin_lock(&sd->input_pkt_queue.lock);
230 static inline void rps_unlock(struct softnet_data *sd)
233 spin_unlock(&sd->input_pkt_queue.lock);
237 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
240 struct netdev_name_node *name_node;
242 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
245 INIT_HLIST_NODE(&name_node->hlist);
246 name_node->dev = dev;
247 name_node->name = name;
251 static struct netdev_name_node *
252 netdev_name_node_head_alloc(struct net_device *dev)
254 struct netdev_name_node *name_node;
256 name_node = netdev_name_node_alloc(dev, dev->name);
259 INIT_LIST_HEAD(&name_node->list);
263 static void netdev_name_node_free(struct netdev_name_node *name_node)
268 static void netdev_name_node_add(struct net *net,
269 struct netdev_name_node *name_node)
271 hlist_add_head_rcu(&name_node->hlist,
272 dev_name_hash(net, name_node->name));
275 static void netdev_name_node_del(struct netdev_name_node *name_node)
277 hlist_del_rcu(&name_node->hlist);
280 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
283 struct hlist_head *head = dev_name_hash(net, name);
284 struct netdev_name_node *name_node;
286 hlist_for_each_entry(name_node, head, hlist)
287 if (!strcmp(name_node->name, name))
292 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
295 struct hlist_head *head = dev_name_hash(net, name);
296 struct netdev_name_node *name_node;
298 hlist_for_each_entry_rcu(name_node, head, hlist)
299 if (!strcmp(name_node->name, name))
304 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
306 struct netdev_name_node *name_node;
307 struct net *net = dev_net(dev);
309 name_node = netdev_name_node_lookup(net, name);
312 name_node = netdev_name_node_alloc(dev, name);
315 netdev_name_node_add(net, name_node);
316 /* The node that holds dev->name acts as a head of per-device list. */
317 list_add_tail(&name_node->list, &dev->name_node->list);
321 EXPORT_SYMBOL(netdev_name_node_alt_create);
323 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
325 list_del(&name_node->list);
326 netdev_name_node_del(name_node);
327 kfree(name_node->name);
328 netdev_name_node_free(name_node);
331 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
333 struct netdev_name_node *name_node;
334 struct net *net = dev_net(dev);
336 name_node = netdev_name_node_lookup(net, name);
339 /* lookup might have found our primary name or a name belonging
342 if (name_node == dev->name_node || name_node->dev != dev)
345 __netdev_name_node_alt_destroy(name_node);
349 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
351 static void netdev_name_node_alt_flush(struct net_device *dev)
353 struct netdev_name_node *name_node, *tmp;
355 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
356 __netdev_name_node_alt_destroy(name_node);
359 /* Device list insertion */
360 static void list_netdevice(struct net_device *dev)
362 struct net *net = dev_net(dev);
366 write_lock_bh(&dev_base_lock);
367 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
368 netdev_name_node_add(net, dev->name_node);
369 hlist_add_head_rcu(&dev->index_hlist,
370 dev_index_hash(net, dev->ifindex));
371 write_unlock_bh(&dev_base_lock);
373 dev_base_seq_inc(net);
376 /* Device list removal
377 * caller must respect a RCU grace period before freeing/reusing dev
379 static void unlist_netdevice(struct net_device *dev)
383 /* Unlink dev from the device chain */
384 write_lock_bh(&dev_base_lock);
385 list_del_rcu(&dev->dev_list);
386 netdev_name_node_del(dev->name_node);
387 hlist_del_rcu(&dev->index_hlist);
388 write_unlock_bh(&dev_base_lock);
390 dev_base_seq_inc(dev_net(dev));
397 static RAW_NOTIFIER_HEAD(netdev_chain);
400 * Device drivers call our routines to queue packets here. We empty the
401 * queue in the local softnet handler.
404 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
405 EXPORT_PER_CPU_SYMBOL(softnet_data);
407 #ifdef CONFIG_LOCKDEP
409 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
410 * according to dev->type
412 static const unsigned short netdev_lock_type[] = {
413 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
414 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
415 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
416 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
417 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
418 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
419 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
420 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
421 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
422 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
423 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
424 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
425 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
426 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
427 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
429 static const char *const netdev_lock_name[] = {
430 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
431 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
432 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
433 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
434 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
435 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
436 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
437 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
438 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
439 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
440 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
441 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
442 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
443 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
444 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
446 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
447 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
449 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
453 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
454 if (netdev_lock_type[i] == dev_type)
456 /* the last key is used by default */
457 return ARRAY_SIZE(netdev_lock_type) - 1;
460 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
461 unsigned short dev_type)
465 i = netdev_lock_pos(dev_type);
466 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
467 netdev_lock_name[i]);
470 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
474 i = netdev_lock_pos(dev->type);
475 lockdep_set_class_and_name(&dev->addr_list_lock,
476 &netdev_addr_lock_key[i],
477 netdev_lock_name[i]);
480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
481 unsigned short dev_type)
485 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
490 /*******************************************************************************
492 * Protocol management and registration routines
494 *******************************************************************************/
498 * Add a protocol ID to the list. Now that the input handler is
499 * smarter we can dispense with all the messy stuff that used to be
502 * BEWARE!!! Protocol handlers, mangling input packets,
503 * MUST BE last in hash buckets and checking protocol handlers
504 * MUST start from promiscuous ptype_all chain in net_bh.
505 * It is true now, do not change it.
506 * Explanation follows: if protocol handler, mangling packet, will
507 * be the first on list, it is not able to sense, that packet
508 * is cloned and should be copied-on-write, so that it will
509 * change it and subsequent readers will get broken packet.
513 static inline struct list_head *ptype_head(const struct packet_type *pt)
515 if (pt->type == htons(ETH_P_ALL))
516 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
518 return pt->dev ? &pt->dev->ptype_specific :
519 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
523 * dev_add_pack - add packet handler
524 * @pt: packet type declaration
526 * Add a protocol handler to the networking stack. The passed &packet_type
527 * is linked into kernel lists and may not be freed until it has been
528 * removed from the kernel lists.
530 * This call does not sleep therefore it can not
531 * guarantee all CPU's that are in middle of receiving packets
532 * will see the new packet type (until the next received packet).
535 void dev_add_pack(struct packet_type *pt)
537 struct list_head *head = ptype_head(pt);
539 spin_lock(&ptype_lock);
540 list_add_rcu(&pt->list, head);
541 spin_unlock(&ptype_lock);
543 EXPORT_SYMBOL(dev_add_pack);
546 * __dev_remove_pack - remove packet handler
547 * @pt: packet type declaration
549 * Remove a protocol handler that was previously added to the kernel
550 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
551 * from the kernel lists and can be freed or reused once this function
554 * The packet type might still be in use by receivers
555 * and must not be freed until after all the CPU's have gone
556 * through a quiescent state.
558 void __dev_remove_pack(struct packet_type *pt)
560 struct list_head *head = ptype_head(pt);
561 struct packet_type *pt1;
563 spin_lock(&ptype_lock);
565 list_for_each_entry(pt1, head, list) {
567 list_del_rcu(&pt->list);
572 pr_warn("dev_remove_pack: %p not found\n", pt);
574 spin_unlock(&ptype_lock);
576 EXPORT_SYMBOL(__dev_remove_pack);
579 * dev_remove_pack - remove packet handler
580 * @pt: packet type declaration
582 * Remove a protocol handler that was previously added to the kernel
583 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
584 * from the kernel lists and can be freed or reused once this function
587 * This call sleeps to guarantee that no CPU is looking at the packet
590 void dev_remove_pack(struct packet_type *pt)
592 __dev_remove_pack(pt);
596 EXPORT_SYMBOL(dev_remove_pack);
600 * dev_add_offload - register offload handlers
601 * @po: protocol offload declaration
603 * Add protocol offload handlers to the networking stack. The passed
604 * &proto_offload is linked into kernel lists and may not be freed until
605 * it has been removed from the kernel lists.
607 * This call does not sleep therefore it can not
608 * guarantee all CPU's that are in middle of receiving packets
609 * will see the new offload handlers (until the next received packet).
611 void dev_add_offload(struct packet_offload *po)
613 struct packet_offload *elem;
615 spin_lock(&offload_lock);
616 list_for_each_entry(elem, &offload_base, list) {
617 if (po->priority < elem->priority)
620 list_add_rcu(&po->list, elem->list.prev);
621 spin_unlock(&offload_lock);
623 EXPORT_SYMBOL(dev_add_offload);
626 * __dev_remove_offload - remove offload handler
627 * @po: packet offload declaration
629 * Remove a protocol offload handler that was previously added to the
630 * kernel offload handlers by dev_add_offload(). The passed &offload_type
631 * is removed from the kernel lists and can be freed or reused once this
634 * The packet type might still be in use by receivers
635 * and must not be freed until after all the CPU's have gone
636 * through a quiescent state.
638 static void __dev_remove_offload(struct packet_offload *po)
640 struct list_head *head = &offload_base;
641 struct packet_offload *po1;
643 spin_lock(&offload_lock);
645 list_for_each_entry(po1, head, list) {
647 list_del_rcu(&po->list);
652 pr_warn("dev_remove_offload: %p not found\n", po);
654 spin_unlock(&offload_lock);
658 * dev_remove_offload - remove packet offload handler
659 * @po: packet offload declaration
661 * Remove a packet offload handler that was previously added to the kernel
662 * offload handlers by dev_add_offload(). The passed &offload_type is
663 * removed from the kernel lists and can be freed or reused once this
666 * This call sleeps to guarantee that no CPU is looking at the packet
669 void dev_remove_offload(struct packet_offload *po)
671 __dev_remove_offload(po);
675 EXPORT_SYMBOL(dev_remove_offload);
677 /******************************************************************************
679 * Device Boot-time Settings Routines
681 ******************************************************************************/
683 /* Boot time configuration table */
684 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
687 * netdev_boot_setup_add - add new setup entry
688 * @name: name of the device
689 * @map: configured settings for the device
691 * Adds new setup entry to the dev_boot_setup list. The function
692 * returns 0 on error and 1 on success. This is a generic routine to
695 static int netdev_boot_setup_add(char *name, struct ifmap *map)
697 struct netdev_boot_setup *s;
701 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
702 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
703 memset(s[i].name, 0, sizeof(s[i].name));
704 strlcpy(s[i].name, name, IFNAMSIZ);
705 memcpy(&s[i].map, map, sizeof(s[i].map));
710 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
714 * netdev_boot_setup_check - check boot time settings
715 * @dev: the netdevice
717 * Check boot time settings for the device.
718 * The found settings are set for the device to be used
719 * later in the device probing.
720 * Returns 0 if no settings found, 1 if they are.
722 int netdev_boot_setup_check(struct net_device *dev)
724 struct netdev_boot_setup *s = dev_boot_setup;
727 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
728 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
729 !strcmp(dev->name, s[i].name)) {
730 dev->irq = s[i].map.irq;
731 dev->base_addr = s[i].map.base_addr;
732 dev->mem_start = s[i].map.mem_start;
733 dev->mem_end = s[i].map.mem_end;
739 EXPORT_SYMBOL(netdev_boot_setup_check);
743 * netdev_boot_base - get address from boot time settings
744 * @prefix: prefix for network device
745 * @unit: id for network device
747 * Check boot time settings for the base address of device.
748 * The found settings are set for the device to be used
749 * later in the device probing.
750 * Returns 0 if no settings found.
752 unsigned long netdev_boot_base(const char *prefix, int unit)
754 const struct netdev_boot_setup *s = dev_boot_setup;
758 sprintf(name, "%s%d", prefix, unit);
761 * If device already registered then return base of 1
762 * to indicate not to probe for this interface
764 if (__dev_get_by_name(&init_net, name))
767 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
768 if (!strcmp(name, s[i].name))
769 return s[i].map.base_addr;
774 * Saves at boot time configured settings for any netdevice.
776 int __init netdev_boot_setup(char *str)
781 str = get_options(str, ARRAY_SIZE(ints), ints);
786 memset(&map, 0, sizeof(map));
790 map.base_addr = ints[2];
792 map.mem_start = ints[3];
794 map.mem_end = ints[4];
796 /* Add new entry to the list */
797 return netdev_boot_setup_add(str, &map);
800 __setup("netdev=", netdev_boot_setup);
802 /*******************************************************************************
804 * Device Interface Subroutines
806 *******************************************************************************/
809 * dev_get_iflink - get 'iflink' value of a interface
810 * @dev: targeted interface
812 * Indicates the ifindex the interface is linked to.
813 * Physical interfaces have the same 'ifindex' and 'iflink' values.
816 int dev_get_iflink(const struct net_device *dev)
818 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
819 return dev->netdev_ops->ndo_get_iflink(dev);
823 EXPORT_SYMBOL(dev_get_iflink);
826 * dev_fill_metadata_dst - Retrieve tunnel egress information.
827 * @dev: targeted interface
830 * For better visibility of tunnel traffic OVS needs to retrieve
831 * egress tunnel information for a packet. Following API allows
832 * user to get this info.
834 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
836 struct ip_tunnel_info *info;
838 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
841 info = skb_tunnel_info_unclone(skb);
844 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
847 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
849 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
852 * __dev_get_by_name - find a device by its name
853 * @net: the applicable net namespace
854 * @name: name to find
856 * Find an interface by name. Must be called under RTNL semaphore
857 * or @dev_base_lock. If the name is found a pointer to the device
858 * is returned. If the name is not found then %NULL is returned. The
859 * reference counters are not incremented so the caller must be
860 * careful with locks.
863 struct net_device *__dev_get_by_name(struct net *net, const char *name)
865 struct netdev_name_node *node_name;
867 node_name = netdev_name_node_lookup(net, name);
868 return node_name ? node_name->dev : NULL;
870 EXPORT_SYMBOL(__dev_get_by_name);
873 * dev_get_by_name_rcu - find a device by its name
874 * @net: the applicable net namespace
875 * @name: name to find
877 * Find an interface by name.
878 * If the name is found a pointer to the device is returned.
879 * If the name is not found then %NULL is returned.
880 * The reference counters are not incremented so the caller must be
881 * careful with locks. The caller must hold RCU lock.
884 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
886 struct netdev_name_node *node_name;
888 node_name = netdev_name_node_lookup_rcu(net, name);
889 return node_name ? node_name->dev : NULL;
891 EXPORT_SYMBOL(dev_get_by_name_rcu);
894 * dev_get_by_name - find a device by its name
895 * @net: the applicable net namespace
896 * @name: name to find
898 * Find an interface by name. This can be called from any
899 * context and does its own locking. The returned handle has
900 * the usage count incremented and the caller must use dev_put() to
901 * release it when it is no longer needed. %NULL is returned if no
902 * matching device is found.
905 struct net_device *dev_get_by_name(struct net *net, const char *name)
907 struct net_device *dev;
910 dev = dev_get_by_name_rcu(net, name);
916 EXPORT_SYMBOL(dev_get_by_name);
919 * __dev_get_by_index - find a device by its ifindex
920 * @net: the applicable net namespace
921 * @ifindex: index of device
923 * Search for an interface by index. Returns %NULL if the device
924 * is not found or a pointer to the device. The device has not
925 * had its reference counter increased so the caller must be careful
926 * about locking. The caller must hold either the RTNL semaphore
930 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
932 struct net_device *dev;
933 struct hlist_head *head = dev_index_hash(net, ifindex);
935 hlist_for_each_entry(dev, head, index_hlist)
936 if (dev->ifindex == ifindex)
941 EXPORT_SYMBOL(__dev_get_by_index);
944 * dev_get_by_index_rcu - find a device by its ifindex
945 * @net: the applicable net namespace
946 * @ifindex: index of device
948 * Search for an interface by index. Returns %NULL if the device
949 * is not found or a pointer to the device. The device has not
950 * had its reference counter increased so the caller must be careful
951 * about locking. The caller must hold RCU lock.
954 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
956 struct net_device *dev;
957 struct hlist_head *head = dev_index_hash(net, ifindex);
959 hlist_for_each_entry_rcu(dev, head, index_hlist)
960 if (dev->ifindex == ifindex)
965 EXPORT_SYMBOL(dev_get_by_index_rcu);
969 * dev_get_by_index - find a device by its ifindex
970 * @net: the applicable net namespace
971 * @ifindex: index of device
973 * Search for an interface by index. Returns NULL if the device
974 * is not found or a pointer to the device. The device returned has
975 * had a reference added and the pointer is safe until the user calls
976 * dev_put to indicate they have finished with it.
979 struct net_device *dev_get_by_index(struct net *net, int ifindex)
981 struct net_device *dev;
984 dev = dev_get_by_index_rcu(net, ifindex);
990 EXPORT_SYMBOL(dev_get_by_index);
993 * dev_get_by_napi_id - find a device by napi_id
994 * @napi_id: ID of the NAPI struct
996 * Search for an interface by NAPI ID. Returns %NULL if the device
997 * is not found or a pointer to the device. The device has not had
998 * its reference counter increased so the caller must be careful
999 * about locking. The caller must hold RCU lock.
1002 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
1004 struct napi_struct *napi;
1006 WARN_ON_ONCE(!rcu_read_lock_held());
1008 if (napi_id < MIN_NAPI_ID)
1011 napi = napi_by_id(napi_id);
1013 return napi ? napi->dev : NULL;
1015 EXPORT_SYMBOL(dev_get_by_napi_id);
1018 * netdev_get_name - get a netdevice name, knowing its ifindex.
1019 * @net: network namespace
1020 * @name: a pointer to the buffer where the name will be stored.
1021 * @ifindex: the ifindex of the interface to get the name from.
1023 int netdev_get_name(struct net *net, char *name, int ifindex)
1025 struct net_device *dev;
1028 down_read(&devnet_rename_sem);
1031 dev = dev_get_by_index_rcu(net, ifindex);
1037 strcpy(name, dev->name);
1042 up_read(&devnet_rename_sem);
1047 * dev_getbyhwaddr_rcu - find a device by its hardware address
1048 * @net: the applicable net namespace
1049 * @type: media type of device
1050 * @ha: hardware address
1052 * Search for an interface by MAC address. Returns NULL if the device
1053 * is not found or a pointer to the device.
1054 * The caller must hold RCU or RTNL.
1055 * The returned device has not had its ref count increased
1056 * and the caller must therefore be careful about locking
1060 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1063 struct net_device *dev;
1065 for_each_netdev_rcu(net, dev)
1066 if (dev->type == type &&
1067 !memcmp(dev->dev_addr, ha, dev->addr_len))
1072 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1074 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1076 struct net_device *dev, *ret = NULL;
1079 for_each_netdev_rcu(net, dev)
1080 if (dev->type == type) {
1088 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1091 * __dev_get_by_flags - find any device with given flags
1092 * @net: the applicable net namespace
1093 * @if_flags: IFF_* values
1094 * @mask: bitmask of bits in if_flags to check
1096 * Search for any interface with the given flags. Returns NULL if a device
1097 * is not found or a pointer to the device. Must be called inside
1098 * rtnl_lock(), and result refcount is unchanged.
1101 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1102 unsigned short mask)
1104 struct net_device *dev, *ret;
1109 for_each_netdev(net, dev) {
1110 if (((dev->flags ^ if_flags) & mask) == 0) {
1117 EXPORT_SYMBOL(__dev_get_by_flags);
1120 * dev_valid_name - check if name is okay for network device
1121 * @name: name string
1123 * Network device names need to be valid file names to
1124 * allow sysfs to work. We also disallow any kind of
1127 bool dev_valid_name(const char *name)
1131 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1133 if (!strcmp(name, ".") || !strcmp(name, ".."))
1137 if (*name == '/' || *name == ':' || isspace(*name))
1143 EXPORT_SYMBOL(dev_valid_name);
1146 * __dev_alloc_name - allocate a name for a device
1147 * @net: network namespace to allocate the device name in
1148 * @name: name format string
1149 * @buf: scratch buffer and result name string
1151 * Passed a format string - eg "lt%d" it will try and find a suitable
1152 * id. It scans list of devices to build up a free map, then chooses
1153 * the first empty slot. The caller must hold the dev_base or rtnl lock
1154 * while allocating the name and adding the device in order to avoid
1156 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1157 * Returns the number of the unit assigned or a negative errno code.
1160 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1164 const int max_netdevices = 8*PAGE_SIZE;
1165 unsigned long *inuse;
1166 struct net_device *d;
1168 if (!dev_valid_name(name))
1171 p = strchr(name, '%');
1174 * Verify the string as this thing may have come from
1175 * the user. There must be either one "%d" and no other "%"
1178 if (p[1] != 'd' || strchr(p + 2, '%'))
1181 /* Use one page as a bit array of possible slots */
1182 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1186 for_each_netdev(net, d) {
1187 struct netdev_name_node *name_node;
1188 list_for_each_entry(name_node, &d->name_node->list, list) {
1189 if (!sscanf(name_node->name, name, &i))
1191 if (i < 0 || i >= max_netdevices)
1194 /* avoid cases where sscanf is not exact inverse of printf */
1195 snprintf(buf, IFNAMSIZ, name, i);
1196 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1199 if (!sscanf(d->name, name, &i))
1201 if (i < 0 || i >= max_netdevices)
1204 /* avoid cases where sscanf is not exact inverse of printf */
1205 snprintf(buf, IFNAMSIZ, name, i);
1206 if (!strncmp(buf, d->name, IFNAMSIZ))
1210 i = find_first_zero_bit(inuse, max_netdevices);
1211 free_page((unsigned long) inuse);
1214 snprintf(buf, IFNAMSIZ, name, i);
1215 if (!__dev_get_by_name(net, buf))
1218 /* It is possible to run out of possible slots
1219 * when the name is long and there isn't enough space left
1220 * for the digits, or if all bits are used.
1225 static int dev_alloc_name_ns(struct net *net,
1226 struct net_device *dev,
1233 ret = __dev_alloc_name(net, name, buf);
1235 strlcpy(dev->name, buf, IFNAMSIZ);
1240 * dev_alloc_name - allocate a name for a device
1242 * @name: name format string
1244 * Passed a format string - eg "lt%d" it will try and find a suitable
1245 * id. It scans list of devices to build up a free map, then chooses
1246 * the first empty slot. The caller must hold the dev_base or rtnl lock
1247 * while allocating the name and adding the device in order to avoid
1249 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1250 * Returns the number of the unit assigned or a negative errno code.
1253 int dev_alloc_name(struct net_device *dev, const char *name)
1255 return dev_alloc_name_ns(dev_net(dev), dev, name);
1257 EXPORT_SYMBOL(dev_alloc_name);
1259 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1264 if (!dev_valid_name(name))
1267 if (strchr(name, '%'))
1268 return dev_alloc_name_ns(net, dev, name);
1269 else if (__dev_get_by_name(net, name))
1271 else if (dev->name != name)
1272 strlcpy(dev->name, name, IFNAMSIZ);
1278 * dev_change_name - change name of a device
1280 * @newname: name (or format string) must be at least IFNAMSIZ
1282 * Change name of a device, can pass format strings "eth%d".
1285 int dev_change_name(struct net_device *dev, const char *newname)
1287 unsigned char old_assign_type;
1288 char oldname[IFNAMSIZ];
1294 BUG_ON(!dev_net(dev));
1298 /* Some auto-enslaved devices e.g. failover slaves are
1299 * special, as userspace might rename the device after
1300 * the interface had been brought up and running since
1301 * the point kernel initiated auto-enslavement. Allow
1302 * live name change even when these slave devices are
1305 * Typically, users of these auto-enslaving devices
1306 * don't actually care about slave name change, as
1307 * they are supposed to operate on master interface
1310 if (dev->flags & IFF_UP &&
1311 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1314 down_write(&devnet_rename_sem);
1316 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1317 up_write(&devnet_rename_sem);
1321 memcpy(oldname, dev->name, IFNAMSIZ);
1323 err = dev_get_valid_name(net, dev, newname);
1325 up_write(&devnet_rename_sem);
1329 if (oldname[0] && !strchr(oldname, '%'))
1330 netdev_info(dev, "renamed from %s\n", oldname);
1332 old_assign_type = dev->name_assign_type;
1333 dev->name_assign_type = NET_NAME_RENAMED;
1336 ret = device_rename(&dev->dev, dev->name);
1338 memcpy(dev->name, oldname, IFNAMSIZ);
1339 dev->name_assign_type = old_assign_type;
1340 up_write(&devnet_rename_sem);
1344 up_write(&devnet_rename_sem);
1346 netdev_adjacent_rename_links(dev, oldname);
1348 write_lock_bh(&dev_base_lock);
1349 netdev_name_node_del(dev->name_node);
1350 write_unlock_bh(&dev_base_lock);
1354 write_lock_bh(&dev_base_lock);
1355 netdev_name_node_add(net, dev->name_node);
1356 write_unlock_bh(&dev_base_lock);
1358 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1359 ret = notifier_to_errno(ret);
1362 /* err >= 0 after dev_alloc_name() or stores the first errno */
1365 down_write(&devnet_rename_sem);
1366 memcpy(dev->name, oldname, IFNAMSIZ);
1367 memcpy(oldname, newname, IFNAMSIZ);
1368 dev->name_assign_type = old_assign_type;
1369 old_assign_type = NET_NAME_RENAMED;
1372 pr_err("%s: name change rollback failed: %d\n",
1381 * dev_set_alias - change ifalias of a device
1383 * @alias: name up to IFALIASZ
1384 * @len: limit of bytes to copy from info
1386 * Set ifalias for a device,
1388 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1390 struct dev_ifalias *new_alias = NULL;
1392 if (len >= IFALIASZ)
1396 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1400 memcpy(new_alias->ifalias, alias, len);
1401 new_alias->ifalias[len] = 0;
1404 mutex_lock(&ifalias_mutex);
1405 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1406 mutex_is_locked(&ifalias_mutex));
1407 mutex_unlock(&ifalias_mutex);
1410 kfree_rcu(new_alias, rcuhead);
1414 EXPORT_SYMBOL(dev_set_alias);
1417 * dev_get_alias - get ifalias of a device
1419 * @name: buffer to store name of ifalias
1420 * @len: size of buffer
1422 * get ifalias for a device. Caller must make sure dev cannot go
1423 * away, e.g. rcu read lock or own a reference count to device.
1425 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1427 const struct dev_ifalias *alias;
1431 alias = rcu_dereference(dev->ifalias);
1433 ret = snprintf(name, len, "%s", alias->ifalias);
1440 * netdev_features_change - device changes features
1441 * @dev: device to cause notification
1443 * Called to indicate a device has changed features.
1445 void netdev_features_change(struct net_device *dev)
1447 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1449 EXPORT_SYMBOL(netdev_features_change);
1452 * netdev_state_change - device changes state
1453 * @dev: device to cause notification
1455 * Called to indicate a device has changed state. This function calls
1456 * the notifier chains for netdev_chain and sends a NEWLINK message
1457 * to the routing socket.
1459 void netdev_state_change(struct net_device *dev)
1461 if (dev->flags & IFF_UP) {
1462 struct netdev_notifier_change_info change_info = {
1466 call_netdevice_notifiers_info(NETDEV_CHANGE,
1468 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1471 EXPORT_SYMBOL(netdev_state_change);
1474 * __netdev_notify_peers - notify network peers about existence of @dev,
1475 * to be called when rtnl lock is already held.
1476 * @dev: network device
1478 * Generate traffic such that interested network peers are aware of
1479 * @dev, such as by generating a gratuitous ARP. This may be used when
1480 * a device wants to inform the rest of the network about some sort of
1481 * reconfiguration such as a failover event or virtual machine
1484 void __netdev_notify_peers(struct net_device *dev)
1487 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1488 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1490 EXPORT_SYMBOL(__netdev_notify_peers);
1493 * netdev_notify_peers - notify network peers about existence of @dev
1494 * @dev: network device
1496 * Generate traffic such that interested network peers are aware of
1497 * @dev, such as by generating a gratuitous ARP. This may be used when
1498 * a device wants to inform the rest of the network about some sort of
1499 * reconfiguration such as a failover event or virtual machine
1502 void netdev_notify_peers(struct net_device *dev)
1505 __netdev_notify_peers(dev);
1508 EXPORT_SYMBOL(netdev_notify_peers);
1510 static int napi_threaded_poll(void *data);
1512 static int napi_kthread_create(struct napi_struct *n)
1516 /* Create and wake up the kthread once to put it in
1517 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1518 * warning and work with loadavg.
1520 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1521 n->dev->name, n->napi_id);
1522 if (IS_ERR(n->thread)) {
1523 err = PTR_ERR(n->thread);
1524 pr_err("kthread_run failed with err %d\n", err);
1531 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1533 const struct net_device_ops *ops = dev->netdev_ops;
1538 if (!netif_device_present(dev)) {
1539 /* may be detached because parent is runtime-suspended */
1540 if (dev->dev.parent)
1541 pm_runtime_resume(dev->dev.parent);
1542 if (!netif_device_present(dev))
1546 /* Block netpoll from trying to do any rx path servicing.
1547 * If we don't do this there is a chance ndo_poll_controller
1548 * or ndo_poll may be running while we open the device
1550 netpoll_poll_disable(dev);
1552 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1553 ret = notifier_to_errno(ret);
1557 set_bit(__LINK_STATE_START, &dev->state);
1559 if (ops->ndo_validate_addr)
1560 ret = ops->ndo_validate_addr(dev);
1562 if (!ret && ops->ndo_open)
1563 ret = ops->ndo_open(dev);
1565 netpoll_poll_enable(dev);
1568 clear_bit(__LINK_STATE_START, &dev->state);
1570 dev->flags |= IFF_UP;
1571 dev_set_rx_mode(dev);
1573 add_device_randomness(dev->dev_addr, dev->addr_len);
1580 * dev_open - prepare an interface for use.
1581 * @dev: device to open
1582 * @extack: netlink extended ack
1584 * Takes a device from down to up state. The device's private open
1585 * function is invoked and then the multicast lists are loaded. Finally
1586 * the device is moved into the up state and a %NETDEV_UP message is
1587 * sent to the netdev notifier chain.
1589 * Calling this function on an active interface is a nop. On a failure
1590 * a negative errno code is returned.
1592 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1596 if (dev->flags & IFF_UP)
1599 ret = __dev_open(dev, extack);
1603 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1604 call_netdevice_notifiers(NETDEV_UP, dev);
1608 EXPORT_SYMBOL(dev_open);
1610 static void __dev_close_many(struct list_head *head)
1612 struct net_device *dev;
1617 list_for_each_entry(dev, head, close_list) {
1618 /* Temporarily disable netpoll until the interface is down */
1619 netpoll_poll_disable(dev);
1621 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1623 clear_bit(__LINK_STATE_START, &dev->state);
1625 /* Synchronize to scheduled poll. We cannot touch poll list, it
1626 * can be even on different cpu. So just clear netif_running().
1628 * dev->stop() will invoke napi_disable() on all of it's
1629 * napi_struct instances on this device.
1631 smp_mb__after_atomic(); /* Commit netif_running(). */
1634 dev_deactivate_many(head);
1636 list_for_each_entry(dev, head, close_list) {
1637 const struct net_device_ops *ops = dev->netdev_ops;
1640 * Call the device specific close. This cannot fail.
1641 * Only if device is UP
1643 * We allow it to be called even after a DETACH hot-plug
1649 dev->flags &= ~IFF_UP;
1650 netpoll_poll_enable(dev);
1654 static void __dev_close(struct net_device *dev)
1658 list_add(&dev->close_list, &single);
1659 __dev_close_many(&single);
1663 void dev_close_many(struct list_head *head, bool unlink)
1665 struct net_device *dev, *tmp;
1667 /* Remove the devices that don't need to be closed */
1668 list_for_each_entry_safe(dev, tmp, head, close_list)
1669 if (!(dev->flags & IFF_UP))
1670 list_del_init(&dev->close_list);
1672 __dev_close_many(head);
1674 list_for_each_entry_safe(dev, tmp, head, close_list) {
1675 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1676 call_netdevice_notifiers(NETDEV_DOWN, dev);
1678 list_del_init(&dev->close_list);
1681 EXPORT_SYMBOL(dev_close_many);
1684 * dev_close - shutdown an interface.
1685 * @dev: device to shutdown
1687 * This function moves an active device into down state. A
1688 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1689 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1692 void dev_close(struct net_device *dev)
1694 if (dev->flags & IFF_UP) {
1697 list_add(&dev->close_list, &single);
1698 dev_close_many(&single, true);
1702 EXPORT_SYMBOL(dev_close);
1706 * dev_disable_lro - disable Large Receive Offload on a device
1709 * Disable Large Receive Offload (LRO) on a net device. Must be
1710 * called under RTNL. This is needed if received packets may be
1711 * forwarded to another interface.
1713 void dev_disable_lro(struct net_device *dev)
1715 struct net_device *lower_dev;
1716 struct list_head *iter;
1718 dev->wanted_features &= ~NETIF_F_LRO;
1719 netdev_update_features(dev);
1721 if (unlikely(dev->features & NETIF_F_LRO))
1722 netdev_WARN(dev, "failed to disable LRO!\n");
1724 netdev_for_each_lower_dev(dev, lower_dev, iter)
1725 dev_disable_lro(lower_dev);
1727 EXPORT_SYMBOL(dev_disable_lro);
1730 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1733 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1734 * called under RTNL. This is needed if Generic XDP is installed on
1737 static void dev_disable_gro_hw(struct net_device *dev)
1739 dev->wanted_features &= ~NETIF_F_GRO_HW;
1740 netdev_update_features(dev);
1742 if (unlikely(dev->features & NETIF_F_GRO_HW))
1743 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1746 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1749 case NETDEV_##val: \
1750 return "NETDEV_" __stringify(val);
1752 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1753 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1754 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1755 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1756 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1757 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1758 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1759 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1760 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1764 return "UNKNOWN_NETDEV_EVENT";
1766 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1768 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1769 struct net_device *dev)
1771 struct netdev_notifier_info info = {
1775 return nb->notifier_call(nb, val, &info);
1778 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1779 struct net_device *dev)
1783 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1784 err = notifier_to_errno(err);
1788 if (!(dev->flags & IFF_UP))
1791 call_netdevice_notifier(nb, NETDEV_UP, dev);
1795 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1796 struct net_device *dev)
1798 if (dev->flags & IFF_UP) {
1799 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1801 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1803 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1806 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1809 struct net_device *dev;
1812 for_each_netdev(net, dev) {
1813 err = call_netdevice_register_notifiers(nb, dev);
1820 for_each_netdev_continue_reverse(net, dev)
1821 call_netdevice_unregister_notifiers(nb, dev);
1825 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1828 struct net_device *dev;
1830 for_each_netdev(net, dev)
1831 call_netdevice_unregister_notifiers(nb, dev);
1834 static int dev_boot_phase = 1;
1837 * register_netdevice_notifier - register a network notifier block
1840 * Register a notifier to be called when network device events occur.
1841 * The notifier passed is linked into the kernel structures and must
1842 * not be reused until it has been unregistered. A negative errno code
1843 * is returned on a failure.
1845 * When registered all registration and up events are replayed
1846 * to the new notifier to allow device to have a race free
1847 * view of the network device list.
1850 int register_netdevice_notifier(struct notifier_block *nb)
1855 /* Close race with setup_net() and cleanup_net() */
1856 down_write(&pernet_ops_rwsem);
1858 err = raw_notifier_chain_register(&netdev_chain, nb);
1864 err = call_netdevice_register_net_notifiers(nb, net);
1871 up_write(&pernet_ops_rwsem);
1875 for_each_net_continue_reverse(net)
1876 call_netdevice_unregister_net_notifiers(nb, net);
1878 raw_notifier_chain_unregister(&netdev_chain, nb);
1881 EXPORT_SYMBOL(register_netdevice_notifier);
1884 * unregister_netdevice_notifier - unregister a network notifier block
1887 * Unregister a notifier previously registered by
1888 * register_netdevice_notifier(). The notifier is unlinked into the
1889 * kernel structures and may then be reused. A negative errno code
1890 * is returned on a failure.
1892 * After unregistering unregister and down device events are synthesized
1893 * for all devices on the device list to the removed notifier to remove
1894 * the need for special case cleanup code.
1897 int unregister_netdevice_notifier(struct notifier_block *nb)
1902 /* Close race with setup_net() and cleanup_net() */
1903 down_write(&pernet_ops_rwsem);
1905 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1910 call_netdevice_unregister_net_notifiers(nb, net);
1914 up_write(&pernet_ops_rwsem);
1917 EXPORT_SYMBOL(unregister_netdevice_notifier);
1919 static int __register_netdevice_notifier_net(struct net *net,
1920 struct notifier_block *nb,
1921 bool ignore_call_fail)
1925 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1931 err = call_netdevice_register_net_notifiers(nb, net);
1932 if (err && !ignore_call_fail)
1933 goto chain_unregister;
1938 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1942 static int __unregister_netdevice_notifier_net(struct net *net,
1943 struct notifier_block *nb)
1947 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1951 call_netdevice_unregister_net_notifiers(nb, net);
1956 * register_netdevice_notifier_net - register a per-netns network notifier block
1957 * @net: network namespace
1960 * Register a notifier to be called when network device events occur.
1961 * The notifier passed is linked into the kernel structures and must
1962 * not be reused until it has been unregistered. A negative errno code
1963 * is returned on a failure.
1965 * When registered all registration and up events are replayed
1966 * to the new notifier to allow device to have a race free
1967 * view of the network device list.
1970 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1975 err = __register_netdevice_notifier_net(net, nb, false);
1979 EXPORT_SYMBOL(register_netdevice_notifier_net);
1982 * unregister_netdevice_notifier_net - unregister a per-netns
1983 * network notifier block
1984 * @net: network namespace
1987 * Unregister a notifier previously registered by
1988 * register_netdevice_notifier(). The notifier is unlinked into the
1989 * kernel structures and may then be reused. A negative errno code
1990 * is returned on a failure.
1992 * After unregistering unregister and down device events are synthesized
1993 * for all devices on the device list to the removed notifier to remove
1994 * the need for special case cleanup code.
1997 int unregister_netdevice_notifier_net(struct net *net,
1998 struct notifier_block *nb)
2003 err = __unregister_netdevice_notifier_net(net, nb);
2007 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
2009 int register_netdevice_notifier_dev_net(struct net_device *dev,
2010 struct notifier_block *nb,
2011 struct netdev_net_notifier *nn)
2016 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
2019 list_add(&nn->list, &dev->net_notifier_list);
2024 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
2026 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
2027 struct notifier_block *nb,
2028 struct netdev_net_notifier *nn)
2033 list_del(&nn->list);
2034 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
2038 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
2040 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
2043 struct netdev_net_notifier *nn;
2045 list_for_each_entry(nn, &dev->net_notifier_list, list) {
2046 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
2047 __register_netdevice_notifier_net(net, nn->nb, true);
2052 * call_netdevice_notifiers_info - call all network notifier blocks
2053 * @val: value passed unmodified to notifier function
2054 * @info: notifier information data
2056 * Call all network notifier blocks. Parameters and return value
2057 * are as for raw_notifier_call_chain().
2060 static int call_netdevice_notifiers_info(unsigned long val,
2061 struct netdev_notifier_info *info)
2063 struct net *net = dev_net(info->dev);
2068 /* Run per-netns notifier block chain first, then run the global one.
2069 * Hopefully, one day, the global one is going to be removed after
2070 * all notifier block registrators get converted to be per-netns.
2072 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2073 if (ret & NOTIFY_STOP_MASK)
2075 return raw_notifier_call_chain(&netdev_chain, val, info);
2078 static int call_netdevice_notifiers_extack(unsigned long val,
2079 struct net_device *dev,
2080 struct netlink_ext_ack *extack)
2082 struct netdev_notifier_info info = {
2087 return call_netdevice_notifiers_info(val, &info);
2091 * call_netdevice_notifiers - call all network notifier blocks
2092 * @val: value passed unmodified to notifier function
2093 * @dev: net_device pointer passed unmodified to notifier function
2095 * Call all network notifier blocks. Parameters and return value
2096 * are as for raw_notifier_call_chain().
2099 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2101 return call_netdevice_notifiers_extack(val, dev, NULL);
2103 EXPORT_SYMBOL(call_netdevice_notifiers);
2106 * call_netdevice_notifiers_mtu - call all network notifier blocks
2107 * @val: value passed unmodified to notifier function
2108 * @dev: net_device pointer passed unmodified to notifier function
2109 * @arg: additional u32 argument passed to the notifier function
2111 * Call all network notifier blocks. Parameters and return value
2112 * are as for raw_notifier_call_chain().
2114 static int call_netdevice_notifiers_mtu(unsigned long val,
2115 struct net_device *dev, u32 arg)
2117 struct netdev_notifier_info_ext info = {
2122 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2124 return call_netdevice_notifiers_info(val, &info.info);
2127 #ifdef CONFIG_NET_INGRESS
2128 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2130 void net_inc_ingress_queue(void)
2132 static_branch_inc(&ingress_needed_key);
2134 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2136 void net_dec_ingress_queue(void)
2138 static_branch_dec(&ingress_needed_key);
2140 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2143 #ifdef CONFIG_NET_EGRESS
2144 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2146 void net_inc_egress_queue(void)
2148 static_branch_inc(&egress_needed_key);
2150 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2152 void net_dec_egress_queue(void)
2154 static_branch_dec(&egress_needed_key);
2156 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2159 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2160 #ifdef CONFIG_JUMP_LABEL
2161 static atomic_t netstamp_needed_deferred;
2162 static atomic_t netstamp_wanted;
2163 static void netstamp_clear(struct work_struct *work)
2165 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2168 wanted = atomic_add_return(deferred, &netstamp_wanted);
2170 static_branch_enable(&netstamp_needed_key);
2172 static_branch_disable(&netstamp_needed_key);
2174 static DECLARE_WORK(netstamp_work, netstamp_clear);
2177 void net_enable_timestamp(void)
2179 #ifdef CONFIG_JUMP_LABEL
2183 wanted = atomic_read(&netstamp_wanted);
2186 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2189 atomic_inc(&netstamp_needed_deferred);
2190 schedule_work(&netstamp_work);
2192 static_branch_inc(&netstamp_needed_key);
2195 EXPORT_SYMBOL(net_enable_timestamp);
2197 void net_disable_timestamp(void)
2199 #ifdef CONFIG_JUMP_LABEL
2203 wanted = atomic_read(&netstamp_wanted);
2206 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2209 atomic_dec(&netstamp_needed_deferred);
2210 schedule_work(&netstamp_work);
2212 static_branch_dec(&netstamp_needed_key);
2215 EXPORT_SYMBOL(net_disable_timestamp);
2217 static inline void net_timestamp_set(struct sk_buff *skb)
2220 if (static_branch_unlikely(&netstamp_needed_key))
2221 __net_timestamp(skb);
2224 #define net_timestamp_check(COND, SKB) \
2225 if (static_branch_unlikely(&netstamp_needed_key)) { \
2226 if ((COND) && !(SKB)->tstamp) \
2227 __net_timestamp(SKB); \
2230 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2232 return __is_skb_forwardable(dev, skb, true);
2234 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2236 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2239 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2242 skb->protocol = eth_type_trans(skb, dev);
2243 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2249 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2251 return __dev_forward_skb2(dev, skb, true);
2253 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2256 * dev_forward_skb - loopback an skb to another netif
2258 * @dev: destination network device
2259 * @skb: buffer to forward
2262 * NET_RX_SUCCESS (no congestion)
2263 * NET_RX_DROP (packet was dropped, but freed)
2265 * dev_forward_skb can be used for injecting an skb from the
2266 * start_xmit function of one device into the receive queue
2267 * of another device.
2269 * The receiving device may be in another namespace, so
2270 * we have to clear all information in the skb that could
2271 * impact namespace isolation.
2273 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2275 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2277 EXPORT_SYMBOL_GPL(dev_forward_skb);
2279 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2281 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2284 static inline int deliver_skb(struct sk_buff *skb,
2285 struct packet_type *pt_prev,
2286 struct net_device *orig_dev)
2288 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2290 refcount_inc(&skb->users);
2291 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2294 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2295 struct packet_type **pt,
2296 struct net_device *orig_dev,
2298 struct list_head *ptype_list)
2300 struct packet_type *ptype, *pt_prev = *pt;
2302 list_for_each_entry_rcu(ptype, ptype_list, list) {
2303 if (ptype->type != type)
2306 deliver_skb(skb, pt_prev, orig_dev);
2312 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2314 if (!ptype->af_packet_priv || !skb->sk)
2317 if (ptype->id_match)
2318 return ptype->id_match(ptype, skb->sk);
2319 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2326 * dev_nit_active - return true if any network interface taps are in use
2328 * @dev: network device to check for the presence of taps
2330 bool dev_nit_active(struct net_device *dev)
2332 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2334 EXPORT_SYMBOL_GPL(dev_nit_active);
2337 * Support routine. Sends outgoing frames to any network
2338 * taps currently in use.
2341 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2343 struct packet_type *ptype;
2344 struct sk_buff *skb2 = NULL;
2345 struct packet_type *pt_prev = NULL;
2346 struct list_head *ptype_list = &ptype_all;
2350 list_for_each_entry_rcu(ptype, ptype_list, list) {
2351 if (ptype->ignore_outgoing)
2354 /* Never send packets back to the socket
2355 * they originated from - MvS (miquels@drinkel.ow.org)
2357 if (skb_loop_sk(ptype, skb))
2361 deliver_skb(skb2, pt_prev, skb->dev);
2366 /* need to clone skb, done only once */
2367 skb2 = skb_clone(skb, GFP_ATOMIC);
2371 net_timestamp_set(skb2);
2373 /* skb->nh should be correctly
2374 * set by sender, so that the second statement is
2375 * just protection against buggy protocols.
2377 skb_reset_mac_header(skb2);
2379 if (skb_network_header(skb2) < skb2->data ||
2380 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2381 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2382 ntohs(skb2->protocol),
2384 skb_reset_network_header(skb2);
2387 skb2->transport_header = skb2->network_header;
2388 skb2->pkt_type = PACKET_OUTGOING;
2392 if (ptype_list == &ptype_all) {
2393 ptype_list = &dev->ptype_all;
2398 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2399 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2405 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2408 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2409 * @dev: Network device
2410 * @txq: number of queues available
2412 * If real_num_tx_queues is changed the tc mappings may no longer be
2413 * valid. To resolve this verify the tc mapping remains valid and if
2414 * not NULL the mapping. With no priorities mapping to this
2415 * offset/count pair it will no longer be used. In the worst case TC0
2416 * is invalid nothing can be done so disable priority mappings. If is
2417 * expected that drivers will fix this mapping if they can before
2418 * calling netif_set_real_num_tx_queues.
2420 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2423 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2425 /* If TC0 is invalidated disable TC mapping */
2426 if (tc->offset + tc->count > txq) {
2427 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2432 /* Invalidated prio to tc mappings set to TC0 */
2433 for (i = 1; i < TC_BITMASK + 1; i++) {
2434 int q = netdev_get_prio_tc_map(dev, i);
2436 tc = &dev->tc_to_txq[q];
2437 if (tc->offset + tc->count > txq) {
2438 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2440 netdev_set_prio_tc_map(dev, i, 0);
2445 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2448 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2451 /* walk through the TCs and see if it falls into any of them */
2452 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2453 if ((txq - tc->offset) < tc->count)
2457 /* didn't find it, just return -1 to indicate no match */
2463 EXPORT_SYMBOL(netdev_txq_to_tc);
2466 struct static_key xps_needed __read_mostly;
2467 EXPORT_SYMBOL(xps_needed);
2468 struct static_key xps_rxqs_needed __read_mostly;
2469 EXPORT_SYMBOL(xps_rxqs_needed);
2470 static DEFINE_MUTEX(xps_map_mutex);
2471 #define xmap_dereference(P) \
2472 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2474 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2477 struct xps_map *map = NULL;
2481 map = xmap_dereference(dev_maps->attr_map[tci]);
2485 for (pos = map->len; pos--;) {
2486 if (map->queues[pos] != index)
2490 map->queues[pos] = map->queues[--map->len];
2494 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2495 kfree_rcu(map, rcu);
2502 static bool remove_xps_queue_cpu(struct net_device *dev,
2503 struct xps_dev_maps *dev_maps,
2504 int cpu, u16 offset, u16 count)
2506 int num_tc = dev->num_tc ? : 1;
2507 bool active = false;
2510 for (tci = cpu * num_tc; num_tc--; tci++) {
2513 for (i = count, j = offset; i--; j++) {
2514 if (!remove_xps_queue(dev_maps, tci, j))
2524 static void reset_xps_maps(struct net_device *dev,
2525 struct xps_dev_maps *dev_maps,
2529 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2530 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2532 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2534 static_key_slow_dec_cpuslocked(&xps_needed);
2535 kfree_rcu(dev_maps, rcu);
2538 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2539 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2540 u16 offset, u16 count, bool is_rxqs_map)
2542 bool active = false;
2545 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2547 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2550 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2553 for (i = offset + (count - 1); count--; i--) {
2554 netdev_queue_numa_node_write(
2555 netdev_get_tx_queue(dev, i),
2561 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2564 const unsigned long *possible_mask = NULL;
2565 struct xps_dev_maps *dev_maps;
2566 unsigned int nr_ids;
2568 if (!static_key_false(&xps_needed))
2572 mutex_lock(&xps_map_mutex);
2574 if (static_key_false(&xps_rxqs_needed)) {
2575 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2577 nr_ids = dev->num_rx_queues;
2578 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2579 offset, count, true);
2583 dev_maps = xmap_dereference(dev->xps_cpus_map);
2587 if (num_possible_cpus() > 1)
2588 possible_mask = cpumask_bits(cpu_possible_mask);
2589 nr_ids = nr_cpu_ids;
2590 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2594 mutex_unlock(&xps_map_mutex);
2598 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2600 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2603 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2604 u16 index, bool is_rxqs_map)
2606 struct xps_map *new_map;
2607 int alloc_len = XPS_MIN_MAP_ALLOC;
2610 for (pos = 0; map && pos < map->len; pos++) {
2611 if (map->queues[pos] != index)
2616 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2618 if (pos < map->alloc_len)
2621 alloc_len = map->alloc_len * 2;
2624 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2628 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2630 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2631 cpu_to_node(attr_index));
2635 for (i = 0; i < pos; i++)
2636 new_map->queues[i] = map->queues[i];
2637 new_map->alloc_len = alloc_len;
2643 /* Must be called under cpus_read_lock */
2644 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2645 u16 index, bool is_rxqs_map)
2647 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2648 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2649 int i, j, tci, numa_node_id = -2;
2650 int maps_sz, num_tc = 1, tc = 0;
2651 struct xps_map *map, *new_map;
2652 bool active = false;
2653 unsigned int nr_ids;
2656 /* Do not allow XPS on subordinate device directly */
2657 num_tc = dev->num_tc;
2661 /* If queue belongs to subordinate dev use its map */
2662 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2664 tc = netdev_txq_to_tc(dev, index);
2669 mutex_lock(&xps_map_mutex);
2671 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2672 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2673 nr_ids = dev->num_rx_queues;
2675 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2676 if (num_possible_cpus() > 1) {
2677 online_mask = cpumask_bits(cpu_online_mask);
2678 possible_mask = cpumask_bits(cpu_possible_mask);
2680 dev_maps = xmap_dereference(dev->xps_cpus_map);
2681 nr_ids = nr_cpu_ids;
2684 if (maps_sz < L1_CACHE_BYTES)
2685 maps_sz = L1_CACHE_BYTES;
2687 /* allocate memory for queue storage */
2688 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2691 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2692 if (!new_dev_maps) {
2693 mutex_unlock(&xps_map_mutex);
2697 tci = j * num_tc + tc;
2698 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2701 map = expand_xps_map(map, j, index, is_rxqs_map);
2705 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2709 goto out_no_new_maps;
2712 /* Increment static keys at most once per type */
2713 static_key_slow_inc_cpuslocked(&xps_needed);
2715 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2718 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2720 /* copy maps belonging to foreign traffic classes */
2721 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2722 /* fill in the new device map from the old device map */
2723 map = xmap_dereference(dev_maps->attr_map[tci]);
2724 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2727 /* We need to explicitly update tci as prevous loop
2728 * could break out early if dev_maps is NULL.
2730 tci = j * num_tc + tc;
2732 if (netif_attr_test_mask(j, mask, nr_ids) &&
2733 netif_attr_test_online(j, online_mask, nr_ids)) {
2734 /* add tx-queue to CPU/rx-queue maps */
2737 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2738 while ((pos < map->len) && (map->queues[pos] != index))
2741 if (pos == map->len)
2742 map->queues[map->len++] = index;
2745 if (numa_node_id == -2)
2746 numa_node_id = cpu_to_node(j);
2747 else if (numa_node_id != cpu_to_node(j))
2751 } else if (dev_maps) {
2752 /* fill in the new device map from the old device map */
2753 map = xmap_dereference(dev_maps->attr_map[tci]);
2754 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2757 /* copy maps belonging to foreign traffic classes */
2758 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2759 /* fill in the new device map from the old device map */
2760 map = xmap_dereference(dev_maps->attr_map[tci]);
2761 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2766 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2768 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2770 /* Cleanup old maps */
2772 goto out_no_old_maps;
2774 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2776 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2777 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2778 map = xmap_dereference(dev_maps->attr_map[tci]);
2779 if (map && map != new_map)
2780 kfree_rcu(map, rcu);
2784 kfree_rcu(dev_maps, rcu);
2787 dev_maps = new_dev_maps;
2792 /* update Tx queue numa node */
2793 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2794 (numa_node_id >= 0) ?
2795 numa_node_id : NUMA_NO_NODE);
2801 /* removes tx-queue from unused CPUs/rx-queues */
2802 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2804 for (i = tc, tci = j * num_tc; i--; tci++)
2805 active |= remove_xps_queue(dev_maps, tci, index);
2806 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2807 !netif_attr_test_online(j, online_mask, nr_ids))
2808 active |= remove_xps_queue(dev_maps, tci, index);
2809 for (i = num_tc - tc, tci++; --i; tci++)
2810 active |= remove_xps_queue(dev_maps, tci, index);
2813 /* free map if not active */
2815 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2818 mutex_unlock(&xps_map_mutex);
2822 /* remove any maps that we added */
2823 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2825 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2826 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2828 xmap_dereference(dev_maps->attr_map[tci]) :
2830 if (new_map && new_map != map)
2835 mutex_unlock(&xps_map_mutex);
2837 kfree(new_dev_maps);
2840 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2842 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2848 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2853 EXPORT_SYMBOL(netif_set_xps_queue);
2856 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2858 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2860 /* Unbind any subordinate channels */
2861 while (txq-- != &dev->_tx[0]) {
2863 netdev_unbind_sb_channel(dev, txq->sb_dev);
2867 void netdev_reset_tc(struct net_device *dev)
2870 netif_reset_xps_queues_gt(dev, 0);
2872 netdev_unbind_all_sb_channels(dev);
2874 /* Reset TC configuration of device */
2876 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2877 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2879 EXPORT_SYMBOL(netdev_reset_tc);
2881 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2883 if (tc >= dev->num_tc)
2887 netif_reset_xps_queues(dev, offset, count);
2889 dev->tc_to_txq[tc].count = count;
2890 dev->tc_to_txq[tc].offset = offset;
2893 EXPORT_SYMBOL(netdev_set_tc_queue);
2895 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2897 if (num_tc > TC_MAX_QUEUE)
2901 netif_reset_xps_queues_gt(dev, 0);
2903 netdev_unbind_all_sb_channels(dev);
2905 dev->num_tc = num_tc;
2908 EXPORT_SYMBOL(netdev_set_num_tc);
2910 void netdev_unbind_sb_channel(struct net_device *dev,
2911 struct net_device *sb_dev)
2913 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2916 netif_reset_xps_queues_gt(sb_dev, 0);
2918 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2919 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2921 while (txq-- != &dev->_tx[0]) {
2922 if (txq->sb_dev == sb_dev)
2926 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2928 int netdev_bind_sb_channel_queue(struct net_device *dev,
2929 struct net_device *sb_dev,
2930 u8 tc, u16 count, u16 offset)
2932 /* Make certain the sb_dev and dev are already configured */
2933 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2936 /* We cannot hand out queues we don't have */
2937 if ((offset + count) > dev->real_num_tx_queues)
2940 /* Record the mapping */
2941 sb_dev->tc_to_txq[tc].count = count;
2942 sb_dev->tc_to_txq[tc].offset = offset;
2944 /* Provide a way for Tx queue to find the tc_to_txq map or
2945 * XPS map for itself.
2948 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2952 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2954 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2956 /* Do not use a multiqueue device to represent a subordinate channel */
2957 if (netif_is_multiqueue(dev))
2960 /* We allow channels 1 - 32767 to be used for subordinate channels.
2961 * Channel 0 is meant to be "native" mode and used only to represent
2962 * the main root device. We allow writing 0 to reset the device back
2963 * to normal mode after being used as a subordinate channel.
2965 if (channel > S16_MAX)
2968 dev->num_tc = -channel;
2972 EXPORT_SYMBOL(netdev_set_sb_channel);
2975 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2976 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2978 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2983 disabling = txq < dev->real_num_tx_queues;
2985 if (txq < 1 || txq > dev->num_tx_queues)
2988 if (dev->reg_state == NETREG_REGISTERED ||
2989 dev->reg_state == NETREG_UNREGISTERING) {
2992 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2998 netif_setup_tc(dev, txq);
3000 dev->real_num_tx_queues = txq;
3004 qdisc_reset_all_tx_gt(dev, txq);
3006 netif_reset_xps_queues_gt(dev, txq);
3010 dev->real_num_tx_queues = txq;
3015 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
3019 * netif_set_real_num_rx_queues - set actual number of RX queues used
3020 * @dev: Network device
3021 * @rxq: Actual number of RX queues
3023 * This must be called either with the rtnl_lock held or before
3024 * registration of the net device. Returns 0 on success, or a
3025 * negative error code. If called before registration, it always
3028 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
3032 if (rxq < 1 || rxq > dev->num_rx_queues)
3035 if (dev->reg_state == NETREG_REGISTERED) {
3038 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3044 dev->real_num_rx_queues = rxq;
3047 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3051 * netif_get_num_default_rss_queues - default number of RSS queues
3053 * This routine should set an upper limit on the number of RSS queues
3054 * used by default by multiqueue devices.
3056 int netif_get_num_default_rss_queues(void)
3058 return is_kdump_kernel() ?
3059 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3061 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3063 static void __netif_reschedule(struct Qdisc *q)
3065 struct softnet_data *sd;
3066 unsigned long flags;
3068 local_irq_save(flags);
3069 sd = this_cpu_ptr(&softnet_data);
3070 q->next_sched = NULL;
3071 *sd->output_queue_tailp = q;
3072 sd->output_queue_tailp = &q->next_sched;
3073 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3074 local_irq_restore(flags);
3077 void __netif_schedule(struct Qdisc *q)
3079 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3080 __netif_reschedule(q);
3082 EXPORT_SYMBOL(__netif_schedule);
3084 struct dev_kfree_skb_cb {
3085 enum skb_free_reason reason;
3088 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3090 return (struct dev_kfree_skb_cb *)skb->cb;
3093 void netif_schedule_queue(struct netdev_queue *txq)
3096 if (!netif_xmit_stopped(txq)) {
3097 struct Qdisc *q = rcu_dereference(txq->qdisc);
3099 __netif_schedule(q);
3103 EXPORT_SYMBOL(netif_schedule_queue);
3105 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3107 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3111 q = rcu_dereference(dev_queue->qdisc);
3112 __netif_schedule(q);
3116 EXPORT_SYMBOL(netif_tx_wake_queue);
3118 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3120 unsigned long flags;
3125 if (likely(refcount_read(&skb->users) == 1)) {
3127 refcount_set(&skb->users, 0);
3128 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3131 get_kfree_skb_cb(skb)->reason = reason;
3132 local_irq_save(flags);
3133 skb->next = __this_cpu_read(softnet_data.completion_queue);
3134 __this_cpu_write(softnet_data.completion_queue, skb);
3135 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3136 local_irq_restore(flags);
3138 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3140 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3142 if (in_irq() || irqs_disabled())
3143 __dev_kfree_skb_irq(skb, reason);
3147 EXPORT_SYMBOL(__dev_kfree_skb_any);
3151 * netif_device_detach - mark device as removed
3152 * @dev: network device
3154 * Mark device as removed from system and therefore no longer available.
3156 void netif_device_detach(struct net_device *dev)
3158 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3159 netif_running(dev)) {
3160 netif_tx_stop_all_queues(dev);
3163 EXPORT_SYMBOL(netif_device_detach);
3166 * netif_device_attach - mark device as attached
3167 * @dev: network device
3169 * Mark device as attached from system and restart if needed.
3171 void netif_device_attach(struct net_device *dev)
3173 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3174 netif_running(dev)) {
3175 netif_tx_wake_all_queues(dev);
3176 __netdev_watchdog_up(dev);
3179 EXPORT_SYMBOL(netif_device_attach);
3182 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3183 * to be used as a distribution range.
3185 static u16 skb_tx_hash(const struct net_device *dev,
3186 const struct net_device *sb_dev,
3187 struct sk_buff *skb)
3191 u16 qcount = dev->real_num_tx_queues;
3194 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3196 qoffset = sb_dev->tc_to_txq[tc].offset;
3197 qcount = sb_dev->tc_to_txq[tc].count;
3200 if (skb_rx_queue_recorded(skb)) {
3201 hash = skb_get_rx_queue(skb);
3202 if (hash >= qoffset)
3204 while (unlikely(hash >= qcount))
3206 return hash + qoffset;
3209 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3212 static void skb_warn_bad_offload(const struct sk_buff *skb)
3214 static const netdev_features_t null_features;
3215 struct net_device *dev = skb->dev;
3216 const char *name = "";
3218 if (!net_ratelimit())
3222 if (dev->dev.parent)
3223 name = dev_driver_string(dev->dev.parent);
3225 name = netdev_name(dev);
3227 skb_dump(KERN_WARNING, skb, false);
3228 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3229 name, dev ? &dev->features : &null_features,
3230 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3234 * Invalidate hardware checksum when packet is to be mangled, and
3235 * complete checksum manually on outgoing path.
3237 int skb_checksum_help(struct sk_buff *skb)
3240 int ret = 0, offset;
3242 if (skb->ip_summed == CHECKSUM_COMPLETE)
3243 goto out_set_summed;
3245 if (unlikely(skb_is_gso(skb))) {
3246 skb_warn_bad_offload(skb);
3250 /* Before computing a checksum, we should make sure no frag could
3251 * be modified by an external entity : checksum could be wrong.
3253 if (skb_has_shared_frag(skb)) {
3254 ret = __skb_linearize(skb);
3259 offset = skb_checksum_start_offset(skb);
3260 BUG_ON(offset >= skb_headlen(skb));
3261 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3263 offset += skb->csum_offset;
3264 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3266 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3270 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3272 skb->ip_summed = CHECKSUM_NONE;
3276 EXPORT_SYMBOL(skb_checksum_help);
3278 int skb_crc32c_csum_help(struct sk_buff *skb)
3281 int ret = 0, offset, start;
3283 if (skb->ip_summed != CHECKSUM_PARTIAL)
3286 if (unlikely(skb_is_gso(skb)))
3289 /* Before computing a checksum, we should make sure no frag could
3290 * be modified by an external entity : checksum could be wrong.
3292 if (unlikely(skb_has_shared_frag(skb))) {
3293 ret = __skb_linearize(skb);
3297 start = skb_checksum_start_offset(skb);
3298 offset = start + offsetof(struct sctphdr, checksum);
3299 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3304 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3308 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3309 skb->len - start, ~(__u32)0,
3311 *(__le32 *)(skb->data + offset) = crc32c_csum;
3312 skb->ip_summed = CHECKSUM_NONE;
3313 skb->csum_not_inet = 0;
3318 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3320 __be16 type = skb->protocol;
3322 /* Tunnel gso handlers can set protocol to ethernet. */
3323 if (type == htons(ETH_P_TEB)) {
3326 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3329 eth = (struct ethhdr *)skb->data;
3330 type = eth->h_proto;
3333 return __vlan_get_protocol(skb, type, depth);
3337 * skb_mac_gso_segment - mac layer segmentation handler.
3338 * @skb: buffer to segment
3339 * @features: features for the output path (see dev->features)
3341 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3342 netdev_features_t features)
3344 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3345 struct packet_offload *ptype;
3346 int vlan_depth = skb->mac_len;
3347 __be16 type = skb_network_protocol(skb, &vlan_depth);
3349 if (unlikely(!type))
3350 return ERR_PTR(-EINVAL);
3352 __skb_pull(skb, vlan_depth);
3355 list_for_each_entry_rcu(ptype, &offload_base, list) {
3356 if (ptype->type == type && ptype->callbacks.gso_segment) {
3357 segs = ptype->callbacks.gso_segment(skb, features);
3363 __skb_push(skb, skb->data - skb_mac_header(skb));
3367 EXPORT_SYMBOL(skb_mac_gso_segment);
3370 /* openvswitch calls this on rx path, so we need a different check.
3372 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3375 return skb->ip_summed != CHECKSUM_PARTIAL &&
3376 skb->ip_summed != CHECKSUM_UNNECESSARY;
3378 return skb->ip_summed == CHECKSUM_NONE;
3382 * __skb_gso_segment - Perform segmentation on skb.
3383 * @skb: buffer to segment
3384 * @features: features for the output path (see dev->features)
3385 * @tx_path: whether it is called in TX path
3387 * This function segments the given skb and returns a list of segments.
3389 * It may return NULL if the skb requires no segmentation. This is
3390 * only possible when GSO is used for verifying header integrity.
3392 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3394 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3395 netdev_features_t features, bool tx_path)
3397 struct sk_buff *segs;
3399 if (unlikely(skb_needs_check(skb, tx_path))) {
3402 /* We're going to init ->check field in TCP or UDP header */
3403 err = skb_cow_head(skb, 0);
3405 return ERR_PTR(err);
3408 /* Only report GSO partial support if it will enable us to
3409 * support segmentation on this frame without needing additional
3412 if (features & NETIF_F_GSO_PARTIAL) {
3413 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3414 struct net_device *dev = skb->dev;
3416 partial_features |= dev->features & dev->gso_partial_features;
3417 if (!skb_gso_ok(skb, features | partial_features))
3418 features &= ~NETIF_F_GSO_PARTIAL;
3421 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3422 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3424 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3425 SKB_GSO_CB(skb)->encap_level = 0;
3427 skb_reset_mac_header(skb);
3428 skb_reset_mac_len(skb);
3430 segs = skb_mac_gso_segment(skb, features);
3432 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3433 skb_warn_bad_offload(skb);
3437 EXPORT_SYMBOL(__skb_gso_segment);
3439 /* Take action when hardware reception checksum errors are detected. */
3441 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3443 if (net_ratelimit()) {
3444 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3445 skb_dump(KERN_ERR, skb, true);
3449 EXPORT_SYMBOL(netdev_rx_csum_fault);
3452 /* XXX: check that highmem exists at all on the given machine. */
3453 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3455 #ifdef CONFIG_HIGHMEM
3458 if (!(dev->features & NETIF_F_HIGHDMA)) {
3459 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3460 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3462 if (PageHighMem(skb_frag_page(frag)))
3470 /* If MPLS offload request, verify we are testing hardware MPLS features
3471 * instead of standard features for the netdev.
3473 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3474 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3475 netdev_features_t features,
3478 if (eth_p_mpls(type))
3479 features &= skb->dev->mpls_features;
3484 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3485 netdev_features_t features,
3492 static netdev_features_t harmonize_features(struct sk_buff *skb,
3493 netdev_features_t features)
3497 type = skb_network_protocol(skb, NULL);
3498 features = net_mpls_features(skb, features, type);
3500 if (skb->ip_summed != CHECKSUM_NONE &&
3501 !can_checksum_protocol(features, type)) {
3502 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3504 if (illegal_highdma(skb->dev, skb))
3505 features &= ~NETIF_F_SG;
3510 netdev_features_t passthru_features_check(struct sk_buff *skb,
3511 struct net_device *dev,
3512 netdev_features_t features)
3516 EXPORT_SYMBOL(passthru_features_check);
3518 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3519 struct net_device *dev,
3520 netdev_features_t features)
3522 return vlan_features_check(skb, features);
3525 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3526 struct net_device *dev,
3527 netdev_features_t features)
3529 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3531 if (gso_segs > dev->gso_max_segs)
3532 return features & ~NETIF_F_GSO_MASK;
3534 if (!skb_shinfo(skb)->gso_type) {
3535 skb_warn_bad_offload(skb);
3536 return features & ~NETIF_F_GSO_MASK;
3539 /* Support for GSO partial features requires software
3540 * intervention before we can actually process the packets
3541 * so we need to strip support for any partial features now
3542 * and we can pull them back in after we have partially
3543 * segmented the frame.
3545 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3546 features &= ~dev->gso_partial_features;
3548 /* Make sure to clear the IPv4 ID mangling feature if the
3549 * IPv4 header has the potential to be fragmented.
3551 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3552 struct iphdr *iph = skb->encapsulation ?
3553 inner_ip_hdr(skb) : ip_hdr(skb);
3555 if (!(iph->frag_off & htons(IP_DF)))
3556 features &= ~NETIF_F_TSO_MANGLEID;
3562 netdev_features_t netif_skb_features(struct sk_buff *skb)
3564 struct net_device *dev = skb->dev;
3565 netdev_features_t features = dev->features;
3567 if (skb_is_gso(skb))
3568 features = gso_features_check(skb, dev, features);
3570 /* If encapsulation offload request, verify we are testing
3571 * hardware encapsulation features instead of standard
3572 * features for the netdev
3574 if (skb->encapsulation)
3575 features &= dev->hw_enc_features;
3577 if (skb_vlan_tagged(skb))
3578 features = netdev_intersect_features(features,
3579 dev->vlan_features |
3580 NETIF_F_HW_VLAN_CTAG_TX |
3581 NETIF_F_HW_VLAN_STAG_TX);
3583 if (dev->netdev_ops->ndo_features_check)
3584 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3587 features &= dflt_features_check(skb, dev, features);
3589 return harmonize_features(skb, features);
3591 EXPORT_SYMBOL(netif_skb_features);
3593 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3594 struct netdev_queue *txq, bool more)
3599 if (dev_nit_active(dev))
3600 dev_queue_xmit_nit(skb, dev);
3603 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3604 trace_net_dev_start_xmit(skb, dev);
3605 rc = netdev_start_xmit(skb, dev, txq, more);
3606 trace_net_dev_xmit(skb, rc, dev, len);
3611 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3612 struct netdev_queue *txq, int *ret)
3614 struct sk_buff *skb = first;
3615 int rc = NETDEV_TX_OK;
3618 struct sk_buff *next = skb->next;
3620 skb_mark_not_on_list(skb);
3621 rc = xmit_one(skb, dev, txq, next != NULL);
3622 if (unlikely(!dev_xmit_complete(rc))) {
3628 if (netif_tx_queue_stopped(txq) && skb) {
3629 rc = NETDEV_TX_BUSY;
3639 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3640 netdev_features_t features)
3642 if (skb_vlan_tag_present(skb) &&
3643 !vlan_hw_offload_capable(features, skb->vlan_proto))
3644 skb = __vlan_hwaccel_push_inside(skb);
3648 int skb_csum_hwoffload_help(struct sk_buff *skb,
3649 const netdev_features_t features)
3651 if (unlikely(skb_csum_is_sctp(skb)))
3652 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3653 skb_crc32c_csum_help(skb);
3655 if (features & NETIF_F_HW_CSUM)
3658 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3659 switch (skb->csum_offset) {
3660 case offsetof(struct tcphdr, check):
3661 case offsetof(struct udphdr, check):
3666 return skb_checksum_help(skb);
3668 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3670 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3672 netdev_features_t features;
3674 features = netif_skb_features(skb);
3675 skb = validate_xmit_vlan(skb, features);
3679 skb = sk_validate_xmit_skb(skb, dev);
3683 if (netif_needs_gso(skb, features)) {
3684 struct sk_buff *segs;
3686 segs = skb_gso_segment(skb, features);
3694 if (skb_needs_linearize(skb, features) &&
3695 __skb_linearize(skb))
3698 /* If packet is not checksummed and device does not
3699 * support checksumming for this protocol, complete
3700 * checksumming here.
3702 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3703 if (skb->encapsulation)
3704 skb_set_inner_transport_header(skb,
3705 skb_checksum_start_offset(skb));
3707 skb_set_transport_header(skb,
3708 skb_checksum_start_offset(skb));
3709 if (skb_csum_hwoffload_help(skb, features))
3714 skb = validate_xmit_xfrm(skb, features, again);
3721 atomic_long_inc(&dev->tx_dropped);
3725 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3727 struct sk_buff *next, *head = NULL, *tail;
3729 for (; skb != NULL; skb = next) {
3731 skb_mark_not_on_list(skb);
3733 /* in case skb wont be segmented, point to itself */
3736 skb = validate_xmit_skb(skb, dev, again);
3744 /* If skb was segmented, skb->prev points to
3745 * the last segment. If not, it still contains skb.
3751 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3753 static void qdisc_pkt_len_init(struct sk_buff *skb)
3755 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3757 qdisc_skb_cb(skb)->pkt_len = skb->len;
3759 /* To get more precise estimation of bytes sent on wire,
3760 * we add to pkt_len the headers size of all segments
3762 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3763 unsigned int hdr_len;
3764 u16 gso_segs = shinfo->gso_segs;
3766 /* mac layer + network layer */
3767 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3769 /* + transport layer */
3770 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3771 const struct tcphdr *th;
3772 struct tcphdr _tcphdr;
3774 th = skb_header_pointer(skb, skb_transport_offset(skb),
3775 sizeof(_tcphdr), &_tcphdr);
3777 hdr_len += __tcp_hdrlen(th);
3779 struct udphdr _udphdr;
3781 if (skb_header_pointer(skb, skb_transport_offset(skb),
3782 sizeof(_udphdr), &_udphdr))
3783 hdr_len += sizeof(struct udphdr);
3786 if (shinfo->gso_type & SKB_GSO_DODGY)
3787 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3790 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3794 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3795 struct net_device *dev,
3796 struct netdev_queue *txq)
3798 spinlock_t *root_lock = qdisc_lock(q);
3799 struct sk_buff *to_free = NULL;
3803 qdisc_calculate_pkt_len(skb, q);
3805 if (q->flags & TCQ_F_NOLOCK) {
3806 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3809 if (unlikely(to_free))
3810 kfree_skb_list(to_free);
3815 * Heuristic to force contended enqueues to serialize on a
3816 * separate lock before trying to get qdisc main lock.
3817 * This permits qdisc->running owner to get the lock more
3818 * often and dequeue packets faster.
3820 contended = qdisc_is_running(q);
3821 if (unlikely(contended))
3822 spin_lock(&q->busylock);
3824 spin_lock(root_lock);
3825 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3826 __qdisc_drop(skb, &to_free);
3828 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3829 qdisc_run_begin(q)) {
3831 * This is a work-conserving queue; there are no old skbs
3832 * waiting to be sent out; and the qdisc is not running -
3833 * xmit the skb directly.
3836 qdisc_bstats_update(q, skb);
3838 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3839 if (unlikely(contended)) {
3840 spin_unlock(&q->busylock);
3847 rc = NET_XMIT_SUCCESS;
3849 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3850 if (qdisc_run_begin(q)) {
3851 if (unlikely(contended)) {
3852 spin_unlock(&q->busylock);
3859 spin_unlock(root_lock);
3860 if (unlikely(to_free))
3861 kfree_skb_list(to_free);
3862 if (unlikely(contended))
3863 spin_unlock(&q->busylock);
3867 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3868 static void skb_update_prio(struct sk_buff *skb)
3870 const struct netprio_map *map;
3871 const struct sock *sk;
3872 unsigned int prioidx;
3876 map = rcu_dereference_bh(skb->dev->priomap);
3879 sk = skb_to_full_sk(skb);
3883 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3885 if (prioidx < map->priomap_len)
3886 skb->priority = map->priomap[prioidx];
3889 #define skb_update_prio(skb)
3893 * dev_loopback_xmit - loop back @skb
3894 * @net: network namespace this loopback is happening in
3895 * @sk: sk needed to be a netfilter okfn
3896 * @skb: buffer to transmit
3898 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3900 skb_reset_mac_header(skb);
3901 __skb_pull(skb, skb_network_offset(skb));
3902 skb->pkt_type = PACKET_LOOPBACK;
3903 skb->ip_summed = CHECKSUM_UNNECESSARY;
3904 WARN_ON(!skb_dst(skb));
3909 EXPORT_SYMBOL(dev_loopback_xmit);
3911 #ifdef CONFIG_NET_EGRESS
3912 static struct sk_buff *
3913 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3915 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3916 struct tcf_result cl_res;
3921 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3922 qdisc_skb_cb(skb)->mru = 0;
3923 qdisc_skb_cb(skb)->post_ct = false;
3924 mini_qdisc_bstats_cpu_update(miniq, skb);
3926 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3928 case TC_ACT_RECLASSIFY:
3929 skb->tc_index = TC_H_MIN(cl_res.classid);
3932 mini_qdisc_qstats_cpu_drop(miniq);
3933 *ret = NET_XMIT_DROP;
3939 *ret = NET_XMIT_SUCCESS;
3942 case TC_ACT_REDIRECT:
3943 /* No need to push/pop skb's mac_header here on egress! */
3944 skb_do_redirect(skb);
3945 *ret = NET_XMIT_SUCCESS;
3953 #endif /* CONFIG_NET_EGRESS */
3956 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3957 struct xps_dev_maps *dev_maps, unsigned int tci)
3959 struct xps_map *map;
3960 int queue_index = -1;
3964 tci += netdev_get_prio_tc_map(dev, skb->priority);
3967 map = rcu_dereference(dev_maps->attr_map[tci]);
3970 queue_index = map->queues[0];
3972 queue_index = map->queues[reciprocal_scale(
3973 skb_get_hash(skb), map->len)];
3974 if (unlikely(queue_index >= dev->real_num_tx_queues))
3981 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3982 struct sk_buff *skb)
3985 struct xps_dev_maps *dev_maps;
3986 struct sock *sk = skb->sk;
3987 int queue_index = -1;
3989 if (!static_key_false(&xps_needed))
3993 if (!static_key_false(&xps_rxqs_needed))
3996 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3998 int tci = sk_rx_queue_get(sk);
4000 if (tci >= 0 && tci < dev->num_rx_queues)
4001 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4006 if (queue_index < 0) {
4007 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
4009 unsigned int tci = skb->sender_cpu - 1;
4011 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4023 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4024 struct net_device *sb_dev)
4028 EXPORT_SYMBOL(dev_pick_tx_zero);
4030 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4031 struct net_device *sb_dev)
4033 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4035 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4037 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4038 struct net_device *sb_dev)
4040 struct sock *sk = skb->sk;
4041 int queue_index = sk_tx_queue_get(sk);
4043 sb_dev = sb_dev ? : dev;
4045 if (queue_index < 0 || skb->ooo_okay ||
4046 queue_index >= dev->real_num_tx_queues) {
4047 int new_index = get_xps_queue(dev, sb_dev, skb);
4050 new_index = skb_tx_hash(dev, sb_dev, skb);
4052 if (queue_index != new_index && sk &&
4054 rcu_access_pointer(sk->sk_dst_cache))
4055 sk_tx_queue_set(sk, new_index);
4057 queue_index = new_index;
4062 EXPORT_SYMBOL(netdev_pick_tx);
4064 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4065 struct sk_buff *skb,
4066 struct net_device *sb_dev)
4068 int queue_index = 0;
4071 u32 sender_cpu = skb->sender_cpu - 1;
4073 if (sender_cpu >= (u32)NR_CPUS)
4074 skb->sender_cpu = raw_smp_processor_id() + 1;
4077 if (dev->real_num_tx_queues != 1) {
4078 const struct net_device_ops *ops = dev->netdev_ops;
4080 if (ops->ndo_select_queue)
4081 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4083 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4085 queue_index = netdev_cap_txqueue(dev, queue_index);
4088 skb_set_queue_mapping(skb, queue_index);
4089 return netdev_get_tx_queue(dev, queue_index);
4093 * __dev_queue_xmit - transmit a buffer
4094 * @skb: buffer to transmit
4095 * @sb_dev: suboordinate device used for L2 forwarding offload
4097 * Queue a buffer for transmission to a network device. The caller must
4098 * have set the device and priority and built the buffer before calling
4099 * this function. The function can be called from an interrupt.
4101 * A negative errno code is returned on a failure. A success does not
4102 * guarantee the frame will be transmitted as it may be dropped due
4103 * to congestion or traffic shaping.
4105 * -----------------------------------------------------------------------------------
4106 * I notice this method can also return errors from the queue disciplines,
4107 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4110 * Regardless of the return value, the skb is consumed, so it is currently
4111 * difficult to retry a send to this method. (You can bump the ref count
4112 * before sending to hold a reference for retry if you are careful.)
4114 * When calling this method, interrupts MUST be enabled. This is because
4115 * the BH enable code must have IRQs enabled so that it will not deadlock.
4118 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4120 struct net_device *dev = skb->dev;
4121 struct netdev_queue *txq;
4126 skb_reset_mac_header(skb);
4128 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4129 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4131 /* Disable soft irqs for various locks below. Also
4132 * stops preemption for RCU.
4136 skb_update_prio(skb);
4138 qdisc_pkt_len_init(skb);
4139 #ifdef CONFIG_NET_CLS_ACT
4140 skb->tc_at_ingress = 0;
4141 # ifdef CONFIG_NET_EGRESS
4142 if (static_branch_unlikely(&egress_needed_key)) {
4143 skb = sch_handle_egress(skb, &rc, dev);
4149 /* If device/qdisc don't need skb->dst, release it right now while
4150 * its hot in this cpu cache.
4152 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4157 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4158 q = rcu_dereference_bh(txq->qdisc);
4160 trace_net_dev_queue(skb);
4162 rc = __dev_xmit_skb(skb, q, dev, txq);
4166 /* The device has no queue. Common case for software devices:
4167 * loopback, all the sorts of tunnels...
4169 * Really, it is unlikely that netif_tx_lock protection is necessary
4170 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4172 * However, it is possible, that they rely on protection
4175 * Check this and shot the lock. It is not prone from deadlocks.
4176 *Either shot noqueue qdisc, it is even simpler 8)
4178 if (dev->flags & IFF_UP) {
4179 int cpu = smp_processor_id(); /* ok because BHs are off */
4181 if (txq->xmit_lock_owner != cpu) {
4182 if (dev_xmit_recursion())
4183 goto recursion_alert;
4185 skb = validate_xmit_skb(skb, dev, &again);
4189 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4190 HARD_TX_LOCK(dev, txq, cpu);
4192 if (!netif_xmit_stopped(txq)) {
4193 dev_xmit_recursion_inc();
4194 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4195 dev_xmit_recursion_dec();
4196 if (dev_xmit_complete(rc)) {
4197 HARD_TX_UNLOCK(dev, txq);
4201 HARD_TX_UNLOCK(dev, txq);
4202 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4205 /* Recursion is detected! It is possible,
4209 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4215 rcu_read_unlock_bh();
4217 atomic_long_inc(&dev->tx_dropped);
4218 kfree_skb_list(skb);
4221 rcu_read_unlock_bh();
4225 int dev_queue_xmit(struct sk_buff *skb)
4227 return __dev_queue_xmit(skb, NULL);
4229 EXPORT_SYMBOL(dev_queue_xmit);
4231 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4233 return __dev_queue_xmit(skb, sb_dev);
4235 EXPORT_SYMBOL(dev_queue_xmit_accel);
4237 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4239 struct net_device *dev = skb->dev;
4240 struct sk_buff *orig_skb = skb;
4241 struct netdev_queue *txq;
4242 int ret = NETDEV_TX_BUSY;
4245 if (unlikely(!netif_running(dev) ||
4246 !netif_carrier_ok(dev)))
4249 skb = validate_xmit_skb_list(skb, dev, &again);
4250 if (skb != orig_skb)
4253 skb_set_queue_mapping(skb, queue_id);
4254 txq = skb_get_tx_queue(dev, skb);
4255 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4259 dev_xmit_recursion_inc();
4260 HARD_TX_LOCK(dev, txq, smp_processor_id());
4261 if (!netif_xmit_frozen_or_drv_stopped(txq))
4262 ret = netdev_start_xmit(skb, dev, txq, false);
4263 HARD_TX_UNLOCK(dev, txq);
4264 dev_xmit_recursion_dec();
4269 atomic_long_inc(&dev->tx_dropped);
4270 kfree_skb_list(skb);
4271 return NET_XMIT_DROP;
4273 EXPORT_SYMBOL(__dev_direct_xmit);
4275 /*************************************************************************
4277 *************************************************************************/
4279 int netdev_max_backlog __read_mostly = 1000;
4280 EXPORT_SYMBOL(netdev_max_backlog);
4282 int netdev_tstamp_prequeue __read_mostly = 1;
4283 int netdev_budget __read_mostly = 300;
4284 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4285 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4286 int weight_p __read_mostly = 64; /* old backlog weight */
4287 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4288 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4289 int dev_rx_weight __read_mostly = 64;
4290 int dev_tx_weight __read_mostly = 64;
4291 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4292 int gro_normal_batch __read_mostly = 8;
4294 /* Called with irq disabled */
4295 static inline void ____napi_schedule(struct softnet_data *sd,
4296 struct napi_struct *napi)
4298 struct task_struct *thread;
4300 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4301 /* Paired with smp_mb__before_atomic() in
4302 * napi_enable()/dev_set_threaded().
4303 * Use READ_ONCE() to guarantee a complete
4304 * read on napi->thread. Only call
4305 * wake_up_process() when it's not NULL.
4307 thread = READ_ONCE(napi->thread);
4309 /* Avoid doing set_bit() if the thread is in
4310 * INTERRUPTIBLE state, cause napi_thread_wait()
4311 * makes sure to proceed with napi polling
4312 * if the thread is explicitly woken from here.
4314 if (READ_ONCE(thread->state) != TASK_INTERRUPTIBLE)
4315 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4316 wake_up_process(thread);
4321 list_add_tail(&napi->poll_list, &sd->poll_list);
4322 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4327 /* One global table that all flow-based protocols share. */
4328 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4329 EXPORT_SYMBOL(rps_sock_flow_table);
4330 u32 rps_cpu_mask __read_mostly;
4331 EXPORT_SYMBOL(rps_cpu_mask);
4333 struct static_key_false rps_needed __read_mostly;
4334 EXPORT_SYMBOL(rps_needed);
4335 struct static_key_false rfs_needed __read_mostly;
4336 EXPORT_SYMBOL(rfs_needed);
4338 static struct rps_dev_flow *
4339 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4340 struct rps_dev_flow *rflow, u16 next_cpu)
4342 if (next_cpu < nr_cpu_ids) {
4343 #ifdef CONFIG_RFS_ACCEL
4344 struct netdev_rx_queue *rxqueue;
4345 struct rps_dev_flow_table *flow_table;
4346 struct rps_dev_flow *old_rflow;
4351 /* Should we steer this flow to a different hardware queue? */
4352 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4353 !(dev->features & NETIF_F_NTUPLE))
4355 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4356 if (rxq_index == skb_get_rx_queue(skb))
4359 rxqueue = dev->_rx + rxq_index;
4360 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4363 flow_id = skb_get_hash(skb) & flow_table->mask;
4364 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4365 rxq_index, flow_id);
4369 rflow = &flow_table->flows[flow_id];
4371 if (old_rflow->filter == rflow->filter)
4372 old_rflow->filter = RPS_NO_FILTER;
4376 per_cpu(softnet_data, next_cpu).input_queue_head;
4379 rflow->cpu = next_cpu;
4384 * get_rps_cpu is called from netif_receive_skb and returns the target
4385 * CPU from the RPS map of the receiving queue for a given skb.
4386 * rcu_read_lock must be held on entry.
4388 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4389 struct rps_dev_flow **rflowp)
4391 const struct rps_sock_flow_table *sock_flow_table;
4392 struct netdev_rx_queue *rxqueue = dev->_rx;
4393 struct rps_dev_flow_table *flow_table;
4394 struct rps_map *map;
4399 if (skb_rx_queue_recorded(skb)) {
4400 u16 index = skb_get_rx_queue(skb);
4402 if (unlikely(index >= dev->real_num_rx_queues)) {
4403 WARN_ONCE(dev->real_num_rx_queues > 1,
4404 "%s received packet on queue %u, but number "
4405 "of RX queues is %u\n",
4406 dev->name, index, dev->real_num_rx_queues);
4412 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4414 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4415 map = rcu_dereference(rxqueue->rps_map);
4416 if (!flow_table && !map)
4419 skb_reset_network_header(skb);
4420 hash = skb_get_hash(skb);
4424 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4425 if (flow_table && sock_flow_table) {
4426 struct rps_dev_flow *rflow;
4430 /* First check into global flow table if there is a match */
4431 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4432 if ((ident ^ hash) & ~rps_cpu_mask)
4435 next_cpu = ident & rps_cpu_mask;
4437 /* OK, now we know there is a match,
4438 * we can look at the local (per receive queue) flow table
4440 rflow = &flow_table->flows[hash & flow_table->mask];
4444 * If the desired CPU (where last recvmsg was done) is
4445 * different from current CPU (one in the rx-queue flow
4446 * table entry), switch if one of the following holds:
4447 * - Current CPU is unset (>= nr_cpu_ids).
4448 * - Current CPU is offline.
4449 * - The current CPU's queue tail has advanced beyond the
4450 * last packet that was enqueued using this table entry.
4451 * This guarantees that all previous packets for the flow
4452 * have been dequeued, thus preserving in order delivery.
4454 if (unlikely(tcpu != next_cpu) &&
4455 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4456 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4457 rflow->last_qtail)) >= 0)) {
4459 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4462 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4472 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4473 if (cpu_online(tcpu)) {
4483 #ifdef CONFIG_RFS_ACCEL
4486 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4487 * @dev: Device on which the filter was set
4488 * @rxq_index: RX queue index
4489 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4490 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4492 * Drivers that implement ndo_rx_flow_steer() should periodically call
4493 * this function for each installed filter and remove the filters for
4494 * which it returns %true.
4496 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4497 u32 flow_id, u16 filter_id)
4499 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4500 struct rps_dev_flow_table *flow_table;
4501 struct rps_dev_flow *rflow;
4506 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4507 if (flow_table && flow_id <= flow_table->mask) {
4508 rflow = &flow_table->flows[flow_id];
4509 cpu = READ_ONCE(rflow->cpu);
4510 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4511 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4512 rflow->last_qtail) <
4513 (int)(10 * flow_table->mask)))
4519 EXPORT_SYMBOL(rps_may_expire_flow);
4521 #endif /* CONFIG_RFS_ACCEL */
4523 /* Called from hardirq (IPI) context */
4524 static void rps_trigger_softirq(void *data)
4526 struct softnet_data *sd = data;
4528 ____napi_schedule(sd, &sd->backlog);
4532 #endif /* CONFIG_RPS */
4535 * Check if this softnet_data structure is another cpu one
4536 * If yes, queue it to our IPI list and return 1
4539 static int rps_ipi_queued(struct softnet_data *sd)
4542 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4545 sd->rps_ipi_next = mysd->rps_ipi_list;
4546 mysd->rps_ipi_list = sd;
4548 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4551 #endif /* CONFIG_RPS */
4555 #ifdef CONFIG_NET_FLOW_LIMIT
4556 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4559 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4561 #ifdef CONFIG_NET_FLOW_LIMIT
4562 struct sd_flow_limit *fl;
4563 struct softnet_data *sd;
4564 unsigned int old_flow, new_flow;
4566 if (qlen < (netdev_max_backlog >> 1))
4569 sd = this_cpu_ptr(&softnet_data);
4572 fl = rcu_dereference(sd->flow_limit);
4574 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4575 old_flow = fl->history[fl->history_head];
4576 fl->history[fl->history_head] = new_flow;
4579 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4581 if (likely(fl->buckets[old_flow]))
4582 fl->buckets[old_flow]--;
4584 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4596 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4597 * queue (may be a remote CPU queue).
4599 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4600 unsigned int *qtail)
4602 struct softnet_data *sd;
4603 unsigned long flags;
4606 sd = &per_cpu(softnet_data, cpu);
4608 local_irq_save(flags);
4611 if (!netif_running(skb->dev))
4613 qlen = skb_queue_len(&sd->input_pkt_queue);
4614 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4617 __skb_queue_tail(&sd->input_pkt_queue, skb);
4618 input_queue_tail_incr_save(sd, qtail);
4620 local_irq_restore(flags);
4621 return NET_RX_SUCCESS;
4624 /* Schedule NAPI for backlog device
4625 * We can use non atomic operation since we own the queue lock
4627 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4628 if (!rps_ipi_queued(sd))
4629 ____napi_schedule(sd, &sd->backlog);
4638 local_irq_restore(flags);
4640 atomic_long_inc(&skb->dev->rx_dropped);
4645 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4647 struct net_device *dev = skb->dev;
4648 struct netdev_rx_queue *rxqueue;
4652 if (skb_rx_queue_recorded(skb)) {
4653 u16 index = skb_get_rx_queue(skb);
4655 if (unlikely(index >= dev->real_num_rx_queues)) {
4656 WARN_ONCE(dev->real_num_rx_queues > 1,
4657 "%s received packet on queue %u, but number "
4658 "of RX queues is %u\n",
4659 dev->name, index, dev->real_num_rx_queues);
4661 return rxqueue; /* Return first rxqueue */
4668 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4669 struct xdp_buff *xdp,
4670 struct bpf_prog *xdp_prog)
4672 void *orig_data, *orig_data_end, *hard_start;
4673 struct netdev_rx_queue *rxqueue;
4674 u32 metalen, act = XDP_DROP;
4675 u32 mac_len, frame_sz;
4676 __be16 orig_eth_type;
4681 /* Reinjected packets coming from act_mirred or similar should
4682 * not get XDP generic processing.
4684 if (skb_is_redirected(skb))
4687 /* XDP packets must be linear and must have sufficient headroom
4688 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4689 * native XDP provides, thus we need to do it here as well.
4691 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4692 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4693 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4694 int troom = skb->tail + skb->data_len - skb->end;
4696 /* In case we have to go down the path and also linearize,
4697 * then lets do the pskb_expand_head() work just once here.
4699 if (pskb_expand_head(skb,
4700 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4701 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4703 if (skb_linearize(skb))
4707 /* The XDP program wants to see the packet starting at the MAC
4710 mac_len = skb->data - skb_mac_header(skb);
4711 hard_start = skb->data - skb_headroom(skb);
4713 /* SKB "head" area always have tailroom for skb_shared_info */
4714 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4715 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4717 rxqueue = netif_get_rxqueue(skb);
4718 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4719 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4720 skb_headlen(skb) + mac_len, true);
4722 orig_data_end = xdp->data_end;
4723 orig_data = xdp->data;
4724 eth = (struct ethhdr *)xdp->data;
4725 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4726 orig_eth_type = eth->h_proto;
4728 act = bpf_prog_run_xdp(xdp_prog, xdp);
4730 /* check if bpf_xdp_adjust_head was used */
4731 off = xdp->data - orig_data;
4734 __skb_pull(skb, off);
4736 __skb_push(skb, -off);
4738 skb->mac_header += off;
4739 skb_reset_network_header(skb);
4742 /* check if bpf_xdp_adjust_tail was used */
4743 off = xdp->data_end - orig_data_end;
4745 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4746 skb->len += off; /* positive on grow, negative on shrink */
4749 /* check if XDP changed eth hdr such SKB needs update */
4750 eth = (struct ethhdr *)xdp->data;
4751 if ((orig_eth_type != eth->h_proto) ||
4752 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4753 __skb_push(skb, ETH_HLEN);
4754 skb->protocol = eth_type_trans(skb, skb->dev);
4760 __skb_push(skb, mac_len);
4763 metalen = xdp->data - xdp->data_meta;
4765 skb_metadata_set(skb, metalen);
4768 bpf_warn_invalid_xdp_action(act);
4771 trace_xdp_exception(skb->dev, xdp_prog, act);
4782 /* When doing generic XDP we have to bypass the qdisc layer and the
4783 * network taps in order to match in-driver-XDP behavior.
4785 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4787 struct net_device *dev = skb->dev;
4788 struct netdev_queue *txq;
4789 bool free_skb = true;
4792 txq = netdev_core_pick_tx(dev, skb, NULL);
4793 cpu = smp_processor_id();
4794 HARD_TX_LOCK(dev, txq, cpu);
4795 if (!netif_xmit_stopped(txq)) {
4796 rc = netdev_start_xmit(skb, dev, txq, 0);
4797 if (dev_xmit_complete(rc))
4800 HARD_TX_UNLOCK(dev, txq);
4802 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4807 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4809 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4812 struct xdp_buff xdp;
4816 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4817 if (act != XDP_PASS) {
4820 err = xdp_do_generic_redirect(skb->dev, skb,
4826 generic_xdp_tx(skb, xdp_prog);
4837 EXPORT_SYMBOL_GPL(do_xdp_generic);
4839 static int netif_rx_internal(struct sk_buff *skb)
4843 net_timestamp_check(netdev_tstamp_prequeue, skb);
4845 trace_netif_rx(skb);
4848 if (static_branch_unlikely(&rps_needed)) {
4849 struct rps_dev_flow voidflow, *rflow = &voidflow;
4855 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4857 cpu = smp_processor_id();
4859 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4868 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4875 * netif_rx - post buffer to the network code
4876 * @skb: buffer to post
4878 * This function receives a packet from a device driver and queues it for
4879 * the upper (protocol) levels to process. It always succeeds. The buffer
4880 * may be dropped during processing for congestion control or by the
4884 * NET_RX_SUCCESS (no congestion)
4885 * NET_RX_DROP (packet was dropped)
4889 int netif_rx(struct sk_buff *skb)
4893 trace_netif_rx_entry(skb);
4895 ret = netif_rx_internal(skb);
4896 trace_netif_rx_exit(ret);
4900 EXPORT_SYMBOL(netif_rx);
4902 int netif_rx_ni(struct sk_buff *skb)
4906 trace_netif_rx_ni_entry(skb);
4909 err = netif_rx_internal(skb);
4910 if (local_softirq_pending())
4913 trace_netif_rx_ni_exit(err);
4917 EXPORT_SYMBOL(netif_rx_ni);
4919 int netif_rx_any_context(struct sk_buff *skb)
4922 * If invoked from contexts which do not invoke bottom half
4923 * processing either at return from interrupt or when softrqs are
4924 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4928 return netif_rx(skb);
4930 return netif_rx_ni(skb);
4932 EXPORT_SYMBOL(netif_rx_any_context);
4934 static __latent_entropy void net_tx_action(struct softirq_action *h)
4936 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4938 if (sd->completion_queue) {
4939 struct sk_buff *clist;
4941 local_irq_disable();
4942 clist = sd->completion_queue;
4943 sd->completion_queue = NULL;
4947 struct sk_buff *skb = clist;
4949 clist = clist->next;
4951 WARN_ON(refcount_read(&skb->users));
4952 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4953 trace_consume_skb(skb);
4955 trace_kfree_skb(skb, net_tx_action);
4957 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4960 __kfree_skb_defer(skb);
4964 if (sd->output_queue) {
4967 local_irq_disable();
4968 head = sd->output_queue;
4969 sd->output_queue = NULL;
4970 sd->output_queue_tailp = &sd->output_queue;
4974 struct Qdisc *q = head;
4975 spinlock_t *root_lock = NULL;
4977 head = head->next_sched;
4979 if (!(q->flags & TCQ_F_NOLOCK)) {
4980 root_lock = qdisc_lock(q);
4981 spin_lock(root_lock);
4983 /* We need to make sure head->next_sched is read
4984 * before clearing __QDISC_STATE_SCHED
4986 smp_mb__before_atomic();
4987 clear_bit(__QDISC_STATE_SCHED, &q->state);
4990 spin_unlock(root_lock);
4994 xfrm_dev_backlog(sd);
4997 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4998 /* This hook is defined here for ATM LANE */
4999 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5000 unsigned char *addr) __read_mostly;
5001 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5004 static inline struct sk_buff *
5005 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5006 struct net_device *orig_dev, bool *another)
5008 #ifdef CONFIG_NET_CLS_ACT
5009 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5010 struct tcf_result cl_res;
5012 /* If there's at least one ingress present somewhere (so
5013 * we get here via enabled static key), remaining devices
5014 * that are not configured with an ingress qdisc will bail
5021 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5025 qdisc_skb_cb(skb)->pkt_len = skb->len;
5026 qdisc_skb_cb(skb)->mru = 0;
5027 qdisc_skb_cb(skb)->post_ct = false;
5028 skb->tc_at_ingress = 1;
5029 mini_qdisc_bstats_cpu_update(miniq, skb);
5031 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
5034 case TC_ACT_RECLASSIFY:
5035 skb->tc_index = TC_H_MIN(cl_res.classid);
5038 mini_qdisc_qstats_cpu_drop(miniq);
5046 case TC_ACT_REDIRECT:
5047 /* skb_mac_header check was done by cls/act_bpf, so
5048 * we can safely push the L2 header back before
5049 * redirecting to another netdev
5051 __skb_push(skb, skb->mac_len);
5052 if (skb_do_redirect(skb) == -EAGAIN) {
5053 __skb_pull(skb, skb->mac_len);
5058 case TC_ACT_CONSUMED:
5063 #endif /* CONFIG_NET_CLS_ACT */
5068 * netdev_is_rx_handler_busy - check if receive handler is registered
5069 * @dev: device to check
5071 * Check if a receive handler is already registered for a given device.
5072 * Return true if there one.
5074 * The caller must hold the rtnl_mutex.
5076 bool netdev_is_rx_handler_busy(struct net_device *dev)
5079 return dev && rtnl_dereference(dev->rx_handler);
5081 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5084 * netdev_rx_handler_register - register receive handler
5085 * @dev: device to register a handler for
5086 * @rx_handler: receive handler to register
5087 * @rx_handler_data: data pointer that is used by rx handler
5089 * Register a receive handler for a device. This handler will then be
5090 * called from __netif_receive_skb. A negative errno code is returned
5093 * The caller must hold the rtnl_mutex.
5095 * For a general description of rx_handler, see enum rx_handler_result.
5097 int netdev_rx_handler_register(struct net_device *dev,
5098 rx_handler_func_t *rx_handler,
5099 void *rx_handler_data)
5101 if (netdev_is_rx_handler_busy(dev))
5104 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5107 /* Note: rx_handler_data must be set before rx_handler */
5108 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5109 rcu_assign_pointer(dev->rx_handler, rx_handler);
5113 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5116 * netdev_rx_handler_unregister - unregister receive handler
5117 * @dev: device to unregister a handler from
5119 * Unregister a receive handler from a device.
5121 * The caller must hold the rtnl_mutex.
5123 void netdev_rx_handler_unregister(struct net_device *dev)
5127 RCU_INIT_POINTER(dev->rx_handler, NULL);
5128 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5129 * section has a guarantee to see a non NULL rx_handler_data
5133 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5135 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5138 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5139 * the special handling of PFMEMALLOC skbs.
5141 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5143 switch (skb->protocol) {
5144 case htons(ETH_P_ARP):
5145 case htons(ETH_P_IP):
5146 case htons(ETH_P_IPV6):
5147 case htons(ETH_P_8021Q):
5148 case htons(ETH_P_8021AD):
5155 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5156 int *ret, struct net_device *orig_dev)
5158 if (nf_hook_ingress_active(skb)) {
5162 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5167 ingress_retval = nf_hook_ingress(skb);
5169 return ingress_retval;
5174 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5175 struct packet_type **ppt_prev)
5177 struct packet_type *ptype, *pt_prev;
5178 rx_handler_func_t *rx_handler;
5179 struct sk_buff *skb = *pskb;
5180 struct net_device *orig_dev;
5181 bool deliver_exact = false;
5182 int ret = NET_RX_DROP;
5185 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5187 trace_netif_receive_skb(skb);
5189 orig_dev = skb->dev;
5191 skb_reset_network_header(skb);
5192 if (!skb_transport_header_was_set(skb))
5193 skb_reset_transport_header(skb);
5194 skb_reset_mac_len(skb);
5199 skb->skb_iif = skb->dev->ifindex;
5201 __this_cpu_inc(softnet_data.processed);
5203 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5207 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5210 if (ret2 != XDP_PASS) {
5214 skb_reset_mac_len(skb);
5217 if (eth_type_vlan(skb->protocol)) {
5218 skb = skb_vlan_untag(skb);
5223 if (skb_skip_tc_classify(skb))
5229 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5231 ret = deliver_skb(skb, pt_prev, orig_dev);
5235 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5237 ret = deliver_skb(skb, pt_prev, orig_dev);
5242 #ifdef CONFIG_NET_INGRESS
5243 if (static_branch_unlikely(&ingress_needed_key)) {
5244 bool another = false;
5246 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5253 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5257 skb_reset_redirect(skb);
5259 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5262 if (skb_vlan_tag_present(skb)) {
5264 ret = deliver_skb(skb, pt_prev, orig_dev);
5267 if (vlan_do_receive(&skb))
5269 else if (unlikely(!skb))
5273 rx_handler = rcu_dereference(skb->dev->rx_handler);
5276 ret = deliver_skb(skb, pt_prev, orig_dev);
5279 switch (rx_handler(&skb)) {
5280 case RX_HANDLER_CONSUMED:
5281 ret = NET_RX_SUCCESS;
5283 case RX_HANDLER_ANOTHER:
5285 case RX_HANDLER_EXACT:
5286 deliver_exact = true;
5287 case RX_HANDLER_PASS:
5294 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5296 if (skb_vlan_tag_get_id(skb)) {
5297 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5300 skb->pkt_type = PACKET_OTHERHOST;
5301 } else if (eth_type_vlan(skb->protocol)) {
5302 /* Outer header is 802.1P with vlan 0, inner header is
5303 * 802.1Q or 802.1AD and vlan_do_receive() above could
5304 * not find vlan dev for vlan id 0.
5306 __vlan_hwaccel_clear_tag(skb);
5307 skb = skb_vlan_untag(skb);
5310 if (vlan_do_receive(&skb))
5311 /* After stripping off 802.1P header with vlan 0
5312 * vlan dev is found for inner header.
5315 else if (unlikely(!skb))
5318 /* We have stripped outer 802.1P vlan 0 header.
5319 * But could not find vlan dev.
5320 * check again for vlan id to set OTHERHOST.
5324 /* Note: we might in the future use prio bits
5325 * and set skb->priority like in vlan_do_receive()
5326 * For the time being, just ignore Priority Code Point
5328 __vlan_hwaccel_clear_tag(skb);
5331 type = skb->protocol;
5333 /* deliver only exact match when indicated */
5334 if (likely(!deliver_exact)) {
5335 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5336 &ptype_base[ntohs(type) &
5340 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5341 &orig_dev->ptype_specific);
5343 if (unlikely(skb->dev != orig_dev)) {
5344 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5345 &skb->dev->ptype_specific);
5349 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5351 *ppt_prev = pt_prev;
5355 atomic_long_inc(&skb->dev->rx_dropped);
5357 atomic_long_inc(&skb->dev->rx_nohandler);
5359 /* Jamal, now you will not able to escape explaining
5360 * me how you were going to use this. :-)
5366 /* The invariant here is that if *ppt_prev is not NULL
5367 * then skb should also be non-NULL.
5369 * Apparently *ppt_prev assignment above holds this invariant due to
5370 * skb dereferencing near it.
5376 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5378 struct net_device *orig_dev = skb->dev;
5379 struct packet_type *pt_prev = NULL;
5382 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5384 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5385 skb->dev, pt_prev, orig_dev);
5390 * netif_receive_skb_core - special purpose version of netif_receive_skb
5391 * @skb: buffer to process
5393 * More direct receive version of netif_receive_skb(). It should
5394 * only be used by callers that have a need to skip RPS and Generic XDP.
5395 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5397 * This function may only be called from softirq context and interrupts
5398 * should be enabled.
5400 * Return values (usually ignored):
5401 * NET_RX_SUCCESS: no congestion
5402 * NET_RX_DROP: packet was dropped
5404 int netif_receive_skb_core(struct sk_buff *skb)
5409 ret = __netif_receive_skb_one_core(skb, false);
5414 EXPORT_SYMBOL(netif_receive_skb_core);
5416 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5417 struct packet_type *pt_prev,
5418 struct net_device *orig_dev)
5420 struct sk_buff *skb, *next;
5424 if (list_empty(head))
5426 if (pt_prev->list_func != NULL)
5427 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5428 ip_list_rcv, head, pt_prev, orig_dev);
5430 list_for_each_entry_safe(skb, next, head, list) {
5431 skb_list_del_init(skb);
5432 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5436 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5438 /* Fast-path assumptions:
5439 * - There is no RX handler.
5440 * - Only one packet_type matches.
5441 * If either of these fails, we will end up doing some per-packet
5442 * processing in-line, then handling the 'last ptype' for the whole
5443 * sublist. This can't cause out-of-order delivery to any single ptype,
5444 * because the 'last ptype' must be constant across the sublist, and all
5445 * other ptypes are handled per-packet.
5447 /* Current (common) ptype of sublist */
5448 struct packet_type *pt_curr = NULL;
5449 /* Current (common) orig_dev of sublist */
5450 struct net_device *od_curr = NULL;
5451 struct list_head sublist;
5452 struct sk_buff *skb, *next;
5454 INIT_LIST_HEAD(&sublist);
5455 list_for_each_entry_safe(skb, next, head, list) {
5456 struct net_device *orig_dev = skb->dev;
5457 struct packet_type *pt_prev = NULL;
5459 skb_list_del_init(skb);
5460 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5463 if (pt_curr != pt_prev || od_curr != orig_dev) {
5464 /* dispatch old sublist */
5465 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5466 /* start new sublist */
5467 INIT_LIST_HEAD(&sublist);
5471 list_add_tail(&skb->list, &sublist);
5474 /* dispatch final sublist */
5475 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5478 static int __netif_receive_skb(struct sk_buff *skb)
5482 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5483 unsigned int noreclaim_flag;
5486 * PFMEMALLOC skbs are special, they should
5487 * - be delivered to SOCK_MEMALLOC sockets only
5488 * - stay away from userspace
5489 * - have bounded memory usage
5491 * Use PF_MEMALLOC as this saves us from propagating the allocation
5492 * context down to all allocation sites.
5494 noreclaim_flag = memalloc_noreclaim_save();
5495 ret = __netif_receive_skb_one_core(skb, true);
5496 memalloc_noreclaim_restore(noreclaim_flag);
5498 ret = __netif_receive_skb_one_core(skb, false);
5503 static void __netif_receive_skb_list(struct list_head *head)
5505 unsigned long noreclaim_flag = 0;
5506 struct sk_buff *skb, *next;
5507 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5509 list_for_each_entry_safe(skb, next, head, list) {
5510 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5511 struct list_head sublist;
5513 /* Handle the previous sublist */
5514 list_cut_before(&sublist, head, &skb->list);
5515 if (!list_empty(&sublist))
5516 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5517 pfmemalloc = !pfmemalloc;
5518 /* See comments in __netif_receive_skb */
5520 noreclaim_flag = memalloc_noreclaim_save();
5522 memalloc_noreclaim_restore(noreclaim_flag);
5525 /* Handle the remaining sublist */
5526 if (!list_empty(head))
5527 __netif_receive_skb_list_core(head, pfmemalloc);
5528 /* Restore pflags */
5530 memalloc_noreclaim_restore(noreclaim_flag);
5533 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5535 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5536 struct bpf_prog *new = xdp->prog;
5542 mutex_lock(&new->aux->used_maps_mutex);
5544 /* generic XDP does not work with DEVMAPs that can
5545 * have a bpf_prog installed on an entry
5547 for (i = 0; i < new->aux->used_map_cnt; i++) {
5548 if (dev_map_can_have_prog(new->aux->used_maps[i]) ||
5549 cpu_map_prog_allowed(new->aux->used_maps[i])) {
5550 mutex_unlock(&new->aux->used_maps_mutex);
5555 mutex_unlock(&new->aux->used_maps_mutex);
5558 switch (xdp->command) {
5559 case XDP_SETUP_PROG:
5560 rcu_assign_pointer(dev->xdp_prog, new);
5565 static_branch_dec(&generic_xdp_needed_key);
5566 } else if (new && !old) {
5567 static_branch_inc(&generic_xdp_needed_key);
5568 dev_disable_lro(dev);
5569 dev_disable_gro_hw(dev);
5581 static int netif_receive_skb_internal(struct sk_buff *skb)
5585 net_timestamp_check(netdev_tstamp_prequeue, skb);
5587 if (skb_defer_rx_timestamp(skb))
5588 return NET_RX_SUCCESS;
5592 if (static_branch_unlikely(&rps_needed)) {
5593 struct rps_dev_flow voidflow, *rflow = &voidflow;
5594 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5597 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5603 ret = __netif_receive_skb(skb);
5608 static void netif_receive_skb_list_internal(struct list_head *head)
5610 struct sk_buff *skb, *next;
5611 struct list_head sublist;
5613 INIT_LIST_HEAD(&sublist);
5614 list_for_each_entry_safe(skb, next, head, list) {
5615 net_timestamp_check(netdev_tstamp_prequeue, skb);
5616 skb_list_del_init(skb);
5617 if (!skb_defer_rx_timestamp(skb))
5618 list_add_tail(&skb->list, &sublist);
5620 list_splice_init(&sublist, head);
5624 if (static_branch_unlikely(&rps_needed)) {
5625 list_for_each_entry_safe(skb, next, head, list) {
5626 struct rps_dev_flow voidflow, *rflow = &voidflow;
5627 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5630 /* Will be handled, remove from list */
5631 skb_list_del_init(skb);
5632 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5637 __netif_receive_skb_list(head);
5642 * netif_receive_skb - process receive buffer from network
5643 * @skb: buffer to process
5645 * netif_receive_skb() is the main receive data processing function.
5646 * It always succeeds. The buffer may be dropped during processing
5647 * for congestion control or by the protocol layers.
5649 * This function may only be called from softirq context and interrupts
5650 * should be enabled.
5652 * Return values (usually ignored):
5653 * NET_RX_SUCCESS: no congestion
5654 * NET_RX_DROP: packet was dropped
5656 int netif_receive_skb(struct sk_buff *skb)
5660 trace_netif_receive_skb_entry(skb);
5662 ret = netif_receive_skb_internal(skb);
5663 trace_netif_receive_skb_exit(ret);
5667 EXPORT_SYMBOL(netif_receive_skb);
5670 * netif_receive_skb_list - process many receive buffers from network
5671 * @head: list of skbs to process.
5673 * Since return value of netif_receive_skb() is normally ignored, and
5674 * wouldn't be meaningful for a list, this function returns void.
5676 * This function may only be called from softirq context and interrupts
5677 * should be enabled.
5679 void netif_receive_skb_list(struct list_head *head)
5681 struct sk_buff *skb;
5683 if (list_empty(head))
5685 if (trace_netif_receive_skb_list_entry_enabled()) {
5686 list_for_each_entry(skb, head, list)
5687 trace_netif_receive_skb_list_entry(skb);
5689 netif_receive_skb_list_internal(head);
5690 trace_netif_receive_skb_list_exit(0);
5692 EXPORT_SYMBOL(netif_receive_skb_list);
5694 static DEFINE_PER_CPU(struct work_struct, flush_works);
5696 /* Network device is going away, flush any packets still pending */
5697 static void flush_backlog(struct work_struct *work)
5699 struct sk_buff *skb, *tmp;
5700 struct softnet_data *sd;
5703 sd = this_cpu_ptr(&softnet_data);
5705 local_irq_disable();
5707 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5708 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5709 __skb_unlink(skb, &sd->input_pkt_queue);
5710 dev_kfree_skb_irq(skb);
5711 input_queue_head_incr(sd);
5717 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5718 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5719 __skb_unlink(skb, &sd->process_queue);
5721 input_queue_head_incr(sd);
5727 static bool flush_required(int cpu)
5729 #if IS_ENABLED(CONFIG_RPS)
5730 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5733 local_irq_disable();
5736 /* as insertion into process_queue happens with the rps lock held,
5737 * process_queue access may race only with dequeue
5739 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5740 !skb_queue_empty_lockless(&sd->process_queue);
5746 /* without RPS we can't safely check input_pkt_queue: during a
5747 * concurrent remote skb_queue_splice() we can detect as empty both
5748 * input_pkt_queue and process_queue even if the latter could end-up
5749 * containing a lot of packets.
5754 static void flush_all_backlogs(void)
5756 static cpumask_t flush_cpus;
5759 /* since we are under rtnl lock protection we can use static data
5760 * for the cpumask and avoid allocating on stack the possibly
5767 cpumask_clear(&flush_cpus);
5768 for_each_online_cpu(cpu) {
5769 if (flush_required(cpu)) {
5770 queue_work_on(cpu, system_highpri_wq,
5771 per_cpu_ptr(&flush_works, cpu));
5772 cpumask_set_cpu(cpu, &flush_cpus);
5776 /* we can have in flight packet[s] on the cpus we are not flushing,
5777 * synchronize_net() in unregister_netdevice_many() will take care of
5780 for_each_cpu(cpu, &flush_cpus)
5781 flush_work(per_cpu_ptr(&flush_works, cpu));
5786 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5787 static void gro_normal_list(struct napi_struct *napi)
5789 if (!napi->rx_count)
5791 netif_receive_skb_list_internal(&napi->rx_list);
5792 INIT_LIST_HEAD(&napi->rx_list);
5796 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5797 * pass the whole batch up to the stack.
5799 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5801 list_add_tail(&skb->list, &napi->rx_list);
5802 napi->rx_count += segs;
5803 if (napi->rx_count >= gro_normal_batch)
5804 gro_normal_list(napi);
5807 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5809 struct packet_offload *ptype;
5810 __be16 type = skb->protocol;
5811 struct list_head *head = &offload_base;
5814 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5816 if (NAPI_GRO_CB(skb)->count == 1) {
5817 skb_shinfo(skb)->gso_size = 0;
5822 list_for_each_entry_rcu(ptype, head, list) {
5823 if (ptype->type != type || !ptype->callbacks.gro_complete)
5826 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5827 ipv6_gro_complete, inet_gro_complete,
5834 WARN_ON(&ptype->list == head);
5836 return NET_RX_SUCCESS;
5840 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5841 return NET_RX_SUCCESS;
5844 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5847 struct list_head *head = &napi->gro_hash[index].list;
5848 struct sk_buff *skb, *p;
5850 list_for_each_entry_safe_reverse(skb, p, head, list) {
5851 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5853 skb_list_del_init(skb);
5854 napi_gro_complete(napi, skb);
5855 napi->gro_hash[index].count--;
5858 if (!napi->gro_hash[index].count)
5859 __clear_bit(index, &napi->gro_bitmask);
5862 /* napi->gro_hash[].list contains packets ordered by age.
5863 * youngest packets at the head of it.
5864 * Complete skbs in reverse order to reduce latencies.
5866 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5868 unsigned long bitmask = napi->gro_bitmask;
5869 unsigned int i, base = ~0U;
5871 while ((i = ffs(bitmask)) != 0) {
5874 __napi_gro_flush_chain(napi, base, flush_old);
5877 EXPORT_SYMBOL(napi_gro_flush);
5879 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5880 struct sk_buff *skb)
5882 unsigned int maclen = skb->dev->hard_header_len;
5883 u32 hash = skb_get_hash_raw(skb);
5884 struct list_head *head;
5887 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5888 list_for_each_entry(p, head, list) {
5889 unsigned long diffs;
5891 NAPI_GRO_CB(p)->flush = 0;
5893 if (hash != skb_get_hash_raw(p)) {
5894 NAPI_GRO_CB(p)->same_flow = 0;
5898 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5899 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5900 if (skb_vlan_tag_present(p))
5901 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5902 diffs |= skb_metadata_dst_cmp(p, skb);
5903 diffs |= skb_metadata_differs(p, skb);
5904 if (maclen == ETH_HLEN)
5905 diffs |= compare_ether_header(skb_mac_header(p),
5906 skb_mac_header(skb));
5908 diffs = memcmp(skb_mac_header(p),
5909 skb_mac_header(skb),
5911 NAPI_GRO_CB(p)->same_flow = !diffs;
5917 static void skb_gro_reset_offset(struct sk_buff *skb)
5919 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5920 const skb_frag_t *frag0 = &pinfo->frags[0];
5922 NAPI_GRO_CB(skb)->data_offset = 0;
5923 NAPI_GRO_CB(skb)->frag0 = NULL;
5924 NAPI_GRO_CB(skb)->frag0_len = 0;
5926 if (!skb_headlen(skb) && pinfo->nr_frags &&
5927 !PageHighMem(skb_frag_page(frag0))) {
5928 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5929 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5930 skb_frag_size(frag0),
5931 skb->end - skb->tail);
5935 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5937 struct skb_shared_info *pinfo = skb_shinfo(skb);
5939 BUG_ON(skb->end - skb->tail < grow);
5941 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5943 skb->data_len -= grow;
5946 skb_frag_off_add(&pinfo->frags[0], grow);
5947 skb_frag_size_sub(&pinfo->frags[0], grow);
5949 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5950 skb_frag_unref(skb, 0);
5951 memmove(pinfo->frags, pinfo->frags + 1,
5952 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5956 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5958 struct sk_buff *oldest;
5960 oldest = list_last_entry(head, struct sk_buff, list);
5962 /* We are called with head length >= MAX_GRO_SKBS, so this is
5965 if (WARN_ON_ONCE(!oldest))
5968 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5971 skb_list_del_init(oldest);
5972 napi_gro_complete(napi, oldest);
5975 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5977 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5978 struct list_head *head = &offload_base;
5979 struct packet_offload *ptype;
5980 __be16 type = skb->protocol;
5981 struct list_head *gro_head;
5982 struct sk_buff *pp = NULL;
5983 enum gro_result ret;
5987 if (netif_elide_gro(skb->dev))
5990 gro_head = gro_list_prepare(napi, skb);
5993 list_for_each_entry_rcu(ptype, head, list) {
5994 if (ptype->type != type || !ptype->callbacks.gro_receive)
5997 skb_set_network_header(skb, skb_gro_offset(skb));
5998 skb_reset_mac_len(skb);
5999 NAPI_GRO_CB(skb)->same_flow = 0;
6000 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6001 NAPI_GRO_CB(skb)->free = 0;
6002 NAPI_GRO_CB(skb)->encap_mark = 0;
6003 NAPI_GRO_CB(skb)->recursion_counter = 0;
6004 NAPI_GRO_CB(skb)->is_fou = 0;
6005 NAPI_GRO_CB(skb)->is_atomic = 1;
6006 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6008 /* Setup for GRO checksum validation */
6009 switch (skb->ip_summed) {
6010 case CHECKSUM_COMPLETE:
6011 NAPI_GRO_CB(skb)->csum = skb->csum;
6012 NAPI_GRO_CB(skb)->csum_valid = 1;
6013 NAPI_GRO_CB(skb)->csum_cnt = 0;
6015 case CHECKSUM_UNNECESSARY:
6016 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6017 NAPI_GRO_CB(skb)->csum_valid = 0;
6020 NAPI_GRO_CB(skb)->csum_cnt = 0;
6021 NAPI_GRO_CB(skb)->csum_valid = 0;
6024 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6025 ipv6_gro_receive, inet_gro_receive,
6031 if (&ptype->list == head)
6034 if (PTR_ERR(pp) == -EINPROGRESS) {
6039 same_flow = NAPI_GRO_CB(skb)->same_flow;
6040 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6043 skb_list_del_init(pp);
6044 napi_gro_complete(napi, pp);
6045 napi->gro_hash[hash].count--;
6051 if (NAPI_GRO_CB(skb)->flush)
6054 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
6055 gro_flush_oldest(napi, gro_head);
6057 napi->gro_hash[hash].count++;
6059 NAPI_GRO_CB(skb)->count = 1;
6060 NAPI_GRO_CB(skb)->age = jiffies;
6061 NAPI_GRO_CB(skb)->last = skb;
6062 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6063 list_add(&skb->list, gro_head);
6067 grow = skb_gro_offset(skb) - skb_headlen(skb);
6069 gro_pull_from_frag0(skb, grow);
6071 if (napi->gro_hash[hash].count) {
6072 if (!test_bit(hash, &napi->gro_bitmask))
6073 __set_bit(hash, &napi->gro_bitmask);
6074 } else if (test_bit(hash, &napi->gro_bitmask)) {
6075 __clear_bit(hash, &napi->gro_bitmask);
6085 struct packet_offload *gro_find_receive_by_type(__be16 type)
6087 struct list_head *offload_head = &offload_base;
6088 struct packet_offload *ptype;
6090 list_for_each_entry_rcu(ptype, offload_head, list) {
6091 if (ptype->type != type || !ptype->callbacks.gro_receive)
6097 EXPORT_SYMBOL(gro_find_receive_by_type);
6099 struct packet_offload *gro_find_complete_by_type(__be16 type)
6101 struct list_head *offload_head = &offload_base;
6102 struct packet_offload *ptype;
6104 list_for_each_entry_rcu(ptype, offload_head, list) {
6105 if (ptype->type != type || !ptype->callbacks.gro_complete)
6111 EXPORT_SYMBOL(gro_find_complete_by_type);
6113 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6114 struct sk_buff *skb,
6119 gro_normal_one(napi, skb, 1);
6122 case GRO_MERGED_FREE:
6123 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6124 napi_skb_free_stolen_head(skb);
6126 __kfree_skb_defer(skb);
6138 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6142 skb_mark_napi_id(skb, napi);
6143 trace_napi_gro_receive_entry(skb);
6145 skb_gro_reset_offset(skb);
6147 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6148 trace_napi_gro_receive_exit(ret);
6152 EXPORT_SYMBOL(napi_gro_receive);
6154 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6156 if (unlikely(skb->pfmemalloc)) {
6160 __skb_pull(skb, skb_headlen(skb));
6161 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6162 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6163 __vlan_hwaccel_clear_tag(skb);
6164 skb->dev = napi->dev;
6167 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6168 skb->pkt_type = PACKET_HOST;
6170 skb->encapsulation = 0;
6171 skb_shinfo(skb)->gso_type = 0;
6172 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6178 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6180 struct sk_buff *skb = napi->skb;
6183 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6186 skb_mark_napi_id(skb, napi);
6191 EXPORT_SYMBOL(napi_get_frags);
6193 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6194 struct sk_buff *skb,
6200 __skb_push(skb, ETH_HLEN);
6201 skb->protocol = eth_type_trans(skb, skb->dev);
6202 if (ret == GRO_NORMAL)
6203 gro_normal_one(napi, skb, 1);
6206 case GRO_MERGED_FREE:
6207 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6208 napi_skb_free_stolen_head(skb);
6210 napi_reuse_skb(napi, skb);
6221 /* Upper GRO stack assumes network header starts at gro_offset=0
6222 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6223 * We copy ethernet header into skb->data to have a common layout.
6225 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6227 struct sk_buff *skb = napi->skb;
6228 const struct ethhdr *eth;
6229 unsigned int hlen = sizeof(*eth);
6233 skb_reset_mac_header(skb);
6234 skb_gro_reset_offset(skb);
6236 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6237 eth = skb_gro_header_slow(skb, hlen, 0);
6238 if (unlikely(!eth)) {
6239 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6240 __func__, napi->dev->name);
6241 napi_reuse_skb(napi, skb);
6245 eth = (const struct ethhdr *)skb->data;
6246 gro_pull_from_frag0(skb, hlen);
6247 NAPI_GRO_CB(skb)->frag0 += hlen;
6248 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6250 __skb_pull(skb, hlen);
6253 * This works because the only protocols we care about don't require
6255 * We'll fix it up properly in napi_frags_finish()
6257 skb->protocol = eth->h_proto;
6262 gro_result_t napi_gro_frags(struct napi_struct *napi)
6265 struct sk_buff *skb = napi_frags_skb(napi);
6267 trace_napi_gro_frags_entry(skb);
6269 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6270 trace_napi_gro_frags_exit(ret);
6274 EXPORT_SYMBOL(napi_gro_frags);
6276 /* Compute the checksum from gro_offset and return the folded value
6277 * after adding in any pseudo checksum.
6279 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6284 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6286 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6287 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6288 /* See comments in __skb_checksum_complete(). */
6290 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6291 !skb->csum_complete_sw)
6292 netdev_rx_csum_fault(skb->dev, skb);
6295 NAPI_GRO_CB(skb)->csum = wsum;
6296 NAPI_GRO_CB(skb)->csum_valid = 1;
6300 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6302 static void net_rps_send_ipi(struct softnet_data *remsd)
6306 struct softnet_data *next = remsd->rps_ipi_next;
6308 if (cpu_online(remsd->cpu))
6309 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6316 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6317 * Note: called with local irq disabled, but exits with local irq enabled.
6319 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6322 struct softnet_data *remsd = sd->rps_ipi_list;
6325 sd->rps_ipi_list = NULL;
6329 /* Send pending IPI's to kick RPS processing on remote cpus. */
6330 net_rps_send_ipi(remsd);
6336 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6339 return sd->rps_ipi_list != NULL;
6345 static int process_backlog(struct napi_struct *napi, int quota)
6347 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6351 /* Check if we have pending ipi, its better to send them now,
6352 * not waiting net_rx_action() end.
6354 if (sd_has_rps_ipi_waiting(sd)) {
6355 local_irq_disable();
6356 net_rps_action_and_irq_enable(sd);
6359 napi->weight = dev_rx_weight;
6361 struct sk_buff *skb;
6363 while ((skb = __skb_dequeue(&sd->process_queue))) {
6365 __netif_receive_skb(skb);
6367 input_queue_head_incr(sd);
6368 if (++work >= quota)
6373 local_irq_disable();
6375 if (skb_queue_empty(&sd->input_pkt_queue)) {
6377 * Inline a custom version of __napi_complete().
6378 * only current cpu owns and manipulates this napi,
6379 * and NAPI_STATE_SCHED is the only possible flag set
6381 * We can use a plain write instead of clear_bit(),
6382 * and we dont need an smp_mb() memory barrier.
6387 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6388 &sd->process_queue);
6398 * __napi_schedule - schedule for receive
6399 * @n: entry to schedule
6401 * The entry's receive function will be scheduled to run.
6402 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6404 void __napi_schedule(struct napi_struct *n)
6406 unsigned long flags;
6408 local_irq_save(flags);
6409 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6410 local_irq_restore(flags);
6412 EXPORT_SYMBOL(__napi_schedule);
6415 * napi_schedule_prep - check if napi can be scheduled
6418 * Test if NAPI routine is already running, and if not mark
6419 * it as running. This is used as a condition variable to
6420 * insure only one NAPI poll instance runs. We also make
6421 * sure there is no pending NAPI disable.
6423 bool napi_schedule_prep(struct napi_struct *n)
6425 unsigned long val, new;
6428 val = READ_ONCE(n->state);
6429 if (unlikely(val & NAPIF_STATE_DISABLE))
6431 new = val | NAPIF_STATE_SCHED;
6433 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6434 * This was suggested by Alexander Duyck, as compiler
6435 * emits better code than :
6436 * if (val & NAPIF_STATE_SCHED)
6437 * new |= NAPIF_STATE_MISSED;
6439 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6441 } while (cmpxchg(&n->state, val, new) != val);
6443 return !(val & NAPIF_STATE_SCHED);
6445 EXPORT_SYMBOL(napi_schedule_prep);
6448 * __napi_schedule_irqoff - schedule for receive
6449 * @n: entry to schedule
6451 * Variant of __napi_schedule() assuming hard irqs are masked
6453 void __napi_schedule_irqoff(struct napi_struct *n)
6455 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6457 EXPORT_SYMBOL(__napi_schedule_irqoff);
6459 bool napi_complete_done(struct napi_struct *n, int work_done)
6461 unsigned long flags, val, new, timeout = 0;
6465 * 1) Don't let napi dequeue from the cpu poll list
6466 * just in case its running on a different cpu.
6467 * 2) If we are busy polling, do nothing here, we have
6468 * the guarantee we will be called later.
6470 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6471 NAPIF_STATE_IN_BUSY_POLL)))
6476 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6477 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6479 if (n->defer_hard_irqs_count > 0) {
6480 n->defer_hard_irqs_count--;
6481 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6485 if (n->gro_bitmask) {
6486 /* When the NAPI instance uses a timeout and keeps postponing
6487 * it, we need to bound somehow the time packets are kept in
6490 napi_gro_flush(n, !!timeout);
6495 if (unlikely(!list_empty(&n->poll_list))) {
6496 /* If n->poll_list is not empty, we need to mask irqs */
6497 local_irq_save(flags);
6498 list_del_init(&n->poll_list);
6499 local_irq_restore(flags);
6503 val = READ_ONCE(n->state);
6505 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6507 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6508 NAPIF_STATE_SCHED_THREADED |
6509 NAPIF_STATE_PREFER_BUSY_POLL);
6511 /* If STATE_MISSED was set, leave STATE_SCHED set,
6512 * because we will call napi->poll() one more time.
6513 * This C code was suggested by Alexander Duyck to help gcc.
6515 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6517 } while (cmpxchg(&n->state, val, new) != val);
6519 if (unlikely(val & NAPIF_STATE_MISSED)) {
6525 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6526 HRTIMER_MODE_REL_PINNED);
6529 EXPORT_SYMBOL(napi_complete_done);
6531 /* must be called under rcu_read_lock(), as we dont take a reference */
6532 static struct napi_struct *napi_by_id(unsigned int napi_id)
6534 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6535 struct napi_struct *napi;
6537 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6538 if (napi->napi_id == napi_id)
6544 #if defined(CONFIG_NET_RX_BUSY_POLL)
6546 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6548 if (!skip_schedule) {
6549 gro_normal_list(napi);
6550 __napi_schedule(napi);
6554 if (napi->gro_bitmask) {
6555 /* flush too old packets
6556 * If HZ < 1000, flush all packets.
6558 napi_gro_flush(napi, HZ >= 1000);
6561 gro_normal_list(napi);
6562 clear_bit(NAPI_STATE_SCHED, &napi->state);
6565 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6568 bool skip_schedule = false;
6569 unsigned long timeout;
6572 /* Busy polling means there is a high chance device driver hard irq
6573 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6574 * set in napi_schedule_prep().
6575 * Since we are about to call napi->poll() once more, we can safely
6576 * clear NAPI_STATE_MISSED.
6578 * Note: x86 could use a single "lock and ..." instruction
6579 * to perform these two clear_bit()
6581 clear_bit(NAPI_STATE_MISSED, &napi->state);
6582 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6586 if (prefer_busy_poll) {
6587 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6588 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6589 if (napi->defer_hard_irqs_count && timeout) {
6590 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6591 skip_schedule = true;
6595 /* All we really want here is to re-enable device interrupts.
6596 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6598 rc = napi->poll(napi, budget);
6599 /* We can't gro_normal_list() here, because napi->poll() might have
6600 * rearmed the napi (napi_complete_done()) in which case it could
6601 * already be running on another CPU.
6603 trace_napi_poll(napi, rc, budget);
6604 netpoll_poll_unlock(have_poll_lock);
6606 __busy_poll_stop(napi, skip_schedule);
6610 void napi_busy_loop(unsigned int napi_id,
6611 bool (*loop_end)(void *, unsigned long),
6612 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6614 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6615 int (*napi_poll)(struct napi_struct *napi, int budget);
6616 void *have_poll_lock = NULL;
6617 struct napi_struct *napi;
6624 napi = napi_by_id(napi_id);
6634 unsigned long val = READ_ONCE(napi->state);
6636 /* If multiple threads are competing for this napi,
6637 * we avoid dirtying napi->state as much as we can.
6639 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6640 NAPIF_STATE_IN_BUSY_POLL)) {
6641 if (prefer_busy_poll)
6642 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6645 if (cmpxchg(&napi->state, val,
6646 val | NAPIF_STATE_IN_BUSY_POLL |
6647 NAPIF_STATE_SCHED) != val) {
6648 if (prefer_busy_poll)
6649 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6652 have_poll_lock = netpoll_poll_lock(napi);
6653 napi_poll = napi->poll;
6655 work = napi_poll(napi, budget);
6656 trace_napi_poll(napi, work, budget);
6657 gro_normal_list(napi);
6660 __NET_ADD_STATS(dev_net(napi->dev),
6661 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6664 if (!loop_end || loop_end(loop_end_arg, start_time))
6667 if (unlikely(need_resched())) {
6669 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6673 if (loop_end(loop_end_arg, start_time))
6680 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6685 EXPORT_SYMBOL(napi_busy_loop);
6687 #endif /* CONFIG_NET_RX_BUSY_POLL */
6689 static void napi_hash_add(struct napi_struct *napi)
6691 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6694 spin_lock(&napi_hash_lock);
6696 /* 0..NR_CPUS range is reserved for sender_cpu use */
6698 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6699 napi_gen_id = MIN_NAPI_ID;
6700 } while (napi_by_id(napi_gen_id));
6701 napi->napi_id = napi_gen_id;
6703 hlist_add_head_rcu(&napi->napi_hash_node,
6704 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6706 spin_unlock(&napi_hash_lock);
6709 /* Warning : caller is responsible to make sure rcu grace period
6710 * is respected before freeing memory containing @napi
6712 static void napi_hash_del(struct napi_struct *napi)
6714 spin_lock(&napi_hash_lock);
6716 hlist_del_init_rcu(&napi->napi_hash_node);
6718 spin_unlock(&napi_hash_lock);
6721 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6723 struct napi_struct *napi;
6725 napi = container_of(timer, struct napi_struct, timer);
6727 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6728 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6730 if (!napi_disable_pending(napi) &&
6731 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6732 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6733 __napi_schedule_irqoff(napi);
6736 return HRTIMER_NORESTART;
6739 static void init_gro_hash(struct napi_struct *napi)
6743 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6744 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6745 napi->gro_hash[i].count = 0;
6747 napi->gro_bitmask = 0;
6750 int dev_set_threaded(struct net_device *dev, bool threaded)
6752 struct napi_struct *napi;
6755 if (dev->threaded == threaded)
6759 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6760 if (!napi->thread) {
6761 err = napi_kthread_create(napi);
6770 dev->threaded = threaded;
6772 /* Make sure kthread is created before THREADED bit
6775 smp_mb__before_atomic();
6777 /* Setting/unsetting threaded mode on a napi might not immediately
6778 * take effect, if the current napi instance is actively being
6779 * polled. In this case, the switch between threaded mode and
6780 * softirq mode will happen in the next round of napi_schedule().
6781 * This should not cause hiccups/stalls to the live traffic.
6783 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6785 set_bit(NAPI_STATE_THREADED, &napi->state);
6787 clear_bit(NAPI_STATE_THREADED, &napi->state);
6793 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6794 int (*poll)(struct napi_struct *, int), int weight)
6796 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6799 INIT_LIST_HEAD(&napi->poll_list);
6800 INIT_HLIST_NODE(&napi->napi_hash_node);
6801 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6802 napi->timer.function = napi_watchdog;
6803 init_gro_hash(napi);
6805 INIT_LIST_HEAD(&napi->rx_list);
6808 if (weight > NAPI_POLL_WEIGHT)
6809 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6811 napi->weight = weight;
6813 #ifdef CONFIG_NETPOLL
6814 napi->poll_owner = -1;
6816 set_bit(NAPI_STATE_SCHED, &napi->state);
6817 set_bit(NAPI_STATE_NPSVC, &napi->state);
6818 list_add_rcu(&napi->dev_list, &dev->napi_list);
6819 napi_hash_add(napi);
6820 /* Create kthread for this napi if dev->threaded is set.
6821 * Clear dev->threaded if kthread creation failed so that
6822 * threaded mode will not be enabled in napi_enable().
6824 if (dev->threaded && napi_kthread_create(napi))
6827 EXPORT_SYMBOL(netif_napi_add);
6829 void napi_disable(struct napi_struct *n)
6832 set_bit(NAPI_STATE_DISABLE, &n->state);
6834 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6836 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6839 hrtimer_cancel(&n->timer);
6841 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6842 clear_bit(NAPI_STATE_DISABLE, &n->state);
6843 clear_bit(NAPI_STATE_THREADED, &n->state);
6845 EXPORT_SYMBOL(napi_disable);
6848 * napi_enable - enable NAPI scheduling
6851 * Resume NAPI from being scheduled on this context.
6852 * Must be paired with napi_disable.
6854 void napi_enable(struct napi_struct *n)
6856 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
6857 smp_mb__before_atomic();
6858 clear_bit(NAPI_STATE_SCHED, &n->state);
6859 clear_bit(NAPI_STATE_NPSVC, &n->state);
6860 if (n->dev->threaded && n->thread)
6861 set_bit(NAPI_STATE_THREADED, &n->state);
6863 EXPORT_SYMBOL(napi_enable);
6865 static void flush_gro_hash(struct napi_struct *napi)
6869 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6870 struct sk_buff *skb, *n;
6872 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6874 napi->gro_hash[i].count = 0;
6878 /* Must be called in process context */
6879 void __netif_napi_del(struct napi_struct *napi)
6881 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6884 napi_hash_del(napi);
6885 list_del_rcu(&napi->dev_list);
6886 napi_free_frags(napi);
6888 flush_gro_hash(napi);
6889 napi->gro_bitmask = 0;
6892 kthread_stop(napi->thread);
6893 napi->thread = NULL;
6896 EXPORT_SYMBOL(__netif_napi_del);
6898 static int __napi_poll(struct napi_struct *n, bool *repoll)
6904 /* This NAPI_STATE_SCHED test is for avoiding a race
6905 * with netpoll's poll_napi(). Only the entity which
6906 * obtains the lock and sees NAPI_STATE_SCHED set will
6907 * actually make the ->poll() call. Therefore we avoid
6908 * accidentally calling ->poll() when NAPI is not scheduled.
6911 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6912 work = n->poll(n, weight);
6913 trace_napi_poll(n, work, weight);
6916 if (unlikely(work > weight))
6917 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6918 n->poll, work, weight);
6920 if (likely(work < weight))
6923 /* Drivers must not modify the NAPI state if they
6924 * consume the entire weight. In such cases this code
6925 * still "owns" the NAPI instance and therefore can
6926 * move the instance around on the list at-will.
6928 if (unlikely(napi_disable_pending(n))) {
6933 /* The NAPI context has more processing work, but busy-polling
6934 * is preferred. Exit early.
6936 if (napi_prefer_busy_poll(n)) {
6937 if (napi_complete_done(n, work)) {
6938 /* If timeout is not set, we need to make sure
6939 * that the NAPI is re-scheduled.
6946 if (n->gro_bitmask) {
6947 /* flush too old packets
6948 * If HZ < 1000, flush all packets.
6950 napi_gro_flush(n, HZ >= 1000);
6955 /* Some drivers may have called napi_schedule
6956 * prior to exhausting their budget.
6958 if (unlikely(!list_empty(&n->poll_list))) {
6959 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6960 n->dev ? n->dev->name : "backlog");
6969 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6971 bool do_repoll = false;
6975 list_del_init(&n->poll_list);
6977 have = netpoll_poll_lock(n);
6979 work = __napi_poll(n, &do_repoll);
6982 list_add_tail(&n->poll_list, repoll);
6984 netpoll_poll_unlock(have);
6989 static int napi_thread_wait(struct napi_struct *napi)
6993 set_current_state(TASK_INTERRUPTIBLE);
6995 while (!kthread_should_stop()) {
6996 /* Testing SCHED_THREADED bit here to make sure the current
6997 * kthread owns this napi and could poll on this napi.
6998 * Testing SCHED bit is not enough because SCHED bit might be
6999 * set by some other busy poll thread or by napi_disable().
7001 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7002 WARN_ON(!list_empty(&napi->poll_list));
7003 __set_current_state(TASK_RUNNING);
7008 /* woken being true indicates this thread owns this napi. */
7010 set_current_state(TASK_INTERRUPTIBLE);
7012 __set_current_state(TASK_RUNNING);
7017 static int napi_threaded_poll(void *data)
7019 struct napi_struct *napi = data;
7022 while (!napi_thread_wait(napi)) {
7024 bool repoll = false;
7028 have = netpoll_poll_lock(napi);
7029 __napi_poll(napi, &repoll);
7030 netpoll_poll_unlock(have);
7043 static __latent_entropy void net_rx_action(struct softirq_action *h)
7045 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7046 unsigned long time_limit = jiffies +
7047 usecs_to_jiffies(netdev_budget_usecs);
7048 int budget = netdev_budget;
7052 local_irq_disable();
7053 list_splice_init(&sd->poll_list, &list);
7057 struct napi_struct *n;
7059 if (list_empty(&list)) {
7060 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7065 n = list_first_entry(&list, struct napi_struct, poll_list);
7066 budget -= napi_poll(n, &repoll);
7068 /* If softirq window is exhausted then punt.
7069 * Allow this to run for 2 jiffies since which will allow
7070 * an average latency of 1.5/HZ.
7072 if (unlikely(budget <= 0 ||
7073 time_after_eq(jiffies, time_limit))) {
7079 local_irq_disable();
7081 list_splice_tail_init(&sd->poll_list, &list);
7082 list_splice_tail(&repoll, &list);
7083 list_splice(&list, &sd->poll_list);
7084 if (!list_empty(&sd->poll_list))
7085 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7087 net_rps_action_and_irq_enable(sd);
7090 struct netdev_adjacent {
7091 struct net_device *dev;
7093 /* upper master flag, there can only be one master device per list */
7096 /* lookup ignore flag */
7099 /* counter for the number of times this device was added to us */
7102 /* private field for the users */
7105 struct list_head list;
7106 struct rcu_head rcu;
7109 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7110 struct list_head *adj_list)
7112 struct netdev_adjacent *adj;
7114 list_for_each_entry(adj, adj_list, list) {
7115 if (adj->dev == adj_dev)
7121 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7122 struct netdev_nested_priv *priv)
7124 struct net_device *dev = (struct net_device *)priv->data;
7126 return upper_dev == dev;
7130 * netdev_has_upper_dev - Check if device is linked to an upper device
7132 * @upper_dev: upper device to check
7134 * Find out if a device is linked to specified upper device and return true
7135 * in case it is. Note that this checks only immediate upper device,
7136 * not through a complete stack of devices. The caller must hold the RTNL lock.
7138 bool netdev_has_upper_dev(struct net_device *dev,
7139 struct net_device *upper_dev)
7141 struct netdev_nested_priv priv = {
7142 .data = (void *)upper_dev,
7147 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7150 EXPORT_SYMBOL(netdev_has_upper_dev);
7153 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7155 * @upper_dev: upper device to check
7157 * Find out if a device is linked to specified upper device and return true
7158 * in case it is. Note that this checks the entire upper device chain.
7159 * The caller must hold rcu lock.
7162 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7163 struct net_device *upper_dev)
7165 struct netdev_nested_priv priv = {
7166 .data = (void *)upper_dev,
7169 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7172 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7175 * netdev_has_any_upper_dev - Check if device is linked to some device
7178 * Find out if a device is linked to an upper device and return true in case
7179 * it is. The caller must hold the RTNL lock.
7181 bool netdev_has_any_upper_dev(struct net_device *dev)
7185 return !list_empty(&dev->adj_list.upper);
7187 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7190 * netdev_master_upper_dev_get - Get master upper device
7193 * Find a master upper device and return pointer to it or NULL in case
7194 * it's not there. The caller must hold the RTNL lock.
7196 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7198 struct netdev_adjacent *upper;
7202 if (list_empty(&dev->adj_list.upper))
7205 upper = list_first_entry(&dev->adj_list.upper,
7206 struct netdev_adjacent, list);
7207 if (likely(upper->master))
7211 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7213 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7215 struct netdev_adjacent *upper;
7219 if (list_empty(&dev->adj_list.upper))
7222 upper = list_first_entry(&dev->adj_list.upper,
7223 struct netdev_adjacent, list);
7224 if (likely(upper->master) && !upper->ignore)
7230 * netdev_has_any_lower_dev - Check if device is linked to some device
7233 * Find out if a device is linked to a lower device and return true in case
7234 * it is. The caller must hold the RTNL lock.
7236 static bool netdev_has_any_lower_dev(struct net_device *dev)
7240 return !list_empty(&dev->adj_list.lower);
7243 void *netdev_adjacent_get_private(struct list_head *adj_list)
7245 struct netdev_adjacent *adj;
7247 adj = list_entry(adj_list, struct netdev_adjacent, list);
7249 return adj->private;
7251 EXPORT_SYMBOL(netdev_adjacent_get_private);
7254 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7256 * @iter: list_head ** of the current position
7258 * Gets the next device from the dev's upper list, starting from iter
7259 * position. The caller must hold RCU read lock.
7261 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7262 struct list_head **iter)
7264 struct netdev_adjacent *upper;
7266 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7268 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7270 if (&upper->list == &dev->adj_list.upper)
7273 *iter = &upper->list;
7277 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7279 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7280 struct list_head **iter,
7283 struct netdev_adjacent *upper;
7285 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7287 if (&upper->list == &dev->adj_list.upper)
7290 *iter = &upper->list;
7291 *ignore = upper->ignore;
7296 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7297 struct list_head **iter)
7299 struct netdev_adjacent *upper;
7301 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7303 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7305 if (&upper->list == &dev->adj_list.upper)
7308 *iter = &upper->list;
7313 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7314 int (*fn)(struct net_device *dev,
7315 struct netdev_nested_priv *priv),
7316 struct netdev_nested_priv *priv)
7318 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7319 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7324 iter = &dev->adj_list.upper;
7328 ret = fn(now, priv);
7335 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7342 niter = &udev->adj_list.upper;
7343 dev_stack[cur] = now;
7344 iter_stack[cur++] = iter;
7351 next = dev_stack[--cur];
7352 niter = iter_stack[cur];
7362 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7363 int (*fn)(struct net_device *dev,
7364 struct netdev_nested_priv *priv),
7365 struct netdev_nested_priv *priv)
7367 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7368 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7372 iter = &dev->adj_list.upper;
7376 ret = fn(now, priv);
7383 udev = netdev_next_upper_dev_rcu(now, &iter);
7388 niter = &udev->adj_list.upper;
7389 dev_stack[cur] = now;
7390 iter_stack[cur++] = iter;
7397 next = dev_stack[--cur];
7398 niter = iter_stack[cur];
7407 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7409 static bool __netdev_has_upper_dev(struct net_device *dev,
7410 struct net_device *upper_dev)
7412 struct netdev_nested_priv priv = {
7414 .data = (void *)upper_dev,
7419 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7424 * netdev_lower_get_next_private - Get the next ->private from the
7425 * lower neighbour list
7427 * @iter: list_head ** of the current position
7429 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7430 * list, starting from iter position. The caller must hold either hold the
7431 * RTNL lock or its own locking that guarantees that the neighbour lower
7432 * list will remain unchanged.
7434 void *netdev_lower_get_next_private(struct net_device *dev,
7435 struct list_head **iter)
7437 struct netdev_adjacent *lower;
7439 lower = list_entry(*iter, struct netdev_adjacent, list);
7441 if (&lower->list == &dev->adj_list.lower)
7444 *iter = lower->list.next;
7446 return lower->private;
7448 EXPORT_SYMBOL(netdev_lower_get_next_private);
7451 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7452 * lower neighbour list, RCU
7455 * @iter: list_head ** of the current position
7457 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7458 * list, starting from iter position. The caller must hold RCU read lock.
7460 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7461 struct list_head **iter)
7463 struct netdev_adjacent *lower;
7465 WARN_ON_ONCE(!rcu_read_lock_held());
7467 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7469 if (&lower->list == &dev->adj_list.lower)
7472 *iter = &lower->list;
7474 return lower->private;
7476 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7479 * netdev_lower_get_next - Get the next device from the lower neighbour
7482 * @iter: list_head ** of the current position
7484 * Gets the next netdev_adjacent from the dev's lower neighbour
7485 * list, starting from iter position. The caller must hold RTNL lock or
7486 * its own locking that guarantees that the neighbour lower
7487 * list will remain unchanged.
7489 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7491 struct netdev_adjacent *lower;
7493 lower = list_entry(*iter, struct netdev_adjacent, list);
7495 if (&lower->list == &dev->adj_list.lower)
7498 *iter = lower->list.next;
7502 EXPORT_SYMBOL(netdev_lower_get_next);
7504 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7505 struct list_head **iter)
7507 struct netdev_adjacent *lower;
7509 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7511 if (&lower->list == &dev->adj_list.lower)
7514 *iter = &lower->list;
7519 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7520 struct list_head **iter,
7523 struct netdev_adjacent *lower;
7525 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7527 if (&lower->list == &dev->adj_list.lower)
7530 *iter = &lower->list;
7531 *ignore = lower->ignore;
7536 int netdev_walk_all_lower_dev(struct net_device *dev,
7537 int (*fn)(struct net_device *dev,
7538 struct netdev_nested_priv *priv),
7539 struct netdev_nested_priv *priv)
7541 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7542 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7546 iter = &dev->adj_list.lower;
7550 ret = fn(now, priv);
7557 ldev = netdev_next_lower_dev(now, &iter);
7562 niter = &ldev->adj_list.lower;
7563 dev_stack[cur] = now;
7564 iter_stack[cur++] = iter;
7571 next = dev_stack[--cur];
7572 niter = iter_stack[cur];
7581 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7583 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7584 int (*fn)(struct net_device *dev,
7585 struct netdev_nested_priv *priv),
7586 struct netdev_nested_priv *priv)
7588 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7589 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7594 iter = &dev->adj_list.lower;
7598 ret = fn(now, priv);
7605 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7612 niter = &ldev->adj_list.lower;
7613 dev_stack[cur] = now;
7614 iter_stack[cur++] = iter;
7621 next = dev_stack[--cur];
7622 niter = iter_stack[cur];
7632 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7633 struct list_head **iter)
7635 struct netdev_adjacent *lower;
7637 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7638 if (&lower->list == &dev->adj_list.lower)
7641 *iter = &lower->list;
7645 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7647 static u8 __netdev_upper_depth(struct net_device *dev)
7649 struct net_device *udev;
7650 struct list_head *iter;
7654 for (iter = &dev->adj_list.upper,
7655 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7657 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7660 if (max_depth < udev->upper_level)
7661 max_depth = udev->upper_level;
7667 static u8 __netdev_lower_depth(struct net_device *dev)
7669 struct net_device *ldev;
7670 struct list_head *iter;
7674 for (iter = &dev->adj_list.lower,
7675 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7677 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7680 if (max_depth < ldev->lower_level)
7681 max_depth = ldev->lower_level;
7687 static int __netdev_update_upper_level(struct net_device *dev,
7688 struct netdev_nested_priv *__unused)
7690 dev->upper_level = __netdev_upper_depth(dev) + 1;
7694 static int __netdev_update_lower_level(struct net_device *dev,
7695 struct netdev_nested_priv *priv)
7697 dev->lower_level = __netdev_lower_depth(dev) + 1;
7699 #ifdef CONFIG_LOCKDEP
7703 if (priv->flags & NESTED_SYNC_IMM)
7704 dev->nested_level = dev->lower_level - 1;
7705 if (priv->flags & NESTED_SYNC_TODO)
7706 net_unlink_todo(dev);
7711 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7712 int (*fn)(struct net_device *dev,
7713 struct netdev_nested_priv *priv),
7714 struct netdev_nested_priv *priv)
7716 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7717 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7721 iter = &dev->adj_list.lower;
7725 ret = fn(now, priv);
7732 ldev = netdev_next_lower_dev_rcu(now, &iter);
7737 niter = &ldev->adj_list.lower;
7738 dev_stack[cur] = now;
7739 iter_stack[cur++] = iter;
7746 next = dev_stack[--cur];
7747 niter = iter_stack[cur];
7756 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7759 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7760 * lower neighbour list, RCU
7764 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7765 * list. The caller must hold RCU read lock.
7767 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7769 struct netdev_adjacent *lower;
7771 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7772 struct netdev_adjacent, list);
7774 return lower->private;
7777 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7780 * netdev_master_upper_dev_get_rcu - Get master upper device
7783 * Find a master upper device and return pointer to it or NULL in case
7784 * it's not there. The caller must hold the RCU read lock.
7786 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7788 struct netdev_adjacent *upper;
7790 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7791 struct netdev_adjacent, list);
7792 if (upper && likely(upper->master))
7796 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7798 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7799 struct net_device *adj_dev,
7800 struct list_head *dev_list)
7802 char linkname[IFNAMSIZ+7];
7804 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7805 "upper_%s" : "lower_%s", adj_dev->name);
7806 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7809 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7811 struct list_head *dev_list)
7813 char linkname[IFNAMSIZ+7];
7815 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7816 "upper_%s" : "lower_%s", name);
7817 sysfs_remove_link(&(dev->dev.kobj), linkname);
7820 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7821 struct net_device *adj_dev,
7822 struct list_head *dev_list)
7824 return (dev_list == &dev->adj_list.upper ||
7825 dev_list == &dev->adj_list.lower) &&
7826 net_eq(dev_net(dev), dev_net(adj_dev));
7829 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7830 struct net_device *adj_dev,
7831 struct list_head *dev_list,
7832 void *private, bool master)
7834 struct netdev_adjacent *adj;
7837 adj = __netdev_find_adj(adj_dev, dev_list);
7841 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7842 dev->name, adj_dev->name, adj->ref_nr);
7847 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7852 adj->master = master;
7854 adj->private = private;
7855 adj->ignore = false;
7858 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7859 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7861 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7862 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7867 /* Ensure that master link is always the first item in list. */
7869 ret = sysfs_create_link(&(dev->dev.kobj),
7870 &(adj_dev->dev.kobj), "master");
7872 goto remove_symlinks;
7874 list_add_rcu(&adj->list, dev_list);
7876 list_add_tail_rcu(&adj->list, dev_list);
7882 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7883 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7891 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7892 struct net_device *adj_dev,
7894 struct list_head *dev_list)
7896 struct netdev_adjacent *adj;
7898 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7899 dev->name, adj_dev->name, ref_nr);
7901 adj = __netdev_find_adj(adj_dev, dev_list);
7904 pr_err("Adjacency does not exist for device %s from %s\n",
7905 dev->name, adj_dev->name);
7910 if (adj->ref_nr > ref_nr) {
7911 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7912 dev->name, adj_dev->name, ref_nr,
7913 adj->ref_nr - ref_nr);
7914 adj->ref_nr -= ref_nr;
7919 sysfs_remove_link(&(dev->dev.kobj), "master");
7921 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7922 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7924 list_del_rcu(&adj->list);
7925 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7926 adj_dev->name, dev->name, adj_dev->name);
7928 kfree_rcu(adj, rcu);
7931 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7932 struct net_device *upper_dev,
7933 struct list_head *up_list,
7934 struct list_head *down_list,
7935 void *private, bool master)
7939 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7944 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7947 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7954 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7955 struct net_device *upper_dev,
7957 struct list_head *up_list,
7958 struct list_head *down_list)
7960 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7961 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7964 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7965 struct net_device *upper_dev,
7966 void *private, bool master)
7968 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7969 &dev->adj_list.upper,
7970 &upper_dev->adj_list.lower,
7974 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7975 struct net_device *upper_dev)
7977 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7978 &dev->adj_list.upper,
7979 &upper_dev->adj_list.lower);
7982 static int __netdev_upper_dev_link(struct net_device *dev,
7983 struct net_device *upper_dev, bool master,
7984 void *upper_priv, void *upper_info,
7985 struct netdev_nested_priv *priv,
7986 struct netlink_ext_ack *extack)
7988 struct netdev_notifier_changeupper_info changeupper_info = {
7993 .upper_dev = upper_dev,
7996 .upper_info = upper_info,
7998 struct net_device *master_dev;
8003 if (dev == upper_dev)
8006 /* To prevent loops, check if dev is not upper device to upper_dev. */
8007 if (__netdev_has_upper_dev(upper_dev, dev))
8010 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8014 if (__netdev_has_upper_dev(dev, upper_dev))
8017 master_dev = __netdev_master_upper_dev_get(dev);
8019 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8022 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8023 &changeupper_info.info);
8024 ret = notifier_to_errno(ret);
8028 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8033 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8034 &changeupper_info.info);
8035 ret = notifier_to_errno(ret);
8039 __netdev_update_upper_level(dev, NULL);
8040 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8042 __netdev_update_lower_level(upper_dev, priv);
8043 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8049 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8055 * netdev_upper_dev_link - Add a link to the upper device
8057 * @upper_dev: new upper device
8058 * @extack: netlink extended ack
8060 * Adds a link to device which is upper to this one. The caller must hold
8061 * the RTNL lock. On a failure a negative errno code is returned.
8062 * On success the reference counts are adjusted and the function
8065 int netdev_upper_dev_link(struct net_device *dev,
8066 struct net_device *upper_dev,
8067 struct netlink_ext_ack *extack)
8069 struct netdev_nested_priv priv = {
8070 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8074 return __netdev_upper_dev_link(dev, upper_dev, false,
8075 NULL, NULL, &priv, extack);
8077 EXPORT_SYMBOL(netdev_upper_dev_link);
8080 * netdev_master_upper_dev_link - Add a master link to the upper device
8082 * @upper_dev: new upper device
8083 * @upper_priv: upper device private
8084 * @upper_info: upper info to be passed down via notifier
8085 * @extack: netlink extended ack
8087 * Adds a link to device which is upper to this one. In this case, only
8088 * one master upper device can be linked, although other non-master devices
8089 * might be linked as well. The caller must hold the RTNL lock.
8090 * On a failure a negative errno code is returned. On success the reference
8091 * counts are adjusted and the function returns zero.
8093 int netdev_master_upper_dev_link(struct net_device *dev,
8094 struct net_device *upper_dev,
8095 void *upper_priv, void *upper_info,
8096 struct netlink_ext_ack *extack)
8098 struct netdev_nested_priv priv = {
8099 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8103 return __netdev_upper_dev_link(dev, upper_dev, true,
8104 upper_priv, upper_info, &priv, extack);
8106 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8108 static void __netdev_upper_dev_unlink(struct net_device *dev,
8109 struct net_device *upper_dev,
8110 struct netdev_nested_priv *priv)
8112 struct netdev_notifier_changeupper_info changeupper_info = {
8116 .upper_dev = upper_dev,
8122 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8124 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8125 &changeupper_info.info);
8127 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8129 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8130 &changeupper_info.info);
8132 __netdev_update_upper_level(dev, NULL);
8133 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8135 __netdev_update_lower_level(upper_dev, priv);
8136 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8141 * netdev_upper_dev_unlink - Removes a link to upper device
8143 * @upper_dev: new upper device
8145 * Removes a link to device which is upper to this one. The caller must hold
8148 void netdev_upper_dev_unlink(struct net_device *dev,
8149 struct net_device *upper_dev)
8151 struct netdev_nested_priv priv = {
8152 .flags = NESTED_SYNC_TODO,
8156 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8158 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8160 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8161 struct net_device *lower_dev,
8164 struct netdev_adjacent *adj;
8166 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8170 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8175 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8176 struct net_device *lower_dev)
8178 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8181 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8182 struct net_device *lower_dev)
8184 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8187 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8188 struct net_device *new_dev,
8189 struct net_device *dev,
8190 struct netlink_ext_ack *extack)
8192 struct netdev_nested_priv priv = {
8201 if (old_dev && new_dev != old_dev)
8202 netdev_adjacent_dev_disable(dev, old_dev);
8203 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8206 if (old_dev && new_dev != old_dev)
8207 netdev_adjacent_dev_enable(dev, old_dev);
8213 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8215 void netdev_adjacent_change_commit(struct net_device *old_dev,
8216 struct net_device *new_dev,
8217 struct net_device *dev)
8219 struct netdev_nested_priv priv = {
8220 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8224 if (!new_dev || !old_dev)
8227 if (new_dev == old_dev)
8230 netdev_adjacent_dev_enable(dev, old_dev);
8231 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8233 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8235 void netdev_adjacent_change_abort(struct net_device *old_dev,
8236 struct net_device *new_dev,
8237 struct net_device *dev)
8239 struct netdev_nested_priv priv = {
8247 if (old_dev && new_dev != old_dev)
8248 netdev_adjacent_dev_enable(dev, old_dev);
8250 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8252 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8255 * netdev_bonding_info_change - Dispatch event about slave change
8257 * @bonding_info: info to dispatch
8259 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8260 * The caller must hold the RTNL lock.
8262 void netdev_bonding_info_change(struct net_device *dev,
8263 struct netdev_bonding_info *bonding_info)
8265 struct netdev_notifier_bonding_info info = {
8269 memcpy(&info.bonding_info, bonding_info,
8270 sizeof(struct netdev_bonding_info));
8271 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8274 EXPORT_SYMBOL(netdev_bonding_info_change);
8277 * netdev_get_xmit_slave - Get the xmit slave of master device
8280 * @all_slaves: assume all the slaves are active
8282 * The reference counters are not incremented so the caller must be
8283 * careful with locks. The caller must hold RCU lock.
8284 * %NULL is returned if no slave is found.
8287 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8288 struct sk_buff *skb,
8291 const struct net_device_ops *ops = dev->netdev_ops;
8293 if (!ops->ndo_get_xmit_slave)
8295 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8297 EXPORT_SYMBOL(netdev_get_xmit_slave);
8299 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8302 const struct net_device_ops *ops = dev->netdev_ops;
8304 if (!ops->ndo_sk_get_lower_dev)
8306 return ops->ndo_sk_get_lower_dev(dev, sk);
8310 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8314 * %NULL is returned if no lower device is found.
8317 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8320 struct net_device *lower;
8322 lower = netdev_sk_get_lower_dev(dev, sk);
8325 lower = netdev_sk_get_lower_dev(dev, sk);
8330 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8332 static void netdev_adjacent_add_links(struct net_device *dev)
8334 struct netdev_adjacent *iter;
8336 struct net *net = dev_net(dev);
8338 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8339 if (!net_eq(net, dev_net(iter->dev)))
8341 netdev_adjacent_sysfs_add(iter->dev, dev,
8342 &iter->dev->adj_list.lower);
8343 netdev_adjacent_sysfs_add(dev, iter->dev,
8344 &dev->adj_list.upper);
8347 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8348 if (!net_eq(net, dev_net(iter->dev)))
8350 netdev_adjacent_sysfs_add(iter->dev, dev,
8351 &iter->dev->adj_list.upper);
8352 netdev_adjacent_sysfs_add(dev, iter->dev,
8353 &dev->adj_list.lower);
8357 static void netdev_adjacent_del_links(struct net_device *dev)
8359 struct netdev_adjacent *iter;
8361 struct net *net = dev_net(dev);
8363 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8364 if (!net_eq(net, dev_net(iter->dev)))
8366 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8367 &iter->dev->adj_list.lower);
8368 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8369 &dev->adj_list.upper);
8372 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8373 if (!net_eq(net, dev_net(iter->dev)))
8375 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8376 &iter->dev->adj_list.upper);
8377 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8378 &dev->adj_list.lower);
8382 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8384 struct netdev_adjacent *iter;
8386 struct net *net = dev_net(dev);
8388 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8389 if (!net_eq(net, dev_net(iter->dev)))
8391 netdev_adjacent_sysfs_del(iter->dev, oldname,
8392 &iter->dev->adj_list.lower);
8393 netdev_adjacent_sysfs_add(iter->dev, dev,
8394 &iter->dev->adj_list.lower);
8397 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8398 if (!net_eq(net, dev_net(iter->dev)))
8400 netdev_adjacent_sysfs_del(iter->dev, oldname,
8401 &iter->dev->adj_list.upper);
8402 netdev_adjacent_sysfs_add(iter->dev, dev,
8403 &iter->dev->adj_list.upper);
8407 void *netdev_lower_dev_get_private(struct net_device *dev,
8408 struct net_device *lower_dev)
8410 struct netdev_adjacent *lower;
8414 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8418 return lower->private;
8420 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8424 * netdev_lower_state_changed - Dispatch event about lower device state change
8425 * @lower_dev: device
8426 * @lower_state_info: state to dispatch
8428 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8429 * The caller must hold the RTNL lock.
8431 void netdev_lower_state_changed(struct net_device *lower_dev,
8432 void *lower_state_info)
8434 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8435 .info.dev = lower_dev,
8439 changelowerstate_info.lower_state_info = lower_state_info;
8440 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8441 &changelowerstate_info.info);
8443 EXPORT_SYMBOL(netdev_lower_state_changed);
8445 static void dev_change_rx_flags(struct net_device *dev, int flags)
8447 const struct net_device_ops *ops = dev->netdev_ops;
8449 if (ops->ndo_change_rx_flags)
8450 ops->ndo_change_rx_flags(dev, flags);
8453 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8455 unsigned int old_flags = dev->flags;
8461 dev->flags |= IFF_PROMISC;
8462 dev->promiscuity += inc;
8463 if (dev->promiscuity == 0) {
8466 * If inc causes overflow, untouch promisc and return error.
8469 dev->flags &= ~IFF_PROMISC;
8471 dev->promiscuity -= inc;
8472 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8477 if (dev->flags != old_flags) {
8478 pr_info("device %s %s promiscuous mode\n",
8480 dev->flags & IFF_PROMISC ? "entered" : "left");
8481 if (audit_enabled) {
8482 current_uid_gid(&uid, &gid);
8483 audit_log(audit_context(), GFP_ATOMIC,
8484 AUDIT_ANOM_PROMISCUOUS,
8485 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8486 dev->name, (dev->flags & IFF_PROMISC),
8487 (old_flags & IFF_PROMISC),
8488 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8489 from_kuid(&init_user_ns, uid),
8490 from_kgid(&init_user_ns, gid),
8491 audit_get_sessionid(current));
8494 dev_change_rx_flags(dev, IFF_PROMISC);
8497 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8502 * dev_set_promiscuity - update promiscuity count on a device
8506 * Add or remove promiscuity from a device. While the count in the device
8507 * remains above zero the interface remains promiscuous. Once it hits zero
8508 * the device reverts back to normal filtering operation. A negative inc
8509 * value is used to drop promiscuity on the device.
8510 * Return 0 if successful or a negative errno code on error.
8512 int dev_set_promiscuity(struct net_device *dev, int inc)
8514 unsigned int old_flags = dev->flags;
8517 err = __dev_set_promiscuity(dev, inc, true);
8520 if (dev->flags != old_flags)
8521 dev_set_rx_mode(dev);
8524 EXPORT_SYMBOL(dev_set_promiscuity);
8526 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8528 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8532 dev->flags |= IFF_ALLMULTI;
8533 dev->allmulti += inc;
8534 if (dev->allmulti == 0) {
8537 * If inc causes overflow, untouch allmulti and return error.
8540 dev->flags &= ~IFF_ALLMULTI;
8542 dev->allmulti -= inc;
8543 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8548 if (dev->flags ^ old_flags) {
8549 dev_change_rx_flags(dev, IFF_ALLMULTI);
8550 dev_set_rx_mode(dev);
8552 __dev_notify_flags(dev, old_flags,
8553 dev->gflags ^ old_gflags);
8559 * dev_set_allmulti - update allmulti count on a device
8563 * Add or remove reception of all multicast frames to a device. While the
8564 * count in the device remains above zero the interface remains listening
8565 * to all interfaces. Once it hits zero the device reverts back to normal
8566 * filtering operation. A negative @inc value is used to drop the counter
8567 * when releasing a resource needing all multicasts.
8568 * Return 0 if successful or a negative errno code on error.
8571 int dev_set_allmulti(struct net_device *dev, int inc)
8573 return __dev_set_allmulti(dev, inc, true);
8575 EXPORT_SYMBOL(dev_set_allmulti);
8578 * Upload unicast and multicast address lists to device and
8579 * configure RX filtering. When the device doesn't support unicast
8580 * filtering it is put in promiscuous mode while unicast addresses
8583 void __dev_set_rx_mode(struct net_device *dev)
8585 const struct net_device_ops *ops = dev->netdev_ops;
8587 /* dev_open will call this function so the list will stay sane. */
8588 if (!(dev->flags&IFF_UP))
8591 if (!netif_device_present(dev))
8594 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8595 /* Unicast addresses changes may only happen under the rtnl,
8596 * therefore calling __dev_set_promiscuity here is safe.
8598 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8599 __dev_set_promiscuity(dev, 1, false);
8600 dev->uc_promisc = true;
8601 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8602 __dev_set_promiscuity(dev, -1, false);
8603 dev->uc_promisc = false;
8607 if (ops->ndo_set_rx_mode)
8608 ops->ndo_set_rx_mode(dev);
8611 void dev_set_rx_mode(struct net_device *dev)
8613 netif_addr_lock_bh(dev);
8614 __dev_set_rx_mode(dev);
8615 netif_addr_unlock_bh(dev);
8619 * dev_get_flags - get flags reported to userspace
8622 * Get the combination of flag bits exported through APIs to userspace.
8624 unsigned int dev_get_flags(const struct net_device *dev)
8628 flags = (dev->flags & ~(IFF_PROMISC |
8633 (dev->gflags & (IFF_PROMISC |
8636 if (netif_running(dev)) {
8637 if (netif_oper_up(dev))
8638 flags |= IFF_RUNNING;
8639 if (netif_carrier_ok(dev))
8640 flags |= IFF_LOWER_UP;
8641 if (netif_dormant(dev))
8642 flags |= IFF_DORMANT;
8647 EXPORT_SYMBOL(dev_get_flags);
8649 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8650 struct netlink_ext_ack *extack)
8652 unsigned int old_flags = dev->flags;
8658 * Set the flags on our device.
8661 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8662 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8664 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8668 * Load in the correct multicast list now the flags have changed.
8671 if ((old_flags ^ flags) & IFF_MULTICAST)
8672 dev_change_rx_flags(dev, IFF_MULTICAST);
8674 dev_set_rx_mode(dev);
8677 * Have we downed the interface. We handle IFF_UP ourselves
8678 * according to user attempts to set it, rather than blindly
8683 if ((old_flags ^ flags) & IFF_UP) {
8684 if (old_flags & IFF_UP)
8687 ret = __dev_open(dev, extack);
8690 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8691 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8692 unsigned int old_flags = dev->flags;
8694 dev->gflags ^= IFF_PROMISC;
8696 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8697 if (dev->flags != old_flags)
8698 dev_set_rx_mode(dev);
8701 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8702 * is important. Some (broken) drivers set IFF_PROMISC, when
8703 * IFF_ALLMULTI is requested not asking us and not reporting.
8705 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8706 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8708 dev->gflags ^= IFF_ALLMULTI;
8709 __dev_set_allmulti(dev, inc, false);
8715 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8716 unsigned int gchanges)
8718 unsigned int changes = dev->flags ^ old_flags;
8721 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8723 if (changes & IFF_UP) {
8724 if (dev->flags & IFF_UP)
8725 call_netdevice_notifiers(NETDEV_UP, dev);
8727 call_netdevice_notifiers(NETDEV_DOWN, dev);
8730 if (dev->flags & IFF_UP &&
8731 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8732 struct netdev_notifier_change_info change_info = {
8736 .flags_changed = changes,
8739 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8744 * dev_change_flags - change device settings
8746 * @flags: device state flags
8747 * @extack: netlink extended ack
8749 * Change settings on device based state flags. The flags are
8750 * in the userspace exported format.
8752 int dev_change_flags(struct net_device *dev, unsigned int flags,
8753 struct netlink_ext_ack *extack)
8756 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8758 ret = __dev_change_flags(dev, flags, extack);
8762 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8763 __dev_notify_flags(dev, old_flags, changes);
8766 EXPORT_SYMBOL(dev_change_flags);
8768 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8770 const struct net_device_ops *ops = dev->netdev_ops;
8772 if (ops->ndo_change_mtu)
8773 return ops->ndo_change_mtu(dev, new_mtu);
8775 /* Pairs with all the lockless reads of dev->mtu in the stack */
8776 WRITE_ONCE(dev->mtu, new_mtu);
8779 EXPORT_SYMBOL(__dev_set_mtu);
8781 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8782 struct netlink_ext_ack *extack)
8784 /* MTU must be positive, and in range */
8785 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8786 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8790 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8791 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8798 * dev_set_mtu_ext - Change maximum transfer unit
8800 * @new_mtu: new transfer unit
8801 * @extack: netlink extended ack
8803 * Change the maximum transfer size of the network device.
8805 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8806 struct netlink_ext_ack *extack)
8810 if (new_mtu == dev->mtu)
8813 err = dev_validate_mtu(dev, new_mtu, extack);
8817 if (!netif_device_present(dev))
8820 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8821 err = notifier_to_errno(err);
8825 orig_mtu = dev->mtu;
8826 err = __dev_set_mtu(dev, new_mtu);
8829 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8831 err = notifier_to_errno(err);
8833 /* setting mtu back and notifying everyone again,
8834 * so that they have a chance to revert changes.
8836 __dev_set_mtu(dev, orig_mtu);
8837 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8844 int dev_set_mtu(struct net_device *dev, int new_mtu)
8846 struct netlink_ext_ack extack;
8849 memset(&extack, 0, sizeof(extack));
8850 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8851 if (err && extack._msg)
8852 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8855 EXPORT_SYMBOL(dev_set_mtu);
8858 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8860 * @new_len: new tx queue length
8862 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8864 unsigned int orig_len = dev->tx_queue_len;
8867 if (new_len != (unsigned int)new_len)
8870 if (new_len != orig_len) {
8871 dev->tx_queue_len = new_len;
8872 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8873 res = notifier_to_errno(res);
8876 res = dev_qdisc_change_tx_queue_len(dev);
8884 netdev_err(dev, "refused to change device tx_queue_len\n");
8885 dev->tx_queue_len = orig_len;
8890 * dev_set_group - Change group this device belongs to
8892 * @new_group: group this device should belong to
8894 void dev_set_group(struct net_device *dev, int new_group)
8896 dev->group = new_group;
8898 EXPORT_SYMBOL(dev_set_group);
8901 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8903 * @addr: new address
8904 * @extack: netlink extended ack
8906 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8907 struct netlink_ext_ack *extack)
8909 struct netdev_notifier_pre_changeaddr_info info = {
8911 .info.extack = extack,
8916 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8917 return notifier_to_errno(rc);
8919 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8922 * dev_set_mac_address - Change Media Access Control Address
8925 * @extack: netlink extended ack
8927 * Change the hardware (MAC) address of the device
8929 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8930 struct netlink_ext_ack *extack)
8932 const struct net_device_ops *ops = dev->netdev_ops;
8935 if (!ops->ndo_set_mac_address)
8937 if (sa->sa_family != dev->type)
8939 if (!netif_device_present(dev))
8941 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8944 err = ops->ndo_set_mac_address(dev, sa);
8947 dev->addr_assign_type = NET_ADDR_SET;
8948 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8949 add_device_randomness(dev->dev_addr, dev->addr_len);
8952 EXPORT_SYMBOL(dev_set_mac_address);
8954 static DECLARE_RWSEM(dev_addr_sem);
8956 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8957 struct netlink_ext_ack *extack)
8961 down_write(&dev_addr_sem);
8962 ret = dev_set_mac_address(dev, sa, extack);
8963 up_write(&dev_addr_sem);
8966 EXPORT_SYMBOL(dev_set_mac_address_user);
8968 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8970 size_t size = sizeof(sa->sa_data);
8971 struct net_device *dev;
8974 down_read(&dev_addr_sem);
8977 dev = dev_get_by_name_rcu(net, dev_name);
8983 memset(sa->sa_data, 0, size);
8985 memcpy(sa->sa_data, dev->dev_addr,
8986 min_t(size_t, size, dev->addr_len));
8987 sa->sa_family = dev->type;
8991 up_read(&dev_addr_sem);
8994 EXPORT_SYMBOL(dev_get_mac_address);
8997 * dev_change_carrier - Change device carrier
8999 * @new_carrier: new value
9001 * Change device carrier
9003 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9005 const struct net_device_ops *ops = dev->netdev_ops;
9007 if (!ops->ndo_change_carrier)
9009 if (!netif_device_present(dev))
9011 return ops->ndo_change_carrier(dev, new_carrier);
9013 EXPORT_SYMBOL(dev_change_carrier);
9016 * dev_get_phys_port_id - Get device physical port ID
9020 * Get device physical port ID
9022 int dev_get_phys_port_id(struct net_device *dev,
9023 struct netdev_phys_item_id *ppid)
9025 const struct net_device_ops *ops = dev->netdev_ops;
9027 if (!ops->ndo_get_phys_port_id)
9029 return ops->ndo_get_phys_port_id(dev, ppid);
9031 EXPORT_SYMBOL(dev_get_phys_port_id);
9034 * dev_get_phys_port_name - Get device physical port name
9037 * @len: limit of bytes to copy to name
9039 * Get device physical port name
9041 int dev_get_phys_port_name(struct net_device *dev,
9042 char *name, size_t len)
9044 const struct net_device_ops *ops = dev->netdev_ops;
9047 if (ops->ndo_get_phys_port_name) {
9048 err = ops->ndo_get_phys_port_name(dev, name, len);
9049 if (err != -EOPNOTSUPP)
9052 return devlink_compat_phys_port_name_get(dev, name, len);
9054 EXPORT_SYMBOL(dev_get_phys_port_name);
9057 * dev_get_port_parent_id - Get the device's port parent identifier
9058 * @dev: network device
9059 * @ppid: pointer to a storage for the port's parent identifier
9060 * @recurse: allow/disallow recursion to lower devices
9062 * Get the devices's port parent identifier
9064 int dev_get_port_parent_id(struct net_device *dev,
9065 struct netdev_phys_item_id *ppid,
9068 const struct net_device_ops *ops = dev->netdev_ops;
9069 struct netdev_phys_item_id first = { };
9070 struct net_device *lower_dev;
9071 struct list_head *iter;
9074 if (ops->ndo_get_port_parent_id) {
9075 err = ops->ndo_get_port_parent_id(dev, ppid);
9076 if (err != -EOPNOTSUPP)
9080 err = devlink_compat_switch_id_get(dev, ppid);
9081 if (!err || err != -EOPNOTSUPP)
9087 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9088 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9093 else if (memcmp(&first, ppid, sizeof(*ppid)))
9099 EXPORT_SYMBOL(dev_get_port_parent_id);
9102 * netdev_port_same_parent_id - Indicate if two network devices have
9103 * the same port parent identifier
9104 * @a: first network device
9105 * @b: second network device
9107 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9109 struct netdev_phys_item_id a_id = { };
9110 struct netdev_phys_item_id b_id = { };
9112 if (dev_get_port_parent_id(a, &a_id, true) ||
9113 dev_get_port_parent_id(b, &b_id, true))
9116 return netdev_phys_item_id_same(&a_id, &b_id);
9118 EXPORT_SYMBOL(netdev_port_same_parent_id);
9121 * dev_change_proto_down - update protocol port state information
9123 * @proto_down: new value
9125 * This info can be used by switch drivers to set the phys state of the
9128 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9130 const struct net_device_ops *ops = dev->netdev_ops;
9132 if (!ops->ndo_change_proto_down)
9134 if (!netif_device_present(dev))
9136 return ops->ndo_change_proto_down(dev, proto_down);
9138 EXPORT_SYMBOL(dev_change_proto_down);
9141 * dev_change_proto_down_generic - generic implementation for
9142 * ndo_change_proto_down that sets carrier according to
9146 * @proto_down: new value
9148 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9151 netif_carrier_off(dev);
9153 netif_carrier_on(dev);
9154 dev->proto_down = proto_down;
9157 EXPORT_SYMBOL(dev_change_proto_down_generic);
9160 * dev_change_proto_down_reason - proto down reason
9163 * @mask: proto down mask
9164 * @value: proto down value
9166 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9172 dev->proto_down_reason = value;
9174 for_each_set_bit(b, &mask, 32) {
9175 if (value & (1 << b))
9176 dev->proto_down_reason |= BIT(b);
9178 dev->proto_down_reason &= ~BIT(b);
9182 EXPORT_SYMBOL(dev_change_proto_down_reason);
9184 struct bpf_xdp_link {
9185 struct bpf_link link;
9186 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9190 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9192 if (flags & XDP_FLAGS_HW_MODE)
9194 if (flags & XDP_FLAGS_DRV_MODE)
9195 return XDP_MODE_DRV;
9196 if (flags & XDP_FLAGS_SKB_MODE)
9197 return XDP_MODE_SKB;
9198 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9201 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9205 return generic_xdp_install;
9208 return dev->netdev_ops->ndo_bpf;
9214 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9215 enum bpf_xdp_mode mode)
9217 return dev->xdp_state[mode].link;
9220 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9221 enum bpf_xdp_mode mode)
9223 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9226 return link->link.prog;
9227 return dev->xdp_state[mode].prog;
9230 static u8 dev_xdp_prog_count(struct net_device *dev)
9235 for (i = 0; i < __MAX_XDP_MODE; i++)
9236 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9241 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9243 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9245 return prog ? prog->aux->id : 0;
9248 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9249 struct bpf_xdp_link *link)
9251 dev->xdp_state[mode].link = link;
9252 dev->xdp_state[mode].prog = NULL;
9255 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9256 struct bpf_prog *prog)
9258 dev->xdp_state[mode].link = NULL;
9259 dev->xdp_state[mode].prog = prog;
9262 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9263 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9264 u32 flags, struct bpf_prog *prog)
9266 struct netdev_bpf xdp;
9269 memset(&xdp, 0, sizeof(xdp));
9270 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9271 xdp.extack = extack;
9275 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9276 * "moved" into driver), so they don't increment it on their own, but
9277 * they do decrement refcnt when program is detached or replaced.
9278 * Given net_device also owns link/prog, we need to bump refcnt here
9279 * to prevent drivers from underflowing it.
9283 err = bpf_op(dev, &xdp);
9290 if (mode != XDP_MODE_HW)
9291 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9296 static void dev_xdp_uninstall(struct net_device *dev)
9298 struct bpf_xdp_link *link;
9299 struct bpf_prog *prog;
9300 enum bpf_xdp_mode mode;
9305 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9306 prog = dev_xdp_prog(dev, mode);
9310 bpf_op = dev_xdp_bpf_op(dev, mode);
9314 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9316 /* auto-detach link from net device */
9317 link = dev_xdp_link(dev, mode);
9323 dev_xdp_set_link(dev, mode, NULL);
9327 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9328 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9329 struct bpf_prog *old_prog, u32 flags)
9331 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9332 struct bpf_prog *cur_prog;
9333 enum bpf_xdp_mode mode;
9339 /* either link or prog attachment, never both */
9340 if (link && (new_prog || old_prog))
9342 /* link supports only XDP mode flags */
9343 if (link && (flags & ~XDP_FLAGS_MODES)) {
9344 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9347 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9348 if (num_modes > 1) {
9349 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9352 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9353 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9354 NL_SET_ERR_MSG(extack,
9355 "More than one program loaded, unset mode is ambiguous");
9358 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9359 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9360 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9364 mode = dev_xdp_mode(dev, flags);
9365 /* can't replace attached link */
9366 if (dev_xdp_link(dev, mode)) {
9367 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9371 cur_prog = dev_xdp_prog(dev, mode);
9372 /* can't replace attached prog with link */
9373 if (link && cur_prog) {
9374 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9377 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9378 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9382 /* put effective new program into new_prog */
9384 new_prog = link->link.prog;
9387 bool offload = mode == XDP_MODE_HW;
9388 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9389 ? XDP_MODE_DRV : XDP_MODE_SKB;
9391 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9392 NL_SET_ERR_MSG(extack, "XDP program already attached");
9395 if (!offload && dev_xdp_prog(dev, other_mode)) {
9396 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9399 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9400 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9403 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9404 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9407 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9408 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9413 /* don't call drivers if the effective program didn't change */
9414 if (new_prog != cur_prog) {
9415 bpf_op = dev_xdp_bpf_op(dev, mode);
9417 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9421 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9427 dev_xdp_set_link(dev, mode, link);
9429 dev_xdp_set_prog(dev, mode, new_prog);
9431 bpf_prog_put(cur_prog);
9436 static int dev_xdp_attach_link(struct net_device *dev,
9437 struct netlink_ext_ack *extack,
9438 struct bpf_xdp_link *link)
9440 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9443 static int dev_xdp_detach_link(struct net_device *dev,
9444 struct netlink_ext_ack *extack,
9445 struct bpf_xdp_link *link)
9447 enum bpf_xdp_mode mode;
9452 mode = dev_xdp_mode(dev, link->flags);
9453 if (dev_xdp_link(dev, mode) != link)
9456 bpf_op = dev_xdp_bpf_op(dev, mode);
9457 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9458 dev_xdp_set_link(dev, mode, NULL);
9462 static void bpf_xdp_link_release(struct bpf_link *link)
9464 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9468 /* if racing with net_device's tear down, xdp_link->dev might be
9469 * already NULL, in which case link was already auto-detached
9471 if (xdp_link->dev) {
9472 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9473 xdp_link->dev = NULL;
9479 static int bpf_xdp_link_detach(struct bpf_link *link)
9481 bpf_xdp_link_release(link);
9485 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9487 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9492 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9493 struct seq_file *seq)
9495 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9500 ifindex = xdp_link->dev->ifindex;
9503 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9506 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9507 struct bpf_link_info *info)
9509 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9514 ifindex = xdp_link->dev->ifindex;
9517 info->xdp.ifindex = ifindex;
9521 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9522 struct bpf_prog *old_prog)
9524 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9525 enum bpf_xdp_mode mode;
9531 /* link might have been auto-released already, so fail */
9532 if (!xdp_link->dev) {
9537 if (old_prog && link->prog != old_prog) {
9541 old_prog = link->prog;
9542 if (old_prog == new_prog) {
9543 /* no-op, don't disturb drivers */
9544 bpf_prog_put(new_prog);
9548 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9549 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9550 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9551 xdp_link->flags, new_prog);
9555 old_prog = xchg(&link->prog, new_prog);
9556 bpf_prog_put(old_prog);
9563 static const struct bpf_link_ops bpf_xdp_link_lops = {
9564 .release = bpf_xdp_link_release,
9565 .dealloc = bpf_xdp_link_dealloc,
9566 .detach = bpf_xdp_link_detach,
9567 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9568 .fill_link_info = bpf_xdp_link_fill_link_info,
9569 .update_prog = bpf_xdp_link_update,
9572 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9574 struct net *net = current->nsproxy->net_ns;
9575 struct bpf_link_primer link_primer;
9576 struct bpf_xdp_link *link;
9577 struct net_device *dev;
9580 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9584 link = kzalloc(sizeof(*link), GFP_USER);
9590 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9592 link->flags = attr->link_create.flags;
9594 err = bpf_link_prime(&link->link, &link_primer);
9601 err = dev_xdp_attach_link(dev, NULL, link);
9605 bpf_link_cleanup(&link_primer);
9609 fd = bpf_link_settle(&link_primer);
9610 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9620 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9622 * @extack: netlink extended ack
9623 * @fd: new program fd or negative value to clear
9624 * @expected_fd: old program fd that userspace expects to replace or clear
9625 * @flags: xdp-related flags
9627 * Set or clear a bpf program for a device
9629 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9630 int fd, int expected_fd, u32 flags)
9632 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9633 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9639 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9640 mode != XDP_MODE_SKB);
9641 if (IS_ERR(new_prog))
9642 return PTR_ERR(new_prog);
9645 if (expected_fd >= 0) {
9646 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9647 mode != XDP_MODE_SKB);
9648 if (IS_ERR(old_prog)) {
9649 err = PTR_ERR(old_prog);
9655 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9658 if (err && new_prog)
9659 bpf_prog_put(new_prog);
9661 bpf_prog_put(old_prog);
9666 * dev_new_index - allocate an ifindex
9667 * @net: the applicable net namespace
9669 * Returns a suitable unique value for a new device interface
9670 * number. The caller must hold the rtnl semaphore or the
9671 * dev_base_lock to be sure it remains unique.
9673 static int dev_new_index(struct net *net)
9675 int ifindex = net->ifindex;
9680 if (!__dev_get_by_index(net, ifindex))
9681 return net->ifindex = ifindex;
9685 /* Delayed registration/unregisteration */
9686 static LIST_HEAD(net_todo_list);
9687 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9689 static void net_set_todo(struct net_device *dev)
9691 list_add_tail(&dev->todo_list, &net_todo_list);
9692 dev_net(dev)->dev_unreg_count++;
9695 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9696 struct net_device *upper, netdev_features_t features)
9698 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9699 netdev_features_t feature;
9702 for_each_netdev_feature(upper_disables, feature_bit) {
9703 feature = __NETIF_F_BIT(feature_bit);
9704 if (!(upper->wanted_features & feature)
9705 && (features & feature)) {
9706 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9707 &feature, upper->name);
9708 features &= ~feature;
9715 static void netdev_sync_lower_features(struct net_device *upper,
9716 struct net_device *lower, netdev_features_t features)
9718 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9719 netdev_features_t feature;
9722 for_each_netdev_feature(upper_disables, feature_bit) {
9723 feature = __NETIF_F_BIT(feature_bit);
9724 if (!(features & feature) && (lower->features & feature)) {
9725 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9726 &feature, lower->name);
9727 lower->wanted_features &= ~feature;
9728 __netdev_update_features(lower);
9730 if (unlikely(lower->features & feature))
9731 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9732 &feature, lower->name);
9734 netdev_features_change(lower);
9739 static netdev_features_t netdev_fix_features(struct net_device *dev,
9740 netdev_features_t features)
9742 /* Fix illegal checksum combinations */
9743 if ((features & NETIF_F_HW_CSUM) &&
9744 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9745 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9746 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9749 /* TSO requires that SG is present as well. */
9750 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9751 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9752 features &= ~NETIF_F_ALL_TSO;
9755 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9756 !(features & NETIF_F_IP_CSUM)) {
9757 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9758 features &= ~NETIF_F_TSO;
9759 features &= ~NETIF_F_TSO_ECN;
9762 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9763 !(features & NETIF_F_IPV6_CSUM)) {
9764 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9765 features &= ~NETIF_F_TSO6;
9768 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9769 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9770 features &= ~NETIF_F_TSO_MANGLEID;
9772 /* TSO ECN requires that TSO is present as well. */
9773 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9774 features &= ~NETIF_F_TSO_ECN;
9776 /* Software GSO depends on SG. */
9777 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9778 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9779 features &= ~NETIF_F_GSO;
9782 /* GSO partial features require GSO partial be set */
9783 if ((features & dev->gso_partial_features) &&
9784 !(features & NETIF_F_GSO_PARTIAL)) {
9786 "Dropping partially supported GSO features since no GSO partial.\n");
9787 features &= ~dev->gso_partial_features;
9790 if (!(features & NETIF_F_RXCSUM)) {
9791 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9792 * successfully merged by hardware must also have the
9793 * checksum verified by hardware. If the user does not
9794 * want to enable RXCSUM, logically, we should disable GRO_HW.
9796 if (features & NETIF_F_GRO_HW) {
9797 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9798 features &= ~NETIF_F_GRO_HW;
9802 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9803 if (features & NETIF_F_RXFCS) {
9804 if (features & NETIF_F_LRO) {
9805 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9806 features &= ~NETIF_F_LRO;
9809 if (features & NETIF_F_GRO_HW) {
9810 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9811 features &= ~NETIF_F_GRO_HW;
9815 if (features & NETIF_F_HW_TLS_TX) {
9816 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9817 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9818 bool hw_csum = features & NETIF_F_HW_CSUM;
9820 if (!ip_csum && !hw_csum) {
9821 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9822 features &= ~NETIF_F_HW_TLS_TX;
9826 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9827 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9828 features &= ~NETIF_F_HW_TLS_RX;
9834 int __netdev_update_features(struct net_device *dev)
9836 struct net_device *upper, *lower;
9837 netdev_features_t features;
9838 struct list_head *iter;
9843 features = netdev_get_wanted_features(dev);
9845 if (dev->netdev_ops->ndo_fix_features)
9846 features = dev->netdev_ops->ndo_fix_features(dev, features);
9848 /* driver might be less strict about feature dependencies */
9849 features = netdev_fix_features(dev, features);
9851 /* some features can't be enabled if they're off on an upper device */
9852 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9853 features = netdev_sync_upper_features(dev, upper, features);
9855 if (dev->features == features)
9858 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9859 &dev->features, &features);
9861 if (dev->netdev_ops->ndo_set_features)
9862 err = dev->netdev_ops->ndo_set_features(dev, features);
9866 if (unlikely(err < 0)) {
9868 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9869 err, &features, &dev->features);
9870 /* return non-0 since some features might have changed and
9871 * it's better to fire a spurious notification than miss it
9877 /* some features must be disabled on lower devices when disabled
9878 * on an upper device (think: bonding master or bridge)
9880 netdev_for_each_lower_dev(dev, lower, iter)
9881 netdev_sync_lower_features(dev, lower, features);
9884 netdev_features_t diff = features ^ dev->features;
9886 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9887 /* udp_tunnel_{get,drop}_rx_info both need
9888 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9889 * device, or they won't do anything.
9890 * Thus we need to update dev->features
9891 * *before* calling udp_tunnel_get_rx_info,
9892 * but *after* calling udp_tunnel_drop_rx_info.
9894 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9895 dev->features = features;
9896 udp_tunnel_get_rx_info(dev);
9898 udp_tunnel_drop_rx_info(dev);
9902 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9903 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9904 dev->features = features;
9905 err |= vlan_get_rx_ctag_filter_info(dev);
9907 vlan_drop_rx_ctag_filter_info(dev);
9911 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9912 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9913 dev->features = features;
9914 err |= vlan_get_rx_stag_filter_info(dev);
9916 vlan_drop_rx_stag_filter_info(dev);
9920 dev->features = features;
9923 return err < 0 ? 0 : 1;
9927 * netdev_update_features - recalculate device features
9928 * @dev: the device to check
9930 * Recalculate dev->features set and send notifications if it
9931 * has changed. Should be called after driver or hardware dependent
9932 * conditions might have changed that influence the features.
9934 void netdev_update_features(struct net_device *dev)
9936 if (__netdev_update_features(dev))
9937 netdev_features_change(dev);
9939 EXPORT_SYMBOL(netdev_update_features);
9942 * netdev_change_features - recalculate device features
9943 * @dev: the device to check
9945 * Recalculate dev->features set and send notifications even
9946 * if they have not changed. Should be called instead of
9947 * netdev_update_features() if also dev->vlan_features might
9948 * have changed to allow the changes to be propagated to stacked
9951 void netdev_change_features(struct net_device *dev)
9953 __netdev_update_features(dev);
9954 netdev_features_change(dev);
9956 EXPORT_SYMBOL(netdev_change_features);
9959 * netif_stacked_transfer_operstate - transfer operstate
9960 * @rootdev: the root or lower level device to transfer state from
9961 * @dev: the device to transfer operstate to
9963 * Transfer operational state from root to device. This is normally
9964 * called when a stacking relationship exists between the root
9965 * device and the device(a leaf device).
9967 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9968 struct net_device *dev)
9970 if (rootdev->operstate == IF_OPER_DORMANT)
9971 netif_dormant_on(dev);
9973 netif_dormant_off(dev);
9975 if (rootdev->operstate == IF_OPER_TESTING)
9976 netif_testing_on(dev);
9978 netif_testing_off(dev);
9980 if (netif_carrier_ok(rootdev))
9981 netif_carrier_on(dev);
9983 netif_carrier_off(dev);
9985 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9987 static int netif_alloc_rx_queues(struct net_device *dev)
9989 unsigned int i, count = dev->num_rx_queues;
9990 struct netdev_rx_queue *rx;
9991 size_t sz = count * sizeof(*rx);
9996 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10002 for (i = 0; i < count; i++) {
10005 /* XDP RX-queue setup */
10006 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10013 /* Rollback successful reg's and free other resources */
10015 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10021 static void netif_free_rx_queues(struct net_device *dev)
10023 unsigned int i, count = dev->num_rx_queues;
10025 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10029 for (i = 0; i < count; i++)
10030 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10035 static void netdev_init_one_queue(struct net_device *dev,
10036 struct netdev_queue *queue, void *_unused)
10038 /* Initialize queue lock */
10039 spin_lock_init(&queue->_xmit_lock);
10040 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10041 queue->xmit_lock_owner = -1;
10042 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10045 dql_init(&queue->dql, HZ);
10049 static void netif_free_tx_queues(struct net_device *dev)
10054 static int netif_alloc_netdev_queues(struct net_device *dev)
10056 unsigned int count = dev->num_tx_queues;
10057 struct netdev_queue *tx;
10058 size_t sz = count * sizeof(*tx);
10060 if (count < 1 || count > 0xffff)
10063 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10069 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10070 spin_lock_init(&dev->tx_global_lock);
10075 void netif_tx_stop_all_queues(struct net_device *dev)
10079 for (i = 0; i < dev->num_tx_queues; i++) {
10080 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10082 netif_tx_stop_queue(txq);
10085 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10088 * register_netdevice - register a network device
10089 * @dev: device to register
10091 * Take a completed network device structure and add it to the kernel
10092 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10093 * chain. 0 is returned on success. A negative errno code is returned
10094 * on a failure to set up the device, or if the name is a duplicate.
10096 * Callers must hold the rtnl semaphore. You may want
10097 * register_netdev() instead of this.
10100 * The locking appears insufficient to guarantee two parallel registers
10101 * will not get the same name.
10104 int register_netdevice(struct net_device *dev)
10107 struct net *net = dev_net(dev);
10109 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10110 NETDEV_FEATURE_COUNT);
10111 BUG_ON(dev_boot_phase);
10116 /* When net_device's are persistent, this will be fatal. */
10117 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10120 ret = ethtool_check_ops(dev->ethtool_ops);
10124 spin_lock_init(&dev->addr_list_lock);
10125 netdev_set_addr_lockdep_class(dev);
10127 ret = dev_get_valid_name(net, dev, dev->name);
10132 dev->name_node = netdev_name_node_head_alloc(dev);
10133 if (!dev->name_node)
10136 /* Init, if this function is available */
10137 if (dev->netdev_ops->ndo_init) {
10138 ret = dev->netdev_ops->ndo_init(dev);
10142 goto err_free_name;
10146 if (((dev->hw_features | dev->features) &
10147 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10148 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10149 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10150 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10157 dev->ifindex = dev_new_index(net);
10158 else if (__dev_get_by_index(net, dev->ifindex))
10161 /* Transfer changeable features to wanted_features and enable
10162 * software offloads (GSO and GRO).
10164 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10165 dev->features |= NETIF_F_SOFT_FEATURES;
10167 if (dev->udp_tunnel_nic_info) {
10168 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10169 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10172 dev->wanted_features = dev->features & dev->hw_features;
10174 if (!(dev->flags & IFF_LOOPBACK))
10175 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10177 /* If IPv4 TCP segmentation offload is supported we should also
10178 * allow the device to enable segmenting the frame with the option
10179 * of ignoring a static IP ID value. This doesn't enable the
10180 * feature itself but allows the user to enable it later.
10182 if (dev->hw_features & NETIF_F_TSO)
10183 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10184 if (dev->vlan_features & NETIF_F_TSO)
10185 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10186 if (dev->mpls_features & NETIF_F_TSO)
10187 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10188 if (dev->hw_enc_features & NETIF_F_TSO)
10189 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10191 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10193 dev->vlan_features |= NETIF_F_HIGHDMA;
10195 /* Make NETIF_F_SG inheritable to tunnel devices.
10197 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10199 /* Make NETIF_F_SG inheritable to MPLS.
10201 dev->mpls_features |= NETIF_F_SG;
10203 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10204 ret = notifier_to_errno(ret);
10208 ret = netdev_register_kobject(dev);
10210 dev->reg_state = NETREG_UNREGISTERED;
10213 dev->reg_state = NETREG_REGISTERED;
10215 __netdev_update_features(dev);
10218 * Default initial state at registry is that the
10219 * device is present.
10222 set_bit(__LINK_STATE_PRESENT, &dev->state);
10224 linkwatch_init_dev(dev);
10226 dev_init_scheduler(dev);
10228 list_netdevice(dev);
10229 add_device_randomness(dev->dev_addr, dev->addr_len);
10231 /* If the device has permanent device address, driver should
10232 * set dev_addr and also addr_assign_type should be set to
10233 * NET_ADDR_PERM (default value).
10235 if (dev->addr_assign_type == NET_ADDR_PERM)
10236 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10238 /* Notify protocols, that a new device appeared. */
10239 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10240 ret = notifier_to_errno(ret);
10242 /* Expect explicit free_netdev() on failure */
10243 dev->needs_free_netdev = false;
10244 unregister_netdevice_queue(dev, NULL);
10248 * Prevent userspace races by waiting until the network
10249 * device is fully setup before sending notifications.
10251 if (!dev->rtnl_link_ops ||
10252 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10253 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10259 if (dev->netdev_ops->ndo_uninit)
10260 dev->netdev_ops->ndo_uninit(dev);
10261 if (dev->priv_destructor)
10262 dev->priv_destructor(dev);
10264 netdev_name_node_free(dev->name_node);
10267 EXPORT_SYMBOL(register_netdevice);
10270 * init_dummy_netdev - init a dummy network device for NAPI
10271 * @dev: device to init
10273 * This takes a network device structure and initialize the minimum
10274 * amount of fields so it can be used to schedule NAPI polls without
10275 * registering a full blown interface. This is to be used by drivers
10276 * that need to tie several hardware interfaces to a single NAPI
10277 * poll scheduler due to HW limitations.
10279 int init_dummy_netdev(struct net_device *dev)
10281 /* Clear everything. Note we don't initialize spinlocks
10282 * are they aren't supposed to be taken by any of the
10283 * NAPI code and this dummy netdev is supposed to be
10284 * only ever used for NAPI polls
10286 memset(dev, 0, sizeof(struct net_device));
10288 /* make sure we BUG if trying to hit standard
10289 * register/unregister code path
10291 dev->reg_state = NETREG_DUMMY;
10293 /* NAPI wants this */
10294 INIT_LIST_HEAD(&dev->napi_list);
10296 /* a dummy interface is started by default */
10297 set_bit(__LINK_STATE_PRESENT, &dev->state);
10298 set_bit(__LINK_STATE_START, &dev->state);
10300 /* napi_busy_loop stats accounting wants this */
10301 dev_net_set(dev, &init_net);
10303 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10304 * because users of this 'device' dont need to change
10310 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10314 * register_netdev - register a network device
10315 * @dev: device to register
10317 * Take a completed network device structure and add it to the kernel
10318 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10319 * chain. 0 is returned on success. A negative errno code is returned
10320 * on a failure to set up the device, or if the name is a duplicate.
10322 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10323 * and expands the device name if you passed a format string to
10326 int register_netdev(struct net_device *dev)
10330 if (rtnl_lock_killable())
10332 err = register_netdevice(dev);
10336 EXPORT_SYMBOL(register_netdev);
10338 int netdev_refcnt_read(const struct net_device *dev)
10342 for_each_possible_cpu(i)
10343 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10346 EXPORT_SYMBOL(netdev_refcnt_read);
10348 #define WAIT_REFS_MIN_MSECS 1
10349 #define WAIT_REFS_MAX_MSECS 250
10351 * netdev_wait_allrefs - wait until all references are gone.
10352 * @dev: target net_device
10354 * This is called when unregistering network devices.
10356 * Any protocol or device that holds a reference should register
10357 * for netdevice notification, and cleanup and put back the
10358 * reference if they receive an UNREGISTER event.
10359 * We can get stuck here if buggy protocols don't correctly
10362 static void netdev_wait_allrefs(struct net_device *dev)
10364 unsigned long rebroadcast_time, warning_time;
10365 int wait = 0, refcnt;
10367 linkwatch_forget_dev(dev);
10369 rebroadcast_time = warning_time = jiffies;
10370 refcnt = netdev_refcnt_read(dev);
10372 while (refcnt != 0) {
10373 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10376 /* Rebroadcast unregister notification */
10377 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10383 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10385 /* We must not have linkwatch events
10386 * pending on unregister. If this
10387 * happens, we simply run the queue
10388 * unscheduled, resulting in a noop
10391 linkwatch_run_queue();
10396 rebroadcast_time = jiffies;
10401 wait = WAIT_REFS_MIN_MSECS;
10404 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10407 refcnt = netdev_refcnt_read(dev);
10409 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
10410 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10411 dev->name, refcnt);
10412 warning_time = jiffies;
10417 /* The sequence is:
10421 * register_netdevice(x1);
10422 * register_netdevice(x2);
10424 * unregister_netdevice(y1);
10425 * unregister_netdevice(y2);
10431 * We are invoked by rtnl_unlock().
10432 * This allows us to deal with problems:
10433 * 1) We can delete sysfs objects which invoke hotplug
10434 * without deadlocking with linkwatch via keventd.
10435 * 2) Since we run with the RTNL semaphore not held, we can sleep
10436 * safely in order to wait for the netdev refcnt to drop to zero.
10438 * We must not return until all unregister events added during
10439 * the interval the lock was held have been completed.
10441 void netdev_run_todo(void)
10443 struct list_head list;
10444 #ifdef CONFIG_LOCKDEP
10445 struct list_head unlink_list;
10447 list_replace_init(&net_unlink_list, &unlink_list);
10449 while (!list_empty(&unlink_list)) {
10450 struct net_device *dev = list_first_entry(&unlink_list,
10453 list_del_init(&dev->unlink_list);
10454 dev->nested_level = dev->lower_level - 1;
10458 /* Snapshot list, allow later requests */
10459 list_replace_init(&net_todo_list, &list);
10464 /* Wait for rcu callbacks to finish before next phase */
10465 if (!list_empty(&list))
10468 while (!list_empty(&list)) {
10469 struct net_device *dev
10470 = list_first_entry(&list, struct net_device, todo_list);
10471 list_del(&dev->todo_list);
10473 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10474 pr_err("network todo '%s' but state %d\n",
10475 dev->name, dev->reg_state);
10480 dev->reg_state = NETREG_UNREGISTERED;
10482 netdev_wait_allrefs(dev);
10485 BUG_ON(netdev_refcnt_read(dev));
10486 BUG_ON(!list_empty(&dev->ptype_all));
10487 BUG_ON(!list_empty(&dev->ptype_specific));
10488 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10489 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10490 #if IS_ENABLED(CONFIG_DECNET)
10491 WARN_ON(dev->dn_ptr);
10493 if (dev->priv_destructor)
10494 dev->priv_destructor(dev);
10495 if (dev->needs_free_netdev)
10498 /* Report a network device has been unregistered */
10500 dev_net(dev)->dev_unreg_count--;
10502 wake_up(&netdev_unregistering_wq);
10504 /* Free network device */
10505 kobject_put(&dev->dev.kobj);
10509 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10510 * all the same fields in the same order as net_device_stats, with only
10511 * the type differing, but rtnl_link_stats64 may have additional fields
10512 * at the end for newer counters.
10514 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10515 const struct net_device_stats *netdev_stats)
10517 #if BITS_PER_LONG == 64
10518 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10519 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10520 /* zero out counters that only exist in rtnl_link_stats64 */
10521 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10522 sizeof(*stats64) - sizeof(*netdev_stats));
10524 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10525 const unsigned long *src = (const unsigned long *)netdev_stats;
10526 u64 *dst = (u64 *)stats64;
10528 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10529 for (i = 0; i < n; i++)
10531 /* zero out counters that only exist in rtnl_link_stats64 */
10532 memset((char *)stats64 + n * sizeof(u64), 0,
10533 sizeof(*stats64) - n * sizeof(u64));
10536 EXPORT_SYMBOL(netdev_stats_to_stats64);
10539 * dev_get_stats - get network device statistics
10540 * @dev: device to get statistics from
10541 * @storage: place to store stats
10543 * Get network statistics from device. Return @storage.
10544 * The device driver may provide its own method by setting
10545 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10546 * otherwise the internal statistics structure is used.
10548 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10549 struct rtnl_link_stats64 *storage)
10551 const struct net_device_ops *ops = dev->netdev_ops;
10553 if (ops->ndo_get_stats64) {
10554 memset(storage, 0, sizeof(*storage));
10555 ops->ndo_get_stats64(dev, storage);
10556 } else if (ops->ndo_get_stats) {
10557 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10559 netdev_stats_to_stats64(storage, &dev->stats);
10561 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10562 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10563 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10566 EXPORT_SYMBOL(dev_get_stats);
10569 * dev_fetch_sw_netstats - get per-cpu network device statistics
10570 * @s: place to store stats
10571 * @netstats: per-cpu network stats to read from
10573 * Read per-cpu network statistics and populate the related fields in @s.
10575 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10576 const struct pcpu_sw_netstats __percpu *netstats)
10580 for_each_possible_cpu(cpu) {
10581 const struct pcpu_sw_netstats *stats;
10582 struct pcpu_sw_netstats tmp;
10583 unsigned int start;
10585 stats = per_cpu_ptr(netstats, cpu);
10587 start = u64_stats_fetch_begin_irq(&stats->syncp);
10588 tmp.rx_packets = stats->rx_packets;
10589 tmp.rx_bytes = stats->rx_bytes;
10590 tmp.tx_packets = stats->tx_packets;
10591 tmp.tx_bytes = stats->tx_bytes;
10592 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10594 s->rx_packets += tmp.rx_packets;
10595 s->rx_bytes += tmp.rx_bytes;
10596 s->tx_packets += tmp.tx_packets;
10597 s->tx_bytes += tmp.tx_bytes;
10600 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10603 * dev_get_tstats64 - ndo_get_stats64 implementation
10604 * @dev: device to get statistics from
10605 * @s: place to store stats
10607 * Populate @s from dev->stats and dev->tstats. Can be used as
10608 * ndo_get_stats64() callback.
10610 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10612 netdev_stats_to_stats64(s, &dev->stats);
10613 dev_fetch_sw_netstats(s, dev->tstats);
10615 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10617 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10619 struct netdev_queue *queue = dev_ingress_queue(dev);
10621 #ifdef CONFIG_NET_CLS_ACT
10624 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10627 netdev_init_one_queue(dev, queue, NULL);
10628 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10629 queue->qdisc_sleeping = &noop_qdisc;
10630 rcu_assign_pointer(dev->ingress_queue, queue);
10635 static const struct ethtool_ops default_ethtool_ops;
10637 void netdev_set_default_ethtool_ops(struct net_device *dev,
10638 const struct ethtool_ops *ops)
10640 if (dev->ethtool_ops == &default_ethtool_ops)
10641 dev->ethtool_ops = ops;
10643 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10645 void netdev_freemem(struct net_device *dev)
10647 char *addr = (char *)dev - dev->padded;
10653 * alloc_netdev_mqs - allocate network device
10654 * @sizeof_priv: size of private data to allocate space for
10655 * @name: device name format string
10656 * @name_assign_type: origin of device name
10657 * @setup: callback to initialize device
10658 * @txqs: the number of TX subqueues to allocate
10659 * @rxqs: the number of RX subqueues to allocate
10661 * Allocates a struct net_device with private data area for driver use
10662 * and performs basic initialization. Also allocates subqueue structs
10663 * for each queue on the device.
10665 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10666 unsigned char name_assign_type,
10667 void (*setup)(struct net_device *),
10668 unsigned int txqs, unsigned int rxqs)
10670 struct net_device *dev;
10671 unsigned int alloc_size;
10672 struct net_device *p;
10674 BUG_ON(strlen(name) >= sizeof(dev->name));
10677 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10682 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10686 alloc_size = sizeof(struct net_device);
10688 /* ensure 32-byte alignment of private area */
10689 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10690 alloc_size += sizeof_priv;
10692 /* ensure 32-byte alignment of whole construct */
10693 alloc_size += NETDEV_ALIGN - 1;
10695 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10699 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10700 dev->padded = (char *)dev - (char *)p;
10702 dev->pcpu_refcnt = alloc_percpu(int);
10703 if (!dev->pcpu_refcnt)
10706 if (dev_addr_init(dev))
10712 dev_net_set(dev, &init_net);
10714 dev->gso_max_size = GSO_MAX_SIZE;
10715 dev->gso_max_segs = GSO_MAX_SEGS;
10716 dev->upper_level = 1;
10717 dev->lower_level = 1;
10718 #ifdef CONFIG_LOCKDEP
10719 dev->nested_level = 0;
10720 INIT_LIST_HEAD(&dev->unlink_list);
10723 INIT_LIST_HEAD(&dev->napi_list);
10724 INIT_LIST_HEAD(&dev->unreg_list);
10725 INIT_LIST_HEAD(&dev->close_list);
10726 INIT_LIST_HEAD(&dev->link_watch_list);
10727 INIT_LIST_HEAD(&dev->adj_list.upper);
10728 INIT_LIST_HEAD(&dev->adj_list.lower);
10729 INIT_LIST_HEAD(&dev->ptype_all);
10730 INIT_LIST_HEAD(&dev->ptype_specific);
10731 INIT_LIST_HEAD(&dev->net_notifier_list);
10732 #ifdef CONFIG_NET_SCHED
10733 hash_init(dev->qdisc_hash);
10735 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10738 if (!dev->tx_queue_len) {
10739 dev->priv_flags |= IFF_NO_QUEUE;
10740 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10743 dev->num_tx_queues = txqs;
10744 dev->real_num_tx_queues = txqs;
10745 if (netif_alloc_netdev_queues(dev))
10748 dev->num_rx_queues = rxqs;
10749 dev->real_num_rx_queues = rxqs;
10750 if (netif_alloc_rx_queues(dev))
10753 strcpy(dev->name, name);
10754 dev->name_assign_type = name_assign_type;
10755 dev->group = INIT_NETDEV_GROUP;
10756 if (!dev->ethtool_ops)
10757 dev->ethtool_ops = &default_ethtool_ops;
10759 nf_hook_ingress_init(dev);
10768 free_percpu(dev->pcpu_refcnt);
10770 netdev_freemem(dev);
10773 EXPORT_SYMBOL(alloc_netdev_mqs);
10776 * free_netdev - free network device
10779 * This function does the last stage of destroying an allocated device
10780 * interface. The reference to the device object is released. If this
10781 * is the last reference then it will be freed.Must be called in process
10784 void free_netdev(struct net_device *dev)
10786 struct napi_struct *p, *n;
10790 /* When called immediately after register_netdevice() failed the unwind
10791 * handling may still be dismantling the device. Handle that case by
10792 * deferring the free.
10794 if (dev->reg_state == NETREG_UNREGISTERING) {
10796 dev->needs_free_netdev = true;
10800 netif_free_tx_queues(dev);
10801 netif_free_rx_queues(dev);
10803 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10805 /* Flush device addresses */
10806 dev_addr_flush(dev);
10808 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10811 free_percpu(dev->pcpu_refcnt);
10812 dev->pcpu_refcnt = NULL;
10813 free_percpu(dev->xdp_bulkq);
10814 dev->xdp_bulkq = NULL;
10816 /* Compatibility with error handling in drivers */
10817 if (dev->reg_state == NETREG_UNINITIALIZED) {
10818 netdev_freemem(dev);
10822 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10823 dev->reg_state = NETREG_RELEASED;
10825 /* will free via device release */
10826 put_device(&dev->dev);
10828 EXPORT_SYMBOL(free_netdev);
10831 * synchronize_net - Synchronize with packet receive processing
10833 * Wait for packets currently being received to be done.
10834 * Does not block later packets from starting.
10836 void synchronize_net(void)
10839 if (rtnl_is_locked())
10840 synchronize_rcu_expedited();
10844 EXPORT_SYMBOL(synchronize_net);
10847 * unregister_netdevice_queue - remove device from the kernel
10851 * This function shuts down a device interface and removes it
10852 * from the kernel tables.
10853 * If head not NULL, device is queued to be unregistered later.
10855 * Callers must hold the rtnl semaphore. You may want
10856 * unregister_netdev() instead of this.
10859 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10864 list_move_tail(&dev->unreg_list, head);
10868 list_add(&dev->unreg_list, &single);
10869 unregister_netdevice_many(&single);
10872 EXPORT_SYMBOL(unregister_netdevice_queue);
10875 * unregister_netdevice_many - unregister many devices
10876 * @head: list of devices
10878 * Note: As most callers use a stack allocated list_head,
10879 * we force a list_del() to make sure stack wont be corrupted later.
10881 void unregister_netdevice_many(struct list_head *head)
10883 struct net_device *dev, *tmp;
10884 LIST_HEAD(close_head);
10886 BUG_ON(dev_boot_phase);
10889 if (list_empty(head))
10892 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10893 /* Some devices call without registering
10894 * for initialization unwind. Remove those
10895 * devices and proceed with the remaining.
10897 if (dev->reg_state == NETREG_UNINITIALIZED) {
10898 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10902 list_del(&dev->unreg_list);
10905 dev->dismantle = true;
10906 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10909 /* If device is running, close it first. */
10910 list_for_each_entry(dev, head, unreg_list)
10911 list_add_tail(&dev->close_list, &close_head);
10912 dev_close_many(&close_head, true);
10914 list_for_each_entry(dev, head, unreg_list) {
10915 /* And unlink it from device chain. */
10916 unlist_netdevice(dev);
10918 dev->reg_state = NETREG_UNREGISTERING;
10920 flush_all_backlogs();
10924 list_for_each_entry(dev, head, unreg_list) {
10925 struct sk_buff *skb = NULL;
10927 /* Shutdown queueing discipline. */
10930 dev_xdp_uninstall(dev);
10932 /* Notify protocols, that we are about to destroy
10933 * this device. They should clean all the things.
10935 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10937 if (!dev->rtnl_link_ops ||
10938 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10939 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10940 GFP_KERNEL, NULL, 0);
10943 * Flush the unicast and multicast chains
10948 netdev_name_node_alt_flush(dev);
10949 netdev_name_node_free(dev->name_node);
10951 if (dev->netdev_ops->ndo_uninit)
10952 dev->netdev_ops->ndo_uninit(dev);
10955 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10957 /* Notifier chain MUST detach us all upper devices. */
10958 WARN_ON(netdev_has_any_upper_dev(dev));
10959 WARN_ON(netdev_has_any_lower_dev(dev));
10961 /* Remove entries from kobject tree */
10962 netdev_unregister_kobject(dev);
10964 /* Remove XPS queueing entries */
10965 netif_reset_xps_queues_gt(dev, 0);
10971 list_for_each_entry(dev, head, unreg_list) {
10978 EXPORT_SYMBOL(unregister_netdevice_many);
10981 * unregister_netdev - remove device from the kernel
10984 * This function shuts down a device interface and removes it
10985 * from the kernel tables.
10987 * This is just a wrapper for unregister_netdevice that takes
10988 * the rtnl semaphore. In general you want to use this and not
10989 * unregister_netdevice.
10991 void unregister_netdev(struct net_device *dev)
10994 unregister_netdevice(dev);
10997 EXPORT_SYMBOL(unregister_netdev);
11000 * dev_change_net_namespace - move device to different nethost namespace
11002 * @net: network namespace
11003 * @pat: If not NULL name pattern to try if the current device name
11004 * is already taken in the destination network namespace.
11006 * This function shuts down a device interface and moves it
11007 * to a new network namespace. On success 0 is returned, on
11008 * a failure a netagive errno code is returned.
11010 * Callers must hold the rtnl semaphore.
11013 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
11015 struct net *net_old = dev_net(dev);
11016 int err, new_nsid, new_ifindex;
11020 /* Don't allow namespace local devices to be moved. */
11022 if (dev->features & NETIF_F_NETNS_LOCAL)
11025 /* Ensure the device has been registrered */
11026 if (dev->reg_state != NETREG_REGISTERED)
11029 /* Get out if there is nothing todo */
11031 if (net_eq(net_old, net))
11034 /* Pick the destination device name, and ensure
11035 * we can use it in the destination network namespace.
11038 if (__dev_get_by_name(net, dev->name)) {
11039 /* We get here if we can't use the current device name */
11042 err = dev_get_valid_name(net, dev, pat);
11048 * And now a mini version of register_netdevice unregister_netdevice.
11051 /* If device is running close it first. */
11054 /* And unlink it from device chain */
11055 unlist_netdevice(dev);
11059 /* Shutdown queueing discipline. */
11062 /* Notify protocols, that we are about to destroy
11063 * this device. They should clean all the things.
11065 * Note that dev->reg_state stays at NETREG_REGISTERED.
11066 * This is wanted because this way 8021q and macvlan know
11067 * the device is just moving and can keep their slaves up.
11069 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11072 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11073 /* If there is an ifindex conflict assign a new one */
11074 if (__dev_get_by_index(net, dev->ifindex))
11075 new_ifindex = dev_new_index(net);
11077 new_ifindex = dev->ifindex;
11079 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11083 * Flush the unicast and multicast chains
11088 /* Send a netdev-removed uevent to the old namespace */
11089 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11090 netdev_adjacent_del_links(dev);
11092 /* Move per-net netdevice notifiers that are following the netdevice */
11093 move_netdevice_notifiers_dev_net(dev, net);
11095 /* Actually switch the network namespace */
11096 dev_net_set(dev, net);
11097 dev->ifindex = new_ifindex;
11099 /* Send a netdev-add uevent to the new namespace */
11100 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11101 netdev_adjacent_add_links(dev);
11103 /* Fixup kobjects */
11104 err = device_rename(&dev->dev, dev->name);
11107 /* Adapt owner in case owning user namespace of target network
11108 * namespace is different from the original one.
11110 err = netdev_change_owner(dev, net_old, net);
11113 /* Add the device back in the hashes */
11114 list_netdevice(dev);
11116 /* Notify protocols, that a new device appeared. */
11117 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11120 * Prevent userspace races by waiting until the network
11121 * device is fully setup before sending notifications.
11123 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11130 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
11132 static int dev_cpu_dead(unsigned int oldcpu)
11134 struct sk_buff **list_skb;
11135 struct sk_buff *skb;
11137 struct softnet_data *sd, *oldsd, *remsd = NULL;
11139 local_irq_disable();
11140 cpu = smp_processor_id();
11141 sd = &per_cpu(softnet_data, cpu);
11142 oldsd = &per_cpu(softnet_data, oldcpu);
11144 /* Find end of our completion_queue. */
11145 list_skb = &sd->completion_queue;
11147 list_skb = &(*list_skb)->next;
11148 /* Append completion queue from offline CPU. */
11149 *list_skb = oldsd->completion_queue;
11150 oldsd->completion_queue = NULL;
11152 /* Append output queue from offline CPU. */
11153 if (oldsd->output_queue) {
11154 *sd->output_queue_tailp = oldsd->output_queue;
11155 sd->output_queue_tailp = oldsd->output_queue_tailp;
11156 oldsd->output_queue = NULL;
11157 oldsd->output_queue_tailp = &oldsd->output_queue;
11159 /* Append NAPI poll list from offline CPU, with one exception :
11160 * process_backlog() must be called by cpu owning percpu backlog.
11161 * We properly handle process_queue & input_pkt_queue later.
11163 while (!list_empty(&oldsd->poll_list)) {
11164 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11165 struct napi_struct,
11168 list_del_init(&napi->poll_list);
11169 if (napi->poll == process_backlog)
11172 ____napi_schedule(sd, napi);
11175 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11176 local_irq_enable();
11179 remsd = oldsd->rps_ipi_list;
11180 oldsd->rps_ipi_list = NULL;
11182 /* send out pending IPI's on offline CPU */
11183 net_rps_send_ipi(remsd);
11185 /* Process offline CPU's input_pkt_queue */
11186 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11188 input_queue_head_incr(oldsd);
11190 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11192 input_queue_head_incr(oldsd);
11199 * netdev_increment_features - increment feature set by one
11200 * @all: current feature set
11201 * @one: new feature set
11202 * @mask: mask feature set
11204 * Computes a new feature set after adding a device with feature set
11205 * @one to the master device with current feature set @all. Will not
11206 * enable anything that is off in @mask. Returns the new feature set.
11208 netdev_features_t netdev_increment_features(netdev_features_t all,
11209 netdev_features_t one, netdev_features_t mask)
11211 if (mask & NETIF_F_HW_CSUM)
11212 mask |= NETIF_F_CSUM_MASK;
11213 mask |= NETIF_F_VLAN_CHALLENGED;
11215 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11216 all &= one | ~NETIF_F_ALL_FOR_ALL;
11218 /* If one device supports hw checksumming, set for all. */
11219 if (all & NETIF_F_HW_CSUM)
11220 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11224 EXPORT_SYMBOL(netdev_increment_features);
11226 static struct hlist_head * __net_init netdev_create_hash(void)
11229 struct hlist_head *hash;
11231 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11233 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11234 INIT_HLIST_HEAD(&hash[i]);
11239 /* Initialize per network namespace state */
11240 static int __net_init netdev_init(struct net *net)
11242 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11243 8 * sizeof_field(struct napi_struct, gro_bitmask));
11245 if (net != &init_net)
11246 INIT_LIST_HEAD(&net->dev_base_head);
11248 net->dev_name_head = netdev_create_hash();
11249 if (net->dev_name_head == NULL)
11252 net->dev_index_head = netdev_create_hash();
11253 if (net->dev_index_head == NULL)
11256 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11261 kfree(net->dev_name_head);
11267 * netdev_drivername - network driver for the device
11268 * @dev: network device
11270 * Determine network driver for device.
11272 const char *netdev_drivername(const struct net_device *dev)
11274 const struct device_driver *driver;
11275 const struct device *parent;
11276 const char *empty = "";
11278 parent = dev->dev.parent;
11282 driver = parent->driver;
11283 if (driver && driver->name)
11284 return driver->name;
11288 static void __netdev_printk(const char *level, const struct net_device *dev,
11289 struct va_format *vaf)
11291 if (dev && dev->dev.parent) {
11292 dev_printk_emit(level[1] - '0',
11295 dev_driver_string(dev->dev.parent),
11296 dev_name(dev->dev.parent),
11297 netdev_name(dev), netdev_reg_state(dev),
11300 printk("%s%s%s: %pV",
11301 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11303 printk("%s(NULL net_device): %pV", level, vaf);
11307 void netdev_printk(const char *level, const struct net_device *dev,
11308 const char *format, ...)
11310 struct va_format vaf;
11313 va_start(args, format);
11318 __netdev_printk(level, dev, &vaf);
11322 EXPORT_SYMBOL(netdev_printk);
11324 #define define_netdev_printk_level(func, level) \
11325 void func(const struct net_device *dev, const char *fmt, ...) \
11327 struct va_format vaf; \
11330 va_start(args, fmt); \
11335 __netdev_printk(level, dev, &vaf); \
11339 EXPORT_SYMBOL(func);
11341 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11342 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11343 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11344 define_netdev_printk_level(netdev_err, KERN_ERR);
11345 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11346 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11347 define_netdev_printk_level(netdev_info, KERN_INFO);
11349 static void __net_exit netdev_exit(struct net *net)
11351 kfree(net->dev_name_head);
11352 kfree(net->dev_index_head);
11353 if (net != &init_net)
11354 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11357 static struct pernet_operations __net_initdata netdev_net_ops = {
11358 .init = netdev_init,
11359 .exit = netdev_exit,
11362 static void __net_exit default_device_exit(struct net *net)
11364 struct net_device *dev, *aux;
11366 * Push all migratable network devices back to the
11367 * initial network namespace
11370 for_each_netdev_safe(net, dev, aux) {
11372 char fb_name[IFNAMSIZ];
11374 /* Ignore unmoveable devices (i.e. loopback) */
11375 if (dev->features & NETIF_F_NETNS_LOCAL)
11378 /* Leave virtual devices for the generic cleanup */
11379 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11382 /* Push remaining network devices to init_net */
11383 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11384 if (__dev_get_by_name(&init_net, fb_name))
11385 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11386 err = dev_change_net_namespace(dev, &init_net, fb_name);
11388 pr_emerg("%s: failed to move %s to init_net: %d\n",
11389 __func__, dev->name, err);
11396 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11398 /* Return with the rtnl_lock held when there are no network
11399 * devices unregistering in any network namespace in net_list.
11402 bool unregistering;
11403 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11405 add_wait_queue(&netdev_unregistering_wq, &wait);
11407 unregistering = false;
11409 list_for_each_entry(net, net_list, exit_list) {
11410 if (net->dev_unreg_count > 0) {
11411 unregistering = true;
11415 if (!unregistering)
11419 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11421 remove_wait_queue(&netdev_unregistering_wq, &wait);
11424 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11426 /* At exit all network devices most be removed from a network
11427 * namespace. Do this in the reverse order of registration.
11428 * Do this across as many network namespaces as possible to
11429 * improve batching efficiency.
11431 struct net_device *dev;
11433 LIST_HEAD(dev_kill_list);
11435 /* To prevent network device cleanup code from dereferencing
11436 * loopback devices or network devices that have been freed
11437 * wait here for all pending unregistrations to complete,
11438 * before unregistring the loopback device and allowing the
11439 * network namespace be freed.
11441 * The netdev todo list containing all network devices
11442 * unregistrations that happen in default_device_exit_batch
11443 * will run in the rtnl_unlock() at the end of
11444 * default_device_exit_batch.
11446 rtnl_lock_unregistering(net_list);
11447 list_for_each_entry(net, net_list, exit_list) {
11448 for_each_netdev_reverse(net, dev) {
11449 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11450 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11452 unregister_netdevice_queue(dev, &dev_kill_list);
11455 unregister_netdevice_many(&dev_kill_list);
11459 static struct pernet_operations __net_initdata default_device_ops = {
11460 .exit = default_device_exit,
11461 .exit_batch = default_device_exit_batch,
11465 * Initialize the DEV module. At boot time this walks the device list and
11466 * unhooks any devices that fail to initialise (normally hardware not
11467 * present) and leaves us with a valid list of present and active devices.
11472 * This is called single threaded during boot, so no need
11473 * to take the rtnl semaphore.
11475 static int __init net_dev_init(void)
11477 int i, rc = -ENOMEM;
11479 BUG_ON(!dev_boot_phase);
11481 if (dev_proc_init())
11484 if (netdev_kobject_init())
11487 INIT_LIST_HEAD(&ptype_all);
11488 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11489 INIT_LIST_HEAD(&ptype_base[i]);
11491 INIT_LIST_HEAD(&offload_base);
11493 if (register_pernet_subsys(&netdev_net_ops))
11497 * Initialise the packet receive queues.
11500 for_each_possible_cpu(i) {
11501 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11502 struct softnet_data *sd = &per_cpu(softnet_data, i);
11504 INIT_WORK(flush, flush_backlog);
11506 skb_queue_head_init(&sd->input_pkt_queue);
11507 skb_queue_head_init(&sd->process_queue);
11508 #ifdef CONFIG_XFRM_OFFLOAD
11509 skb_queue_head_init(&sd->xfrm_backlog);
11511 INIT_LIST_HEAD(&sd->poll_list);
11512 sd->output_queue_tailp = &sd->output_queue;
11514 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11518 init_gro_hash(&sd->backlog);
11519 sd->backlog.poll = process_backlog;
11520 sd->backlog.weight = weight_p;
11523 dev_boot_phase = 0;
11525 /* The loopback device is special if any other network devices
11526 * is present in a network namespace the loopback device must
11527 * be present. Since we now dynamically allocate and free the
11528 * loopback device ensure this invariant is maintained by
11529 * keeping the loopback device as the first device on the
11530 * list of network devices. Ensuring the loopback devices
11531 * is the first device that appears and the last network device
11534 if (register_pernet_device(&loopback_net_ops))
11537 if (register_pernet_device(&default_device_ops))
11540 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11541 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11543 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11544 NULL, dev_cpu_dead);
11551 subsys_initcall(net_dev_init);